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 * Work queue call back to started compression on a file and pages.
819 * This is done inside an ordered work queue, and the compression is spread
820 * across many cpus. The actual IO submission is step two, and the ordered work
821 * queue takes care of making sure that happens in the same order things were
822 * put onto the queue by writepages and friends.
824 * If this code finds it can't get good compression, it puts an entry onto the
825 * work queue to write the uncompressed bytes. This makes sure that both
826 * compressed inodes and uncompressed inodes are written in the same order that
827 * the flusher thread sent them down.
829 static void compress_file_range(struct btrfs_work *work)
831 struct async_chunk *async_chunk =
832 container_of(work, struct async_chunk, work);
833 struct btrfs_inode *inode = async_chunk->inode;
834 struct btrfs_fs_info *fs_info = inode->root->fs_info;
835 struct address_space *mapping = inode->vfs_inode.i_mapping;
836 u64 blocksize = fs_info->sectorsize;
837 u64 start = async_chunk->start;
838 u64 end = async_chunk->end;
842 struct page **pages = NULL;
843 unsigned long nr_pages;
844 unsigned long total_compressed = 0;
845 unsigned long total_in = 0;
848 int compress_type = fs_info->compress_type;
851 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
854 * We need to save i_size before now because it could change in between
855 * us evaluating the size and assigning it. This is because we lock and
856 * unlock the page in truncate and fallocate, and then modify the i_size
859 * The barriers are to emulate READ_ONCE, remove that once i_size_read
863 i_size = i_size_read(&inode->vfs_inode);
865 actual_end = min_t(u64, i_size, end + 1);
868 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
869 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
872 * we don't want to send crud past the end of i_size through
873 * compression, that's just a waste of CPU time. So, if the
874 * end of the file is before the start of our current
875 * requested range of bytes, we bail out to the uncompressed
876 * cleanup code that can deal with all of this.
878 * It isn't really the fastest way to fix things, but this is a
879 * very uncommon corner.
881 if (actual_end <= start)
882 goto cleanup_and_bail_uncompressed;
884 total_compressed = actual_end - start;
887 * Skip compression for a small file range(<=blocksize) that
888 * isn't an inline extent, since it doesn't save disk space at all.
890 if (total_compressed <= blocksize &&
891 (start > 0 || end + 1 < inode->disk_i_size))
892 goto cleanup_and_bail_uncompressed;
895 * For subpage case, we require full page alignment for the sector
897 * Thus we must also check against @actual_end, not just @end.
899 if (blocksize < PAGE_SIZE) {
900 if (!PAGE_ALIGNED(start) ||
901 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
902 goto cleanup_and_bail_uncompressed;
905 total_compressed = min_t(unsigned long, total_compressed,
906 BTRFS_MAX_UNCOMPRESSED);
911 * we do compression for mount -o compress and when the
912 * inode has not been flagged as nocompress. This flag can
913 * change at any time if we discover bad compression ratios.
915 if (inode_need_compress(inode, start, end)) {
917 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
919 /* just bail out to the uncompressed code */
924 if (inode->defrag_compress)
925 compress_type = inode->defrag_compress;
926 else if (inode->prop_compress)
927 compress_type = inode->prop_compress;
930 * we need to call clear_page_dirty_for_io on each
931 * page in the range. Otherwise applications with the file
932 * mmap'd can wander in and change the page contents while
933 * we are compressing them.
935 * If the compression fails for any reason, we set the pages
936 * dirty again later on.
938 * Note that the remaining part is redirtied, the start pointer
939 * has moved, the end is the original one.
942 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
946 /* Compression level is applied here and only here */
947 ret = btrfs_compress_pages(
948 compress_type | (fs_info->compress_level << 4),
956 unsigned long offset = offset_in_page(total_compressed);
957 struct page *page = pages[nr_pages - 1];
959 /* zero the tail end of the last page, we might be
960 * sending it down to disk
963 memzero_page(page, offset, PAGE_SIZE - offset);
969 * Check cow_file_range() for why we don't even try to create inline
970 * extent for subpage case.
972 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
973 /* lets try to make an inline extent */
974 if (ret || total_in < actual_end) {
975 /* we didn't compress the entire range, try
976 * to make an uncompressed inline extent.
978 ret = cow_file_range_inline(inode, actual_end,
979 0, BTRFS_COMPRESS_NONE,
982 /* try making a compressed inline extent */
983 ret = cow_file_range_inline(inode, actual_end,
985 compress_type, pages,
989 unsigned long clear_flags = EXTENT_DELALLOC |
990 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
991 EXTENT_DO_ACCOUNTING;
994 mapping_set_error(mapping, -EIO);
997 * inline extent creation worked or returned error,
998 * we don't need to create any more async work items.
999 * Unlock and free up our temp pages.
1001 * We use DO_ACCOUNTING here because we need the
1002 * delalloc_release_metadata to be done _after_ we drop
1003 * our outstanding extent for clearing delalloc for this
1006 extent_clear_unlock_delalloc(inode, start, end,
1010 PAGE_START_WRITEBACK |
1011 PAGE_END_WRITEBACK);
1014 * Ensure we only free the compressed pages if we have
1015 * them allocated, as we can still reach here with
1016 * inode_need_compress() == false.
1019 for (i = 0; i < nr_pages; i++) {
1020 WARN_ON(pages[i]->mapping);
1029 if (will_compress) {
1031 * we aren't doing an inline extent round the compressed size
1032 * up to a block size boundary so the allocator does sane
1035 total_compressed = ALIGN(total_compressed, blocksize);
1038 * one last check to make sure the compression is really a
1039 * win, compare the page count read with the blocks on disk,
1040 * compression must free at least one sector size
1042 total_in = round_up(total_in, fs_info->sectorsize);
1043 if (total_compressed + blocksize <= total_in) {
1045 * The async work queues will take care of doing actual
1046 * allocation on disk for these compressed pages, and
1047 * will submit them to the elevator.
1049 add_async_extent(async_chunk, start, total_in,
1050 total_compressed, pages, nr_pages,
1053 if (start + total_in < end) {
1064 * the compression code ran but failed to make things smaller,
1065 * free any pages it allocated and our page pointer array
1067 for (i = 0; i < nr_pages; i++) {
1068 WARN_ON(pages[i]->mapping);
1073 total_compressed = 0;
1076 /* flag the file so we don't compress in the future */
1077 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
1078 !(inode->prop_compress)) {
1079 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1082 cleanup_and_bail_uncompressed:
1084 * No compression, but we still need to write the pages in the file
1085 * we've been given so far. redirty the locked page if it corresponds
1086 * to our extent and set things up for the async work queue to run
1087 * cow_file_range to do the normal delalloc dance.
1089 if (async_chunk->locked_page &&
1090 (page_offset(async_chunk->locked_page) >= start &&
1091 page_offset(async_chunk->locked_page)) <= end) {
1092 __set_page_dirty_nobuffers(async_chunk->locked_page);
1093 /* unlocked later on in the async handlers */
1097 extent_range_redirty_for_io(&inode->vfs_inode, start, end);
1098 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1099 BTRFS_COMPRESS_NONE);
1102 static void free_async_extent_pages(struct async_extent *async_extent)
1106 if (!async_extent->pages)
1109 for (i = 0; i < async_extent->nr_pages; i++) {
1110 WARN_ON(async_extent->pages[i]->mapping);
1111 put_page(async_extent->pages[i]);
1113 kfree(async_extent->pages);
1114 async_extent->nr_pages = 0;
1115 async_extent->pages = NULL;
1118 static void submit_uncompressed_range(struct btrfs_inode *inode,
1119 struct async_extent *async_extent,
1120 struct page *locked_page)
1122 u64 start = async_extent->start;
1123 u64 end = async_extent->start + async_extent->ram_size - 1;
1125 struct writeback_control wbc = {
1126 .sync_mode = WB_SYNC_ALL,
1127 .range_start = start,
1129 .no_cgroup_owner = 1,
1133 * Call cow_file_range() to run the delalloc range directly, since we
1134 * won't go to NOCOW or async path again.
1136 * Also we call cow_file_range() with @unlock_page == 0, so that we
1137 * can directly submit them without interruption.
1139 ret = cow_file_range(inode, locked_page, start, end, NULL, true, false);
1140 /* Inline extent inserted, page gets unlocked and everything is done */
1145 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1147 const u64 page_start = page_offset(locked_page);
1149 set_page_writeback(locked_page);
1150 end_page_writeback(locked_page);
1151 btrfs_mark_ordered_io_finished(inode, locked_page,
1152 page_start, PAGE_SIZE,
1154 btrfs_page_clear_uptodate(inode->root->fs_info,
1155 locked_page, page_start,
1157 mapping_set_error(locked_page->mapping, ret);
1158 unlock_page(locked_page);
1163 /* All pages will be unlocked, including @locked_page */
1164 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1165 extent_write_locked_range(&inode->vfs_inode, start, end, &wbc);
1166 wbc_detach_inode(&wbc);
1169 static void submit_one_async_extent(struct async_chunk *async_chunk,
1170 struct async_extent *async_extent,
1173 struct btrfs_inode *inode = async_chunk->inode;
1174 struct extent_io_tree *io_tree = &inode->io_tree;
1175 struct btrfs_root *root = inode->root;
1176 struct btrfs_fs_info *fs_info = root->fs_info;
1177 struct btrfs_ordered_extent *ordered;
1178 struct btrfs_key ins;
1179 struct page *locked_page = NULL;
1180 struct extent_map *em;
1182 u64 start = async_extent->start;
1183 u64 end = async_extent->start + async_extent->ram_size - 1;
1185 if (async_chunk->blkcg_css)
1186 kthread_associate_blkcg(async_chunk->blkcg_css);
1189 * If async_chunk->locked_page is in the async_extent range, we need to
1192 if (async_chunk->locked_page) {
1193 u64 locked_page_start = page_offset(async_chunk->locked_page);
1194 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1196 if (!(start >= locked_page_end || end <= locked_page_start))
1197 locked_page = async_chunk->locked_page;
1199 lock_extent(io_tree, start, end, NULL);
1201 if (async_extent->compress_type == BTRFS_COMPRESS_NONE) {
1202 submit_uncompressed_range(inode, async_extent, locked_page);
1206 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1207 async_extent->compressed_size,
1208 async_extent->compressed_size,
1209 0, *alloc_hint, &ins, 1, 1);
1212 * Here we used to try again by going back to non-compressed
1213 * path for ENOSPC. But we can't reserve space even for
1214 * compressed size, how could it work for uncompressed size
1215 * which requires larger size? So here we directly go error
1221 /* Here we're doing allocation and writeback of the compressed pages */
1222 em = create_io_em(inode, start,
1223 async_extent->ram_size, /* len */
1224 start, /* orig_start */
1225 ins.objectid, /* block_start */
1226 ins.offset, /* block_len */
1227 ins.offset, /* orig_block_len */
1228 async_extent->ram_size, /* ram_bytes */
1229 async_extent->compress_type,
1230 BTRFS_ORDERED_COMPRESSED);
1233 goto out_free_reserve;
1235 free_extent_map(em);
1237 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1238 async_extent->ram_size, /* num_bytes */
1239 async_extent->ram_size, /* ram_bytes */
1240 ins.objectid, /* disk_bytenr */
1241 ins.offset, /* disk_num_bytes */
1243 1 << BTRFS_ORDERED_COMPRESSED,
1244 async_extent->compress_type);
1245 if (IS_ERR(ordered)) {
1246 btrfs_drop_extent_map_range(inode, start, end, false);
1247 ret = PTR_ERR(ordered);
1248 goto out_free_reserve;
1250 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1252 /* Clear dirty, set writeback and unlock the pages. */
1253 extent_clear_unlock_delalloc(inode, start, end,
1254 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1255 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1256 btrfs_submit_compressed_write(ordered,
1257 async_extent->pages, /* compressed_pages */
1258 async_extent->nr_pages,
1259 async_chunk->write_flags, true);
1260 *alloc_hint = ins.objectid + ins.offset;
1262 if (async_chunk->blkcg_css)
1263 kthread_associate_blkcg(NULL);
1264 kfree(async_extent);
1268 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1269 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1271 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1272 extent_clear_unlock_delalloc(inode, start, end,
1273 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1274 EXTENT_DELALLOC_NEW |
1275 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1276 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1277 PAGE_END_WRITEBACK);
1278 free_async_extent_pages(async_extent);
1279 if (async_chunk->blkcg_css)
1280 kthread_associate_blkcg(NULL);
1281 btrfs_debug(fs_info,
1282 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1283 root->root_key.objectid, btrfs_ino(inode), start,
1284 async_extent->ram_size, ret);
1285 kfree(async_extent);
1288 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1291 struct extent_map_tree *em_tree = &inode->extent_tree;
1292 struct extent_map *em;
1295 read_lock(&em_tree->lock);
1296 em = search_extent_mapping(em_tree, start, num_bytes);
1299 * if block start isn't an actual block number then find the
1300 * first block in this inode and use that as a hint. If that
1301 * block is also bogus then just don't worry about it.
1303 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1304 free_extent_map(em);
1305 em = search_extent_mapping(em_tree, 0, 0);
1306 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1307 alloc_hint = em->block_start;
1309 free_extent_map(em);
1311 alloc_hint = em->block_start;
1312 free_extent_map(em);
1315 read_unlock(&em_tree->lock);
1321 * when extent_io.c finds a delayed allocation range in the file,
1322 * the call backs end up in this code. The basic idea is to
1323 * allocate extents on disk for the range, and create ordered data structs
1324 * in ram to track those extents.
1326 * locked_page is the page that writepage had locked already. We use
1327 * it to make sure we don't do extra locks or unlocks.
1329 * When this function fails, it unlocks all pages except @locked_page.
1331 * When this function successfully creates an inline extent, it returns 1 and
1332 * unlocks all pages including locked_page and starts I/O on them.
1333 * (In reality inline extents are limited to a single page, so locked_page is
1334 * the only page handled anyway).
1336 * When this function succeed and creates a normal extent, the page locking
1337 * status depends on the passed in flags:
1339 * - If @keep_locked is set, all pages are kept locked.
1340 * - Else all pages except for @locked_page are unlocked.
1342 * When a failure happens in the second or later iteration of the
1343 * while-loop, the ordered extents created in previous iterations are kept
1344 * intact. So, the caller must clean them up by calling
1345 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1348 static noinline int cow_file_range(struct btrfs_inode *inode,
1349 struct page *locked_page, u64 start, u64 end,
1351 bool keep_locked, bool no_inline)
1353 struct btrfs_root *root = inode->root;
1354 struct btrfs_fs_info *fs_info = root->fs_info;
1356 u64 orig_start = start;
1358 unsigned long ram_size;
1359 u64 cur_alloc_size = 0;
1361 u64 blocksize = fs_info->sectorsize;
1362 struct btrfs_key ins;
1363 struct extent_map *em;
1364 unsigned clear_bits;
1365 unsigned long page_ops;
1366 bool extent_reserved = false;
1369 if (btrfs_is_free_space_inode(inode)) {
1374 num_bytes = ALIGN(end - start + 1, blocksize);
1375 num_bytes = max(blocksize, num_bytes);
1376 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1378 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1381 * Due to the page size limit, for subpage we can only trigger the
1382 * writeback for the dirty sectors of page, that means data writeback
1383 * is doing more writeback than what we want.
1385 * This is especially unexpected for some call sites like fallocate,
1386 * where we only increase i_size after everything is done.
1387 * This means we can trigger inline extent even if we didn't want to.
1388 * So here we skip inline extent creation completely.
1390 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1391 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1394 /* lets try to make an inline extent */
1395 ret = cow_file_range_inline(inode, actual_end, 0,
1396 BTRFS_COMPRESS_NONE, NULL, false);
1399 * We use DO_ACCOUNTING here because we need the
1400 * delalloc_release_metadata to be run _after_ we drop
1401 * our outstanding extent for clearing delalloc for this
1404 extent_clear_unlock_delalloc(inode, start, end,
1406 EXTENT_LOCKED | EXTENT_DELALLOC |
1407 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1408 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1409 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1411 * locked_page is locked by the caller of
1412 * writepage_delalloc(), not locked by
1413 * __process_pages_contig().
1415 * We can't let __process_pages_contig() to unlock it,
1416 * as it doesn't have any subpage::writers recorded.
1418 * Here we manually unlock the page, since the caller
1419 * can't determine if it's an inline extent or a
1420 * compressed extent.
1422 unlock_page(locked_page);
1424 } else if (ret < 0) {
1429 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1432 * Relocation relies on the relocated extents to have exactly the same
1433 * size as the original extents. Normally writeback for relocation data
1434 * extents follows a NOCOW path because relocation preallocates the
1435 * extents. However, due to an operation such as scrub turning a block
1436 * group to RO mode, it may fallback to COW mode, so we must make sure
1437 * an extent allocated during COW has exactly the requested size and can
1438 * not be split into smaller extents, otherwise relocation breaks and
1439 * fails during the stage where it updates the bytenr of file extent
1442 if (btrfs_is_data_reloc_root(root))
1443 min_alloc_size = num_bytes;
1445 min_alloc_size = fs_info->sectorsize;
1447 while (num_bytes > 0) {
1448 struct btrfs_ordered_extent *ordered;
1450 cur_alloc_size = num_bytes;
1451 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1452 min_alloc_size, 0, alloc_hint,
1456 cur_alloc_size = ins.offset;
1457 extent_reserved = true;
1459 ram_size = ins.offset;
1460 em = create_io_em(inode, start, ins.offset, /* len */
1461 start, /* orig_start */
1462 ins.objectid, /* block_start */
1463 ins.offset, /* block_len */
1464 ins.offset, /* orig_block_len */
1465 ram_size, /* ram_bytes */
1466 BTRFS_COMPRESS_NONE, /* compress_type */
1467 BTRFS_ORDERED_REGULAR /* type */);
1472 free_extent_map(em);
1474 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1475 ram_size, ins.objectid, cur_alloc_size,
1476 0, 1 << BTRFS_ORDERED_REGULAR,
1477 BTRFS_COMPRESS_NONE);
1478 if (IS_ERR(ordered)) {
1479 ret = PTR_ERR(ordered);
1480 goto out_drop_extent_cache;
1483 if (btrfs_is_data_reloc_root(root)) {
1484 ret = btrfs_reloc_clone_csums(ordered);
1487 * Only drop cache here, and process as normal.
1489 * We must not allow extent_clear_unlock_delalloc()
1490 * at out_unlock label to free meta of this ordered
1491 * extent, as its meta should be freed by
1492 * btrfs_finish_ordered_io().
1494 * So we must continue until @start is increased to
1495 * skip current ordered extent.
1498 btrfs_drop_extent_map_range(inode, start,
1499 start + ram_size - 1,
1502 btrfs_put_ordered_extent(ordered);
1504 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1507 * We're not doing compressed IO, don't unlock the first page
1508 * (which the caller expects to stay locked), don't clear any
1509 * dirty bits and don't set any writeback bits
1511 * Do set the Ordered (Private2) bit so we know this page was
1512 * properly setup for writepage.
1514 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1515 page_ops |= PAGE_SET_ORDERED;
1517 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1519 EXTENT_LOCKED | EXTENT_DELALLOC,
1521 if (num_bytes < cur_alloc_size)
1524 num_bytes -= cur_alloc_size;
1525 alloc_hint = ins.objectid + ins.offset;
1526 start += cur_alloc_size;
1527 extent_reserved = false;
1530 * btrfs_reloc_clone_csums() error, since start is increased
1531 * extent_clear_unlock_delalloc() at out_unlock label won't
1532 * free metadata of current ordered extent, we're OK to exit.
1539 out_drop_extent_cache:
1540 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1542 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1543 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1546 * If done_offset is non-NULL and ret == -EAGAIN, we expect the
1547 * caller to write out the successfully allocated region and retry.
1549 if (done_offset && ret == -EAGAIN) {
1550 if (orig_start < start)
1551 *done_offset = start - 1;
1553 *done_offset = start;
1555 } else if (ret == -EAGAIN) {
1556 /* Convert to -ENOSPC since the caller cannot retry. */
1561 * Now, we have three regions to clean up:
1563 * |-------(1)----|---(2)---|-------------(3)----------|
1564 * `- orig_start `- start `- start + cur_alloc_size `- end
1566 * We process each region below.
1569 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1570 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1571 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1574 * For the range (1). We have already instantiated the ordered extents
1575 * for this region. They are cleaned up by
1576 * btrfs_cleanup_ordered_extents() in e.g,
1577 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1578 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1579 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1582 * However, in case of @keep_locked, we still need to unlock the pages
1583 * (except @locked_page) to ensure all the pages are unlocked.
1585 if (keep_locked && orig_start < start) {
1587 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1588 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1589 locked_page, 0, page_ops);
1593 * For the range (2). If we reserved an extent for our delalloc range
1594 * (or a subrange) and failed to create the respective ordered extent,
1595 * then it means that when we reserved the extent we decremented the
1596 * extent's size from the data space_info's bytes_may_use counter and
1597 * incremented the space_info's bytes_reserved counter by the same
1598 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1599 * to decrement again the data space_info's bytes_may_use counter,
1600 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1602 if (extent_reserved) {
1603 extent_clear_unlock_delalloc(inode, start,
1604 start + cur_alloc_size - 1,
1608 start += cur_alloc_size;
1612 * For the range (3). We never touched the region. In addition to the
1613 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1614 * space_info's bytes_may_use counter, reserved in
1615 * btrfs_check_data_free_space().
1618 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1619 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1620 clear_bits, page_ops);
1626 * Phase two of compressed writeback. This is the ordered portion of the code,
1627 * which only gets called in the order the work was queued. We walk all the
1628 * async extents created by compress_file_range and send them down to the disk.
1630 static noinline void submit_compressed_extents(struct btrfs_work *work)
1632 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1634 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1635 struct async_extent *async_extent;
1636 unsigned long nr_pages;
1639 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1642 while (!list_empty(&async_chunk->extents)) {
1643 async_extent = list_entry(async_chunk->extents.next,
1644 struct async_extent, list);
1645 list_del(&async_extent->list);
1646 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1649 /* atomic_sub_return implies a barrier */
1650 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1652 cond_wake_up_nomb(&fs_info->async_submit_wait);
1655 static noinline void async_cow_free(struct btrfs_work *work)
1657 struct async_chunk *async_chunk;
1658 struct async_cow *async_cow;
1660 async_chunk = container_of(work, struct async_chunk, work);
1661 btrfs_add_delayed_iput(async_chunk->inode);
1662 if (async_chunk->blkcg_css)
1663 css_put(async_chunk->blkcg_css);
1665 async_cow = async_chunk->async_cow;
1666 if (atomic_dec_and_test(&async_cow->num_chunks))
1670 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1671 struct page *locked_page, u64 start,
1672 u64 end, struct writeback_control *wbc)
1674 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1675 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1676 struct async_cow *ctx;
1677 struct async_chunk *async_chunk;
1678 unsigned long nr_pages;
1679 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1682 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1684 nofs_flag = memalloc_nofs_save();
1685 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1686 memalloc_nofs_restore(nofs_flag);
1690 unlock_extent(&inode->io_tree, start, end, NULL);
1691 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1693 async_chunk = ctx->chunks;
1694 atomic_set(&ctx->num_chunks, num_chunks);
1696 for (i = 0; i < num_chunks; i++) {
1697 u64 cur_end = min(end, start + SZ_512K - 1);
1700 * igrab is called higher up in the call chain, take only the
1701 * lightweight reference for the callback lifetime
1703 ihold(&inode->vfs_inode);
1704 async_chunk[i].async_cow = ctx;
1705 async_chunk[i].inode = inode;
1706 async_chunk[i].start = start;
1707 async_chunk[i].end = cur_end;
1708 async_chunk[i].write_flags = write_flags;
1709 INIT_LIST_HEAD(&async_chunk[i].extents);
1712 * The locked_page comes all the way from writepage and its
1713 * the original page we were actually given. As we spread
1714 * this large delalloc region across multiple async_chunk
1715 * structs, only the first struct needs a pointer to locked_page
1717 * This way we don't need racey decisions about who is supposed
1722 * Depending on the compressibility, the pages might or
1723 * might not go through async. We want all of them to
1724 * be accounted against wbc once. Let's do it here
1725 * before the paths diverge. wbc accounting is used
1726 * only for foreign writeback detection and doesn't
1727 * need full accuracy. Just account the whole thing
1728 * against the first page.
1730 wbc_account_cgroup_owner(wbc, locked_page,
1732 async_chunk[i].locked_page = locked_page;
1735 async_chunk[i].locked_page = NULL;
1738 if (blkcg_css != blkcg_root_css) {
1740 async_chunk[i].blkcg_css = blkcg_css;
1741 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1743 async_chunk[i].blkcg_css = NULL;
1746 btrfs_init_work(&async_chunk[i].work, compress_file_range,
1747 submit_compressed_extents, async_cow_free);
1749 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1750 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1752 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1754 start = cur_end + 1;
1759 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1760 struct page *locked_page, u64 start,
1761 u64 end, struct writeback_control *wbc)
1763 u64 done_offset = end;
1765 bool locked_page_done = false;
1767 while (start <= end) {
1768 ret = cow_file_range(inode, locked_page, start, end, &done_offset,
1770 if (ret && ret != -EAGAIN)
1776 if (done_offset == start) {
1777 wait_on_bit_io(&inode->root->fs_info->flags,
1778 BTRFS_FS_NEED_ZONE_FINISH,
1779 TASK_UNINTERRUPTIBLE);
1783 if (!locked_page_done) {
1784 __set_page_dirty_nobuffers(locked_page);
1785 account_page_redirty(locked_page);
1787 locked_page_done = true;
1788 extent_write_locked_range(&inode->vfs_inode, start, done_offset,
1790 start = done_offset + 1;
1796 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1797 u64 bytenr, u64 num_bytes, bool nowait)
1799 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1800 struct btrfs_ordered_sum *sums;
1804 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1806 if (ret == 0 && list_empty(&list))
1809 while (!list_empty(&list)) {
1810 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1811 list_del(&sums->list);
1819 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1820 const u64 start, const u64 end)
1822 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1823 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1824 const u64 range_bytes = end + 1 - start;
1825 struct extent_io_tree *io_tree = &inode->io_tree;
1826 u64 range_start = start;
1831 * If EXTENT_NORESERVE is set it means that when the buffered write was
1832 * made we had not enough available data space and therefore we did not
1833 * reserve data space for it, since we though we could do NOCOW for the
1834 * respective file range (either there is prealloc extent or the inode
1835 * has the NOCOW bit set).
1837 * However when we need to fallback to COW mode (because for example the
1838 * block group for the corresponding extent was turned to RO mode by a
1839 * scrub or relocation) we need to do the following:
1841 * 1) We increment the bytes_may_use counter of the data space info.
1842 * If COW succeeds, it allocates a new data extent and after doing
1843 * that it decrements the space info's bytes_may_use counter and
1844 * increments its bytes_reserved counter by the same amount (we do
1845 * this at btrfs_add_reserved_bytes()). So we need to increment the
1846 * bytes_may_use counter to compensate (when space is reserved at
1847 * buffered write time, the bytes_may_use counter is incremented);
1849 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1850 * that if the COW path fails for any reason, it decrements (through
1851 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1852 * data space info, which we incremented in the step above.
1854 * If we need to fallback to cow and the inode corresponds to a free
1855 * space cache inode or an inode of the data relocation tree, we must
1856 * also increment bytes_may_use of the data space_info for the same
1857 * reason. Space caches and relocated data extents always get a prealloc
1858 * extent for them, however scrub or balance may have set the block
1859 * group that contains that extent to RO mode and therefore force COW
1860 * when starting writeback.
1862 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1863 EXTENT_NORESERVE, 0, NULL);
1864 if (count > 0 || is_space_ino || is_reloc_ino) {
1866 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1867 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1869 if (is_space_ino || is_reloc_ino)
1870 bytes = range_bytes;
1872 spin_lock(&sinfo->lock);
1873 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1874 spin_unlock(&sinfo->lock);
1877 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1882 * Don't try to create inline extents, as a mix of inline extent that
1883 * is written out and unlocked directly and a normal NOCOW extent
1886 ret = cow_file_range(inode, locked_page, start, end, NULL, false, true);
1891 struct can_nocow_file_extent_args {
1894 /* Start file offset of the range we want to NOCOW. */
1896 /* End file offset (inclusive) of the range we want to NOCOW. */
1898 bool writeback_path;
1901 * Free the path passed to can_nocow_file_extent() once it's not needed
1906 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1911 /* Number of bytes that can be written to in NOCOW mode. */
1916 * Check if we can NOCOW the file extent that the path points to.
1917 * This function may return with the path released, so the caller should check
1918 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1920 * Returns: < 0 on error
1921 * 0 if we can not NOCOW
1924 static int can_nocow_file_extent(struct btrfs_path *path,
1925 struct btrfs_key *key,
1926 struct btrfs_inode *inode,
1927 struct can_nocow_file_extent_args *args)
1929 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1930 struct extent_buffer *leaf = path->nodes[0];
1931 struct btrfs_root *root = inode->root;
1932 struct btrfs_file_extent_item *fi;
1937 bool nowait = path->nowait;
1939 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1940 extent_type = btrfs_file_extent_type(leaf, fi);
1942 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1945 /* Can't access these fields unless we know it's not an inline extent. */
1946 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1947 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1948 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1950 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1951 extent_type == BTRFS_FILE_EXTENT_REG)
1955 * If the extent was created before the generation where the last snapshot
1956 * for its subvolume was created, then this implies the extent is shared,
1957 * hence we must COW.
1959 if (!args->strict &&
1960 btrfs_file_extent_generation(leaf, fi) <=
1961 btrfs_root_last_snapshot(&root->root_item))
1964 /* An explicit hole, must COW. */
1965 if (args->disk_bytenr == 0)
1968 /* Compressed/encrypted/encoded extents must be COWed. */
1969 if (btrfs_file_extent_compression(leaf, fi) ||
1970 btrfs_file_extent_encryption(leaf, fi) ||
1971 btrfs_file_extent_other_encoding(leaf, fi))
1974 extent_end = btrfs_file_extent_end(path);
1977 * The following checks can be expensive, as they need to take other
1978 * locks and do btree or rbtree searches, so release the path to avoid
1979 * blocking other tasks for too long.
