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);
128 static noinline int cow_file_range(struct btrfs_inode *inode,
129 struct page *locked_page,
130 u64 start, u64 end, u64 *done_offset,
131 bool keep_locked, bool no_inline);
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 btrfs_mark_ordered_io_finished
427 * through btrfs_mark_ordered_io_finished() on it
428 * in run_delalloc_range() for the error handling, which will
429 * clear page Ordered and run the ordered extent accounting.
431 * Here we can't just clear the Ordered bit, or
432 * btrfs_mark_ordered_io_finished() would skip the accounting
433 * for the page range, and the ordered extent will never finish.
435 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
439 page = find_get_page(inode->vfs_inode.i_mapping, index);
445 * Here we just clear all Ordered bits for every page in the
446 * range, then btrfs_mark_ordered_io_finished() will handle
447 * the ordered extent accounting for the range.
449 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
455 /* The locked page covers the full range, nothing needs to be done */
456 if (bytes + offset <= page_start + PAGE_SIZE)
459 * In case this page belongs to the delalloc range being
460 * instantiated then skip it, since the first page of a range is
461 * going to be properly cleaned up by the caller of
464 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
465 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
466 offset = page_offset(locked_page) + PAGE_SIZE;
470 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
473 static int btrfs_dirty_inode(struct btrfs_inode *inode);
475 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
476 struct btrfs_new_inode_args *args)
480 if (args->default_acl) {
481 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
487 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
491 if (!args->default_acl && !args->acl)
492 cache_no_acl(args->inode);
493 return btrfs_xattr_security_init(trans, args->inode, args->dir,
494 &args->dentry->d_name);
498 * this does all the hard work for inserting an inline extent into
499 * the btree. The caller should have done a btrfs_drop_extents so that
500 * no overlapping inline items exist in the btree
502 static int insert_inline_extent(struct btrfs_trans_handle *trans,
503 struct btrfs_path *path,
504 struct btrfs_inode *inode, bool extent_inserted,
505 size_t size, size_t compressed_size,
507 struct page **compressed_pages,
510 struct btrfs_root *root = inode->root;
511 struct extent_buffer *leaf;
512 struct page *page = NULL;
515 struct btrfs_file_extent_item *ei;
517 size_t cur_size = size;
520 ASSERT((compressed_size > 0 && compressed_pages) ||
521 (compressed_size == 0 && !compressed_pages));
523 if (compressed_size && compressed_pages)
524 cur_size = compressed_size;
526 if (!extent_inserted) {
527 struct btrfs_key key;
530 key.objectid = btrfs_ino(inode);
532 key.type = BTRFS_EXTENT_DATA_KEY;
534 datasize = btrfs_file_extent_calc_inline_size(cur_size);
535 ret = btrfs_insert_empty_item(trans, root, path, &key,
540 leaf = path->nodes[0];
541 ei = btrfs_item_ptr(leaf, path->slots[0],
542 struct btrfs_file_extent_item);
543 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
544 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
545 btrfs_set_file_extent_encryption(leaf, ei, 0);
546 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
547 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
548 ptr = btrfs_file_extent_inline_start(ei);
550 if (compress_type != BTRFS_COMPRESS_NONE) {
553 while (compressed_size > 0) {
554 cpage = compressed_pages[i];
555 cur_size = min_t(unsigned long, compressed_size,
558 kaddr = kmap_local_page(cpage);
559 write_extent_buffer(leaf, kaddr, ptr, cur_size);
564 compressed_size -= cur_size;
566 btrfs_set_file_extent_compression(leaf, ei,
569 page = find_get_page(inode->vfs_inode.i_mapping, 0);
570 btrfs_set_file_extent_compression(leaf, ei, 0);
571 kaddr = kmap_local_page(page);
572 write_extent_buffer(leaf, kaddr, ptr, size);
576 btrfs_mark_buffer_dirty(leaf);
577 btrfs_release_path(path);
580 * We align size to sectorsize for inline extents just for simplicity
583 ret = btrfs_inode_set_file_extent_range(inode, 0,
584 ALIGN(size, root->fs_info->sectorsize));
589 * We're an inline extent, so nobody can extend the file past i_size
590 * without locking a page we already have locked.
592 * We must do any i_size and inode updates before we unlock the pages.
593 * Otherwise we could end up racing with unlink.
595 i_size = i_size_read(&inode->vfs_inode);
596 if (update_i_size && size > i_size) {
597 i_size_write(&inode->vfs_inode, size);
600 inode->disk_i_size = i_size;
608 * conditionally insert an inline extent into the file. This
609 * does the checks required to make sure the data is small enough
610 * to fit as an inline extent.
612 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
613 size_t compressed_size,
615 struct page **compressed_pages,
618 struct btrfs_drop_extents_args drop_args = { 0 };
619 struct btrfs_root *root = inode->root;
620 struct btrfs_fs_info *fs_info = root->fs_info;
621 struct btrfs_trans_handle *trans;
622 u64 data_len = (compressed_size ?: size);
624 struct btrfs_path *path;
627 * We can create an inline extent if it ends at or beyond the current
628 * i_size, is no larger than a sector (decompressed), and the (possibly
629 * compressed) data fits in a leaf and the configured maximum inline
632 if (size < i_size_read(&inode->vfs_inode) ||
633 size > fs_info->sectorsize ||
634 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
635 data_len > fs_info->max_inline)
638 path = btrfs_alloc_path();
642 trans = btrfs_join_transaction(root);
644 btrfs_free_path(path);
645 return PTR_ERR(trans);
647 trans->block_rsv = &inode->block_rsv;
649 drop_args.path = path;
651 drop_args.end = fs_info->sectorsize;
652 drop_args.drop_cache = true;
653 drop_args.replace_extent = true;
654 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
655 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
657 btrfs_abort_transaction(trans, ret);
661 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
662 size, compressed_size, compress_type,
663 compressed_pages, update_i_size);
664 if (ret && ret != -ENOSPC) {
665 btrfs_abort_transaction(trans, ret);
667 } else if (ret == -ENOSPC) {
672 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
673 ret = btrfs_update_inode(trans, root, inode);
674 if (ret && ret != -ENOSPC) {
675 btrfs_abort_transaction(trans, ret);
677 } else if (ret == -ENOSPC) {
682 btrfs_set_inode_full_sync(inode);
685 * Don't forget to free the reserved space, as for inlined extent
686 * it won't count as data extent, free them directly here.
687 * And at reserve time, it's always aligned to page size, so
688 * just free one page here.
690 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
691 btrfs_free_path(path);
692 btrfs_end_transaction(trans);
696 struct async_extent {
701 unsigned long nr_pages;
703 struct list_head list;
707 struct btrfs_inode *inode;
708 struct page *locked_page;
711 blk_opf_t write_flags;
712 struct list_head extents;
713 struct cgroup_subsys_state *blkcg_css;
714 struct btrfs_work work;
715 struct async_cow *async_cow;
720 struct async_chunk chunks[];
723 static noinline int add_async_extent(struct async_chunk *cow,
724 u64 start, u64 ram_size,
727 unsigned long nr_pages,
730 struct async_extent *async_extent;
732 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
733 BUG_ON(!async_extent); /* -ENOMEM */
734 async_extent->start = start;
735 async_extent->ram_size = ram_size;
736 async_extent->compressed_size = compressed_size;
737 async_extent->pages = pages;
738 async_extent->nr_pages = nr_pages;
739 async_extent->compress_type = compress_type;
740 list_add_tail(&async_extent->list, &cow->extents);
745 * Check if the inode needs to be submitted to compression, based on mount
746 * options, defragmentation, properties or heuristics.
748 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
751 struct btrfs_fs_info *fs_info = inode->root->fs_info;
753 if (!btrfs_inode_can_compress(inode)) {
754 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
755 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
760 * Special check for subpage.
762 * We lock the full page then run each delalloc range in the page, thus
763 * for the following case, we will hit some subpage specific corner case:
766 * | |///////| |///////|
769 * In above case, both range A and range B will try to unlock the full
770 * page [0, 64K), causing the one finished later will have page
771 * unlocked already, triggering various page lock requirement BUG_ON()s.
773 * So here we add an artificial limit that subpage compression can only
774 * if the range is fully page aligned.
776 * In theory we only need to ensure the first page is fully covered, but
777 * the tailing partial page will be locked until the full compression
778 * finishes, delaying the write of other range.
780 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
781 * first to prevent any submitted async extent to unlock the full page.
782 * By this, we can ensure for subpage case that only the last async_cow
783 * will unlock the full page.
785 if (fs_info->sectorsize < PAGE_SIZE) {
786 if (!PAGE_ALIGNED(start) ||
787 !PAGE_ALIGNED(end + 1))
792 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
795 if (inode->defrag_compress)
797 /* bad compression ratios */
798 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
800 if (btrfs_test_opt(fs_info, COMPRESS) ||
801 inode->flags & BTRFS_INODE_COMPRESS ||
802 inode->prop_compress)
803 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
807 static inline void inode_should_defrag(struct btrfs_inode *inode,
808 u64 start, u64 end, u64 num_bytes, u32 small_write)
810 /* If this is a small write inside eof, kick off a defrag */
811 if (num_bytes < small_write &&
812 (start > 0 || end + 1 < inode->disk_i_size))
813 btrfs_add_inode_defrag(NULL, inode, small_write);
817 * we create compressed extents in two phases. The first
818 * phase compresses a range of pages that have already been
819 * locked (both pages and state bits are locked).
821 * This is done inside an ordered work queue, and the compression
822 * is spread across many cpus. The actual IO submission is step
823 * two, and the ordered work queue takes care of making sure that
824 * happens in the same order things were put onto the queue by
825 * writepages and friends.
827 * If this code finds it can't get good compression, it puts an
828 * entry onto the work queue to write the uncompressed bytes. This
829 * makes sure that both compressed inodes and uncompressed inodes
830 * are written in the same order that the flusher thread sent them
833 static noinline int compress_file_range(struct async_chunk *async_chunk)
835 struct btrfs_inode *inode = async_chunk->inode;
836 struct btrfs_fs_info *fs_info = inode->root->fs_info;
837 struct address_space *mapping = inode->vfs_inode.i_mapping;
838 u64 blocksize = fs_info->sectorsize;
839 u64 start = async_chunk->start;
840 u64 end = async_chunk->end;
844 struct page **pages = NULL;
845 unsigned long nr_pages;
846 unsigned long total_compressed = 0;
847 unsigned long total_in = 0;
850 int compress_type = fs_info->compress_type;
851 int compressed_extents = 0;
854 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
857 * We need to save i_size before now because it could change in between
858 * us evaluating the size and assigning it. This is because we lock and
859 * unlock the page in truncate and fallocate, and then modify the i_size
862 * The barriers are to emulate READ_ONCE, remove that once i_size_read
866 i_size = i_size_read(&inode->vfs_inode);
868 actual_end = min_t(u64, i_size, end + 1);
871 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
872 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
875 * we don't want to send crud past the end of i_size through
876 * compression, that's just a waste of CPU time. So, if the
877 * end of the file is before the start of our current
878 * requested range of bytes, we bail out to the uncompressed
879 * cleanup code that can deal with all of this.
881 * It isn't really the fastest way to fix things, but this is a
882 * very uncommon corner.
884 if (actual_end <= start)
885 goto cleanup_and_bail_uncompressed;
887 total_compressed = actual_end - start;
890 * Skip compression for a small file range(<=blocksize) that
891 * isn't an inline extent, since it doesn't save disk space at all.
893 if (total_compressed <= blocksize &&
894 (start > 0 || end + 1 < inode->disk_i_size))
895 goto cleanup_and_bail_uncompressed;
898 * For subpage case, we require full page alignment for the sector
900 * Thus we must also check against @actual_end, not just @end.
902 if (blocksize < PAGE_SIZE) {
903 if (!PAGE_ALIGNED(start) ||
904 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
905 goto cleanup_and_bail_uncompressed;
908 total_compressed = min_t(unsigned long, total_compressed,
909 BTRFS_MAX_UNCOMPRESSED);
914 * we do compression for mount -o compress and when the
915 * inode has not been flagged as nocompress. This flag can
916 * change at any time if we discover bad compression ratios.
918 if (inode_need_compress(inode, start, end)) {
920 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
922 /* just bail out to the uncompressed code */
927 if (inode->defrag_compress)
928 compress_type = inode->defrag_compress;
929 else if (inode->prop_compress)
930 compress_type = inode->prop_compress;
933 * we need to call clear_page_dirty_for_io on each
934 * page in the range. Otherwise applications with the file
935 * mmap'd can wander in and change the page contents while
936 * we are compressing them.
938 * If the compression fails for any reason, we set the pages
939 * dirty again later on.
941 * Note that the remaining part is redirtied, the start pointer
942 * has moved, the end is the original one.
945 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
949 /* Compression level is applied here and only here */
950 ret = btrfs_compress_pages(
951 compress_type | (fs_info->compress_level << 4),
959 unsigned long offset = offset_in_page(total_compressed);
960 struct page *page = pages[nr_pages - 1];
962 /* zero the tail end of the last page, we might be
963 * sending it down to disk
966 memzero_page(page, offset, PAGE_SIZE - offset);
972 * Check cow_file_range() for why we don't even try to create inline
973 * extent for subpage case.
975 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
976 /* lets try to make an inline extent */
977 if (ret || total_in < actual_end) {
978 /* we didn't compress the entire range, try
979 * to make an uncompressed inline extent.
981 ret = cow_file_range_inline(inode, actual_end,
982 0, BTRFS_COMPRESS_NONE,
985 /* try making a compressed inline extent */
986 ret = cow_file_range_inline(inode, actual_end,
988 compress_type, pages,
992 unsigned long clear_flags = EXTENT_DELALLOC |
993 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
994 EXTENT_DO_ACCOUNTING;
997 mapping_set_error(mapping, -EIO);
1000 * inline extent creation worked or returned error,
1001 * we don't need to create any more async work items.
1002 * Unlock and free up our temp pages.
1004 * We use DO_ACCOUNTING here because we need the
1005 * delalloc_release_metadata to be done _after_ we drop
1006 * our outstanding extent for clearing delalloc for this
1009 extent_clear_unlock_delalloc(inode, start, end,
1013 PAGE_START_WRITEBACK |
1014 PAGE_END_WRITEBACK);
1017 * Ensure we only free the compressed pages if we have
1018 * them allocated, as we can still reach here with
1019 * inode_need_compress() == false.
1022 for (i = 0; i < nr_pages; i++) {
1023 WARN_ON(pages[i]->mapping);
1032 if (will_compress) {
1034 * we aren't doing an inline extent round the compressed size
1035 * up to a block size boundary so the allocator does sane
1038 total_compressed = ALIGN(total_compressed, blocksize);
1041 * one last check to make sure the compression is really a
1042 * win, compare the page count read with the blocks on disk,
1043 * compression must free at least one sector size
1045 total_in = round_up(total_in, fs_info->sectorsize);
1046 if (total_compressed + blocksize <= total_in) {
1047 compressed_extents++;
1050 * The async work queues will take care of doing actual
1051 * allocation on disk for these compressed pages, and
1052 * will submit them to the elevator.
1054 add_async_extent(async_chunk, start, total_in,
1055 total_compressed, pages, nr_pages,
1058 if (start + total_in < end) {
1064 return compressed_extents;
1069 * the compression code ran but failed to make things smaller,
1070 * free any pages it allocated and our page pointer array
1072 for (i = 0; i < nr_pages; i++) {
1073 WARN_ON(pages[i]->mapping);
1078 total_compressed = 0;
1081 /* flag the file so we don't compress in the future */
1082 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
1083 !(inode->prop_compress)) {
1084 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1087 cleanup_and_bail_uncompressed:
1089 * No compression, but we still need to write the pages in the file
1090 * we've been given so far. redirty the locked page if it corresponds
1091 * to our extent and set things up for the async work queue to run
1092 * cow_file_range to do the normal delalloc dance.
1094 if (async_chunk->locked_page &&
1095 (page_offset(async_chunk->locked_page) >= start &&
1096 page_offset(async_chunk->locked_page)) <= end) {
1097 __set_page_dirty_nobuffers(async_chunk->locked_page);
1098 /* unlocked later on in the async handlers */
1102 extent_range_redirty_for_io(&inode->vfs_inode, start, end);
1103 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1104 BTRFS_COMPRESS_NONE);
1105 compressed_extents++;
1107 return compressed_extents;
1110 static void free_async_extent_pages(struct async_extent *async_extent)
1114 if (!async_extent->pages)
1117 for (i = 0; i < async_extent->nr_pages; i++) {
1118 WARN_ON(async_extent->pages[i]->mapping);
1119 put_page(async_extent->pages[i]);
1121 kfree(async_extent->pages);
1122 async_extent->nr_pages = 0;
1123 async_extent->pages = NULL;
1126 static void submit_uncompressed_range(struct btrfs_inode *inode,
1127 struct async_extent *async_extent,
1128 struct page *locked_page)
1130 u64 start = async_extent->start;
1131 u64 end = async_extent->start + async_extent->ram_size - 1;
1133 struct writeback_control wbc = {
1134 .sync_mode = WB_SYNC_ALL,
1135 .range_start = start,
1137 .no_cgroup_owner = 1,
1141 * Call cow_file_range() to run the delalloc range directly, since we
1142 * won't go to NOCOW or async path again.
1144 * Also we call cow_file_range() with @unlock_page == 0, so that we
1145 * can directly submit them without interruption.
1147 ret = cow_file_range(inode, locked_page, start, end, NULL, true, false);
1148 /* Inline extent inserted, page gets unlocked and everything is done */
1153 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1155 const u64 page_start = page_offset(locked_page);
1157 set_page_writeback(locked_page);
1158 end_page_writeback(locked_page);
1159 btrfs_mark_ordered_io_finished(inode, locked_page,
1160 page_start, PAGE_SIZE,
1162 btrfs_page_clear_uptodate(inode->root->fs_info,
1163 locked_page, page_start,
1165 mapping_set_error(locked_page->mapping, ret);
1166 unlock_page(locked_page);
1171 /* All pages will be unlocked, including @locked_page */
1172 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1173 extent_write_locked_range(&inode->vfs_inode, start, end, &wbc);
1174 wbc_detach_inode(&wbc);
1177 static void submit_one_async_extent(struct async_chunk *async_chunk,
1178 struct async_extent *async_extent,
1181 struct btrfs_inode *inode = async_chunk->inode;
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 submit_uncompressed_range(inode, async_extent, locked_page);
1215 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1216 async_extent->compressed_size,
1217 async_extent->compressed_size,
1218 0, *alloc_hint, &ins, 1, 1);
1221 * Here we used to try again by going back to non-compressed
1222 * path for ENOSPC. But we can't reserve space even for
1223 * compressed size, how could it work for uncompressed size
1224 * which requires larger size? So here we directly go error
1230 /* Here we're doing allocation and writeback of the compressed pages */
1231 em = create_io_em(inode, start,
1232 async_extent->ram_size, /* len */
1233 start, /* orig_start */
1234 ins.objectid, /* block_start */
1235 ins.offset, /* block_len */
1236 ins.offset, /* orig_block_len */
1237 async_extent->ram_size, /* ram_bytes */
1238 async_extent->compress_type,
1239 BTRFS_ORDERED_COMPRESSED);
1242 goto out_free_reserve;
1244 free_extent_map(em);
1246 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1247 async_extent->ram_size, /* num_bytes */
1248 async_extent->ram_size, /* ram_bytes */
1249 ins.objectid, /* disk_bytenr */
1250 ins.offset, /* disk_num_bytes */
1252 1 << BTRFS_ORDERED_COMPRESSED,
1253 async_extent->compress_type);
1254 if (IS_ERR(ordered)) {
1255 btrfs_drop_extent_map_range(inode, start, end, false);
1256 ret = PTR_ERR(ordered);
1257 goto out_free_reserve;
1259 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1261 /* Clear dirty, set writeback and unlock the pages. */
1262 extent_clear_unlock_delalloc(inode, start, end,
1263 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1264 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1265 btrfs_submit_compressed_write(ordered,
1266 async_extent->pages, /* compressed_pages */
1267 async_extent->nr_pages,
1268 async_chunk->write_flags, true);
1269 *alloc_hint = ins.objectid + ins.offset;
1271 if (async_chunk->blkcg_css)
1272 kthread_associate_blkcg(NULL);
1273 kfree(async_extent);
1277 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1278 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1280 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1281 extent_clear_unlock_delalloc(inode, start, end,
1282 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1283 EXTENT_DELALLOC_NEW |
1284 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1285 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1286 PAGE_END_WRITEBACK);
1287 free_async_extent_pages(async_extent);
1288 if (async_chunk->blkcg_css)
1289 kthread_associate_blkcg(NULL);
1290 btrfs_debug(fs_info,
1291 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1292 root->root_key.objectid, btrfs_ino(inode), start,
1293 async_extent->ram_size, ret);
1294 kfree(async_extent);
1298 * Phase two of compressed writeback. This is the ordered portion of the code,
1299 * which only gets called in the order the work was queued. We walk all the
1300 * async extents created by compress_file_range and send them down to the disk.
1302 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1304 struct async_extent *async_extent;
1307 while (!list_empty(&async_chunk->extents)) {
1308 async_extent = list_entry(async_chunk->extents.next,
1309 struct async_extent, list);
1310 list_del(&async_extent->list);
1312 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1316 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1319 struct extent_map_tree *em_tree = &inode->extent_tree;
1320 struct extent_map *em;
1323 read_lock(&em_tree->lock);
1324 em = search_extent_mapping(em_tree, start, num_bytes);
1327 * if block start isn't an actual block number then find the
1328 * first block in this inode and use that as a hint. If that
1329 * block is also bogus then just don't worry about it.
1331 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1332 free_extent_map(em);
1333 em = search_extent_mapping(em_tree, 0, 0);
1334 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1335 alloc_hint = em->block_start;
1337 free_extent_map(em);
1339 alloc_hint = em->block_start;
1340 free_extent_map(em);
1343 read_unlock(&em_tree->lock);
1349 * when extent_io.c finds a delayed allocation range in the file,
1350 * the call backs end up in this code. The basic idea is to
1351 * allocate extents on disk for the range, and create ordered data structs
1352 * in ram to track those extents.
1354 * locked_page is the page that writepage had locked already. We use
1355 * it to make sure we don't do extra locks or unlocks.
1357 * When this function fails, it unlocks all pages except @locked_page.
1359 * When this function successfully creates an inline extent, it returns 1 and
1360 * unlocks all pages including locked_page and starts I/O on them.
1361 * (In reality inline extents are limited to a single page, so locked_page is
1362 * the only page handled anyway).
1364 * When this function succeed and creates a normal extent, the page locking
1365 * status depends on the passed in flags:
1367 * - If @keep_locked is set, all pages are kept locked.
1368 * - Else all pages except for @locked_page are unlocked.
1370 * When a failure happens in the second or later iteration of the
1371 * while-loop, the ordered extents created in previous iterations are kept
1372 * intact. So, the caller must clean them up by calling
1373 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1376 static noinline int cow_file_range(struct btrfs_inode *inode,
1377 struct page *locked_page, u64 start, u64 end,
1379 bool keep_locked, bool no_inline)
1381 struct btrfs_root *root = inode->root;
1382 struct btrfs_fs_info *fs_info = root->fs_info;
1384 u64 orig_start = start;
1386 unsigned long ram_size;
1387 u64 cur_alloc_size = 0;
1389 u64 blocksize = fs_info->sectorsize;
1390 struct btrfs_key ins;
1391 struct extent_map *em;
1392 unsigned clear_bits;
1393 unsigned long page_ops;
1394 bool extent_reserved = false;
1397 if (btrfs_is_free_space_inode(inode)) {
1402 num_bytes = ALIGN(end - start + 1, blocksize);
1403 num_bytes = max(blocksize, num_bytes);
1404 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1406 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1409 * Due to the page size limit, for subpage we can only trigger the
1410 * writeback for the dirty sectors of page, that means data writeback
1411 * is doing more writeback than what we want.
1413 * This is especially unexpected for some call sites like fallocate,
1414 * where we only increase i_size after everything is done.
1415 * This means we can trigger inline extent even if we didn't want to.
1416 * So here we skip inline extent creation completely.
1418 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1419 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1422 /* lets try to make an inline extent */
1423 ret = cow_file_range_inline(inode, actual_end, 0,
1424 BTRFS_COMPRESS_NONE, NULL, false);
1427 * We use DO_ACCOUNTING here because we need the
1428 * delalloc_release_metadata to be run _after_ we drop
1429 * our outstanding extent for clearing delalloc for this
1432 extent_clear_unlock_delalloc(inode, start, end,
1434 EXTENT_LOCKED | EXTENT_DELALLOC |
1435 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1436 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1437 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1439 * locked_page is locked by the caller of
1440 * writepage_delalloc(), not locked by
1441 * __process_pages_contig().
1443 * We can't let __process_pages_contig() to unlock it,
1444 * as it doesn't have any subpage::writers recorded.
1446 * Here we manually unlock the page, since the caller
1447 * can't determine if it's an inline extent or a
1448 * compressed extent.
1450 unlock_page(locked_page);
1452 } else if (ret < 0) {
1457 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1460 * Relocation relies on the relocated extents to have exactly the same
1461 * size as the original extents. Normally writeback for relocation data
1462 * extents follows a NOCOW path because relocation preallocates the
1463 * extents. However, due to an operation such as scrub turning a block
1464 * group to RO mode, it may fallback to COW mode, so we must make sure
1465 * an extent allocated during COW has exactly the requested size and can
1466 * not be split into smaller extents, otherwise relocation breaks and
1467 * fails during the stage where it updates the bytenr of file extent
1470 if (btrfs_is_data_reloc_root(root))
1471 min_alloc_size = num_bytes;
1473 min_alloc_size = fs_info->sectorsize;
1475 while (num_bytes > 0) {
1476 struct btrfs_ordered_extent *ordered;
1478 cur_alloc_size = num_bytes;
1479 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1480 min_alloc_size, 0, alloc_hint,
1484 cur_alloc_size = ins.offset;
1485 extent_reserved = true;
1487 ram_size = ins.offset;
1488 em = create_io_em(inode, start, ins.offset, /* len */
1489 start, /* orig_start */
1490 ins.objectid, /* block_start */
1491 ins.offset, /* block_len */
1492 ins.offset, /* orig_block_len */
1493 ram_size, /* ram_bytes */
1494 BTRFS_COMPRESS_NONE, /* compress_type */
1495 BTRFS_ORDERED_REGULAR /* type */);
1500 free_extent_map(em);
1502 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1503 ram_size, ins.objectid, cur_alloc_size,
1504 0, 1 << BTRFS_ORDERED_REGULAR,
1505 BTRFS_COMPRESS_NONE);
1506 if (IS_ERR(ordered)) {
1507 ret = PTR_ERR(ordered);
1508 goto out_drop_extent_cache;
1511 if (btrfs_is_data_reloc_root(root)) {
1512 ret = btrfs_reloc_clone_csums(ordered);
1515 * Only drop cache here, and process as normal.
1517 * We must not allow extent_clear_unlock_delalloc()
1518 * at out_unlock label to free meta of this ordered
1519 * extent, as its meta should be freed by
1520 * btrfs_finish_ordered_io().
1522 * So we must continue until @start is increased to
1523 * skip current ordered extent.
1526 btrfs_drop_extent_map_range(inode, start,
1527 start + ram_size - 1,
1530 btrfs_put_ordered_extent(ordered);
1532 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1535 * We're not doing compressed IO, don't unlock the first page
1536 * (which the caller expects to stay locked), don't clear any
1537 * dirty bits and don't set any writeback bits
1539 * Do set the Ordered (Private2) bit so we know this page was
1540 * properly setup for writepage.
1542 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1543 page_ops |= PAGE_SET_ORDERED;
1545 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1547 EXTENT_LOCKED | EXTENT_DELALLOC,
1549 if (num_bytes < cur_alloc_size)
1552 num_bytes -= cur_alloc_size;
1553 alloc_hint = ins.objectid + ins.offset;
1554 start += cur_alloc_size;
1555 extent_reserved = false;
1558 * btrfs_reloc_clone_csums() error, since start is increased
1559 * extent_clear_unlock_delalloc() at out_unlock label won't
1560 * free metadata of current ordered extent, we're OK to exit.
1567 out_drop_extent_cache:
1568 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1570 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1571 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1574 * If done_offset is non-NULL and ret == -EAGAIN, we expect the
1575 * caller to write out the successfully allocated region and retry.
1577 if (done_offset && ret == -EAGAIN) {
1578 if (orig_start < start)
1579 *done_offset = start - 1;
1581 *done_offset = start;
1583 } else if (ret == -EAGAIN) {
1584 /* Convert to -ENOSPC since the caller cannot retry. */
1589 * Now, we have three regions to clean up:
1591 * |-------(1)----|---(2)---|-------------(3)----------|
1592 * `- orig_start `- start `- start + cur_alloc_size `- end
1594 * We process each region below.
1597 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1598 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1599 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1602 * For the range (1). We have already instantiated the ordered extents
1603 * for this region. They are cleaned up by
1604 * btrfs_cleanup_ordered_extents() in e.g,
1605 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1606 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1607 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1610 * However, in case of @keep_locked, we still need to unlock the pages
1611 * (except @locked_page) to ensure all the pages are unlocked.
1613 if (keep_locked && orig_start < start) {
1615 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1616 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1617 locked_page, 0, page_ops);
1621 * For the range (2). If we reserved an extent for our delalloc range
1622 * (or a subrange) and failed to create the respective ordered extent,
1623 * then it means that when we reserved the extent we decremented the
1624 * extent's size from the data space_info's bytes_may_use counter and
1625 * incremented the space_info's bytes_reserved counter by the same
1626 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1627 * to decrement again the data space_info's bytes_may_use counter,
1628 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1630 if (extent_reserved) {
1631 extent_clear_unlock_delalloc(inode, start,
1632 start + cur_alloc_size - 1,
1636 start += cur_alloc_size;
1640 * For the range (3). We never touched the region. In addition to the
1641 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1642 * space_info's bytes_may_use counter, reserved in
1643 * btrfs_check_data_free_space().
