1	// SPDX-License-Identifier: GPL-2.0
2	/*
3	* Copyright (C) 2007 Oracle. All rights reserved.
4	*/
5
6	#include <crypto/hash.h>
7	#include <linux/kernel.h>
8	#include <linux/bio.h>
9	#include <linux/blk-cgroup.h>
10	#include <linux/file.h>
11	#include <linux/fs.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>
37	#include "misc.h"
38	#include "ctree.h"
39	#include "disk-io.h"
40	#include "transaction.h"
41	#include "btrfs_inode.h"
42	#include "ordered-data.h"
43	#include "xattr.h"
44	#include "tree-log.h"
45	#include "bio.h"
46	#include "compression.h"
47	#include "locking.h"
48	#include "props.h"
49	#include "qgroup.h"
50	#include "delalloc-space.h"
51	#include "block-group.h"
52	#include "space-info.h"
53	#include "zoned.h"
54	#include "subpage.h"
55	#include "inode-item.h"
56	#include "fs.h"
57	#include "accessors.h"
58	#include "extent-tree.h"
59	#include "root-tree.h"
60	#include "defrag.h"
61	#include "dir-item.h"
62	#include "file-item.h"
63	#include "uuid-tree.h"
64	#include "ioctl.h"
65	#include "file.h"
66	#include "acl.h"
67	#include "relocation.h"
68	#include "verity.h"
69	#include "super.h"
70	#include "orphan.h"
71	#include "backref.h"
72	#include "raid-stripe-tree.h"
73
74	struct btrfs_iget_args {
75	u64 ino;
76	struct btrfs_root *root;
77	};
78
79	struct btrfs_dio_data {
80	ssize_t submitted;
81	struct extent_changeset *data_reserved;
82	struct btrfs_ordered_extent *ordered;
83	bool data_space_reserved;
84	bool nocow_done;
85	};
86
87	struct btrfs_dio_private {
88	/* Range of I/O */
89	u64 file_offset;
90	u32 bytes;
91
92	/* This must be last */
93	struct btrfs_bio bbio;
94	};
95
96	static struct bio_set btrfs_dio_bioset;
97
98	struct btrfs_rename_ctx {
99	/* Output field. Stores the index number of the old directory entry. */
100	u64 index;
101	};
102
103	/*
104	* Used by data_reloc_print_warning_inode() to pass needed info for filename
105	* resolution and output of error message.
106	*/
107	struct data_reloc_warn {
108	struct btrfs_path path;
109	struct btrfs_fs_info *fs_info;
110	u64 extent_item_size;
111	u64 logical;
112	int mirror_num;
113	};
114
115	/*
116	* For the file_extent_tree, we want to hold the inode lock when we lookup and
117	* update the disk_i_size, but lockdep will complain because our io_tree we hold
118	* the tree lock and get the inode lock when setting delalloc. These two things
119	* are unrelated, so make a class for the file_extent_tree so we don't get the
120	* two locking patterns mixed up.
121	*/
122	static struct lock_class_key file_extent_tree_class;
123
124	static const struct inode_operations btrfs_dir_inode_operations;
125	static const struct inode_operations btrfs_symlink_inode_operations;
126	static const struct inode_operations btrfs_special_inode_operations;
127	static const struct inode_operations btrfs_file_inode_operations;
128	static const struct address_space_operations btrfs_aops;
129	static const struct file_operations btrfs_dir_file_operations;
130
131	static struct kmem_cache *btrfs_inode_cachep;
132
133	static int btrfs_setsize(struct inode *inode, struct iattr *attr);
134	static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback);
135
136	static noinline int run_delalloc_cow(struct btrfs_inode *inode,
137	struct page *locked_page, u64 start,
138	u64 end, struct writeback_control *wbc,
139	bool pages_dirty);
140	static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
141	u64 len, u64 orig_start, u64 block_start,
142	u64 block_len, u64 orig_block_len,
143	u64 ram_bytes, int compress_type,
144	int type);
145
146	static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
147	u64 root, void *warn_ctx)
148	{
149	struct data_reloc_warn *warn = warn_ctx;
150	struct btrfs_fs_info *fs_info = warn->fs_info;
151	struct extent_buffer *eb;
152	struct btrfs_inode_item *inode_item;
153	struct inode_fs_paths *ipath = NULL;
154	struct btrfs_root *local_root;
155	struct btrfs_key key;
156	unsigned int nofs_flag;
157	u32 nlink;
158	int ret;
159
160	local_root = btrfs_get_fs_root(fs_info, objectid: root, check_ref: true);
161	if (IS_ERR(ptr: local_root)) {
162	ret = PTR_ERR(ptr: local_root);
163	goto err;
164	}
165
166	/* This makes the path point to (inum INODE_ITEM ioff). */
167	key.objectid = inum;
168	key.type = BTRFS_INODE_ITEM_KEY;
169	key.offset = 0;
170
171	ret = btrfs_search_slot(NULL, root: local_root, key: &key, p: &warn->path, ins_len: 0, cow: 0);
172	if (ret) {
173	btrfs_put_root(root: local_root);
174	btrfs_release_path(p: &warn->path);
175	goto err;
176	}
177
178	eb = warn->path.nodes[0];
179	inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item);
180	nlink = btrfs_inode_nlink(eb, s: inode_item);
181	btrfs_release_path(p: &warn->path);
182
183	nofs_flag = memalloc_nofs_save();
184	ipath = init_ipath(total_bytes: 4096, fs_root: local_root, path: &warn->path);
185	memalloc_nofs_restore(flags: nofs_flag);
186	if (IS_ERR(ptr: ipath)) {
187	btrfs_put_root(root: local_root);
188	ret = PTR_ERR(ptr: ipath);
189	ipath = NULL;
190	/*
191	* -ENOMEM, not a critical error, just output an generic error
192	* without filename.
193	*/
194	btrfs_warn(fs_info,
195	"checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu",
196	warn->logical, warn->mirror_num, root, inum, offset);
197	return ret;
198	}
199	ret = paths_from_inode(inum, ipath);
200	if (ret < 0)
201	goto err;
202
203	/*
204	* We deliberately ignore the bit ipath might have been too small to
205	* hold all of the paths here
206	*/
207	for (int i = 0; i < ipath->fspath->elem_cnt; i++) {
208	btrfs_warn(fs_info,
209	"checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)",
210	warn->logical, warn->mirror_num, root, inum, offset,
211	fs_info->sectorsize, nlink,
212	(char *)(unsigned long)ipath->fspath->val[i]);
213	}
214
215	btrfs_put_root(root: local_root);
216	free_ipath(ipath);
217	return 0;
218
219	err:
220	btrfs_warn(fs_info,
221	"checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d",
222	warn->logical, warn->mirror_num, root, inum, offset, ret);
223
224	free_ipath(ipath);
225	return ret;
226	}
227
228	/*
229	* Do extra user-friendly error output (e.g. lookup all the affected files).
230	*
231	* Return true if we succeeded doing the backref lookup.
232	* Return false if such lookup failed, and has to fallback to the old error message.