1981 btrfs_release_path(path);
1983 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1984 key->offset - args->extent_offset,
1985 args->disk_bytenr, args->strict, path);
1986 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1990 if (args->free_path) {
1992 * We don't need the path anymore, plus through the
1993 * csum_exist_in_range() call below we will end up allocating
1994 * another path. So free the path to avoid unnecessary extra
1997 btrfs_free_path(path);
2001 /* If there are pending snapshots for this root, we must COW. */
2002 if (args->writeback_path && !is_freespace_inode &&
2003 atomic_read(&root->snapshot_force_cow))
2006 args->disk_bytenr += args->extent_offset;
2007 args->disk_bytenr += args->start - key->offset;
2008 args->num_bytes = min(args->end + 1, extent_end) - args->start;
2011 * Force COW if csums exist in the range. This ensures that csums for a
2012 * given extent are either valid or do not exist.
2014 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
2016 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2022 if (args->free_path && path)
2023 btrfs_free_path(path);
2025 return ret < 0 ? ret : can_nocow;
2029 * when nowcow writeback call back. This checks for snapshots or COW copies
2030 * of the extents that exist in the file, and COWs the file as required.
2032 * If no cow copies or snapshots exist, we write directly to the existing
2035 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
2036 struct page *locked_page,
2037 const u64 start, const u64 end)
2039 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2040 struct btrfs_root *root = inode->root;
2041 struct btrfs_path *path;
2042 u64 cow_start = (u64)-1;
2043 u64 cur_offset = start;
2045 bool check_prev = true;
2046 u64 ino = btrfs_ino(inode);
2047 struct btrfs_block_group *bg;
2049 struct can_nocow_file_extent_args nocow_args = { 0 };
2051 path = btrfs_alloc_path();
2053 extent_clear_unlock_delalloc(inode, start, end, locked_page,
2054 EXTENT_LOCKED | EXTENT_DELALLOC |
2055 EXTENT_DO_ACCOUNTING |
2056 EXTENT_DEFRAG, PAGE_UNLOCK |
2057 PAGE_START_WRITEBACK |
2058 PAGE_END_WRITEBACK);
2062 nocow_args.end = end;
2063 nocow_args.writeback_path = true;
2066 struct btrfs_ordered_extent *ordered;
2067 struct btrfs_key found_key;
2068 struct btrfs_file_extent_item *fi;
2069 struct extent_buffer *leaf;
2078 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2084 * If there is no extent for our range when doing the initial
2085 * search, then go back to the previous slot as it will be the
2086 * one containing the search offset
2088 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2089 leaf = path->nodes[0];
2090 btrfs_item_key_to_cpu(leaf, &found_key,
2091 path->slots[0] - 1);
2092 if (found_key.objectid == ino &&
2093 found_key.type == BTRFS_EXTENT_DATA_KEY)
2098 /* Go to next leaf if we have exhausted the current one */
2099 leaf = path->nodes[0];
2100 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2101 ret = btrfs_next_leaf(root, path);
2103 if (cow_start != (u64)-1)
2104 cur_offset = cow_start;
2109 leaf = path->nodes[0];
2112 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2114 /* Didn't find anything for our INO */
2115 if (found_key.objectid > ino)
2118 * Keep searching until we find an EXTENT_ITEM or there are no
2119 * more extents for this inode
2121 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2122 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2127 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2128 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2129 found_key.offset > end)
2133 * If the found extent starts after requested offset, then
2134 * adjust extent_end to be right before this extent begins
2136 if (found_key.offset > cur_offset) {
2137 extent_end = found_key.offset;
2143 * Found extent which begins before our range and potentially
2146 fi = btrfs_item_ptr(leaf, path->slots[0],
2147 struct btrfs_file_extent_item);
2148 extent_type = btrfs_file_extent_type(leaf, fi);
2149 /* If this is triggered then we have a memory corruption. */
2150 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2151 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2155 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2156 extent_end = btrfs_file_extent_end(path);
2159 * If the extent we got ends before our current offset, skip to
2162 if (extent_end <= cur_offset) {
2167 nocow_args.start = cur_offset;
2168 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2170 if (cow_start != (u64)-1)
2171 cur_offset = cow_start;
2173 } else if (ret == 0) {
2178 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2183 * If nocow is false then record the beginning of the range
2184 * that needs to be COWed
2187 if (cow_start == (u64)-1)
2188 cow_start = cur_offset;
2189 cur_offset = extent_end;
2190 if (cur_offset > end)
2192 if (!path->nodes[0])
2199 * COW range from cow_start to found_key.offset - 1. As the key
2200 * will contain the beginning of the first extent that can be
2201 * NOCOW, following one which needs to be COW'ed
2203 if (cow_start != (u64)-1) {
2204 ret = fallback_to_cow(inode, locked_page,
2205 cow_start, found_key.offset - 1);
2208 cow_start = (u64)-1;
2211 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2212 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2214 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2215 struct extent_map *em;
2217 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2219 nocow_args.disk_bytenr, /* block_start */
2220 nocow_args.num_bytes, /* block_len */
2221 nocow_args.disk_num_bytes, /* orig_block_len */
2222 ram_bytes, BTRFS_COMPRESS_NONE,
2223 BTRFS_ORDERED_PREALLOC);
2228 free_extent_map(em);
2231 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2232 nocow_args.num_bytes, nocow_args.num_bytes,
2233 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2235 ? (1 << BTRFS_ORDERED_PREALLOC)
2236 : (1 << BTRFS_ORDERED_NOCOW),
2237 BTRFS_COMPRESS_NONE);
2238 if (IS_ERR(ordered)) {
2240 btrfs_drop_extent_map_range(inode, cur_offset,
2243 ret = PTR_ERR(ordered);
2248 btrfs_dec_nocow_writers(bg);
2252 if (btrfs_is_data_reloc_root(root))
2254 * Error handled later, as we must prevent
2255 * extent_clear_unlock_delalloc() in error handler
2256 * from freeing metadata of created ordered extent.
2258 ret = btrfs_reloc_clone_csums(ordered);
2259 btrfs_put_ordered_extent(ordered);
2261 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2262 locked_page, EXTENT_LOCKED |
2264 EXTENT_CLEAR_DATA_RESV,
2265 PAGE_UNLOCK | PAGE_SET_ORDERED);
2267 cur_offset = extent_end;
2270 * btrfs_reloc_clone_csums() error, now we're OK to call error
2271 * handler, as metadata for created ordered extent will only
2272 * be freed by btrfs_finish_ordered_io().
2276 if (cur_offset > end)
2279 btrfs_release_path(path);
2281 if (cur_offset <= end && cow_start == (u64)-1)
2282 cow_start = cur_offset;
2284 if (cow_start != (u64)-1) {
2286 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2293 btrfs_dec_nocow_writers(bg);
2295 if (ret && cur_offset < end)
2296 extent_clear_unlock_delalloc(inode, cur_offset, end,
2297 locked_page, EXTENT_LOCKED |
2298 EXTENT_DELALLOC | EXTENT_DEFRAG |
2299 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2300 PAGE_START_WRITEBACK |
2301 PAGE_END_WRITEBACK);
2302 btrfs_free_path(path);
2306 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2308 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2309 if (inode->defrag_bytes &&
2310 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2319 * Function to process delayed allocation (create CoW) for ranges which are
2320 * being touched for the first time.
2322 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2323 u64 start, u64 end, struct writeback_control *wbc)
2325 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2329 * The range must cover part of the @locked_page, or a return of 1
2330 * can confuse the caller.
2332 ASSERT(!(end <= page_offset(locked_page) ||
2333 start >= page_offset(locked_page) + PAGE_SIZE));
2335 if (should_nocow(inode, start, end)) {
2337 * Normally on a zoned device we're only doing COW writes, but
2338 * in case of relocation on a zoned filesystem we have taken
2339 * precaution, that we're only writing sequentially. It's safe
2340 * to use run_delalloc_nocow() here, like for regular
2341 * preallocated inodes.
2343 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2344 ret = run_delalloc_nocow(inode, locked_page, start, end);
2348 if (btrfs_inode_can_compress(inode) &&
2349 inode_need_compress(inode, start, end) &&
2350 run_delalloc_compressed(inode, locked_page, start, end, wbc))
2354 ret = run_delalloc_zoned(inode, locked_page, start, end, wbc);
2356 ret = cow_file_range(inode, locked_page, start, end, NULL,
2361 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2366 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2367 struct extent_state *orig, u64 split)
2369 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2372 /* not delalloc, ignore it */
2373 if (!(orig->state & EXTENT_DELALLOC))
2376 size = orig->end - orig->start + 1;
2377 if (size > fs_info->max_extent_size) {
2382 * See the explanation in btrfs_merge_delalloc_extent, the same
2383 * applies here, just in reverse.
2385 new_size = orig->end - split + 1;
2386 num_extents = count_max_extents(fs_info, new_size);
2387 new_size = split - orig->start;
2388 num_extents += count_max_extents(fs_info, new_size);
2389 if (count_max_extents(fs_info, size) >= num_extents)
2393 spin_lock(&inode->lock);
2394 btrfs_mod_outstanding_extents(inode, 1);
2395 spin_unlock(&inode->lock);
2399 * Handle merged delayed allocation extents so we can keep track of new extents
2400 * that are just merged onto old extents, such as when we are doing sequential
2401 * writes, so we can properly account for the metadata space we'll need.
2403 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2404 struct extent_state *other)
2406 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2407 u64 new_size, old_size;
2410 /* not delalloc, ignore it */
2411 if (!(other->state & EXTENT_DELALLOC))
2414 if (new->start > other->start)
2415 new_size = new->end - other->start + 1;
2417 new_size = other->end - new->start + 1;
2419 /* we're not bigger than the max, unreserve the space and go */
2420 if (new_size <= fs_info->max_extent_size) {
2421 spin_lock(&inode->lock);
2422 btrfs_mod_outstanding_extents(inode, -1);
2423 spin_unlock(&inode->lock);
2428 * We have to add up either side to figure out how many extents were
2429 * accounted for before we merged into one big extent. If the number of
2430 * extents we accounted for is <= the amount we need for the new range
2431 * then we can return, otherwise drop. Think of it like this
2435 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2436 * need 2 outstanding extents, on one side we have 1 and the other side
2437 * we have 1 so they are == and we can return. But in this case
2439 * [MAX_SIZE+4k][MAX_SIZE+4k]
2441 * Each range on their own accounts for 2 extents, but merged together
2442 * they are only 3 extents worth of accounting, so we need to drop in
2445 old_size = other->end - other->start + 1;
2446 num_extents = count_max_extents(fs_info, old_size);
2447 old_size = new->end - new->start + 1;
2448 num_extents += count_max_extents(fs_info, old_size);
2449 if (count_max_extents(fs_info, new_size) >= num_extents)
2452 spin_lock(&inode->lock);
2453 btrfs_mod_outstanding_extents(inode, -1);
2454 spin_unlock(&inode->lock);
2457 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2458 struct btrfs_inode *inode)
2460 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2462 spin_lock(&root->delalloc_lock);
2463 if (list_empty(&inode->delalloc_inodes)) {
2464 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2465 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2466 root->nr_delalloc_inodes++;
2467 if (root->nr_delalloc_inodes == 1) {
2468 spin_lock(&fs_info->delalloc_root_lock);
2469 BUG_ON(!list_empty(&root->delalloc_root));
2470 list_add_tail(&root->delalloc_root,
2471 &fs_info->delalloc_roots);
2472 spin_unlock(&fs_info->delalloc_root_lock);
2475 spin_unlock(&root->delalloc_lock);
2478 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2479 struct btrfs_inode *inode)
2481 struct btrfs_fs_info *fs_info = root->fs_info;
2483 if (!list_empty(&inode->delalloc_inodes)) {
2484 list_del_init(&inode->delalloc_inodes);
2485 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2486 &inode->runtime_flags);
2487 root->nr_delalloc_inodes--;
2488 if (!root->nr_delalloc_inodes) {
2489 ASSERT(list_empty(&root->delalloc_inodes));
2490 spin_lock(&fs_info->delalloc_root_lock);
2491 BUG_ON(list_empty(&root->delalloc_root));
2492 list_del_init(&root->delalloc_root);
2493 spin_unlock(&fs_info->delalloc_root_lock);
2498 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2499 struct btrfs_inode *inode)
2501 spin_lock(&root->delalloc_lock);
2502 __btrfs_del_delalloc_inode(root, inode);
2503 spin_unlock(&root->delalloc_lock);
2507 * Properly track delayed allocation bytes in the inode and to maintain the
2508 * list of inodes that have pending delalloc work to be done.
2510 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2513 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2515 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2518 * set_bit and clear bit hooks normally require _irqsave/restore
2519 * but in this case, we are only testing for the DELALLOC
2520 * bit, which is only set or cleared with irqs on
2522 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2523 struct btrfs_root *root = inode->root;
2524 u64 len = state->end + 1 - state->start;
2525 u32 num_extents = count_max_extents(fs_info, len);
2526 bool do_list = !btrfs_is_free_space_inode(inode);
2528 spin_lock(&inode->lock);
2529 btrfs_mod_outstanding_extents(inode, num_extents);
2530 spin_unlock(&inode->lock);
2532 /* For sanity tests */
2533 if (btrfs_is_testing(fs_info))
2536 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2537 fs_info->delalloc_batch);
2538 spin_lock(&inode->lock);
2539 inode->delalloc_bytes += len;
2540 if (bits & EXTENT_DEFRAG)
2541 inode->defrag_bytes += len;
2542 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2543 &inode->runtime_flags))
2544 btrfs_add_delalloc_inodes(root, inode);
2545 spin_unlock(&inode->lock);
2548 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2549 (bits & EXTENT_DELALLOC_NEW)) {
2550 spin_lock(&inode->lock);
2551 inode->new_delalloc_bytes += state->end + 1 - state->start;
2552 spin_unlock(&inode->lock);
2557 * Once a range is no longer delalloc this function ensures that proper
2558 * accounting happens.
2560 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2561 struct extent_state *state, u32 bits)
2563 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2564 u64 len = state->end + 1 - state->start;
2565 u32 num_extents = count_max_extents(fs_info, len);
2567 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2568 spin_lock(&inode->lock);
2569 inode->defrag_bytes -= len;
2570 spin_unlock(&inode->lock);
2574 * set_bit and clear bit hooks normally require _irqsave/restore
2575 * but in this case, we are only testing for the DELALLOC
2576 * bit, which is only set or cleared with irqs on
2578 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2579 struct btrfs_root *root = inode->root;
2580 bool do_list = !btrfs_is_free_space_inode(inode);
2582 spin_lock(&inode->lock);
2583 btrfs_mod_outstanding_extents(inode, -num_extents);
2584 spin_unlock(&inode->lock);
2587 * We don't reserve metadata space for space cache inodes so we
2588 * don't need to call delalloc_release_metadata if there is an
2591 if (bits & EXTENT_CLEAR_META_RESV &&
2592 root != fs_info->tree_root)
2593 btrfs_delalloc_release_metadata(inode, len, false);
2595 /* For sanity tests. */
2596 if (btrfs_is_testing(fs_info))
2599 if (!btrfs_is_data_reloc_root(root) &&
2600 do_list && !(state->state & EXTENT_NORESERVE) &&
2601 (bits & EXTENT_CLEAR_DATA_RESV))
2602 btrfs_free_reserved_data_space_noquota(fs_info, len);
2604 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2605 fs_info->delalloc_batch);
2606 spin_lock(&inode->lock);
2607 inode->delalloc_bytes -= len;
2608 if (do_list && inode->delalloc_bytes == 0 &&
2609 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2610 &inode->runtime_flags))
2611 btrfs_del_delalloc_inode(root, inode);
2612 spin_unlock(&inode->lock);
2615 if ((state->state & EXTENT_DELALLOC_NEW) &&
2616 (bits & EXTENT_DELALLOC_NEW)) {
2617 spin_lock(&inode->lock);
2618 ASSERT(inode->new_delalloc_bytes >= len);
2619 inode->new_delalloc_bytes -= len;
2620 if (bits & EXTENT_ADD_INODE_BYTES)
2621 inode_add_bytes(&inode->vfs_inode, len);
2622 spin_unlock(&inode->lock);
2626 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2627 struct btrfs_ordered_extent *ordered)
2629 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2630 u64 len = bbio->bio.bi_iter.bi_size;
2631 struct btrfs_ordered_extent *new;
2634 /* Must always be called for the beginning of an ordered extent. */
2635 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2638 /* No need to split if the ordered extent covers the entire bio. */
2639 if (ordered->disk_num_bytes == len) {
2640 refcount_inc(&ordered->refs);
2641 bbio->ordered = ordered;
2646 * Don't split the extent_map for NOCOW extents, as we're writing into
2647 * a pre-existing one.
2649 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2650 ret = split_extent_map(bbio->inode, bbio->file_offset,
2651 ordered->num_bytes, len,
2652 ordered->disk_bytenr);
2657 new = btrfs_split_ordered_extent(ordered, len);
2659 return PTR_ERR(new);
2660 bbio->ordered = new;
2665 * given a list of ordered sums record them in the inode. This happens
2666 * at IO completion time based on sums calculated at bio submission time.
2668 static int add_pending_csums(struct btrfs_trans_handle *trans,
2669 struct list_head *list)
2671 struct btrfs_ordered_sum *sum;
2672 struct btrfs_root *csum_root = NULL;
2675 list_for_each_entry(sum, list, list) {
2676 trans->adding_csums = true;
2678 csum_root = btrfs_csum_root(trans->fs_info,
2680 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2681 trans->adding_csums = false;
2688 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2691 struct extent_state **cached_state)
2693 u64 search_start = start;
2694 const u64 end = start + len - 1;
2696 while (search_start < end) {
2697 const u64 search_len = end - search_start + 1;
2698 struct extent_map *em;
2702 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2706 if (em->block_start != EXTENT_MAP_HOLE)
2710 if (em->start < search_start)
2711 em_len -= search_start - em->start;
2712 if (em_len > search_len)
2713 em_len = search_len;
2715 ret = set_extent_bit(&inode->io_tree, search_start,
2716 search_start + em_len - 1,
2717 EXTENT_DELALLOC_NEW, cached_state);
2719 search_start = extent_map_end(em);
2720 free_extent_map(em);
2727 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2728 unsigned int extra_bits,
2729 struct extent_state **cached_state)
2731 WARN_ON(PAGE_ALIGNED(end));
2733 if (start >= i_size_read(&inode->vfs_inode) &&
2734 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2736 * There can't be any extents following eof in this case so just
2737 * set the delalloc new bit for the range directly.
2739 extra_bits |= EXTENT_DELALLOC_NEW;
2743 ret = btrfs_find_new_delalloc_bytes(inode, start,
2750 return set_extent_bit(&inode->io_tree, start, end,
2751 EXTENT_DELALLOC | extra_bits, cached_state);
2754 /* see btrfs_writepage_start_hook for details on why this is required */
2755 struct btrfs_writepage_fixup {
2757 struct btrfs_inode *inode;
2758 struct btrfs_work work;
2761 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2763 struct btrfs_writepage_fixup *fixup =
2764 container_of(work, struct btrfs_writepage_fixup, work);
2765 struct btrfs_ordered_extent *ordered;
2766 struct extent_state *cached_state = NULL;
2767 struct extent_changeset *data_reserved = NULL;
2768 struct page *page = fixup->page;
2769 struct btrfs_inode *inode = fixup->inode;
2770 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2771 u64 page_start = page_offset(page);
2772 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2774 bool free_delalloc_space = true;
2777 * This is similar to page_mkwrite, we need to reserve the space before
2778 * we take the page lock.
2780 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2786 * Before we queued this fixup, we took a reference on the page.
2787 * page->mapping may go NULL, but it shouldn't be moved to a different
2790 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2792 * Unfortunately this is a little tricky, either
2794 * 1) We got here and our page had already been dealt with and
2795 * we reserved our space, thus ret == 0, so we need to just
2796 * drop our space reservation and bail. This can happen the
2797 * first time we come into the fixup worker, or could happen
2798 * while waiting for the ordered extent.
2799 * 2) Our page was already dealt with, but we happened to get an
2800 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2801 * this case we obviously don't have anything to release, but
2802 * because the page was already dealt with we don't want to
2803 * mark the page with an error, so make sure we're resetting
2804 * ret to 0. This is why we have this check _before_ the ret
2805 * check, because we do not want to have a surprise ENOSPC
2806 * when the page was already properly dealt with.
2809 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2810 btrfs_delalloc_release_space(inode, data_reserved,
2811 page_start, PAGE_SIZE,
2819 * We can't mess with the page state unless it is locked, so now that
2820 * it is locked bail if we failed to make our space reservation.
2825 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2827 /* already ordered? We're done */
2828 if (PageOrdered(page))
2831 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2833 unlock_extent(&inode->io_tree, page_start, page_end,
2836 btrfs_start_ordered_extent(ordered);
2837 btrfs_put_ordered_extent(ordered);
2841 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2847 * Everything went as planned, we're now the owner of a dirty page with
2848 * delayed allocation bits set and space reserved for our COW
2851 * The page was dirty when we started, nothing should have cleaned it.
2853 BUG_ON(!PageDirty(page));
2854 free_delalloc_space = false;
2856 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2857 if (free_delalloc_space)
2858 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2860 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2864 * We hit ENOSPC or other errors. Update the mapping and page
2865 * to reflect the errors and clean the page.
2867 mapping_set_error(page->mapping, ret);
2868 btrfs_mark_ordered_io_finished(inode, page, page_start,
2870 btrfs_page_clear_uptodate(fs_info, page, page_start, PAGE_SIZE);
2871 clear_page_dirty_for_io(page);
2873 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
2877 extent_changeset_free(data_reserved);
2879 * As a precaution, do a delayed iput in case it would be the last iput
2880 * that could need flushing space. Recursing back to fixup worker would
2883 btrfs_add_delayed_iput(inode);
2887 * There are a few paths in the higher layers of the kernel that directly
2888 * set the page dirty bit without asking the filesystem if it is a
2889 * good idea. This causes problems because we want to make sure COW
2890 * properly happens and the data=ordered rules are followed.
2892 * In our case any range that doesn't have the ORDERED bit set
2893 * hasn't been properly setup for IO. We kick off an async process
2894 * to fix it up. The async helper will wait for ordered extents, set
2895 * the delalloc bit and make it safe to write the page.
2897 int btrfs_writepage_cow_fixup(struct page *page)
2899 struct inode *inode = page->mapping->host;
2900 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2901 struct btrfs_writepage_fixup *fixup;
2903 /* This page has ordered extent covering it already */
2904 if (PageOrdered(page))
2908 * PageChecked is set below when we create a fixup worker for this page,
2909 * don't try to create another one if we're already PageChecked()
2911 * The extent_io writepage code will redirty the page if we send back
2914 if (PageChecked(page))
2917 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2922 * We are already holding a reference to this inode from
2923 * write_cache_pages. We need to hold it because the space reservation
2924 * takes place outside of the page lock, and we can't trust
2925 * page->mapping outside of the page lock.
2928 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2930 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2932 fixup->inode = BTRFS_I(inode);
2933 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2938 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2939 struct btrfs_inode *inode, u64 file_pos,
2940 struct btrfs_file_extent_item *stack_fi,
2941 const bool update_inode_bytes,
2942 u64 qgroup_reserved)
2944 struct btrfs_root *root = inode->root;
2945 const u64 sectorsize = root->fs_info->sectorsize;
2946 struct btrfs_path *path;
2947 struct extent_buffer *leaf;
2948 struct btrfs_key ins;
2949 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2950 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2951 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2952 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2953 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2954 struct btrfs_drop_extents_args drop_args = { 0 };
2957 path = btrfs_alloc_path();
2962 * we may be replacing one extent in the tree with another.
2963 * The new extent is pinned in the extent map, and we don't want
2964 * to drop it from the cache until it is completely in the btree.
2966 * So, tell btrfs_drop_extents to leave this extent in the cache.
2967 * the caller is expected to unpin it and allow it to be merged
2970 drop_args.path = path;
2971 drop_args.start = file_pos;
2972 drop_args.end = file_pos + num_bytes;
2973 drop_args.replace_extent = true;
2974 drop_args.extent_item_size = sizeof(*stack_fi);
2975 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2979 if (!drop_args.extent_inserted) {
2980 ins.objectid = btrfs_ino(inode);
2981 ins.offset = file_pos;
2982 ins.type = BTRFS_EXTENT_DATA_KEY;
2984 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2989 leaf = path->nodes[0];
2990 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2991 write_extent_buffer(leaf, stack_fi,
2992 btrfs_item_ptr_offset(leaf, path->slots[0]),
2993 sizeof(struct btrfs_file_extent_item));
2995 btrfs_mark_buffer_dirty(leaf);
2996 btrfs_release_path(path);
2999 * If we dropped an inline extent here, we know the range where it is
3000 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
3001 * number of bytes only for that range containing the inline extent.
3002 * The remaining of the range will be processed when clearning the
3003 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
3005 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
3006 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
3008 inline_size = drop_args.bytes_found - inline_size;
3009 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
3010 drop_args.bytes_found -= inline_size;
3011 num_bytes -= sectorsize;
3014 if (update_inode_bytes)
3015 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
3017 ins.objectid = disk_bytenr;
3018 ins.offset = disk_num_bytes;
3019 ins.type = BTRFS_EXTENT_ITEM_KEY;
3021 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3025 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3027 qgroup_reserved, &ins);
3029 btrfs_free_path(path);
3034 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3037 struct btrfs_block_group *cache;
3039 cache = btrfs_lookup_block_group(fs_info, start);
3042 spin_lock(&cache->lock);
3043 cache->delalloc_bytes -= len;
3044 spin_unlock(&cache->lock);
3046 btrfs_put_block_group(cache);
3049 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3050 struct btrfs_ordered_extent *oe)
3052 struct btrfs_file_extent_item stack_fi;
3053 bool update_inode_bytes;
3054 u64 num_bytes = oe->num_bytes;
3055 u64 ram_bytes = oe->ram_bytes;
3057 memset(&stack_fi, 0, sizeof(stack_fi));
3058 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3059 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3060 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3061 oe->disk_num_bytes);
3062 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3063 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3064 num_bytes = oe->truncated_len;
3065 ram_bytes = num_bytes;
3067 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3068 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3069 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3070 /* Encryption and other encoding is reserved and all 0 */
3073 * For delalloc, when completing an ordered extent we update the inode's
3074 * bytes when clearing the range in the inode's io tree, so pass false
3075 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3076 * except if the ordered extent was truncated.
3078 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3079 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3080 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3082 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3083 oe->file_offset, &stack_fi,
3084 update_inode_bytes, oe->qgroup_rsv);
3088 * As ordered data IO finishes, this gets called so we can finish
3089 * an ordered extent if the range of bytes in the file it covers are
3092 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3094 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3095 struct btrfs_root *root = inode->root;
3096 struct btrfs_fs_info *fs_info = root->fs_info;
3097 struct btrfs_trans_handle *trans = NULL;
3098 struct extent_io_tree *io_tree = &inode->io_tree;
3099 struct extent_state *cached_state = NULL;
3101 int compress_type = 0;
3103 u64 logical_len = ordered_extent->num_bytes;
3104 bool freespace_inode;
3105 bool truncated = false;
3106 bool clear_reserved_extent = true;
3107 unsigned int clear_bits = EXTENT_DEFRAG;
3109 start = ordered_extent->file_offset;
3110 end = start + ordered_extent->num_bytes - 1;
3112 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3113 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3114 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3115 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3116 clear_bits |= EXTENT_DELALLOC_NEW;
3118 freespace_inode = btrfs_is_free_space_inode(inode);
3119 if (!freespace_inode)
3120 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3122 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3127 if (btrfs_is_zoned(fs_info))
3128 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3129 ordered_extent->disk_num_bytes);
3131 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3133 logical_len = ordered_extent->truncated_len;
3134 /* Truncated the entire extent, don't bother adding */
3139 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3140 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3142 btrfs_inode_safe_disk_i_size_write(inode, 0);
3143 if (freespace_inode)
3144 trans = btrfs_join_transaction_spacecache(root);
3146 trans = btrfs_join_transaction(root);
3147 if (IS_ERR(trans)) {
3148 ret = PTR_ERR(trans);
3152 trans->block_rsv = &inode->block_rsv;
3153 ret = btrfs_update_inode_fallback(trans, root, inode);
3154 if (ret) /* -ENOMEM or corruption */
3155 btrfs_abort_transaction(trans, ret);
3159 clear_bits |= EXTENT_LOCKED;
3160 lock_extent(io_tree, start, end, &cached_state);
3162 if (freespace_inode)
3163 trans = btrfs_join_transaction_spacecache(root);
3165 trans = btrfs_join_transaction(root);
3166 if (IS_ERR(trans)) {
3167 ret = PTR_ERR(trans);
3172 trans->block_rsv = &inode->block_rsv;
3174 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3175 compress_type = ordered_extent->compress_type;
3176 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3177 BUG_ON(compress_type);
3178 ret = btrfs_mark_extent_written(trans, inode,
3179 ordered_extent->file_offset,
3180 ordered_extent->file_offset +
3182 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3183 ordered_extent->disk_num_bytes);
3185 BUG_ON(root == fs_info->tree_root);
3186 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3188 clear_reserved_extent = false;
3189 btrfs_release_delalloc_bytes(fs_info,
3190 ordered_extent->disk_bytenr,
3191 ordered_extent->disk_num_bytes);
3194 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3195 ordered_extent->num_bytes, trans->transid);
3197 btrfs_abort_transaction(trans, ret);
3201 ret = add_pending_csums(trans, &ordered_extent->list);
3203 btrfs_abort_transaction(trans, ret);
3208 * If this is a new delalloc range, clear its new delalloc flag to
3209 * update the inode's number of bytes. This needs to be done first
3210 * before updating the inode item.
3212 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3213 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3214 clear_extent_bit(&inode->io_tree, start, end,
3215 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3218 btrfs_inode_safe_disk_i_size_write(inode, 0);
3219 ret = btrfs_update_inode_fallback(trans, root, inode);
3220 if (ret) { /* -ENOMEM or corruption */
3221 btrfs_abort_transaction(trans, ret);
3226 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3230 btrfs_end_transaction(trans);
3232 if (ret || truncated) {
3233 u64 unwritten_start = start;
3236 * If we failed to finish this ordered extent for any reason we
3237 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3238 * extent, and mark the inode with the error if it wasn't
3239 * already set. Any error during writeback would have already
3240 * set the mapping error, so we need to set it if we're the ones
3241 * marking this ordered extent as failed.