1646 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1647 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1648 clear_bits, page_ops);
1654 * work queue call back to started compression on a file and pages
1656 static noinline void async_cow_start(struct btrfs_work *work)
1658 struct async_chunk *async_chunk;
1659 int compressed_extents;
1661 async_chunk = container_of(work, struct async_chunk, work);
1663 compressed_extents = compress_file_range(async_chunk);
1664 if (compressed_extents == 0) {
1665 btrfs_add_delayed_iput(async_chunk->inode);
1666 async_chunk->inode = NULL;
1671 * work queue call back to submit previously compressed pages
1673 static noinline void async_cow_submit(struct btrfs_work *work)
1675 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1677 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1678 unsigned long nr_pages;
1680 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1684 * ->inode could be NULL if async_chunk_start has failed to compress,
1685 * in which case we don't have anything to submit, yet we need to
1686 * always adjust ->async_delalloc_pages as its paired with the init
1687 * happening in run_delalloc_compressed
1689 if (async_chunk->inode)
1690 submit_compressed_extents(async_chunk);
1692 /* atomic_sub_return implies a barrier */
1693 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1695 cond_wake_up_nomb(&fs_info->async_submit_wait);
1698 static noinline void async_cow_free(struct btrfs_work *work)
1700 struct async_chunk *async_chunk;
1701 struct async_cow *async_cow;
1703 async_chunk = container_of(work, struct async_chunk, work);
1704 if (async_chunk->inode)
1705 btrfs_add_delayed_iput(async_chunk->inode);
1706 if (async_chunk->blkcg_css)
1707 css_put(async_chunk->blkcg_css);
1709 async_cow = async_chunk->async_cow;
1710 if (atomic_dec_and_test(&async_cow->num_chunks))
1714 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1715 struct page *locked_page, u64 start,
1716 u64 end, struct writeback_control *wbc)
1718 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1719 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1720 struct async_cow *ctx;
1721 struct async_chunk *async_chunk;
1722 unsigned long nr_pages;
1723 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1726 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1728 nofs_flag = memalloc_nofs_save();
1729 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1730 memalloc_nofs_restore(nofs_flag);
1734 unlock_extent(&inode->io_tree, start, end, NULL);
1735 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1737 async_chunk = ctx->chunks;
1738 atomic_set(&ctx->num_chunks, num_chunks);
1740 for (i = 0; i < num_chunks; i++) {
1741 u64 cur_end = min(end, start + SZ_512K - 1);
1744 * igrab is called higher up in the call chain, take only the
1745 * lightweight reference for the callback lifetime
1747 ihold(&inode->vfs_inode);
1748 async_chunk[i].async_cow = ctx;
1749 async_chunk[i].inode = inode;
1750 async_chunk[i].start = start;
1751 async_chunk[i].end = cur_end;
1752 async_chunk[i].write_flags = write_flags;
1753 INIT_LIST_HEAD(&async_chunk[i].extents);
1756 * The locked_page comes all the way from writepage and its
1757 * the original page we were actually given. As we spread
1758 * this large delalloc region across multiple async_chunk
1759 * structs, only the first struct needs a pointer to locked_page
1761 * This way we don't need racey decisions about who is supposed
1766 * Depending on the compressibility, the pages might or
1767 * might not go through async. We want all of them to
1768 * be accounted against wbc once. Let's do it here
1769 * before the paths diverge. wbc accounting is used
1770 * only for foreign writeback detection and doesn't
1771 * need full accuracy. Just account the whole thing
1772 * against the first page.
1774 wbc_account_cgroup_owner(wbc, locked_page,
1776 async_chunk[i].locked_page = locked_page;
1779 async_chunk[i].locked_page = NULL;
1782 if (blkcg_css != blkcg_root_css) {
1784 async_chunk[i].blkcg_css = blkcg_css;
1785 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1787 async_chunk[i].blkcg_css = NULL;
1790 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1791 async_cow_submit, async_cow_free);
1793 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1794 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1796 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1798 start = cur_end + 1;
1803 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1804 struct page *locked_page, u64 start,
1805 u64 end, struct writeback_control *wbc)
1807 u64 done_offset = end;
1809 bool locked_page_done = false;
1811 while (start <= end) {
1812 ret = cow_file_range(inode, locked_page, start, end, &done_offset,
1814 if (ret && ret != -EAGAIN)
1820 if (done_offset == start) {
1821 wait_on_bit_io(&inode->root->fs_info->flags,
1822 BTRFS_FS_NEED_ZONE_FINISH,
1823 TASK_UNINTERRUPTIBLE);
1827 if (!locked_page_done) {
1828 __set_page_dirty_nobuffers(locked_page);
1829 account_page_redirty(locked_page);
1831 locked_page_done = true;
1832 extent_write_locked_range(&inode->vfs_inode, start, done_offset,
1834 start = done_offset + 1;
1840 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1841 u64 bytenr, u64 num_bytes, bool nowait)
1843 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1844 struct btrfs_ordered_sum *sums;
1848 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1850 if (ret == 0 && list_empty(&list))
1853 while (!list_empty(&list)) {
1854 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1855 list_del(&sums->list);
1863 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1864 const u64 start, const u64 end)
1866 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1867 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1868 const u64 range_bytes = end + 1 - start;
1869 struct extent_io_tree *io_tree = &inode->io_tree;
1870 u64 range_start = start;
1875 * If EXTENT_NORESERVE is set it means that when the buffered write was
1876 * made we had not enough available data space and therefore we did not
1877 * reserve data space for it, since we though we could do NOCOW for the
1878 * respective file range (either there is prealloc extent or the inode
1879 * has the NOCOW bit set).
1881 * However when we need to fallback to COW mode (because for example the
1882 * block group for the corresponding extent was turned to RO mode by a
1883 * scrub or relocation) we need to do the following:
1885 * 1) We increment the bytes_may_use counter of the data space info.
1886 * If COW succeeds, it allocates a new data extent and after doing
1887 * that it decrements the space info's bytes_may_use counter and
1888 * increments its bytes_reserved counter by the same amount (we do
1889 * this at btrfs_add_reserved_bytes()). So we need to increment the
1890 * bytes_may_use counter to compensate (when space is reserved at
1891 * buffered write time, the bytes_may_use counter is incremented);
1893 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1894 * that if the COW path fails for any reason, it decrements (through
1895 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1896 * data space info, which we incremented in the step above.
1898 * If we need to fallback to cow and the inode corresponds to a free
1899 * space cache inode or an inode of the data relocation tree, we must
1900 * also increment bytes_may_use of the data space_info for the same
1901 * reason. Space caches and relocated data extents always get a prealloc
1902 * extent for them, however scrub or balance may have set the block
1903 * group that contains that extent to RO mode and therefore force COW
1904 * when starting writeback.
1906 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1907 EXTENT_NORESERVE, 0, NULL);
1908 if (count > 0 || is_space_ino || is_reloc_ino) {
1910 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1911 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1913 if (is_space_ino || is_reloc_ino)
1914 bytes = range_bytes;
1916 spin_lock(&sinfo->lock);
1917 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1918 spin_unlock(&sinfo->lock);
1921 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1926 * Don't try to create inline extents, as a mix of inline extent that
1927 * is written out and unlocked directly and a normal NOCOW extent
1930 ret = cow_file_range(inode, locked_page, start, end, NULL, false, true);
1935 struct can_nocow_file_extent_args {
1938 /* Start file offset of the range we want to NOCOW. */
1940 /* End file offset (inclusive) of the range we want to NOCOW. */
1942 bool writeback_path;
1945 * Free the path passed to can_nocow_file_extent() once it's not needed
1950 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1955 /* Number of bytes that can be written to in NOCOW mode. */
1960 * Check if we can NOCOW the file extent that the path points to.
1961 * This function may return with the path released, so the caller should check
1962 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1964 * Returns: < 0 on error
1965 * 0 if we can not NOCOW
1968 static int can_nocow_file_extent(struct btrfs_path *path,
1969 struct btrfs_key *key,
1970 struct btrfs_inode *inode,
1971 struct can_nocow_file_extent_args *args)
1973 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1974 struct extent_buffer *leaf = path->nodes[0];
1975 struct btrfs_root *root = inode->root;
1976 struct btrfs_file_extent_item *fi;
1981 bool nowait = path->nowait;
1983 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1984 extent_type = btrfs_file_extent_type(leaf, fi);
1986 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1989 /* Can't access these fields unless we know it's not an inline extent. */
1990 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1991 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1992 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1994 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1995 extent_type == BTRFS_FILE_EXTENT_REG)
1999 * If the extent was created before the generation where the last snapshot
2000 * for its subvolume was created, then this implies the extent is shared,
2001 * hence we must COW.
2003 if (!args->strict &&
2004 btrfs_file_extent_generation(leaf, fi) <=
2005 btrfs_root_last_snapshot(&root->root_item))
2008 /* An explicit hole, must COW. */
2009 if (args->disk_bytenr == 0)
2012 /* Compressed/encrypted/encoded extents must be COWed. */
2013 if (btrfs_file_extent_compression(leaf, fi) ||
2014 btrfs_file_extent_encryption(leaf, fi) ||
2015 btrfs_file_extent_other_encoding(leaf, fi))
2018 extent_end = btrfs_file_extent_end(path);
2021 * The following checks can be expensive, as they need to take other
2022 * locks and do btree or rbtree searches, so release the path to avoid
2023 * blocking other tasks for too long.
2025 btrfs_release_path(path);
2027 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
2028 key->offset - args->extent_offset,
2029 args->disk_bytenr, args->strict, path);
2030 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2034 if (args->free_path) {
2036 * We don't need the path anymore, plus through the
2037 * csum_exist_in_range() call below we will end up allocating
2038 * another path. So free the path to avoid unnecessary extra
2041 btrfs_free_path(path);
2045 /* If there are pending snapshots for this root, we must COW. */
2046 if (args->writeback_path && !is_freespace_inode &&
2047 atomic_read(&root->snapshot_force_cow))
2050 args->disk_bytenr += args->extent_offset;
2051 args->disk_bytenr += args->start - key->offset;
2052 args->num_bytes = min(args->end + 1, extent_end) - args->start;
2055 * Force COW if csums exist in the range. This ensures that csums for a
2056 * given extent are either valid or do not exist.
2058 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
2060 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2066 if (args->free_path && path)
2067 btrfs_free_path(path);
2069 return ret < 0 ? ret : can_nocow;
2073 * when nowcow writeback call back. This checks for snapshots or COW copies
2074 * of the extents that exist in the file, and COWs the file as required.
2076 * If no cow copies or snapshots exist, we write directly to the existing
2079 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
2080 struct page *locked_page,
2081 const u64 start, const u64 end)
2083 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2084 struct btrfs_root *root = inode->root;
2085 struct btrfs_path *path;
2086 u64 cow_start = (u64)-1;
2087 u64 cur_offset = start;
2089 bool check_prev = true;
2090 u64 ino = btrfs_ino(inode);
2091 struct btrfs_block_group *bg;
2093 struct can_nocow_file_extent_args nocow_args = { 0 };
2095 path = btrfs_alloc_path();
2097 extent_clear_unlock_delalloc(inode, start, end, locked_page,
2098 EXTENT_LOCKED | EXTENT_DELALLOC |
2099 EXTENT_DO_ACCOUNTING |
2100 EXTENT_DEFRAG, PAGE_UNLOCK |
2101 PAGE_START_WRITEBACK |
2102 PAGE_END_WRITEBACK);
2106 nocow_args.end = end;
2107 nocow_args.writeback_path = true;
2110 struct btrfs_ordered_extent *ordered;
2111 struct btrfs_key found_key;
2112 struct btrfs_file_extent_item *fi;
2113 struct extent_buffer *leaf;
2122 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2128 * If there is no extent for our range when doing the initial
2129 * search, then go back to the previous slot as it will be the
2130 * one containing the search offset
2132 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2133 leaf = path->nodes[0];
2134 btrfs_item_key_to_cpu(leaf, &found_key,
2135 path->slots[0] - 1);
2136 if (found_key.objectid == ino &&
2137 found_key.type == BTRFS_EXTENT_DATA_KEY)
2142 /* Go to next leaf if we have exhausted the current one */
2143 leaf = path->nodes[0];
2144 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2145 ret = btrfs_next_leaf(root, path);
2147 if (cow_start != (u64)-1)
2148 cur_offset = cow_start;
2153 leaf = path->nodes[0];
2156 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2158 /* Didn't find anything for our INO */
2159 if (found_key.objectid > ino)
2162 * Keep searching until we find an EXTENT_ITEM or there are no
2163 * more extents for this inode
2165 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2166 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2171 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2172 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2173 found_key.offset > end)
2177 * If the found extent starts after requested offset, then
2178 * adjust extent_end to be right before this extent begins
2180 if (found_key.offset > cur_offset) {
2181 extent_end = found_key.offset;
2187 * Found extent which begins before our range and potentially
2190 fi = btrfs_item_ptr(leaf, path->slots[0],
2191 struct btrfs_file_extent_item);
2192 extent_type = btrfs_file_extent_type(leaf, fi);
2193 /* If this is triggered then we have a memory corruption. */
2194 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2195 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2199 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2200 extent_end = btrfs_file_extent_end(path);
2203 * If the extent we got ends before our current offset, skip to
2206 if (extent_end <= cur_offset) {
2211 nocow_args.start = cur_offset;
2212 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2214 if (cow_start != (u64)-1)
2215 cur_offset = cow_start;
2217 } else if (ret == 0) {
2222 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2227 * If nocow is false then record the beginning of the range
2228 * that needs to be COWed
2231 if (cow_start == (u64)-1)
2232 cow_start = cur_offset;
2233 cur_offset = extent_end;
2234 if (cur_offset > end)
2236 if (!path->nodes[0])
2243 * COW range from cow_start to found_key.offset - 1. As the key
2244 * will contain the beginning of the first extent that can be
2245 * NOCOW, following one which needs to be COW'ed
2247 if (cow_start != (u64)-1) {
2248 ret = fallback_to_cow(inode, locked_page,
2249 cow_start, found_key.offset - 1);
2252 cow_start = (u64)-1;
2255 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2256 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2258 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2259 struct extent_map *em;
2261 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2263 nocow_args.disk_bytenr, /* block_start */
2264 nocow_args.num_bytes, /* block_len */
2265 nocow_args.disk_num_bytes, /* orig_block_len */
2266 ram_bytes, BTRFS_COMPRESS_NONE,
2267 BTRFS_ORDERED_PREALLOC);
2272 free_extent_map(em);
2275 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2276 nocow_args.num_bytes, nocow_args.num_bytes,
2277 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2279 ? (1 << BTRFS_ORDERED_PREALLOC)
2280 : (1 << BTRFS_ORDERED_NOCOW),
2281 BTRFS_COMPRESS_NONE);
2282 if (IS_ERR(ordered)) {
2284 btrfs_drop_extent_map_range(inode, cur_offset,
2287 ret = PTR_ERR(ordered);
2292 btrfs_dec_nocow_writers(bg);
2296 if (btrfs_is_data_reloc_root(root))
2298 * Error handled later, as we must prevent
2299 * extent_clear_unlock_delalloc() in error handler
2300 * from freeing metadata of created ordered extent.
2302 ret = btrfs_reloc_clone_csums(ordered);
2303 btrfs_put_ordered_extent(ordered);
2305 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2306 locked_page, EXTENT_LOCKED |
2308 EXTENT_CLEAR_DATA_RESV,
2309 PAGE_UNLOCK | PAGE_SET_ORDERED);
2311 cur_offset = extent_end;
2314 * btrfs_reloc_clone_csums() error, now we're OK to call error
2315 * handler, as metadata for created ordered extent will only
2316 * be freed by btrfs_finish_ordered_io().
2320 if (cur_offset > end)
2323 btrfs_release_path(path);
2325 if (cur_offset <= end && cow_start == (u64)-1)
2326 cow_start = cur_offset;
2328 if (cow_start != (u64)-1) {
2330 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2337 btrfs_dec_nocow_writers(bg);
2339 if (ret && cur_offset < end)
2340 extent_clear_unlock_delalloc(inode, cur_offset, end,
2341 locked_page, EXTENT_LOCKED |
2342 EXTENT_DELALLOC | EXTENT_DEFRAG |
2343 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2344 PAGE_START_WRITEBACK |
2345 PAGE_END_WRITEBACK);
2346 btrfs_free_path(path);
2350 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2352 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2353 if (inode->defrag_bytes &&
2354 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2363 * Function to process delayed allocation (create CoW) for ranges which are
2364 * being touched for the first time.
2366 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2367 u64 start, u64 end, struct writeback_control *wbc)
2369 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2373 * The range must cover part of the @locked_page, or a return of 1
2374 * can confuse the caller.
2376 ASSERT(!(end <= page_offset(locked_page) ||
2377 start >= page_offset(locked_page) + PAGE_SIZE));
2379 if (should_nocow(inode, start, end)) {
2381 * Normally on a zoned device we're only doing COW writes, but
2382 * in case of relocation on a zoned filesystem we have taken
2383 * precaution, that we're only writing sequentially. It's safe
2384 * to use run_delalloc_nocow() here, like for regular
2385 * preallocated inodes.
2387 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2388 ret = run_delalloc_nocow(inode, locked_page, start, end);
2392 if (btrfs_inode_can_compress(inode) &&
2393 inode_need_compress(inode, start, end) &&
2394 run_delalloc_compressed(inode, locked_page, start, end, wbc))
2398 ret = run_delalloc_zoned(inode, locked_page, start, end, wbc);
2400 ret = cow_file_range(inode, locked_page, start, end, NULL,
2405 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2410 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2411 struct extent_state *orig, u64 split)
2413 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2416 /* not delalloc, ignore it */
2417 if (!(orig->state & EXTENT_DELALLOC))
2420 size = orig->end - orig->start + 1;
2421 if (size > fs_info->max_extent_size) {
2426 * See the explanation in btrfs_merge_delalloc_extent, the same
2427 * applies here, just in reverse.
2429 new_size = orig->end - split + 1;
2430 num_extents = count_max_extents(fs_info, new_size);
2431 new_size = split - orig->start;
2432 num_extents += count_max_extents(fs_info, new_size);
2433 if (count_max_extents(fs_info, size) >= num_extents)
2437 spin_lock(&inode->lock);
2438 btrfs_mod_outstanding_extents(inode, 1);
2439 spin_unlock(&inode->lock);
2443 * Handle merged delayed allocation extents so we can keep track of new extents
2444 * that are just merged onto old extents, such as when we are doing sequential
2445 * writes, so we can properly account for the metadata space we'll need.
2447 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2448 struct extent_state *other)
2450 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2451 u64 new_size, old_size;
2454 /* not delalloc, ignore it */
2455 if (!(other->state & EXTENT_DELALLOC))
2458 if (new->start > other->start)
2459 new_size = new->end - other->start + 1;
2461 new_size = other->end - new->start + 1;
2463 /* we're not bigger than the max, unreserve the space and go */
2464 if (new_size <= fs_info->max_extent_size) {
2465 spin_lock(&inode->lock);
2466 btrfs_mod_outstanding_extents(inode, -1);
2467 spin_unlock(&inode->lock);
2472 * We have to add up either side to figure out how many extents were
2473 * accounted for before we merged into one big extent. If the number of
2474 * extents we accounted for is <= the amount we need for the new range
2475 * then we can return, otherwise drop. Think of it like this
2479 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2480 * need 2 outstanding extents, on one side we have 1 and the other side
2481 * we have 1 so they are == and we can return. But in this case
2483 * [MAX_SIZE+4k][MAX_SIZE+4k]
2485 * Each range on their own accounts for 2 extents, but merged together
2486 * they are only 3 extents worth of accounting, so we need to drop in
2489 old_size = other->end - other->start + 1;
2490 num_extents = count_max_extents(fs_info, old_size);
2491 old_size = new->end - new->start + 1;
2492 num_extents += count_max_extents(fs_info, old_size);
2493 if (count_max_extents(fs_info, new_size) >= num_extents)
2496 spin_lock(&inode->lock);
2497 btrfs_mod_outstanding_extents(inode, -1);
2498 spin_unlock(&inode->lock);
2501 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2502 struct btrfs_inode *inode)
2504 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2506 spin_lock(&root->delalloc_lock);
2507 if (list_empty(&inode->delalloc_inodes)) {
2508 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2509 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2510 root->nr_delalloc_inodes++;
2511 if (root->nr_delalloc_inodes == 1) {
2512 spin_lock(&fs_info->delalloc_root_lock);
2513 BUG_ON(!list_empty(&root->delalloc_root));
2514 list_add_tail(&root->delalloc_root,
2515 &fs_info->delalloc_roots);
2516 spin_unlock(&fs_info->delalloc_root_lock);
2519 spin_unlock(&root->delalloc_lock);
2522 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2523 struct btrfs_inode *inode)
2525 struct btrfs_fs_info *fs_info = root->fs_info;
2527 if (!list_empty(&inode->delalloc_inodes)) {
2528 list_del_init(&inode->delalloc_inodes);
2529 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2530 &inode->runtime_flags);
2531 root->nr_delalloc_inodes--;
2532 if (!root->nr_delalloc_inodes) {
2533 ASSERT(list_empty(&root->delalloc_inodes));
2534 spin_lock(&fs_info->delalloc_root_lock);
2535 BUG_ON(list_empty(&root->delalloc_root));
2536 list_del_init(&root->delalloc_root);
2537 spin_unlock(&fs_info->delalloc_root_lock);
2542 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2543 struct btrfs_inode *inode)
2545 spin_lock(&root->delalloc_lock);
2546 __btrfs_del_delalloc_inode(root, inode);
2547 spin_unlock(&root->delalloc_lock);
2551 * Properly track delayed allocation bytes in the inode and to maintain the
2552 * list of inodes that have pending delalloc work to be done.
2554 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2557 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2559 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2562 * set_bit and clear bit hooks normally require _irqsave/restore
2563 * but in this case, we are only testing for the DELALLOC
2564 * bit, which is only set or cleared with irqs on
2566 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2567 struct btrfs_root *root = inode->root;
2568 u64 len = state->end + 1 - state->start;
2569 u32 num_extents = count_max_extents(fs_info, len);
2570 bool do_list = !btrfs_is_free_space_inode(inode);
2572 spin_lock(&inode->lock);
2573 btrfs_mod_outstanding_extents(inode, num_extents);
2574 spin_unlock(&inode->lock);
2576 /* For sanity tests */
2577 if (btrfs_is_testing(fs_info))
2580 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2581 fs_info->delalloc_batch);
2582 spin_lock(&inode->lock);
2583 inode->delalloc_bytes += len;
2584 if (bits & EXTENT_DEFRAG)
2585 inode->defrag_bytes += len;
2586 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2587 &inode->runtime_flags))
2588 btrfs_add_delalloc_inodes(root, inode);
2589 spin_unlock(&inode->lock);
2592 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2593 (bits & EXTENT_DELALLOC_NEW)) {
2594 spin_lock(&inode->lock);
2595 inode->new_delalloc_bytes += state->end + 1 - state->start;
2596 spin_unlock(&inode->lock);
2601 * Once a range is no longer delalloc this function ensures that proper
2602 * accounting happens.
2604 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2605 struct extent_state *state, u32 bits)
2607 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2608 u64 len = state->end + 1 - state->start;
2609 u32 num_extents = count_max_extents(fs_info, len);
2611 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2612 spin_lock(&inode->lock);
2613 inode->defrag_bytes -= len;
2614 spin_unlock(&inode->lock);
2618 * set_bit and clear bit hooks normally require _irqsave/restore
2619 * but in this case, we are only testing for the DELALLOC
2620 * bit, which is only set or cleared with irqs on
2622 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2623 struct btrfs_root *root = inode->root;
2624 bool do_list = !btrfs_is_free_space_inode(inode);
2626 spin_lock(&inode->lock);
2627 btrfs_mod_outstanding_extents(inode, -num_extents);
2628 spin_unlock(&inode->lock);
2631 * We don't reserve metadata space for space cache inodes so we
2632 * don't need to call delalloc_release_metadata if there is an
2635 if (bits & EXTENT_CLEAR_META_RESV &&
2636 root != fs_info->tree_root)
2637 btrfs_delalloc_release_metadata(inode, len, false);
2639 /* For sanity tests. */
2640 if (btrfs_is_testing(fs_info))
2643 if (!btrfs_is_data_reloc_root(root) &&
2644 do_list && !(state->state & EXTENT_NORESERVE) &&
2645 (bits & EXTENT_CLEAR_DATA_RESV))
2646 btrfs_free_reserved_data_space_noquota(fs_info, len);
2648 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2649 fs_info->delalloc_batch);
2650 spin_lock(&inode->lock);
2651 inode->delalloc_bytes -= len;
2652 if (do_list && inode->delalloc_bytes == 0 &&
2653 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2654 &inode->runtime_flags))
2655 btrfs_del_delalloc_inode(root, inode);
2656 spin_unlock(&inode->lock);
2659 if ((state->state & EXTENT_DELALLOC_NEW) &&
2660 (bits & EXTENT_DELALLOC_NEW)) {
2661 spin_lock(&inode->lock);
2662 ASSERT(inode->new_delalloc_bytes >= len);
2663 inode->new_delalloc_bytes -= len;
2664 if (bits & EXTENT_ADD_INODE_BYTES)
2665 inode_add_bytes(&inode->vfs_inode, len);
2666 spin_unlock(&inode->lock);
2670 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2671 struct btrfs_ordered_extent *ordered)
2673 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2674 u64 len = bbio->bio.bi_iter.bi_size;
2675 struct btrfs_ordered_extent *new;
2678 /* Must always be called for the beginning of an ordered extent. */
2679 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2682 /* No need to split if the ordered extent covers the entire bio. */
2683 if (ordered->disk_num_bytes == len) {
2684 refcount_inc(&ordered->refs);
2685 bbio->ordered = ordered;
2690 * Don't split the extent_map for NOCOW extents, as we're writing into
2691 * a pre-existing one.
2693 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2694 ret = split_extent_map(bbio->inode, bbio->file_offset,
2695 ordered->num_bytes, len,
2696 ordered->disk_bytenr);
2701 new = btrfs_split_ordered_extent(ordered, len);
2703 return PTR_ERR(new);
2704 bbio->ordered = new;
2709 * given a list of ordered sums record them in the inode. This happens
2710 * at IO completion time based on sums calculated at bio submission time.
2712 static int add_pending_csums(struct btrfs_trans_handle *trans,
2713 struct list_head *list)
2715 struct btrfs_ordered_sum *sum;
2716 struct btrfs_root *csum_root = NULL;
2719 list_for_each_entry(sum, list, list) {
2720 trans->adding_csums = true;
2722 csum_root = btrfs_csum_root(trans->fs_info,
2724 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2725 trans->adding_csums = false;
2732 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2735 struct extent_state **cached_state)
2737 u64 search_start = start;
2738 const u64 end = start + len - 1;
2740 while (search_start < end) {
2741 const u64 search_len = end - search_start + 1;
2742 struct extent_map *em;
2746 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2750 if (em->block_start != EXTENT_MAP_HOLE)
2754 if (em->start < search_start)
2755 em_len -= search_start - em->start;
2756 if (em_len > search_len)
2757 em_len = search_len;
2759 ret = set_extent_bit(&inode->io_tree, search_start,
2760 search_start + em_len - 1,
2761 EXTENT_DELALLOC_NEW, cached_state);
2763 search_start = extent_map_end(em);
2764 free_extent_map(em);
2771 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2772 unsigned int extra_bits,
2773 struct extent_state **cached_state)
2775 WARN_ON(PAGE_ALIGNED(end));
2777 if (start >= i_size_read(&inode->vfs_inode) &&
2778 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2780 * There can't be any extents following eof in this case so just
2781 * set the delalloc new bit for the range directly.
2783 extra_bits |= EXTENT_DELALLOC_NEW;
2787 ret = btrfs_find_new_delalloc_bytes(inode, start,
2794 return set_extent_bit(&inode->io_tree, start, end,
2795 EXTENT_DELALLOC | extra_bits, cached_state);
2798 /* see btrfs_writepage_start_hook for details on why this is required */
2799 struct btrfs_writepage_fixup {
2801 struct btrfs_inode *inode;
2802 struct btrfs_work work;
2805 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2807 struct btrfs_writepage_fixup *fixup =
2808 container_of(work, struct btrfs_writepage_fixup, work);
2809 struct btrfs_ordered_extent *ordered;
2810 struct extent_state *cached_state = NULL;
2811 struct extent_changeset *data_reserved = NULL;
2812 struct page *page = fixup->page;
2813 struct btrfs_inode *inode = fixup->inode;
2814 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2815 u64 page_start = page_offset(page);
2816 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2818 bool free_delalloc_space = true;
2821 * This is similar to page_mkwrite, we need to reserve the space before
2822 * we take the page lock.
2824 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2830 * Before we queued this fixup, we took a reference on the page.
2831 * page->mapping may go NULL, but it shouldn't be moved to a different
2834 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2836 * Unfortunately this is a little tricky, either
2838 * 1) We got here and our page had already been dealt with and
2839 * we reserved our space, thus ret == 0, so we need to just
2840 * drop our space reservation and bail. This can happen the
2841 * first time we come into the fixup worker, or could happen
2842 * while waiting for the ordered extent.
2843 * 2) Our page was already dealt with, but we happened to get an
2844 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2845 * this case we obviously don't have anything to release, but
2846 * because the page was already dealt with we don't want to
2847 * mark the page with an error, so make sure we're resetting
2848 * ret to 0. This is why we have this check _before_ the ret
2849 * check, because we do not want to have a surprise ENOSPC
2850 * when the page was already properly dealt with.
2853 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2854 btrfs_delalloc_release_space(inode, data_reserved,
2855 page_start, PAGE_SIZE,
2863 * We can't mess with the page state unless it is locked, so now that
2864 * it is locked bail if we failed to make our space reservation.
2869 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2871 /* already ordered? We're done */
2872 if (PageOrdered(page))
2875 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2877 unlock_extent(&inode->io_tree, page_start, page_end,
2880 btrfs_start_ordered_extent(ordered);
2881 btrfs_put_ordered_extent(ordered);
2885 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2891 * Everything went as planned, we're now the owner of a dirty page with
2892 * delayed allocation bits set and space reserved for our COW
2895 * The page was dirty when we started, nothing should have cleaned it.
2897 BUG_ON(!PageDirty(page));
2898 free_delalloc_space = false;
2900 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2901 if (free_delalloc_space)
2902 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2904 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2908 * We hit ENOSPC or other errors. Update the mapping and page
2909 * to reflect the errors and clean the page.
2911 mapping_set_error(page->mapping, ret);
2912 btrfs_mark_ordered_io_finished(inode, page, page_start,
2914 btrfs_page_clear_uptodate(fs_info, page, page_start, PAGE_SIZE);
2915 clear_page_dirty_for_io(page);
2917 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
2921 extent_changeset_free(data_reserved);
2923 * As a precaution, do a delayed iput in case it would be the last iput
2924 * that could need flushing space. Recursing back to fixup worker would
2927 btrfs_add_delayed_iput(inode);
2931 * There are a few paths in the higher layers of the kernel that directly
2932 * set the page dirty bit without asking the filesystem if it is a
2933 * good idea. This causes problems because we want to make sure COW
2934 * properly happens and the data=ordered rules are followed.
2936 * In our case any range that doesn't have the ORDERED bit set
2937 * hasn't been properly setup for IO. We kick off an async process
2938 * to fix it up. The async helper will wait for ordered extents, set
2939 * the delalloc bit and make it safe to write the page.
2941 int btrfs_writepage_cow_fixup(struct page *page)
2943 struct inode *inode = page->mapping->host;
2944 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2945 struct btrfs_writepage_fixup *fixup;
2947 /* This page has ordered extent covering it already */
2948 if (PageOrdered(page))
2952 * PageChecked is set below when we create a fixup worker for this page,
2953 * don't try to create another one if we're already PageChecked()
2955 * The extent_io writepage code will redirty the page if we send back
2958 if (PageChecked(page))
2961 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2966 * We are already holding a reference to this inode from
2967 * write_cache_pages. We need to hold it because the space reservation
2968 * takes place outside of the page lock, and we can't trust
2969 * page->mapping outside of the page lock.
2972 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2974 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2976 fixup->inode = BTRFS_I(inode);
2977 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2982 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2983 struct btrfs_inode *inode, u64 file_pos,
2984 struct btrfs_file_extent_item *stack_fi,
2985 const bool update_inode_bytes,
2986 u64 qgroup_reserved)
2988 struct btrfs_root *root = inode->root;
2989 const u64 sectorsize = root->fs_info->sectorsize;
2990 struct btrfs_path *path;
2991 struct extent_buffer *leaf;
2992 struct btrfs_key ins;
2993 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2994 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2995 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2996 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2997 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2998 struct btrfs_drop_extents_args drop_args = { 0 };
3001 path = btrfs_alloc_path();
3006 * we may be replacing one extent in the tree with another.