233	*/
234	static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off,
235	const u8 *csum, const u8 *csum_expected,
236	int mirror_num)
237	{
238	struct btrfs_fs_info *fs_info = inode->root->fs_info;
239	struct btrfs_path path = { 0 };
240	struct btrfs_key found_key = { 0 };
241	struct extent_buffer *eb;
242	struct btrfs_extent_item *ei;
243	const u32 csum_size = fs_info->csum_size;
244	u64 logical;
245	u64 flags;
246	u32 item_size;
247	int ret;
248
249	mutex_lock(&fs_info->reloc_mutex);
250	logical = btrfs_get_reloc_bg_bytenr(fs_info);
251	mutex_unlock(lock: &fs_info->reloc_mutex);
252
253	if (logical == U64_MAX) {
254	btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation");
255	btrfs_warn_rl(fs_info,
256	"csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
257	inode->root->root_key.objectid, btrfs_ino(inode), file_off,
258	CSUM_FMT_VALUE(csum_size, csum),
259	CSUM_FMT_VALUE(csum_size, csum_expected),
260	mirror_num);
261	return;
262	}
263
264	logical += file_off;
265	btrfs_warn_rl(fs_info,
266	"csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
267	inode->root->root_key.objectid,
268	btrfs_ino(inode), file_off, logical,
269	CSUM_FMT_VALUE(csum_size, csum),
270	CSUM_FMT_VALUE(csum_size, csum_expected),
271	mirror_num);
272
273	ret = extent_from_logical(fs_info, logical, path: &path, found_key: &found_key, flags: &flags);
274	if (ret < 0) {
275	btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d",
276	logical, ret);
277	return;
278	}
279	eb = path.nodes[0];
280	ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
281	item_size = btrfs_item_size(eb, slot: path.slots[0]);
282	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
283	unsigned long ptr = 0;
284	u64 ref_root;
285	u8 ref_level;
286
287	while (true) {
288	ret = tree_backref_for_extent(ptr: &ptr, eb, key: &found_key, ei,
289	item_size, out_root: &ref_root,
290	out_level: &ref_level);
291	if (ret < 0) {
292	btrfs_warn_rl(fs_info,
293	"failed to resolve tree backref for logical %llu: %d",
294	logical, ret);
295	break;
296	}
297	if (ret > 0)
298	break;
299
300	btrfs_warn_rl(fs_info,
301	"csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu",
302	logical, mirror_num,
303	(ref_level ? "node" : "leaf"),
304	ref_level, ref_root);
305	}
306	btrfs_release_path(p: &path);
307	} else {
308	struct btrfs_backref_walk_ctx ctx = { 0 };
309	struct data_reloc_warn reloc_warn = { 0 };
310
311	btrfs_release_path(p: &path);
312
313	ctx.bytenr = found_key.objectid;
314	ctx.extent_item_pos = logical - found_key.objectid;
315	ctx.fs_info = fs_info;
316
317	reloc_warn.logical = logical;
318	reloc_warn.extent_item_size = found_key.offset;
319	reloc_warn.mirror_num = mirror_num;
320	reloc_warn.fs_info = fs_info;
321
322	iterate_extent_inodes(ctx: &ctx, search_commit_root: true,
323	iterate: data_reloc_print_warning_inode, user_ctx: &reloc_warn);
324	}
325	}
326
327	static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode,
328	u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
329	{
330	struct btrfs_root *root = inode->root;
331	const u32 csum_size = root->fs_info->csum_size;
332
333	/* For data reloc tree, it's better to do a backref lookup instead. */
334	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
335	return print_data_reloc_error(inode, file_off: logical_start, csum,
336	csum_expected, mirror_num);
337
338	/* Output without objectid, which is more meaningful */
339	if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) {
340	btrfs_warn_rl(root->fs_info,
341	"csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
342	root->root_key.objectid, btrfs_ino(inode),
343	logical_start,
344	CSUM_FMT_VALUE(csum_size, csum),
345	CSUM_FMT_VALUE(csum_size, csum_expected),
346	mirror_num);
347	} else {
348	btrfs_warn_rl(root->fs_info,
349	"csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
350	root->root_key.objectid, btrfs_ino(inode),
351	logical_start,
352	CSUM_FMT_VALUE(csum_size, csum),
353	CSUM_FMT_VALUE(csum_size, csum_expected),
354	mirror_num);
355	}
356	}
357
358	/*
359	* Lock inode i_rwsem based on arguments passed.
360	*
361	* ilock_flags can have the following bit set:
362	*
363	* BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
364	* BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
365	* return -EAGAIN
366	* BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
367	*/
368	int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
369	{
370	if (ilock_flags & BTRFS_ILOCK_SHARED) {
371	if (ilock_flags & BTRFS_ILOCK_TRY) {
372	if (!inode_trylock_shared(inode: &inode->vfs_inode))
373	return -EAGAIN;
374	else
375	return 0;
376	}
377	inode_lock_shared(inode: &inode->vfs_inode);
378	} else {
379	if (ilock_flags & BTRFS_ILOCK_TRY) {
380	if (!inode_trylock(inode: &inode->vfs_inode))
381	return -EAGAIN;
382	else
383	return 0;
384	}
385	inode_lock(inode: &inode->vfs_inode);
386	}
387	if (ilock_flags & BTRFS_ILOCK_MMAP)
388	down_write(sem: &inode->i_mmap_lock);
389	return 0;
390	}
391
392	/*
393	* Unock inode i_rwsem.
394	*
395	* ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
396	* to decide whether the lock acquired is shared or exclusive.
397	*/
398	void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
399	{
400	if (ilock_flags & BTRFS_ILOCK_MMAP)
401	up_write(sem: &inode->i_mmap_lock);
402	if (ilock_flags & BTRFS_ILOCK_SHARED)
403	inode_unlock_shared(inode: &inode->vfs_inode);
404	else
405	inode_unlock(inode: &inode->vfs_inode);
406	}
407
408	/*
409	* Cleanup all submitted ordered extents in specified range to handle errors
410	* from the btrfs_run_delalloc_range() callback.
411	*
412	* NOTE: caller must ensure that when an error happens, it can not call
413	* extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
414	* and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
415	* to be released, which we want to happen only when finishing the ordered
416	* extent (btrfs_finish_ordered_io()).
417	*/
418	static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
419	struct page *locked_page,
420	u64 offset, u64 bytes)
421	{
422	unsigned long index = offset >> PAGE_SHIFT;
423	unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
424	u64 page_start = 0, page_end = 0;
425	struct page *page;
426
427	if (locked_page) {
428	page_start = page_offset(page: locked_page);
429	page_end = page_start + PAGE_SIZE - 1;
430	}
431
432	while (index <= end_index) {
433	/*
434	* For locked page, we will call btrfs_mark_ordered_io_finished
435	* through btrfs_mark_ordered_io_finished() on it
436	* in run_delalloc_range() for the error handling, which will
437	* clear page Ordered and run the ordered extent accounting.
438	*
439	* Here we can't just clear the Ordered bit, or
440	* btrfs_mark_ordered_io_finished() would skip the accounting
441	* for the page range, and the ordered extent will never finish.
442	*/
443	if (locked_page && index == (page_start >> PAGE_SHIFT)) {
444	index++;
445	continue;
446	}
447	page = find_get_page(mapping: inode->vfs_inode.i_mapping, offset: index);
448	index++;
449	if (!page)
450	continue;
451
452	/*
453	* Here we just clear all Ordered bits for every page in the
454	* range, then btrfs_mark_ordered_io_finished() will handle
455	* the ordered extent accounting for the range.