3243 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3244 &ordered_extent->flags))
3245 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3248 unwritten_start += logical_len;
3249 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3251 /* Drop extent maps for the part of the extent we didn't write. */
3252 btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3255 * If the ordered extent had an IOERR or something else went
3256 * wrong we need to return the space for this ordered extent
3257 * back to the allocator. We only free the extent in the
3258 * truncated case if we didn't write out the extent at all.
3260 * If we made it past insert_reserved_file_extent before we
3261 * errored out then we don't need to do this as the accounting
3262 * has already been done.
3264 if ((ret || !logical_len) &&
3265 clear_reserved_extent &&
3266 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3267 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3269 * Discard the range before returning it back to the
3272 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3273 btrfs_discard_extent(fs_info,
3274 ordered_extent->disk_bytenr,
3275 ordered_extent->disk_num_bytes,
3277 btrfs_free_reserved_extent(fs_info,
3278 ordered_extent->disk_bytenr,
3279 ordered_extent->disk_num_bytes, 1);
3281 * Actually free the qgroup rsv which was released when
3282 * the ordered extent was created.
3284 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3285 ordered_extent->qgroup_rsv,
3286 BTRFS_QGROUP_RSV_DATA);
3291 * This needs to be done to make sure anybody waiting knows we are done
3292 * updating everything for this ordered extent.
3294 btrfs_remove_ordered_extent(inode, ordered_extent);
3297 btrfs_put_ordered_extent(ordered_extent);
3298 /* once for the tree */
3299 btrfs_put_ordered_extent(ordered_extent);
3304 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3306 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3307 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
3308 btrfs_finish_ordered_zoned(ordered);
3309 return btrfs_finish_one_ordered(ordered);
3313 * Verify the checksum for a single sector without any extra action that depend
3314 * on the type of I/O.
3316 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3317 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3319 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3322 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3324 shash->tfm = fs_info->csum_shash;
3326 kaddr = kmap_local_page(page) + pgoff;
3327 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3328 kunmap_local(kaddr);
3330 if (memcmp(csum, csum_expected, fs_info->csum_size))
3336 * Verify the checksum of a single data sector.
3338 * @bbio: btrfs_io_bio which contains the csum
3339 * @dev: device the sector is on
3340 * @bio_offset: offset to the beginning of the bio (in bytes)
3341 * @bv: bio_vec to check
3343 * Check if the checksum on a data block is valid. When a checksum mismatch is
3344 * detected, report the error and fill the corrupted range with zero.
3346 * Return %true if the sector is ok or had no checksum to start with, else %false.
3348 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3349 u32 bio_offset, struct bio_vec *bv)
3351 struct btrfs_inode *inode = bbio->inode;
3352 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3353 u64 file_offset = bbio->file_offset + bio_offset;
3354 u64 end = file_offset + bv->bv_len - 1;
3356 u8 csum[BTRFS_CSUM_SIZE];
3358 ASSERT(bv->bv_len == fs_info->sectorsize);
3363 if (btrfs_is_data_reloc_root(inode->root) &&
3364 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3366 /* Skip the range without csum for data reloc inode */
3367 clear_extent_bits(&inode->io_tree, file_offset, end,
3372 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3374 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3380 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3383 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3389 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3391 * @inode: The inode we want to perform iput on
3393 * This function uses the generic vfs_inode::i_count to track whether we should
3394 * just decrement it (in case it's > 1) or if this is the last iput then link
3395 * the inode to the delayed iput machinery. Delayed iputs are processed at
3396 * transaction commit time/superblock commit/cleaner kthread.
3398 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3400 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3401 unsigned long flags;
3403 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3406 atomic_inc(&fs_info->nr_delayed_iputs);
3408 * Need to be irq safe here because we can be called from either an irq
3409 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3412 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3413 ASSERT(list_empty(&inode->delayed_iput));
3414 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3415 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3416 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3417 wake_up_process(fs_info->cleaner_kthread);
3420 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3421 struct btrfs_inode *inode)
3423 list_del_init(&inode->delayed_iput);
3424 spin_unlock_irq(&fs_info->delayed_iput_lock);
3425 iput(&inode->vfs_inode);
3426 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3427 wake_up(&fs_info->delayed_iputs_wait);
3428 spin_lock_irq(&fs_info->delayed_iput_lock);
3431 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3432 struct btrfs_inode *inode)
3434 if (!list_empty(&inode->delayed_iput)) {
3435 spin_lock_irq(&fs_info->delayed_iput_lock);
3436 if (!list_empty(&inode->delayed_iput))
3437 run_delayed_iput_locked(fs_info, inode);
3438 spin_unlock_irq(&fs_info->delayed_iput_lock);
3442 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3445 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3446 * calls btrfs_add_delayed_iput() and that needs to lock
3447 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3448 * prevent a deadlock.
3450 spin_lock_irq(&fs_info->delayed_iput_lock);
3451 while (!list_empty(&fs_info->delayed_iputs)) {
3452 struct btrfs_inode *inode;
3454 inode = list_first_entry(&fs_info->delayed_iputs,
3455 struct btrfs_inode, delayed_iput);
3456 run_delayed_iput_locked(fs_info, inode);
3457 if (need_resched()) {
3458 spin_unlock_irq(&fs_info->delayed_iput_lock);
3460 spin_lock_irq(&fs_info->delayed_iput_lock);
3463 spin_unlock_irq(&fs_info->delayed_iput_lock);
3467 * Wait for flushing all delayed iputs
3469 * @fs_info: the filesystem
3471 * This will wait on any delayed iputs that are currently running with KILLABLE
3472 * set. Once they are all done running we will return, unless we are killed in
3473 * which case we return EINTR. This helps in user operations like fallocate etc
3474 * that might get blocked on the iputs.
3476 * Return EINTR if we were killed, 0 if nothing's pending
3478 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3480 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3481 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3488 * This creates an orphan entry for the given inode in case something goes wrong
3489 * in the middle of an unlink.
3491 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3492 struct btrfs_inode *inode)
3496 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3497 if (ret && ret != -EEXIST) {
3498 btrfs_abort_transaction(trans, ret);
3506 * We have done the delete so we can go ahead and remove the orphan item for
3507 * this particular inode.
3509 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3510 struct btrfs_inode *inode)
3512 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3516 * this cleans up any orphans that may be left on the list from the last use
3519 int btrfs_orphan_cleanup(struct btrfs_root *root)
3521 struct btrfs_fs_info *fs_info = root->fs_info;
3522 struct btrfs_path *path;
3523 struct extent_buffer *leaf;
3524 struct btrfs_key key, found_key;
3525 struct btrfs_trans_handle *trans;
3526 struct inode *inode;
3527 u64 last_objectid = 0;
3528 int ret = 0, nr_unlink = 0;
3530 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3533 path = btrfs_alloc_path();
3538 path->reada = READA_BACK;
3540 key.objectid = BTRFS_ORPHAN_OBJECTID;
3541 key.type = BTRFS_ORPHAN_ITEM_KEY;
3542 key.offset = (u64)-1;
3545 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3550 * if ret == 0 means we found what we were searching for, which
3551 * is weird, but possible, so only screw with path if we didn't
3552 * find the key and see if we have stuff that matches
3556 if (path->slots[0] == 0)
3561 /* pull out the item */
3562 leaf = path->nodes[0];
3563 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3565 /* make sure the item matches what we want */
3566 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3568 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3571 /* release the path since we're done with it */
3572 btrfs_release_path(path);
3575 * this is where we are basically btrfs_lookup, without the
3576 * crossing root thing. we store the inode number in the
3577 * offset of the orphan item.
3580 if (found_key.offset == last_objectid) {
3582 "Error removing orphan entry, stopping orphan cleanup");
3587 last_objectid = found_key.offset;
3589 found_key.objectid = found_key.offset;
3590 found_key.type = BTRFS_INODE_ITEM_KEY;
3591 found_key.offset = 0;
3592 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3593 if (IS_ERR(inode)) {
3594 ret = PTR_ERR(inode);
3600 if (!inode && root == fs_info->tree_root) {
3601 struct btrfs_root *dead_root;
3602 int is_dead_root = 0;
3605 * This is an orphan in the tree root. Currently these
3606 * could come from 2 sources:
3607 * a) a root (snapshot/subvolume) deletion in progress
3608 * b) a free space cache inode
3609 * We need to distinguish those two, as the orphan item
3610 * for a root must not get deleted before the deletion
3611 * of the snapshot/subvolume's tree completes.
3613 * btrfs_find_orphan_roots() ran before us, which has
3614 * found all deleted roots and loaded them into
3615 * fs_info->fs_roots_radix. So here we can find if an
3616 * orphan item corresponds to a deleted root by looking
3617 * up the root from that radix tree.
3620 spin_lock(&fs_info->fs_roots_radix_lock);
3621 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3622 (unsigned long)found_key.objectid);
3623 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3625 spin_unlock(&fs_info->fs_roots_radix_lock);
3628 /* prevent this orphan from being found again */
3629 key.offset = found_key.objectid - 1;
3636 * If we have an inode with links, there are a couple of
3639 * 1. We were halfway through creating fsverity metadata for the
3640 * file. In that case, the orphan item represents incomplete
3641 * fsverity metadata which must be cleaned up with
3642 * btrfs_drop_verity_items and deleting the orphan item.
3644 * 2. Old kernels (before v3.12) used to create an
3645 * orphan item for truncate indicating that there were possibly
3646 * extent items past i_size that needed to be deleted. In v3.12,
3647 * truncate was changed to update i_size in sync with the extent
3648 * items, but the (useless) orphan item was still created. Since
3649 * v4.18, we don't create the orphan item for truncate at all.
3651 * So, this item could mean that we need to do a truncate, but
3652 * only if this filesystem was last used on a pre-v3.12 kernel
3653 * and was not cleanly unmounted. The odds of that are quite
3654 * slim, and it's a pain to do the truncate now, so just delete
3657 * It's also possible that this orphan item was supposed to be
3658 * deleted but wasn't. The inode number may have been reused,
3659 * but either way, we can delete the orphan item.
3661 if (!inode || inode->i_nlink) {
3663 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3669 trans = btrfs_start_transaction(root, 1);
3670 if (IS_ERR(trans)) {
3671 ret = PTR_ERR(trans);
3674 btrfs_debug(fs_info, "auto deleting %Lu",
3675 found_key.objectid);
3676 ret = btrfs_del_orphan_item(trans, root,
3677 found_key.objectid);
3678 btrfs_end_transaction(trans);
3686 /* this will do delete_inode and everything for us */
3689 /* release the path since we're done with it */
3690 btrfs_release_path(path);
3692 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3693 trans = btrfs_join_transaction(root);
3695 btrfs_end_transaction(trans);
3699 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3703 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3704 btrfs_free_path(path);
3709 * very simple check to peek ahead in the leaf looking for xattrs. If we
3710 * don't find any xattrs, we know there can't be any acls.
3712 * slot is the slot the inode is in, objectid is the objectid of the inode
3714 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3715 int slot, u64 objectid,
3716 int *first_xattr_slot)
3718 u32 nritems = btrfs_header_nritems(leaf);
3719 struct btrfs_key found_key;
3720 static u64 xattr_access = 0;
3721 static u64 xattr_default = 0;
3724 if (!xattr_access) {
3725 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3726 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3727 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3728 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3732 *first_xattr_slot = -1;
3733 while (slot < nritems) {
3734 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3736 /* we found a different objectid, there must not be acls */
3737 if (found_key.objectid != objectid)
3740 /* we found an xattr, assume we've got an acl */
3741 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3742 if (*first_xattr_slot == -1)
3743 *first_xattr_slot = slot;
3744 if (found_key.offset == xattr_access ||
3745 found_key.offset == xattr_default)
3750 * we found a key greater than an xattr key, there can't
3751 * be any acls later on
3753 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3760 * it goes inode, inode backrefs, xattrs, extents,
3761 * so if there are a ton of hard links to an inode there can
3762 * be a lot of backrefs. Don't waste time searching too hard,
3763 * this is just an optimization
3768 /* we hit the end of the leaf before we found an xattr or
3769 * something larger than an xattr. We have to assume the inode
3772 if (*first_xattr_slot == -1)
3773 *first_xattr_slot = slot;
3778 * read an inode from the btree into the in-memory inode
3780 static int btrfs_read_locked_inode(struct inode *inode,
3781 struct btrfs_path *in_path)
3783 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3784 struct btrfs_path *path = in_path;
3785 struct extent_buffer *leaf;
3786 struct btrfs_inode_item *inode_item;
3787 struct btrfs_root *root = BTRFS_I(inode)->root;
3788 struct btrfs_key location;
3793 bool filled = false;
3794 int first_xattr_slot;
3796 ret = btrfs_fill_inode(inode, &rdev);
3801 path = btrfs_alloc_path();
3806 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3808 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3810 if (path != in_path)
3811 btrfs_free_path(path);
3815 leaf = path->nodes[0];
3820 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3821 struct btrfs_inode_item);
3822 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3823 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3824 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3825 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3826 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3827 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3828 round_up(i_size_read(inode), fs_info->sectorsize));
3830 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3831 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3833 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3834 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3836 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3837 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3839 BTRFS_I(inode)->i_otime.tv_sec =
3840 btrfs_timespec_sec(leaf, &inode_item->otime);
3841 BTRFS_I(inode)->i_otime.tv_nsec =
3842 btrfs_timespec_nsec(leaf, &inode_item->otime);
3844 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3845 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3846 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3848 inode_set_iversion_queried(inode,
3849 btrfs_inode_sequence(leaf, inode_item));
3850 inode->i_generation = BTRFS_I(inode)->generation;
3852 rdev = btrfs_inode_rdev(leaf, inode_item);
3854 BTRFS_I(inode)->index_cnt = (u64)-1;
3855 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3856 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3860 * If we were modified in the current generation and evicted from memory
3861 * and then re-read we need to do a full sync since we don't have any
3862 * idea about which extents were modified before we were evicted from
3865 * This is required for both inode re-read from disk and delayed inode
3866 * in delayed_nodes_tree.
3868 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3869 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3870 &BTRFS_I(inode)->runtime_flags);
3873 * We don't persist the id of the transaction where an unlink operation
3874 * against the inode was last made. So here we assume the inode might
3875 * have been evicted, and therefore the exact value of last_unlink_trans
3876 * lost, and set it to last_trans to avoid metadata inconsistencies
3877 * between the inode and its parent if the inode is fsync'ed and the log
3878 * replayed. For example, in the scenario:
3881 * ln mydir/foo mydir/bar
3884 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3885 * xfs_io -c fsync mydir/foo
3887 * mount fs, triggers fsync log replay
3889 * We must make sure that when we fsync our inode foo we also log its
3890 * parent inode, otherwise after log replay the parent still has the
3891 * dentry with the "bar" name but our inode foo has a link count of 1
3892 * and doesn't have an inode ref with the name "bar" anymore.
3894 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3895 * but it guarantees correctness at the expense of occasional full
3896 * transaction commits on fsync if our inode is a directory, or if our
3897 * inode is not a directory, logging its parent unnecessarily.
3899 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3902 * Same logic as for last_unlink_trans. We don't persist the generation
3903 * of the last transaction where this inode was used for a reflink
3904 * operation, so after eviction and reloading the inode we must be
3905 * pessimistic and assume the last transaction that modified the inode.
3907 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3910 if (inode->i_nlink != 1 ||
3911 path->slots[0] >= btrfs_header_nritems(leaf))
3914 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3915 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3918 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3919 if (location.type == BTRFS_INODE_REF_KEY) {
3920 struct btrfs_inode_ref *ref;
3922 ref = (struct btrfs_inode_ref *)ptr;
3923 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3924 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3925 struct btrfs_inode_extref *extref;
3927 extref = (struct btrfs_inode_extref *)ptr;
3928 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3933 * try to precache a NULL acl entry for files that don't have
3934 * any xattrs or acls
3936 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3937 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3938 if (first_xattr_slot != -1) {
3939 path->slots[0] = first_xattr_slot;
3940 ret = btrfs_load_inode_props(inode, path);
3943 "error loading props for ino %llu (root %llu): %d",
3944 btrfs_ino(BTRFS_I(inode)),
3945 root->root_key.objectid, ret);
3947 if (path != in_path)
3948 btrfs_free_path(path);
3951 cache_no_acl(inode);
3953 switch (inode->i_mode & S_IFMT) {
3955 inode->i_mapping->a_ops = &btrfs_aops;
3956 inode->i_fop = &btrfs_file_operations;
3957 inode->i_op = &btrfs_file_inode_operations;
3960 inode->i_fop = &btrfs_dir_file_operations;
3961 inode->i_op = &btrfs_dir_inode_operations;
3964 inode->i_op = &btrfs_symlink_inode_operations;
3965 inode_nohighmem(inode);
3966 inode->i_mapping->a_ops = &btrfs_aops;
3969 inode->i_op = &btrfs_special_inode_operations;
3970 init_special_inode(inode, inode->i_mode, rdev);
3974 btrfs_sync_inode_flags_to_i_flags(inode);
3979 * given a leaf and an inode, copy the inode fields into the leaf
3981 static void fill_inode_item(struct btrfs_trans_handle *trans,
3982 struct extent_buffer *leaf,
3983 struct btrfs_inode_item *item,
3984 struct inode *inode)
3986 struct btrfs_map_token token;
3989 btrfs_init_map_token(&token, leaf);
3991 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3992 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3993 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3994 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3995 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3997 btrfs_set_token_timespec_sec(&token, &item->atime,
3998 inode->i_atime.tv_sec);
3999 btrfs_set_token_timespec_nsec(&token, &item->atime,
4000 inode->i_atime.tv_nsec);
4002 btrfs_set_token_timespec_sec(&token, &item->mtime,
4003 inode->i_mtime.tv_sec);
4004 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4005 inode->i_mtime.tv_nsec);
4007 btrfs_set_token_timespec_sec(&token, &item->ctime,
4008 inode->i_ctime.tv_sec);
4009 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4010 inode->i_ctime.tv_nsec);
4012 btrfs_set_token_timespec_sec(&token, &item->otime,
4013 BTRFS_I(inode)->i_otime.tv_sec);
4014 btrfs_set_token_timespec_nsec(&token, &item->otime,
4015 BTRFS_I(inode)->i_otime.tv_nsec);
4017 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4018 btrfs_set_token_inode_generation(&token, item,
4019 BTRFS_I(inode)->generation);
4020 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4021 btrfs_set_token_inode_transid(&token, item, trans->transid);
4022 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4023 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4024 BTRFS_I(inode)->ro_flags);
4025 btrfs_set_token_inode_flags(&token, item, flags);
4026 btrfs_set_token_inode_block_group(&token, item, 0);
4030 * copy everything in the in-memory inode into the btree.
4032 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4033 struct btrfs_root *root,
4034 struct btrfs_inode *inode)
4036 struct btrfs_inode_item *inode_item;
4037 struct btrfs_path *path;
4038 struct extent_buffer *leaf;
4041 path = btrfs_alloc_path();
4045 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4052 leaf = path->nodes[0];
4053 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4054 struct btrfs_inode_item);
4056 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4057 btrfs_mark_buffer_dirty(leaf);
4058 btrfs_set_inode_last_trans(trans, inode);
4061 btrfs_free_path(path);
4066 * copy everything in the in-memory inode into the btree.
4068 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4069 struct btrfs_root *root,
4070 struct btrfs_inode *inode)
4072 struct btrfs_fs_info *fs_info = root->fs_info;
4076 * If the inode is a free space inode, we can deadlock during commit
4077 * if we put it into the delayed code.
4079 * The data relocation inode should also be directly updated
4082 if (!btrfs_is_free_space_inode(inode)
4083 && !btrfs_is_data_reloc_root(root)
4084 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4085 btrfs_update_root_times(trans, root);
4087 ret = btrfs_delayed_update_inode(trans, root, inode);
4089 btrfs_set_inode_last_trans(trans, inode);
4093 return btrfs_update_inode_item(trans, root, inode);
4096 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4097 struct btrfs_root *root, struct btrfs_inode *inode)
4101 ret = btrfs_update_inode(trans, root, inode);
4103 return btrfs_update_inode_item(trans, root, inode);
4108 * unlink helper that gets used here in inode.c and in the tree logging
4109 * recovery code. It remove a link in a directory with a given name, and
4110 * also drops the back refs in the inode to the directory
4112 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4113 struct btrfs_inode *dir,
4114 struct btrfs_inode *inode,
4115 const struct fscrypt_str *name,
4116 struct btrfs_rename_ctx *rename_ctx)
4118 struct btrfs_root *root = dir->root;
4119 struct btrfs_fs_info *fs_info = root->fs_info;
4120 struct btrfs_path *path;
4122 struct btrfs_dir_item *di;
4124 u64 ino = btrfs_ino(inode);
4125 u64 dir_ino = btrfs_ino(dir);
4127 path = btrfs_alloc_path();
4133 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4134 if (IS_ERR_OR_NULL(di)) {
4135 ret = di ? PTR_ERR(di) : -ENOENT;
4138 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4141 btrfs_release_path(path);
4144 * If we don't have dir index, we have to get it by looking up
4145 * the inode ref, since we get the inode ref, remove it directly,
4146 * it is unnecessary to do delayed deletion.
4148 * But if we have dir index, needn't search inode ref to get it.
4149 * Since the inode ref is close to the inode item, it is better
4150 * that we delay to delete it, and just do this deletion when
4151 * we update the inode item.
4153 if (inode->dir_index) {
4154 ret = btrfs_delayed_delete_inode_ref(inode);
4156 index = inode->dir_index;
4161 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4164 "failed to delete reference to %.*s, inode %llu parent %llu",
4165 name->len, name->name, ino, dir_ino);
4166 btrfs_abort_transaction(trans, ret);
4171 rename_ctx->index = index;
4173 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4175 btrfs_abort_transaction(trans, ret);
4180 * If we are in a rename context, we don't need to update anything in the
4181 * log. That will be done later during the rename by btrfs_log_new_name().
4182 * Besides that, doing it here would only cause extra unnecessary btree
4183 * operations on the log tree, increasing latency for applications.
4186 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4187 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4191 * If we have a pending delayed iput we could end up with the final iput
4192 * being run in btrfs-cleaner context. If we have enough of these built
4193 * up we can end up burning a lot of time in btrfs-cleaner without any
4194 * way to throttle the unlinks. Since we're currently holding a ref on
4195 * the inode we can run the delayed iput here without any issues as the
4196 * final iput won't be done until after we drop the ref we're currently
4199 btrfs_run_delayed_iput(fs_info, inode);
4201 btrfs_free_path(path);
4205 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4206 inode_inc_iversion(&inode->vfs_inode);
4207 inode_inc_iversion(&dir->vfs_inode);
4208 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4209 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4210 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4211 ret = btrfs_update_inode(trans, root, dir);
4216 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4217 struct btrfs_inode *dir, struct btrfs_inode *inode,
4218 const struct fscrypt_str *name)
4222 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4224 drop_nlink(&inode->vfs_inode);
4225 ret = btrfs_update_inode(trans, inode->root, inode);
4231 * helper to start transaction for unlink and rmdir.
4233 * unlink and rmdir are special in btrfs, they do not always free space, so
4234 * if we cannot make our reservations the normal way try and see if there is
4235 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4236 * allow the unlink to occur.
4238 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4240 struct btrfs_root *root = dir->root;
4242 return btrfs_start_transaction_fallback_global_rsv(root,
4243 BTRFS_UNLINK_METADATA_UNITS);
4246 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4248 struct btrfs_trans_handle *trans;
4249 struct inode *inode = d_inode(dentry);
4251 struct fscrypt_name fname;
4253 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4257 /* This needs to handle no-key deletions later on */
4259 trans = __unlink_start_trans(BTRFS_I(dir));
4260 if (IS_ERR(trans)) {
4261 ret = PTR_ERR(trans);
4265 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4268 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4273 if (inode->i_nlink == 0) {
4274 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4280 btrfs_end_transaction(trans);
4281 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4283 fscrypt_free_filename(&fname);
4287 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4288 struct btrfs_inode *dir, struct dentry *dentry)
4290 struct btrfs_root *root = dir->root;
4291 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4292 struct btrfs_path *path;
4293 struct extent_buffer *leaf;
4294 struct btrfs_dir_item *di;
4295 struct btrfs_key key;
4299 u64 dir_ino = btrfs_ino(dir);
4300 struct fscrypt_name fname;
4302 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4306 /* This needs to handle no-key deletions later on */
4308 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4309 objectid = inode->root->root_key.objectid;
4310 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4311 objectid = inode->location.objectid;
4314 fscrypt_free_filename(&fname);
4318 path = btrfs_alloc_path();
4324 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4325 &fname.disk_name, -1);
4326 if (IS_ERR_OR_NULL(di)) {
4327 ret = di ? PTR_ERR(di) : -ENOENT;
4331 leaf = path->nodes[0];
4332 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4333 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4334 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4336 btrfs_abort_transaction(trans, ret);
4339 btrfs_release_path(path);
4342 * This is a placeholder inode for a subvolume we didn't have a
4343 * reference to at the time of the snapshot creation. In the meantime
4344 * we could have renamed the real subvol link into our snapshot, so
4345 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4346 * Instead simply lookup the dir_index_item for this entry so we can
4347 * remove it. Otherwise we know we have a ref to the root and we can
4348 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4350 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4351 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4352 if (IS_ERR_OR_NULL(di)) {
4357 btrfs_abort_transaction(trans, ret);
4361 leaf = path->nodes[0];
4362 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4364 btrfs_release_path(path);
4366 ret = btrfs_del_root_ref(trans, objectid,
4367 root->root_key.objectid, dir_ino,
4368 &index, &fname.disk_name);
4370 btrfs_abort_transaction(trans, ret);
4375 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4377 btrfs_abort_transaction(trans, ret);
4381 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4382 inode_inc_iversion(&dir->vfs_inode);
4383 dir->vfs_inode.i_mtime = current_time(&dir->vfs_inode);
4384 dir->vfs_inode.i_ctime = dir->vfs_inode.i_mtime;
4385 ret = btrfs_update_inode_fallback(trans, root, dir);
4387 btrfs_abort_transaction(trans, ret);
4389 btrfs_free_path(path);
4390 fscrypt_free_filename(&fname);
4395 * Helper to check if the subvolume references other subvolumes or if it's
4398 static noinline int may_destroy_subvol(struct btrfs_root *root)
4400 struct btrfs_fs_info *fs_info = root->fs_info;
4401 struct btrfs_path *path;
4402 struct btrfs_dir_item *di;
4403 struct btrfs_key key;
4404 struct fscrypt_str name = FSTR_INIT("default", 7);
4408 path = btrfs_alloc_path();
4412 /* Make sure this root isn't set as the default subvol */
4413 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4414 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4416 if (di && !IS_ERR(di)) {
4417 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4418 if (key.objectid == root->root_key.objectid) {
4421 "deleting default subvolume %llu is not allowed",
4425 btrfs_release_path(path);
4428 key.objectid = root->root_key.objectid;
4429 key.type = BTRFS_ROOT_REF_KEY;
4430 key.offset = (u64)-1;
4432 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4438 if (path->slots[0] > 0) {
4440 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4441 if (key.objectid == root->root_key.objectid &&
4442 key.type == BTRFS_ROOT_REF_KEY)
4446 btrfs_free_path(path);
4450 /* Delete all dentries for inodes belonging to the root */
4451 static void btrfs_prune_dentries(struct btrfs_root *root)
4453 struct btrfs_fs_info *fs_info = root->fs_info;
4454 struct rb_node *node;
4455 struct rb_node *prev;
4456 struct btrfs_inode *entry;
4457 struct inode *inode;
4460 if (!BTRFS_FS_ERROR(fs_info))
4461 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4463 spin_lock(&root->inode_lock);
4465 node = root->inode_tree.rb_node;
4469 entry = rb_entry(node, struct btrfs_inode, rb_node);
4471 if (objectid < btrfs_ino(entry))
4472 node = node->rb_left;
4473 else if (objectid > btrfs_ino(entry))
4474 node = node->rb_right;
4480 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4481 if (objectid <= btrfs_ino(entry)) {
4485 prev = rb_next(prev);
4489 entry = rb_entry(node, struct btrfs_inode, rb_node);
4490 objectid = btrfs_ino(entry) + 1;
4491 inode = igrab(&entry->vfs_inode);
4493 spin_unlock(&root->inode_lock);
4494 if (atomic_read(&inode->i_count) > 1)
4495 d_prune_aliases(inode);
4497 * btrfs_drop_inode will have it removed from the inode
4498 * cache when its usage count hits zero.
4502 spin_lock(&root->inode_lock);
4506 if (cond_resched_lock(&root->inode_lock))
4509 node = rb_next(node);
4511 spin_unlock(&root->inode_lock);
4514 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4516 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4517 struct btrfs_root *root = dir->root;
4518 struct inode *inode = d_inode(dentry);
4519 struct btrfs_root *dest = BTRFS_I(inode)->root;
4520 struct btrfs_trans_handle *trans;
4521 struct btrfs_block_rsv block_rsv;
4526 * Don't allow to delete a subvolume with send in progress. This is
4527 * inside the inode lock so the error handling that has to drop the bit
4528 * again is not run concurrently.
4530 spin_lock(&dest->root_item_lock);
4531 if (dest->send_in_progress) {
4532 spin_unlock(&dest->root_item_lock);
4534 "attempt to delete subvolume %llu during send",
4535 dest->root_key.objectid);
4538 if (atomic_read(&dest->nr_swapfiles)) {
4539 spin_unlock(&dest->root_item_lock);
4541 "attempt to delete subvolume %llu with active swapfile",
4542 root->root_key.objectid);
4545 root_flags = btrfs_root_flags(&dest->root_item);
4546 btrfs_set_root_flags(&dest->root_item,
4547 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4548 spin_unlock(&dest->root_item_lock);
4550 down_write(&fs_info->subvol_sem);
4552 ret = may_destroy_subvol(dest);
4556 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4558 * One for dir inode,
4559 * two for dir entries,
4560 * two for root ref/backref.