3007 * The new extent is pinned in the extent map, and we don't want
3008 * to drop it from the cache until it is completely in the btree.
3010 * So, tell btrfs_drop_extents to leave this extent in the cache.
3011 * the caller is expected to unpin it and allow it to be merged
3014 drop_args.path = path;
3015 drop_args.start = file_pos;
3016 drop_args.end = file_pos + num_bytes;
3017 drop_args.replace_extent = true;
3018 drop_args.extent_item_size = sizeof(*stack_fi);
3019 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
3023 if (!drop_args.extent_inserted) {
3024 ins.objectid = btrfs_ino(inode);
3025 ins.offset = file_pos;
3026 ins.type = BTRFS_EXTENT_DATA_KEY;
3028 ret = btrfs_insert_empty_item(trans, root, path, &ins,
3033 leaf = path->nodes[0];
3034 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
3035 write_extent_buffer(leaf, stack_fi,
3036 btrfs_item_ptr_offset(leaf, path->slots[0]),
3037 sizeof(struct btrfs_file_extent_item));
3039 btrfs_mark_buffer_dirty(leaf);
3040 btrfs_release_path(path);
3043 * If we dropped an inline extent here, we know the range where it is
3044 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
3045 * number of bytes only for that range containing the inline extent.
3046 * The remaining of the range will be processed when clearning the
3047 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
3049 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
3050 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
3052 inline_size = drop_args.bytes_found - inline_size;
3053 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
3054 drop_args.bytes_found -= inline_size;
3055 num_bytes -= sectorsize;
3058 if (update_inode_bytes)
3059 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
3061 ins.objectid = disk_bytenr;
3062 ins.offset = disk_num_bytes;
3063 ins.type = BTRFS_EXTENT_ITEM_KEY;
3065 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3069 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3071 qgroup_reserved, &ins);
3073 btrfs_free_path(path);
3078 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3081 struct btrfs_block_group *cache;
3083 cache = btrfs_lookup_block_group(fs_info, start);
3086 spin_lock(&cache->lock);
3087 cache->delalloc_bytes -= len;
3088 spin_unlock(&cache->lock);
3090 btrfs_put_block_group(cache);
3093 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3094 struct btrfs_ordered_extent *oe)
3096 struct btrfs_file_extent_item stack_fi;
3097 bool update_inode_bytes;
3098 u64 num_bytes = oe->num_bytes;
3099 u64 ram_bytes = oe->ram_bytes;
3101 memset(&stack_fi, 0, sizeof(stack_fi));
3102 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3103 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3104 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3105 oe->disk_num_bytes);
3106 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3107 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3108 num_bytes = oe->truncated_len;
3109 ram_bytes = num_bytes;
3111 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3112 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3113 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3114 /* Encryption and other encoding is reserved and all 0 */
3117 * For delalloc, when completing an ordered extent we update the inode's
3118 * bytes when clearing the range in the inode's io tree, so pass false
3119 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3120 * except if the ordered extent was truncated.
3122 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3123 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3124 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3126 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3127 oe->file_offset, &stack_fi,
3128 update_inode_bytes, oe->qgroup_rsv);
3132 * As ordered data IO finishes, this gets called so we can finish
3133 * an ordered extent if the range of bytes in the file it covers are
3136 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3138 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3139 struct btrfs_root *root = inode->root;
3140 struct btrfs_fs_info *fs_info = root->fs_info;
3141 struct btrfs_trans_handle *trans = NULL;
3142 struct extent_io_tree *io_tree = &inode->io_tree;
3143 struct extent_state *cached_state = NULL;
3145 int compress_type = 0;
3147 u64 logical_len = ordered_extent->num_bytes;
3148 bool freespace_inode;
3149 bool truncated = false;
3150 bool clear_reserved_extent = true;
3151 unsigned int clear_bits = EXTENT_DEFRAG;
3153 start = ordered_extent->file_offset;
3154 end = start + ordered_extent->num_bytes - 1;
3156 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3157 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3158 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3159 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3160 clear_bits |= EXTENT_DELALLOC_NEW;
3162 freespace_inode = btrfs_is_free_space_inode(inode);
3163 if (!freespace_inode)
3164 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3166 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3171 if (btrfs_is_zoned(fs_info))
3172 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3173 ordered_extent->disk_num_bytes);
3175 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3177 logical_len = ordered_extent->truncated_len;
3178 /* Truncated the entire extent, don't bother adding */
3183 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3184 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3186 btrfs_inode_safe_disk_i_size_write(inode, 0);
3187 if (freespace_inode)
3188 trans = btrfs_join_transaction_spacecache(root);
3190 trans = btrfs_join_transaction(root);
3191 if (IS_ERR(trans)) {
3192 ret = PTR_ERR(trans);
3196 trans->block_rsv = &inode->block_rsv;
3197 ret = btrfs_update_inode_fallback(trans, root, inode);
3198 if (ret) /* -ENOMEM or corruption */
3199 btrfs_abort_transaction(trans, ret);
3203 clear_bits |= EXTENT_LOCKED;
3204 lock_extent(io_tree, start, end, &cached_state);
3206 if (freespace_inode)
3207 trans = btrfs_join_transaction_spacecache(root);
3209 trans = btrfs_join_transaction(root);
3210 if (IS_ERR(trans)) {
3211 ret = PTR_ERR(trans);
3216 trans->block_rsv = &inode->block_rsv;
3218 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3219 compress_type = ordered_extent->compress_type;
3220 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3221 BUG_ON(compress_type);
3222 ret = btrfs_mark_extent_written(trans, inode,
3223 ordered_extent->file_offset,
3224 ordered_extent->file_offset +
3226 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3227 ordered_extent->disk_num_bytes);
3229 BUG_ON(root == fs_info->tree_root);
3230 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3232 clear_reserved_extent = false;
3233 btrfs_release_delalloc_bytes(fs_info,
3234 ordered_extent->disk_bytenr,
3235 ordered_extent->disk_num_bytes);
3238 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3239 ordered_extent->num_bytes, trans->transid);
3241 btrfs_abort_transaction(trans, ret);
3245 ret = add_pending_csums(trans, &ordered_extent->list);
3247 btrfs_abort_transaction(trans, ret);
3252 * If this is a new delalloc range, clear its new delalloc flag to
3253 * update the inode's number of bytes. This needs to be done first
3254 * before updating the inode item.
3256 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3257 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3258 clear_extent_bit(&inode->io_tree, start, end,
3259 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3262 btrfs_inode_safe_disk_i_size_write(inode, 0);
3263 ret = btrfs_update_inode_fallback(trans, root, inode);
3264 if (ret) { /* -ENOMEM or corruption */
3265 btrfs_abort_transaction(trans, ret);
3270 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3274 btrfs_end_transaction(trans);
3276 if (ret || truncated) {
3277 u64 unwritten_start = start;
3280 * If we failed to finish this ordered extent for any reason we
3281 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3282 * extent, and mark the inode with the error if it wasn't
3283 * already set. Any error during writeback would have already
3284 * set the mapping error, so we need to set it if we're the ones
3285 * marking this ordered extent as failed.
3287 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3288 &ordered_extent->flags))
3289 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3292 unwritten_start += logical_len;
3293 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3295 /* Drop extent maps for the part of the extent we didn't write. */
3296 btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3299 * If the ordered extent had an IOERR or something else went
3300 * wrong we need to return the space for this ordered extent
3301 * back to the allocator. We only free the extent in the
3302 * truncated case if we didn't write out the extent at all.
3304 * If we made it past insert_reserved_file_extent before we
3305 * errored out then we don't need to do this as the accounting
3306 * has already been done.
3308 if ((ret || !logical_len) &&
3309 clear_reserved_extent &&
3310 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3311 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3313 * Discard the range before returning it back to the
3316 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3317 btrfs_discard_extent(fs_info,
3318 ordered_extent->disk_bytenr,
3319 ordered_extent->disk_num_bytes,
3321 btrfs_free_reserved_extent(fs_info,
3322 ordered_extent->disk_bytenr,
3323 ordered_extent->disk_num_bytes, 1);
3325 * Actually free the qgroup rsv which was released when
3326 * the ordered extent was created.
3328 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3329 ordered_extent->qgroup_rsv,
3330 BTRFS_QGROUP_RSV_DATA);
3335 * This needs to be done to make sure anybody waiting knows we are done
3336 * updating everything for this ordered extent.
3338 btrfs_remove_ordered_extent(inode, ordered_extent);
3341 btrfs_put_ordered_extent(ordered_extent);
3342 /* once for the tree */
3343 btrfs_put_ordered_extent(ordered_extent);
3348 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3350 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3351 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
3352 btrfs_finish_ordered_zoned(ordered);
3353 return btrfs_finish_one_ordered(ordered);
3357 * Verify the checksum for a single sector without any extra action that depend
3358 * on the type of I/O.
3360 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3361 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3363 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3366 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3368 shash->tfm = fs_info->csum_shash;
3370 kaddr = kmap_local_page(page) + pgoff;
3371 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3372 kunmap_local(kaddr);
3374 if (memcmp(csum, csum_expected, fs_info->csum_size))
3380 * Verify the checksum of a single data sector.
3382 * @bbio: btrfs_io_bio which contains the csum
3383 * @dev: device the sector is on
3384 * @bio_offset: offset to the beginning of the bio (in bytes)
3385 * @bv: bio_vec to check
3387 * Check if the checksum on a data block is valid. When a checksum mismatch is
3388 * detected, report the error and fill the corrupted range with zero.
3390 * Return %true if the sector is ok or had no checksum to start with, else %false.
3392 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3393 u32 bio_offset, struct bio_vec *bv)
3395 struct btrfs_inode *inode = bbio->inode;
3396 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3397 u64 file_offset = bbio->file_offset + bio_offset;
3398 u64 end = file_offset + bv->bv_len - 1;
3400 u8 csum[BTRFS_CSUM_SIZE];
3402 ASSERT(bv->bv_len == fs_info->sectorsize);
3407 if (btrfs_is_data_reloc_root(inode->root) &&
3408 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3410 /* Skip the range without csum for data reloc inode */
3411 clear_extent_bits(&inode->io_tree, file_offset, end,
3416 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3418 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3424 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3427 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3433 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3435 * @inode: The inode we want to perform iput on
3437 * This function uses the generic vfs_inode::i_count to track whether we should
3438 * just decrement it (in case it's > 1) or if this is the last iput then link
3439 * the inode to the delayed iput machinery. Delayed iputs are processed at
3440 * transaction commit time/superblock commit/cleaner kthread.
3442 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3444 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3445 unsigned long flags;
3447 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3450 atomic_inc(&fs_info->nr_delayed_iputs);
3452 * Need to be irq safe here because we can be called from either an irq
3453 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3456 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3457 ASSERT(list_empty(&inode->delayed_iput));
3458 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3459 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3460 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3461 wake_up_process(fs_info->cleaner_kthread);
3464 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3465 struct btrfs_inode *inode)
3467 list_del_init(&inode->delayed_iput);
3468 spin_unlock_irq(&fs_info->delayed_iput_lock);
3469 iput(&inode->vfs_inode);
3470 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3471 wake_up(&fs_info->delayed_iputs_wait);
3472 spin_lock_irq(&fs_info->delayed_iput_lock);
3475 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3476 struct btrfs_inode *inode)
3478 if (!list_empty(&inode->delayed_iput)) {
3479 spin_lock_irq(&fs_info->delayed_iput_lock);
3480 if (!list_empty(&inode->delayed_iput))
3481 run_delayed_iput_locked(fs_info, inode);
3482 spin_unlock_irq(&fs_info->delayed_iput_lock);
3486 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3489 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3490 * calls btrfs_add_delayed_iput() and that needs to lock
3491 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3492 * prevent a deadlock.
3494 spin_lock_irq(&fs_info->delayed_iput_lock);
3495 while (!list_empty(&fs_info->delayed_iputs)) {
3496 struct btrfs_inode *inode;
3498 inode = list_first_entry(&fs_info->delayed_iputs,
3499 struct btrfs_inode, delayed_iput);
3500 run_delayed_iput_locked(fs_info, inode);
3501 if (need_resched()) {
3502 spin_unlock_irq(&fs_info->delayed_iput_lock);
3504 spin_lock_irq(&fs_info->delayed_iput_lock);
3507 spin_unlock_irq(&fs_info->delayed_iput_lock);
3511 * Wait for flushing all delayed iputs
3513 * @fs_info: the filesystem
3515 * This will wait on any delayed iputs that are currently running with KILLABLE
3516 * set. Once they are all done running we will return, unless we are killed in
3517 * which case we return EINTR. This helps in user operations like fallocate etc
3518 * that might get blocked on the iputs.
3520 * Return EINTR if we were killed, 0 if nothing's pending
3522 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3524 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3525 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3532 * This creates an orphan entry for the given inode in case something goes wrong
3533 * in the middle of an unlink.
3535 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3536 struct btrfs_inode *inode)
3540 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3541 if (ret && ret != -EEXIST) {
3542 btrfs_abort_transaction(trans, ret);
3550 * We have done the delete so we can go ahead and remove the orphan item for
3551 * this particular inode.
3553 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3554 struct btrfs_inode *inode)
3556 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3560 * this cleans up any orphans that may be left on the list from the last use
3563 int btrfs_orphan_cleanup(struct btrfs_root *root)
3565 struct btrfs_fs_info *fs_info = root->fs_info;
3566 struct btrfs_path *path;
3567 struct extent_buffer *leaf;
3568 struct btrfs_key key, found_key;
3569 struct btrfs_trans_handle *trans;
3570 struct inode *inode;
3571 u64 last_objectid = 0;
3572 int ret = 0, nr_unlink = 0;
3574 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3577 path = btrfs_alloc_path();
3582 path->reada = READA_BACK;
3584 key.objectid = BTRFS_ORPHAN_OBJECTID;
3585 key.type = BTRFS_ORPHAN_ITEM_KEY;
3586 key.offset = (u64)-1;
3589 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3594 * if ret == 0 means we found what we were searching for, which
3595 * is weird, but possible, so only screw with path if we didn't
3596 * find the key and see if we have stuff that matches
3600 if (path->slots[0] == 0)
3605 /* pull out the item */
3606 leaf = path->nodes[0];
3607 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3609 /* make sure the item matches what we want */
3610 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3612 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3615 /* release the path since we're done with it */
3616 btrfs_release_path(path);
3619 * this is where we are basically btrfs_lookup, without the
3620 * crossing root thing. we store the inode number in the
3621 * offset of the orphan item.
3624 if (found_key.offset == last_objectid) {
3626 "Error removing orphan entry, stopping orphan cleanup");
3631 last_objectid = found_key.offset;
3633 found_key.objectid = found_key.offset;
3634 found_key.type = BTRFS_INODE_ITEM_KEY;
3635 found_key.offset = 0;
3636 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3637 if (IS_ERR(inode)) {
3638 ret = PTR_ERR(inode);
3644 if (!inode && root == fs_info->tree_root) {
3645 struct btrfs_root *dead_root;
3646 int is_dead_root = 0;
3649 * This is an orphan in the tree root. Currently these
3650 * could come from 2 sources:
3651 * a) a root (snapshot/subvolume) deletion in progress
3652 * b) a free space cache inode
3653 * We need to distinguish those two, as the orphan item
3654 * for a root must not get deleted before the deletion
3655 * of the snapshot/subvolume's tree completes.
3657 * btrfs_find_orphan_roots() ran before us, which has
3658 * found all deleted roots and loaded them into
3659 * fs_info->fs_roots_radix. So here we can find if an
3660 * orphan item corresponds to a deleted root by looking
3661 * up the root from that radix tree.
3664 spin_lock(&fs_info->fs_roots_radix_lock);
3665 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3666 (unsigned long)found_key.objectid);
3667 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3669 spin_unlock(&fs_info->fs_roots_radix_lock);
3672 /* prevent this orphan from being found again */
3673 key.offset = found_key.objectid - 1;
3680 * If we have an inode with links, there are a couple of
3683 * 1. We were halfway through creating fsverity metadata for the
3684 * file. In that case, the orphan item represents incomplete
3685 * fsverity metadata which must be cleaned up with
3686 * btrfs_drop_verity_items and deleting the orphan item.
3688 * 2. Old kernels (before v3.12) used to create an
3689 * orphan item for truncate indicating that there were possibly
3690 * extent items past i_size that needed to be deleted. In v3.12,
3691 * truncate was changed to update i_size in sync with the extent
3692 * items, but the (useless) orphan item was still created. Since
3693 * v4.18, we don't create the orphan item for truncate at all.
3695 * So, this item could mean that we need to do a truncate, but
3696 * only if this filesystem was last used on a pre-v3.12 kernel
3697 * and was not cleanly unmounted. The odds of that are quite
3698 * slim, and it's a pain to do the truncate now, so just delete
3701 * It's also possible that this orphan item was supposed to be
3702 * deleted but wasn't. The inode number may have been reused,
3703 * but either way, we can delete the orphan item.
3705 if (!inode || inode->i_nlink) {
3707 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3713 trans = btrfs_start_transaction(root, 1);
3714 if (IS_ERR(trans)) {
3715 ret = PTR_ERR(trans);
3718 btrfs_debug(fs_info, "auto deleting %Lu",
3719 found_key.objectid);
3720 ret = btrfs_del_orphan_item(trans, root,
3721 found_key.objectid);
3722 btrfs_end_transaction(trans);
3730 /* this will do delete_inode and everything for us */
3733 /* release the path since we're done with it */
3734 btrfs_release_path(path);
3736 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3737 trans = btrfs_join_transaction(root);
3739 btrfs_end_transaction(trans);
3743 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3747 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3748 btrfs_free_path(path);
3753 * very simple check to peek ahead in the leaf looking for xattrs. If we
3754 * don't find any xattrs, we know there can't be any acls.
3756 * slot is the slot the inode is in, objectid is the objectid of the inode
3758 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3759 int slot, u64 objectid,
3760 int *first_xattr_slot)
3762 u32 nritems = btrfs_header_nritems(leaf);
3763 struct btrfs_key found_key;
3764 static u64 xattr_access = 0;
3765 static u64 xattr_default = 0;
3768 if (!xattr_access) {
3769 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3770 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3771 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3772 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3776 *first_xattr_slot = -1;
3777 while (slot < nritems) {
3778 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3780 /* we found a different objectid, there must not be acls */
3781 if (found_key.objectid != objectid)
3784 /* we found an xattr, assume we've got an acl */
3785 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3786 if (*first_xattr_slot == -1)
3787 *first_xattr_slot = slot;
3788 if (found_key.offset == xattr_access ||
3789 found_key.offset == xattr_default)
3794 * we found a key greater than an xattr key, there can't
3795 * be any acls later on
3797 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3804 * it goes inode, inode backrefs, xattrs, extents,
3805 * so if there are a ton of hard links to an inode there can
3806 * be a lot of backrefs. Don't waste time searching too hard,
3807 * this is just an optimization
3812 /* we hit the end of the leaf before we found an xattr or
3813 * something larger than an xattr. We have to assume the inode
3816 if (*first_xattr_slot == -1)
3817 *first_xattr_slot = slot;
3822 * read an inode from the btree into the in-memory inode
3824 static int btrfs_read_locked_inode(struct inode *inode,
3825 struct btrfs_path *in_path)
3827 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3828 struct btrfs_path *path = in_path;
3829 struct extent_buffer *leaf;
3830 struct btrfs_inode_item *inode_item;
3831 struct btrfs_root *root = BTRFS_I(inode)->root;
3832 struct btrfs_key location;
3837 bool filled = false;
3838 int first_xattr_slot;
3840 ret = btrfs_fill_inode(inode, &rdev);
3845 path = btrfs_alloc_path();
3850 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3852 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3854 if (path != in_path)
3855 btrfs_free_path(path);
3859 leaf = path->nodes[0];
3864 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3865 struct btrfs_inode_item);
3866 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3867 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3868 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3869 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3870 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3871 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3872 round_up(i_size_read(inode), fs_info->sectorsize));
3874 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3875 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3877 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3878 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3880 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3881 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3883 BTRFS_I(inode)->i_otime.tv_sec =
3884 btrfs_timespec_sec(leaf, &inode_item->otime);
3885 BTRFS_I(inode)->i_otime.tv_nsec =
3886 btrfs_timespec_nsec(leaf, &inode_item->otime);
3888 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3889 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3890 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3892 inode_set_iversion_queried(inode,
3893 btrfs_inode_sequence(leaf, inode_item));
3894 inode->i_generation = BTRFS_I(inode)->generation;
3896 rdev = btrfs_inode_rdev(leaf, inode_item);
3898 BTRFS_I(inode)->index_cnt = (u64)-1;
3899 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3900 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3904 * If we were modified in the current generation and evicted from memory
3905 * and then re-read we need to do a full sync since we don't have any
3906 * idea about which extents were modified before we were evicted from
3909 * This is required for both inode re-read from disk and delayed inode
3910 * in delayed_nodes_tree.
3912 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3913 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3914 &BTRFS_I(inode)->runtime_flags);
3917 * We don't persist the id of the transaction where an unlink operation
3918 * against the inode was last made. So here we assume the inode might
3919 * have been evicted, and therefore the exact value of last_unlink_trans
3920 * lost, and set it to last_trans to avoid metadata inconsistencies
3921 * between the inode and its parent if the inode is fsync'ed and the log
3922 * replayed. For example, in the scenario:
3925 * ln mydir/foo mydir/bar
3928 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3929 * xfs_io -c fsync mydir/foo
3931 * mount fs, triggers fsync log replay
3933 * We must make sure that when we fsync our inode foo we also log its
3934 * parent inode, otherwise after log replay the parent still has the
3935 * dentry with the "bar" name but our inode foo has a link count of 1
3936 * and doesn't have an inode ref with the name "bar" anymore.
3938 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3939 * but it guarantees correctness at the expense of occasional full
3940 * transaction commits on fsync if our inode is a directory, or if our
3941 * inode is not a directory, logging its parent unnecessarily.
3943 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3946 * Same logic as for last_unlink_trans. We don't persist the generation
3947 * of the last transaction where this inode was used for a reflink
3948 * operation, so after eviction and reloading the inode we must be
3949 * pessimistic and assume the last transaction that modified the inode.
3951 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3954 if (inode->i_nlink != 1 ||
3955 path->slots[0] >= btrfs_header_nritems(leaf))
3958 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3959 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3962 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3963 if (location.type == BTRFS_INODE_REF_KEY) {
3964 struct btrfs_inode_ref *ref;
3966 ref = (struct btrfs_inode_ref *)ptr;
3967 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3968 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3969 struct btrfs_inode_extref *extref;
3971 extref = (struct btrfs_inode_extref *)ptr;
3972 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3977 * try to precache a NULL acl entry for files that don't have
3978 * any xattrs or acls
3980 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3981 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3982 if (first_xattr_slot != -1) {
3983 path->slots[0] = first_xattr_slot;
3984 ret = btrfs_load_inode_props(inode, path);
3987 "error loading props for ino %llu (root %llu): %d",
3988 btrfs_ino(BTRFS_I(inode)),
3989 root->root_key.objectid, ret);
3991 if (path != in_path)
3992 btrfs_free_path(path);
3995 cache_no_acl(inode);
3997 switch (inode->i_mode & S_IFMT) {
3999 inode->i_mapping->a_ops = &btrfs_aops;
4000 inode->i_fop = &btrfs_file_operations;
4001 inode->i_op = &btrfs_file_inode_operations;
4004 inode->i_fop = &btrfs_dir_file_operations;
4005 inode->i_op = &btrfs_dir_inode_operations;
4008 inode->i_op = &btrfs_symlink_inode_operations;
4009 inode_nohighmem(inode);
4010 inode->i_mapping->a_ops = &btrfs_aops;
4013 inode->i_op = &btrfs_special_inode_operations;
4014 init_special_inode(inode, inode->i_mode, rdev);
4018 btrfs_sync_inode_flags_to_i_flags(inode);
4023 * given a leaf and an inode, copy the inode fields into the leaf
4025 static void fill_inode_item(struct btrfs_trans_handle *trans,
4026 struct extent_buffer *leaf,
4027 struct btrfs_inode_item *item,
4028 struct inode *inode)
4030 struct btrfs_map_token token;
4033 btrfs_init_map_token(&token, leaf);
4035 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4036 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4037 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
4038 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4039 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4041 btrfs_set_token_timespec_sec(&token, &item->atime,
4042 inode->i_atime.tv_sec);
4043 btrfs_set_token_timespec_nsec(&token, &item->atime,
4044 inode->i_atime.tv_nsec);
4046 btrfs_set_token_timespec_sec(&token, &item->mtime,
4047 inode->i_mtime.tv_sec);
4048 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4049 inode->i_mtime.tv_nsec);
4051 btrfs_set_token_timespec_sec(&token, &item->ctime,
4052 inode->i_ctime.tv_sec);
4053 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4054 inode->i_ctime.tv_nsec);
4056 btrfs_set_token_timespec_sec(&token, &item->otime,
4057 BTRFS_I(inode)->i_otime.tv_sec);
4058 btrfs_set_token_timespec_nsec(&token, &item->otime,
4059 BTRFS_I(inode)->i_otime.tv_nsec);
4061 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4062 btrfs_set_token_inode_generation(&token, item,
4063 BTRFS_I(inode)->generation);
4064 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4065 btrfs_set_token_inode_transid(&token, item, trans->transid);
4066 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4067 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4068 BTRFS_I(inode)->ro_flags);
4069 btrfs_set_token_inode_flags(&token, item, flags);
4070 btrfs_set_token_inode_block_group(&token, item, 0);
4074 * copy everything in the in-memory inode into the btree.
4076 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4077 struct btrfs_root *root,
4078 struct btrfs_inode *inode)
4080 struct btrfs_inode_item *inode_item;
4081 struct btrfs_path *path;
4082 struct extent_buffer *leaf;
4085 path = btrfs_alloc_path();
4089 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4096 leaf = path->nodes[0];
4097 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4098 struct btrfs_inode_item);
4100 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4101 btrfs_mark_buffer_dirty(leaf);
4102 btrfs_set_inode_last_trans(trans, inode);
4105 btrfs_free_path(path);
4110 * copy everything in the in-memory inode into the btree.
4112 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4113 struct btrfs_root *root,
4114 struct btrfs_inode *inode)
4116 struct btrfs_fs_info *fs_info = root->fs_info;
4120 * If the inode is a free space inode, we can deadlock during commit
4121 * if we put it into the delayed code.
4123 * The data relocation inode should also be directly updated
4126 if (!btrfs_is_free_space_inode(inode)
4127 && !btrfs_is_data_reloc_root(root)
4128 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4129 btrfs_update_root_times(trans, root);
4131 ret = btrfs_delayed_update_inode(trans, root, inode);
4133 btrfs_set_inode_last_trans(trans, inode);
4137 return btrfs_update_inode_item(trans, root, inode);
4140 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4141 struct btrfs_root *root, struct btrfs_inode *inode)
4145 ret = btrfs_update_inode(trans, root, inode);
4147 return btrfs_update_inode_item(trans, root, inode);
4152 * unlink helper that gets used here in inode.c and in the tree logging
4153 * recovery code. It remove a link in a directory with a given name, and
4154 * also drops the back refs in the inode to the directory
4156 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4157 struct btrfs_inode *dir,
4158 struct btrfs_inode *inode,
4159 const struct fscrypt_str *name,
4160 struct btrfs_rename_ctx *rename_ctx)
4162 struct btrfs_root *root = dir->root;
4163 struct btrfs_fs_info *fs_info = root->fs_info;
4164 struct btrfs_path *path;
4166 struct btrfs_dir_item *di;
4168 u64 ino = btrfs_ino(inode);
4169 u64 dir_ino = btrfs_ino(dir);
4171 path = btrfs_alloc_path();
4177 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4178 if (IS_ERR_OR_NULL(di)) {
4179 ret = di ? PTR_ERR(di) : -ENOENT;
4182 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4185 btrfs_release_path(path);
4188 * If we don't have dir index, we have to get it by looking up
4189 * the inode ref, since we get the inode ref, remove it directly,
4190 * it is unnecessary to do delayed deletion.
4192 * But if we have dir index, needn't search inode ref to get it.
4193 * Since the inode ref is close to the inode item, it is better
4194 * that we delay to delete it, and just do this deletion when
4195 * we update the inode item.
4197 if (inode->dir_index) {
4198 ret = btrfs_delayed_delete_inode_ref(inode);
4200 index = inode->dir_index;
4205 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4208 "failed to delete reference to %.*s, inode %llu parent %llu",
4209 name->len, name->name, ino, dir_ino);
4210 btrfs_abort_transaction(trans, ret);
4215 rename_ctx->index = index;
4217 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4219 btrfs_abort_transaction(trans, ret);
4224 * If we are in a rename context, we don't need to update anything in the
4225 * log. That will be done later during the rename by btrfs_log_new_name().
4226 * Besides that, doing it here would only cause extra unnecessary btree
4227 * operations on the log tree, increasing latency for applications.
4230 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4231 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4235 * If we have a pending delayed iput we could end up with the final iput
4236 * being run in btrfs-cleaner context. If we have enough of these built
4237 * up we can end up burning a lot of time in btrfs-cleaner without any
4238 * way to throttle the unlinks. Since we're currently holding a ref on
4239 * the inode we can run the delayed iput here without any issues as the
4240 * final iput won't be done until after we drop the ref we're currently
4243 btrfs_run_delayed_iput(fs_info, inode);
4245 btrfs_free_path(path);
4249 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4250 inode_inc_iversion(&inode->vfs_inode);
4251 inode_inc_iversion(&dir->vfs_inode);
4252 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4253 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4254 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4255 ret = btrfs_update_inode(trans, root, dir);
4260 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4261 struct btrfs_inode *dir, struct btrfs_inode *inode,
4262 const struct fscrypt_str *name)
4266 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4268 drop_nlink(&inode->vfs_inode);
4269 ret = btrfs_update_inode(trans, inode->root, inode);
4275 * helper to start transaction for unlink and rmdir.
4277 * unlink and rmdir are special in btrfs, they do not always free space, so
4278 * if we cannot make our reservations the normal way try and see if there is
4279 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4280 * allow the unlink to occur.