456	*/
457	btrfs_folio_clamp_clear_ordered(fs_info: inode->root->fs_info,
458	page_folio(page), start: offset, len: bytes);
459	put_page(page);
460	}
461
462	if (locked_page) {
463	/* The locked page covers the full range, nothing needs to be done */
464	if (bytes + offset <= page_start + PAGE_SIZE)
465	return;
466	/*
467	* In case this page belongs to the delalloc range being
468	* instantiated then skip it, since the first page of a range is
469	* going to be properly cleaned up by the caller of
470	* run_delalloc_range
471	*/
472	if (page_start >= offset && page_end <= (offset + bytes - 1)) {
473	bytes = offset + bytes - page_offset(page: locked_page) - PAGE_SIZE;
474	offset = page_offset(page: locked_page) + PAGE_SIZE;
475	}
476	}
477
478	return btrfs_mark_ordered_io_finished(inode, NULL, file_offset: offset, num_bytes: bytes, uptodate: false);
479	}
480
481	static int btrfs_dirty_inode(struct btrfs_inode *inode);
482
483	static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
484	struct btrfs_new_inode_args *args)
485	{
486	int err;
487
488	if (args->default_acl) {
489	err = __btrfs_set_acl(trans, inode: args->inode, acl: args->default_acl,
490	ACL_TYPE_DEFAULT);
491	if (err)
492	return err;
493	}
494	if (args->acl) {
495	err = __btrfs_set_acl(trans, inode: args->inode, acl: args->acl, ACL_TYPE_ACCESS);
496	if (err)
497	return err;
498	}
499	if (!args->default_acl && !args->acl)
500	cache_no_acl(inode: args->inode);
501	return btrfs_xattr_security_init(trans, inode: args->inode, dir: args->dir,
502	qstr: &args->dentry->d_name);
503	}
504
505	/*
506	* this does all the hard work for inserting an inline extent into
507	* the btree. The caller should have done a btrfs_drop_extents so that
508	* no overlapping inline items exist in the btree
509	*/
510	static int insert_inline_extent(struct btrfs_trans_handle *trans,
511	struct btrfs_path *path,
512	struct btrfs_inode *inode, bool extent_inserted,
513	size_t size, size_t compressed_size,
514	int compress_type,
515	struct page **compressed_pages,
516	bool update_i_size)
517	{
518	struct btrfs_root *root = inode->root;
519	struct extent_buffer *leaf;
520	struct page *page = NULL;
521	char *kaddr;
522	unsigned long ptr;
523	struct btrfs_file_extent_item *ei;
524	int ret;
525	size_t cur_size = size;
526	u64 i_size;
527
528	ASSERT((compressed_size > 0 && compressed_pages) ||
529	(compressed_size == 0 && !compressed_pages));
530
531	if (compressed_size && compressed_pages)
532	cur_size = compressed_size;
533
534	if (!extent_inserted) {
535	struct btrfs_key key;
536	size_t datasize;
537
538	key.objectid = btrfs_ino(inode);
539	key.offset = 0;
540	key.type = BTRFS_EXTENT_DATA_KEY;
541
542	datasize = btrfs_file_extent_calc_inline_size(datasize: cur_size);
543	ret = btrfs_insert_empty_item(trans, root, path, key: &key,
544	data_size: datasize);
545	if (ret)
546	goto fail;
547	}
548	leaf = path->nodes[0];
549	ei = btrfs_item_ptr(leaf, path->slots[0],
550	struct btrfs_file_extent_item);
551	btrfs_set_file_extent_generation(eb: leaf, s: ei, val: trans->transid);
552	btrfs_set_file_extent_type(eb: leaf, s: ei, val: BTRFS_FILE_EXTENT_INLINE);
553	btrfs_set_file_extent_encryption(eb: leaf, s: ei, val: 0);
554	btrfs_set_file_extent_other_encoding(eb: leaf, s: ei, val: 0);
555	btrfs_set_file_extent_ram_bytes(eb: leaf, s: ei, val: size);
556	ptr = btrfs_file_extent_inline_start(e: ei);
557
558	if (compress_type != BTRFS_COMPRESS_NONE) {
559	struct page *cpage;
560	int i = 0;
561	while (compressed_size > 0) {
562	cpage = compressed_pages[i];
563	cur_size = min_t(unsigned long, compressed_size,
564	PAGE_SIZE);
565
566	kaddr = kmap_local_page(page: cpage);
567	write_extent_buffer(eb: leaf, src: kaddr, start: ptr, len: cur_size);
568	kunmap_local(kaddr);
569
570	i++;
571	ptr += cur_size;
572	compressed_size -= cur_size;
573	}
574	btrfs_set_file_extent_compression(eb: leaf, s: ei,
575	val: compress_type);
576	} else {
577	page = find_get_page(mapping: inode->vfs_inode.i_mapping, offset: 0);
578	btrfs_set_file_extent_compression(eb: leaf, s: ei, val: 0);
579	kaddr = kmap_local_page(page);
580	write_extent_buffer(eb: leaf, src: kaddr, start: ptr, len: size);
581	kunmap_local(kaddr);
582	put_page(page);
583	}
584	btrfs_mark_buffer_dirty(trans, buf: leaf);
585	btrfs_release_path(p: path);
586
587	/*
588	* We align size to sectorsize for inline extents just for simplicity
589	* sake.
590	*/
591	ret = btrfs_inode_set_file_extent_range(inode, start: 0,
592	ALIGN(size, root->fs_info->sectorsize));
593	if (ret)
594	goto fail;
595
596	/*
597	* We're an inline extent, so nobody can extend the file past i_size
598	* without locking a page we already have locked.
599	*
600	* We must do any i_size and inode updates before we unlock the pages.
601	* Otherwise we could end up racing with unlink.
602	*/
603	i_size = i_size_read(inode: &inode->vfs_inode);
604	if (update_i_size && size > i_size) {
605	i_size_write(inode: &inode->vfs_inode, i_size: size);
606	i_size = size;
607	}
608	inode->disk_i_size = i_size;
609
610	fail:
611	return ret;
612	}
613
614
615	/*
616	* conditionally insert an inline extent into the file. This
617	* does the checks required to make sure the data is small enough
618	* to fit as an inline extent.
619	*/
620	static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
621	size_t compressed_size,
622	int compress_type,
623	struct page **compressed_pages,
624	bool update_i_size)
625	{
626	struct btrfs_drop_extents_args drop_args = { 0 };
627	struct btrfs_root *root = inode->root;
628	struct btrfs_fs_info *fs_info = root->fs_info;
629	struct btrfs_trans_handle *trans;
630	u64 data_len = (compressed_size ?: size);
631	int ret;
632	struct btrfs_path *path;
633
634	/*
635	* We can create an inline extent if it ends at or beyond the current
636	* i_size, is no larger than a sector (decompressed), and the (possibly
637	* compressed) data fits in a leaf and the configured maximum inline
638	* size.
639	*/
640	if (size < i_size_read(inode: &inode->vfs_inode) ||
641	size > fs_info->sectorsize ||
642	data_len > BTRFS_MAX_INLINE_DATA_SIZE(info: fs_info) ||
643	data_len > fs_info->max_inline)
644	return 1;
645
646	path = btrfs_alloc_path();
647	if (!path)
648	return -ENOMEM;
649
650	trans = btrfs_join_transaction(root);
651	if (IS_ERR(ptr: trans)) {
652	btrfs_free_path(p: path);
653	return PTR_ERR(ptr: trans);
654	}
655	trans->block_rsv = &inode->block_rsv;
656
657	drop_args.path = path;
658	drop_args.start = 0;
659	drop_args.end = fs_info->sectorsize;
660	drop_args.drop_cache = true;
661	drop_args.replace_extent = true;
662	drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(datasize: data_len);
663	ret = btrfs_drop_extents(trans, root, inode, args: &drop_args);
664	if (ret) {
665	btrfs_abort_transaction(trans, ret);
666	goto out;
667	}
668
669	ret = insert_inline_extent(trans, path, inode, extent_inserted: drop_args.extent_inserted,
670	size, compressed_size, compress_type,
671	compressed_pages, update_i_size);
672	if (ret && ret != -ENOSPC) {
673	btrfs_abort_transaction(trans, ret);
674	goto out;
675	} else if (ret == -ENOSPC) {
676	ret = 1;
677	goto out;
678	}
679
680	btrfs_update_inode_bytes(inode, add_bytes: size, del_bytes: drop_args.bytes_found);
681	ret = btrfs_update_inode(trans, inode);
682	if (ret && ret != -ENOSPC) {
683	btrfs_abort_transaction(trans, ret);
684	goto out;
685	} else if (ret == -ENOSPC) {
686	ret = 1;
687	goto out;
688	}
689
690	btrfs_set_inode_full_sync(inode);
691	out:
692	/*
693	* Don't forget to free the reserved space, as for inlined extent
694	* it won't count as data extent, free them directly here.
695	* And at reserve time, it's always aligned to page size, so
696	* just free one page here.