4562 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4566 trans = btrfs_start_transaction(root, 0);
4567 if (IS_ERR(trans)) {
4568 ret = PTR_ERR(trans);
4571 trans->block_rsv = &block_rsv;
4572 trans->bytes_reserved = block_rsv.size;
4574 btrfs_record_snapshot_destroy(trans, dir);
4576 ret = btrfs_unlink_subvol(trans, dir, dentry);
4578 btrfs_abort_transaction(trans, ret);
4582 ret = btrfs_record_root_in_trans(trans, dest);
4584 btrfs_abort_transaction(trans, ret);
4588 memset(&dest->root_item.drop_progress, 0,
4589 sizeof(dest->root_item.drop_progress));
4590 btrfs_set_root_drop_level(&dest->root_item, 0);
4591 btrfs_set_root_refs(&dest->root_item, 0);
4593 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4594 ret = btrfs_insert_orphan_item(trans,
4596 dest->root_key.objectid);
4598 btrfs_abort_transaction(trans, ret);
4603 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4604 BTRFS_UUID_KEY_SUBVOL,
4605 dest->root_key.objectid);
4606 if (ret && ret != -ENOENT) {
4607 btrfs_abort_transaction(trans, ret);
4610 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4611 ret = btrfs_uuid_tree_remove(trans,
4612 dest->root_item.received_uuid,
4613 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4614 dest->root_key.objectid);
4615 if (ret && ret != -ENOENT) {
4616 btrfs_abort_transaction(trans, ret);
4621 free_anon_bdev(dest->anon_dev);
4624 trans->block_rsv = NULL;
4625 trans->bytes_reserved = 0;
4626 ret = btrfs_end_transaction(trans);
4627 inode->i_flags |= S_DEAD;
4629 btrfs_subvolume_release_metadata(root, &block_rsv);
4631 up_write(&fs_info->subvol_sem);
4633 spin_lock(&dest->root_item_lock);
4634 root_flags = btrfs_root_flags(&dest->root_item);
4635 btrfs_set_root_flags(&dest->root_item,
4636 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4637 spin_unlock(&dest->root_item_lock);
4639 d_invalidate(dentry);
4640 btrfs_prune_dentries(dest);
4641 ASSERT(dest->send_in_progress == 0);
4647 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4649 struct inode *inode = d_inode(dentry);
4650 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4652 struct btrfs_trans_handle *trans;
4653 u64 last_unlink_trans;
4654 struct fscrypt_name fname;
4656 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4658 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4659 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4661 "extent tree v2 doesn't support snapshot deletion yet");
4664 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4667 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4671 /* This needs to handle no-key deletions later on */
4673 trans = __unlink_start_trans(BTRFS_I(dir));
4674 if (IS_ERR(trans)) {
4675 err = PTR_ERR(trans);
4679 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4680 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4684 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4688 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4690 /* now the directory is empty */
4691 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4694 btrfs_i_size_write(BTRFS_I(inode), 0);
4696 * Propagate the last_unlink_trans value of the deleted dir to
4697 * its parent directory. This is to prevent an unrecoverable
4698 * log tree in the case we do something like this:
4700 * 2) create snapshot under dir foo
4701 * 3) delete the snapshot
4704 * 6) fsync foo or some file inside foo
4706 if (last_unlink_trans >= trans->transid)
4707 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4710 btrfs_end_transaction(trans);
4712 btrfs_btree_balance_dirty(fs_info);
4713 fscrypt_free_filename(&fname);
4719 * btrfs_truncate_block - read, zero a chunk and write a block
4720 * @inode - inode that we're zeroing
4721 * @from - the offset to start zeroing
4722 * @len - the length to zero, 0 to zero the entire range respective to the
4724 * @front - zero up to the offset instead of from the offset on
4726 * This will find the block for the "from" offset and cow the block and zero the
4727 * part we want to zero. This is used with truncate and hole punching.
4729 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4732 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4733 struct address_space *mapping = inode->vfs_inode.i_mapping;
4734 struct extent_io_tree *io_tree = &inode->io_tree;
4735 struct btrfs_ordered_extent *ordered;
4736 struct extent_state *cached_state = NULL;
4737 struct extent_changeset *data_reserved = NULL;
4738 bool only_release_metadata = false;
4739 u32 blocksize = fs_info->sectorsize;
4740 pgoff_t index = from >> PAGE_SHIFT;
4741 unsigned offset = from & (blocksize - 1);
4743 gfp_t mask = btrfs_alloc_write_mask(mapping);
4744 size_t write_bytes = blocksize;
4749 if (IS_ALIGNED(offset, blocksize) &&
4750 (!len || IS_ALIGNED(len, blocksize)))
4753 block_start = round_down(from, blocksize);
4754 block_end = block_start + blocksize - 1;
4756 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4759 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4760 /* For nocow case, no need to reserve data space */
4761 only_release_metadata = true;
4766 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4768 if (!only_release_metadata)
4769 btrfs_free_reserved_data_space(inode, data_reserved,
4770 block_start, blocksize);
4774 page = find_or_create_page(mapping, index, mask);
4776 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4778 btrfs_delalloc_release_extents(inode, blocksize);
4783 if (!PageUptodate(page)) {
4784 ret = btrfs_read_folio(NULL, page_folio(page));
4786 if (page->mapping != mapping) {
4791 if (!PageUptodate(page)) {
4798 * We unlock the page after the io is completed and then re-lock it
4799 * above. release_folio() could have come in between that and cleared
4800 * PagePrivate(), but left the page in the mapping. Set the page mapped
4801 * here to make sure it's properly set for the subpage stuff.
4803 ret = set_page_extent_mapped(page);
4807 wait_on_page_writeback(page);
4809 lock_extent(io_tree, block_start, block_end, &cached_state);
4811 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4813 unlock_extent(io_tree, block_start, block_end, &cached_state);
4816 btrfs_start_ordered_extent(ordered);
4817 btrfs_put_ordered_extent(ordered);
4821 clear_extent_bit(&inode->io_tree, block_start, block_end,
4822 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4825 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4828 unlock_extent(io_tree, block_start, block_end, &cached_state);
4832 if (offset != blocksize) {
4834 len = blocksize - offset;
4836 memzero_page(page, (block_start - page_offset(page)),
4839 memzero_page(page, (block_start - page_offset(page)) + offset,
4842 btrfs_page_clear_checked(fs_info, page, block_start,
4843 block_end + 1 - block_start);
4844 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4845 unlock_extent(io_tree, block_start, block_end, &cached_state);
4847 if (only_release_metadata)
4848 set_extent_bit(&inode->io_tree, block_start, block_end,
4849 EXTENT_NORESERVE, NULL);
4853 if (only_release_metadata)
4854 btrfs_delalloc_release_metadata(inode, blocksize, true);
4856 btrfs_delalloc_release_space(inode, data_reserved,
4857 block_start, blocksize, true);
4859 btrfs_delalloc_release_extents(inode, blocksize);
4863 if (only_release_metadata)
4864 btrfs_check_nocow_unlock(inode);
4865 extent_changeset_free(data_reserved);
4869 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4870 u64 offset, u64 len)
4872 struct btrfs_fs_info *fs_info = root->fs_info;
4873 struct btrfs_trans_handle *trans;
4874 struct btrfs_drop_extents_args drop_args = { 0 };
4878 * If NO_HOLES is enabled, we don't need to do anything.
4879 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4880 * or btrfs_update_inode() will be called, which guarantee that the next
4881 * fsync will know this inode was changed and needs to be logged.
4883 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4887 * 1 - for the one we're dropping
4888 * 1 - for the one we're adding
4889 * 1 - for updating the inode.
4891 trans = btrfs_start_transaction(root, 3);
4893 return PTR_ERR(trans);
4895 drop_args.start = offset;
4896 drop_args.end = offset + len;
4897 drop_args.drop_cache = true;
4899 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4901 btrfs_abort_transaction(trans, ret);
4902 btrfs_end_transaction(trans);
4906 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4908 btrfs_abort_transaction(trans, ret);
4910 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4911 btrfs_update_inode(trans, root, inode);
4913 btrfs_end_transaction(trans);
4918 * This function puts in dummy file extents for the area we're creating a hole
4919 * for. So if we are truncating this file to a larger size we need to insert
4920 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4921 * the range between oldsize and size
4923 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4925 struct btrfs_root *root = inode->root;
4926 struct btrfs_fs_info *fs_info = root->fs_info;
4927 struct extent_io_tree *io_tree = &inode->io_tree;
4928 struct extent_map *em = NULL;
4929 struct extent_state *cached_state = NULL;
4930 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4931 u64 block_end = ALIGN(size, fs_info->sectorsize);
4938 * If our size started in the middle of a block we need to zero out the
4939 * rest of the block before we expand the i_size, otherwise we could
4940 * expose stale data.
4942 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4946 if (size <= hole_start)
4949 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4951 cur_offset = hole_start;
4953 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4954 block_end - cur_offset);
4960 last_byte = min(extent_map_end(em), block_end);
4961 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4962 hole_size = last_byte - cur_offset;
4964 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4965 struct extent_map *hole_em;
4967 err = maybe_insert_hole(root, inode, cur_offset,
4972 err = btrfs_inode_set_file_extent_range(inode,
4973 cur_offset, hole_size);
4977 hole_em = alloc_extent_map();
4979 btrfs_drop_extent_map_range(inode, cur_offset,
4980 cur_offset + hole_size - 1,
4982 btrfs_set_inode_full_sync(inode);
4985 hole_em->start = cur_offset;
4986 hole_em->len = hole_size;
4987 hole_em->orig_start = cur_offset;
4989 hole_em->block_start = EXTENT_MAP_HOLE;
4990 hole_em->block_len = 0;
4991 hole_em->orig_block_len = 0;
4992 hole_em->ram_bytes = hole_size;
4993 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4994 hole_em->generation = fs_info->generation;
4996 err = btrfs_replace_extent_map_range(inode, hole_em, true);
4997 free_extent_map(hole_em);
4999 err = btrfs_inode_set_file_extent_range(inode,
5000 cur_offset, hole_size);
5005 free_extent_map(em);
5007 cur_offset = last_byte;
5008 if (cur_offset >= block_end)
5011 free_extent_map(em);
5012 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
5016 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5018 struct btrfs_root *root = BTRFS_I(inode)->root;
5019 struct btrfs_trans_handle *trans;
5020 loff_t oldsize = i_size_read(inode);
5021 loff_t newsize = attr->ia_size;
5022 int mask = attr->ia_valid;
5026 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5027 * special case where we need to update the times despite not having
5028 * these flags set. For all other operations the VFS set these flags
5029 * explicitly if it wants a timestamp update.
5031 if (newsize != oldsize) {
5032 inode_inc_iversion(inode);
5033 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5034 inode->i_mtime = current_time(inode);
5035 inode->i_ctime = inode->i_mtime;
5039 if (newsize > oldsize) {
5041 * Don't do an expanding truncate while snapshotting is ongoing.
5042 * This is to ensure the snapshot captures a fully consistent
5043 * state of this file - if the snapshot captures this expanding
5044 * truncation, it must capture all writes that happened before
5047 btrfs_drew_write_lock(&root->snapshot_lock);
5048 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5050 btrfs_drew_write_unlock(&root->snapshot_lock);
5054 trans = btrfs_start_transaction(root, 1);
5055 if (IS_ERR(trans)) {
5056 btrfs_drew_write_unlock(&root->snapshot_lock);
5057 return PTR_ERR(trans);
5060 i_size_write(inode, newsize);
5061 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5062 pagecache_isize_extended(inode, oldsize, newsize);
5063 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5064 btrfs_drew_write_unlock(&root->snapshot_lock);
5065 btrfs_end_transaction(trans);
5067 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5069 if (btrfs_is_zoned(fs_info)) {
5070 ret = btrfs_wait_ordered_range(inode,
5071 ALIGN(newsize, fs_info->sectorsize),
5078 * We're truncating a file that used to have good data down to
5079 * zero. Make sure any new writes to the file get on disk
5083 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5084 &BTRFS_I(inode)->runtime_flags);
5086 truncate_setsize(inode, newsize);
5088 inode_dio_wait(inode);
5090 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5091 if (ret && inode->i_nlink) {
5095 * Truncate failed, so fix up the in-memory size. We
5096 * adjusted disk_i_size down as we removed extents, so
5097 * wait for disk_i_size to be stable and then update the
5098 * in-memory size to match.
5100 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5103 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5110 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5113 struct inode *inode = d_inode(dentry);
5114 struct btrfs_root *root = BTRFS_I(inode)->root;
5117 if (btrfs_root_readonly(root))
5120 err = setattr_prepare(idmap, dentry, attr);
5124 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5125 err = btrfs_setsize(inode, attr);
5130 if (attr->ia_valid) {
5131 setattr_copy(idmap, inode, attr);
5132 inode_inc_iversion(inode);
5133 err = btrfs_dirty_inode(BTRFS_I(inode));
5135 if (!err && attr->ia_valid & ATTR_MODE)
5136 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5143 * While truncating the inode pages during eviction, we get the VFS
5144 * calling btrfs_invalidate_folio() against each folio of the inode. This
5145 * is slow because the calls to btrfs_invalidate_folio() result in a
5146 * huge amount of calls to lock_extent() and clear_extent_bit(),
5147 * which keep merging and splitting extent_state structures over and over,
5148 * wasting lots of time.
5150 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5151 * skip all those expensive operations on a per folio basis and do only
5152 * the ordered io finishing, while we release here the extent_map and
5153 * extent_state structures, without the excessive merging and splitting.
5155 static void evict_inode_truncate_pages(struct inode *inode)
5157 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5158 struct rb_node *node;
5160 ASSERT(inode->i_state & I_FREEING);
5161 truncate_inode_pages_final(&inode->i_data);
5163 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5166 * Keep looping until we have no more ranges in the io tree.
5167 * We can have ongoing bios started by readahead that have
5168 * their endio callback (extent_io.c:end_bio_extent_readpage)
5169 * still in progress (unlocked the pages in the bio but did not yet
5170 * unlocked the ranges in the io tree). Therefore this means some
5171 * ranges can still be locked and eviction started because before
5172 * submitting those bios, which are executed by a separate task (work
5173 * queue kthread), inode references (inode->i_count) were not taken
5174 * (which would be dropped in the end io callback of each bio).
5175 * Therefore here we effectively end up waiting for those bios and
5176 * anyone else holding locked ranges without having bumped the inode's
5177 * reference count - if we don't do it, when they access the inode's
5178 * io_tree to unlock a range it may be too late, leading to an
5179 * use-after-free issue.
5181 spin_lock(&io_tree->lock);
5182 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5183 struct extent_state *state;
5184 struct extent_state *cached_state = NULL;
5187 unsigned state_flags;
5189 node = rb_first(&io_tree->state);
5190 state = rb_entry(node, struct extent_state, rb_node);
5191 start = state->start;
5193 state_flags = state->state;
5194 spin_unlock(&io_tree->lock);
5196 lock_extent(io_tree, start, end, &cached_state);
5199 * If still has DELALLOC flag, the extent didn't reach disk,
5200 * and its reserved space won't be freed by delayed_ref.
5201 * So we need to free its reserved space here.
5202 * (Refer to comment in btrfs_invalidate_folio, case 2)
5204 * Note, end is the bytenr of last byte, so we need + 1 here.
5206 if (state_flags & EXTENT_DELALLOC)
5207 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5210 clear_extent_bit(io_tree, start, end,
5211 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5215 spin_lock(&io_tree->lock);
5217 spin_unlock(&io_tree->lock);
5220 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5221 struct btrfs_block_rsv *rsv)
5223 struct btrfs_fs_info *fs_info = root->fs_info;
5224 struct btrfs_trans_handle *trans;
5225 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5229 * Eviction should be taking place at some place safe because of our
5230 * delayed iputs. However the normal flushing code will run delayed
5231 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5233 * We reserve the delayed_refs_extra here again because we can't use
5234 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5235 * above. We reserve our extra bit here because we generate a ton of
5236 * delayed refs activity by truncating.
5238 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5239 * if we fail to make this reservation we can re-try without the
5240 * delayed_refs_extra so we can make some forward progress.
5242 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5243 BTRFS_RESERVE_FLUSH_EVICT);
5245 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5246 BTRFS_RESERVE_FLUSH_EVICT);
5249 "could not allocate space for delete; will truncate on mount");
5250 return ERR_PTR(-ENOSPC);
5252 delayed_refs_extra = 0;
5255 trans = btrfs_join_transaction(root);
5259 if (delayed_refs_extra) {
5260 trans->block_rsv = &fs_info->trans_block_rsv;
5261 trans->bytes_reserved = delayed_refs_extra;
5262 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5263 delayed_refs_extra, true);
5268 void btrfs_evict_inode(struct inode *inode)
5270 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5271 struct btrfs_trans_handle *trans;
5272 struct btrfs_root *root = BTRFS_I(inode)->root;
5273 struct btrfs_block_rsv *rsv = NULL;
5276 trace_btrfs_inode_evict(inode);
5279 fsverity_cleanup_inode(inode);
5284 evict_inode_truncate_pages(inode);
5286 if (inode->i_nlink &&
5287 ((btrfs_root_refs(&root->root_item) != 0 &&
5288 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5289 btrfs_is_free_space_inode(BTRFS_I(inode))))
5292 if (is_bad_inode(inode))
5295 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5298 if (inode->i_nlink > 0) {
5299 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5300 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5305 * This makes sure the inode item in tree is uptodate and the space for
5306 * the inode update is released.
5308 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5313 * This drops any pending insert or delete operations we have for this
5314 * inode. We could have a delayed dir index deletion queued up, but
5315 * we're removing the inode completely so that'll be taken care of in
5318 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5320 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5323 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5324 rsv->failfast = true;
5326 btrfs_i_size_write(BTRFS_I(inode), 0);
5329 struct btrfs_truncate_control control = {
5330 .inode = BTRFS_I(inode),
5331 .ino = btrfs_ino(BTRFS_I(inode)),
5336 trans = evict_refill_and_join(root, rsv);
5340 trans->block_rsv = rsv;
5342 ret = btrfs_truncate_inode_items(trans, root, &control);
5343 trans->block_rsv = &fs_info->trans_block_rsv;
5344 btrfs_end_transaction(trans);
5346 * We have not added new delayed items for our inode after we
5347 * have flushed its delayed items, so no need to throttle on
5348 * delayed items. However we have modified extent buffers.
5350 btrfs_btree_balance_dirty_nodelay(fs_info);
5351 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5358 * Errors here aren't a big deal, it just means we leave orphan items in
5359 * the tree. They will be cleaned up on the next mount. If the inode
5360 * number gets reused, cleanup deletes the orphan item without doing
5361 * anything, and unlink reuses the existing orphan item.
5363 * If it turns out that we are dropping too many of these, we might want
5364 * to add a mechanism for retrying these after a commit.
5366 trans = evict_refill_and_join(root, rsv);
5367 if (!IS_ERR(trans)) {
5368 trans->block_rsv = rsv;
5369 btrfs_orphan_del(trans, BTRFS_I(inode));
5370 trans->block_rsv = &fs_info->trans_block_rsv;
5371 btrfs_end_transaction(trans);
5375 btrfs_free_block_rsv(fs_info, rsv);
5377 * If we didn't successfully delete, the orphan item will still be in
5378 * the tree and we'll retry on the next mount. Again, we might also want
5379 * to retry these periodically in the future.
5381 btrfs_remove_delayed_node(BTRFS_I(inode));
5382 fsverity_cleanup_inode(inode);
5387 * Return the key found in the dir entry in the location pointer, fill @type
5388 * with BTRFS_FT_*, and return 0.
5390 * If no dir entries were found, returns -ENOENT.
5391 * If found a corrupted location in dir entry, returns -EUCLEAN.
5393 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5394 struct btrfs_key *location, u8 *type)
5396 struct btrfs_dir_item *di;
5397 struct btrfs_path *path;
5398 struct btrfs_root *root = dir->root;
5400 struct fscrypt_name fname;
5402 path = btrfs_alloc_path();
5406 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5410 * fscrypt_setup_filename() should never return a positive value, but
5411 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5415 /* This needs to handle no-key deletions later on */
5417 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5418 &fname.disk_name, 0);
5419 if (IS_ERR_OR_NULL(di)) {
5420 ret = di ? PTR_ERR(di) : -ENOENT;
5424 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5425 if (location->type != BTRFS_INODE_ITEM_KEY &&
5426 location->type != BTRFS_ROOT_ITEM_KEY) {
5428 btrfs_warn(root->fs_info,
5429 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5430 __func__, fname.disk_name.name, btrfs_ino(dir),
5431 location->objectid, location->type, location->offset);
5434 *type = btrfs_dir_ftype(path->nodes[0], di);
5436 fscrypt_free_filename(&fname);
5437 btrfs_free_path(path);
5442 * when we hit a tree root in a directory, the btrfs part of the inode
5443 * needs to be changed to reflect the root directory of the tree root. This
5444 * is kind of like crossing a mount point.
5446 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5447 struct btrfs_inode *dir,
5448 struct dentry *dentry,
5449 struct btrfs_key *location,
5450 struct btrfs_root **sub_root)
5452 struct btrfs_path *path;
5453 struct btrfs_root *new_root;
5454 struct btrfs_root_ref *ref;
5455 struct extent_buffer *leaf;
5456 struct btrfs_key key;
5459 struct fscrypt_name fname;
5461 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5465 path = btrfs_alloc_path();
5472 key.objectid = dir->root->root_key.objectid;
5473 key.type = BTRFS_ROOT_REF_KEY;
5474 key.offset = location->objectid;
5476 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5483 leaf = path->nodes[0];
5484 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5485 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5486 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5489 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5490 (unsigned long)(ref + 1), fname.disk_name.len);
5494 btrfs_release_path(path);
5496 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5497 if (IS_ERR(new_root)) {
5498 err = PTR_ERR(new_root);
5502 *sub_root = new_root;
5503 location->objectid = btrfs_root_dirid(&new_root->root_item);
5504 location->type = BTRFS_INODE_ITEM_KEY;
5505 location->offset = 0;
5508 btrfs_free_path(path);
5509 fscrypt_free_filename(&fname);
5513 static void inode_tree_add(struct btrfs_inode *inode)
5515 struct btrfs_root *root = inode->root;
5516 struct btrfs_inode *entry;
5518 struct rb_node *parent;
5519 struct rb_node *new = &inode->rb_node;
5520 u64 ino = btrfs_ino(inode);
5522 if (inode_unhashed(&inode->vfs_inode))
5525 spin_lock(&root->inode_lock);
5526 p = &root->inode_tree.rb_node;
5529 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5531 if (ino < btrfs_ino(entry))
5532 p = &parent->rb_left;
5533 else if (ino > btrfs_ino(entry))
5534 p = &parent->rb_right;
5536 WARN_ON(!(entry->vfs_inode.i_state &
5537 (I_WILL_FREE | I_FREEING)));
5538 rb_replace_node(parent, new, &root->inode_tree);
5539 RB_CLEAR_NODE(parent);
5540 spin_unlock(&root->inode_lock);
5544 rb_link_node(new, parent, p);
5545 rb_insert_color(new, &root->inode_tree);
5546 spin_unlock(&root->inode_lock);
5549 static void inode_tree_del(struct btrfs_inode *inode)
5551 struct btrfs_root *root = inode->root;
5554 spin_lock(&root->inode_lock);
5555 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5556 rb_erase(&inode->rb_node, &root->inode_tree);
5557 RB_CLEAR_NODE(&inode->rb_node);
5558 empty = RB_EMPTY_ROOT(&root->inode_tree);
5560 spin_unlock(&root->inode_lock);
5562 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5563 spin_lock(&root->inode_lock);
5564 empty = RB_EMPTY_ROOT(&root->inode_tree);
5565 spin_unlock(&root->inode_lock);
5567 btrfs_add_dead_root(root);
5572 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5574 struct btrfs_iget_args *args = p;
5576 inode->i_ino = args->ino;
5577 BTRFS_I(inode)->location.objectid = args->ino;
5578 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5579 BTRFS_I(inode)->location.offset = 0;
5580 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5581 BUG_ON(args->root && !BTRFS_I(inode)->root);
5583 if (args->root && args->root == args->root->fs_info->tree_root &&
5584 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5585 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5586 &BTRFS_I(inode)->runtime_flags);
5590 static int btrfs_find_actor(struct inode *inode, void *opaque)
5592 struct btrfs_iget_args *args = opaque;
5594 return args->ino == BTRFS_I(inode)->location.objectid &&
5595 args->root == BTRFS_I(inode)->root;
5598 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5599 struct btrfs_root *root)
5601 struct inode *inode;
5602 struct btrfs_iget_args args;
5603 unsigned long hashval = btrfs_inode_hash(ino, root);
5608 inode = iget5_locked(s, hashval, btrfs_find_actor,
5609 btrfs_init_locked_inode,
5615 * Get an inode object given its inode number and corresponding root.
5616 * Path can be preallocated to prevent recursing back to iget through
5617 * allocator. NULL is also valid but may require an additional allocation
5620 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5621 struct btrfs_root *root, struct btrfs_path *path)
5623 struct inode *inode;
5625 inode = btrfs_iget_locked(s, ino, root);
5627 return ERR_PTR(-ENOMEM);
5629 if (inode->i_state & I_NEW) {
5632 ret = btrfs_read_locked_inode(inode, path);
5634 inode_tree_add(BTRFS_I(inode));
5635 unlock_new_inode(inode);
5639 * ret > 0 can come from btrfs_search_slot called by
5640 * btrfs_read_locked_inode, this means the inode item
5645 inode = ERR_PTR(ret);
5652 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5654 return btrfs_iget_path(s, ino, root, NULL);
5657 static struct inode *new_simple_dir(struct super_block *s,
5658 struct btrfs_key *key,
5659 struct btrfs_root *root)
5661 struct inode *inode = new_inode(s);
5664 return ERR_PTR(-ENOMEM);
5666 BTRFS_I(inode)->root = btrfs_grab_root(root);
5667 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5668 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5670 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5672 * We only need lookup, the rest is read-only and there's no inode
5673 * associated with the dentry
5675 inode->i_op = &simple_dir_inode_operations;
5676 inode->i_opflags &= ~IOP_XATTR;
5677 inode->i_fop = &simple_dir_operations;
5678 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5679 inode->i_mtime = current_time(inode);
5680 inode->i_atime = inode->i_mtime;
5681 inode->i_ctime = inode->i_mtime;
5682 BTRFS_I(inode)->i_otime = inode->i_mtime;
5687 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5688 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5689 static_assert(BTRFS_FT_DIR == FT_DIR);
5690 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5691 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5692 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5693 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5694 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5696 static inline u8 btrfs_inode_type(struct inode *inode)
5698 return fs_umode_to_ftype(inode->i_mode);
5701 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5703 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5704 struct inode *inode;
5705 struct btrfs_root *root = BTRFS_I(dir)->root;
5706 struct btrfs_root *sub_root = root;
5707 struct btrfs_key location;
5711 if (dentry->d_name.len > BTRFS_NAME_LEN)
5712 return ERR_PTR(-ENAMETOOLONG);
5714 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5716 return ERR_PTR(ret);
5718 if (location.type == BTRFS_INODE_ITEM_KEY) {
5719 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5723 /* Do extra check against inode mode with di_type */
5724 if (btrfs_inode_type(inode) != di_type) {
5726 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5727 inode->i_mode, btrfs_inode_type(inode),
5730 return ERR_PTR(-EUCLEAN);
5735 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5736 &location, &sub_root);
5739 inode = ERR_PTR(ret);
5741 inode = new_simple_dir(dir->i_sb, &location, root);
5743 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5744 btrfs_put_root(sub_root);
5749 down_read(&fs_info->cleanup_work_sem);
5750 if (!sb_rdonly(inode->i_sb))
5751 ret = btrfs_orphan_cleanup(sub_root);
5752 up_read(&fs_info->cleanup_work_sem);
5755 inode = ERR_PTR(ret);
5762 static int btrfs_dentry_delete(const struct dentry *dentry)
5764 struct btrfs_root *root;
5765 struct inode *inode = d_inode(dentry);
5767 if (!inode && !IS_ROOT(dentry))
5768 inode = d_inode(dentry->d_parent);
5771 root = BTRFS_I(inode)->root;
5772 if (btrfs_root_refs(&root->root_item) == 0)
5775 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5781 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5784 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5786 if (inode == ERR_PTR(-ENOENT))
5788 return d_splice_alias(inode, dentry);
5792 * Find the highest existing sequence number in a directory and then set the
5793 * in-memory index_cnt variable to the first free sequence number.