4282 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4284 struct btrfs_root *root = dir->root;
4286 return btrfs_start_transaction_fallback_global_rsv(root,
4287 BTRFS_UNLINK_METADATA_UNITS);
4290 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4292 struct btrfs_trans_handle *trans;
4293 struct inode *inode = d_inode(dentry);
4295 struct fscrypt_name fname;
4297 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4301 /* This needs to handle no-key deletions later on */
4303 trans = __unlink_start_trans(BTRFS_I(dir));
4304 if (IS_ERR(trans)) {
4305 ret = PTR_ERR(trans);
4309 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4312 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4317 if (inode->i_nlink == 0) {
4318 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4324 btrfs_end_transaction(trans);
4325 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4327 fscrypt_free_filename(&fname);
4331 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4332 struct btrfs_inode *dir, struct dentry *dentry)
4334 struct btrfs_root *root = dir->root;
4335 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4336 struct btrfs_path *path;
4337 struct extent_buffer *leaf;
4338 struct btrfs_dir_item *di;
4339 struct btrfs_key key;
4343 u64 dir_ino = btrfs_ino(dir);
4344 struct fscrypt_name fname;
4346 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4350 /* This needs to handle no-key deletions later on */
4352 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4353 objectid = inode->root->root_key.objectid;
4354 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4355 objectid = inode->location.objectid;
4358 fscrypt_free_filename(&fname);
4362 path = btrfs_alloc_path();
4368 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4369 &fname.disk_name, -1);
4370 if (IS_ERR_OR_NULL(di)) {
4371 ret = di ? PTR_ERR(di) : -ENOENT;
4375 leaf = path->nodes[0];
4376 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4377 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4378 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4380 btrfs_abort_transaction(trans, ret);
4383 btrfs_release_path(path);
4386 * This is a placeholder inode for a subvolume we didn't have a
4387 * reference to at the time of the snapshot creation. In the meantime
4388 * we could have renamed the real subvol link into our snapshot, so
4389 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4390 * Instead simply lookup the dir_index_item for this entry so we can
4391 * remove it. Otherwise we know we have a ref to the root and we can
4392 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4394 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4395 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4396 if (IS_ERR_OR_NULL(di)) {
4401 btrfs_abort_transaction(trans, ret);
4405 leaf = path->nodes[0];
4406 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4408 btrfs_release_path(path);
4410 ret = btrfs_del_root_ref(trans, objectid,
4411 root->root_key.objectid, dir_ino,
4412 &index, &fname.disk_name);
4414 btrfs_abort_transaction(trans, ret);
4419 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4421 btrfs_abort_transaction(trans, ret);
4425 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4426 inode_inc_iversion(&dir->vfs_inode);
4427 dir->vfs_inode.i_mtime = current_time(&dir->vfs_inode);
4428 dir->vfs_inode.i_ctime = dir->vfs_inode.i_mtime;
4429 ret = btrfs_update_inode_fallback(trans, root, dir);
4431 btrfs_abort_transaction(trans, ret);
4433 btrfs_free_path(path);
4434 fscrypt_free_filename(&fname);
4439 * Helper to check if the subvolume references other subvolumes or if it's
4442 static noinline int may_destroy_subvol(struct btrfs_root *root)
4444 struct btrfs_fs_info *fs_info = root->fs_info;
4445 struct btrfs_path *path;
4446 struct btrfs_dir_item *di;
4447 struct btrfs_key key;
4448 struct fscrypt_str name = FSTR_INIT("default", 7);
4452 path = btrfs_alloc_path();
4456 /* Make sure this root isn't set as the default subvol */
4457 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4458 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4460 if (di && !IS_ERR(di)) {
4461 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4462 if (key.objectid == root->root_key.objectid) {
4465 "deleting default subvolume %llu is not allowed",
4469 btrfs_release_path(path);
4472 key.objectid = root->root_key.objectid;
4473 key.type = BTRFS_ROOT_REF_KEY;
4474 key.offset = (u64)-1;
4476 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4482 if (path->slots[0] > 0) {
4484 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4485 if (key.objectid == root->root_key.objectid &&
4486 key.type == BTRFS_ROOT_REF_KEY)
4490 btrfs_free_path(path);
4494 /* Delete all dentries for inodes belonging to the root */
4495 static void btrfs_prune_dentries(struct btrfs_root *root)
4497 struct btrfs_fs_info *fs_info = root->fs_info;
4498 struct rb_node *node;
4499 struct rb_node *prev;
4500 struct btrfs_inode *entry;
4501 struct inode *inode;
4504 if (!BTRFS_FS_ERROR(fs_info))
4505 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4507 spin_lock(&root->inode_lock);
4509 node = root->inode_tree.rb_node;
4513 entry = rb_entry(node, struct btrfs_inode, rb_node);
4515 if (objectid < btrfs_ino(entry))
4516 node = node->rb_left;
4517 else if (objectid > btrfs_ino(entry))
4518 node = node->rb_right;
4524 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4525 if (objectid <= btrfs_ino(entry)) {
4529 prev = rb_next(prev);
4533 entry = rb_entry(node, struct btrfs_inode, rb_node);
4534 objectid = btrfs_ino(entry) + 1;
4535 inode = igrab(&entry->vfs_inode);
4537 spin_unlock(&root->inode_lock);
4538 if (atomic_read(&inode->i_count) > 1)
4539 d_prune_aliases(inode);
4541 * btrfs_drop_inode will have it removed from the inode
4542 * cache when its usage count hits zero.
4546 spin_lock(&root->inode_lock);
4550 if (cond_resched_lock(&root->inode_lock))
4553 node = rb_next(node);
4555 spin_unlock(&root->inode_lock);
4558 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4560 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4561 struct btrfs_root *root = dir->root;
4562 struct inode *inode = d_inode(dentry);
4563 struct btrfs_root *dest = BTRFS_I(inode)->root;
4564 struct btrfs_trans_handle *trans;
4565 struct btrfs_block_rsv block_rsv;
4570 * Don't allow to delete a subvolume with send in progress. This is
4571 * inside the inode lock so the error handling that has to drop the bit
4572 * again is not run concurrently.
4574 spin_lock(&dest->root_item_lock);
4575 if (dest->send_in_progress) {
4576 spin_unlock(&dest->root_item_lock);
4578 "attempt to delete subvolume %llu during send",
4579 dest->root_key.objectid);
4582 if (atomic_read(&dest->nr_swapfiles)) {
4583 spin_unlock(&dest->root_item_lock);
4585 "attempt to delete subvolume %llu with active swapfile",
4586 root->root_key.objectid);
4589 root_flags = btrfs_root_flags(&dest->root_item);
4590 btrfs_set_root_flags(&dest->root_item,
4591 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4592 spin_unlock(&dest->root_item_lock);
4594 down_write(&fs_info->subvol_sem);
4596 ret = may_destroy_subvol(dest);
4600 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4602 * One for dir inode,
4603 * two for dir entries,
4604 * two for root ref/backref.
4606 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4610 trans = btrfs_start_transaction(root, 0);
4611 if (IS_ERR(trans)) {
4612 ret = PTR_ERR(trans);
4615 trans->block_rsv = &block_rsv;
4616 trans->bytes_reserved = block_rsv.size;
4618 btrfs_record_snapshot_destroy(trans, dir);
4620 ret = btrfs_unlink_subvol(trans, dir, dentry);
4622 btrfs_abort_transaction(trans, ret);
4626 ret = btrfs_record_root_in_trans(trans, dest);
4628 btrfs_abort_transaction(trans, ret);
4632 memset(&dest->root_item.drop_progress, 0,
4633 sizeof(dest->root_item.drop_progress));
4634 btrfs_set_root_drop_level(&dest->root_item, 0);
4635 btrfs_set_root_refs(&dest->root_item, 0);
4637 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4638 ret = btrfs_insert_orphan_item(trans,
4640 dest->root_key.objectid);
4642 btrfs_abort_transaction(trans, ret);
4647 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4648 BTRFS_UUID_KEY_SUBVOL,
4649 dest->root_key.objectid);
4650 if (ret && ret != -ENOENT) {
4651 btrfs_abort_transaction(trans, ret);
4654 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4655 ret = btrfs_uuid_tree_remove(trans,
4656 dest->root_item.received_uuid,
4657 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4658 dest->root_key.objectid);
4659 if (ret && ret != -ENOENT) {
4660 btrfs_abort_transaction(trans, ret);
4665 free_anon_bdev(dest->anon_dev);
4668 trans->block_rsv = NULL;
4669 trans->bytes_reserved = 0;
4670 ret = btrfs_end_transaction(trans);
4671 inode->i_flags |= S_DEAD;
4673 btrfs_subvolume_release_metadata(root, &block_rsv);
4675 up_write(&fs_info->subvol_sem);
4677 spin_lock(&dest->root_item_lock);
4678 root_flags = btrfs_root_flags(&dest->root_item);
4679 btrfs_set_root_flags(&dest->root_item,
4680 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4681 spin_unlock(&dest->root_item_lock);
4683 d_invalidate(dentry);
4684 btrfs_prune_dentries(dest);
4685 ASSERT(dest->send_in_progress == 0);
4691 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4693 struct inode *inode = d_inode(dentry);
4694 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4696 struct btrfs_trans_handle *trans;
4697 u64 last_unlink_trans;
4698 struct fscrypt_name fname;
4700 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4702 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4703 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4705 "extent tree v2 doesn't support snapshot deletion yet");
4708 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4711 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4715 /* This needs to handle no-key deletions later on */
4717 trans = __unlink_start_trans(BTRFS_I(dir));
4718 if (IS_ERR(trans)) {
4719 err = PTR_ERR(trans);
4723 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4724 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4728 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4732 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4734 /* now the directory is empty */
4735 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4738 btrfs_i_size_write(BTRFS_I(inode), 0);
4740 * Propagate the last_unlink_trans value of the deleted dir to
4741 * its parent directory. This is to prevent an unrecoverable
4742 * log tree in the case we do something like this:
4744 * 2) create snapshot under dir foo
4745 * 3) delete the snapshot
4748 * 6) fsync foo or some file inside foo
4750 if (last_unlink_trans >= trans->transid)
4751 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4754 btrfs_end_transaction(trans);
4756 btrfs_btree_balance_dirty(fs_info);
4757 fscrypt_free_filename(&fname);
4763 * btrfs_truncate_block - read, zero a chunk and write a block
4764 * @inode - inode that we're zeroing
4765 * @from - the offset to start zeroing
4766 * @len - the length to zero, 0 to zero the entire range respective to the
4768 * @front - zero up to the offset instead of from the offset on
4770 * This will find the block for the "from" offset and cow the block and zero the
4771 * part we want to zero. This is used with truncate and hole punching.
4773 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4776 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4777 struct address_space *mapping = inode->vfs_inode.i_mapping;
4778 struct extent_io_tree *io_tree = &inode->io_tree;
4779 struct btrfs_ordered_extent *ordered;
4780 struct extent_state *cached_state = NULL;
4781 struct extent_changeset *data_reserved = NULL;
4782 bool only_release_metadata = false;
4783 u32 blocksize = fs_info->sectorsize;
4784 pgoff_t index = from >> PAGE_SHIFT;
4785 unsigned offset = from & (blocksize - 1);
4787 gfp_t mask = btrfs_alloc_write_mask(mapping);
4788 size_t write_bytes = blocksize;
4793 if (IS_ALIGNED(offset, blocksize) &&
4794 (!len || IS_ALIGNED(len, blocksize)))
4797 block_start = round_down(from, blocksize);
4798 block_end = block_start + blocksize - 1;
4800 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4803 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4804 /* For nocow case, no need to reserve data space */
4805 only_release_metadata = true;
4810 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4812 if (!only_release_metadata)
4813 btrfs_free_reserved_data_space(inode, data_reserved,
4814 block_start, blocksize);
4818 page = find_or_create_page(mapping, index, mask);
4820 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4822 btrfs_delalloc_release_extents(inode, blocksize);
4827 if (!PageUptodate(page)) {
4828 ret = btrfs_read_folio(NULL, page_folio(page));
4830 if (page->mapping != mapping) {
4835 if (!PageUptodate(page)) {
4842 * We unlock the page after the io is completed and then re-lock it
4843 * above. release_folio() could have come in between that and cleared
4844 * PagePrivate(), but left the page in the mapping. Set the page mapped
4845 * here to make sure it's properly set for the subpage stuff.
4847 ret = set_page_extent_mapped(page);
4851 wait_on_page_writeback(page);
4853 lock_extent(io_tree, block_start, block_end, &cached_state);
4855 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4857 unlock_extent(io_tree, block_start, block_end, &cached_state);
4860 btrfs_start_ordered_extent(ordered);
4861 btrfs_put_ordered_extent(ordered);
4865 clear_extent_bit(&inode->io_tree, block_start, block_end,
4866 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4869 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4872 unlock_extent(io_tree, block_start, block_end, &cached_state);
4876 if (offset != blocksize) {
4878 len = blocksize - offset;
4880 memzero_page(page, (block_start - page_offset(page)),
4883 memzero_page(page, (block_start - page_offset(page)) + offset,
4886 btrfs_page_clear_checked(fs_info, page, block_start,
4887 block_end + 1 - block_start);
4888 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4889 unlock_extent(io_tree, block_start, block_end, &cached_state);
4891 if (only_release_metadata)
4892 set_extent_bit(&inode->io_tree, block_start, block_end,
4893 EXTENT_NORESERVE, NULL);
4897 if (only_release_metadata)
4898 btrfs_delalloc_release_metadata(inode, blocksize, true);
4900 btrfs_delalloc_release_space(inode, data_reserved,
4901 block_start, blocksize, true);
4903 btrfs_delalloc_release_extents(inode, blocksize);
4907 if (only_release_metadata)
4908 btrfs_check_nocow_unlock(inode);
4909 extent_changeset_free(data_reserved);
4913 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4914 u64 offset, u64 len)
4916 struct btrfs_fs_info *fs_info = root->fs_info;
4917 struct btrfs_trans_handle *trans;
4918 struct btrfs_drop_extents_args drop_args = { 0 };
4922 * If NO_HOLES is enabled, we don't need to do anything.
4923 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4924 * or btrfs_update_inode() will be called, which guarantee that the next
4925 * fsync will know this inode was changed and needs to be logged.
4927 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4931 * 1 - for the one we're dropping
4932 * 1 - for the one we're adding
4933 * 1 - for updating the inode.
4935 trans = btrfs_start_transaction(root, 3);
4937 return PTR_ERR(trans);
4939 drop_args.start = offset;
4940 drop_args.end = offset + len;
4941 drop_args.drop_cache = true;
4943 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4945 btrfs_abort_transaction(trans, ret);
4946 btrfs_end_transaction(trans);
4950 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4952 btrfs_abort_transaction(trans, ret);
4954 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4955 btrfs_update_inode(trans, root, inode);
4957 btrfs_end_transaction(trans);
4962 * This function puts in dummy file extents for the area we're creating a hole
4963 * for. So if we are truncating this file to a larger size we need to insert
4964 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4965 * the range between oldsize and size
4967 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4969 struct btrfs_root *root = inode->root;
4970 struct btrfs_fs_info *fs_info = root->fs_info;
4971 struct extent_io_tree *io_tree = &inode->io_tree;
4972 struct extent_map *em = NULL;
4973 struct extent_state *cached_state = NULL;
4974 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4975 u64 block_end = ALIGN(size, fs_info->sectorsize);
4982 * If our size started in the middle of a block we need to zero out the
4983 * rest of the block before we expand the i_size, otherwise we could
4984 * expose stale data.
4986 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4990 if (size <= hole_start)
4993 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4995 cur_offset = hole_start;
4997 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4998 block_end - cur_offset);
5004 last_byte = min(extent_map_end(em), block_end);
5005 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5006 hole_size = last_byte - cur_offset;
5008 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5009 struct extent_map *hole_em;
5011 err = maybe_insert_hole(root, inode, cur_offset,
5016 err = btrfs_inode_set_file_extent_range(inode,
5017 cur_offset, hole_size);
5021 hole_em = alloc_extent_map();
5023 btrfs_drop_extent_map_range(inode, cur_offset,
5024 cur_offset + hole_size - 1,
5026 btrfs_set_inode_full_sync(inode);
5029 hole_em->start = cur_offset;
5030 hole_em->len = hole_size;
5031 hole_em->orig_start = cur_offset;
5033 hole_em->block_start = EXTENT_MAP_HOLE;
5034 hole_em->block_len = 0;
5035 hole_em->orig_block_len = 0;
5036 hole_em->ram_bytes = hole_size;
5037 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5038 hole_em->generation = fs_info->generation;
5040 err = btrfs_replace_extent_map_range(inode, hole_em, true);
5041 free_extent_map(hole_em);
5043 err = btrfs_inode_set_file_extent_range(inode,
5044 cur_offset, hole_size);
5049 free_extent_map(em);
5051 cur_offset = last_byte;
5052 if (cur_offset >= block_end)
5055 free_extent_map(em);
5056 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
5060 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5062 struct btrfs_root *root = BTRFS_I(inode)->root;
5063 struct btrfs_trans_handle *trans;
5064 loff_t oldsize = i_size_read(inode);
5065 loff_t newsize = attr->ia_size;
5066 int mask = attr->ia_valid;
5070 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5071 * special case where we need to update the times despite not having
5072 * these flags set. For all other operations the VFS set these flags
5073 * explicitly if it wants a timestamp update.
5075 if (newsize != oldsize) {
5076 inode_inc_iversion(inode);
5077 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5078 inode->i_mtime = current_time(inode);
5079 inode->i_ctime = inode->i_mtime;
5083 if (newsize > oldsize) {
5085 * Don't do an expanding truncate while snapshotting is ongoing.
5086 * This is to ensure the snapshot captures a fully consistent
5087 * state of this file - if the snapshot captures this expanding
5088 * truncation, it must capture all writes that happened before
5091 btrfs_drew_write_lock(&root->snapshot_lock);
5092 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5094 btrfs_drew_write_unlock(&root->snapshot_lock);
5098 trans = btrfs_start_transaction(root, 1);
5099 if (IS_ERR(trans)) {
5100 btrfs_drew_write_unlock(&root->snapshot_lock);
5101 return PTR_ERR(trans);
5104 i_size_write(inode, newsize);
5105 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5106 pagecache_isize_extended(inode, oldsize, newsize);
5107 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5108 btrfs_drew_write_unlock(&root->snapshot_lock);
5109 btrfs_end_transaction(trans);
5111 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5113 if (btrfs_is_zoned(fs_info)) {
5114 ret = btrfs_wait_ordered_range(inode,
5115 ALIGN(newsize, fs_info->sectorsize),
5122 * We're truncating a file that used to have good data down to
5123 * zero. Make sure any new writes to the file get on disk
5127 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5128 &BTRFS_I(inode)->runtime_flags);
5130 truncate_setsize(inode, newsize);
5132 inode_dio_wait(inode);
5134 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5135 if (ret && inode->i_nlink) {
5139 * Truncate failed, so fix up the in-memory size. We
5140 * adjusted disk_i_size down as we removed extents, so
5141 * wait for disk_i_size to be stable and then update the
5142 * in-memory size to match.
5144 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5147 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5154 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5157 struct inode *inode = d_inode(dentry);
5158 struct btrfs_root *root = BTRFS_I(inode)->root;
5161 if (btrfs_root_readonly(root))
5164 err = setattr_prepare(idmap, dentry, attr);
5168 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5169 err = btrfs_setsize(inode, attr);
5174 if (attr->ia_valid) {
5175 setattr_copy(idmap, inode, attr);
5176 inode_inc_iversion(inode);
5177 err = btrfs_dirty_inode(BTRFS_I(inode));
5179 if (!err && attr->ia_valid & ATTR_MODE)
5180 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5187 * While truncating the inode pages during eviction, we get the VFS
5188 * calling btrfs_invalidate_folio() against each folio of the inode. This
5189 * is slow because the calls to btrfs_invalidate_folio() result in a
5190 * huge amount of calls to lock_extent() and clear_extent_bit(),
5191 * which keep merging and splitting extent_state structures over and over,
5192 * wasting lots of time.
5194 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5195 * skip all those expensive operations on a per folio basis and do only
5196 * the ordered io finishing, while we release here the extent_map and
5197 * extent_state structures, without the excessive merging and splitting.
5199 static void evict_inode_truncate_pages(struct inode *inode)
5201 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5202 struct rb_node *node;
5204 ASSERT(inode->i_state & I_FREEING);
5205 truncate_inode_pages_final(&inode->i_data);
5207 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5210 * Keep looping until we have no more ranges in the io tree.
5211 * We can have ongoing bios started by readahead that have
5212 * their endio callback (extent_io.c:end_bio_extent_readpage)
5213 * still in progress (unlocked the pages in the bio but did not yet
5214 * unlocked the ranges in the io tree). Therefore this means some
5215 * ranges can still be locked and eviction started because before
5216 * submitting those bios, which are executed by a separate task (work
5217 * queue kthread), inode references (inode->i_count) were not taken
5218 * (which would be dropped in the end io callback of each bio).
5219 * Therefore here we effectively end up waiting for those bios and
5220 * anyone else holding locked ranges without having bumped the inode's
5221 * reference count - if we don't do it, when they access the inode's
5222 * io_tree to unlock a range it may be too late, leading to an
5223 * use-after-free issue.
5225 spin_lock(&io_tree->lock);
5226 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5227 struct extent_state *state;
5228 struct extent_state *cached_state = NULL;
5231 unsigned state_flags;
5233 node = rb_first(&io_tree->state);
5234 state = rb_entry(node, struct extent_state, rb_node);
5235 start = state->start;
5237 state_flags = state->state;
5238 spin_unlock(&io_tree->lock);
5240 lock_extent(io_tree, start, end, &cached_state);
5243 * If still has DELALLOC flag, the extent didn't reach disk,
5244 * and its reserved space won't be freed by delayed_ref.
5245 * So we need to free its reserved space here.
5246 * (Refer to comment in btrfs_invalidate_folio, case 2)
5248 * Note, end is the bytenr of last byte, so we need + 1 here.
5250 if (state_flags & EXTENT_DELALLOC)
5251 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5254 clear_extent_bit(io_tree, start, end,
5255 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5259 spin_lock(&io_tree->lock);
5261 spin_unlock(&io_tree->lock);
5264 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5265 struct btrfs_block_rsv *rsv)
5267 struct btrfs_fs_info *fs_info = root->fs_info;
5268 struct btrfs_trans_handle *trans;
5269 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5273 * Eviction should be taking place at some place safe because of our
5274 * delayed iputs. However the normal flushing code will run delayed
5275 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5277 * We reserve the delayed_refs_extra here again because we can't use
5278 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5279 * above. We reserve our extra bit here because we generate a ton of
5280 * delayed refs activity by truncating.
5282 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5283 * if we fail to make this reservation we can re-try without the
5284 * delayed_refs_extra so we can make some forward progress.
5286 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5287 BTRFS_RESERVE_FLUSH_EVICT);
5289 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5290 BTRFS_RESERVE_FLUSH_EVICT);
5293 "could not allocate space for delete; will truncate on mount");
5294 return ERR_PTR(-ENOSPC);
5296 delayed_refs_extra = 0;
5299 trans = btrfs_join_transaction(root);
5303 if (delayed_refs_extra) {
5304 trans->block_rsv = &fs_info->trans_block_rsv;
5305 trans->bytes_reserved = delayed_refs_extra;
5306 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5307 delayed_refs_extra, true);
5312 void btrfs_evict_inode(struct inode *inode)
5314 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5315 struct btrfs_trans_handle *trans;
5316 struct btrfs_root *root = BTRFS_I(inode)->root;
5317 struct btrfs_block_rsv *rsv = NULL;
5320 trace_btrfs_inode_evict(inode);
5323 fsverity_cleanup_inode(inode);
5328 evict_inode_truncate_pages(inode);
5330 if (inode->i_nlink &&
5331 ((btrfs_root_refs(&root->root_item) != 0 &&
5332 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5333 btrfs_is_free_space_inode(BTRFS_I(inode))))
5336 if (is_bad_inode(inode))
5339 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5342 if (inode->i_nlink > 0) {
5343 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5344 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5349 * This makes sure the inode item in tree is uptodate and the space for
5350 * the inode update is released.
5352 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5357 * This drops any pending insert or delete operations we have for this
5358 * inode. We could have a delayed dir index deletion queued up, but
5359 * we're removing the inode completely so that'll be taken care of in
5362 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5364 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5367 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5368 rsv->failfast = true;
5370 btrfs_i_size_write(BTRFS_I(inode), 0);
5373 struct btrfs_truncate_control control = {
5374 .inode = BTRFS_I(inode),
5375 .ino = btrfs_ino(BTRFS_I(inode)),
5380 trans = evict_refill_and_join(root, rsv);
5384 trans->block_rsv = rsv;
5386 ret = btrfs_truncate_inode_items(trans, root, &control);
5387 trans->block_rsv = &fs_info->trans_block_rsv;
5388 btrfs_end_transaction(trans);
5390 * We have not added new delayed items for our inode after we
5391 * have flushed its delayed items, so no need to throttle on
5392 * delayed items. However we have modified extent buffers.
5394 btrfs_btree_balance_dirty_nodelay(fs_info);
5395 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5402 * Errors here aren't a big deal, it just means we leave orphan items in
5403 * the tree. They will be cleaned up on the next mount. If the inode
5404 * number gets reused, cleanup deletes the orphan item without doing
5405 * anything, and unlink reuses the existing orphan item.
5407 * If it turns out that we are dropping too many of these, we might want
5408 * to add a mechanism for retrying these after a commit.
5410 trans = evict_refill_and_join(root, rsv);
5411 if (!IS_ERR(trans)) {
5412 trans->block_rsv = rsv;
5413 btrfs_orphan_del(trans, BTRFS_I(inode));
5414 trans->block_rsv = &fs_info->trans_block_rsv;
5415 btrfs_end_transaction(trans);
5419 btrfs_free_block_rsv(fs_info, rsv);
5421 * If we didn't successfully delete, the orphan item will still be in
5422 * the tree and we'll retry on the next mount. Again, we might also want
5423 * to retry these periodically in the future.
5425 btrfs_remove_delayed_node(BTRFS_I(inode));
5426 fsverity_cleanup_inode(inode);
5431 * Return the key found in the dir entry in the location pointer, fill @type
5432 * with BTRFS_FT_*, and return 0.
5434 * If no dir entries were found, returns -ENOENT.
5435 * If found a corrupted location in dir entry, returns -EUCLEAN.
5437 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5438 struct btrfs_key *location, u8 *type)
5440 struct btrfs_dir_item *di;
5441 struct btrfs_path *path;
5442 struct btrfs_root *root = dir->root;
5444 struct fscrypt_name fname;
5446 path = btrfs_alloc_path();
5450 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5454 * fscrypt_setup_filename() should never return a positive value, but
5455 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5459 /* This needs to handle no-key deletions later on */
5461 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5462 &fname.disk_name, 0);
5463 if (IS_ERR_OR_NULL(di)) {
5464 ret = di ? PTR_ERR(di) : -ENOENT;
5468 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5469 if (location->type != BTRFS_INODE_ITEM_KEY &&
5470 location->type != BTRFS_ROOT_ITEM_KEY) {
5472 btrfs_warn(root->fs_info,
5473 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5474 __func__, fname.disk_name.name, btrfs_ino(dir),
5475 location->objectid, location->type, location->offset);
5478 *type = btrfs_dir_ftype(path->nodes[0], di);
5480 fscrypt_free_filename(&fname);
5481 btrfs_free_path(path);
5486 * when we hit a tree root in a directory, the btrfs part of the inode
5487 * needs to be changed to reflect the root directory of the tree root. This
5488 * is kind of like crossing a mount point.
5490 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5491 struct btrfs_inode *dir,
5492 struct dentry *dentry,
5493 struct btrfs_key *location,
5494 struct btrfs_root **sub_root)
5496 struct btrfs_path *path;
5497 struct btrfs_root *new_root;
5498 struct btrfs_root_ref *ref;
5499 struct extent_buffer *leaf;
5500 struct btrfs_key key;
5503 struct fscrypt_name fname;
5505 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5509 path = btrfs_alloc_path();
5516 key.objectid = dir->root->root_key.objectid;
5517 key.type = BTRFS_ROOT_REF_KEY;
5518 key.offset = location->objectid;
5520 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5527 leaf = path->nodes[0];
5528 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5529 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5530 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5533 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5534 (unsigned long)(ref + 1), fname.disk_name.len);
5538 btrfs_release_path(path);
5540 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5541 if (IS_ERR(new_root)) {
5542 err = PTR_ERR(new_root);
5546 *sub_root = new_root;
5547 location->objectid = btrfs_root_dirid(&new_root->root_item);
5548 location->type = BTRFS_INODE_ITEM_KEY;
5549 location->offset = 0;
5552 btrfs_free_path(path);
5553 fscrypt_free_filename(&fname);
5557 static void inode_tree_add(struct btrfs_inode *inode)
5559 struct btrfs_root *root = inode->root;
5560 struct btrfs_inode *entry;
5562 struct rb_node *parent;
5563 struct rb_node *new = &inode->rb_node;
5564 u64 ino = btrfs_ino(inode);
5566 if (inode_unhashed(&inode->vfs_inode))
5569 spin_lock(&root->inode_lock);
5570 p = &root->inode_tree.rb_node;
5573 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5575 if (ino < btrfs_ino(entry))
5576 p = &parent->rb_left;
5577 else if (ino > btrfs_ino(entry))
5578 p = &parent->rb_right;
5580 WARN_ON(!(entry->vfs_inode.i_state &
5581 (I_WILL_FREE | I_FREEING)));
5582 rb_replace_node(parent, new, &root->inode_tree);
5583 RB_CLEAR_NODE(parent);
5584 spin_unlock(&root->inode_lock);
5588 rb_link_node(new, parent, p);
5589 rb_insert_color(new, &root->inode_tree);
5590 spin_unlock(&root->inode_lock);
5593 static void inode_tree_del(struct btrfs_inode *inode)
5595 struct btrfs_root *root = inode->root;
5598 spin_lock(&root->inode_lock);
5599 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5600 rb_erase(&inode->rb_node, &root->inode_tree);
5601 RB_CLEAR_NODE(&inode->rb_node);
5602 empty = RB_EMPTY_ROOT(&root->inode_tree);
5604 spin_unlock(&root->inode_lock);
5606 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5607 spin_lock(&root->inode_lock);
5608 empty = RB_EMPTY_ROOT(&root->inode_tree);
5609 spin_unlock(&root->inode_lock);
5611 btrfs_add_dead_root(root);
5616 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5618 struct btrfs_iget_args *args = p;
5620 inode->i_ino = args->ino;
5621 BTRFS_I(inode)->location.objectid = args->ino;
5622 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5623 BTRFS_I(inode)->location.offset = 0;
5624 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5625 BUG_ON(args->root && !BTRFS_I(inode)->root);
5627 if (args->root && args->root == args->root->fs_info->tree_root &&
5628 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5629 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5630 &BTRFS_I(inode)->runtime_flags);
5634 static int btrfs_find_actor(struct inode *inode, void *opaque)
5636 struct btrfs_iget_args *args = opaque;
5638 return args->ino == BTRFS_I(inode)->location.objectid &&
5639 args->root == BTRFS_I(inode)->root;
5642 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5643 struct btrfs_root *root)
5645 struct inode *inode;
5646 struct btrfs_iget_args args;
5647 unsigned long hashval = btrfs_inode_hash(ino, root);
5652 inode = iget5_locked(s, hashval, btrfs_find_actor,
5653 btrfs_init_locked_inode,
5659 * Get an inode object given its inode number and corresponding root.