697	*/
698	btrfs_qgroup_free_data(inode, NULL, start: 0, PAGE_SIZE, NULL);
699	btrfs_free_path(p: path);
700	btrfs_end_transaction(trans);
701	return ret;
702	}
703
704	struct async_extent {
705	u64 start;
706	u64 ram_size;
707	u64 compressed_size;
708	struct page **pages;
709	unsigned long nr_pages;
710	int compress_type;
711	struct list_head list;
712	};
713
714	struct async_chunk {
715	struct btrfs_inode *inode;
716	struct page *locked_page;
717	u64 start;
718	u64 end;
719	blk_opf_t write_flags;
720	struct list_head extents;
721	struct cgroup_subsys_state *blkcg_css;
722	struct btrfs_work work;
723	struct async_cow *async_cow;
724	};
725
726	struct async_cow {
727	atomic_t num_chunks;
728	struct async_chunk chunks[];
729	};
730
731	static noinline int add_async_extent(struct async_chunk *cow,
732	u64 start, u64 ram_size,
733	u64 compressed_size,
734	struct page **pages,
735	unsigned long nr_pages,
736	int compress_type)
737	{
738	struct async_extent *async_extent;
739
740	async_extent = kmalloc(size: sizeof(*async_extent), GFP_NOFS);
741	if (!async_extent)
742	return -ENOMEM;
743	async_extent->start = start;
744	async_extent->ram_size = ram_size;
745	async_extent->compressed_size = compressed_size;
746	async_extent->pages = pages;
747	async_extent->nr_pages = nr_pages;
748	async_extent->compress_type = compress_type;
749	list_add_tail(new: &async_extent->list, head: &cow->extents);
750	return 0;
751	}
752
753	/*
754	* Check if the inode needs to be submitted to compression, based on mount
755	* options, defragmentation, properties or heuristics.
756	*/
757	static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
758	u64 end)
759	{
760	struct btrfs_fs_info *fs_info = inode->root->fs_info;
761
762	if (!btrfs_inode_can_compress(inode)) {
763	WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
764	KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
765	btrfs_ino(inode));
766	return 0;
767	}
768	/*
769	* Special check for subpage.
770	*
771	* We lock the full page then run each delalloc range in the page, thus
772	* for the following case, we will hit some subpage specific corner case:
773	*
774	* 0 32K 64K
775	* | |///////| |///////|
776	* \- A \- B
777	*
778	* In above case, both range A and range B will try to unlock the full
779	* page [0, 64K), causing the one finished later will have page
780	* unlocked already, triggering various page lock requirement BUG_ON()s.
781	*
782	* So here we add an artificial limit that subpage compression can only
783	* if the range is fully page aligned.
784	*
785	* In theory we only need to ensure the first page is fully covered, but
786	* the tailing partial page will be locked until the full compression
787	* finishes, delaying the write of other range.
788	*
789	* TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
790	* first to prevent any submitted async extent to unlock the full page.
791	* By this, we can ensure for subpage case that only the last async_cow
792	* will unlock the full page.
793	*/
794	if (fs_info->sectorsize < PAGE_SIZE) {
795	if (!PAGE_ALIGNED(start) ||
796	!PAGE_ALIGNED(end + 1))
797	return 0;
798	}
799
800	/* force compress */
801	if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
802	return 1;
803	/* defrag ioctl */
804	if (inode->defrag_compress)
805	return 1;
806	/* bad compression ratios */
807	if (inode->flags & BTRFS_INODE_NOCOMPRESS)
808	return 0;
809	if (btrfs_test_opt(fs_info, COMPRESS) ||
810	inode->flags & BTRFS_INODE_COMPRESS ||
811	inode->prop_compress)
812	return btrfs_compress_heuristic(inode: &inode->vfs_inode, start, end);
813	return 0;
814	}
815
816	static inline void inode_should_defrag(struct btrfs_inode *inode,
817	u64 start, u64 end, u64 num_bytes, u32 small_write)
818	{
819	/* If this is a small write inside eof, kick off a defrag */
820	if (num_bytes < small_write &&
821	(start > 0 || end + 1 < inode->disk_i_size))
822	btrfs_add_inode_defrag(NULL, inode, extent_thresh: small_write);
823	}
824
825	/*
826	* Work queue call back to started compression on a file and pages.
827	*
828	* This is done inside an ordered work queue, and the compression is spread
829	* across many cpus. The actual IO submission is step two, and the ordered work
830	* queue takes care of making sure that happens in the same order things were
831	* put onto the queue by writepages and friends.
832	*
833	* If this code finds it can't get good compression, it puts an entry onto the
834	* work queue to write the uncompressed bytes. This makes sure that both
835	* compressed inodes and uncompressed inodes are written in the same order that
836	* the flusher thread sent them down.
837	*/
838	static void compress_file_range(struct btrfs_work *work)
839	{
840	struct async_chunk *async_chunk =
841	container_of(work, struct async_chunk, work);
842	struct btrfs_inode *inode = async_chunk->inode;
843	struct btrfs_fs_info *fs_info = inode->root->fs_info;
844	struct address_space *mapping = inode->vfs_inode.i_mapping;
845	u64 blocksize = fs_info->sectorsize;
846	u64 start = async_chunk->start;
847	u64 end = async_chunk->end;
848	u64 actual_end;
849	u64 i_size;
850	int ret = 0;
851	struct page **pages;
852	unsigned long nr_pages;
853	unsigned long total_compressed = 0;
854	unsigned long total_in = 0;
855	unsigned int poff;
856	int i;
857	int compress_type = fs_info->compress_type;
858
859	inode_should_defrag(inode, start, end, num_bytes: end - start + 1, SZ_16K);
860
861	/*
862	* We need to call clear_page_dirty_for_io on each page in the range.
863	* Otherwise applications with the file mmap'd can wander in and change
864	* the page contents while we are compressing them.
865	*/
866	extent_range_clear_dirty_for_io(inode: &inode->vfs_inode, start, end);
867
868	/*
869	* We need to save i_size before now because it could change in between
870	* us evaluating the size and assigning it. This is because we lock and
871	* unlock the page in truncate and fallocate, and then modify the i_size
872	* later on.
873	*
874	* The barriers are to emulate READ_ONCE, remove that once i_size_read
875	* does that for us.
876	*/
877	barrier();
878	i_size = i_size_read(inode: &inode->vfs_inode);
879	barrier();
880	actual_end = min_t(u64, i_size, end + 1);
881	again:
882	pages = NULL;
883	nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
884	nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
885
886	/*
887	* we don't want to send crud past the end of i_size through
888	* compression, that's just a waste of CPU time. So, if the
889	* end of the file is before the start of our current
890	* requested range of bytes, we bail out to the uncompressed
891	* cleanup code that can deal with all of this.
892	*
893	* It isn't really the fastest way to fix things, but this is a
894	* very uncommon corner.
895	*/
896	if (actual_end <= start)
897	goto cleanup_and_bail_uncompressed;
898
899	total_compressed = actual_end - start;
900
901	/*
902	* Skip compression for a small file range(<=blocksize) that
903	* isn't an inline extent, since it doesn't save disk space at all.
904	*/
905	if (total_compressed <= blocksize &&
906	(start > 0 || end + 1 < inode->disk_i_size))
907	goto cleanup_and_bail_uncompressed;
908
909	/*
910	* For subpage case, we require full page alignment for the sector
911	* aligned range.
912	* Thus we must also check against @actual_end, not just @end.
913	*/
914	if (blocksize < PAGE_SIZE) {
915	if (!PAGE_ALIGNED(start) ||
916	!PAGE_ALIGNED(round_up(actual_end, blocksize)))
917	goto cleanup_and_bail_uncompressed;
918	}
919
920	total_compressed = min_t(unsigned long, total_compressed,
921	BTRFS_MAX_UNCOMPRESSED);
922	total_in = 0;
923	ret = 0;
924
925	/*
926	* We do compression for mount -o compress and when the inode has not
927	* been flagged as NOCOMPRESS. This flag can change at any time if we
928	* discover bad compression ratios.
929	*/
930	if (!inode_need_compress(inode, start, end))
931	goto cleanup_and_bail_uncompressed;
932
933	pages = kcalloc(n: nr_pages, size: sizeof(struct page *), GFP_NOFS);
934	if (!pages) {
935	/*
936	* Memory allocation failure is not a fatal error, we can fall
937	* back to uncompressed code.