5795 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5797 struct btrfs_root *root = inode->root;
5798 struct btrfs_key key, found_key;
5799 struct btrfs_path *path;
5800 struct extent_buffer *leaf;
5803 key.objectid = btrfs_ino(inode);
5804 key.type = BTRFS_DIR_INDEX_KEY;
5805 key.offset = (u64)-1;
5807 path = btrfs_alloc_path();
5811 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5814 /* FIXME: we should be able to handle this */
5819 if (path->slots[0] == 0) {
5820 inode->index_cnt = BTRFS_DIR_START_INDEX;
5826 leaf = path->nodes[0];
5827 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5829 if (found_key.objectid != btrfs_ino(inode) ||
5830 found_key.type != BTRFS_DIR_INDEX_KEY) {
5831 inode->index_cnt = BTRFS_DIR_START_INDEX;
5835 inode->index_cnt = found_key.offset + 1;
5837 btrfs_free_path(path);
5841 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5843 if (dir->index_cnt == (u64)-1) {
5846 ret = btrfs_inode_delayed_dir_index_count(dir);
5848 ret = btrfs_set_inode_index_count(dir);
5854 *index = dir->index_cnt;
5860 * All this infrastructure exists because dir_emit can fault, and we are holding
5861 * the tree lock when doing readdir. For now just allocate a buffer and copy
5862 * our information into that, and then dir_emit from the buffer. This is
5863 * similar to what NFS does, only we don't keep the buffer around in pagecache
5864 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5865 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5868 static int btrfs_opendir(struct inode *inode, struct file *file)
5870 struct btrfs_file_private *private;
5874 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5878 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5881 private->last_index = last_index;
5882 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5883 if (!private->filldir_buf) {
5887 file->private_data = private;
5898 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5901 struct dir_entry *entry = addr;
5902 char *name = (char *)(entry + 1);
5904 ctx->pos = get_unaligned(&entry->offset);
5905 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5906 get_unaligned(&entry->ino),
5907 get_unaligned(&entry->type)))
5909 addr += sizeof(struct dir_entry) +
5910 get_unaligned(&entry->name_len);
5916 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5918 struct inode *inode = file_inode(file);
5919 struct btrfs_root *root = BTRFS_I(inode)->root;
5920 struct btrfs_file_private *private = file->private_data;
5921 struct btrfs_dir_item *di;
5922 struct btrfs_key key;
5923 struct btrfs_key found_key;
5924 struct btrfs_path *path;
5926 struct list_head ins_list;
5927 struct list_head del_list;
5934 struct btrfs_key location;
5936 if (!dir_emit_dots(file, ctx))
5939 path = btrfs_alloc_path();
5943 addr = private->filldir_buf;
5944 path->reada = READA_FORWARD;
5946 INIT_LIST_HEAD(&ins_list);
5947 INIT_LIST_HEAD(&del_list);
5948 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5949 &ins_list, &del_list);
5952 key.type = BTRFS_DIR_INDEX_KEY;
5953 key.offset = ctx->pos;
5954 key.objectid = btrfs_ino(BTRFS_I(inode));
5956 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5957 struct dir_entry *entry;
5958 struct extent_buffer *leaf = path->nodes[0];
5961 if (found_key.objectid != key.objectid)
5963 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5965 if (found_key.offset < ctx->pos)
5967 if (found_key.offset > private->last_index)
5969 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5971 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5972 name_len = btrfs_dir_name_len(leaf, di);
5973 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5975 btrfs_release_path(path);
5976 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5979 addr = private->filldir_buf;
5985 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5987 name_ptr = (char *)(entry + 1);
5988 read_extent_buffer(leaf, name_ptr,
5989 (unsigned long)(di + 1), name_len);
5990 put_unaligned(name_len, &entry->name_len);
5991 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5992 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5993 put_unaligned(location.objectid, &entry->ino);
5994 put_unaligned(found_key.offset, &entry->offset);
5996 addr += sizeof(struct dir_entry) + name_len;
5997 total_len += sizeof(struct dir_entry) + name_len;
5999 /* Catch error encountered during iteration */
6003 btrfs_release_path(path);
6005 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6009 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6014 * Stop new entries from being returned after we return the last
6017 * New directory entries are assigned a strictly increasing
6018 * offset. This means that new entries created during readdir
6019 * are *guaranteed* to be seen in the future by that readdir.
6020 * This has broken buggy programs which operate on names as
6021 * they're returned by readdir. Until we re-use freed offsets
6022 * we have this hack to stop new entries from being returned
6023 * under the assumption that they'll never reach this huge
6026 * This is being careful not to overflow 32bit loff_t unless the
6027 * last entry requires it because doing so has broken 32bit apps
6030 if (ctx->pos >= INT_MAX)
6031 ctx->pos = LLONG_MAX;
6038 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6039 btrfs_free_path(path);
6044 * This is somewhat expensive, updating the tree every time the
6045 * inode changes. But, it is most likely to find the inode in cache.
6046 * FIXME, needs more benchmarking...there are no reasons other than performance
6047 * to keep or drop this code.
6049 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6051 struct btrfs_root *root = inode->root;
6052 struct btrfs_fs_info *fs_info = root->fs_info;
6053 struct btrfs_trans_handle *trans;
6056 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6059 trans = btrfs_join_transaction(root);
6061 return PTR_ERR(trans);
6063 ret = btrfs_update_inode(trans, root, inode);
6064 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6065 /* whoops, lets try again with the full transaction */
6066 btrfs_end_transaction(trans);
6067 trans = btrfs_start_transaction(root, 1);
6069 return PTR_ERR(trans);
6071 ret = btrfs_update_inode(trans, root, inode);
6073 btrfs_end_transaction(trans);
6074 if (inode->delayed_node)
6075 btrfs_balance_delayed_items(fs_info);
6081 * This is a copy of file_update_time. We need this so we can return error on
6082 * ENOSPC for updating the inode in the case of file write and mmap writes.
6084 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6087 struct btrfs_root *root = BTRFS_I(inode)->root;
6088 bool dirty = flags & ~S_VERSION;
6090 if (btrfs_root_readonly(root))
6093 if (flags & S_VERSION)
6094 dirty |= inode_maybe_inc_iversion(inode, dirty);
6095 if (flags & S_CTIME)
6096 inode->i_ctime = *now;
6097 if (flags & S_MTIME)
6098 inode->i_mtime = *now;
6099 if (flags & S_ATIME)
6100 inode->i_atime = *now;
6101 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6105 * helper to find a free sequence number in a given directory. This current
6106 * code is very simple, later versions will do smarter things in the btree
6108 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6112 if (dir->index_cnt == (u64)-1) {
6113 ret = btrfs_inode_delayed_dir_index_count(dir);
6115 ret = btrfs_set_inode_index_count(dir);
6121 *index = dir->index_cnt;
6127 static int btrfs_insert_inode_locked(struct inode *inode)
6129 struct btrfs_iget_args args;
6131 args.ino = BTRFS_I(inode)->location.objectid;
6132 args.root = BTRFS_I(inode)->root;
6134 return insert_inode_locked4(inode,
6135 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6136 btrfs_find_actor, &args);
6139 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6140 unsigned int *trans_num_items)
6142 struct inode *dir = args->dir;
6143 struct inode *inode = args->inode;
6146 if (!args->orphan) {
6147 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6153 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6155 fscrypt_free_filename(&args->fname);
6159 /* 1 to add inode item */
6160 *trans_num_items = 1;
6161 /* 1 to add compression property */
6162 if (BTRFS_I(dir)->prop_compress)
6163 (*trans_num_items)++;
6164 /* 1 to add default ACL xattr */
6165 if (args->default_acl)
6166 (*trans_num_items)++;
6167 /* 1 to add access ACL xattr */
6169 (*trans_num_items)++;
6170 #ifdef CONFIG_SECURITY
6171 /* 1 to add LSM xattr */
6172 if (dir->i_security)
6173 (*trans_num_items)++;
6176 /* 1 to add orphan item */
6177 (*trans_num_items)++;
6181 * 1 to add dir index
6182 * 1 to update parent inode item
6184 * No need for 1 unit for the inode ref item because it is
6185 * inserted in a batch together with the inode item at
6186 * btrfs_create_new_inode().
6188 *trans_num_items += 3;
6193 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6195 posix_acl_release(args->acl);
6196 posix_acl_release(args->default_acl);
6197 fscrypt_free_filename(&args->fname);
6201 * Inherit flags from the parent inode.
6203 * Currently only the compression flags and the cow flags are inherited.
6205 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6211 if (flags & BTRFS_INODE_NOCOMPRESS) {
6212 inode->flags &= ~BTRFS_INODE_COMPRESS;
6213 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6214 } else if (flags & BTRFS_INODE_COMPRESS) {
6215 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6216 inode->flags |= BTRFS_INODE_COMPRESS;
6219 if (flags & BTRFS_INODE_NODATACOW) {
6220 inode->flags |= BTRFS_INODE_NODATACOW;
6221 if (S_ISREG(inode->vfs_inode.i_mode))
6222 inode->flags |= BTRFS_INODE_NODATASUM;
6225 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6228 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6229 struct btrfs_new_inode_args *args)
6231 struct inode *dir = args->dir;
6232 struct inode *inode = args->inode;
6233 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6234 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6235 struct btrfs_root *root;
6236 struct btrfs_inode_item *inode_item;
6237 struct btrfs_key *location;
6238 struct btrfs_path *path;
6240 struct btrfs_inode_ref *ref;
6241 struct btrfs_key key[2];
6243 struct btrfs_item_batch batch;
6247 path = btrfs_alloc_path();
6252 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6253 root = BTRFS_I(inode)->root;
6255 ret = btrfs_get_free_objectid(root, &objectid);
6258 inode->i_ino = objectid;
6262 * O_TMPFILE, set link count to 0, so that after this point, we
6263 * fill in an inode item with the correct link count.
6265 set_nlink(inode, 0);
6267 trace_btrfs_inode_request(dir);
6269 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6273 /* index_cnt is ignored for everything but a dir. */
6274 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6275 BTRFS_I(inode)->generation = trans->transid;
6276 inode->i_generation = BTRFS_I(inode)->generation;
6279 * Subvolumes don't inherit flags from their parent directory.
6280 * Originally this was probably by accident, but we probably can't
6281 * change it now without compatibility issues.
6284 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6286 if (S_ISREG(inode->i_mode)) {
6287 if (btrfs_test_opt(fs_info, NODATASUM))
6288 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6289 if (btrfs_test_opt(fs_info, NODATACOW))
6290 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6291 BTRFS_INODE_NODATASUM;
6294 location = &BTRFS_I(inode)->location;
6295 location->objectid = objectid;
6296 location->offset = 0;
6297 location->type = BTRFS_INODE_ITEM_KEY;
6299 ret = btrfs_insert_inode_locked(inode);
6302 BTRFS_I(dir)->index_cnt--;
6307 * We could have gotten an inode number from somebody who was fsynced
6308 * and then removed in this same transaction, so let's just set full
6309 * sync since it will be a full sync anyway and this will blow away the
6310 * old info in the log.
6312 btrfs_set_inode_full_sync(BTRFS_I(inode));
6314 key[0].objectid = objectid;
6315 key[0].type = BTRFS_INODE_ITEM_KEY;
6318 sizes[0] = sizeof(struct btrfs_inode_item);
6320 if (!args->orphan) {
6322 * Start new inodes with an inode_ref. This is slightly more
6323 * efficient for small numbers of hard links since they will
6324 * be packed into one item. Extended refs will kick in if we
6325 * add more hard links than can fit in the ref item.
6327 key[1].objectid = objectid;
6328 key[1].type = BTRFS_INODE_REF_KEY;
6330 key[1].offset = objectid;
6331 sizes[1] = 2 + sizeof(*ref);
6333 key[1].offset = btrfs_ino(BTRFS_I(dir));
6334 sizes[1] = name->len + sizeof(*ref);
6338 batch.keys = &key[0];
6339 batch.data_sizes = &sizes[0];
6340 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6341 batch.nr = args->orphan ? 1 : 2;
6342 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6344 btrfs_abort_transaction(trans, ret);
6348 inode->i_mtime = current_time(inode);
6349 inode->i_atime = inode->i_mtime;
6350 inode->i_ctime = inode->i_mtime;
6351 BTRFS_I(inode)->i_otime = inode->i_mtime;
6354 * We're going to fill the inode item now, so at this point the inode
6355 * must be fully initialized.
6358 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6359 struct btrfs_inode_item);
6360 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6361 sizeof(*inode_item));
6362 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6364 if (!args->orphan) {
6365 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6366 struct btrfs_inode_ref);
6367 ptr = (unsigned long)(ref + 1);
6369 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6370 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6371 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6373 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6375 btrfs_set_inode_ref_index(path->nodes[0], ref,
6376 BTRFS_I(inode)->dir_index);
6377 write_extent_buffer(path->nodes[0], name->name, ptr,
6382 btrfs_mark_buffer_dirty(path->nodes[0]);
6384 * We don't need the path anymore, plus inheriting properties, adding
6385 * ACLs, security xattrs, orphan item or adding the link, will result in
6386 * allocating yet another path. So just free our path.
6388 btrfs_free_path(path);
6392 struct inode *parent;
6395 * Subvolumes inherit properties from their parent subvolume,
6396 * not the directory they were created in.
6398 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6399 BTRFS_I(dir)->root);
6400 if (IS_ERR(parent)) {
6401 ret = PTR_ERR(parent);
6403 ret = btrfs_inode_inherit_props(trans, inode, parent);
6407 ret = btrfs_inode_inherit_props(trans, inode, dir);
6411 "error inheriting props for ino %llu (root %llu): %d",
6412 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6417 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6420 if (!args->subvol) {
6421 ret = btrfs_init_inode_security(trans, args);
6423 btrfs_abort_transaction(trans, ret);
6428 inode_tree_add(BTRFS_I(inode));
6430 trace_btrfs_inode_new(inode);
6431 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6433 btrfs_update_root_times(trans, root);
6436 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6438 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6439 0, BTRFS_I(inode)->dir_index);
6442 btrfs_abort_transaction(trans, ret);
6450 * discard_new_inode() calls iput(), but the caller owns the reference
6454 discard_new_inode(inode);
6456 btrfs_free_path(path);
6461 * utility function to add 'inode' into 'parent_inode' with
6462 * a give name and a given sequence number.
6463 * if 'add_backref' is true, also insert a backref from the
6464 * inode to the parent directory.
6466 int btrfs_add_link(struct btrfs_trans_handle *trans,
6467 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6468 const struct fscrypt_str *name, int add_backref, u64 index)
6471 struct btrfs_key key;
6472 struct btrfs_root *root = parent_inode->root;
6473 u64 ino = btrfs_ino(inode);
6474 u64 parent_ino = btrfs_ino(parent_inode);
6476 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6477 memcpy(&key, &inode->root->root_key, sizeof(key));
6480 key.type = BTRFS_INODE_ITEM_KEY;
6484 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6485 ret = btrfs_add_root_ref(trans, key.objectid,
6486 root->root_key.objectid, parent_ino,
6488 } else if (add_backref) {
6489 ret = btrfs_insert_inode_ref(trans, root, name,
6490 ino, parent_ino, index);
6493 /* Nothing to clean up yet */
6497 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6498 btrfs_inode_type(&inode->vfs_inode), index);
6499 if (ret == -EEXIST || ret == -EOVERFLOW)
6502 btrfs_abort_transaction(trans, ret);
6506 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6508 inode_inc_iversion(&parent_inode->vfs_inode);
6510 * If we are replaying a log tree, we do not want to update the mtime
6511 * and ctime of the parent directory with the current time, since the
6512 * log replay procedure is responsible for setting them to their correct
6513 * values (the ones it had when the fsync was done).
6515 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6516 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6518 parent_inode->vfs_inode.i_mtime = now;
6519 parent_inode->vfs_inode.i_ctime = now;
6521 ret = btrfs_update_inode(trans, root, parent_inode);
6523 btrfs_abort_transaction(trans, ret);
6527 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6530 err = btrfs_del_root_ref(trans, key.objectid,
6531 root->root_key.objectid, parent_ino,
6532 &local_index, name);
6534 btrfs_abort_transaction(trans, err);
6535 } else if (add_backref) {
6539 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6542 btrfs_abort_transaction(trans, err);
6545 /* Return the original error code */
6549 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6550 struct inode *inode)
6552 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6553 struct btrfs_root *root = BTRFS_I(dir)->root;
6554 struct btrfs_new_inode_args new_inode_args = {
6559 unsigned int trans_num_items;
6560 struct btrfs_trans_handle *trans;
6563 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6567 trans = btrfs_start_transaction(root, trans_num_items);
6568 if (IS_ERR(trans)) {
6569 err = PTR_ERR(trans);
6570 goto out_new_inode_args;
6573 err = btrfs_create_new_inode(trans, &new_inode_args);
6575 d_instantiate_new(dentry, inode);
6577 btrfs_end_transaction(trans);
6578 btrfs_btree_balance_dirty(fs_info);
6580 btrfs_new_inode_args_destroy(&new_inode_args);
6587 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6588 struct dentry *dentry, umode_t mode, dev_t rdev)
6590 struct inode *inode;
6592 inode = new_inode(dir->i_sb);
6595 inode_init_owner(idmap, inode, dir, mode);
6596 inode->i_op = &btrfs_special_inode_operations;
6597 init_special_inode(inode, inode->i_mode, rdev);
6598 return btrfs_create_common(dir, dentry, inode);
6601 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6602 struct dentry *dentry, umode_t mode, bool excl)
6604 struct inode *inode;
6606 inode = new_inode(dir->i_sb);
6609 inode_init_owner(idmap, inode, dir, mode);
6610 inode->i_fop = &btrfs_file_operations;
6611 inode->i_op = &btrfs_file_inode_operations;
6612 inode->i_mapping->a_ops = &btrfs_aops;
6613 return btrfs_create_common(dir, dentry, inode);
6616 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6617 struct dentry *dentry)
6619 struct btrfs_trans_handle *trans = NULL;
6620 struct btrfs_root *root = BTRFS_I(dir)->root;
6621 struct inode *inode = d_inode(old_dentry);
6622 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6623 struct fscrypt_name fname;
6628 /* do not allow sys_link's with other subvols of the same device */
6629 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6632 if (inode->i_nlink >= BTRFS_LINK_MAX)
6635 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6639 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6644 * 2 items for inode and inode ref
6645 * 2 items for dir items
6646 * 1 item for parent inode
6647 * 1 item for orphan item deletion if O_TMPFILE
6649 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6650 if (IS_ERR(trans)) {
6651 err = PTR_ERR(trans);
6656 /* There are several dir indexes for this inode, clear the cache. */
6657 BTRFS_I(inode)->dir_index = 0ULL;
6659 inode_inc_iversion(inode);
6660 inode->i_ctime = current_time(inode);
6662 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6664 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6665 &fname.disk_name, 1, index);
6670 struct dentry *parent = dentry->d_parent;
6672 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6675 if (inode->i_nlink == 1) {
6677 * If new hard link count is 1, it's a file created
6678 * with open(2) O_TMPFILE flag.
6680 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6684 d_instantiate(dentry, inode);
6685 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6689 fscrypt_free_filename(&fname);
6691 btrfs_end_transaction(trans);
6693 inode_dec_link_count(inode);
6696 btrfs_btree_balance_dirty(fs_info);
6700 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6701 struct dentry *dentry, umode_t mode)
6703 struct inode *inode;
6705 inode = new_inode(dir->i_sb);
6708 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6709 inode->i_op = &btrfs_dir_inode_operations;
6710 inode->i_fop = &btrfs_dir_file_operations;
6711 return btrfs_create_common(dir, dentry, inode);
6714 static noinline int uncompress_inline(struct btrfs_path *path,
6716 struct btrfs_file_extent_item *item)
6719 struct extent_buffer *leaf = path->nodes[0];
6722 unsigned long inline_size;
6726 compress_type = btrfs_file_extent_compression(leaf, item);
6727 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6728 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6729 tmp = kmalloc(inline_size, GFP_NOFS);
6732 ptr = btrfs_file_extent_inline_start(item);
6734 read_extent_buffer(leaf, tmp, ptr, inline_size);
6736 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6737 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6740 * decompression code contains a memset to fill in any space between the end
6741 * of the uncompressed data and the end of max_size in case the decompressed
6742 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6743 * the end of an inline extent and the beginning of the next block, so we
6744 * cover that region here.
6747 if (max_size < PAGE_SIZE)
6748 memzero_page(page, max_size, PAGE_SIZE - max_size);
6753 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6756 struct btrfs_file_extent_item *fi;
6760 if (!page || PageUptodate(page))
6763 ASSERT(page_offset(page) == 0);
6765 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6766 struct btrfs_file_extent_item);
6767 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6768 return uncompress_inline(path, page, fi);
6770 copy_size = min_t(u64, PAGE_SIZE,
6771 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6772 kaddr = kmap_local_page(page);
6773 read_extent_buffer(path->nodes[0], kaddr,
6774 btrfs_file_extent_inline_start(fi), copy_size);
6775 kunmap_local(kaddr);
6776 if (copy_size < PAGE_SIZE)
6777 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6782 * Lookup the first extent overlapping a range in a file.
6784 * @inode: file to search in
6785 * @page: page to read extent data into if the extent is inline
6786 * @pg_offset: offset into @page to copy to
6787 * @start: file offset
6788 * @len: length of range starting at @start
6790 * Return the first &struct extent_map which overlaps the given range, reading
6791 * it from the B-tree and caching it if necessary. Note that there may be more
6792 * extents which overlap the given range after the returned extent_map.
6794 * If @page is not NULL and the extent is inline, this also reads the extent
6795 * data directly into the page and marks the extent up to date in the io_tree.
6797 * Return: ERR_PTR on error, non-NULL extent_map on success.
6799 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6800 struct page *page, size_t pg_offset,
6803 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6805 u64 extent_start = 0;
6807 u64 objectid = btrfs_ino(inode);
6808 int extent_type = -1;
6809 struct btrfs_path *path = NULL;
6810 struct btrfs_root *root = inode->root;
6811 struct btrfs_file_extent_item *item;
6812 struct extent_buffer *leaf;
6813 struct btrfs_key found_key;
6814 struct extent_map *em = NULL;
6815 struct extent_map_tree *em_tree = &inode->extent_tree;
6817 read_lock(&em_tree->lock);
6818 em = lookup_extent_mapping(em_tree, start, len);
6819 read_unlock(&em_tree->lock);
6822 if (em->start > start || em->start + em->len <= start)
6823 free_extent_map(em);
6824 else if (em->block_start == EXTENT_MAP_INLINE && page)
6825 free_extent_map(em);
6829 em = alloc_extent_map();
6834 em->start = EXTENT_MAP_HOLE;
6835 em->orig_start = EXTENT_MAP_HOLE;
6837 em->block_len = (u64)-1;
6839 path = btrfs_alloc_path();
6845 /* Chances are we'll be called again, so go ahead and do readahead */
6846 path->reada = READA_FORWARD;
6849 * The same explanation in load_free_space_cache applies here as well,
6850 * we only read when we're loading the free space cache, and at that
6851 * point the commit_root has everything we need.
6853 if (btrfs_is_free_space_inode(inode)) {
6854 path->search_commit_root = 1;
6855 path->skip_locking = 1;
6858 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6861 } else if (ret > 0) {
6862 if (path->slots[0] == 0)
6868 leaf = path->nodes[0];
6869 item = btrfs_item_ptr(leaf, path->slots[0],
6870 struct btrfs_file_extent_item);
6871 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6872 if (found_key.objectid != objectid ||
6873 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6875 * If we backup past the first extent we want to move forward
6876 * and see if there is an extent in front of us, otherwise we'll
6877 * say there is a hole for our whole search range which can
6884 extent_type = btrfs_file_extent_type(leaf, item);
6885 extent_start = found_key.offset;
6886 extent_end = btrfs_file_extent_end(path);
6887 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6888 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6889 /* Only regular file could have regular/prealloc extent */
6890 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6893 "regular/prealloc extent found for non-regular inode %llu",
6897 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6899 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6900 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6905 if (start >= extent_end) {
6907 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6908 ret = btrfs_next_leaf(root, path);
6914 leaf = path->nodes[0];
6916 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6917 if (found_key.objectid != objectid ||
6918 found_key.type != BTRFS_EXTENT_DATA_KEY)
6920 if (start + len <= found_key.offset)
6922 if (start > found_key.offset)
6925 /* New extent overlaps with existing one */
6927 em->orig_start = start;
6928 em->len = found_key.offset - start;
6929 em->block_start = EXTENT_MAP_HOLE;
6933 btrfs_extent_item_to_extent_map(inode, path, item, em);
6935 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6936 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6938 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6940 * Inline extent can only exist at file offset 0. This is
6941 * ensured by tree-checker and inline extent creation path.
6942 * Thus all members representing file offsets should be zero.
6944 ASSERT(pg_offset == 0);
6945 ASSERT(extent_start == 0);
6946 ASSERT(em->start == 0);
6949 * btrfs_extent_item_to_extent_map() should have properly
6950 * initialized em members already.
6952 * Other members are not utilized for inline extents.
6954 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6955 ASSERT(em->len == fs_info->sectorsize);
6957 ret = read_inline_extent(inode, path, page);
6964 em->orig_start = start;
6966 em->block_start = EXTENT_MAP_HOLE;
6969 btrfs_release_path(path);
6970 if (em->start > start || extent_map_end(em) <= start) {
6972 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6973 em->start, em->len, start, len);
6978 write_lock(&em_tree->lock);
6979 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6980 write_unlock(&em_tree->lock);
6982 btrfs_free_path(path);
6984 trace_btrfs_get_extent(root, inode, em);
6987 free_extent_map(em);
6988 return ERR_PTR(ret);
6993 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6994 struct btrfs_dio_data *dio_data,
6997 const u64 orig_start,
6998 const u64 block_start,
6999 const u64 block_len,
7000 const u64 orig_block_len,
7001 const u64 ram_bytes,
7004 struct extent_map *em = NULL;
7005 struct btrfs_ordered_extent *ordered;
7007 if (type != BTRFS_ORDERED_NOCOW) {
7008 em = create_io_em(inode, start, len, orig_start, block_start,
7009 block_len, orig_block_len, ram_bytes,
7010 BTRFS_COMPRESS_NONE, /* compress_type */
7015 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
7016 block_start, block_len, 0,
7018 (1 << BTRFS_ORDERED_DIRECT),
7019 BTRFS_COMPRESS_NONE);
7020 if (IS_ERR(ordered)) {
7022 free_extent_map(em);
7023 btrfs_drop_extent_map_range(inode, start,
7024 start + len - 1, false);
7026 em = ERR_CAST(ordered);
7028 ASSERT(!dio_data->ordered);
7029 dio_data->ordered = ordered;
7036 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7037 struct btrfs_dio_data *dio_data,
7040 struct btrfs_root *root = inode->root;
7041 struct btrfs_fs_info *fs_info = root->fs_info;
7042 struct extent_map *em;
7043 struct btrfs_key ins;
7047 alloc_hint = get_extent_allocation_hint(inode, start, len);
7048 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7049 0, alloc_hint, &ins, 1, 1);
7051 return ERR_PTR(ret);
7053 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7054 ins.objectid, ins.offset, ins.offset,
7055 ins.offset, BTRFS_ORDERED_REGULAR);
7056 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7058 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7064 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7066 struct btrfs_block_group *block_group;
7067 bool readonly = false;
7069 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7070 if (!block_group || block_group->ro)
7073 btrfs_put_block_group(block_group);
7078 * Check if we can do nocow write into the range [@offset, @offset + @len)
7080 * @offset: File offset
7081 * @len: The length to write, will be updated to the nocow writeable
7083 * @orig_start: (optional) Return the original file offset of the file extent
7084 * @orig_len: (optional) Return the original on-disk length of the file extent
7085 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7086 * @strict: if true, omit optimizations that might force us into unnecessary
7087 * cow. e.g., don't trust generation number.
7090 * >0 and update @len if we can do nocow write
7091 * 0 if we can't do nocow write
7092 * <0 if error happened
7094 * NOTE: This only checks the file extents, caller is responsible to wait for
7095 * any ordered extents.
7097 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7098 u64 *orig_start, u64 *orig_block_len,
7099 u64 *ram_bytes, bool nowait, bool strict)
7101 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7102 struct can_nocow_file_extent_args nocow_args = { 0 };
7103 struct btrfs_path *path;
7105 struct extent_buffer *leaf;
7106 struct btrfs_root *root = BTRFS_I(inode)->root;
7107 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7108 struct btrfs_file_extent_item *fi;
7109 struct btrfs_key key;
7112 path = btrfs_alloc_path();
7115 path->nowait = nowait;
7117 ret = btrfs_lookup_file_extent(NULL, root, path,
7118 btrfs_ino(BTRFS_I(inode)), offset, 0);
7123 if (path->slots[0] == 0) {
7124 /* can't find the item, must cow */
7131 leaf = path->nodes[0];
7132 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7133 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7134 key.type != BTRFS_EXTENT_DATA_KEY) {
7135 /* not our file or wrong item type, must cow */
7139 if (key.offset > offset) {
7140 /* Wrong offset, must cow */
7144 if (btrfs_file_extent_end(path) <= offset)
7147 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7148 found_type = btrfs_file_extent_type(leaf, fi);
7150 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7152 nocow_args.start = offset;
7153 nocow_args.end = offset + *len - 1;
7154 nocow_args.strict = strict;
7155 nocow_args.free_path = true;
7157 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7158 /* can_nocow_file_extent() has freed the path. */
7162 /* Treat errors as not being able to NOCOW. */
7168 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7171 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7172 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7175 range_end = round_up(offset + nocow_args.num_bytes,
7176 root->fs_info->sectorsize) - 1;
7177 ret = test_range_bit(io_tree, offset, range_end,
7178 EXTENT_DELALLOC, 0, NULL);
7186 *orig_start = key.offset - nocow_args.extent_offset;
7188 *orig_block_len = nocow_args.disk_num_bytes;
7190 *len = nocow_args.num_bytes;
7193 btrfs_free_path(path);
7197 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7198 struct extent_state **cached_state,
7199 unsigned int iomap_flags)
7201 const bool writing = (iomap_flags & IOMAP_WRITE);
7202 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7203 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7204 struct btrfs_ordered_extent *ordered;
7209 if (!try_lock_extent(io_tree, lockstart, lockend,
7213 lock_extent(io_tree, lockstart, lockend, cached_state);
7216 * We're concerned with the entire range that we're going to be
7217 * doing DIO to, so we need to make sure there's no ordered
7218 * extents in this range.
7220 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7221 lockend - lockstart + 1);
7224 * We need to make sure there are no buffered pages in this
7225 * range either, we could have raced between the invalidate in
7226 * generic_file_direct_write and locking the extent. The
7227 * invalidate needs to happen so that reads after a write do not
7231 (!writing || !filemap_range_has_page(inode->i_mapping,
7232 lockstart, lockend)))
7235 unlock_extent(io_tree, lockstart, lockend, cached_state);
7239 btrfs_put_ordered_extent(ordered);
7244 * If we are doing a DIO read and the ordered extent we
7245 * found is for a buffered write, we can not wait for it
7246 * to complete and retry, because if we do so we can
7247 * deadlock with concurrent buffered writes on page
7248 * locks. This happens only if our DIO read covers more
7249 * than one extent map, if at this point has already
7250 * created an ordered extent for a previous extent map
7251 * and locked its range in the inode's io tree, and a
7252 * concurrent write against that previous extent map's
7253 * range and this range started (we unlock the ranges
7254 * in the io tree only when the bios complete and
7255 * buffered writes always lock pages before attempting
7256 * to lock range in the io tree).