5660 * Path can be preallocated to prevent recursing back to iget through
5661 * allocator. NULL is also valid but may require an additional allocation
5664 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5665 struct btrfs_root *root, struct btrfs_path *path)
5667 struct inode *inode;
5669 inode = btrfs_iget_locked(s, ino, root);
5671 return ERR_PTR(-ENOMEM);
5673 if (inode->i_state & I_NEW) {
5676 ret = btrfs_read_locked_inode(inode, path);
5678 inode_tree_add(BTRFS_I(inode));
5679 unlock_new_inode(inode);
5683 * ret > 0 can come from btrfs_search_slot called by
5684 * btrfs_read_locked_inode, this means the inode item
5689 inode = ERR_PTR(ret);
5696 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5698 return btrfs_iget_path(s, ino, root, NULL);
5701 static struct inode *new_simple_dir(struct super_block *s,
5702 struct btrfs_key *key,
5703 struct btrfs_root *root)
5705 struct inode *inode = new_inode(s);
5708 return ERR_PTR(-ENOMEM);
5710 BTRFS_I(inode)->root = btrfs_grab_root(root);
5711 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5712 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5714 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5716 * We only need lookup, the rest is read-only and there's no inode
5717 * associated with the dentry
5719 inode->i_op = &simple_dir_inode_operations;
5720 inode->i_opflags &= ~IOP_XATTR;
5721 inode->i_fop = &simple_dir_operations;
5722 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5723 inode->i_mtime = current_time(inode);
5724 inode->i_atime = inode->i_mtime;
5725 inode->i_ctime = inode->i_mtime;
5726 BTRFS_I(inode)->i_otime = inode->i_mtime;
5731 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5732 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5733 static_assert(BTRFS_FT_DIR == FT_DIR);
5734 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5735 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5736 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5737 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5738 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5740 static inline u8 btrfs_inode_type(struct inode *inode)
5742 return fs_umode_to_ftype(inode->i_mode);
5745 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5747 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5748 struct inode *inode;
5749 struct btrfs_root *root = BTRFS_I(dir)->root;
5750 struct btrfs_root *sub_root = root;
5751 struct btrfs_key location;
5755 if (dentry->d_name.len > BTRFS_NAME_LEN)
5756 return ERR_PTR(-ENAMETOOLONG);
5758 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5760 return ERR_PTR(ret);
5762 if (location.type == BTRFS_INODE_ITEM_KEY) {
5763 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5767 /* Do extra check against inode mode with di_type */
5768 if (btrfs_inode_type(inode) != di_type) {
5770 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5771 inode->i_mode, btrfs_inode_type(inode),
5774 return ERR_PTR(-EUCLEAN);
5779 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5780 &location, &sub_root);
5783 inode = ERR_PTR(ret);
5785 inode = new_simple_dir(dir->i_sb, &location, root);
5787 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5788 btrfs_put_root(sub_root);
5793 down_read(&fs_info->cleanup_work_sem);
5794 if (!sb_rdonly(inode->i_sb))
5795 ret = btrfs_orphan_cleanup(sub_root);
5796 up_read(&fs_info->cleanup_work_sem);
5799 inode = ERR_PTR(ret);
5806 static int btrfs_dentry_delete(const struct dentry *dentry)
5808 struct btrfs_root *root;
5809 struct inode *inode = d_inode(dentry);
5811 if (!inode && !IS_ROOT(dentry))
5812 inode = d_inode(dentry->d_parent);
5815 root = BTRFS_I(inode)->root;
5816 if (btrfs_root_refs(&root->root_item) == 0)
5819 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5825 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5828 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5830 if (inode == ERR_PTR(-ENOENT))
5832 return d_splice_alias(inode, dentry);
5836 * Find the highest existing sequence number in a directory and then set the
5837 * in-memory index_cnt variable to the first free sequence number.
5839 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5841 struct btrfs_root *root = inode->root;
5842 struct btrfs_key key, found_key;
5843 struct btrfs_path *path;
5844 struct extent_buffer *leaf;
5847 key.objectid = btrfs_ino(inode);
5848 key.type = BTRFS_DIR_INDEX_KEY;
5849 key.offset = (u64)-1;
5851 path = btrfs_alloc_path();
5855 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5858 /* FIXME: we should be able to handle this */
5863 if (path->slots[0] == 0) {
5864 inode->index_cnt = BTRFS_DIR_START_INDEX;
5870 leaf = path->nodes[0];
5871 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5873 if (found_key.objectid != btrfs_ino(inode) ||
5874 found_key.type != BTRFS_DIR_INDEX_KEY) {
5875 inode->index_cnt = BTRFS_DIR_START_INDEX;
5879 inode->index_cnt = found_key.offset + 1;
5881 btrfs_free_path(path);
5885 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5887 if (dir->index_cnt == (u64)-1) {
5890 ret = btrfs_inode_delayed_dir_index_count(dir);
5892 ret = btrfs_set_inode_index_count(dir);
5898 *index = dir->index_cnt;
5904 * All this infrastructure exists because dir_emit can fault, and we are holding
5905 * the tree lock when doing readdir. For now just allocate a buffer and copy
5906 * our information into that, and then dir_emit from the buffer. This is
5907 * similar to what NFS does, only we don't keep the buffer around in pagecache
5908 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5909 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5912 static int btrfs_opendir(struct inode *inode, struct file *file)
5914 struct btrfs_file_private *private;
5918 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5922 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5925 private->last_index = last_index;
5926 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5927 if (!private->filldir_buf) {
5931 file->private_data = private;
5942 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5945 struct dir_entry *entry = addr;
5946 char *name = (char *)(entry + 1);
5948 ctx->pos = get_unaligned(&entry->offset);
5949 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5950 get_unaligned(&entry->ino),
5951 get_unaligned(&entry->type)))
5953 addr += sizeof(struct dir_entry) +
5954 get_unaligned(&entry->name_len);
5960 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5962 struct inode *inode = file_inode(file);
5963 struct btrfs_root *root = BTRFS_I(inode)->root;
5964 struct btrfs_file_private *private = file->private_data;
5965 struct btrfs_dir_item *di;
5966 struct btrfs_key key;
5967 struct btrfs_key found_key;
5968 struct btrfs_path *path;
5970 struct list_head ins_list;
5971 struct list_head del_list;
5978 struct btrfs_key location;
5980 if (!dir_emit_dots(file, ctx))
5983 path = btrfs_alloc_path();
5987 addr = private->filldir_buf;
5988 path->reada = READA_FORWARD;
5990 INIT_LIST_HEAD(&ins_list);
5991 INIT_LIST_HEAD(&del_list);
5992 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5993 &ins_list, &del_list);
5996 key.type = BTRFS_DIR_INDEX_KEY;
5997 key.offset = ctx->pos;
5998 key.objectid = btrfs_ino(BTRFS_I(inode));
6000 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
6001 struct dir_entry *entry;
6002 struct extent_buffer *leaf = path->nodes[0];
6005 if (found_key.objectid != key.objectid)
6007 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6009 if (found_key.offset < ctx->pos)
6011 if (found_key.offset > private->last_index)
6013 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6015 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
6016 name_len = btrfs_dir_name_len(leaf, di);
6017 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6019 btrfs_release_path(path);
6020 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6023 addr = private->filldir_buf;
6029 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
6031 name_ptr = (char *)(entry + 1);
6032 read_extent_buffer(leaf, name_ptr,
6033 (unsigned long)(di + 1), name_len);
6034 put_unaligned(name_len, &entry->name_len);
6035 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
6036 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6037 put_unaligned(location.objectid, &entry->ino);
6038 put_unaligned(found_key.offset, &entry->offset);
6040 addr += sizeof(struct dir_entry) + name_len;
6041 total_len += sizeof(struct dir_entry) + name_len;
6043 /* Catch error encountered during iteration */
6047 btrfs_release_path(path);
6049 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6053 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6058 * Stop new entries from being returned after we return the last
6061 * New directory entries are assigned a strictly increasing
6062 * offset. This means that new entries created during readdir
6063 * are *guaranteed* to be seen in the future by that readdir.
6064 * This has broken buggy programs which operate on names as
6065 * they're returned by readdir. Until we re-use freed offsets
6066 * we have this hack to stop new entries from being returned
6067 * under the assumption that they'll never reach this huge
6070 * This is being careful not to overflow 32bit loff_t unless the
6071 * last entry requires it because doing so has broken 32bit apps
6074 if (ctx->pos >= INT_MAX)
6075 ctx->pos = LLONG_MAX;
6082 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6083 btrfs_free_path(path);
6088 * This is somewhat expensive, updating the tree every time the
6089 * inode changes. But, it is most likely to find the inode in cache.
6090 * FIXME, needs more benchmarking...there are no reasons other than performance
6091 * to keep or drop this code.
6093 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6095 struct btrfs_root *root = inode->root;
6096 struct btrfs_fs_info *fs_info = root->fs_info;
6097 struct btrfs_trans_handle *trans;
6100 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6103 trans = btrfs_join_transaction(root);
6105 return PTR_ERR(trans);
6107 ret = btrfs_update_inode(trans, root, inode);
6108 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6109 /* whoops, lets try again with the full transaction */
6110 btrfs_end_transaction(trans);
6111 trans = btrfs_start_transaction(root, 1);
6113 return PTR_ERR(trans);
6115 ret = btrfs_update_inode(trans, root, inode);
6117 btrfs_end_transaction(trans);
6118 if (inode->delayed_node)
6119 btrfs_balance_delayed_items(fs_info);
6125 * This is a copy of file_update_time. We need this so we can return error on
6126 * ENOSPC for updating the inode in the case of file write and mmap writes.
6128 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6131 struct btrfs_root *root = BTRFS_I(inode)->root;
6132 bool dirty = flags & ~S_VERSION;
6134 if (btrfs_root_readonly(root))
6137 if (flags & S_VERSION)
6138 dirty |= inode_maybe_inc_iversion(inode, dirty);
6139 if (flags & S_CTIME)
6140 inode->i_ctime = *now;
6141 if (flags & S_MTIME)
6142 inode->i_mtime = *now;
6143 if (flags & S_ATIME)
6144 inode->i_atime = *now;
6145 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6149 * helper to find a free sequence number in a given directory. This current
6150 * code is very simple, later versions will do smarter things in the btree
6152 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6156 if (dir->index_cnt == (u64)-1) {
6157 ret = btrfs_inode_delayed_dir_index_count(dir);
6159 ret = btrfs_set_inode_index_count(dir);
6165 *index = dir->index_cnt;
6171 static int btrfs_insert_inode_locked(struct inode *inode)
6173 struct btrfs_iget_args args;
6175 args.ino = BTRFS_I(inode)->location.objectid;
6176 args.root = BTRFS_I(inode)->root;
6178 return insert_inode_locked4(inode,
6179 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6180 btrfs_find_actor, &args);
6183 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6184 unsigned int *trans_num_items)
6186 struct inode *dir = args->dir;
6187 struct inode *inode = args->inode;
6190 if (!args->orphan) {
6191 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6197 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6199 fscrypt_free_filename(&args->fname);
6203 /* 1 to add inode item */
6204 *trans_num_items = 1;
6205 /* 1 to add compression property */
6206 if (BTRFS_I(dir)->prop_compress)
6207 (*trans_num_items)++;
6208 /* 1 to add default ACL xattr */
6209 if (args->default_acl)
6210 (*trans_num_items)++;
6211 /* 1 to add access ACL xattr */
6213 (*trans_num_items)++;
6214 #ifdef CONFIG_SECURITY
6215 /* 1 to add LSM xattr */
6216 if (dir->i_security)
6217 (*trans_num_items)++;
6220 /* 1 to add orphan item */
6221 (*trans_num_items)++;
6225 * 1 to add dir index
6226 * 1 to update parent inode item
6228 * No need for 1 unit for the inode ref item because it is
6229 * inserted in a batch together with the inode item at
6230 * btrfs_create_new_inode().
6232 *trans_num_items += 3;
6237 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6239 posix_acl_release(args->acl);
6240 posix_acl_release(args->default_acl);
6241 fscrypt_free_filename(&args->fname);
6245 * Inherit flags from the parent inode.
6247 * Currently only the compression flags and the cow flags are inherited.
6249 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6255 if (flags & BTRFS_INODE_NOCOMPRESS) {
6256 inode->flags &= ~BTRFS_INODE_COMPRESS;
6257 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6258 } else if (flags & BTRFS_INODE_COMPRESS) {
6259 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6260 inode->flags |= BTRFS_INODE_COMPRESS;
6263 if (flags & BTRFS_INODE_NODATACOW) {
6264 inode->flags |= BTRFS_INODE_NODATACOW;
6265 if (S_ISREG(inode->vfs_inode.i_mode))
6266 inode->flags |= BTRFS_INODE_NODATASUM;
6269 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6272 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6273 struct btrfs_new_inode_args *args)
6275 struct inode *dir = args->dir;
6276 struct inode *inode = args->inode;
6277 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6278 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6279 struct btrfs_root *root;
6280 struct btrfs_inode_item *inode_item;
6281 struct btrfs_key *location;
6282 struct btrfs_path *path;
6284 struct btrfs_inode_ref *ref;
6285 struct btrfs_key key[2];
6287 struct btrfs_item_batch batch;
6291 path = btrfs_alloc_path();
6296 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6297 root = BTRFS_I(inode)->root;
6299 ret = btrfs_get_free_objectid(root, &objectid);
6302 inode->i_ino = objectid;
6306 * O_TMPFILE, set link count to 0, so that after this point, we
6307 * fill in an inode item with the correct link count.
6309 set_nlink(inode, 0);
6311 trace_btrfs_inode_request(dir);
6313 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6317 /* index_cnt is ignored for everything but a dir. */
6318 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6319 BTRFS_I(inode)->generation = trans->transid;
6320 inode->i_generation = BTRFS_I(inode)->generation;
6323 * Subvolumes don't inherit flags from their parent directory.
6324 * Originally this was probably by accident, but we probably can't
6325 * change it now without compatibility issues.
6328 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6330 if (S_ISREG(inode->i_mode)) {
6331 if (btrfs_test_opt(fs_info, NODATASUM))
6332 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6333 if (btrfs_test_opt(fs_info, NODATACOW))
6334 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6335 BTRFS_INODE_NODATASUM;
6338 location = &BTRFS_I(inode)->location;
6339 location->objectid = objectid;
6340 location->offset = 0;
6341 location->type = BTRFS_INODE_ITEM_KEY;
6343 ret = btrfs_insert_inode_locked(inode);
6346 BTRFS_I(dir)->index_cnt--;
6351 * We could have gotten an inode number from somebody who was fsynced
6352 * and then removed in this same transaction, so let's just set full
6353 * sync since it will be a full sync anyway and this will blow away the
6354 * old info in the log.
6356 btrfs_set_inode_full_sync(BTRFS_I(inode));
6358 key[0].objectid = objectid;
6359 key[0].type = BTRFS_INODE_ITEM_KEY;
6362 sizes[0] = sizeof(struct btrfs_inode_item);
6364 if (!args->orphan) {
6366 * Start new inodes with an inode_ref. This is slightly more
6367 * efficient for small numbers of hard links since they will
6368 * be packed into one item. Extended refs will kick in if we
6369 * add more hard links than can fit in the ref item.
6371 key[1].objectid = objectid;
6372 key[1].type = BTRFS_INODE_REF_KEY;
6374 key[1].offset = objectid;
6375 sizes[1] = 2 + sizeof(*ref);
6377 key[1].offset = btrfs_ino(BTRFS_I(dir));
6378 sizes[1] = name->len + sizeof(*ref);
6382 batch.keys = &key[0];
6383 batch.data_sizes = &sizes[0];
6384 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6385 batch.nr = args->orphan ? 1 : 2;
6386 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6388 btrfs_abort_transaction(trans, ret);
6392 inode->i_mtime = current_time(inode);
6393 inode->i_atime = inode->i_mtime;
6394 inode->i_ctime = inode->i_mtime;
6395 BTRFS_I(inode)->i_otime = inode->i_mtime;
6398 * We're going to fill the inode item now, so at this point the inode
6399 * must be fully initialized.
6402 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6403 struct btrfs_inode_item);
6404 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6405 sizeof(*inode_item));
6406 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6408 if (!args->orphan) {
6409 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6410 struct btrfs_inode_ref);
6411 ptr = (unsigned long)(ref + 1);
6413 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6414 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6415 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6417 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6419 btrfs_set_inode_ref_index(path->nodes[0], ref,
6420 BTRFS_I(inode)->dir_index);
6421 write_extent_buffer(path->nodes[0], name->name, ptr,
6426 btrfs_mark_buffer_dirty(path->nodes[0]);
6428 * We don't need the path anymore, plus inheriting properties, adding
6429 * ACLs, security xattrs, orphan item or adding the link, will result in
6430 * allocating yet another path. So just free our path.
6432 btrfs_free_path(path);
6436 struct inode *parent;
6439 * Subvolumes inherit properties from their parent subvolume,
6440 * not the directory they were created in.
6442 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6443 BTRFS_I(dir)->root);
6444 if (IS_ERR(parent)) {
6445 ret = PTR_ERR(parent);
6447 ret = btrfs_inode_inherit_props(trans, inode, parent);
6451 ret = btrfs_inode_inherit_props(trans, inode, dir);
6455 "error inheriting props for ino %llu (root %llu): %d",
6456 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6461 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6464 if (!args->subvol) {
6465 ret = btrfs_init_inode_security(trans, args);
6467 btrfs_abort_transaction(trans, ret);
6472 inode_tree_add(BTRFS_I(inode));
6474 trace_btrfs_inode_new(inode);
6475 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6477 btrfs_update_root_times(trans, root);
6480 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6482 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6483 0, BTRFS_I(inode)->dir_index);
6486 btrfs_abort_transaction(trans, ret);
6494 * discard_new_inode() calls iput(), but the caller owns the reference
6498 discard_new_inode(inode);
6500 btrfs_free_path(path);
6505 * utility function to add 'inode' into 'parent_inode' with
6506 * a give name and a given sequence number.
6507 * if 'add_backref' is true, also insert a backref from the
6508 * inode to the parent directory.
6510 int btrfs_add_link(struct btrfs_trans_handle *trans,
6511 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6512 const struct fscrypt_str *name, int add_backref, u64 index)
6515 struct btrfs_key key;
6516 struct btrfs_root *root = parent_inode->root;
6517 u64 ino = btrfs_ino(inode);
6518 u64 parent_ino = btrfs_ino(parent_inode);
6520 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6521 memcpy(&key, &inode->root->root_key, sizeof(key));
6524 key.type = BTRFS_INODE_ITEM_KEY;
6528 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6529 ret = btrfs_add_root_ref(trans, key.objectid,
6530 root->root_key.objectid, parent_ino,
6532 } else if (add_backref) {
6533 ret = btrfs_insert_inode_ref(trans, root, name,
6534 ino, parent_ino, index);
6537 /* Nothing to clean up yet */
6541 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6542 btrfs_inode_type(&inode->vfs_inode), index);
6543 if (ret == -EEXIST || ret == -EOVERFLOW)
6546 btrfs_abort_transaction(trans, ret);
6550 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6552 inode_inc_iversion(&parent_inode->vfs_inode);
6554 * If we are replaying a log tree, we do not want to update the mtime
6555 * and ctime of the parent directory with the current time, since the
6556 * log replay procedure is responsible for setting them to their correct
6557 * values (the ones it had when the fsync was done).
6559 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6560 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6562 parent_inode->vfs_inode.i_mtime = now;
6563 parent_inode->vfs_inode.i_ctime = now;
6565 ret = btrfs_update_inode(trans, root, parent_inode);
6567 btrfs_abort_transaction(trans, ret);
6571 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6574 err = btrfs_del_root_ref(trans, key.objectid,
6575 root->root_key.objectid, parent_ino,
6576 &local_index, name);
6578 btrfs_abort_transaction(trans, err);
6579 } else if (add_backref) {
6583 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6586 btrfs_abort_transaction(trans, err);
6589 /* Return the original error code */
6593 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6594 struct inode *inode)
6596 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6597 struct btrfs_root *root = BTRFS_I(dir)->root;
6598 struct btrfs_new_inode_args new_inode_args = {
6603 unsigned int trans_num_items;
6604 struct btrfs_trans_handle *trans;
6607 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6611 trans = btrfs_start_transaction(root, trans_num_items);
6612 if (IS_ERR(trans)) {
6613 err = PTR_ERR(trans);
6614 goto out_new_inode_args;
6617 err = btrfs_create_new_inode(trans, &new_inode_args);
6619 d_instantiate_new(dentry, inode);
6621 btrfs_end_transaction(trans);
6622 btrfs_btree_balance_dirty(fs_info);
6624 btrfs_new_inode_args_destroy(&new_inode_args);
6631 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6632 struct dentry *dentry, umode_t mode, dev_t rdev)
6634 struct inode *inode;
6636 inode = new_inode(dir->i_sb);
6639 inode_init_owner(idmap, inode, dir, mode);
6640 inode->i_op = &btrfs_special_inode_operations;
6641 init_special_inode(inode, inode->i_mode, rdev);
6642 return btrfs_create_common(dir, dentry, inode);
6645 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6646 struct dentry *dentry, umode_t mode, bool excl)
6648 struct inode *inode;
6650 inode = new_inode(dir->i_sb);
6653 inode_init_owner(idmap, inode, dir, mode);
6654 inode->i_fop = &btrfs_file_operations;
6655 inode->i_op = &btrfs_file_inode_operations;
6656 inode->i_mapping->a_ops = &btrfs_aops;
6657 return btrfs_create_common(dir, dentry, inode);
6660 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6661 struct dentry *dentry)
6663 struct btrfs_trans_handle *trans = NULL;
6664 struct btrfs_root *root = BTRFS_I(dir)->root;
6665 struct inode *inode = d_inode(old_dentry);
6666 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6667 struct fscrypt_name fname;
6672 /* do not allow sys_link's with other subvols of the same device */
6673 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6676 if (inode->i_nlink >= BTRFS_LINK_MAX)
6679 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6683 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6688 * 2 items for inode and inode ref
6689 * 2 items for dir items
6690 * 1 item for parent inode
6691 * 1 item for orphan item deletion if O_TMPFILE
6693 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6694 if (IS_ERR(trans)) {
6695 err = PTR_ERR(trans);
6700 /* There are several dir indexes for this inode, clear the cache. */
6701 BTRFS_I(inode)->dir_index = 0ULL;
6703 inode_inc_iversion(inode);
6704 inode->i_ctime = current_time(inode);
6706 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6708 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6709 &fname.disk_name, 1, index);
6714 struct dentry *parent = dentry->d_parent;
6716 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6719 if (inode->i_nlink == 1) {
6721 * If new hard link count is 1, it's a file created
6722 * with open(2) O_TMPFILE flag.
6724 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6728 d_instantiate(dentry, inode);
6729 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6733 fscrypt_free_filename(&fname);
6735 btrfs_end_transaction(trans);
6737 inode_dec_link_count(inode);
6740 btrfs_btree_balance_dirty(fs_info);
6744 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6745 struct dentry *dentry, umode_t mode)
6747 struct inode *inode;
6749 inode = new_inode(dir->i_sb);
6752 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6753 inode->i_op = &btrfs_dir_inode_operations;
6754 inode->i_fop = &btrfs_dir_file_operations;
6755 return btrfs_create_common(dir, dentry, inode);
6758 static noinline int uncompress_inline(struct btrfs_path *path,
6760 struct btrfs_file_extent_item *item)
6763 struct extent_buffer *leaf = path->nodes[0];
6766 unsigned long inline_size;
6770 compress_type = btrfs_file_extent_compression(leaf, item);
6771 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6772 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6773 tmp = kmalloc(inline_size, GFP_NOFS);
6776 ptr = btrfs_file_extent_inline_start(item);
6778 read_extent_buffer(leaf, tmp, ptr, inline_size);
6780 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6781 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6784 * decompression code contains a memset to fill in any space between the end
6785 * of the uncompressed data and the end of max_size in case the decompressed
6786 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6787 * the end of an inline extent and the beginning of the next block, so we
6788 * cover that region here.
6791 if (max_size < PAGE_SIZE)
6792 memzero_page(page, max_size, PAGE_SIZE - max_size);
6797 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6800 struct btrfs_file_extent_item *fi;
6804 if (!page || PageUptodate(page))
6807 ASSERT(page_offset(page) == 0);
6809 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6810 struct btrfs_file_extent_item);
6811 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6812 return uncompress_inline(path, page, fi);
6814 copy_size = min_t(u64, PAGE_SIZE,
6815 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6816 kaddr = kmap_local_page(page);
6817 read_extent_buffer(path->nodes[0], kaddr,
6818 btrfs_file_extent_inline_start(fi), copy_size);
6819 kunmap_local(kaddr);
6820 if (copy_size < PAGE_SIZE)
6821 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6826 * Lookup the first extent overlapping a range in a file.
6828 * @inode: file to search in
6829 * @page: page to read extent data into if the extent is inline
6830 * @pg_offset: offset into @page to copy to
6831 * @start: file offset
6832 * @len: length of range starting at @start
6834 * Return the first &struct extent_map which overlaps the given range, reading
6835 * it from the B-tree and caching it if necessary. Note that there may be more
6836 * extents which overlap the given range after the returned extent_map.
6838 * If @page is not NULL and the extent is inline, this also reads the extent
6839 * data directly into the page and marks the extent up to date in the io_tree.
6841 * Return: ERR_PTR on error, non-NULL extent_map on success.
6843 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6844 struct page *page, size_t pg_offset,
6847 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6849 u64 extent_start = 0;
6851 u64 objectid = btrfs_ino(inode);
6852 int extent_type = -1;
6853 struct btrfs_path *path = NULL;
6854 struct btrfs_root *root = inode->root;
6855 struct btrfs_file_extent_item *item;
6856 struct extent_buffer *leaf;
6857 struct btrfs_key found_key;
6858 struct extent_map *em = NULL;
6859 struct extent_map_tree *em_tree = &inode->extent_tree;
6861 read_lock(&em_tree->lock);
6862 em = lookup_extent_mapping(em_tree, start, len);
6863 read_unlock(&em_tree->lock);
6866 if (em->start > start || em->start + em->len <= start)
6867 free_extent_map(em);
6868 else if (em->block_start == EXTENT_MAP_INLINE && page)
6869 free_extent_map(em);
6873 em = alloc_extent_map();
6878 em->start = EXTENT_MAP_HOLE;
6879 em->orig_start = EXTENT_MAP_HOLE;
6881 em->block_len = (u64)-1;
6883 path = btrfs_alloc_path();
6889 /* Chances are we'll be called again, so go ahead and do readahead */
6890 path->reada = READA_FORWARD;
6893 * The same explanation in load_free_space_cache applies here as well,
6894 * we only read when we're loading the free space cache, and at that
6895 * point the commit_root has everything we need.
6897 if (btrfs_is_free_space_inode(inode)) {
6898 path->search_commit_root = 1;
6899 path->skip_locking = 1;
6902 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6905 } else if (ret > 0) {
6906 if (path->slots[0] == 0)
6912 leaf = path->nodes[0];
6913 item = btrfs_item_ptr(leaf, path->slots[0],
6914 struct btrfs_file_extent_item);
6915 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6916 if (found_key.objectid != objectid ||
6917 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6919 * If we backup past the first extent we want to move forward
6920 * and see if there is an extent in front of us, otherwise we'll
6921 * say there is a hole for our whole search range which can
6928 extent_type = btrfs_file_extent_type(leaf, item);
6929 extent_start = found_key.offset;
6930 extent_end = btrfs_file_extent_end(path);
6931 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6932 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6933 /* Only regular file could have regular/prealloc extent */
6934 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6937 "regular/prealloc extent found for non-regular inode %llu",
6941 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6943 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6944 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6949 if (start >= extent_end) {
6951 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6952 ret = btrfs_next_leaf(root, path);
6958 leaf = path->nodes[0];
6960 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6961 if (found_key.objectid != objectid ||
6962 found_key.type != BTRFS_EXTENT_DATA_KEY)
6964 if (start + len <= found_key.offset)
6966 if (start > found_key.offset)
6969 /* New extent overlaps with existing one */
6971 em->orig_start = start;
6972 em->len = found_key.offset - start;
6973 em->block_start = EXTENT_MAP_HOLE;
6977 btrfs_extent_item_to_extent_map(inode, path, item, em);
6979 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6980 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6982 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6984 * Inline extent can only exist at file offset 0. This is
6985 * ensured by tree-checker and inline extent creation path.
6986 * Thus all members representing file offsets should be zero.
6988 ASSERT(pg_offset == 0);
6989 ASSERT(extent_start == 0);
6990 ASSERT(em->start == 0);
6993 * btrfs_extent_item_to_extent_map() should have properly
6994 * initialized em members already.
6996 * Other members are not utilized for inline extents.
6998 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6999 ASSERT(em->len == fs_info->sectorsize);
7001 ret = read_inline_extent(inode, path, page);
7008 em->orig_start = start;
7010 em->block_start = EXTENT_MAP_HOLE;
7013 btrfs_release_path(path);
7014 if (em->start > start || extent_map_end(em) <= start) {
7016 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7017 em->start, em->len, start, len);
7022 write_lock(&em_tree->lock);
7023 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7024 write_unlock(&em_tree->lock);
7026 btrfs_free_path(path);
7028 trace_btrfs_get_extent(root, inode, em);
7031 free_extent_map(em);
7032 return ERR_PTR(ret);
7037 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7038 struct btrfs_dio_data *dio_data,
7041 const u64 orig_start,
7042 const u64 block_start,
7043 const u64 block_len,
7044 const u64 orig_block_len,
7045 const u64 ram_bytes,
7048 struct extent_map *em = NULL;
7049 struct btrfs_ordered_extent *ordered;
7051 if (type != BTRFS_ORDERED_NOCOW) {
7052 em = create_io_em(inode, start, len, orig_start, block_start,
7053 block_len, orig_block_len, ram_bytes,
7054 BTRFS_COMPRESS_NONE, /* compress_type */
7059 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
7060 block_start, block_len, 0,
7062 (1 << BTRFS_ORDERED_DIRECT),
7063 BTRFS_COMPRESS_NONE);
7064 if (IS_ERR(ordered)) {
7066 free_extent_map(em);
7067 btrfs_drop_extent_map_range(inode, start,
7068 start + len - 1, false);
7070 em = ERR_CAST(ordered);
7072 ASSERT(!dio_data->ordered);
7073 dio_data->ordered = ordered;
7080 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7081 struct btrfs_dio_data *dio_data,
7084 struct btrfs_root *root = inode->root;
7085 struct btrfs_fs_info *fs_info = root->fs_info;
7086 struct extent_map *em;
7087 struct btrfs_key ins;
7091 alloc_hint = get_extent_allocation_hint(inode, start, len);
7092 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7093 0, alloc_hint, &ins, 1, 1);
7095 return ERR_PTR(ret);
7097 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7098 ins.objectid, ins.offset, ins.offset,
7099 ins.offset, BTRFS_ORDERED_REGULAR);
7100 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7102 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7108 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7110 struct btrfs_block_group *block_group;
7111 bool readonly = false;
7113 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7114 if (!block_group || block_group->ro)
7117 btrfs_put_block_group(block_group);
7122 * Check if we can do nocow write into the range [@offset, @offset + @len)
7124 * @offset: File offset
7125 * @len: The length to write, will be updated to the nocow writeable
7127 * @orig_start: (optional) Return the original file offset of the file extent
7128 * @orig_len: (optional) Return the original on-disk length of the file extent
7129 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7130 * @strict: if true, omit optimizations that might force us into unnecessary
7131 * cow. e.g., don't trust generation number.
7134 * >0 and update @len if we can do nocow write
7135 * 0 if we can't do nocow write
7136 * <0 if error happened
7138 * NOTE: This only checks the file extents, caller is responsible to wait for
7139 * any ordered extents.