938	*/
939	goto cleanup_and_bail_uncompressed;
940	}
941
942	if (inode->defrag_compress)
943	compress_type = inode->defrag_compress;
944	else if (inode->prop_compress)
945	compress_type = inode->prop_compress;
946
947	/* Compression level is applied here. */
948	ret = btrfs_compress_pages(type_level: compress_type | (fs_info->compress_level << 4),
949	mapping, start, pages, out_pages: &nr_pages, total_in: &total_in,
950	total_out: &total_compressed);
951	if (ret)
952	goto mark_incompressible;
953
954	/*
955	* Zero the tail end of the last page, as we might be sending it down
956	* to disk.
957	*/
958	poff = offset_in_page(total_compressed);
959	if (poff)
960	memzero_page(page: pages[nr_pages - 1], offset: poff, PAGE_SIZE - poff);
961
962	/*
963	* Try to create an inline extent.
964	*
965	* If we didn't compress the entire range, try to create an uncompressed
966	* inline extent, else a compressed one.
967	*
968	* Check cow_file_range() for why we don't even try to create inline
969	* extent for the subpage case.
970	*/
971	if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
972	if (total_in < actual_end) {
973	ret = cow_file_range_inline(inode, size: actual_end, compressed_size: 0,
974	compress_type: BTRFS_COMPRESS_NONE, NULL,
975	update_i_size: false);
976	} else {
977	ret = cow_file_range_inline(inode, size: actual_end,
978	compressed_size: total_compressed,
979	compress_type, compressed_pages: pages,
980	update_i_size: false);
981	}
982	if (ret <= 0) {
983	unsigned long clear_flags = EXTENT_DELALLOC |
984	EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
985	EXTENT_DO_ACCOUNTING;
986
987	if (ret < 0)
988	mapping_set_error(mapping, error: -EIO);
989
990	/*
991	* inline extent creation worked or returned error,
992	* we don't need to create any more async work items.
993	* Unlock and free up our temp pages.
994	*
995	* We use DO_ACCOUNTING here because we need the
996	* delalloc_release_metadata to be done _after_ we drop
997	* our outstanding extent for clearing delalloc for this
998	* range.
999	*/
1000	extent_clear_unlock_delalloc(inode, start, end,
1001	NULL,
1002	bits_to_clear: clear_flags,
1003	page_ops: PAGE_UNLOCK |
1004	PAGE_START_WRITEBACK |
1005	PAGE_END_WRITEBACK);
1006	goto free_pages;
1007	}
1008	}
1009
1010	/*
1011	* We aren't doing an inline extent. Round the compressed size up to a
1012	* block size boundary so the allocator does sane things.
1013	*/
1014	total_compressed = ALIGN(total_compressed, blocksize);
1015
1016	/*
1017	* One last check to make sure the compression is really a win, compare
1018	* the page count read with the blocks on disk, compression must free at
1019	* least one sector.
1020	*/
1021	total_in = round_up(total_in, fs_info->sectorsize);
1022	if (total_compressed + blocksize > total_in)
1023	goto mark_incompressible;
1024
1025	/*
1026	* The async work queues will take care of doing actual allocation on
1027	* disk for these compressed pages, and will submit the bios.
1028	*/
1029	ret = add_async_extent(cow: async_chunk, start, ram_size: total_in, compressed_size: total_compressed, pages,
1030	nr_pages, compress_type);
1031	BUG_ON(ret);
1032	if (start + total_in < end) {
1033	start += total_in;
1034	cond_resched();
1035	goto again;
1036	}
1037	return;
1038
1039	mark_incompressible:
1040	if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && !inode->prop_compress)
1041	inode->flags |= BTRFS_INODE_NOCOMPRESS;
1042	cleanup_and_bail_uncompressed:
1043	ret = add_async_extent(cow: async_chunk, start, ram_size: end - start + 1, compressed_size: 0, NULL, nr_pages: 0,
1044	compress_type: BTRFS_COMPRESS_NONE);
1045	BUG_ON(ret);
1046	free_pages:
1047	if (pages) {
1048	for (i = 0; i < nr_pages; i++) {
1049	WARN_ON(pages[i]->mapping);
1050	btrfs_free_compr_page(page: pages[i]);
1051	}
1052	kfree(objp: pages);
1053	}
1054	}
1055
1056	static void free_async_extent_pages(struct async_extent *async_extent)
1057	{
1058	int i;
1059
1060	if (!async_extent->pages)
1061	return;
1062
1063	for (i = 0; i < async_extent->nr_pages; i++) {
1064	WARN_ON(async_extent->pages[i]->mapping);
1065	btrfs_free_compr_page(page: async_extent->pages[i]);
1066	}
1067	kfree(objp: async_extent->pages);
1068	async_extent->nr_pages = 0;
1069	async_extent->pages = NULL;
1070	}
1071
1072	static void submit_uncompressed_range(struct btrfs_inode *inode,
1073	struct async_extent *async_extent,
1074	struct page *locked_page)
1075	{
1076	u64 start = async_extent->start;
1077	u64 end = async_extent->start + async_extent->ram_size - 1;
1078	int ret;
1079	struct writeback_control wbc = {
1080	.sync_mode = WB_SYNC_ALL,
1081	.range_start = start,
1082	.range_end = end,
1083	.no_cgroup_owner = 1,
1084	};
1085
1086	wbc_attach_fdatawrite_inode(wbc: &wbc, inode: &inode->vfs_inode);
1087	ret = run_delalloc_cow(inode, locked_page, start, end, wbc: &wbc, pages_dirty: false);
1088	wbc_detach_inode(wbc: &wbc);
1089	if (ret < 0) {
1090	btrfs_cleanup_ordered_extents(inode, locked_page, offset: start, bytes: end - start + 1);
1091	if (locked_page) {
1092	const u64 page_start = page_offset(page: locked_page);
1093
1094	set_page_writeback(locked_page);
1095	end_page_writeback(page: locked_page);
1096	btrfs_mark_ordered_io_finished(inode, page: locked_page,
1097	file_offset: page_start, PAGE_SIZE,
1098	uptodate: !ret);
1099	mapping_set_error(mapping: locked_page->mapping, error: ret);
1100	unlock_page(page: locked_page);
1101	}
1102	}
1103	}
1104
1105	static void submit_one_async_extent(struct async_chunk *async_chunk,
1106	struct async_extent *async_extent,
1107	u64 *alloc_hint)
1108	{
1109	struct btrfs_inode *inode = async_chunk->inode;
1110	struct extent_io_tree *io_tree = &inode->io_tree;
1111	struct btrfs_root *root = inode->root;
1112	struct btrfs_fs_info *fs_info = root->fs_info;
1113	struct btrfs_ordered_extent *ordered;
1114	struct btrfs_key ins;
1115	struct page *locked_page = NULL;
1116	struct extent_map *em;
1117	int ret = 0;
1118	u64 start = async_extent->start;
1119	u64 end = async_extent->start + async_extent->ram_size - 1;
1120
1121	if (async_chunk->blkcg_css)
1122	kthread_associate_blkcg(css: async_chunk->blkcg_css);
1123
1124	/*
1125	* If async_chunk->locked_page is in the async_extent range, we need to
1126	* handle it.
1127	*/
1128	if (async_chunk->locked_page) {
1129	u64 locked_page_start = page_offset(page: async_chunk->locked_page);
1130	u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1131
1132	if (!(start >= locked_page_end || end <= locked_page_start))
1133	locked_page = async_chunk->locked_page;
1134	}
1135	lock_extent(tree: io_tree, start, end, NULL);
1136
1137	if (async_extent->compress_type == BTRFS_COMPRESS_NONE) {
1138	submit_uncompressed_range(inode, async_extent, locked_page);
1139	goto done;
1140	}
1141
1142	ret = btrfs_reserve_extent(root, ram_bytes: async_extent->ram_size,
1143	num_bytes: async_extent->compressed_size,
1144	min_alloc_size: async_extent->compressed_size,
1145	empty_size: 0, hint_byte: *alloc_hint, ins: &ins, is_data: 1, delalloc: 1);
1146	if (ret) {
1147	/*
1148	* Here we used to try again by going back to non-compressed
1149	* path for ENOSPC. But we can't reserve space even for
1150	* compressed size, how could it work for uncompressed size
1151	* which requires larger size? So here we directly go error
1152	* path.