7259 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7260 btrfs_start_ordered_extent(ordered);
7262 ret = nowait ? -EAGAIN : -ENOTBLK;
7263 btrfs_put_ordered_extent(ordered);
7266 * We could trigger writeback for this range (and wait
7267 * for it to complete) and then invalidate the pages for
7268 * this range (through invalidate_inode_pages2_range()),
7269 * but that can lead us to a deadlock with a concurrent
7270 * call to readahead (a buffered read or a defrag call
7271 * triggered a readahead) on a page lock due to an
7272 * ordered dio extent we created before but did not have
7273 * yet a corresponding bio submitted (whence it can not
7274 * complete), which makes readahead wait for that
7275 * ordered extent to complete while holding a lock on
7278 ret = nowait ? -EAGAIN : -ENOTBLK;
7290 /* The callers of this must take lock_extent() */
7291 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7292 u64 len, u64 orig_start, u64 block_start,
7293 u64 block_len, u64 orig_block_len,
7294 u64 ram_bytes, int compress_type,
7297 struct extent_map *em;
7300 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7301 type == BTRFS_ORDERED_COMPRESSED ||
7302 type == BTRFS_ORDERED_NOCOW ||
7303 type == BTRFS_ORDERED_REGULAR);
7305 em = alloc_extent_map();
7307 return ERR_PTR(-ENOMEM);
7310 em->orig_start = orig_start;
7312 em->block_len = block_len;
7313 em->block_start = block_start;
7314 em->orig_block_len = orig_block_len;
7315 em->ram_bytes = ram_bytes;
7316 em->generation = -1;
7317 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7318 if (type == BTRFS_ORDERED_PREALLOC) {
7319 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7320 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7321 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7322 em->compress_type = compress_type;
7325 ret = btrfs_replace_extent_map_range(inode, em, true);
7327 free_extent_map(em);
7328 return ERR_PTR(ret);
7331 /* em got 2 refs now, callers needs to do free_extent_map once. */
7336 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7337 struct inode *inode,
7338 struct btrfs_dio_data *dio_data,
7339 u64 start, u64 *lenp,
7340 unsigned int iomap_flags)
7342 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7343 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7344 struct extent_map *em = *map;
7346 u64 block_start, orig_start, orig_block_len, ram_bytes;
7347 struct btrfs_block_group *bg;
7348 bool can_nocow = false;
7349 bool space_reserved = false;
7355 * We don't allocate a new extent in the following cases
7357 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7359 * 2) The extent is marked as PREALLOC. We're good to go here and can
7360 * just use the extent.
7363 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7364 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7365 em->block_start != EXTENT_MAP_HOLE)) {
7366 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7367 type = BTRFS_ORDERED_PREALLOC;
7369 type = BTRFS_ORDERED_NOCOW;
7370 len = min(len, em->len - (start - em->start));
7371 block_start = em->block_start + (start - em->start);
7373 if (can_nocow_extent(inode, start, &len, &orig_start,
7374 &orig_block_len, &ram_bytes, false, false) == 1) {
7375 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7383 struct extent_map *em2;
7385 /* We can NOCOW, so only need to reserve metadata space. */
7386 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7389 /* Our caller expects us to free the input extent map. */
7390 free_extent_map(em);
7392 btrfs_dec_nocow_writers(bg);
7393 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7397 space_reserved = true;
7399 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7400 orig_start, block_start,
7401 len, orig_block_len,
7403 btrfs_dec_nocow_writers(bg);
7404 if (type == BTRFS_ORDERED_PREALLOC) {
7405 free_extent_map(em);
7415 dio_data->nocow_done = true;
7417 /* Our caller expects us to free the input extent map. */
7418 free_extent_map(em);
7427 * If we could not allocate data space before locking the file
7428 * range and we can't do a NOCOW write, then we have to fail.
7430 if (!dio_data->data_space_reserved) {
7436 * We have to COW and we have already reserved data space before,
7437 * so now we reserve only metadata.
7439 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7443 space_reserved = true;
7445 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7451 len = min(len, em->len - (start - em->start));
7453 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7454 prev_len - len, true);
7458 * We have created our ordered extent, so we can now release our reservation
7459 * for an outstanding extent.
7461 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7464 * Need to update the i_size under the extent lock so buffered
7465 * readers will get the updated i_size when we unlock.
7467 if (start + len > i_size_read(inode))
7468 i_size_write(inode, start + len);
7470 if (ret && space_reserved) {
7471 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7472 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7478 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7479 loff_t length, unsigned int flags, struct iomap *iomap,
7480 struct iomap *srcmap)
7482 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7483 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7484 struct extent_map *em;
7485 struct extent_state *cached_state = NULL;
7486 struct btrfs_dio_data *dio_data = iter->private;
7487 u64 lockstart, lockend;
7488 const bool write = !!(flags & IOMAP_WRITE);
7491 const u64 data_alloc_len = length;
7492 bool unlock_extents = false;
7495 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7496 * we're NOWAIT we may submit a bio for a partial range and return
7497 * EIOCBQUEUED, which would result in an errant short read.
7499 * The best way to handle this would be to allow for partial completions
7500 * of iocb's, so we could submit the partial bio, return and fault in
7501 * the rest of the pages, and then submit the io for the rest of the
7502 * range. However we don't have that currently, so simply return
7503 * -EAGAIN at this point so that the normal path is used.
7505 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7509 * Cap the size of reads to that usually seen in buffered I/O as we need
7510 * to allocate a contiguous array for the checksums.
7513 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7516 lockend = start + len - 1;
7519 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7520 * enough if we've written compressed pages to this area, so we need to
7521 * flush the dirty pages again to make absolutely sure that any
7522 * outstanding dirty pages are on disk - the first flush only starts
7523 * compression on the data, while keeping the pages locked, so by the
7524 * time the second flush returns we know bios for the compressed pages
7525 * were submitted and finished, and the pages no longer under writeback.
7527 * If we have a NOWAIT request and we have any pages in the range that
7528 * are locked, likely due to compression still in progress, we don't want
7529 * to block on page locks. We also don't want to block on pages marked as
7530 * dirty or under writeback (same as for the non-compression case).
7531 * iomap_dio_rw() did the same check, but after that and before we got
7532 * here, mmap'ed writes may have happened or buffered reads started
7533 * (readpage() and readahead(), which lock pages), as we haven't locked
7534 * the file range yet.
7536 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7537 &BTRFS_I(inode)->runtime_flags)) {
7538 if (flags & IOMAP_NOWAIT) {
7539 if (filemap_range_needs_writeback(inode->i_mapping,
7540 lockstart, lockend))
7543 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7544 start + length - 1);
7550 memset(dio_data, 0, sizeof(*dio_data));
7553 * We always try to allocate data space and must do it before locking
7554 * the file range, to avoid deadlocks with concurrent writes to the same
7555 * range if the range has several extents and the writes don't expand the
7556 * current i_size (the inode lock is taken in shared mode). If we fail to
7557 * allocate data space here we continue and later, after locking the
7558 * file range, we fail with ENOSPC only if we figure out we can not do a
7561 if (write && !(flags & IOMAP_NOWAIT)) {
7562 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7563 &dio_data->data_reserved,
7564 start, data_alloc_len, false);
7566 dio_data->data_space_reserved = true;
7567 else if (ret && !(BTRFS_I(inode)->flags &
7568 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7573 * If this errors out it's because we couldn't invalidate pagecache for
7574 * this range and we need to fallback to buffered IO, or we are doing a
7575 * NOWAIT read/write and we need to block.
7577 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7581 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7588 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7589 * io. INLINE is special, and we could probably kludge it in here, but
7590 * it's still buffered so for safety lets just fall back to the generic
7593 * For COMPRESSED we _have_ to read the entire extent in so we can
7594 * decompress it, so there will be buffering required no matter what we
7595 * do, so go ahead and fallback to buffered.
7597 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7598 * to buffered IO. Don't blame me, this is the price we pay for using
7601 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7602 em->block_start == EXTENT_MAP_INLINE) {
7603 free_extent_map(em);
7605 * If we are in a NOWAIT context, return -EAGAIN in order to
7606 * fallback to buffered IO. This is not only because we can
7607 * block with buffered IO (no support for NOWAIT semantics at
7608 * the moment) but also to avoid returning short reads to user
7609 * space - this happens if we were able to read some data from
7610 * previous non-compressed extents and then when we fallback to
7611 * buffered IO, at btrfs_file_read_iter() by calling
7612 * filemap_read(), we fail to fault in pages for the read buffer,
7613 * in which case filemap_read() returns a short read (the number
7614 * of bytes previously read is > 0, so it does not return -EFAULT).
7616 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7620 len = min(len, em->len - (start - em->start));
7623 * If we have a NOWAIT request and the range contains multiple extents
7624 * (or a mix of extents and holes), then we return -EAGAIN to make the
7625 * caller fallback to a context where it can do a blocking (without
7626 * NOWAIT) request. This way we avoid doing partial IO and returning
7627 * success to the caller, which is not optimal for writes and for reads
7628 * it can result in unexpected behaviour for an application.
7630 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7631 * iomap_dio_rw(), we can end up returning less data then what the caller
7632 * asked for, resulting in an unexpected, and incorrect, short read.
7633 * That is, the caller asked to read N bytes and we return less than that,
7634 * which is wrong unless we are crossing EOF. This happens if we get a
7635 * page fault error when trying to fault in pages for the buffer that is
7636 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7637 * have previously submitted bios for other extents in the range, in
7638 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7639 * those bios have completed by the time we get the page fault error,
7640 * which we return back to our caller - we should only return EIOCBQUEUED
7641 * after we have submitted bios for all the extents in the range.
7643 if ((flags & IOMAP_NOWAIT) && len < length) {
7644 free_extent_map(em);
7650 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7651 start, &len, flags);
7654 unlock_extents = true;
7655 /* Recalc len in case the new em is smaller than requested */
7656 len = min(len, em->len - (start - em->start));
7657 if (dio_data->data_space_reserved) {
7659 u64 release_len = 0;
7661 if (dio_data->nocow_done) {
7662 release_offset = start;
7663 release_len = data_alloc_len;
7664 } else if (len < data_alloc_len) {
7665 release_offset = start + len;
7666 release_len = data_alloc_len - len;
7669 if (release_len > 0)
7670 btrfs_free_reserved_data_space(BTRFS_I(inode),
7671 dio_data->data_reserved,
7677 * We need to unlock only the end area that we aren't using.
7678 * The rest is going to be unlocked by the endio routine.
7680 lockstart = start + len;
7681 if (lockstart < lockend)
7682 unlock_extents = true;
7686 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7689 free_extent_state(cached_state);
7692 * Translate extent map information to iomap.
7693 * We trim the extents (and move the addr) even though iomap code does
7694 * that, since we have locked only the parts we are performing I/O in.
7696 if ((em->block_start == EXTENT_MAP_HOLE) ||
7697 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7698 iomap->addr = IOMAP_NULL_ADDR;
7699 iomap->type = IOMAP_HOLE;
7701 iomap->addr = em->block_start + (start - em->start);
7702 iomap->type = IOMAP_MAPPED;
7704 iomap->offset = start;
7705 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7706 iomap->length = len;
7707 free_extent_map(em);
7712 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7715 if (dio_data->data_space_reserved) {
7716 btrfs_free_reserved_data_space(BTRFS_I(inode),
7717 dio_data->data_reserved,
7718 start, data_alloc_len);
7719 extent_changeset_free(dio_data->data_reserved);
7725 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7726 ssize_t written, unsigned int flags, struct iomap *iomap)
7728 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7729 struct btrfs_dio_data *dio_data = iter->private;
7730 size_t submitted = dio_data->submitted;
7731 const bool write = !!(flags & IOMAP_WRITE);
7734 if (!write && (iomap->type == IOMAP_HOLE)) {
7735 /* If reading from a hole, unlock and return */
7736 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7741 if (submitted < length) {
7743 length -= submitted;
7745 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7746 pos, length, false);
7748 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7749 pos + length - 1, NULL);
7753 btrfs_put_ordered_extent(dio_data->ordered);
7754 dio_data->ordered = NULL;
7758 extent_changeset_free(dio_data->data_reserved);
7762 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7764 struct btrfs_dio_private *dip =
7765 container_of(bbio, struct btrfs_dio_private, bbio);
7766 struct btrfs_inode *inode = bbio->inode;
7767 struct bio *bio = &bbio->bio;
7769 if (bio->bi_status) {
7770 btrfs_warn(inode->root->fs_info,
7771 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7772 btrfs_ino(inode), bio->bi_opf,
7773 dip->file_offset, dip->bytes, bio->bi_status);
7776 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7777 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7778 dip->file_offset, dip->bytes,
7781 unlock_extent(&inode->io_tree, dip->file_offset,
7782 dip->file_offset + dip->bytes - 1, NULL);
7785 bbio->bio.bi_private = bbio->private;
7786 iomap_dio_bio_end_io(bio);
7789 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7792 struct btrfs_bio *bbio = btrfs_bio(bio);
7793 struct btrfs_dio_private *dip =
7794 container_of(bbio, struct btrfs_dio_private, bbio);
7795 struct btrfs_dio_data *dio_data = iter->private;
7797 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7798 btrfs_dio_end_io, bio->bi_private);
7799 bbio->inode = BTRFS_I(iter->inode);
7800 bbio->file_offset = file_offset;
7802 dip->file_offset = file_offset;
7803 dip->bytes = bio->bi_iter.bi_size;
7805 dio_data->submitted += bio->bi_iter.bi_size;
7808 * Check if we are doing a partial write. If we are, we need to split
7809 * the ordered extent to match the submitted bio. Hang on to the
7810 * remaining unfinishable ordered_extent in dio_data so that it can be
7811 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7812 * remaining pages is blocked on the outstanding ordered extent.
7814 if (iter->flags & IOMAP_WRITE) {
7817 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7819 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7820 file_offset, dip->bytes,
7822 bio->bi_status = errno_to_blk_status(ret);
7823 iomap_dio_bio_end_io(bio);
7828 btrfs_submit_bio(bbio, 0);
7831 static const struct iomap_ops btrfs_dio_iomap_ops = {
7832 .iomap_begin = btrfs_dio_iomap_begin,
7833 .iomap_end = btrfs_dio_iomap_end,
7836 static const struct iomap_dio_ops btrfs_dio_ops = {
7837 .submit_io = btrfs_dio_submit_io,
7838 .bio_set = &btrfs_dio_bioset,
7841 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7843 struct btrfs_dio_data data = { 0 };
7845 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7846 IOMAP_DIO_PARTIAL, &data, done_before);
7849 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7852 struct btrfs_dio_data data = { 0 };
7854 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7855 IOMAP_DIO_PARTIAL, &data, done_before);
7858 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7863 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7868 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7869 * file range (0 to LLONG_MAX), but that is not enough if we have
7870 * compression enabled. The first filemap_fdatawrite_range() only kicks
7871 * in the compression of data (in an async thread) and will return
7872 * before the compression is done and writeback is started. A second
7873 * filemap_fdatawrite_range() is needed to wait for the compression to
7874 * complete and writeback to start. We also need to wait for ordered
7875 * extents to complete, because our fiemap implementation uses mainly
7876 * file extent items to list the extents, searching for extent maps
7877 * only for file ranges with holes or prealloc extents to figure out
7878 * if we have delalloc in those ranges.
7880 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7881 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7886 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7889 static int btrfs_writepages(struct address_space *mapping,
7890 struct writeback_control *wbc)
7892 return extent_writepages(mapping, wbc);
7895 static void btrfs_readahead(struct readahead_control *rac)
7897 extent_readahead(rac);
7901 * For release_folio() and invalidate_folio() we have a race window where
7902 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7903 * If we continue to release/invalidate the page, we could cause use-after-free
7904 * for subpage spinlock. So this function is to spin and wait for subpage
7907 static void wait_subpage_spinlock(struct page *page)
7909 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7910 struct btrfs_subpage *subpage;
7912 if (!btrfs_is_subpage(fs_info, page))
7915 ASSERT(PagePrivate(page) && page->private);
7916 subpage = (struct btrfs_subpage *)page->private;
7919 * This may look insane as we just acquire the spinlock and release it,
7920 * without doing anything. But we just want to make sure no one is
7921 * still holding the subpage spinlock.
7922 * And since the page is not dirty nor writeback, and we have page
7923 * locked, the only possible way to hold a spinlock is from the endio
7924 * function to clear page writeback.
7926 * Here we just acquire the spinlock so that all existing callers
7927 * should exit and we're safe to release/invalidate the page.
7929 spin_lock_irq(&subpage->lock);
7930 spin_unlock_irq(&subpage->lock);
7933 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7935 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7938 wait_subpage_spinlock(&folio->page);
7939 clear_page_extent_mapped(&folio->page);
7944 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7946 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7948 return __btrfs_release_folio(folio, gfp_flags);
7951 #ifdef CONFIG_MIGRATION
7952 static int btrfs_migrate_folio(struct address_space *mapping,
7953 struct folio *dst, struct folio *src,
7954 enum migrate_mode mode)
7956 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7958 if (ret != MIGRATEPAGE_SUCCESS)
7961 if (folio_test_ordered(src)) {
7962 folio_clear_ordered(src);
7963 folio_set_ordered(dst);
7966 return MIGRATEPAGE_SUCCESS;
7969 #define btrfs_migrate_folio NULL
7972 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
7975 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
7976 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7977 struct extent_io_tree *tree = &inode->io_tree;
7978 struct extent_state *cached_state = NULL;
7979 u64 page_start = folio_pos(folio);
7980 u64 page_end = page_start + folio_size(folio) - 1;
7982 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
7985 * We have folio locked so no new ordered extent can be created on this
7986 * page, nor bio can be submitted for this folio.
7988 * But already submitted bio can still be finished on this folio.
7989 * Furthermore, endio function won't skip folio which has Ordered
7990 * (Private2) already cleared, so it's possible for endio and
7991 * invalidate_folio to do the same ordered extent accounting twice
7994 * So here we wait for any submitted bios to finish, so that we won't
7995 * do double ordered extent accounting on the same folio.
7997 folio_wait_writeback(folio);
7998 wait_subpage_spinlock(&folio->page);
8001 * For subpage case, we have call sites like
8002 * btrfs_punch_hole_lock_range() which passes range not aligned to
8004 * If the range doesn't cover the full folio, we don't need to and
8005 * shouldn't clear page extent mapped, as folio->private can still
8006 * record subpage dirty bits for other part of the range.
8008 * For cases that invalidate the full folio even the range doesn't
8009 * cover the full folio, like invalidating the last folio, we're
8010 * still safe to wait for ordered extent to finish.
8012 if (!(offset == 0 && length == folio_size(folio))) {
8013 btrfs_release_folio(folio, GFP_NOFS);
8017 if (!inode_evicting)
8018 lock_extent(tree, page_start, page_end, &cached_state);
8021 while (cur < page_end) {
8022 struct btrfs_ordered_extent *ordered;
8025 u32 extra_flags = 0;
8027 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8028 page_end + 1 - cur);
8030 range_end = page_end;
8032 * No ordered extent covering this range, we are safe
8033 * to delete all extent states in the range.
8035 extra_flags = EXTENT_CLEAR_ALL_BITS;
8038 if (ordered->file_offset > cur) {
8040 * There is a range between [cur, oe->file_offset) not
8041 * covered by any ordered extent.
8042 * We are safe to delete all extent states, and handle
8043 * the ordered extent in the next iteration.
8045 range_end = ordered->file_offset - 1;
8046 extra_flags = EXTENT_CLEAR_ALL_BITS;
8050 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8052 ASSERT(range_end + 1 - cur < U32_MAX);
8053 range_len = range_end + 1 - cur;
8054 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8056 * If Ordered (Private2) is cleared, it means endio has
8057 * already been executed for the range.
8058 * We can't delete the extent states as
8059 * btrfs_finish_ordered_io() may still use some of them.
8063 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8066 * IO on this page will never be started, so we need to account
8067 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8068 * here, must leave that up for the ordered extent completion.
8070 * This will also unlock the range for incoming
8071 * btrfs_finish_ordered_io().
8073 if (!inode_evicting)
8074 clear_extent_bit(tree, cur, range_end,
8076 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8077 EXTENT_DEFRAG, &cached_state);
8079 spin_lock_irq(&inode->ordered_tree.lock);
8080 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8081 ordered->truncated_len = min(ordered->truncated_len,
8082 cur - ordered->file_offset);
8083 spin_unlock_irq(&inode->ordered_tree.lock);
8086 * If the ordered extent has finished, we're safe to delete all
8087 * the extent states of the range, otherwise
8088 * btrfs_finish_ordered_io() will get executed by endio for
8089 * other pages, so we can't delete extent states.
8091 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8092 cur, range_end + 1 - cur)) {
8093 btrfs_finish_ordered_io(ordered);
8095 * The ordered extent has finished, now we're again
8096 * safe to delete all extent states of the range.
8098 extra_flags = EXTENT_CLEAR_ALL_BITS;
8102 btrfs_put_ordered_extent(ordered);
8104 * Qgroup reserved space handler
8105 * Sector(s) here will be either:
8107 * 1) Already written to disk or bio already finished
8108 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8109 * Qgroup will be handled by its qgroup_record then.
8110 * btrfs_qgroup_free_data() call will do nothing here.
8112 * 2) Not written to disk yet
8113 * Then btrfs_qgroup_free_data() call will clear the
8114 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8115 * reserved data space.
8116 * Since the IO will never happen for this page.
8118 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8119 if (!inode_evicting) {
8120 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8121 EXTENT_DELALLOC | EXTENT_UPTODATE |
8122 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8123 extra_flags, &cached_state);
8125 cur = range_end + 1;
8128 * We have iterated through all ordered extents of the page, the page
8129 * should not have Ordered (Private2) anymore, or the above iteration
8130 * did something wrong.
8132 ASSERT(!folio_test_ordered(folio));
8133 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8134 if (!inode_evicting)
8135 __btrfs_release_folio(folio, GFP_NOFS);
8136 clear_page_extent_mapped(&folio->page);
8140 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8141 * called from a page fault handler when a page is first dirtied. Hence we must
8142 * be careful to check for EOF conditions here. We set the page up correctly
8143 * for a written page which means we get ENOSPC checking when writing into
8144 * holes and correct delalloc and unwritten extent mapping on filesystems that
8145 * support these features.
8147 * We are not allowed to take the i_mutex here so we have to play games to
8148 * protect against truncate races as the page could now be beyond EOF. Because
8149 * truncate_setsize() writes the inode size before removing pages, once we have
8150 * the page lock we can determine safely if the page is beyond EOF. If it is not
8151 * beyond EOF, then the page is guaranteed safe against truncation until we
8154 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8156 struct page *page = vmf->page;
8157 struct inode *inode = file_inode(vmf->vma->vm_file);
8158 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8159 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8160 struct btrfs_ordered_extent *ordered;
8161 struct extent_state *cached_state = NULL;
8162 struct extent_changeset *data_reserved = NULL;
8163 unsigned long zero_start;
8173 reserved_space = PAGE_SIZE;
8175 sb_start_pagefault(inode->i_sb);
8176 page_start = page_offset(page);
8177 page_end = page_start + PAGE_SIZE - 1;
8181 * Reserving delalloc space after obtaining the page lock can lead to
8182 * deadlock. For example, if a dirty page is locked by this function
8183 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8184 * dirty page write out, then the btrfs_writepages() function could
8185 * end up waiting indefinitely to get a lock on the page currently
8186 * being processed by btrfs_page_mkwrite() function.
8188 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8189 page_start, reserved_space);
8191 ret2 = file_update_time(vmf->vma->vm_file);
8195 ret = vmf_error(ret2);
8201 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8203 down_read(&BTRFS_I(inode)->i_mmap_lock);
8205 size = i_size_read(inode);
8207 if ((page->mapping != inode->i_mapping) ||
8208 (page_start >= size)) {
8209 /* page got truncated out from underneath us */
8212 wait_on_page_writeback(page);
8214 lock_extent(io_tree, page_start, page_end, &cached_state);
8215 ret2 = set_page_extent_mapped(page);
8217 ret = vmf_error(ret2);
8218 unlock_extent(io_tree, page_start, page_end, &cached_state);
8223 * we can't set the delalloc bits if there are pending ordered
8224 * extents. Drop our locks and wait for them to finish
8226 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8229 unlock_extent(io_tree, page_start, page_end, &cached_state);
8231 up_read(&BTRFS_I(inode)->i_mmap_lock);
8232 btrfs_start_ordered_extent(ordered);
8233 btrfs_put_ordered_extent(ordered);
8237 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8238 reserved_space = round_up(size - page_start,
8239 fs_info->sectorsize);
8240 if (reserved_space < PAGE_SIZE) {
8241 end = page_start + reserved_space - 1;
8242 btrfs_delalloc_release_space(BTRFS_I(inode),
8243 data_reserved, page_start,
8244 PAGE_SIZE - reserved_space, true);
8249 * page_mkwrite gets called when the page is firstly dirtied after it's
8250 * faulted in, but write(2) could also dirty a page and set delalloc
8251 * bits, thus in this case for space account reason, we still need to
8252 * clear any delalloc bits within this page range since we have to
8253 * reserve data&meta space before lock_page() (see above comments).
8255 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8256 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8257 EXTENT_DEFRAG, &cached_state);
8259 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8262 unlock_extent(io_tree, page_start, page_end, &cached_state);
8263 ret = VM_FAULT_SIGBUS;
8267 /* page is wholly or partially inside EOF */
8268 if (page_start + PAGE_SIZE > size)
8269 zero_start = offset_in_page(size);
8271 zero_start = PAGE_SIZE;
8273 if (zero_start != PAGE_SIZE)
8274 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8276 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8277 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8278 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8280 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8282 unlock_extent(io_tree, page_start, page_end, &cached_state);
8283 up_read(&BTRFS_I(inode)->i_mmap_lock);
8285 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8286 sb_end_pagefault(inode->i_sb);
8287 extent_changeset_free(data_reserved);
8288 return VM_FAULT_LOCKED;
8292 up_read(&BTRFS_I(inode)->i_mmap_lock);
8294 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8295 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8296 reserved_space, (ret != 0));
8298 sb_end_pagefault(inode->i_sb);
8299 extent_changeset_free(data_reserved);
8303 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8305 struct btrfs_truncate_control control = {
8307 .ino = btrfs_ino(inode),
8308 .min_type = BTRFS_EXTENT_DATA_KEY,
8309 .clear_extent_range = true,
8311 struct btrfs_root *root = inode->root;
8312 struct btrfs_fs_info *fs_info = root->fs_info;
8313 struct btrfs_block_rsv *rsv;
8315 struct btrfs_trans_handle *trans;
8316 u64 mask = fs_info->sectorsize - 1;
8317 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8319 if (!skip_writeback) {
8320 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8321 inode->vfs_inode.i_size & (~mask),
8328 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8329 * things going on here:
8331 * 1) We need to reserve space to update our inode.
8333 * 2) We need to have something to cache all the space that is going to
8334 * be free'd up by the truncate operation, but also have some slack
8335 * space reserved in case it uses space during the truncate (thank you
8336 * very much snapshotting).
8338 * And we need these to be separate. The fact is we can use a lot of
8339 * space doing the truncate, and we have no earthly idea how much space
8340 * we will use, so we need the truncate reservation to be separate so it
8341 * doesn't end up using space reserved for updating the inode. We also
8342 * need to be able to stop the transaction and start a new one, which
8343 * means we need to be able to update the inode several times, and we
8344 * have no idea of knowing how many times that will be, so we can't just
8345 * reserve 1 item for the entirety of the operation, so that has to be
8346 * done separately as well.
8348 * So that leaves us with
8350 * 1) rsv - for the truncate reservation, which we will steal from the
8351 * transaction reservation.
8352 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8353 * updating the inode.
8355 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8358 rsv->size = min_size;
8359 rsv->failfast = true;
8362 * 1 for the truncate slack space
8363 * 1 for updating the inode.
8365 trans = btrfs_start_transaction(root, 2);
8366 if (IS_ERR(trans)) {
8367 ret = PTR_ERR(trans);
8371 /* Migrate the slack space for the truncate to our reserve */
8372 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8375 * We have reserved 2 metadata units when we started the transaction and
8376 * min_size matches 1 unit, so this should never fail, but if it does,
8377 * it's not critical we just fail truncation.
8380 btrfs_end_transaction(trans);
8384 trans->block_rsv = rsv;
8387 struct extent_state *cached_state = NULL;
8388 const u64 new_size = inode->vfs_inode.i_size;
8389 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8391 control.new_size = new_size;
8392 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8394 * We want to drop from the next block forward in case this new
8395 * size is not block aligned since we will be keeping the last
8396 * block of the extent just the way it is.
8398 btrfs_drop_extent_map_range(inode,
8399 ALIGN(new_size, fs_info->sectorsize),
8402 ret = btrfs_truncate_inode_items(trans, root, &control);
8404 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8405 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8407 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8409 trans->block_rsv = &fs_info->trans_block_rsv;
8410 if (ret != -ENOSPC && ret != -EAGAIN)
8413 ret = btrfs_update_inode(trans, root, inode);
8417 btrfs_end_transaction(trans);
8418 btrfs_btree_balance_dirty(fs_info);
8420 trans = btrfs_start_transaction(root, 2);
8421 if (IS_ERR(trans)) {
8422 ret = PTR_ERR(trans);
8427 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8428 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8429 rsv, min_size, false);
8431 * We have reserved 2 metadata units when we started the
8432 * transaction and min_size matches 1 unit, so this should never
8433 * fail, but if it does, it's not critical we just fail truncation.
8438 trans->block_rsv = rsv;
8442 * We can't call btrfs_truncate_block inside a trans handle as we could
8443 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8444 * know we've truncated everything except the last little bit, and can
8445 * do btrfs_truncate_block and then update the disk_i_size.
8447 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8448 btrfs_end_transaction(trans);
8449 btrfs_btree_balance_dirty(fs_info);
8451 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8454 trans = btrfs_start_transaction(root, 1);
8455 if (IS_ERR(trans)) {
8456 ret = PTR_ERR(trans);
8459 btrfs_inode_safe_disk_i_size_write(inode, 0);
8465 trans->block_rsv = &fs_info->trans_block_rsv;
8466 ret2 = btrfs_update_inode(trans, root, inode);
8470 ret2 = btrfs_end_transaction(trans);
8473 btrfs_btree_balance_dirty(fs_info);
8476 btrfs_free_block_rsv(fs_info, rsv);
8478 * So if we truncate and then write and fsync we normally would just
8479 * write the extents that changed, which is a problem if we need to
8480 * first truncate that entire inode. So set this flag so we write out
8481 * all of the extents in the inode to the sync log so we're completely
8484 * If no extents were dropped or trimmed we don't need to force the next
8485 * fsync to truncate all the inode's items from the log and re-log them
8486 * all. This means the truncate operation did not change the file size,
8487 * or changed it to a smaller size but there was only an implicit hole
8488 * between the old i_size and the new i_size, and there were no prealloc
8489 * extents beyond i_size to drop.