7141 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7142 u64 *orig_start, u64 *orig_block_len,
7143 u64 *ram_bytes, bool nowait, bool strict)
7145 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7146 struct can_nocow_file_extent_args nocow_args = { 0 };
7147 struct btrfs_path *path;
7149 struct extent_buffer *leaf;
7150 struct btrfs_root *root = BTRFS_I(inode)->root;
7151 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7152 struct btrfs_file_extent_item *fi;
7153 struct btrfs_key key;
7156 path = btrfs_alloc_path();
7159 path->nowait = nowait;
7161 ret = btrfs_lookup_file_extent(NULL, root, path,
7162 btrfs_ino(BTRFS_I(inode)), offset, 0);
7167 if (path->slots[0] == 0) {
7168 /* can't find the item, must cow */
7175 leaf = path->nodes[0];
7176 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7177 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7178 key.type != BTRFS_EXTENT_DATA_KEY) {
7179 /* not our file or wrong item type, must cow */
7183 if (key.offset > offset) {
7184 /* Wrong offset, must cow */
7188 if (btrfs_file_extent_end(path) <= offset)
7191 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7192 found_type = btrfs_file_extent_type(leaf, fi);
7194 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7196 nocow_args.start = offset;
7197 nocow_args.end = offset + *len - 1;
7198 nocow_args.strict = strict;
7199 nocow_args.free_path = true;
7201 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7202 /* can_nocow_file_extent() has freed the path. */
7206 /* Treat errors as not being able to NOCOW. */
7212 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7215 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7216 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7219 range_end = round_up(offset + nocow_args.num_bytes,
7220 root->fs_info->sectorsize) - 1;
7221 ret = test_range_bit(io_tree, offset, range_end,
7222 EXTENT_DELALLOC, 0, NULL);
7230 *orig_start = key.offset - nocow_args.extent_offset;
7232 *orig_block_len = nocow_args.disk_num_bytes;
7234 *len = nocow_args.num_bytes;
7237 btrfs_free_path(path);
7241 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7242 struct extent_state **cached_state,
7243 unsigned int iomap_flags)
7245 const bool writing = (iomap_flags & IOMAP_WRITE);
7246 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7247 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7248 struct btrfs_ordered_extent *ordered;
7253 if (!try_lock_extent(io_tree, lockstart, lockend,
7257 lock_extent(io_tree, lockstart, lockend, cached_state);
7260 * We're concerned with the entire range that we're going to be
7261 * doing DIO to, so we need to make sure there's no ordered
7262 * extents in this range.
7264 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7265 lockend - lockstart + 1);
7268 * We need to make sure there are no buffered pages in this
7269 * range either, we could have raced between the invalidate in
7270 * generic_file_direct_write and locking the extent. The
7271 * invalidate needs to happen so that reads after a write do not
7275 (!writing || !filemap_range_has_page(inode->i_mapping,
7276 lockstart, lockend)))
7279 unlock_extent(io_tree, lockstart, lockend, cached_state);
7283 btrfs_put_ordered_extent(ordered);
7288 * If we are doing a DIO read and the ordered extent we
7289 * found is for a buffered write, we can not wait for it
7290 * to complete and retry, because if we do so we can
7291 * deadlock with concurrent buffered writes on page
7292 * locks. This happens only if our DIO read covers more
7293 * than one extent map, if at this point has already
7294 * created an ordered extent for a previous extent map
7295 * and locked its range in the inode's io tree, and a
7296 * concurrent write against that previous extent map's
7297 * range and this range started (we unlock the ranges
7298 * in the io tree only when the bios complete and
7299 * buffered writes always lock pages before attempting
7300 * to lock range in the io tree).
7303 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7304 btrfs_start_ordered_extent(ordered);
7306 ret = nowait ? -EAGAIN : -ENOTBLK;
7307 btrfs_put_ordered_extent(ordered);
7310 * We could trigger writeback for this range (and wait
7311 * for it to complete) and then invalidate the pages for
7312 * this range (through invalidate_inode_pages2_range()),
7313 * but that can lead us to a deadlock with a concurrent
7314 * call to readahead (a buffered read or a defrag call
7315 * triggered a readahead) on a page lock due to an
7316 * ordered dio extent we created before but did not have
7317 * yet a corresponding bio submitted (whence it can not
7318 * complete), which makes readahead wait for that
7319 * ordered extent to complete while holding a lock on
7322 ret = nowait ? -EAGAIN : -ENOTBLK;
7334 /* The callers of this must take lock_extent() */
7335 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7336 u64 len, u64 orig_start, u64 block_start,
7337 u64 block_len, u64 orig_block_len,
7338 u64 ram_bytes, int compress_type,
7341 struct extent_map *em;
7344 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7345 type == BTRFS_ORDERED_COMPRESSED ||
7346 type == BTRFS_ORDERED_NOCOW ||
7347 type == BTRFS_ORDERED_REGULAR);
7349 em = alloc_extent_map();
7351 return ERR_PTR(-ENOMEM);
7354 em->orig_start = orig_start;
7356 em->block_len = block_len;
7357 em->block_start = block_start;
7358 em->orig_block_len = orig_block_len;
7359 em->ram_bytes = ram_bytes;
7360 em->generation = -1;
7361 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7362 if (type == BTRFS_ORDERED_PREALLOC) {
7363 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7364 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7365 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7366 em->compress_type = compress_type;
7369 ret = btrfs_replace_extent_map_range(inode, em, true);
7371 free_extent_map(em);
7372 return ERR_PTR(ret);
7375 /* em got 2 refs now, callers needs to do free_extent_map once. */
7380 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7381 struct inode *inode,
7382 struct btrfs_dio_data *dio_data,
7383 u64 start, u64 *lenp,
7384 unsigned int iomap_flags)
7386 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7387 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7388 struct extent_map *em = *map;
7390 u64 block_start, orig_start, orig_block_len, ram_bytes;
7391 struct btrfs_block_group *bg;
7392 bool can_nocow = false;
7393 bool space_reserved = false;
7399 * We don't allocate a new extent in the following cases
7401 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7403 * 2) The extent is marked as PREALLOC. We're good to go here and can
7404 * just use the extent.
7407 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7408 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7409 em->block_start != EXTENT_MAP_HOLE)) {
7410 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7411 type = BTRFS_ORDERED_PREALLOC;
7413 type = BTRFS_ORDERED_NOCOW;
7414 len = min(len, em->len - (start - em->start));
7415 block_start = em->block_start + (start - em->start);
7417 if (can_nocow_extent(inode, start, &len, &orig_start,
7418 &orig_block_len, &ram_bytes, false, false) == 1) {
7419 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7427 struct extent_map *em2;
7429 /* We can NOCOW, so only need to reserve metadata space. */
7430 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7433 /* Our caller expects us to free the input extent map. */
7434 free_extent_map(em);
7436 btrfs_dec_nocow_writers(bg);
7437 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7441 space_reserved = true;
7443 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7444 orig_start, block_start,
7445 len, orig_block_len,
7447 btrfs_dec_nocow_writers(bg);
7448 if (type == BTRFS_ORDERED_PREALLOC) {
7449 free_extent_map(em);
7459 dio_data->nocow_done = true;
7461 /* Our caller expects us to free the input extent map. */
7462 free_extent_map(em);
7471 * If we could not allocate data space before locking the file
7472 * range and we can't do a NOCOW write, then we have to fail.
7474 if (!dio_data->data_space_reserved) {
7480 * We have to COW and we have already reserved data space before,
7481 * so now we reserve only metadata.
7483 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7487 space_reserved = true;
7489 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7495 len = min(len, em->len - (start - em->start));
7497 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7498 prev_len - len, true);
7502 * We have created our ordered extent, so we can now release our reservation
7503 * for an outstanding extent.
7505 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7508 * Need to update the i_size under the extent lock so buffered
7509 * readers will get the updated i_size when we unlock.
7511 if (start + len > i_size_read(inode))
7512 i_size_write(inode, start + len);
7514 if (ret && space_reserved) {
7515 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7516 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7522 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7523 loff_t length, unsigned int flags, struct iomap *iomap,
7524 struct iomap *srcmap)
7526 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7527 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7528 struct extent_map *em;
7529 struct extent_state *cached_state = NULL;
7530 struct btrfs_dio_data *dio_data = iter->private;
7531 u64 lockstart, lockend;
7532 const bool write = !!(flags & IOMAP_WRITE);
7535 const u64 data_alloc_len = length;
7536 bool unlock_extents = false;
7539 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7540 * we're NOWAIT we may submit a bio for a partial range and return
7541 * EIOCBQUEUED, which would result in an errant short read.
7543 * The best way to handle this would be to allow for partial completions
7544 * of iocb's, so we could submit the partial bio, return and fault in
7545 * the rest of the pages, and then submit the io for the rest of the
7546 * range. However we don't have that currently, so simply return
7547 * -EAGAIN at this point so that the normal path is used.
7549 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7553 * Cap the size of reads to that usually seen in buffered I/O as we need
7554 * to allocate a contiguous array for the checksums.
7557 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7560 lockend = start + len - 1;
7563 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7564 * enough if we've written compressed pages to this area, so we need to
7565 * flush the dirty pages again to make absolutely sure that any
7566 * outstanding dirty pages are on disk - the first flush only starts
7567 * compression on the data, while keeping the pages locked, so by the
7568 * time the second flush returns we know bios for the compressed pages
7569 * were submitted and finished, and the pages no longer under writeback.
7571 * If we have a NOWAIT request and we have any pages in the range that
7572 * are locked, likely due to compression still in progress, we don't want
7573 * to block on page locks. We also don't want to block on pages marked as
7574 * dirty or under writeback (same as for the non-compression case).
7575 * iomap_dio_rw() did the same check, but after that and before we got
7576 * here, mmap'ed writes may have happened or buffered reads started
7577 * (readpage() and readahead(), which lock pages), as we haven't locked
7578 * the file range yet.
7580 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7581 &BTRFS_I(inode)->runtime_flags)) {
7582 if (flags & IOMAP_NOWAIT) {
7583 if (filemap_range_needs_writeback(inode->i_mapping,
7584 lockstart, lockend))
7587 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7588 start + length - 1);
7594 memset(dio_data, 0, sizeof(*dio_data));
7597 * We always try to allocate data space and must do it before locking
7598 * the file range, to avoid deadlocks with concurrent writes to the same
7599 * range if the range has several extents and the writes don't expand the
7600 * current i_size (the inode lock is taken in shared mode). If we fail to
7601 * allocate data space here we continue and later, after locking the
7602 * file range, we fail with ENOSPC only if we figure out we can not do a
7605 if (write && !(flags & IOMAP_NOWAIT)) {
7606 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7607 &dio_data->data_reserved,
7608 start, data_alloc_len, false);
7610 dio_data->data_space_reserved = true;
7611 else if (ret && !(BTRFS_I(inode)->flags &
7612 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7617 * If this errors out it's because we couldn't invalidate pagecache for
7618 * this range and we need to fallback to buffered IO, or we are doing a
7619 * NOWAIT read/write and we need to block.
7621 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7625 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7632 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7633 * io. INLINE is special, and we could probably kludge it in here, but
7634 * it's still buffered so for safety lets just fall back to the generic
7637 * For COMPRESSED we _have_ to read the entire extent in so we can
7638 * decompress it, so there will be buffering required no matter what we
7639 * do, so go ahead and fallback to buffered.
7641 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7642 * to buffered IO. Don't blame me, this is the price we pay for using
7645 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7646 em->block_start == EXTENT_MAP_INLINE) {
7647 free_extent_map(em);
7649 * If we are in a NOWAIT context, return -EAGAIN in order to
7650 * fallback to buffered IO. This is not only because we can
7651 * block with buffered IO (no support for NOWAIT semantics at
7652 * the moment) but also to avoid returning short reads to user
7653 * space - this happens if we were able to read some data from
7654 * previous non-compressed extents and then when we fallback to
7655 * buffered IO, at btrfs_file_read_iter() by calling
7656 * filemap_read(), we fail to fault in pages for the read buffer,
7657 * in which case filemap_read() returns a short read (the number
7658 * of bytes previously read is > 0, so it does not return -EFAULT).
7660 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7664 len = min(len, em->len - (start - em->start));
7667 * If we have a NOWAIT request and the range contains multiple extents
7668 * (or a mix of extents and holes), then we return -EAGAIN to make the
7669 * caller fallback to a context where it can do a blocking (without
7670 * NOWAIT) request. This way we avoid doing partial IO and returning
7671 * success to the caller, which is not optimal for writes and for reads
7672 * it can result in unexpected behaviour for an application.
7674 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7675 * iomap_dio_rw(), we can end up returning less data then what the caller
7676 * asked for, resulting in an unexpected, and incorrect, short read.
7677 * That is, the caller asked to read N bytes and we return less than that,
7678 * which is wrong unless we are crossing EOF. This happens if we get a
7679 * page fault error when trying to fault in pages for the buffer that is
7680 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7681 * have previously submitted bios for other extents in the range, in
7682 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7683 * those bios have completed by the time we get the page fault error,
7684 * which we return back to our caller - we should only return EIOCBQUEUED
7685 * after we have submitted bios for all the extents in the range.
7687 if ((flags & IOMAP_NOWAIT) && len < length) {
7688 free_extent_map(em);
7694 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7695 start, &len, flags);
7698 unlock_extents = true;
7699 /* Recalc len in case the new em is smaller than requested */
7700 len = min(len, em->len - (start - em->start));
7701 if (dio_data->data_space_reserved) {
7703 u64 release_len = 0;
7705 if (dio_data->nocow_done) {
7706 release_offset = start;
7707 release_len = data_alloc_len;
7708 } else if (len < data_alloc_len) {
7709 release_offset = start + len;
7710 release_len = data_alloc_len - len;
7713 if (release_len > 0)
7714 btrfs_free_reserved_data_space(BTRFS_I(inode),
7715 dio_data->data_reserved,
7721 * We need to unlock only the end area that we aren't using.
7722 * The rest is going to be unlocked by the endio routine.
7724 lockstart = start + len;
7725 if (lockstart < lockend)
7726 unlock_extents = true;
7730 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7733 free_extent_state(cached_state);
7736 * Translate extent map information to iomap.
7737 * We trim the extents (and move the addr) even though iomap code does
7738 * that, since we have locked only the parts we are performing I/O in.
7740 if ((em->block_start == EXTENT_MAP_HOLE) ||
7741 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7742 iomap->addr = IOMAP_NULL_ADDR;
7743 iomap->type = IOMAP_HOLE;
7745 iomap->addr = em->block_start + (start - em->start);
7746 iomap->type = IOMAP_MAPPED;
7748 iomap->offset = start;
7749 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7750 iomap->length = len;
7751 free_extent_map(em);
7756 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7759 if (dio_data->data_space_reserved) {
7760 btrfs_free_reserved_data_space(BTRFS_I(inode),
7761 dio_data->data_reserved,
7762 start, data_alloc_len);
7763 extent_changeset_free(dio_data->data_reserved);
7769 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7770 ssize_t written, unsigned int flags, struct iomap *iomap)
7772 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7773 struct btrfs_dio_data *dio_data = iter->private;
7774 size_t submitted = dio_data->submitted;
7775 const bool write = !!(flags & IOMAP_WRITE);
7778 if (!write && (iomap->type == IOMAP_HOLE)) {
7779 /* If reading from a hole, unlock and return */
7780 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7785 if (submitted < length) {
7787 length -= submitted;
7789 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7790 pos, length, false);
7792 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7793 pos + length - 1, NULL);
7797 btrfs_put_ordered_extent(dio_data->ordered);
7798 dio_data->ordered = NULL;
7802 extent_changeset_free(dio_data->data_reserved);
7806 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7808 struct btrfs_dio_private *dip =
7809 container_of(bbio, struct btrfs_dio_private, bbio);
7810 struct btrfs_inode *inode = bbio->inode;
7811 struct bio *bio = &bbio->bio;
7813 if (bio->bi_status) {
7814 btrfs_warn(inode->root->fs_info,
7815 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7816 btrfs_ino(inode), bio->bi_opf,
7817 dip->file_offset, dip->bytes, bio->bi_status);
7820 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7821 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7822 dip->file_offset, dip->bytes,
7825 unlock_extent(&inode->io_tree, dip->file_offset,
7826 dip->file_offset + dip->bytes - 1, NULL);
7829 bbio->bio.bi_private = bbio->private;
7830 iomap_dio_bio_end_io(bio);
7833 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7836 struct btrfs_bio *bbio = btrfs_bio(bio);
7837 struct btrfs_dio_private *dip =
7838 container_of(bbio, struct btrfs_dio_private, bbio);
7839 struct btrfs_dio_data *dio_data = iter->private;
7841 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7842 btrfs_dio_end_io, bio->bi_private);
7843 bbio->inode = BTRFS_I(iter->inode);
7844 bbio->file_offset = file_offset;
7846 dip->file_offset = file_offset;
7847 dip->bytes = bio->bi_iter.bi_size;
7849 dio_data->submitted += bio->bi_iter.bi_size;
7852 * Check if we are doing a partial write. If we are, we need to split
7853 * the ordered extent to match the submitted bio. Hang on to the
7854 * remaining unfinishable ordered_extent in dio_data so that it can be
7855 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7856 * remaining pages is blocked on the outstanding ordered extent.
7858 if (iter->flags & IOMAP_WRITE) {
7861 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7863 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7864 file_offset, dip->bytes,
7866 bio->bi_status = errno_to_blk_status(ret);
7867 iomap_dio_bio_end_io(bio);
7872 btrfs_submit_bio(bbio, 0);
7875 static const struct iomap_ops btrfs_dio_iomap_ops = {
7876 .iomap_begin = btrfs_dio_iomap_begin,
7877 .iomap_end = btrfs_dio_iomap_end,
7880 static const struct iomap_dio_ops btrfs_dio_ops = {
7881 .submit_io = btrfs_dio_submit_io,
7882 .bio_set = &btrfs_dio_bioset,
7885 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7887 struct btrfs_dio_data data = { 0 };
7889 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7890 IOMAP_DIO_PARTIAL, &data, done_before);
7893 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7896 struct btrfs_dio_data data = { 0 };
7898 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7899 IOMAP_DIO_PARTIAL, &data, done_before);
7902 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7907 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7912 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7913 * file range (0 to LLONG_MAX), but that is not enough if we have
7914 * compression enabled. The first filemap_fdatawrite_range() only kicks
7915 * in the compression of data (in an async thread) and will return
7916 * before the compression is done and writeback is started. A second
7917 * filemap_fdatawrite_range() is needed to wait for the compression to
7918 * complete and writeback to start. We also need to wait for ordered
7919 * extents to complete, because our fiemap implementation uses mainly
7920 * file extent items to list the extents, searching for extent maps
7921 * only for file ranges with holes or prealloc extents to figure out
7922 * if we have delalloc in those ranges.
7924 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7925 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7930 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7933 static int btrfs_writepages(struct address_space *mapping,
7934 struct writeback_control *wbc)
7936 return extent_writepages(mapping, wbc);
7939 static void btrfs_readahead(struct readahead_control *rac)
7941 extent_readahead(rac);
7945 * For release_folio() and invalidate_folio() we have a race window where
7946 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7947 * If we continue to release/invalidate the page, we could cause use-after-free
7948 * for subpage spinlock. So this function is to spin and wait for subpage
7951 static void wait_subpage_spinlock(struct page *page)
7953 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7954 struct btrfs_subpage *subpage;
7956 if (!btrfs_is_subpage(fs_info, page))
7959 ASSERT(PagePrivate(page) && page->private);
7960 subpage = (struct btrfs_subpage *)page->private;
7963 * This may look insane as we just acquire the spinlock and release it,
7964 * without doing anything. But we just want to make sure no one is
7965 * still holding the subpage spinlock.
7966 * And since the page is not dirty nor writeback, and we have page
7967 * locked, the only possible way to hold a spinlock is from the endio
7968 * function to clear page writeback.
7970 * Here we just acquire the spinlock so that all existing callers
7971 * should exit and we're safe to release/invalidate the page.
7973 spin_lock_irq(&subpage->lock);
7974 spin_unlock_irq(&subpage->lock);
7977 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7979 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7982 wait_subpage_spinlock(&folio->page);
7983 clear_page_extent_mapped(&folio->page);
7988 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7990 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7992 return __btrfs_release_folio(folio, gfp_flags);
7995 #ifdef CONFIG_MIGRATION
7996 static int btrfs_migrate_folio(struct address_space *mapping,
7997 struct folio *dst, struct folio *src,
7998 enum migrate_mode mode)
8000 int ret = filemap_migrate_folio(mapping, dst, src, mode);
8002 if (ret != MIGRATEPAGE_SUCCESS)
8005 if (folio_test_ordered(src)) {
8006 folio_clear_ordered(src);
8007 folio_set_ordered(dst);
8010 return MIGRATEPAGE_SUCCESS;
8013 #define btrfs_migrate_folio NULL
8016 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8019 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8020 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8021 struct extent_io_tree *tree = &inode->io_tree;
8022 struct extent_state *cached_state = NULL;
8023 u64 page_start = folio_pos(folio);
8024 u64 page_end = page_start + folio_size(folio) - 1;
8026 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8029 * We have folio locked so no new ordered extent can be created on this
8030 * page, nor bio can be submitted for this folio.
8032 * But already submitted bio can still be finished on this folio.
8033 * Furthermore, endio function won't skip folio which has Ordered
8034 * (Private2) already cleared, so it's possible for endio and
8035 * invalidate_folio to do the same ordered extent accounting twice
8038 * So here we wait for any submitted bios to finish, so that we won't
8039 * do double ordered extent accounting on the same folio.
8041 folio_wait_writeback(folio);
8042 wait_subpage_spinlock(&folio->page);
8045 * For subpage case, we have call sites like
8046 * btrfs_punch_hole_lock_range() which passes range not aligned to
8048 * If the range doesn't cover the full folio, we don't need to and
8049 * shouldn't clear page extent mapped, as folio->private can still
8050 * record subpage dirty bits for other part of the range.
8052 * For cases that invalidate the full folio even the range doesn't
8053 * cover the full folio, like invalidating the last folio, we're
8054 * still safe to wait for ordered extent to finish.
8056 if (!(offset == 0 && length == folio_size(folio))) {
8057 btrfs_release_folio(folio, GFP_NOFS);
8061 if (!inode_evicting)
8062 lock_extent(tree, page_start, page_end, &cached_state);
8065 while (cur < page_end) {
8066 struct btrfs_ordered_extent *ordered;
8069 u32 extra_flags = 0;
8071 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8072 page_end + 1 - cur);
8074 range_end = page_end;
8076 * No ordered extent covering this range, we are safe
8077 * to delete all extent states in the range.
8079 extra_flags = EXTENT_CLEAR_ALL_BITS;
8082 if (ordered->file_offset > cur) {
8084 * There is a range between [cur, oe->file_offset) not
8085 * covered by any ordered extent.
8086 * We are safe to delete all extent states, and handle
8087 * the ordered extent in the next iteration.
8089 range_end = ordered->file_offset - 1;
8090 extra_flags = EXTENT_CLEAR_ALL_BITS;
8094 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8096 ASSERT(range_end + 1 - cur < U32_MAX);
8097 range_len = range_end + 1 - cur;
8098 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8100 * If Ordered (Private2) is cleared, it means endio has
8101 * already been executed for the range.
8102 * We can't delete the extent states as
8103 * btrfs_finish_ordered_io() may still use some of them.
8107 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8110 * IO on this page will never be started, so we need to account
8111 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8112 * here, must leave that up for the ordered extent completion.
8114 * This will also unlock the range for incoming
8115 * btrfs_finish_ordered_io().
8117 if (!inode_evicting)
8118 clear_extent_bit(tree, cur, range_end,
8120 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8121 EXTENT_DEFRAG, &cached_state);
8123 spin_lock_irq(&inode->ordered_tree.lock);
8124 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8125 ordered->truncated_len = min(ordered->truncated_len,
8126 cur - ordered->file_offset);
8127 spin_unlock_irq(&inode->ordered_tree.lock);
8130 * If the ordered extent has finished, we're safe to delete all
8131 * the extent states of the range, otherwise
8132 * btrfs_finish_ordered_io() will get executed by endio for
8133 * other pages, so we can't delete extent states.
8135 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8136 cur, range_end + 1 - cur)) {
8137 btrfs_finish_ordered_io(ordered);
8139 * The ordered extent has finished, now we're again
8140 * safe to delete all extent states of the range.
8142 extra_flags = EXTENT_CLEAR_ALL_BITS;
8146 btrfs_put_ordered_extent(ordered);
8148 * Qgroup reserved space handler
8149 * Sector(s) here will be either:
8151 * 1) Already written to disk or bio already finished
8152 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8153 * Qgroup will be handled by its qgroup_record then.
8154 * btrfs_qgroup_free_data() call will do nothing here.
8156 * 2) Not written to disk yet
8157 * Then btrfs_qgroup_free_data() call will clear the
8158 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8159 * reserved data space.
8160 * Since the IO will never happen for this page.
8162 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8163 if (!inode_evicting) {
8164 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8165 EXTENT_DELALLOC | EXTENT_UPTODATE |
8166 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8167 extra_flags, &cached_state);
8169 cur = range_end + 1;
8172 * We have iterated through all ordered extents of the page, the page
8173 * should not have Ordered (Private2) anymore, or the above iteration
8174 * did something wrong.
8176 ASSERT(!folio_test_ordered(folio));
8177 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8178 if (!inode_evicting)
8179 __btrfs_release_folio(folio, GFP_NOFS);
8180 clear_page_extent_mapped(&folio->page);
8184 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8185 * called from a page fault handler when a page is first dirtied. Hence we must
8186 * be careful to check for EOF conditions here. We set the page up correctly
8187 * for a written page which means we get ENOSPC checking when writing into
8188 * holes and correct delalloc and unwritten extent mapping on filesystems that
8189 * support these features.
8191 * We are not allowed to take the i_mutex here so we have to play games to
8192 * protect against truncate races as the page could now be beyond EOF. Because
8193 * truncate_setsize() writes the inode size before removing pages, once we have
8194 * the page lock we can determine safely if the page is beyond EOF. If it is not
8195 * beyond EOF, then the page is guaranteed safe against truncation until we
8198 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8200 struct page *page = vmf->page;
8201 struct inode *inode = file_inode(vmf->vma->vm_file);
8202 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8203 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8204 struct btrfs_ordered_extent *ordered;
8205 struct extent_state *cached_state = NULL;
8206 struct extent_changeset *data_reserved = NULL;
8207 unsigned long zero_start;
8217 reserved_space = PAGE_SIZE;
8219 sb_start_pagefault(inode->i_sb);
8220 page_start = page_offset(page);
8221 page_end = page_start + PAGE_SIZE - 1;
8225 * Reserving delalloc space after obtaining the page lock can lead to
8226 * deadlock. For example, if a dirty page is locked by this function
8227 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8228 * dirty page write out, then the btrfs_writepages() function could
8229 * end up waiting indefinitely to get a lock on the page currently
8230 * being processed by btrfs_page_mkwrite() function.
8232 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8233 page_start, reserved_space);
8235 ret2 = file_update_time(vmf->vma->vm_file);
8239 ret = vmf_error(ret2);
8245 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8247 down_read(&BTRFS_I(inode)->i_mmap_lock);
8249 size = i_size_read(inode);
8251 if ((page->mapping != inode->i_mapping) ||
8252 (page_start >= size)) {
8253 /* page got truncated out from underneath us */
8256 wait_on_page_writeback(page);
8258 lock_extent(io_tree, page_start, page_end, &cached_state);
8259 ret2 = set_page_extent_mapped(page);
8261 ret = vmf_error(ret2);
8262 unlock_extent(io_tree, page_start, page_end, &cached_state);
8267 * we can't set the delalloc bits if there are pending ordered
8268 * extents. Drop our locks and wait for them to finish
8270 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8273 unlock_extent(io_tree, page_start, page_end, &cached_state);
8275 up_read(&BTRFS_I(inode)->i_mmap_lock);
8276 btrfs_start_ordered_extent(ordered);
8277 btrfs_put_ordered_extent(ordered);
8281 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8282 reserved_space = round_up(size - page_start,
8283 fs_info->sectorsize);
8284 if (reserved_space < PAGE_SIZE) {
8285 end = page_start + reserved_space - 1;
8286 btrfs_delalloc_release_space(BTRFS_I(inode),
8287 data_reserved, page_start,
8288 PAGE_SIZE - reserved_space, true);
8293 * page_mkwrite gets called when the page is firstly dirtied after it's
8294 * faulted in, but write(2) could also dirty a page and set delalloc
8295 * bits, thus in this case for space account reason, we still need to
8296 * clear any delalloc bits within this page range since we have to
8297 * reserve data&meta space before lock_page() (see above comments).
8299 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8300 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8301 EXTENT_DEFRAG, &cached_state);
8303 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8306 unlock_extent(io_tree, page_start, page_end, &cached_state);
8307 ret = VM_FAULT_SIGBUS;
8311 /* page is wholly or partially inside EOF */
8312 if (page_start + PAGE_SIZE > size)
8313 zero_start = offset_in_page(size);
8315 zero_start = PAGE_SIZE;
8317 if (zero_start != PAGE_SIZE)
8318 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8320 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8321 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8322 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8324 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8326 unlock_extent(io_tree, page_start, page_end, &cached_state);
8327 up_read(&BTRFS_I(inode)->i_mmap_lock);
8329 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8330 sb_end_pagefault(inode->i_sb);
8331 extent_changeset_free(data_reserved);
8332 return VM_FAULT_LOCKED;
8336 up_read(&BTRFS_I(inode)->i_mmap_lock);
8338 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8339 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8340 reserved_space, (ret != 0));
8342 sb_end_pagefault(inode->i_sb);
8343 extent_changeset_free(data_reserved);
8347 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8349 struct btrfs_truncate_control control = {
8351 .ino = btrfs_ino(inode),
8352 .min_type = BTRFS_EXTENT_DATA_KEY,
8353 .clear_extent_range = true,
8355 struct btrfs_root *root = inode->root;
8356 struct btrfs_fs_info *fs_info = root->fs_info;
8357 struct btrfs_block_rsv *rsv;
8359 struct btrfs_trans_handle *trans;
8360 u64 mask = fs_info->sectorsize - 1;
8361 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8363 if (!skip_writeback) {
8364 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8365 inode->vfs_inode.i_size & (~mask),
8372 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8373 * things going on here:
8375 * 1) We need to reserve space to update our inode.
8377 * 2) We need to have something to cache all the space that is going to
8378 * be free'd up by the truncate operation, but also have some slack
8379 * space reserved in case it uses space during the truncate (thank you
8380 * very much snapshotting).
8382 * And we need these to be separate. The fact is we can use a lot of
8383 * space doing the truncate, and we have no earthly idea how much space
8384 * we will use, so we need the truncate reservation to be separate so it
8385 * doesn't end up using space reserved for updating the inode. We also
8386 * need to be able to stop the transaction and start a new one, which
8387 * means we need to be able to update the inode several times, and we
8388 * have no idea of knowing how many times that will be, so we can't just
8389 * reserve 1 item for the entirety of the operation, so that has to be
8390 * done separately as well.
8392 * So that leaves us with
8394 * 1) rsv - for the truncate reservation, which we will steal from the
8395 * transaction reservation.
8396 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8397 * updating the inode.
8399 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8402 rsv->size = min_size;
8403 rsv->failfast = true;
8406 * 1 for the truncate slack space
8407 * 1 for updating the inode.
8409 trans = btrfs_start_transaction(root, 2);
8410 if (IS_ERR(trans)) {
8411 ret = PTR_ERR(trans);
8415 /* Migrate the slack space for the truncate to our reserve */
8416 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8419 * We have reserved 2 metadata units when we started the transaction and
8420 * min_size matches 1 unit, so this should never fail, but if it does,
8421 * it's not critical we just fail truncation.