1153	*/
1154	goto out_free;
1155	}
1156
1157	/* Here we're doing allocation and writeback of the compressed pages */
1158	em = create_io_em(inode, start,
1159	len: async_extent->ram_size, /* len */
1160	orig_start: start, /* orig_start */
1161	block_start: ins.objectid, /* block_start */
1162	block_len: ins.offset, /* block_len */
1163	orig_block_len: ins.offset, /* orig_block_len */
1164	ram_bytes: async_extent->ram_size, /* ram_bytes */
1165	compress_type: async_extent->compress_type,
1166	type: BTRFS_ORDERED_COMPRESSED);
1167	if (IS_ERR(ptr: em)) {
1168	ret = PTR_ERR(ptr: em);
1169	goto out_free_reserve;
1170	}
1171	free_extent_map(em);
1172
1173	ordered = btrfs_alloc_ordered_extent(inode, file_offset: start, /* file_offset */
1174	num_bytes: async_extent->ram_size, /* num_bytes */
1175	ram_bytes: async_extent->ram_size, /* ram_bytes */
1176	disk_bytenr: ins.objectid, /* disk_bytenr */
1177	disk_num_bytes: ins.offset, /* disk_num_bytes */
1178	offset: 0, /* offset */
1179	flags: 1 << BTRFS_ORDERED_COMPRESSED,
1180	compress_type: async_extent->compress_type);
1181	if (IS_ERR(ptr: ordered)) {
1182	btrfs_drop_extent_map_range(inode, start, end, skip_pinned: false);
1183	ret = PTR_ERR(ptr: ordered);
1184	goto out_free_reserve;
1185	}
1186	btrfs_dec_block_group_reservations(fs_info, start: ins.objectid);
1187
1188	/* Clear dirty, set writeback and unlock the pages. */
1189	extent_clear_unlock_delalloc(inode, start, end,
1190	NULL, bits_to_clear: EXTENT_LOCKED | EXTENT_DELALLOC,
1191	page_ops: PAGE_UNLOCK | PAGE_START_WRITEBACK);
1192	btrfs_submit_compressed_write(ordered,
1193	compressed_pages: async_extent->pages, /* compressed_pages */
1194	nr_pages: async_extent->nr_pages,
1195	write_flags: async_chunk->write_flags, writeback: true);
1196	*alloc_hint = ins.objectid + ins.offset;
1197	done:
1198	if (async_chunk->blkcg_css)
1199	kthread_associate_blkcg(NULL);
1200	kfree(objp: async_extent);
1201	return;
1202
1203	out_free_reserve:
1204	btrfs_dec_block_group_reservations(fs_info, start: ins.objectid);
1205	btrfs_free_reserved_extent(fs_info, start: ins.objectid, len: ins.offset, delalloc: 1);
1206	out_free:
1207	mapping_set_error(mapping: inode->vfs_inode.i_mapping, error: -EIO);
1208	extent_clear_unlock_delalloc(inode, start, end,
1209	NULL, bits_to_clear: EXTENT_LOCKED | EXTENT_DELALLOC |
1210	EXTENT_DELALLOC_NEW |
1211	EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1212	page_ops: PAGE_UNLOCK | PAGE_START_WRITEBACK |
1213	PAGE_END_WRITEBACK);
1214	free_async_extent_pages(async_extent);
1215	if (async_chunk->blkcg_css)
1216	kthread_associate_blkcg(NULL);
1217	btrfs_debug(fs_info,
1218	"async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1219	root->root_key.objectid, btrfs_ino(inode), start,
1220	async_extent->ram_size, ret);
1221	kfree(objp: async_extent);
1222	}
1223
1224	static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1225	u64 num_bytes)
1226	{
1227	struct extent_map_tree *em_tree = &inode->extent_tree;
1228	struct extent_map *em;
1229	u64 alloc_hint = 0;
1230
1231	read_lock(&em_tree->lock);
1232	em = search_extent_mapping(tree: em_tree, start, len: num_bytes);
1233	if (em) {
1234	/*
1235	* if block start isn't an actual block number then find the
1236	* first block in this inode and use that as a hint. If that
1237	* block is also bogus then just don't worry about it.
1238	*/
1239	if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1240	free_extent_map(em);
1241	em = search_extent_mapping(tree: em_tree, start: 0, len: 0);
1242	if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1243	alloc_hint = em->block_start;
1244	if (em)
1245	free_extent_map(em);
1246	} else {
1247	alloc_hint = em->block_start;
1248	free_extent_map(em);
1249	}
1250	}
1251	read_unlock(&em_tree->lock);
1252
1253	return alloc_hint;
1254	}
1255
1256	/*
1257	* when extent_io.c finds a delayed allocation range in the file,
1258	* the call backs end up in this code. The basic idea is to
1259	* allocate extents on disk for the range, and create ordered data structs
1260	* in ram to track those extents.
1261	*
1262	* locked_page is the page that writepage had locked already. We use
1263	* it to make sure we don't do extra locks or unlocks.
1264	*
1265	* When this function fails, it unlocks all pages except @locked_page.
1266	*
1267	* When this function successfully creates an inline extent, it returns 1 and
1268	* unlocks all pages including locked_page and starts I/O on them.
1269	* (In reality inline extents are limited to a single page, so locked_page is
1270	* the only page handled anyway).
1271	*
1272	* When this function succeed and creates a normal extent, the page locking
1273	* status depends on the passed in flags:
1274	*
1275	* - If @keep_locked is set, all pages are kept locked.
1276	* - Else all pages except for @locked_page are unlocked.
1277	*
1278	* When a failure happens in the second or later iteration of the
1279	* while-loop, the ordered extents created in previous iterations are kept
1280	* intact. So, the caller must clean them up by calling
1281	* btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1282	* example.
1283	*/
1284	static noinline int cow_file_range(struct btrfs_inode *inode,
1285	struct page *locked_page, u64 start, u64 end,
1286	u64 *done_offset,
1287	bool keep_locked, bool no_inline)
1288	{
1289	struct btrfs_root *root = inode->root;
1290	struct btrfs_fs_info *fs_info = root->fs_info;
1291	u64 alloc_hint = 0;
1292	u64 orig_start = start;
1293	u64 num_bytes;
1294	unsigned long ram_size;
1295	u64 cur_alloc_size = 0;
1296	u64 min_alloc_size;
1297	u64 blocksize = fs_info->sectorsize;
1298	struct btrfs_key ins;
1299	struct extent_map *em;
1300	unsigned clear_bits;
1301	unsigned long page_ops;
1302	bool extent_reserved = false;
1303	int ret = 0;
1304
1305	if (btrfs_is_free_space_inode(inode)) {
1306	ret = -EINVAL;
1307	goto out_unlock;
1308	}
1309
1310	num_bytes = ALIGN(end - start + 1, blocksize);
1311	num_bytes = max(blocksize, num_bytes);
1312	ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1313
1314	inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1315
1316	/*
1317	* Due to the page size limit, for subpage we can only trigger the
1318	* writeback for the dirty sectors of page, that means data writeback
1319	* is doing more writeback than what we want.
1320	*
1321	* This is especially unexpected for some call sites like fallocate,
1322	* where we only increase i_size after everything is done.
1323	* This means we can trigger inline extent even if we didn't want to.
1324	* So here we skip inline extent creation completely.
1325	*/
1326	if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1327	u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1328	end + 1);
1329
1330	/* lets try to make an inline extent */
1331	ret = cow_file_range_inline(inode, size: actual_end, compressed_size: 0,
1332	compress_type: BTRFS_COMPRESS_NONE, NULL, update_i_size: false);
1333	if (ret == 0) {
1334	/*
1335	* We use DO_ACCOUNTING here because we need the
1336	* delalloc_release_metadata to be run _after_ we drop
1337	* our outstanding extent for clearing delalloc for this
1338	* range.
1339	*/
1340	extent_clear_unlock_delalloc(inode, start, end,
1341	locked_page,
1342	bits_to_clear: EXTENT_LOCKED | EXTENT_DELALLOC |
1343	EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1344	EXTENT_DO_ACCOUNTING, page_ops: PAGE_UNLOCK |
1345	PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1346	/*
1347	* locked_page is locked by the caller of
1348	* writepage_delalloc(), not locked by
1349	* __process_pages_contig().