8491 if (control.extents_found > 0)
8492 btrfs_set_inode_full_sync(inode);
8497 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8500 struct inode *inode;
8502 inode = new_inode(dir->i_sb);
8505 * Subvolumes don't inherit the sgid bit or the parent's gid if
8506 * the parent's sgid bit is set. This is probably a bug.
8508 inode_init_owner(idmap, inode, NULL,
8509 S_IFDIR | (~current_umask() & S_IRWXUGO));
8510 inode->i_op = &btrfs_dir_inode_operations;
8511 inode->i_fop = &btrfs_dir_file_operations;
8516 struct inode *btrfs_alloc_inode(struct super_block *sb)
8518 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8519 struct btrfs_inode *ei;
8520 struct inode *inode;
8522 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8529 ei->last_sub_trans = 0;
8530 ei->logged_trans = 0;
8531 ei->delalloc_bytes = 0;
8532 ei->new_delalloc_bytes = 0;
8533 ei->defrag_bytes = 0;
8534 ei->disk_i_size = 0;
8538 ei->index_cnt = (u64)-1;
8540 ei->last_unlink_trans = 0;
8541 ei->last_reflink_trans = 0;
8542 ei->last_log_commit = 0;
8544 spin_lock_init(&ei->lock);
8545 ei->outstanding_extents = 0;
8546 if (sb->s_magic != BTRFS_TEST_MAGIC)
8547 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8548 BTRFS_BLOCK_RSV_DELALLOC);
8549 ei->runtime_flags = 0;
8550 ei->prop_compress = BTRFS_COMPRESS_NONE;
8551 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8553 ei->delayed_node = NULL;
8555 ei->i_otime.tv_sec = 0;
8556 ei->i_otime.tv_nsec = 0;
8558 inode = &ei->vfs_inode;
8559 extent_map_tree_init(&ei->extent_tree);
8560 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8561 ei->io_tree.inode = ei;
8562 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8563 IO_TREE_INODE_FILE_EXTENT);
8564 mutex_init(&ei->log_mutex);
8565 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8566 INIT_LIST_HEAD(&ei->delalloc_inodes);
8567 INIT_LIST_HEAD(&ei->delayed_iput);
8568 RB_CLEAR_NODE(&ei->rb_node);
8569 init_rwsem(&ei->i_mmap_lock);
8574 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8575 void btrfs_test_destroy_inode(struct inode *inode)
8577 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8578 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8582 void btrfs_free_inode(struct inode *inode)
8584 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8587 void btrfs_destroy_inode(struct inode *vfs_inode)
8589 struct btrfs_ordered_extent *ordered;
8590 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8591 struct btrfs_root *root = inode->root;
8592 bool freespace_inode;
8594 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8595 WARN_ON(vfs_inode->i_data.nrpages);
8596 WARN_ON(inode->block_rsv.reserved);
8597 WARN_ON(inode->block_rsv.size);
8598 WARN_ON(inode->outstanding_extents);
8599 if (!S_ISDIR(vfs_inode->i_mode)) {
8600 WARN_ON(inode->delalloc_bytes);
8601 WARN_ON(inode->new_delalloc_bytes);
8603 WARN_ON(inode->csum_bytes);
8604 WARN_ON(inode->defrag_bytes);
8607 * This can happen where we create an inode, but somebody else also
8608 * created the same inode and we need to destroy the one we already
8615 * If this is a free space inode do not take the ordered extents lockdep
8618 freespace_inode = btrfs_is_free_space_inode(inode);
8621 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8625 btrfs_err(root->fs_info,
8626 "found ordered extent %llu %llu on inode cleanup",
8627 ordered->file_offset, ordered->num_bytes);
8629 if (!freespace_inode)
8630 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8632 btrfs_remove_ordered_extent(inode, ordered);
8633 btrfs_put_ordered_extent(ordered);
8634 btrfs_put_ordered_extent(ordered);
8637 btrfs_qgroup_check_reserved_leak(inode);
8638 inode_tree_del(inode);
8639 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8640 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8641 btrfs_put_root(inode->root);
8644 int btrfs_drop_inode(struct inode *inode)
8646 struct btrfs_root *root = BTRFS_I(inode)->root;
8651 /* the snap/subvol tree is on deleting */
8652 if (btrfs_root_refs(&root->root_item) == 0)
8655 return generic_drop_inode(inode);
8658 static void init_once(void *foo)
8660 struct btrfs_inode *ei = foo;
8662 inode_init_once(&ei->vfs_inode);
8665 void __cold btrfs_destroy_cachep(void)
8668 * Make sure all delayed rcu free inodes are flushed before we
8672 bioset_exit(&btrfs_dio_bioset);
8673 kmem_cache_destroy(btrfs_inode_cachep);
8676 int __init btrfs_init_cachep(void)
8678 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8679 sizeof(struct btrfs_inode), 0,
8680 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8682 if (!btrfs_inode_cachep)
8685 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8686 offsetof(struct btrfs_dio_private, bbio.bio),
8692 btrfs_destroy_cachep();
8696 static int btrfs_getattr(struct mnt_idmap *idmap,
8697 const struct path *path, struct kstat *stat,
8698 u32 request_mask, unsigned int flags)
8702 struct inode *inode = d_inode(path->dentry);
8703 u32 blocksize = inode->i_sb->s_blocksize;
8704 u32 bi_flags = BTRFS_I(inode)->flags;
8705 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8707 stat->result_mask |= STATX_BTIME;
8708 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8709 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8710 if (bi_flags & BTRFS_INODE_APPEND)
8711 stat->attributes |= STATX_ATTR_APPEND;
8712 if (bi_flags & BTRFS_INODE_COMPRESS)
8713 stat->attributes |= STATX_ATTR_COMPRESSED;
8714 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8715 stat->attributes |= STATX_ATTR_IMMUTABLE;
8716 if (bi_flags & BTRFS_INODE_NODUMP)
8717 stat->attributes |= STATX_ATTR_NODUMP;
8718 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8719 stat->attributes |= STATX_ATTR_VERITY;
8721 stat->attributes_mask |= (STATX_ATTR_APPEND |
8722 STATX_ATTR_COMPRESSED |
8723 STATX_ATTR_IMMUTABLE |
8726 generic_fillattr(idmap, inode, stat);
8727 stat->dev = BTRFS_I(inode)->root->anon_dev;
8729 spin_lock(&BTRFS_I(inode)->lock);
8730 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8731 inode_bytes = inode_get_bytes(inode);
8732 spin_unlock(&BTRFS_I(inode)->lock);
8733 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8734 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8738 static int btrfs_rename_exchange(struct inode *old_dir,
8739 struct dentry *old_dentry,
8740 struct inode *new_dir,
8741 struct dentry *new_dentry)
8743 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8744 struct btrfs_trans_handle *trans;
8745 unsigned int trans_num_items;
8746 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8747 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8748 struct inode *new_inode = new_dentry->d_inode;
8749 struct inode *old_inode = old_dentry->d_inode;
8750 struct timespec64 ctime = current_time(old_inode);
8751 struct btrfs_rename_ctx old_rename_ctx;
8752 struct btrfs_rename_ctx new_rename_ctx;
8753 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8754 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8759 bool need_abort = false;
8760 struct fscrypt_name old_fname, new_fname;
8761 struct fscrypt_str *old_name, *new_name;
8764 * For non-subvolumes allow exchange only within one subvolume, in the
8765 * same inode namespace. Two subvolumes (represented as directory) can
8766 * be exchanged as they're a logical link and have a fixed inode number.
8769 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8770 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8773 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8777 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8779 fscrypt_free_filename(&old_fname);
8783 old_name = &old_fname.disk_name;
8784 new_name = &new_fname.disk_name;
8786 /* close the race window with snapshot create/destroy ioctl */
8787 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8788 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8789 down_read(&fs_info->subvol_sem);
8793 * 1 to remove old dir item
8794 * 1 to remove old dir index
8795 * 1 to add new dir item
8796 * 1 to add new dir index
8797 * 1 to update parent inode
8799 * If the parents are the same, we only need to account for one
8801 trans_num_items = (old_dir == new_dir ? 9 : 10);
8802 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8804 * 1 to remove old root ref
8805 * 1 to remove old root backref
8806 * 1 to add new root ref
8807 * 1 to add new root backref
8809 trans_num_items += 4;
8812 * 1 to update inode item
8813 * 1 to remove old inode ref
8814 * 1 to add new inode ref
8816 trans_num_items += 3;
8818 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8819 trans_num_items += 4;
8821 trans_num_items += 3;
8822 trans = btrfs_start_transaction(root, trans_num_items);
8823 if (IS_ERR(trans)) {
8824 ret = PTR_ERR(trans);
8829 ret = btrfs_record_root_in_trans(trans, dest);
8835 * We need to find a free sequence number both in the source and
8836 * in the destination directory for the exchange.
8838 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8841 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8845 BTRFS_I(old_inode)->dir_index = 0ULL;
8846 BTRFS_I(new_inode)->dir_index = 0ULL;
8848 /* Reference for the source. */
8849 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8850 /* force full log commit if subvolume involved. */
8851 btrfs_set_log_full_commit(trans);
8853 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8854 btrfs_ino(BTRFS_I(new_dir)),
8861 /* And now for the dest. */
8862 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8863 /* force full log commit if subvolume involved. */
8864 btrfs_set_log_full_commit(trans);
8866 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8867 btrfs_ino(BTRFS_I(old_dir)),
8871 btrfs_abort_transaction(trans, ret);
8876 /* Update inode version and ctime/mtime. */
8877 inode_inc_iversion(old_dir);
8878 inode_inc_iversion(new_dir);
8879 inode_inc_iversion(old_inode);
8880 inode_inc_iversion(new_inode);
8881 old_dir->i_mtime = ctime;
8882 old_dir->i_ctime = ctime;
8883 new_dir->i_mtime = ctime;
8884 new_dir->i_ctime = ctime;
8885 old_inode->i_ctime = ctime;
8886 new_inode->i_ctime = ctime;
8888 if (old_dentry->d_parent != new_dentry->d_parent) {
8889 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8890 BTRFS_I(old_inode), true);
8891 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8892 BTRFS_I(new_inode), true);
8895 /* src is a subvolume */
8896 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8897 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8898 } else { /* src is an inode */
8899 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8900 BTRFS_I(old_dentry->d_inode),
8901 old_name, &old_rename_ctx);
8903 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8906 btrfs_abort_transaction(trans, ret);
8910 /* dest is a subvolume */
8911 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8912 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8913 } else { /* dest is an inode */
8914 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8915 BTRFS_I(new_dentry->d_inode),
8916 new_name, &new_rename_ctx);
8918 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
8921 btrfs_abort_transaction(trans, ret);
8925 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8926 new_name, 0, old_idx);
8928 btrfs_abort_transaction(trans, ret);
8932 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8933 old_name, 0, new_idx);
8935 btrfs_abort_transaction(trans, ret);
8939 if (old_inode->i_nlink == 1)
8940 BTRFS_I(old_inode)->dir_index = old_idx;
8941 if (new_inode->i_nlink == 1)
8942 BTRFS_I(new_inode)->dir_index = new_idx;
8945 * Now pin the logs of the roots. We do it to ensure that no other task
8946 * can sync the logs while we are in progress with the rename, because
8947 * that could result in an inconsistency in case any of the inodes that
8948 * are part of this rename operation were logged before.
8950 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8951 btrfs_pin_log_trans(root);
8952 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8953 btrfs_pin_log_trans(dest);
8955 /* Do the log updates for all inodes. */
8956 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8957 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
8958 old_rename_ctx.index, new_dentry->d_parent);
8959 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8960 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
8961 new_rename_ctx.index, old_dentry->d_parent);
8963 /* Now unpin the logs. */
8964 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8965 btrfs_end_log_trans(root);
8966 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8967 btrfs_end_log_trans(dest);
8969 ret2 = btrfs_end_transaction(trans);
8970 ret = ret ? ret : ret2;
8972 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
8973 old_ino == BTRFS_FIRST_FREE_OBJECTID)
8974 up_read(&fs_info->subvol_sem);
8976 fscrypt_free_filename(&new_fname);
8977 fscrypt_free_filename(&old_fname);
8981 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
8984 struct inode *inode;
8986 inode = new_inode(dir->i_sb);
8988 inode_init_owner(idmap, inode, dir,
8989 S_IFCHR | WHITEOUT_MODE);
8990 inode->i_op = &btrfs_special_inode_operations;
8991 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
8996 static int btrfs_rename(struct mnt_idmap *idmap,
8997 struct inode *old_dir, struct dentry *old_dentry,
8998 struct inode *new_dir, struct dentry *new_dentry,
9001 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9002 struct btrfs_new_inode_args whiteout_args = {
9004 .dentry = old_dentry,
9006 struct btrfs_trans_handle *trans;
9007 unsigned int trans_num_items;
9008 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9009 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9010 struct inode *new_inode = d_inode(new_dentry);
9011 struct inode *old_inode = d_inode(old_dentry);
9012 struct btrfs_rename_ctx rename_ctx;
9016 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9017 struct fscrypt_name old_fname, new_fname;
9019 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9022 /* we only allow rename subvolume link between subvolumes */
9023 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9026 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9027 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9030 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9031 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9034 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9038 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9040 fscrypt_free_filename(&old_fname);
9044 /* check for collisions, even if the name isn't there */
9045 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9047 if (ret == -EEXIST) {
9049 * eexist without a new_inode */
9050 if (WARN_ON(!new_inode)) {
9051 goto out_fscrypt_names;
9054 /* maybe -EOVERFLOW */
9055 goto out_fscrypt_names;
9061 * we're using rename to replace one file with another. Start IO on it
9062 * now so we don't add too much work to the end of the transaction
9064 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9065 filemap_flush(old_inode->i_mapping);
9067 if (flags & RENAME_WHITEOUT) {
9068 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9069 if (!whiteout_args.inode) {
9071 goto out_fscrypt_names;
9073 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9075 goto out_whiteout_inode;
9077 /* 1 to update the old parent inode. */
9078 trans_num_items = 1;
9081 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9082 /* Close the race window with snapshot create/destroy ioctl */
9083 down_read(&fs_info->subvol_sem);
9085 * 1 to remove old root ref
9086 * 1 to remove old root backref
9087 * 1 to add new root ref
9088 * 1 to add new root backref
9090 trans_num_items += 4;
9094 * 1 to remove old inode ref
9095 * 1 to add new inode ref
9097 trans_num_items += 3;
9100 * 1 to remove old dir item
9101 * 1 to remove old dir index
9102 * 1 to add new dir item
9103 * 1 to add new dir index
9105 trans_num_items += 4;
9106 /* 1 to update new parent inode if it's not the same as the old parent */
9107 if (new_dir != old_dir)
9112 * 1 to remove inode ref
9113 * 1 to remove dir item
9114 * 1 to remove dir index
9115 * 1 to possibly add orphan item
9117 trans_num_items += 5;
9119 trans = btrfs_start_transaction(root, trans_num_items);
9120 if (IS_ERR(trans)) {
9121 ret = PTR_ERR(trans);
9126 ret = btrfs_record_root_in_trans(trans, dest);
9131 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9135 BTRFS_I(old_inode)->dir_index = 0ULL;
9136 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9137 /* force full log commit if subvolume involved. */
9138 btrfs_set_log_full_commit(trans);
9140 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9141 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9147 inode_inc_iversion(old_dir);
9148 inode_inc_iversion(new_dir);
9149 inode_inc_iversion(old_inode);
9150 old_dir->i_mtime = current_time(old_dir);
9151 old_dir->i_ctime = old_dir->i_mtime;
9152 new_dir->i_mtime = old_dir->i_mtime;
9153 new_dir->i_ctime = old_dir->i_mtime;
9154 old_inode->i_ctime = old_dir->i_mtime;
9156 if (old_dentry->d_parent != new_dentry->d_parent)
9157 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9158 BTRFS_I(old_inode), true);
9160 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9161 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9163 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9164 BTRFS_I(d_inode(old_dentry)),
9165 &old_fname.disk_name, &rename_ctx);
9167 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9170 btrfs_abort_transaction(trans, ret);
9175 inode_inc_iversion(new_inode);
9176 new_inode->i_ctime = current_time(new_inode);
9177 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9178 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9179 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9180 BUG_ON(new_inode->i_nlink == 0);
9182 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9183 BTRFS_I(d_inode(new_dentry)),
9184 &new_fname.disk_name);
9186 if (!ret && new_inode->i_nlink == 0)
9187 ret = btrfs_orphan_add(trans,
9188 BTRFS_I(d_inode(new_dentry)));
9190 btrfs_abort_transaction(trans, ret);
9195 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9196 &new_fname.disk_name, 0, index);
9198 btrfs_abort_transaction(trans, ret);
9202 if (old_inode->i_nlink == 1)
9203 BTRFS_I(old_inode)->dir_index = index;
9205 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9206 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9207 rename_ctx.index, new_dentry->d_parent);
9209 if (flags & RENAME_WHITEOUT) {
9210 ret = btrfs_create_new_inode(trans, &whiteout_args);
9212 btrfs_abort_transaction(trans, ret);
9215 unlock_new_inode(whiteout_args.inode);
9216 iput(whiteout_args.inode);
9217 whiteout_args.inode = NULL;
9221 ret2 = btrfs_end_transaction(trans);
9222 ret = ret ? ret : ret2;
9224 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9225 up_read(&fs_info->subvol_sem);
9226 if (flags & RENAME_WHITEOUT)
9227 btrfs_new_inode_args_destroy(&whiteout_args);
9229 if (flags & RENAME_WHITEOUT)
9230 iput(whiteout_args.inode);
9232 fscrypt_free_filename(&old_fname);
9233 fscrypt_free_filename(&new_fname);
9237 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9238 struct dentry *old_dentry, struct inode *new_dir,
9239 struct dentry *new_dentry, unsigned int flags)
9243 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9246 if (flags & RENAME_EXCHANGE)
9247 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9250 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9253 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9258 struct btrfs_delalloc_work {
9259 struct inode *inode;
9260 struct completion completion;
9261 struct list_head list;
9262 struct btrfs_work work;
9265 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9267 struct btrfs_delalloc_work *delalloc_work;
9268 struct inode *inode;
9270 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9272 inode = delalloc_work->inode;
9273 filemap_flush(inode->i_mapping);
9274 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9275 &BTRFS_I(inode)->runtime_flags))
9276 filemap_flush(inode->i_mapping);
9279 complete(&delalloc_work->completion);
9282 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9284 struct btrfs_delalloc_work *work;
9286 work = kmalloc(sizeof(*work), GFP_NOFS);
9290 init_completion(&work->completion);
9291 INIT_LIST_HEAD(&work->list);
9292 work->inode = inode;
9293 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9299 * some fairly slow code that needs optimization. This walks the list
9300 * of all the inodes with pending delalloc and forces them to disk.
9302 static int start_delalloc_inodes(struct btrfs_root *root,
9303 struct writeback_control *wbc, bool snapshot,
9304 bool in_reclaim_context)
9306 struct btrfs_inode *binode;
9307 struct inode *inode;
9308 struct btrfs_delalloc_work *work, *next;
9309 struct list_head works;
9310 struct list_head splice;
9312 bool full_flush = wbc->nr_to_write == LONG_MAX;
9314 INIT_LIST_HEAD(&works);
9315 INIT_LIST_HEAD(&splice);
9317 mutex_lock(&root->delalloc_mutex);
9318 spin_lock(&root->delalloc_lock);
9319 list_splice_init(&root->delalloc_inodes, &splice);
9320 while (!list_empty(&splice)) {
9321 binode = list_entry(splice.next, struct btrfs_inode,
9324 list_move_tail(&binode->delalloc_inodes,
9325 &root->delalloc_inodes);
9327 if (in_reclaim_context &&
9328 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9331 inode = igrab(&binode->vfs_inode);
9333 cond_resched_lock(&root->delalloc_lock);
9336 spin_unlock(&root->delalloc_lock);
9339 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9340 &binode->runtime_flags);
9342 work = btrfs_alloc_delalloc_work(inode);
9348 list_add_tail(&work->list, &works);
9349 btrfs_queue_work(root->fs_info->flush_workers,
9352 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9353 btrfs_add_delayed_iput(BTRFS_I(inode));
9354 if (ret || wbc->nr_to_write <= 0)
9358 spin_lock(&root->delalloc_lock);
9360 spin_unlock(&root->delalloc_lock);
9363 list_for_each_entry_safe(work, next, &works, list) {
9364 list_del_init(&work->list);
9365 wait_for_completion(&work->completion);
9369 if (!list_empty(&splice)) {
9370 spin_lock(&root->delalloc_lock);
9371 list_splice_tail(&splice, &root->delalloc_inodes);
9372 spin_unlock(&root->delalloc_lock);
9374 mutex_unlock(&root->delalloc_mutex);
9378 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9380 struct writeback_control wbc = {
9381 .nr_to_write = LONG_MAX,
9382 .sync_mode = WB_SYNC_NONE,
9384 .range_end = LLONG_MAX,
9386 struct btrfs_fs_info *fs_info = root->fs_info;
9388 if (BTRFS_FS_ERROR(fs_info))
9391 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9394 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9395 bool in_reclaim_context)
9397 struct writeback_control wbc = {
9399 .sync_mode = WB_SYNC_NONE,
9401 .range_end = LLONG_MAX,
9403 struct btrfs_root *root;
9404 struct list_head splice;
9407 if (BTRFS_FS_ERROR(fs_info))
9410 INIT_LIST_HEAD(&splice);
9412 mutex_lock(&fs_info->delalloc_root_mutex);
9413 spin_lock(&fs_info->delalloc_root_lock);
9414 list_splice_init(&fs_info->delalloc_roots, &splice);
9415 while (!list_empty(&splice)) {
9417 * Reset nr_to_write here so we know that we're doing a full
9421 wbc.nr_to_write = LONG_MAX;
9423 root = list_first_entry(&splice, struct btrfs_root,
9425 root = btrfs_grab_root(root);
9427 list_move_tail(&root->delalloc_root,
9428 &fs_info->delalloc_roots);
9429 spin_unlock(&fs_info->delalloc_root_lock);
9431 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9432 btrfs_put_root(root);
9433 if (ret < 0 || wbc.nr_to_write <= 0)
9435 spin_lock(&fs_info->delalloc_root_lock);
9437 spin_unlock(&fs_info->delalloc_root_lock);
9441 if (!list_empty(&splice)) {
9442 spin_lock(&fs_info->delalloc_root_lock);
9443 list_splice_tail(&splice, &fs_info->delalloc_roots);
9444 spin_unlock(&fs_info->delalloc_root_lock);
9446 mutex_unlock(&fs_info->delalloc_root_mutex);
9450 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9451 struct dentry *dentry, const char *symname)
9453 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9454 struct btrfs_trans_handle *trans;
9455 struct btrfs_root *root = BTRFS_I(dir)->root;
9456 struct btrfs_path *path;
9457 struct btrfs_key key;
9458 struct inode *inode;
9459 struct btrfs_new_inode_args new_inode_args = {
9463 unsigned int trans_num_items;
9468 struct btrfs_file_extent_item *ei;
9469 struct extent_buffer *leaf;
9471 name_len = strlen(symname);
9472 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9473 return -ENAMETOOLONG;
9475 inode = new_inode(dir->i_sb);
9478 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9479 inode->i_op = &btrfs_symlink_inode_operations;
9480 inode_nohighmem(inode);
9481 inode->i_mapping->a_ops = &btrfs_aops;
9482 btrfs_i_size_write(BTRFS_I(inode), name_len);
9483 inode_set_bytes(inode, name_len);
9485 new_inode_args.inode = inode;
9486 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9489 /* 1 additional item for the inline extent */
9492 trans = btrfs_start_transaction(root, trans_num_items);
9493 if (IS_ERR(trans)) {
9494 err = PTR_ERR(trans);
9495 goto out_new_inode_args;
9498 err = btrfs_create_new_inode(trans, &new_inode_args);
9502 path = btrfs_alloc_path();
9505 btrfs_abort_transaction(trans, err);
9506 discard_new_inode(inode);
9510 key.objectid = btrfs_ino(BTRFS_I(inode));
9512 key.type = BTRFS_EXTENT_DATA_KEY;
9513 datasize = btrfs_file_extent_calc_inline_size(name_len);
9514 err = btrfs_insert_empty_item(trans, root, path, &key,
9517 btrfs_abort_transaction(trans, err);
9518 btrfs_free_path(path);
9519 discard_new_inode(inode);
9523 leaf = path->nodes[0];
9524 ei = btrfs_item_ptr(leaf, path->slots[0],
9525 struct btrfs_file_extent_item);
9526 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9527 btrfs_set_file_extent_type(leaf, ei,
9528 BTRFS_FILE_EXTENT_INLINE);
9529 btrfs_set_file_extent_encryption(leaf, ei, 0);
9530 btrfs_set_file_extent_compression(leaf, ei, 0);
9531 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9532 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9534 ptr = btrfs_file_extent_inline_start(ei);
9535 write_extent_buffer(leaf, symname, ptr, name_len);
9536 btrfs_mark_buffer_dirty(leaf);
9537 btrfs_free_path(path);
9539 d_instantiate_new(dentry, inode);
9542 btrfs_end_transaction(trans);
9543 btrfs_btree_balance_dirty(fs_info);
9545 btrfs_new_inode_args_destroy(&new_inode_args);
9552 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9553 struct btrfs_trans_handle *trans_in,
9554 struct btrfs_inode *inode,
9555 struct btrfs_key *ins,
9558 struct btrfs_file_extent_item stack_fi;
9559 struct btrfs_replace_extent_info extent_info;
9560 struct btrfs_trans_handle *trans = trans_in;
9561 struct btrfs_path *path;
9562 u64 start = ins->objectid;
9563 u64 len = ins->offset;
9564 int qgroup_released;
9567 memset(&stack_fi, 0, sizeof(stack_fi));
9569 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9570 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9571 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9572 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9573 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9574 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9575 /* Encryption and other encoding is reserved and all 0 */
9577 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9578 if (qgroup_released < 0)
9579 return ERR_PTR(qgroup_released);
9582 ret = insert_reserved_file_extent(trans, inode,
9583 file_offset, &stack_fi,
9584 true, qgroup_released);
9590 extent_info.disk_offset = start;
9591 extent_info.disk_len = len;
9592 extent_info.data_offset = 0;
9593 extent_info.data_len = len;
9594 extent_info.file_offset = file_offset;
9595 extent_info.extent_buf = (char *)&stack_fi;
9596 extent_info.is_new_extent = true;
9597 extent_info.update_times = true;
9598 extent_info.qgroup_reserved = qgroup_released;
9599 extent_info.insertions = 0;
9601 path = btrfs_alloc_path();
9607 ret = btrfs_replace_file_extents(inode, path, file_offset,
9608 file_offset + len - 1, &extent_info,
9610 btrfs_free_path(path);
9617 * We have released qgroup data range at the beginning of the function,
9618 * and normally qgroup_released bytes will be freed when committing
9620 * But if we error out early, we have to free what we have released
9621 * or we leak qgroup data reservation.
9623 btrfs_qgroup_free_refroot(inode->root->fs_info,
9624 inode->root->root_key.objectid, qgroup_released,
9625 BTRFS_QGROUP_RSV_DATA);
9626 return ERR_PTR(ret);
9629 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9630 u64 start, u64 num_bytes, u64 min_size,
9631 loff_t actual_len, u64 *alloc_hint,
9632 struct btrfs_trans_handle *trans)
9634 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9635 struct extent_map *em;
9636 struct btrfs_root *root = BTRFS_I(inode)->root;
9637 struct btrfs_key ins;
9638 u64 cur_offset = start;
9639 u64 clear_offset = start;
9642 u64 last_alloc = (u64)-1;
9644 bool own_trans = true;
9645 u64 end = start + num_bytes - 1;
9649 while (num_bytes > 0) {
9650 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9651 cur_bytes = max(cur_bytes, min_size);
9653 * If we are severely fragmented we could end up with really
9654 * small allocations, so if the allocator is returning small
9655 * chunks lets make its job easier by only searching for those
9658 cur_bytes = min(cur_bytes, last_alloc);
9659 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9660 min_size, 0, *alloc_hint, &ins, 1, 0);
9665 * We've reserved this space, and thus converted it from
9666 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9667 * from here on out we will only need to clear our reservation
9668 * for the remaining unreserved area, so advance our
9669 * clear_offset by our extent size.
9671 clear_offset += ins.offset;
9673 last_alloc = ins.offset;
9674 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9677 * Now that we inserted the prealloc extent we can finally
9678 * decrement the number of reservations in the block group.
9679 * If we did it before, we could race with relocation and have
9680 * relocation miss the reserved extent, making it fail later.