8424 btrfs_end_transaction(trans);
8428 trans->block_rsv = rsv;
8431 struct extent_state *cached_state = NULL;
8432 const u64 new_size = inode->vfs_inode.i_size;
8433 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8435 control.new_size = new_size;
8436 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8438 * We want to drop from the next block forward in case this new
8439 * size is not block aligned since we will be keeping the last
8440 * block of the extent just the way it is.
8442 btrfs_drop_extent_map_range(inode,
8443 ALIGN(new_size, fs_info->sectorsize),
8446 ret = btrfs_truncate_inode_items(trans, root, &control);
8448 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8449 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8451 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8453 trans->block_rsv = &fs_info->trans_block_rsv;
8454 if (ret != -ENOSPC && ret != -EAGAIN)
8457 ret = btrfs_update_inode(trans, root, inode);
8461 btrfs_end_transaction(trans);
8462 btrfs_btree_balance_dirty(fs_info);
8464 trans = btrfs_start_transaction(root, 2);
8465 if (IS_ERR(trans)) {
8466 ret = PTR_ERR(trans);
8471 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8472 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8473 rsv, min_size, false);
8475 * We have reserved 2 metadata units when we started the
8476 * transaction and min_size matches 1 unit, so this should never
8477 * fail, but if it does, it's not critical we just fail truncation.
8482 trans->block_rsv = rsv;
8486 * We can't call btrfs_truncate_block inside a trans handle as we could
8487 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8488 * know we've truncated everything except the last little bit, and can
8489 * do btrfs_truncate_block and then update the disk_i_size.
8491 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8492 btrfs_end_transaction(trans);
8493 btrfs_btree_balance_dirty(fs_info);
8495 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8498 trans = btrfs_start_transaction(root, 1);
8499 if (IS_ERR(trans)) {
8500 ret = PTR_ERR(trans);
8503 btrfs_inode_safe_disk_i_size_write(inode, 0);
8509 trans->block_rsv = &fs_info->trans_block_rsv;
8510 ret2 = btrfs_update_inode(trans, root, inode);
8514 ret2 = btrfs_end_transaction(trans);
8517 btrfs_btree_balance_dirty(fs_info);
8520 btrfs_free_block_rsv(fs_info, rsv);
8522 * So if we truncate and then write and fsync we normally would just
8523 * write the extents that changed, which is a problem if we need to
8524 * first truncate that entire inode. So set this flag so we write out
8525 * all of the extents in the inode to the sync log so we're completely
8528 * If no extents were dropped or trimmed we don't need to force the next
8529 * fsync to truncate all the inode's items from the log and re-log them
8530 * all. This means the truncate operation did not change the file size,
8531 * or changed it to a smaller size but there was only an implicit hole
8532 * between the old i_size and the new i_size, and there were no prealloc
8533 * extents beyond i_size to drop.
8535 if (control.extents_found > 0)
8536 btrfs_set_inode_full_sync(inode);
8541 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8544 struct inode *inode;
8546 inode = new_inode(dir->i_sb);
8549 * Subvolumes don't inherit the sgid bit or the parent's gid if
8550 * the parent's sgid bit is set. This is probably a bug.
8552 inode_init_owner(idmap, inode, NULL,
8553 S_IFDIR | (~current_umask() & S_IRWXUGO));
8554 inode->i_op = &btrfs_dir_inode_operations;
8555 inode->i_fop = &btrfs_dir_file_operations;
8560 struct inode *btrfs_alloc_inode(struct super_block *sb)
8562 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8563 struct btrfs_inode *ei;
8564 struct inode *inode;
8566 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8573 ei->last_sub_trans = 0;
8574 ei->logged_trans = 0;
8575 ei->delalloc_bytes = 0;
8576 ei->new_delalloc_bytes = 0;
8577 ei->defrag_bytes = 0;
8578 ei->disk_i_size = 0;
8582 ei->index_cnt = (u64)-1;
8584 ei->last_unlink_trans = 0;
8585 ei->last_reflink_trans = 0;
8586 ei->last_log_commit = 0;
8588 spin_lock_init(&ei->lock);
8589 ei->outstanding_extents = 0;
8590 if (sb->s_magic != BTRFS_TEST_MAGIC)
8591 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8592 BTRFS_BLOCK_RSV_DELALLOC);
8593 ei->runtime_flags = 0;
8594 ei->prop_compress = BTRFS_COMPRESS_NONE;
8595 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8597 ei->delayed_node = NULL;
8599 ei->i_otime.tv_sec = 0;
8600 ei->i_otime.tv_nsec = 0;
8602 inode = &ei->vfs_inode;
8603 extent_map_tree_init(&ei->extent_tree);
8604 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8605 ei->io_tree.inode = ei;
8606 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8607 IO_TREE_INODE_FILE_EXTENT);
8608 mutex_init(&ei->log_mutex);
8609 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8610 INIT_LIST_HEAD(&ei->delalloc_inodes);
8611 INIT_LIST_HEAD(&ei->delayed_iput);
8612 RB_CLEAR_NODE(&ei->rb_node);
8613 init_rwsem(&ei->i_mmap_lock);
8618 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8619 void btrfs_test_destroy_inode(struct inode *inode)
8621 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8622 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8626 void btrfs_free_inode(struct inode *inode)
8628 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8631 void btrfs_destroy_inode(struct inode *vfs_inode)
8633 struct btrfs_ordered_extent *ordered;
8634 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8635 struct btrfs_root *root = inode->root;
8636 bool freespace_inode;
8638 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8639 WARN_ON(vfs_inode->i_data.nrpages);
8640 WARN_ON(inode->block_rsv.reserved);
8641 WARN_ON(inode->block_rsv.size);
8642 WARN_ON(inode->outstanding_extents);
8643 if (!S_ISDIR(vfs_inode->i_mode)) {
8644 WARN_ON(inode->delalloc_bytes);
8645 WARN_ON(inode->new_delalloc_bytes);
8647 WARN_ON(inode->csum_bytes);
8648 WARN_ON(inode->defrag_bytes);
8651 * This can happen where we create an inode, but somebody else also
8652 * created the same inode and we need to destroy the one we already
8659 * If this is a free space inode do not take the ordered extents lockdep
8662 freespace_inode = btrfs_is_free_space_inode(inode);
8665 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8669 btrfs_err(root->fs_info,
8670 "found ordered extent %llu %llu on inode cleanup",
8671 ordered->file_offset, ordered->num_bytes);
8673 if (!freespace_inode)
8674 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8676 btrfs_remove_ordered_extent(inode, ordered);
8677 btrfs_put_ordered_extent(ordered);
8678 btrfs_put_ordered_extent(ordered);
8681 btrfs_qgroup_check_reserved_leak(inode);
8682 inode_tree_del(inode);
8683 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8684 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8685 btrfs_put_root(inode->root);
8688 int btrfs_drop_inode(struct inode *inode)
8690 struct btrfs_root *root = BTRFS_I(inode)->root;
8695 /* the snap/subvol tree is on deleting */
8696 if (btrfs_root_refs(&root->root_item) == 0)
8699 return generic_drop_inode(inode);
8702 static void init_once(void *foo)
8704 struct btrfs_inode *ei = foo;
8706 inode_init_once(&ei->vfs_inode);
8709 void __cold btrfs_destroy_cachep(void)
8712 * Make sure all delayed rcu free inodes are flushed before we
8716 bioset_exit(&btrfs_dio_bioset);
8717 kmem_cache_destroy(btrfs_inode_cachep);
8720 int __init btrfs_init_cachep(void)
8722 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8723 sizeof(struct btrfs_inode), 0,
8724 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8726 if (!btrfs_inode_cachep)
8729 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8730 offsetof(struct btrfs_dio_private, bbio.bio),
8736 btrfs_destroy_cachep();
8740 static int btrfs_getattr(struct mnt_idmap *idmap,
8741 const struct path *path, struct kstat *stat,
8742 u32 request_mask, unsigned int flags)
8746 struct inode *inode = d_inode(path->dentry);
8747 u32 blocksize = inode->i_sb->s_blocksize;
8748 u32 bi_flags = BTRFS_I(inode)->flags;
8749 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8751 stat->result_mask |= STATX_BTIME;
8752 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8753 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8754 if (bi_flags & BTRFS_INODE_APPEND)
8755 stat->attributes |= STATX_ATTR_APPEND;
8756 if (bi_flags & BTRFS_INODE_COMPRESS)
8757 stat->attributes |= STATX_ATTR_COMPRESSED;
8758 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8759 stat->attributes |= STATX_ATTR_IMMUTABLE;
8760 if (bi_flags & BTRFS_INODE_NODUMP)
8761 stat->attributes |= STATX_ATTR_NODUMP;
8762 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8763 stat->attributes |= STATX_ATTR_VERITY;
8765 stat->attributes_mask |= (STATX_ATTR_APPEND |
8766 STATX_ATTR_COMPRESSED |
8767 STATX_ATTR_IMMUTABLE |
8770 generic_fillattr(idmap, inode, stat);
8771 stat->dev = BTRFS_I(inode)->root->anon_dev;
8773 spin_lock(&BTRFS_I(inode)->lock);
8774 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8775 inode_bytes = inode_get_bytes(inode);
8776 spin_unlock(&BTRFS_I(inode)->lock);
8777 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8778 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8782 static int btrfs_rename_exchange(struct inode *old_dir,
8783 struct dentry *old_dentry,
8784 struct inode *new_dir,
8785 struct dentry *new_dentry)
8787 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8788 struct btrfs_trans_handle *trans;
8789 unsigned int trans_num_items;
8790 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8791 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8792 struct inode *new_inode = new_dentry->d_inode;
8793 struct inode *old_inode = old_dentry->d_inode;
8794 struct timespec64 ctime = current_time(old_inode);
8795 struct btrfs_rename_ctx old_rename_ctx;
8796 struct btrfs_rename_ctx new_rename_ctx;
8797 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8798 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8803 bool need_abort = false;
8804 struct fscrypt_name old_fname, new_fname;
8805 struct fscrypt_str *old_name, *new_name;
8808 * For non-subvolumes allow exchange only within one subvolume, in the
8809 * same inode namespace. Two subvolumes (represented as directory) can
8810 * be exchanged as they're a logical link and have a fixed inode number.
8813 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8814 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8817 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8821 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8823 fscrypt_free_filename(&old_fname);
8827 old_name = &old_fname.disk_name;
8828 new_name = &new_fname.disk_name;
8830 /* close the race window with snapshot create/destroy ioctl */
8831 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8832 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8833 down_read(&fs_info->subvol_sem);
8837 * 1 to remove old dir item
8838 * 1 to remove old dir index
8839 * 1 to add new dir item
8840 * 1 to add new dir index
8841 * 1 to update parent inode
8843 * If the parents are the same, we only need to account for one
8845 trans_num_items = (old_dir == new_dir ? 9 : 10);
8846 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8848 * 1 to remove old root ref
8849 * 1 to remove old root backref
8850 * 1 to add new root ref
8851 * 1 to add new root backref
8853 trans_num_items += 4;
8856 * 1 to update inode item
8857 * 1 to remove old inode ref
8858 * 1 to add new inode ref
8860 trans_num_items += 3;
8862 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8863 trans_num_items += 4;
8865 trans_num_items += 3;
8866 trans = btrfs_start_transaction(root, trans_num_items);
8867 if (IS_ERR(trans)) {
8868 ret = PTR_ERR(trans);
8873 ret = btrfs_record_root_in_trans(trans, dest);
8879 * We need to find a free sequence number both in the source and
8880 * in the destination directory for the exchange.
8882 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8885 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8889 BTRFS_I(old_inode)->dir_index = 0ULL;
8890 BTRFS_I(new_inode)->dir_index = 0ULL;
8892 /* Reference for the source. */
8893 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8894 /* force full log commit if subvolume involved. */
8895 btrfs_set_log_full_commit(trans);
8897 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8898 btrfs_ino(BTRFS_I(new_dir)),
8905 /* And now for the dest. */
8906 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8907 /* force full log commit if subvolume involved. */
8908 btrfs_set_log_full_commit(trans);
8910 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8911 btrfs_ino(BTRFS_I(old_dir)),
8915 btrfs_abort_transaction(trans, ret);
8920 /* Update inode version and ctime/mtime. */
8921 inode_inc_iversion(old_dir);
8922 inode_inc_iversion(new_dir);
8923 inode_inc_iversion(old_inode);
8924 inode_inc_iversion(new_inode);
8925 old_dir->i_mtime = ctime;
8926 old_dir->i_ctime = ctime;
8927 new_dir->i_mtime = ctime;
8928 new_dir->i_ctime = ctime;
8929 old_inode->i_ctime = ctime;
8930 new_inode->i_ctime = ctime;
8932 if (old_dentry->d_parent != new_dentry->d_parent) {
8933 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8934 BTRFS_I(old_inode), true);
8935 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8936 BTRFS_I(new_inode), true);
8939 /* src is a subvolume */
8940 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8941 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8942 } else { /* src is an inode */
8943 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8944 BTRFS_I(old_dentry->d_inode),
8945 old_name, &old_rename_ctx);
8947 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8950 btrfs_abort_transaction(trans, ret);
8954 /* dest is a subvolume */
8955 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8956 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8957 } else { /* dest is an inode */
8958 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8959 BTRFS_I(new_dentry->d_inode),
8960 new_name, &new_rename_ctx);
8962 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
8965 btrfs_abort_transaction(trans, ret);
8969 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8970 new_name, 0, old_idx);
8972 btrfs_abort_transaction(trans, ret);
8976 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8977 old_name, 0, new_idx);
8979 btrfs_abort_transaction(trans, ret);
8983 if (old_inode->i_nlink == 1)
8984 BTRFS_I(old_inode)->dir_index = old_idx;
8985 if (new_inode->i_nlink == 1)
8986 BTRFS_I(new_inode)->dir_index = new_idx;
8989 * Now pin the logs of the roots. We do it to ensure that no other task
8990 * can sync the logs while we are in progress with the rename, because
8991 * that could result in an inconsistency in case any of the inodes that
8992 * are part of this rename operation were logged before.
8994 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8995 btrfs_pin_log_trans(root);
8996 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8997 btrfs_pin_log_trans(dest);
8999 /* Do the log updates for all inodes. */
9000 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9001 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9002 old_rename_ctx.index, new_dentry->d_parent);
9003 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9004 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9005 new_rename_ctx.index, old_dentry->d_parent);
9007 /* Now unpin the logs. */
9008 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9009 btrfs_end_log_trans(root);
9010 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9011 btrfs_end_log_trans(dest);
9013 ret2 = btrfs_end_transaction(trans);
9014 ret = ret ? ret : ret2;
9016 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9017 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9018 up_read(&fs_info->subvol_sem);
9020 fscrypt_free_filename(&new_fname);
9021 fscrypt_free_filename(&old_fname);
9025 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
9028 struct inode *inode;
9030 inode = new_inode(dir->i_sb);
9032 inode_init_owner(idmap, inode, dir,
9033 S_IFCHR | WHITEOUT_MODE);
9034 inode->i_op = &btrfs_special_inode_operations;
9035 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9040 static int btrfs_rename(struct mnt_idmap *idmap,
9041 struct inode *old_dir, struct dentry *old_dentry,
9042 struct inode *new_dir, struct dentry *new_dentry,
9045 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9046 struct btrfs_new_inode_args whiteout_args = {
9048 .dentry = old_dentry,
9050 struct btrfs_trans_handle *trans;
9051 unsigned int trans_num_items;
9052 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9053 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9054 struct inode *new_inode = d_inode(new_dentry);
9055 struct inode *old_inode = d_inode(old_dentry);
9056 struct btrfs_rename_ctx rename_ctx;
9060 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9061 struct fscrypt_name old_fname, new_fname;
9063 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9066 /* we only allow rename subvolume link between subvolumes */
9067 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9070 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9071 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9074 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9075 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9078 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9082 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9084 fscrypt_free_filename(&old_fname);
9088 /* check for collisions, even if the name isn't there */
9089 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9091 if (ret == -EEXIST) {
9093 * eexist without a new_inode */
9094 if (WARN_ON(!new_inode)) {
9095 goto out_fscrypt_names;
9098 /* maybe -EOVERFLOW */
9099 goto out_fscrypt_names;
9105 * we're using rename to replace one file with another. Start IO on it
9106 * now so we don't add too much work to the end of the transaction
9108 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9109 filemap_flush(old_inode->i_mapping);
9111 if (flags & RENAME_WHITEOUT) {
9112 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9113 if (!whiteout_args.inode) {
9115 goto out_fscrypt_names;
9117 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9119 goto out_whiteout_inode;
9121 /* 1 to update the old parent inode. */
9122 trans_num_items = 1;
9125 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9126 /* Close the race window with snapshot create/destroy ioctl */
9127 down_read(&fs_info->subvol_sem);
9129 * 1 to remove old root ref
9130 * 1 to remove old root backref
9131 * 1 to add new root ref
9132 * 1 to add new root backref
9134 trans_num_items += 4;
9138 * 1 to remove old inode ref
9139 * 1 to add new inode ref
9141 trans_num_items += 3;
9144 * 1 to remove old dir item
9145 * 1 to remove old dir index
9146 * 1 to add new dir item
9147 * 1 to add new dir index
9149 trans_num_items += 4;
9150 /* 1 to update new parent inode if it's not the same as the old parent */
9151 if (new_dir != old_dir)
9156 * 1 to remove inode ref
9157 * 1 to remove dir item
9158 * 1 to remove dir index
9159 * 1 to possibly add orphan item
9161 trans_num_items += 5;
9163 trans = btrfs_start_transaction(root, trans_num_items);
9164 if (IS_ERR(trans)) {
9165 ret = PTR_ERR(trans);
9170 ret = btrfs_record_root_in_trans(trans, dest);
9175 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9179 BTRFS_I(old_inode)->dir_index = 0ULL;
9180 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9181 /* force full log commit if subvolume involved. */
9182 btrfs_set_log_full_commit(trans);
9184 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9185 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9191 inode_inc_iversion(old_dir);
9192 inode_inc_iversion(new_dir);
9193 inode_inc_iversion(old_inode);
9194 old_dir->i_mtime = current_time(old_dir);
9195 old_dir->i_ctime = old_dir->i_mtime;
9196 new_dir->i_mtime = old_dir->i_mtime;
9197 new_dir->i_ctime = old_dir->i_mtime;
9198 old_inode->i_ctime = old_dir->i_mtime;
9200 if (old_dentry->d_parent != new_dentry->d_parent)
9201 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9202 BTRFS_I(old_inode), true);
9204 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9205 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9207 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9208 BTRFS_I(d_inode(old_dentry)),
9209 &old_fname.disk_name, &rename_ctx);
9211 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9214 btrfs_abort_transaction(trans, ret);
9219 inode_inc_iversion(new_inode);
9220 new_inode->i_ctime = current_time(new_inode);
9221 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9222 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9223 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9224 BUG_ON(new_inode->i_nlink == 0);
9226 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9227 BTRFS_I(d_inode(new_dentry)),
9228 &new_fname.disk_name);
9230 if (!ret && new_inode->i_nlink == 0)
9231 ret = btrfs_orphan_add(trans,
9232 BTRFS_I(d_inode(new_dentry)));
9234 btrfs_abort_transaction(trans, ret);
9239 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9240 &new_fname.disk_name, 0, index);
9242 btrfs_abort_transaction(trans, ret);
9246 if (old_inode->i_nlink == 1)
9247 BTRFS_I(old_inode)->dir_index = index;
9249 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9250 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9251 rename_ctx.index, new_dentry->d_parent);
9253 if (flags & RENAME_WHITEOUT) {
9254 ret = btrfs_create_new_inode(trans, &whiteout_args);
9256 btrfs_abort_transaction(trans, ret);
9259 unlock_new_inode(whiteout_args.inode);
9260 iput(whiteout_args.inode);
9261 whiteout_args.inode = NULL;
9265 ret2 = btrfs_end_transaction(trans);
9266 ret = ret ? ret : ret2;
9268 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9269 up_read(&fs_info->subvol_sem);
9270 if (flags & RENAME_WHITEOUT)
9271 btrfs_new_inode_args_destroy(&whiteout_args);
9273 if (flags & RENAME_WHITEOUT)
9274 iput(whiteout_args.inode);
9276 fscrypt_free_filename(&old_fname);
9277 fscrypt_free_filename(&new_fname);
9281 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9282 struct dentry *old_dentry, struct inode *new_dir,
9283 struct dentry *new_dentry, unsigned int flags)
9287 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9290 if (flags & RENAME_EXCHANGE)
9291 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9294 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9297 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9302 struct btrfs_delalloc_work {
9303 struct inode *inode;
9304 struct completion completion;
9305 struct list_head list;
9306 struct btrfs_work work;
9309 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9311 struct btrfs_delalloc_work *delalloc_work;
9312 struct inode *inode;
9314 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9316 inode = delalloc_work->inode;
9317 filemap_flush(inode->i_mapping);
9318 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9319 &BTRFS_I(inode)->runtime_flags))
9320 filemap_flush(inode->i_mapping);
9323 complete(&delalloc_work->completion);
9326 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9328 struct btrfs_delalloc_work *work;
9330 work = kmalloc(sizeof(*work), GFP_NOFS);
9334 init_completion(&work->completion);
9335 INIT_LIST_HEAD(&work->list);
9336 work->inode = inode;
9337 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9343 * some fairly slow code that needs optimization. This walks the list
9344 * of all the inodes with pending delalloc and forces them to disk.
9346 static int start_delalloc_inodes(struct btrfs_root *root,
9347 struct writeback_control *wbc, bool snapshot,
9348 bool in_reclaim_context)
9350 struct btrfs_inode *binode;
9351 struct inode *inode;
9352 struct btrfs_delalloc_work *work, *next;
9353 struct list_head works;
9354 struct list_head splice;
9356 bool full_flush = wbc->nr_to_write == LONG_MAX;
9358 INIT_LIST_HEAD(&works);
9359 INIT_LIST_HEAD(&splice);
9361 mutex_lock(&root->delalloc_mutex);
9362 spin_lock(&root->delalloc_lock);
9363 list_splice_init(&root->delalloc_inodes, &splice);
9364 while (!list_empty(&splice)) {
9365 binode = list_entry(splice.next, struct btrfs_inode,
9368 list_move_tail(&binode->delalloc_inodes,
9369 &root->delalloc_inodes);
9371 if (in_reclaim_context &&
9372 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9375 inode = igrab(&binode->vfs_inode);
9377 cond_resched_lock(&root->delalloc_lock);
9380 spin_unlock(&root->delalloc_lock);
9383 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9384 &binode->runtime_flags);
9386 work = btrfs_alloc_delalloc_work(inode);
9392 list_add_tail(&work->list, &works);
9393 btrfs_queue_work(root->fs_info->flush_workers,
9396 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9397 btrfs_add_delayed_iput(BTRFS_I(inode));
9398 if (ret || wbc->nr_to_write <= 0)
9402 spin_lock(&root->delalloc_lock);
9404 spin_unlock(&root->delalloc_lock);
9407 list_for_each_entry_safe(work, next, &works, list) {
9408 list_del_init(&work->list);
9409 wait_for_completion(&work->completion);
9413 if (!list_empty(&splice)) {
9414 spin_lock(&root->delalloc_lock);
9415 list_splice_tail(&splice, &root->delalloc_inodes);
9416 spin_unlock(&root->delalloc_lock);
9418 mutex_unlock(&root->delalloc_mutex);
9422 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9424 struct writeback_control wbc = {
9425 .nr_to_write = LONG_MAX,
9426 .sync_mode = WB_SYNC_NONE,
9428 .range_end = LLONG_MAX,
9430 struct btrfs_fs_info *fs_info = root->fs_info;
9432 if (BTRFS_FS_ERROR(fs_info))
9435 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9438 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9439 bool in_reclaim_context)
9441 struct writeback_control wbc = {
9443 .sync_mode = WB_SYNC_NONE,
9445 .range_end = LLONG_MAX,
9447 struct btrfs_root *root;
9448 struct list_head splice;
9451 if (BTRFS_FS_ERROR(fs_info))
9454 INIT_LIST_HEAD(&splice);
9456 mutex_lock(&fs_info->delalloc_root_mutex);
9457 spin_lock(&fs_info->delalloc_root_lock);
9458 list_splice_init(&fs_info->delalloc_roots, &splice);
9459 while (!list_empty(&splice)) {
9461 * Reset nr_to_write here so we know that we're doing a full
9465 wbc.nr_to_write = LONG_MAX;
9467 root = list_first_entry(&splice, struct btrfs_root,
9469 root = btrfs_grab_root(root);
9471 list_move_tail(&root->delalloc_root,
9472 &fs_info->delalloc_roots);
9473 spin_unlock(&fs_info->delalloc_root_lock);
9475 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9476 btrfs_put_root(root);
9477 if (ret < 0 || wbc.nr_to_write <= 0)
9479 spin_lock(&fs_info->delalloc_root_lock);
9481 spin_unlock(&fs_info->delalloc_root_lock);
9485 if (!list_empty(&splice)) {
9486 spin_lock(&fs_info->delalloc_root_lock);
9487 list_splice_tail(&splice, &fs_info->delalloc_roots);
9488 spin_unlock(&fs_info->delalloc_root_lock);
9490 mutex_unlock(&fs_info->delalloc_root_mutex);
9494 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9495 struct dentry *dentry, const char *symname)
9497 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9498 struct btrfs_trans_handle *trans;
9499 struct btrfs_root *root = BTRFS_I(dir)->root;
9500 struct btrfs_path *path;
9501 struct btrfs_key key;
9502 struct inode *inode;
9503 struct btrfs_new_inode_args new_inode_args = {
9507 unsigned int trans_num_items;
9512 struct btrfs_file_extent_item *ei;
9513 struct extent_buffer *leaf;
9515 name_len = strlen(symname);
9516 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9517 return -ENAMETOOLONG;
9519 inode = new_inode(dir->i_sb);
9522 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9523 inode->i_op = &btrfs_symlink_inode_operations;
9524 inode_nohighmem(inode);
9525 inode->i_mapping->a_ops = &btrfs_aops;
9526 btrfs_i_size_write(BTRFS_I(inode), name_len);
9527 inode_set_bytes(inode, name_len);
9529 new_inode_args.inode = inode;
9530 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9533 /* 1 additional item for the inline extent */
9536 trans = btrfs_start_transaction(root, trans_num_items);
9537 if (IS_ERR(trans)) {
9538 err = PTR_ERR(trans);
9539 goto out_new_inode_args;
9542 err = btrfs_create_new_inode(trans, &new_inode_args);
9546 path = btrfs_alloc_path();
9549 btrfs_abort_transaction(trans, err);
9550 discard_new_inode(inode);
9554 key.objectid = btrfs_ino(BTRFS_I(inode));
9556 key.type = BTRFS_EXTENT_DATA_KEY;
9557 datasize = btrfs_file_extent_calc_inline_size(name_len);
9558 err = btrfs_insert_empty_item(trans, root, path, &key,
9561 btrfs_abort_transaction(trans, err);
9562 btrfs_free_path(path);
9563 discard_new_inode(inode);
9567 leaf = path->nodes[0];
9568 ei = btrfs_item_ptr(leaf, path->slots[0],
9569 struct btrfs_file_extent_item);
9570 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9571 btrfs_set_file_extent_type(leaf, ei,
9572 BTRFS_FILE_EXTENT_INLINE);
9573 btrfs_set_file_extent_encryption(leaf, ei, 0);
9574 btrfs_set_file_extent_compression(leaf, ei, 0);
9575 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9576 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9578 ptr = btrfs_file_extent_inline_start(ei);
9579 write_extent_buffer(leaf, symname, ptr, name_len);
9580 btrfs_mark_buffer_dirty(leaf);
9581 btrfs_free_path(path);
9583 d_instantiate_new(dentry, inode);
9586 btrfs_end_transaction(trans);
9587 btrfs_btree_balance_dirty(fs_info);
9589 btrfs_new_inode_args_destroy(&new_inode_args);
9596 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9597 struct btrfs_trans_handle *trans_in,
9598 struct btrfs_inode *inode,
9599 struct btrfs_key *ins,
9602 struct btrfs_file_extent_item stack_fi;
9603 struct btrfs_replace_extent_info extent_info;
9604 struct btrfs_trans_handle *trans = trans_in;
9605 struct btrfs_path *path;
9606 u64 start = ins->objectid;
9607 u64 len = ins->offset;
9608 int qgroup_released;
9611 memset(&stack_fi, 0, sizeof(stack_fi));
9613 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9614 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9615 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9616 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9617 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9618 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9619 /* Encryption and other encoding is reserved and all 0 */
9621 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9622 if (qgroup_released < 0)
9623 return ERR_PTR(qgroup_released);
9626 ret = insert_reserved_file_extent(trans, inode,
9627 file_offset, &stack_fi,
9628 true, qgroup_released);
9634 extent_info.disk_offset = start;
9635 extent_info.disk_len = len;
9636 extent_info.data_offset = 0;
9637 extent_info.data_len = len;
9638 extent_info.file_offset = file_offset;
9639 extent_info.extent_buf = (char *)&stack_fi;
9640 extent_info.is_new_extent = true;
9641 extent_info.update_times = true;
9642 extent_info.qgroup_reserved = qgroup_released;
9643 extent_info.insertions = 0;
9645 path = btrfs_alloc_path();
9651 ret = btrfs_replace_file_extents(inode, path, file_offset,
9652 file_offset + len - 1, &extent_info,
9654 btrfs_free_path(path);
9661 * We have released qgroup data range at the beginning of the function,
9662 * and normally qgroup_released bytes will be freed when committing
9664 * But if we error out early, we have to free what we have released
9665 * or we leak qgroup data reservation.
9667 btrfs_qgroup_free_refroot(inode->root->fs_info,
9668 inode->root->root_key.objectid, qgroup_released,
9669 BTRFS_QGROUP_RSV_DATA);
9670 return ERR_PTR(ret);
9673 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9674 u64 start, u64 num_bytes, u64 min_size,
9675 loff_t actual_len, u64 *alloc_hint,
9676 struct btrfs_trans_handle *trans)
9678 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9679 struct extent_map *em;
9680 struct btrfs_root *root = BTRFS_I(inode)->root;
9681 struct btrfs_key ins;
9682 u64 cur_offset = start;
9683 u64 clear_offset = start;
9686 u64 last_alloc = (u64)-1;
9688 bool own_trans = true;
9689 u64 end = start + num_bytes - 1;
9693 while (num_bytes > 0) {
9694 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9695 cur_bytes = max(cur_bytes, min_size);
9697 * If we are severely fragmented we could end up with really
9698 * small allocations, so if the allocator is returning small
9699 * chunks lets make its job easier by only searching for those
9702 cur_bytes = min(cur_bytes, last_alloc);
9703 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9704 min_size, 0, *alloc_hint, &ins, 1, 0);
9709 * We've reserved this space, and thus converted it from
9710 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9711 * from here on out we will only need to clear our reservation
9712 * for the remaining unreserved area, so advance our
9713 * clear_offset by our extent size.
9715 clear_offset += ins.offset;
9717 last_alloc = ins.offset;
9718 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9721 * Now that we inserted the prealloc extent we can finally
9722 * decrement the number of reservations in the block group.
9723 * If we did it before, we could race with relocation and have
9724 * relocation miss the reserved extent, making it fail later.