1350	*
1351	* We can't let __process_pages_contig() to unlock it,
1352	* as it doesn't have any subpage::writers recorded.
1353	*
1354	* Here we manually unlock the page, since the caller
1355	* can't determine if it's an inline extent or a
1356	* compressed extent.
1357	*/
1358	unlock_page(page: locked_page);
1359	ret = 1;
1360	goto done;
1361	} else if (ret < 0) {
1362	goto out_unlock;
1363	}
1364	}
1365
1366	alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1367
1368	/*
1369	* Relocation relies on the relocated extents to have exactly the same
1370	* size as the original extents. Normally writeback for relocation data
1371	* extents follows a NOCOW path because relocation preallocates the
1372	* extents. However, due to an operation such as scrub turning a block
1373	* group to RO mode, it may fallback to COW mode, so we must make sure
1374	* an extent allocated during COW has exactly the requested size and can
1375	* not be split into smaller extents, otherwise relocation breaks and
1376	* fails during the stage where it updates the bytenr of file extent
1377	* items.
1378	*/
1379	if (btrfs_is_data_reloc_root(root))
1380	min_alloc_size = num_bytes;
1381	else
1382	min_alloc_size = fs_info->sectorsize;
1383
1384	while (num_bytes > 0) {
1385	struct btrfs_ordered_extent *ordered;
1386
1387	cur_alloc_size = num_bytes;
1388	ret = btrfs_reserve_extent(root, ram_bytes: cur_alloc_size, num_bytes: cur_alloc_size,
1389	min_alloc_size, empty_size: 0, hint_byte: alloc_hint,
1390	ins: &ins, is_data: 1, delalloc: 1);
1391	if (ret == -EAGAIN) {
1392	/*
1393	* btrfs_reserve_extent only returns -EAGAIN for zoned
1394	* file systems, which is an indication that there are
1395	* no active zones to allocate from at the moment.
1396	*
1397	* If this is the first loop iteration, wait for at
1398	* least one zone to finish before retrying the
1399	* allocation. Otherwise ask the caller to write out
1400	* the already allocated blocks before coming back to
1401	* us, or return -ENOSPC if it can't handle retries.
1402	*/
1403	ASSERT(btrfs_is_zoned(fs_info));
1404	if (start == orig_start) {
1405	wait_on_bit_io(word: &inode->root->fs_info->flags,
1406	bit: BTRFS_FS_NEED_ZONE_FINISH,
1407	TASK_UNINTERRUPTIBLE);
1408	continue;
1409	}
1410	if (done_offset) {
1411	*done_offset = start - 1;
1412	return 0;
1413	}
1414	ret = -ENOSPC;
1415	}
1416	if (ret < 0)
1417	goto out_unlock;
1418	cur_alloc_size = ins.offset;
1419	extent_reserved = true;
1420
1421	ram_size = ins.offset;
1422	em = create_io_em(inode, start, len: ins.offset, /* len */
1423	orig_start: start, /* orig_start */
1424	block_start: ins.objectid, /* block_start */
1425	block_len: ins.offset, /* block_len */
1426	orig_block_len: ins.offset, /* orig_block_len */
1427	ram_bytes: ram_size, /* ram_bytes */
1428	compress_type: BTRFS_COMPRESS_NONE, /* compress_type */
1429	type: BTRFS_ORDERED_REGULAR /* type */);
1430	if (IS_ERR(ptr: em)) {
1431	ret = PTR_ERR(ptr: em);
1432	goto out_reserve;
1433	}
1434	free_extent_map(em);
1435
1436	ordered = btrfs_alloc_ordered_extent(inode, file_offset: start, num_bytes: ram_size,
1437	ram_bytes: ram_size, disk_bytenr: ins.objectid, disk_num_bytes: cur_alloc_size,
1438	offset: 0, flags: 1 << BTRFS_ORDERED_REGULAR,
1439	compress_type: BTRFS_COMPRESS_NONE);
1440	if (IS_ERR(ptr: ordered)) {
1441	ret = PTR_ERR(ptr: ordered);
1442	goto out_drop_extent_cache;
1443	}
1444
1445	if (btrfs_is_data_reloc_root(root)) {
1446	ret = btrfs_reloc_clone_csums(ordered);
1447
1448	/*
1449	* Only drop cache here, and process as normal.
1450	*
1451	* We must not allow extent_clear_unlock_delalloc()
1452	* at out_unlock label to free meta of this ordered
1453	* extent, as its meta should be freed by
1454	* btrfs_finish_ordered_io().
1455	*
1456	* So we must continue until @start is increased to
1457	* skip current ordered extent.
1458	*/
1459	if (ret)
1460	btrfs_drop_extent_map_range(inode, start,
1461	end: start + ram_size - 1,
1462	skip_pinned: false);
1463	}
1464	btrfs_put_ordered_extent(entry: ordered);
1465
1466	btrfs_dec_block_group_reservations(fs_info, start: ins.objectid);
1467
1468	/*
1469	* We're not doing compressed IO, don't unlock the first page
1470	* (which the caller expects to stay locked), don't clear any
1471	* dirty bits and don't set any writeback bits
1472	*
1473	* Do set the Ordered (Private2) bit so we know this page was
1474	* properly setup for writepage.
1475	*/
1476	page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1477	page_ops |= PAGE_SET_ORDERED;
1478
1479	extent_clear_unlock_delalloc(inode, start, end: start + ram_size - 1,
1480	locked_page,
1481	bits_to_clear: EXTENT_LOCKED | EXTENT_DELALLOC,
1482	page_ops);
1483	if (num_bytes < cur_alloc_size)
1484	num_bytes = 0;
1485	else
1486	num_bytes -= cur_alloc_size;
1487	alloc_hint = ins.objectid + ins.offset;
1488	start += cur_alloc_size;
1489	extent_reserved = false;
1490
1491	/*
1492	* btrfs_reloc_clone_csums() error, since start is increased
1493	* extent_clear_unlock_delalloc() at out_unlock label won't
1494	* free metadata of current ordered extent, we're OK to exit.
1495	*/
1496	if (ret)
1497	goto out_unlock;
1498	}
1499	done:
1500	if (done_offset)
1501	*done_offset = end;
1502	return ret;
1503
1504	out_drop_extent_cache:
1505	btrfs_drop_extent_map_range(inode, start, end: start + ram_size - 1, skip_pinned: false);
1506	out_reserve:
1507	btrfs_dec_block_group_reservations(fs_info, start: ins.objectid);
1508	btrfs_free_reserved_extent(fs_info, start: ins.objectid, len: ins.offset, delalloc: 1);
1509	out_unlock:
1510	/*
1511	* Now, we have three regions to clean up:
1512	*
1513	* |-------(1)----|---(2)---|-------------(3)----------|
1514	* `- orig_start `- start `- start + cur_alloc_size `- end
1515	*
1516	* We process each region below.
1517	*/
1518
1519	clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1520	EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1521	page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1522
1523	/*
1524	* For the range (1). We have already instantiated the ordered extents
1525	* for this region. They are cleaned up by
1526	* btrfs_cleanup_ordered_extents() in e.g,
1527	* btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1528	* already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1529	* EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1530	* function.
1531	*
1532	* However, in case of @keep_locked, we still need to unlock the pages
1533	* (except @locked_page) to ensure all the pages are unlocked.