9682 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9683 if (IS_ERR(trans)) {
9684 ret = PTR_ERR(trans);
9685 btrfs_free_reserved_extent(fs_info, ins.objectid,
9690 em = alloc_extent_map();
9692 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9693 cur_offset + ins.offset - 1, false);
9694 btrfs_set_inode_full_sync(BTRFS_I(inode));
9698 em->start = cur_offset;
9699 em->orig_start = cur_offset;
9700 em->len = ins.offset;
9701 em->block_start = ins.objectid;
9702 em->block_len = ins.offset;
9703 em->orig_block_len = ins.offset;
9704 em->ram_bytes = ins.offset;
9705 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9706 em->generation = trans->transid;
9708 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9709 free_extent_map(em);
9711 num_bytes -= ins.offset;
9712 cur_offset += ins.offset;
9713 *alloc_hint = ins.objectid + ins.offset;
9715 inode_inc_iversion(inode);
9716 inode->i_ctime = current_time(inode);
9717 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9718 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9719 (actual_len > inode->i_size) &&
9720 (cur_offset > inode->i_size)) {
9721 if (cur_offset > actual_len)
9722 i_size = actual_len;
9724 i_size = cur_offset;
9725 i_size_write(inode, i_size);
9726 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9729 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9732 btrfs_abort_transaction(trans, ret);
9734 btrfs_end_transaction(trans);
9739 btrfs_end_transaction(trans);
9743 if (clear_offset < end)
9744 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9745 end - clear_offset + 1);
9749 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9750 u64 start, u64 num_bytes, u64 min_size,
9751 loff_t actual_len, u64 *alloc_hint)
9753 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9754 min_size, actual_len, alloc_hint,
9758 int btrfs_prealloc_file_range_trans(struct inode *inode,
9759 struct btrfs_trans_handle *trans, int mode,
9760 u64 start, u64 num_bytes, u64 min_size,
9761 loff_t actual_len, u64 *alloc_hint)
9763 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9764 min_size, actual_len, alloc_hint, trans);
9767 static int btrfs_permission(struct mnt_idmap *idmap,
9768 struct inode *inode, int mask)
9770 struct btrfs_root *root = BTRFS_I(inode)->root;
9771 umode_t mode = inode->i_mode;
9773 if (mask & MAY_WRITE &&
9774 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9775 if (btrfs_root_readonly(root))
9777 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9780 return generic_permission(idmap, inode, mask);
9783 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9784 struct file *file, umode_t mode)
9786 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9787 struct btrfs_trans_handle *trans;
9788 struct btrfs_root *root = BTRFS_I(dir)->root;
9789 struct inode *inode;
9790 struct btrfs_new_inode_args new_inode_args = {
9792 .dentry = file->f_path.dentry,
9795 unsigned int trans_num_items;
9798 inode = new_inode(dir->i_sb);
9801 inode_init_owner(idmap, inode, dir, mode);
9802 inode->i_fop = &btrfs_file_operations;
9803 inode->i_op = &btrfs_file_inode_operations;
9804 inode->i_mapping->a_ops = &btrfs_aops;
9806 new_inode_args.inode = inode;
9807 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9811 trans = btrfs_start_transaction(root, trans_num_items);
9812 if (IS_ERR(trans)) {
9813 ret = PTR_ERR(trans);
9814 goto out_new_inode_args;
9817 ret = btrfs_create_new_inode(trans, &new_inode_args);
9820 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9821 * set it to 1 because d_tmpfile() will issue a warning if the count is
9824 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9826 set_nlink(inode, 1);
9829 d_tmpfile(file, inode);
9830 unlock_new_inode(inode);
9831 mark_inode_dirty(inode);
9834 btrfs_end_transaction(trans);
9835 btrfs_btree_balance_dirty(fs_info);
9837 btrfs_new_inode_args_destroy(&new_inode_args);
9841 return finish_open_simple(file, ret);
9844 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9846 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9847 unsigned long index = start >> PAGE_SHIFT;
9848 unsigned long end_index = end >> PAGE_SHIFT;
9852 ASSERT(end + 1 - start <= U32_MAX);
9853 len = end + 1 - start;
9854 while (index <= end_index) {
9855 page = find_get_page(inode->vfs_inode.i_mapping, index);
9856 ASSERT(page); /* Pages should be in the extent_io_tree */
9858 btrfs_page_set_writeback(fs_info, page, start, len);
9864 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9867 switch (compress_type) {
9868 case BTRFS_COMPRESS_NONE:
9869 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9870 case BTRFS_COMPRESS_ZLIB:
9871 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9872 case BTRFS_COMPRESS_LZO:
9874 * The LZO format depends on the sector size. 64K is the maximum
9875 * sector size that we support.
9877 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9879 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9880 (fs_info->sectorsize_bits - 12);
9881 case BTRFS_COMPRESS_ZSTD:
9882 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9888 static ssize_t btrfs_encoded_read_inline(
9890 struct iov_iter *iter, u64 start,
9892 struct extent_state **cached_state,
9893 u64 extent_start, size_t count,
9894 struct btrfs_ioctl_encoded_io_args *encoded,
9897 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9898 struct btrfs_root *root = inode->root;
9899 struct btrfs_fs_info *fs_info = root->fs_info;
9900 struct extent_io_tree *io_tree = &inode->io_tree;
9901 struct btrfs_path *path;
9902 struct extent_buffer *leaf;
9903 struct btrfs_file_extent_item *item;
9909 path = btrfs_alloc_path();
9914 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9918 /* The extent item disappeared? */
9923 leaf = path->nodes[0];
9924 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9926 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9927 ptr = btrfs_file_extent_inline_start(item);
9929 encoded->len = min_t(u64, extent_start + ram_bytes,
9930 inode->vfs_inode.i_size) - iocb->ki_pos;
9931 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9932 btrfs_file_extent_compression(leaf, item));
9935 encoded->compression = ret;
9936 if (encoded->compression) {
9939 inline_size = btrfs_file_extent_inline_item_len(leaf,
9941 if (inline_size > count) {
9945 count = inline_size;
9946 encoded->unencoded_len = ram_bytes;
9947 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9949 count = min_t(u64, count, encoded->len);
9950 encoded->len = count;
9951 encoded->unencoded_len = count;
9952 ptr += iocb->ki_pos - extent_start;
9955 tmp = kmalloc(count, GFP_NOFS);
9960 read_extent_buffer(leaf, tmp, ptr, count);
9961 btrfs_release_path(path);
9962 unlock_extent(io_tree, start, lockend, cached_state);
9963 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9966 ret = copy_to_iter(tmp, count, iter);
9971 btrfs_free_path(path);
9975 struct btrfs_encoded_read_private {
9976 wait_queue_head_t wait;
9978 blk_status_t status;
9981 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
9983 struct btrfs_encoded_read_private *priv = bbio->private;
9985 if (bbio->bio.bi_status) {
9987 * The memory barrier implied by the atomic_dec_return() here
9988 * pairs with the memory barrier implied by the
9989 * atomic_dec_return() or io_wait_event() in
9990 * btrfs_encoded_read_regular_fill_pages() to ensure that this
9991 * write is observed before the load of status in
9992 * btrfs_encoded_read_regular_fill_pages().
9994 WRITE_ONCE(priv->status, bbio->bio.bi_status);
9996 if (!atomic_dec_return(&priv->pending))
9997 wake_up(&priv->wait);
9998 bio_put(&bbio->bio);
10001 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10002 u64 file_offset, u64 disk_bytenr,
10003 u64 disk_io_size, struct page **pages)
10005 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10006 struct btrfs_encoded_read_private priv = {
10007 .pending = ATOMIC_INIT(1),
10009 unsigned long i = 0;
10010 struct btrfs_bio *bbio;
10012 init_waitqueue_head(&priv.wait);
10014 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10015 btrfs_encoded_read_endio, &priv);
10016 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10017 bbio->inode = inode;
10020 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
10022 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
10023 atomic_inc(&priv.pending);
10024 btrfs_submit_bio(bbio, 0);
10026 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10027 btrfs_encoded_read_endio, &priv);
10028 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10029 bbio->inode = inode;
10034 disk_bytenr += bytes;
10035 disk_io_size -= bytes;
10036 } while (disk_io_size);
10038 atomic_inc(&priv.pending);
10039 btrfs_submit_bio(bbio, 0);
10041 if (atomic_dec_return(&priv.pending))
10042 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10043 /* See btrfs_encoded_read_endio() for ordering. */
10044 return blk_status_to_errno(READ_ONCE(priv.status));
10047 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10048 struct iov_iter *iter,
10049 u64 start, u64 lockend,
10050 struct extent_state **cached_state,
10051 u64 disk_bytenr, u64 disk_io_size,
10052 size_t count, bool compressed,
10055 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10056 struct extent_io_tree *io_tree = &inode->io_tree;
10057 struct page **pages;
10058 unsigned long nr_pages, i;
10060 size_t page_offset;
10063 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10064 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10067 ret = btrfs_alloc_page_array(nr_pages, pages);
10073 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10074 disk_io_size, pages);
10078 unlock_extent(io_tree, start, lockend, cached_state);
10079 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10086 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10087 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10090 while (cur < count) {
10091 size_t bytes = min_t(size_t, count - cur,
10092 PAGE_SIZE - page_offset);
10094 if (copy_page_to_iter(pages[i], page_offset, bytes,
10105 for (i = 0; i < nr_pages; i++) {
10107 __free_page(pages[i]);
10113 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10114 struct btrfs_ioctl_encoded_io_args *encoded)
10116 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10117 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10118 struct extent_io_tree *io_tree = &inode->io_tree;
10120 size_t count = iov_iter_count(iter);
10121 u64 start, lockend, disk_bytenr, disk_io_size;
10122 struct extent_state *cached_state = NULL;
10123 struct extent_map *em;
10124 bool unlocked = false;
10126 file_accessed(iocb->ki_filp);
10128 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10130 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10131 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10134 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10136 * We don't know how long the extent containing iocb->ki_pos is, but if
10137 * it's compressed we know that it won't be longer than this.
10139 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10142 struct btrfs_ordered_extent *ordered;
10144 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10145 lockend - start + 1);
10147 goto out_unlock_inode;
10148 lock_extent(io_tree, start, lockend, &cached_state);
10149 ordered = btrfs_lookup_ordered_range(inode, start,
10150 lockend - start + 1);
10153 btrfs_put_ordered_extent(ordered);
10154 unlock_extent(io_tree, start, lockend, &cached_state);
10158 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10161 goto out_unlock_extent;
10164 if (em->block_start == EXTENT_MAP_INLINE) {
10165 u64 extent_start = em->start;
10168 * For inline extents we get everything we need out of the
10171 free_extent_map(em);
10173 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10174 &cached_state, extent_start,
10175 count, encoded, &unlocked);
10180 * We only want to return up to EOF even if the extent extends beyond
10183 encoded->len = min_t(u64, extent_map_end(em),
10184 inode->vfs_inode.i_size) - iocb->ki_pos;
10185 if (em->block_start == EXTENT_MAP_HOLE ||
10186 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10187 disk_bytenr = EXTENT_MAP_HOLE;
10188 count = min_t(u64, count, encoded->len);
10189 encoded->len = count;
10190 encoded->unencoded_len = count;
10191 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10192 disk_bytenr = em->block_start;
10194 * Bail if the buffer isn't large enough to return the whole
10195 * compressed extent.
10197 if (em->block_len > count) {
10201 disk_io_size = em->block_len;
10202 count = em->block_len;
10203 encoded->unencoded_len = em->ram_bytes;
10204 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10205 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10206 em->compress_type);
10209 encoded->compression = ret;
10211 disk_bytenr = em->block_start + (start - em->start);
10212 if (encoded->len > count)
10213 encoded->len = count;
10215 * Don't read beyond what we locked. This also limits the page
10216 * allocations that we'll do.
10218 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10219 count = start + disk_io_size - iocb->ki_pos;
10220 encoded->len = count;
10221 encoded->unencoded_len = count;
10222 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10224 free_extent_map(em);
10227 if (disk_bytenr == EXTENT_MAP_HOLE) {
10228 unlock_extent(io_tree, start, lockend, &cached_state);
10229 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10231 ret = iov_iter_zero(count, iter);
10235 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10236 &cached_state, disk_bytenr,
10237 disk_io_size, count,
10238 encoded->compression,
10244 iocb->ki_pos += encoded->len;
10246 free_extent_map(em);
10249 unlock_extent(io_tree, start, lockend, &cached_state);
10252 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10256 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10257 const struct btrfs_ioctl_encoded_io_args *encoded)
10259 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10260 struct btrfs_root *root = inode->root;
10261 struct btrfs_fs_info *fs_info = root->fs_info;
10262 struct extent_io_tree *io_tree = &inode->io_tree;
10263 struct extent_changeset *data_reserved = NULL;
10264 struct extent_state *cached_state = NULL;
10265 struct btrfs_ordered_extent *ordered;
10269 u64 num_bytes, ram_bytes, disk_num_bytes;
10270 unsigned long nr_pages, i;
10271 struct page **pages;
10272 struct btrfs_key ins;
10273 bool extent_reserved = false;
10274 struct extent_map *em;
10277 switch (encoded->compression) {
10278 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10279 compression = BTRFS_COMPRESS_ZLIB;
10281 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10282 compression = BTRFS_COMPRESS_ZSTD;
10284 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10285 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10286 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10287 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10288 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10289 /* The sector size must match for LZO. */
10290 if (encoded->compression -
10291 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10292 fs_info->sectorsize_bits)
10294 compression = BTRFS_COMPRESS_LZO;
10299 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10302 orig_count = iov_iter_count(from);
10304 /* The extent size must be sane. */
10305 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10306 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10310 * The compressed data must be smaller than the decompressed data.
10312 * It's of course possible for data to compress to larger or the same
10313 * size, but the buffered I/O path falls back to no compression for such
10314 * data, and we don't want to break any assumptions by creating these
10317 * Note that this is less strict than the current check we have that the
10318 * compressed data must be at least one sector smaller than the
10319 * decompressed data. We only want to enforce the weaker requirement
10320 * from old kernels that it is at least one byte smaller.
10322 if (orig_count >= encoded->unencoded_len)
10325 /* The extent must start on a sector boundary. */
10326 start = iocb->ki_pos;
10327 if (!IS_ALIGNED(start, fs_info->sectorsize))
10331 * The extent must end on a sector boundary. However, we allow a write
10332 * which ends at or extends i_size to have an unaligned length; we round
10333 * up the extent size and set i_size to the unaligned end.
10335 if (start + encoded->len < inode->vfs_inode.i_size &&
10336 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10339 /* Finally, the offset in the unencoded data must be sector-aligned. */
10340 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10343 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10344 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10345 end = start + num_bytes - 1;
10348 * If the extent cannot be inline, the compressed data on disk must be
10349 * sector-aligned. For convenience, we extend it with zeroes if it
10352 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10353 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10354 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10357 for (i = 0; i < nr_pages; i++) {
10358 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10361 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10366 kaddr = kmap_local_page(pages[i]);
10367 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10368 kunmap_local(kaddr);
10372 if (bytes < PAGE_SIZE)
10373 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10374 kunmap_local(kaddr);
10378 struct btrfs_ordered_extent *ordered;
10380 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10383 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10384 start >> PAGE_SHIFT,
10385 end >> PAGE_SHIFT);
10388 lock_extent(io_tree, start, end, &cached_state);
10389 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10391 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10394 btrfs_put_ordered_extent(ordered);
10395 unlock_extent(io_tree, start, end, &cached_state);
10400 * We don't use the higher-level delalloc space functions because our
10401 * num_bytes and disk_num_bytes are different.
10403 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10406 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10408 goto out_free_data_space;
10409 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10412 goto out_qgroup_free_data;
10414 /* Try an inline extent first. */
10415 if (start == 0 && encoded->unencoded_len == encoded->len &&
10416 encoded->unencoded_offset == 0) {
10417 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10418 compression, pages, true);
10422 goto out_delalloc_release;
10426 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10427 disk_num_bytes, 0, 0, &ins, 1, 1);
10429 goto out_delalloc_release;
10430 extent_reserved = true;
10432 em = create_io_em(inode, start, num_bytes,
10433 start - encoded->unencoded_offset, ins.objectid,
10434 ins.offset, ins.offset, ram_bytes, compression,
10435 BTRFS_ORDERED_COMPRESSED);
10438 goto out_free_reserved;
10440 free_extent_map(em);
10442 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10443 ins.objectid, ins.offset,
10444 encoded->unencoded_offset,
10445 (1 << BTRFS_ORDERED_ENCODED) |
10446 (1 << BTRFS_ORDERED_COMPRESSED),
10448 if (IS_ERR(ordered)) {
10449 btrfs_drop_extent_map_range(inode, start, end, false);
10450 ret = PTR_ERR(ordered);
10451 goto out_free_reserved;
10453 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10455 if (start + encoded->len > inode->vfs_inode.i_size)
10456 i_size_write(&inode->vfs_inode, start + encoded->len);
10458 unlock_extent(io_tree, start, end, &cached_state);
10460 btrfs_delalloc_release_extents(inode, num_bytes);
10462 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10467 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10468 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10469 out_delalloc_release:
10470 btrfs_delalloc_release_extents(inode, num_bytes);
10471 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10472 out_qgroup_free_data:
10474 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10475 out_free_data_space:
10477 * If btrfs_reserve_extent() succeeded, then we already decremented
10480 if (!extent_reserved)
10481 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10483 unlock_extent(io_tree, start, end, &cached_state);
10485 for (i = 0; i < nr_pages; i++) {
10487 __free_page(pages[i]);
10492 iocb->ki_pos += encoded->len;
10498 * Add an entry indicating a block group or device which is pinned by a
10499 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10500 * negative errno on failure.
10502 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10503 bool is_block_group)
10505 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10506 struct btrfs_swapfile_pin *sp, *entry;
10507 struct rb_node **p;
10508 struct rb_node *parent = NULL;
10510 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10515 sp->is_block_group = is_block_group;
10516 sp->bg_extent_count = 1;
10518 spin_lock(&fs_info->swapfile_pins_lock);
10519 p = &fs_info->swapfile_pins.rb_node;
10522 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10523 if (sp->ptr < entry->ptr ||
10524 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10525 p = &(*p)->rb_left;
10526 } else if (sp->ptr > entry->ptr ||
10527 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10528 p = &(*p)->rb_right;
10530 if (is_block_group)
10531 entry->bg_extent_count++;
10532 spin_unlock(&fs_info->swapfile_pins_lock);
10537 rb_link_node(&sp->node, parent, p);
10538 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10539 spin_unlock(&fs_info->swapfile_pins_lock);
10543 /* Free all of the entries pinned by this swapfile. */
10544 static void btrfs_free_swapfile_pins(struct inode *inode)
10546 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10547 struct btrfs_swapfile_pin *sp;
10548 struct rb_node *node, *next;
10550 spin_lock(&fs_info->swapfile_pins_lock);
10551 node = rb_first(&fs_info->swapfile_pins);
10553 next = rb_next(node);
10554 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10555 if (sp->inode == inode) {
10556 rb_erase(&sp->node, &fs_info->swapfile_pins);
10557 if (sp->is_block_group) {
10558 btrfs_dec_block_group_swap_extents(sp->ptr,
10559 sp->bg_extent_count);
10560 btrfs_put_block_group(sp->ptr);
10566 spin_unlock(&fs_info->swapfile_pins_lock);
10569 struct btrfs_swap_info {
10575 unsigned long nr_pages;
10579 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10580 struct btrfs_swap_info *bsi)
10582 unsigned long nr_pages;
10583 unsigned long max_pages;
10584 u64 first_ppage, first_ppage_reported, next_ppage;
10588 * Our swapfile may have had its size extended after the swap header was
10589 * written. In that case activating the swapfile should not go beyond
10590 * the max size set in the swap header.
10592 if (bsi->nr_pages >= sis->max)
10595 max_pages = sis->max - bsi->nr_pages;
10596 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10597 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10599 if (first_ppage >= next_ppage)
10601 nr_pages = next_ppage - first_ppage;
10602 nr_pages = min(nr_pages, max_pages);
10604 first_ppage_reported = first_ppage;
10605 if (bsi->start == 0)
10606 first_ppage_reported++;
10607 if (bsi->lowest_ppage > first_ppage_reported)
10608 bsi->lowest_ppage = first_ppage_reported;
10609 if (bsi->highest_ppage < (next_ppage - 1))
10610 bsi->highest_ppage = next_ppage - 1;
10612 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10615 bsi->nr_extents += ret;
10616 bsi->nr_pages += nr_pages;
10620 static void btrfs_swap_deactivate(struct file *file)
10622 struct inode *inode = file_inode(file);
10624 btrfs_free_swapfile_pins(inode);
10625 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10628 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10631 struct inode *inode = file_inode(file);
10632 struct btrfs_root *root = BTRFS_I(inode)->root;
10633 struct btrfs_fs_info *fs_info = root->fs_info;
10634 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10635 struct extent_state *cached_state = NULL;
10636 struct extent_map *em = NULL;
10637 struct btrfs_device *device = NULL;
10638 struct btrfs_swap_info bsi = {
10639 .lowest_ppage = (sector_t)-1ULL,
10646 * If the swap file was just created, make sure delalloc is done. If the
10647 * file changes again after this, the user is doing something stupid and
10648 * we don't really care.
10650 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10655 * The inode is locked, so these flags won't change after we check them.
10657 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10658 btrfs_warn(fs_info, "swapfile must not be compressed");
10661 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10662 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10665 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10666 btrfs_warn(fs_info, "swapfile must not be checksummed");
10671 * Balance or device remove/replace/resize can move stuff around from
10672 * under us. The exclop protection makes sure they aren't running/won't
10673 * run concurrently while we are mapping the swap extents, and
10674 * fs_info->swapfile_pins prevents them from running while the swap
10675 * file is active and moving the extents. Note that this also prevents
10676 * a concurrent device add which isn't actually necessary, but it's not
10677 * really worth the trouble to allow it.
10679 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10680 btrfs_warn(fs_info,
10681 "cannot activate swapfile while exclusive operation is running");
10686 * Prevent snapshot creation while we are activating the swap file.
10687 * We do not want to race with snapshot creation. If snapshot creation
10688 * already started before we bumped nr_swapfiles from 0 to 1 and
10689 * completes before the first write into the swap file after it is
10690 * activated, than that write would fallback to COW.
10692 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10693 btrfs_exclop_finish(fs_info);
10694 btrfs_warn(fs_info,
10695 "cannot activate swapfile because snapshot creation is in progress");
10699 * Snapshots can create extents which require COW even if NODATACOW is
10700 * set. We use this counter to prevent snapshots. We must increment it
10701 * before walking the extents because we don't want a concurrent
10702 * snapshot to run after we've already checked the extents.
10704 * It is possible that subvolume is marked for deletion but still not
10705 * removed yet. To prevent this race, we check the root status before
10706 * activating the swapfile.
10708 spin_lock(&root->root_item_lock);
10709 if (btrfs_root_dead(root)) {
10710 spin_unlock(&root->root_item_lock);
10712 btrfs_exclop_finish(fs_info);
10713 btrfs_warn(fs_info,
10714 "cannot activate swapfile because subvolume %llu is being deleted",
10715 root->root_key.objectid);
10718 atomic_inc(&root->nr_swapfiles);
10719 spin_unlock(&root->root_item_lock);
10721 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10723 lock_extent(io_tree, 0, isize - 1, &cached_state);
10725 while (start < isize) {
10726 u64 logical_block_start, physical_block_start;
10727 struct btrfs_block_group *bg;
10728 u64 len = isize - start;
10730 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10736 if (em->block_start == EXTENT_MAP_HOLE) {
10737 btrfs_warn(fs_info, "swapfile must not have holes");
10741 if (em->block_start == EXTENT_MAP_INLINE) {
10743 * It's unlikely we'll ever actually find ourselves
10744 * here, as a file small enough to fit inline won't be
10745 * big enough to store more than the swap header, but in
10746 * case something changes in the future, let's catch it
10747 * here rather than later.
10749 btrfs_warn(fs_info, "swapfile must not be inline");
10753 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10754 btrfs_warn(fs_info, "swapfile must not be compressed");
10759 logical_block_start = em->block_start + (start - em->start);
10760 len = min(len, em->len - (start - em->start));
10761 free_extent_map(em);
10764 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10770 btrfs_warn(fs_info,
10771 "swapfile must not be copy-on-write");
10776 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10782 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10783 btrfs_warn(fs_info,
10784 "swapfile must have single data profile");
10789 if (device == NULL) {
10790 device = em->map_lookup->stripes[0].dev;
10791 ret = btrfs_add_swapfile_pin(inode, device, false);
10796 } else if (device != em->map_lookup->stripes[0].dev) {
10797 btrfs_warn(fs_info, "swapfile must be on one device");
10802 physical_block_start = (em->map_lookup->stripes[0].physical +
10803 (logical_block_start - em->start));
10804 len = min(len, em->len - (logical_block_start - em->start));
10805 free_extent_map(em);
10808 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10810 btrfs_warn(fs_info,
10811 "could not find block group containing swapfile");
10816 if (!btrfs_inc_block_group_swap_extents(bg)) {
10817 btrfs_warn(fs_info,
10818 "block group for swapfile at %llu is read-only%s",
10820 atomic_read(&fs_info->scrubs_running) ?
10821 " (scrub running)" : "");
10822 btrfs_put_block_group(bg);
10827 ret = btrfs_add_swapfile_pin(inode, bg, true);
10829 btrfs_put_block_group(bg);
10836 if (bsi.block_len &&
10837 bsi.block_start + bsi.block_len == physical_block_start) {
10838 bsi.block_len += len;
10840 if (bsi.block_len) {
10841 ret = btrfs_add_swap_extent(sis, &bsi);
10846 bsi.block_start = physical_block_start;
10847 bsi.block_len = len;
10854 ret = btrfs_add_swap_extent(sis, &bsi);
10857 if (!IS_ERR_OR_NULL(em))
10858 free_extent_map(em);
10860 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10863 btrfs_swap_deactivate(file);
10865 btrfs_drew_write_unlock(&root->snapshot_lock);
10867 btrfs_exclop_finish(fs_info);
10873 sis->bdev = device->bdev;
10874 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10875 sis->max = bsi.nr_pages;
10876 sis->pages = bsi.nr_pages - 1;
10877 sis->highest_bit = bsi.nr_pages - 1;
10878 return bsi.nr_extents;
10881 static void btrfs_swap_deactivate(struct file *file)
10885 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10888 return -EOPNOTSUPP;
10893 * Update the number of bytes used in the VFS' inode. When we replace extents in
10894 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10895 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10896 * always get a correct value.
10898 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10899 const u64 add_bytes,
10900 const u64 del_bytes)
10902 if (add_bytes == del_bytes)
10905 spin_lock(&inode->lock);
10907 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10909 inode_add_bytes(&inode->vfs_inode, add_bytes);
10910 spin_unlock(&inode->lock);
10914 * Verify that there are no ordered extents for a given file range.
10916 * @inode: The target inode.
10917 * @start: Start offset of the file range, should be sector size aligned.
10918 * @end: End offset (inclusive) of the file range, its value +1 should be
10919 * sector size aligned.
10921 * This should typically be used for cases where we locked an inode's VFS lock in
10922 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10923 * we have flushed all delalloc in the range, we have waited for all ordered
10924 * extents in the range to complete and finally we have locked the file range in
10925 * the inode's io_tree.
10927 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10929 struct btrfs_root *root = inode->root;
10930 struct btrfs_ordered_extent *ordered;
10932 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10935 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10937 btrfs_err(root->fs_info,
10938 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10939 start, end, btrfs_ino(inode), root->root_key.objectid,
10940 ordered->file_offset,
10941 ordered->file_offset + ordered->num_bytes - 1);
10942 btrfs_put_ordered_extent(ordered);
10945 ASSERT(ordered == NULL);
10948 static const struct inode_operations btrfs_dir_inode_operations = {
10949 .getattr = btrfs_getattr,
10950 .lookup = btrfs_lookup,
10951 .create = btrfs_create,
10952 .unlink = btrfs_unlink,
10953 .link = btrfs_link,
10954 .mkdir = btrfs_mkdir,
10955 .rmdir = btrfs_rmdir,
10956 .rename = btrfs_rename2,
10957 .symlink = btrfs_symlink,
10958 .setattr = btrfs_setattr,
10959 .mknod = btrfs_mknod,
10960 .listxattr = btrfs_listxattr,
10961 .permission = btrfs_permission,
10962 .get_inode_acl = btrfs_get_acl,
10963 .set_acl = btrfs_set_acl,
10964 .update_time = btrfs_update_time,
10965 .tmpfile = btrfs_tmpfile,
10966 .fileattr_get = btrfs_fileattr_get,
10967 .fileattr_set = btrfs_fileattr_set,
10970 static const struct file_operations btrfs_dir_file_operations = {
10971 .llseek = generic_file_llseek,
10972 .read = generic_read_dir,
10973 .iterate_shared = btrfs_real_readdir,
10974 .open = btrfs_opendir,
10975 .unlocked_ioctl = btrfs_ioctl,
10976 #ifdef CONFIG_COMPAT
10977 .compat_ioctl = btrfs_compat_ioctl,
10979 .release = btrfs_release_file,
10980 .fsync = btrfs_sync_file,
10984 * btrfs doesn't support the bmap operation because swapfiles
10985 * use bmap to make a mapping of extents in the file. They assume
10986 * these extents won't change over the life of the file and they
10987 * use the bmap result to do IO directly to the drive.
10989 * the btrfs bmap call would return logical addresses that aren't
10990 * suitable for IO and they also will change frequently as COW
10991 * operations happen. So, swapfile + btrfs == corruption.
10993 * For now we're avoiding this by dropping bmap.
10995 static const struct address_space_operations btrfs_aops = {
10996 .read_folio = btrfs_read_folio,
10997 .writepages = btrfs_writepages,
10998 .readahead = btrfs_readahead,
10999 .invalidate_folio = btrfs_invalidate_folio,
11000 .release_folio = btrfs_release_folio,
11001 .migrate_folio = btrfs_migrate_folio,
11002 .dirty_folio = filemap_dirty_folio,
11003 .error_remove_page = generic_error_remove_page,
11004 .swap_activate = btrfs_swap_activate,
11005 .swap_deactivate = btrfs_swap_deactivate,
11008 static const struct inode_operations btrfs_file_inode_operations = {
11009 .getattr = btrfs_getattr,
11010 .setattr = btrfs_setattr,
11011 .listxattr = btrfs_listxattr,
11012 .permission = btrfs_permission,
11013 .fiemap = btrfs_fiemap,
11014 .get_inode_acl = btrfs_get_acl,
11015 .set_acl = btrfs_set_acl,
11016 .update_time = btrfs_update_time,
11017 .fileattr_get = btrfs_fileattr_get,
11018 .fileattr_set = btrfs_fileattr_set,
11020 static const struct inode_operations btrfs_special_inode_operations = {
11021 .getattr = btrfs_getattr,
11022 .setattr = btrfs_setattr,
11023 .permission = btrfs_permission,
11024 .listxattr = btrfs_listxattr,
11025 .get_inode_acl = btrfs_get_acl,
11026 .set_acl = btrfs_set_acl,
11027 .update_time = btrfs_update_time,
11029 static const struct inode_operations btrfs_symlink_inode_operations = {
11030 .get_link = page_get_link,
11031 .getattr = btrfs_getattr,
11032 .setattr = btrfs_setattr,
11033 .permission = btrfs_permission,
11034 .listxattr = btrfs_listxattr,
11035 .update_time = btrfs_update_time,
11038 const struct dentry_operations btrfs_dentry_operations = {
11039 .d_delete = btrfs_dentry_delete,