9726 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9727 if (IS_ERR(trans)) {
9728 ret = PTR_ERR(trans);
9729 btrfs_free_reserved_extent(fs_info, ins.objectid,
9734 em = alloc_extent_map();
9736 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9737 cur_offset + ins.offset - 1, false);
9738 btrfs_set_inode_full_sync(BTRFS_I(inode));
9742 em->start = cur_offset;
9743 em->orig_start = cur_offset;
9744 em->len = ins.offset;
9745 em->block_start = ins.objectid;
9746 em->block_len = ins.offset;
9747 em->orig_block_len = ins.offset;
9748 em->ram_bytes = ins.offset;
9749 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9750 em->generation = trans->transid;
9752 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9753 free_extent_map(em);
9755 num_bytes -= ins.offset;
9756 cur_offset += ins.offset;
9757 *alloc_hint = ins.objectid + ins.offset;
9759 inode_inc_iversion(inode);
9760 inode->i_ctime = current_time(inode);
9761 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9762 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9763 (actual_len > inode->i_size) &&
9764 (cur_offset > inode->i_size)) {
9765 if (cur_offset > actual_len)
9766 i_size = actual_len;
9768 i_size = cur_offset;
9769 i_size_write(inode, i_size);
9770 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9773 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9776 btrfs_abort_transaction(trans, ret);
9778 btrfs_end_transaction(trans);
9783 btrfs_end_transaction(trans);
9787 if (clear_offset < end)
9788 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9789 end - clear_offset + 1);
9793 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9794 u64 start, u64 num_bytes, u64 min_size,
9795 loff_t actual_len, u64 *alloc_hint)
9797 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9798 min_size, actual_len, alloc_hint,
9802 int btrfs_prealloc_file_range_trans(struct inode *inode,
9803 struct btrfs_trans_handle *trans, int mode,
9804 u64 start, u64 num_bytes, u64 min_size,
9805 loff_t actual_len, u64 *alloc_hint)
9807 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9808 min_size, actual_len, alloc_hint, trans);
9811 static int btrfs_permission(struct mnt_idmap *idmap,
9812 struct inode *inode, int mask)
9814 struct btrfs_root *root = BTRFS_I(inode)->root;
9815 umode_t mode = inode->i_mode;
9817 if (mask & MAY_WRITE &&
9818 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9819 if (btrfs_root_readonly(root))
9821 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9824 return generic_permission(idmap, inode, mask);
9827 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9828 struct file *file, umode_t mode)
9830 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9831 struct btrfs_trans_handle *trans;
9832 struct btrfs_root *root = BTRFS_I(dir)->root;
9833 struct inode *inode;
9834 struct btrfs_new_inode_args new_inode_args = {
9836 .dentry = file->f_path.dentry,
9839 unsigned int trans_num_items;
9842 inode = new_inode(dir->i_sb);
9845 inode_init_owner(idmap, inode, dir, mode);
9846 inode->i_fop = &btrfs_file_operations;
9847 inode->i_op = &btrfs_file_inode_operations;
9848 inode->i_mapping->a_ops = &btrfs_aops;
9850 new_inode_args.inode = inode;
9851 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9855 trans = btrfs_start_transaction(root, trans_num_items);
9856 if (IS_ERR(trans)) {
9857 ret = PTR_ERR(trans);
9858 goto out_new_inode_args;
9861 ret = btrfs_create_new_inode(trans, &new_inode_args);
9864 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9865 * set it to 1 because d_tmpfile() will issue a warning if the count is
9868 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9870 set_nlink(inode, 1);
9873 d_tmpfile(file, inode);
9874 unlock_new_inode(inode);
9875 mark_inode_dirty(inode);
9878 btrfs_end_transaction(trans);
9879 btrfs_btree_balance_dirty(fs_info);
9881 btrfs_new_inode_args_destroy(&new_inode_args);
9885 return finish_open_simple(file, ret);
9888 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9890 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9891 unsigned long index = start >> PAGE_SHIFT;
9892 unsigned long end_index = end >> PAGE_SHIFT;
9896 ASSERT(end + 1 - start <= U32_MAX);
9897 len = end + 1 - start;
9898 while (index <= end_index) {
9899 page = find_get_page(inode->vfs_inode.i_mapping, index);
9900 ASSERT(page); /* Pages should be in the extent_io_tree */
9902 btrfs_page_set_writeback(fs_info, page, start, len);
9908 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9911 switch (compress_type) {
9912 case BTRFS_COMPRESS_NONE:
9913 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9914 case BTRFS_COMPRESS_ZLIB:
9915 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9916 case BTRFS_COMPRESS_LZO:
9918 * The LZO format depends on the sector size. 64K is the maximum
9919 * sector size that we support.
9921 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9923 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9924 (fs_info->sectorsize_bits - 12);
9925 case BTRFS_COMPRESS_ZSTD:
9926 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9932 static ssize_t btrfs_encoded_read_inline(
9934 struct iov_iter *iter, u64 start,
9936 struct extent_state **cached_state,
9937 u64 extent_start, size_t count,
9938 struct btrfs_ioctl_encoded_io_args *encoded,
9941 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9942 struct btrfs_root *root = inode->root;
9943 struct btrfs_fs_info *fs_info = root->fs_info;
9944 struct extent_io_tree *io_tree = &inode->io_tree;
9945 struct btrfs_path *path;
9946 struct extent_buffer *leaf;
9947 struct btrfs_file_extent_item *item;
9953 path = btrfs_alloc_path();
9958 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9962 /* The extent item disappeared? */
9967 leaf = path->nodes[0];
9968 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9970 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9971 ptr = btrfs_file_extent_inline_start(item);
9973 encoded->len = min_t(u64, extent_start + ram_bytes,
9974 inode->vfs_inode.i_size) - iocb->ki_pos;
9975 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9976 btrfs_file_extent_compression(leaf, item));
9979 encoded->compression = ret;
9980 if (encoded->compression) {
9983 inline_size = btrfs_file_extent_inline_item_len(leaf,
9985 if (inline_size > count) {
9989 count = inline_size;
9990 encoded->unencoded_len = ram_bytes;
9991 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9993 count = min_t(u64, count, encoded->len);
9994 encoded->len = count;
9995 encoded->unencoded_len = count;
9996 ptr += iocb->ki_pos - extent_start;
9999 tmp = kmalloc(count, GFP_NOFS);
10004 read_extent_buffer(leaf, tmp, ptr, count);
10005 btrfs_release_path(path);
10006 unlock_extent(io_tree, start, lockend, cached_state);
10007 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10010 ret = copy_to_iter(tmp, count, iter);
10015 btrfs_free_path(path);
10019 struct btrfs_encoded_read_private {
10020 wait_queue_head_t wait;
10022 blk_status_t status;
10025 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
10027 struct btrfs_encoded_read_private *priv = bbio->private;
10029 if (bbio->bio.bi_status) {
10031 * The memory barrier implied by the atomic_dec_return() here
10032 * pairs with the memory barrier implied by the
10033 * atomic_dec_return() or io_wait_event() in
10034 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10035 * write is observed before the load of status in
10036 * btrfs_encoded_read_regular_fill_pages().
10038 WRITE_ONCE(priv->status, bbio->bio.bi_status);
10040 if (!atomic_dec_return(&priv->pending))
10041 wake_up(&priv->wait);
10042 bio_put(&bbio->bio);
10045 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10046 u64 file_offset, u64 disk_bytenr,
10047 u64 disk_io_size, struct page **pages)
10049 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10050 struct btrfs_encoded_read_private priv = {
10051 .pending = ATOMIC_INIT(1),
10053 unsigned long i = 0;
10054 struct btrfs_bio *bbio;
10056 init_waitqueue_head(&priv.wait);
10058 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10059 btrfs_encoded_read_endio, &priv);
10060 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10061 bbio->inode = inode;
10064 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
10066 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
10067 atomic_inc(&priv.pending);
10068 btrfs_submit_bio(bbio, 0);
10070 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10071 btrfs_encoded_read_endio, &priv);
10072 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10073 bbio->inode = inode;
10078 disk_bytenr += bytes;
10079 disk_io_size -= bytes;
10080 } while (disk_io_size);
10082 atomic_inc(&priv.pending);
10083 btrfs_submit_bio(bbio, 0);
10085 if (atomic_dec_return(&priv.pending))
10086 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10087 /* See btrfs_encoded_read_endio() for ordering. */
10088 return blk_status_to_errno(READ_ONCE(priv.status));
10091 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10092 struct iov_iter *iter,
10093 u64 start, u64 lockend,
10094 struct extent_state **cached_state,
10095 u64 disk_bytenr, u64 disk_io_size,
10096 size_t count, bool compressed,
10099 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10100 struct extent_io_tree *io_tree = &inode->io_tree;
10101 struct page **pages;
10102 unsigned long nr_pages, i;
10104 size_t page_offset;
10107 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10108 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10111 ret = btrfs_alloc_page_array(nr_pages, pages);
10117 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10118 disk_io_size, pages);
10122 unlock_extent(io_tree, start, lockend, cached_state);
10123 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10130 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10131 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10134 while (cur < count) {
10135 size_t bytes = min_t(size_t, count - cur,
10136 PAGE_SIZE - page_offset);
10138 if (copy_page_to_iter(pages[i], page_offset, bytes,
10149 for (i = 0; i < nr_pages; i++) {
10151 __free_page(pages[i]);
10157 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10158 struct btrfs_ioctl_encoded_io_args *encoded)
10160 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10161 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10162 struct extent_io_tree *io_tree = &inode->io_tree;
10164 size_t count = iov_iter_count(iter);
10165 u64 start, lockend, disk_bytenr, disk_io_size;
10166 struct extent_state *cached_state = NULL;
10167 struct extent_map *em;
10168 bool unlocked = false;
10170 file_accessed(iocb->ki_filp);
10172 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10174 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10175 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10178 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10180 * We don't know how long the extent containing iocb->ki_pos is, but if
10181 * it's compressed we know that it won't be longer than this.
10183 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10186 struct btrfs_ordered_extent *ordered;
10188 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10189 lockend - start + 1);
10191 goto out_unlock_inode;
10192 lock_extent(io_tree, start, lockend, &cached_state);
10193 ordered = btrfs_lookup_ordered_range(inode, start,
10194 lockend - start + 1);
10197 btrfs_put_ordered_extent(ordered);
10198 unlock_extent(io_tree, start, lockend, &cached_state);
10202 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10205 goto out_unlock_extent;
10208 if (em->block_start == EXTENT_MAP_INLINE) {
10209 u64 extent_start = em->start;
10212 * For inline extents we get everything we need out of the
10215 free_extent_map(em);
10217 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10218 &cached_state, extent_start,
10219 count, encoded, &unlocked);
10224 * We only want to return up to EOF even if the extent extends beyond
10227 encoded->len = min_t(u64, extent_map_end(em),
10228 inode->vfs_inode.i_size) - iocb->ki_pos;
10229 if (em->block_start == EXTENT_MAP_HOLE ||
10230 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10231 disk_bytenr = EXTENT_MAP_HOLE;
10232 count = min_t(u64, count, encoded->len);
10233 encoded->len = count;
10234 encoded->unencoded_len = count;
10235 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10236 disk_bytenr = em->block_start;
10238 * Bail if the buffer isn't large enough to return the whole
10239 * compressed extent.
10241 if (em->block_len > count) {
10245 disk_io_size = em->block_len;
10246 count = em->block_len;
10247 encoded->unencoded_len = em->ram_bytes;
10248 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10249 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10250 em->compress_type);
10253 encoded->compression = ret;
10255 disk_bytenr = em->block_start + (start - em->start);
10256 if (encoded->len > count)
10257 encoded->len = count;
10259 * Don't read beyond what we locked. This also limits the page
10260 * allocations that we'll do.
10262 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10263 count = start + disk_io_size - iocb->ki_pos;
10264 encoded->len = count;
10265 encoded->unencoded_len = count;
10266 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10268 free_extent_map(em);
10271 if (disk_bytenr == EXTENT_MAP_HOLE) {
10272 unlock_extent(io_tree, start, lockend, &cached_state);
10273 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10275 ret = iov_iter_zero(count, iter);
10279 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10280 &cached_state, disk_bytenr,
10281 disk_io_size, count,
10282 encoded->compression,
10288 iocb->ki_pos += encoded->len;
10290 free_extent_map(em);
10293 unlock_extent(io_tree, start, lockend, &cached_state);
10296 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10300 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10301 const struct btrfs_ioctl_encoded_io_args *encoded)
10303 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10304 struct btrfs_root *root = inode->root;
10305 struct btrfs_fs_info *fs_info = root->fs_info;
10306 struct extent_io_tree *io_tree = &inode->io_tree;
10307 struct extent_changeset *data_reserved = NULL;
10308 struct extent_state *cached_state = NULL;
10309 struct btrfs_ordered_extent *ordered;
10313 u64 num_bytes, ram_bytes, disk_num_bytes;
10314 unsigned long nr_pages, i;
10315 struct page **pages;
10316 struct btrfs_key ins;
10317 bool extent_reserved = false;
10318 struct extent_map *em;
10321 switch (encoded->compression) {
10322 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10323 compression = BTRFS_COMPRESS_ZLIB;
10325 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10326 compression = BTRFS_COMPRESS_ZSTD;
10328 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10329 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10330 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10331 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10332 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10333 /* The sector size must match for LZO. */
10334 if (encoded->compression -
10335 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10336 fs_info->sectorsize_bits)
10338 compression = BTRFS_COMPRESS_LZO;
10343 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10346 orig_count = iov_iter_count(from);
10348 /* The extent size must be sane. */
10349 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10350 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10354 * The compressed data must be smaller than the decompressed data.
10356 * It's of course possible for data to compress to larger or the same
10357 * size, but the buffered I/O path falls back to no compression for such
10358 * data, and we don't want to break any assumptions by creating these
10361 * Note that this is less strict than the current check we have that the
10362 * compressed data must be at least one sector smaller than the
10363 * decompressed data. We only want to enforce the weaker requirement
10364 * from old kernels that it is at least one byte smaller.
10366 if (orig_count >= encoded->unencoded_len)
10369 /* The extent must start on a sector boundary. */
10370 start = iocb->ki_pos;
10371 if (!IS_ALIGNED(start, fs_info->sectorsize))
10375 * The extent must end on a sector boundary. However, we allow a write
10376 * which ends at or extends i_size to have an unaligned length; we round
10377 * up the extent size and set i_size to the unaligned end.
10379 if (start + encoded->len < inode->vfs_inode.i_size &&
10380 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10383 /* Finally, the offset in the unencoded data must be sector-aligned. */
10384 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10387 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10388 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10389 end = start + num_bytes - 1;
10392 * If the extent cannot be inline, the compressed data on disk must be
10393 * sector-aligned. For convenience, we extend it with zeroes if it
10396 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10397 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10398 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10401 for (i = 0; i < nr_pages; i++) {
10402 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10405 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10410 kaddr = kmap_local_page(pages[i]);
10411 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10412 kunmap_local(kaddr);
10416 if (bytes < PAGE_SIZE)
10417 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10418 kunmap_local(kaddr);
10422 struct btrfs_ordered_extent *ordered;
10424 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10427 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10428 start >> PAGE_SHIFT,
10429 end >> PAGE_SHIFT);
10432 lock_extent(io_tree, start, end, &cached_state);
10433 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10435 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10438 btrfs_put_ordered_extent(ordered);
10439 unlock_extent(io_tree, start, end, &cached_state);
10444 * We don't use the higher-level delalloc space functions because our
10445 * num_bytes and disk_num_bytes are different.
10447 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10450 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10452 goto out_free_data_space;
10453 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10456 goto out_qgroup_free_data;
10458 /* Try an inline extent first. */
10459 if (start == 0 && encoded->unencoded_len == encoded->len &&
10460 encoded->unencoded_offset == 0) {
10461 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10462 compression, pages, true);
10466 goto out_delalloc_release;
10470 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10471 disk_num_bytes, 0, 0, &ins, 1, 1);
10473 goto out_delalloc_release;
10474 extent_reserved = true;
10476 em = create_io_em(inode, start, num_bytes,
10477 start - encoded->unencoded_offset, ins.objectid,
10478 ins.offset, ins.offset, ram_bytes, compression,
10479 BTRFS_ORDERED_COMPRESSED);
10482 goto out_free_reserved;
10484 free_extent_map(em);
10486 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10487 ins.objectid, ins.offset,
10488 encoded->unencoded_offset,
10489 (1 << BTRFS_ORDERED_ENCODED) |
10490 (1 << BTRFS_ORDERED_COMPRESSED),
10492 if (IS_ERR(ordered)) {
10493 btrfs_drop_extent_map_range(inode, start, end, false);
10494 ret = PTR_ERR(ordered);
10495 goto out_free_reserved;
10497 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10499 if (start + encoded->len > inode->vfs_inode.i_size)
10500 i_size_write(&inode->vfs_inode, start + encoded->len);
10502 unlock_extent(io_tree, start, end, &cached_state);
10504 btrfs_delalloc_release_extents(inode, num_bytes);
10506 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10511 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10512 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10513 out_delalloc_release:
10514 btrfs_delalloc_release_extents(inode, num_bytes);
10515 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10516 out_qgroup_free_data:
10518 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10519 out_free_data_space:
10521 * If btrfs_reserve_extent() succeeded, then we already decremented
10524 if (!extent_reserved)
10525 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10527 unlock_extent(io_tree, start, end, &cached_state);
10529 for (i = 0; i < nr_pages; i++) {
10531 __free_page(pages[i]);
10536 iocb->ki_pos += encoded->len;
10542 * Add an entry indicating a block group or device which is pinned by a
10543 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10544 * negative errno on failure.
10546 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10547 bool is_block_group)
10549 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10550 struct btrfs_swapfile_pin *sp, *entry;
10551 struct rb_node **p;
10552 struct rb_node *parent = NULL;
10554 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10559 sp->is_block_group = is_block_group;
10560 sp->bg_extent_count = 1;
10562 spin_lock(&fs_info->swapfile_pins_lock);
10563 p = &fs_info->swapfile_pins.rb_node;
10566 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10567 if (sp->ptr < entry->ptr ||
10568 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10569 p = &(*p)->rb_left;
10570 } else if (sp->ptr > entry->ptr ||
10571 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10572 p = &(*p)->rb_right;
10574 if (is_block_group)
10575 entry->bg_extent_count++;
10576 spin_unlock(&fs_info->swapfile_pins_lock);
10581 rb_link_node(&sp->node, parent, p);
10582 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10583 spin_unlock(&fs_info->swapfile_pins_lock);
10587 /* Free all of the entries pinned by this swapfile. */
10588 static void btrfs_free_swapfile_pins(struct inode *inode)
10590 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10591 struct btrfs_swapfile_pin *sp;
10592 struct rb_node *node, *next;
10594 spin_lock(&fs_info->swapfile_pins_lock);
10595 node = rb_first(&fs_info->swapfile_pins);
10597 next = rb_next(node);
10598 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10599 if (sp->inode == inode) {
10600 rb_erase(&sp->node, &fs_info->swapfile_pins);
10601 if (sp->is_block_group) {
10602 btrfs_dec_block_group_swap_extents(sp->ptr,
10603 sp->bg_extent_count);
10604 btrfs_put_block_group(sp->ptr);
10610 spin_unlock(&fs_info->swapfile_pins_lock);
10613 struct btrfs_swap_info {
10619 unsigned long nr_pages;
10623 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10624 struct btrfs_swap_info *bsi)
10626 unsigned long nr_pages;
10627 unsigned long max_pages;
10628 u64 first_ppage, first_ppage_reported, next_ppage;
10632 * Our swapfile may have had its size extended after the swap header was
10633 * written. In that case activating the swapfile should not go beyond
10634 * the max size set in the swap header.
10636 if (bsi->nr_pages >= sis->max)
10639 max_pages = sis->max - bsi->nr_pages;
10640 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10641 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10643 if (first_ppage >= next_ppage)
10645 nr_pages = next_ppage - first_ppage;
10646 nr_pages = min(nr_pages, max_pages);
10648 first_ppage_reported = first_ppage;
10649 if (bsi->start == 0)
10650 first_ppage_reported++;
10651 if (bsi->lowest_ppage > first_ppage_reported)
10652 bsi->lowest_ppage = first_ppage_reported;
10653 if (bsi->highest_ppage < (next_ppage - 1))
10654 bsi->highest_ppage = next_ppage - 1;
10656 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10659 bsi->nr_extents += ret;
10660 bsi->nr_pages += nr_pages;
10664 static void btrfs_swap_deactivate(struct file *file)
10666 struct inode *inode = file_inode(file);
10668 btrfs_free_swapfile_pins(inode);
10669 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10672 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10675 struct inode *inode = file_inode(file);
10676 struct btrfs_root *root = BTRFS_I(inode)->root;
10677 struct btrfs_fs_info *fs_info = root->fs_info;
10678 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10679 struct extent_state *cached_state = NULL;
10680 struct extent_map *em = NULL;
10681 struct btrfs_device *device = NULL;
10682 struct btrfs_swap_info bsi = {
10683 .lowest_ppage = (sector_t)-1ULL,
10690 * If the swap file was just created, make sure delalloc is done. If the
10691 * file changes again after this, the user is doing something stupid and
10692 * we don't really care.
10694 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10699 * The inode is locked, so these flags won't change after we check them.
10701 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10702 btrfs_warn(fs_info, "swapfile must not be compressed");
10705 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10706 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10709 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10710 btrfs_warn(fs_info, "swapfile must not be checksummed");
10715 * Balance or device remove/replace/resize can move stuff around from
10716 * under us. The exclop protection makes sure they aren't running/won't
10717 * run concurrently while we are mapping the swap extents, and
10718 * fs_info->swapfile_pins prevents them from running while the swap
10719 * file is active and moving the extents. Note that this also prevents
10720 * a concurrent device add which isn't actually necessary, but it's not
10721 * really worth the trouble to allow it.
10723 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10724 btrfs_warn(fs_info,
10725 "cannot activate swapfile while exclusive operation is running");
10730 * Prevent snapshot creation while we are activating the swap file.
10731 * We do not want to race with snapshot creation. If snapshot creation
10732 * already started before we bumped nr_swapfiles from 0 to 1 and
10733 * completes before the first write into the swap file after it is
10734 * activated, than that write would fallback to COW.
10736 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10737 btrfs_exclop_finish(fs_info);
10738 btrfs_warn(fs_info,
10739 "cannot activate swapfile because snapshot creation is in progress");
10743 * Snapshots can create extents which require COW even if NODATACOW is
10744 * set. We use this counter to prevent snapshots. We must increment it
10745 * before walking the extents because we don't want a concurrent
10746 * snapshot to run after we've already checked the extents.
10748 * It is possible that subvolume is marked for deletion but still not
10749 * removed yet. To prevent this race, we check the root status before
10750 * activating the swapfile.
10752 spin_lock(&root->root_item_lock);
10753 if (btrfs_root_dead(root)) {
10754 spin_unlock(&root->root_item_lock);
10756 btrfs_exclop_finish(fs_info);
10757 btrfs_warn(fs_info,
10758 "cannot activate swapfile because subvolume %llu is being deleted",
10759 root->root_key.objectid);
10762 atomic_inc(&root->nr_swapfiles);
10763 spin_unlock(&root->root_item_lock);
10765 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10767 lock_extent(io_tree, 0, isize - 1, &cached_state);
10769 while (start < isize) {
10770 u64 logical_block_start, physical_block_start;
10771 struct btrfs_block_group *bg;
10772 u64 len = isize - start;
10774 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10780 if (em->block_start == EXTENT_MAP_HOLE) {
10781 btrfs_warn(fs_info, "swapfile must not have holes");
10785 if (em->block_start == EXTENT_MAP_INLINE) {
10787 * It's unlikely we'll ever actually find ourselves
10788 * here, as a file small enough to fit inline won't be
10789 * big enough to store more than the swap header, but in
10790 * case something changes in the future, let's catch it
10791 * here rather than later.
10793 btrfs_warn(fs_info, "swapfile must not be inline");
10797 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10798 btrfs_warn(fs_info, "swapfile must not be compressed");
10803 logical_block_start = em->block_start + (start - em->start);
10804 len = min(len, em->len - (start - em->start));
10805 free_extent_map(em);
10808 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10814 btrfs_warn(fs_info,
10815 "swapfile must not be copy-on-write");
10820 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10826 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10827 btrfs_warn(fs_info,
10828 "swapfile must have single data profile");
10833 if (device == NULL) {
10834 device = em->map_lookup->stripes[0].dev;
10835 ret = btrfs_add_swapfile_pin(inode, device, false);
10840 } else if (device != em->map_lookup->stripes[0].dev) {
10841 btrfs_warn(fs_info, "swapfile must be on one device");
10846 physical_block_start = (em->map_lookup->stripes[0].physical +
10847 (logical_block_start - em->start));
10848 len = min(len, em->len - (logical_block_start - em->start));
10849 free_extent_map(em);
10852 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10854 btrfs_warn(fs_info,
10855 "could not find block group containing swapfile");
10860 if (!btrfs_inc_block_group_swap_extents(bg)) {
10861 btrfs_warn(fs_info,
10862 "block group for swapfile at %llu is read-only%s",
10864 atomic_read(&fs_info->scrubs_running) ?
10865 " (scrub running)" : "");
10866 btrfs_put_block_group(bg);
10871 ret = btrfs_add_swapfile_pin(inode, bg, true);
10873 btrfs_put_block_group(bg);
10880 if (bsi.block_len &&
10881 bsi.block_start + bsi.block_len == physical_block_start) {
10882 bsi.block_len += len;
10884 if (bsi.block_len) {
10885 ret = btrfs_add_swap_extent(sis, &bsi);
10890 bsi.block_start = physical_block_start;
10891 bsi.block_len = len;
10898 ret = btrfs_add_swap_extent(sis, &bsi);
10901 if (!IS_ERR_OR_NULL(em))
10902 free_extent_map(em);
10904 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10907 btrfs_swap_deactivate(file);
10909 btrfs_drew_write_unlock(&root->snapshot_lock);
10911 btrfs_exclop_finish(fs_info);
10917 sis->bdev = device->bdev;
10918 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10919 sis->max = bsi.nr_pages;
10920 sis->pages = bsi.nr_pages - 1;
10921 sis->highest_bit = bsi.nr_pages - 1;
10922 return bsi.nr_extents;
10925 static void btrfs_swap_deactivate(struct file *file)
10929 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10932 return -EOPNOTSUPP;
10937 * Update the number of bytes used in the VFS' inode. When we replace extents in
10938 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10939 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10940 * always get a correct value.
10942 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10943 const u64 add_bytes,
10944 const u64 del_bytes)
10946 if (add_bytes == del_bytes)
10949 spin_lock(&inode->lock);
10951 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10953 inode_add_bytes(&inode->vfs_inode, add_bytes);
10954 spin_unlock(&inode->lock);
10958 * Verify that there are no ordered extents for a given file range.
10960 * @inode: The target inode.
10961 * @start: Start offset of the file range, should be sector size aligned.
10962 * @end: End offset (inclusive) of the file range, its value +1 should be
10963 * sector size aligned.
10965 * This should typically be used for cases where we locked an inode's VFS lock in
10966 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10967 * we have flushed all delalloc in the range, we have waited for all ordered
10968 * extents in the range to complete and finally we have locked the file range in
10969 * the inode's io_tree.
10971 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10973 struct btrfs_root *root = inode->root;
10974 struct btrfs_ordered_extent *ordered;
10976 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10979 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10981 btrfs_err(root->fs_info,
10982 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10983 start, end, btrfs_ino(inode), root->root_key.objectid,
10984 ordered->file_offset,
10985 ordered->file_offset + ordered->num_bytes - 1);
10986 btrfs_put_ordered_extent(ordered);
10989 ASSERT(ordered == NULL);
10992 static const struct inode_operations btrfs_dir_inode_operations = {
10993 .getattr = btrfs_getattr,
10994 .lookup = btrfs_lookup,
10995 .create = btrfs_create,
10996 .unlink = btrfs_unlink,
10997 .link = btrfs_link,
10998 .mkdir = btrfs_mkdir,
10999 .rmdir = btrfs_rmdir,
11000 .rename = btrfs_rename2,
11001 .symlink = btrfs_symlink,
11002 .setattr = btrfs_setattr,
11003 .mknod = btrfs_mknod,
11004 .listxattr = btrfs_listxattr,
11005 .permission = btrfs_permission,
11006 .get_inode_acl = btrfs_get_acl,
11007 .set_acl = btrfs_set_acl,
11008 .update_time = btrfs_update_time,
11009 .tmpfile = btrfs_tmpfile,
11010 .fileattr_get = btrfs_fileattr_get,
11011 .fileattr_set = btrfs_fileattr_set,
11014 static const struct file_operations btrfs_dir_file_operations = {
11015 .llseek = generic_file_llseek,
11016 .read = generic_read_dir,
11017 .iterate_shared = btrfs_real_readdir,
11018 .open = btrfs_opendir,
11019 .unlocked_ioctl = btrfs_ioctl,
11020 #ifdef CONFIG_COMPAT
11021 .compat_ioctl = btrfs_compat_ioctl,
11023 .release = btrfs_release_file,
11024 .fsync = btrfs_sync_file,
11028 * btrfs doesn't support the bmap operation because swapfiles
11029 * use bmap to make a mapping of extents in the file. They assume
11030 * these extents won't change over the life of the file and they
11031 * use the bmap result to do IO directly to the drive.
11033 * the btrfs bmap call would return logical addresses that aren't
11034 * suitable for IO and they also will change frequently as COW
11035 * operations happen. So, swapfile + btrfs == corruption.
11037 * For now we're avoiding this by dropping bmap.
11039 static const struct address_space_operations btrfs_aops = {
11040 .read_folio = btrfs_read_folio,
11041 .writepages = btrfs_writepages,
11042 .readahead = btrfs_readahead,
11043 .invalidate_folio = btrfs_invalidate_folio,
11044 .release_folio = btrfs_release_folio,
11045 .migrate_folio = btrfs_migrate_folio,
11046 .dirty_folio = filemap_dirty_folio,
11047 .error_remove_page = generic_error_remove_page,
11048 .swap_activate = btrfs_swap_activate,
11049 .swap_deactivate = btrfs_swap_deactivate,
11052 static const struct inode_operations btrfs_file_inode_operations = {
11053 .getattr = btrfs_getattr,
11054 .setattr = btrfs_setattr,
11055 .listxattr = btrfs_listxattr,
11056 .permission = btrfs_permission,
11057 .fiemap = btrfs_fiemap,
11058 .get_inode_acl = btrfs_get_acl,
11059 .set_acl = btrfs_set_acl,
11060 .update_time = btrfs_update_time,
11061 .fileattr_get = btrfs_fileattr_get,
11062 .fileattr_set = btrfs_fileattr_set,
11064 static const struct inode_operations btrfs_special_inode_operations = {
11065 .getattr = btrfs_getattr,
11066 .setattr = btrfs_setattr,
11067 .permission = btrfs_permission,
11068 .listxattr = btrfs_listxattr,
11069 .get_inode_acl = btrfs_get_acl,
11070 .set_acl = btrfs_set_acl,
11071 .update_time = btrfs_update_time,
11073 static const struct inode_operations btrfs_symlink_inode_operations = {
11074 .get_link = page_get_link,
11075 .getattr = btrfs_getattr,
11076 .setattr = btrfs_setattr,
11077 .permission = btrfs_permission,
11078 .listxattr = btrfs_listxattr,
11079 .update_time = btrfs_update_time,
11082 const struct dentry_operations btrfs_dentry_operations = {
11083 .d_delete = btrfs_dentry_delete,