1534	*/
1535	if (keep_locked && orig_start < start) {
1536	if (!locked_page)
1537	mapping_set_error(mapping: inode->vfs_inode.i_mapping, error: ret);
1538	extent_clear_unlock_delalloc(inode, start: orig_start, end: start - 1,
1539	locked_page, bits_to_clear: 0, page_ops);
1540	}
1541
1542	/*
1543	* For the range (2). If we reserved an extent for our delalloc range
1544	* (or a subrange) and failed to create the respective ordered extent,
1545	* then it means that when we reserved the extent we decremented the
1546	* extent's size from the data space_info's bytes_may_use counter and
1547	* incremented the space_info's bytes_reserved counter by the same
1548	* amount. We must make sure extent_clear_unlock_delalloc() does not try
1549	* to decrement again the data space_info's bytes_may_use counter,
1550	* therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1551	*/
1552	if (extent_reserved) {
1553	extent_clear_unlock_delalloc(inode, start,
1554	end: start + cur_alloc_size - 1,
1555	locked_page,
1556	bits_to_clear: clear_bits,
1557	page_ops);
1558	start += cur_alloc_size;
1559	}
1560
1561	/*
1562	* For the range (3). We never touched the region. In addition to the
1563	* clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1564	* space_info's bytes_may_use counter, reserved in
1565	* btrfs_check_data_free_space().
1566	*/
1567	if (start < end) {
1568	clear_bits |= EXTENT_CLEAR_DATA_RESV;
1569	extent_clear_unlock_delalloc(inode, start, end, locked_page,
1570	bits_to_clear: clear_bits, page_ops);
1571	}
1572	return ret;
1573	}
1574
1575	/*
1576	* Phase two of compressed writeback. This is the ordered portion of the code,
1577	* which only gets called in the order the work was queued. We walk all the
1578	* async extents created by compress_file_range and send them down to the disk.
1579	*
1580	* If called with @do_free == true then it'll try to finish the work and free
1581	* the work struct eventually.
1582	*/
1583	static noinline void submit_compressed_extents(struct btrfs_work *work, bool do_free)
1584	{
1585	struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1586	work);
1587	struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1588	struct async_extent *async_extent;
1589	unsigned long nr_pages;
1590	u64 alloc_hint = 0;
1591
1592	if (do_free) {
1593	struct async_chunk *async_chunk;
1594	struct async_cow *async_cow;
1595
1596	async_chunk = container_of(work, struct async_chunk, work);
1597	btrfs_add_delayed_iput(inode: async_chunk->inode);
1598	if (async_chunk->blkcg_css)
1599	css_put(css: async_chunk->blkcg_css);
1600
1601	async_cow = async_chunk->async_cow;
1602	if (atomic_dec_and_test(v: &async_cow->num_chunks))
1603	kvfree(addr: async_cow);
1604	return;
1605	}
1606
1607	nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1608	PAGE_SHIFT;
1609
1610	while (!list_empty(head: &async_chunk->extents)) {
1611	async_extent = list_entry(async_chunk->extents.next,
1612	struct async_extent, list);
1613	list_del(entry: &async_extent->list);
1614	submit_one_async_extent(async_chunk, async_extent, alloc_hint: &alloc_hint);
1615	}
1616
1617	/* atomic_sub_return implies a barrier */
1618	if (atomic_sub_return(i: nr_pages, v: &fs_info->async_delalloc_pages) <
1619	5 * SZ_1M)
1620	cond_wake_up_nomb(wq: &fs_info->async_submit_wait);
1621	}
1622
1623	static bool run_delalloc_compressed(struct btrfs_inode *inode,
1624	struct page *locked_page, u64 start,
1625	u64 end, struct writeback_control *wbc)
1626	{
1627	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1628	struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1629	struct async_cow *ctx;
1630	struct async_chunk *async_chunk;
1631	unsigned long nr_pages;
1632	u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1633	int i;
1634	unsigned nofs_flag;
1635	const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1636
1637	nofs_flag = memalloc_nofs_save();
1638	ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1639	memalloc_nofs_restore(flags: nofs_flag);
1640	if (!ctx)
1641	return false;
1642
1643	unlock_extent(tree: &inode->io_tree, start, end, NULL);
1644	set_bit(nr: BTRFS_INODE_HAS_ASYNC_EXTENT, addr: &inode->runtime_flags);
1645
1646	async_chunk = ctx->chunks;
1647	atomic_set(v: &ctx->num_chunks, i: num_chunks);
1648
1649	for (i = 0; i < num_chunks; i++) {
1650	u64 cur_end = min(end, start + SZ_512K - 1);
1651
1652	/*
1653	* igrab is called higher up in the call chain, take only the
1654	* lightweight reference for the callback lifetime
1655	*/
1656	ihold(inode: &inode->vfs_inode);
1657	async_chunk[i].async_cow = ctx;
1658	async_chunk[i].inode = inode;
1659	async_chunk[i].start = start;
1660	async_chunk[i].end = cur_end;
1661	async_chunk[i].write_flags = write_flags;
1662	INIT_LIST_HEAD(list: &async_chunk[i].extents);
1663
1664	/*
1665	* The locked_page comes all the way from writepage and its
1666	* the original page we were actually given. As we spread
1667	* this large delalloc region across multiple async_chunk
1668	* structs, only the first struct needs a pointer to locked_page
1669	*
1670	* This way we don't need racey decisions about who is supposed
1671	* to unlock it.
1672	*/
1673	if (locked_page) {
1674	/*
1675	* Depending on the compressibility, the pages might or
1676	* might not go through async. We want all of them to
1677	* be accounted against wbc once. Let's do it here
1678	* before the paths diverge. wbc accounting is used
1679	* only for foreign writeback detection and doesn't
1680	* need full accuracy. Just account the whole thing
1681	* against the first page.
1682	*/
1683	wbc_account_cgroup_owner(wbc, page: locked_page,
1684	bytes: cur_end - start);
1685	async_chunk[i].locked_page = locked_page;
1686	locked_page = NULL;
1687	} else {
1688	async_chunk[i].locked_page = NULL;
1689	}
1690
1691	if (blkcg_css != blkcg_root_css) {
1692	css_get(css: blkcg_css);
1693	async_chunk[i].blkcg_css = blkcg_css;
1694	async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1695	} else {
1696	async_chunk[i].blkcg_css = NULL;
1697	}
1698
1699	btrfs_init_work(work: &async_chunk[i].work, func: compress_file_range,
1700	ordered_func: submit_compressed_extents);
1701
1702	nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1703	atomic_add(i: nr_pages, v: &fs_info->async_delalloc_pages);
1704
1705	btrfs_queue_work(wq: fs_info->delalloc_workers, work: &async_chunk[i].work);
1706
1707	start = cur_end + 1;
1708	}
1709	return true;
1710	}
1711
1712	/*
1713	* Run the delalloc range from start to end, and write back any dirty pages
1714	* covered by the range.
1715	*/
1716	static noinline int run_delalloc_cow(struct btrfs_inode *inode,
1717	struct page *locked_page, u64 start,
1718	u64 end, struct writeback_control *wbc,
1719	bool pages_dirty)
1720	{
1721	u64 done_offset = end;
1722	int ret;
1723
1724	while (start <= end) {
1725	ret = cow_file_range(inode, locked_page, start, end, done_offset: &done_offset,
1726	keep_locked: true, no_inline: false);
1727	if (ret)
1728	return ret;
1729	extent_write_locked_range(inode: &inode->vfs_inode, locked_page, start,
1730	end: done_offset, wbc, pages_dirty);
1731	start = done_offset + 1;
1732	}
1733
1734	return 1;
1735	}
1736
1737	static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1738	u64 bytenr, u64 num_bytes, bool nowait)
1739	{
1740	struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1741	struct btrfs_ordered_sum *sums;
1742	int ret;
1743	LIST_HEAD(list);
1744
1745	ret = btrfs_lookup_csums_list(root: csum_root, start: bytenr, end: bytenr + num_bytes - 1,
1746	list: &list, search_commit: 0, nowait);
1747	if (ret == 0 && list_empty(head: &list))
1748	return 0;
1749
1750	while (!list_empty(head: &list)) {
1751	sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1752	list_del(entry: &sums->list);
1753	kfree(objp: sums);
1754	}
1755	if (ret < 0)
1756	return ret;
1757	return 1;
1758	}
1759
1760	static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1761	const u64 start, const u64 end)
1762	{
1763	const bool is_space_ino = btrfs_is_free_space_inode(inode);
1764	const bool is_reloc_ino = btrfs_is_data_reloc_root(root: inode->root);
1765	const u64 range_bytes = end + 1 - start;
1766	struct extent_io_tree *io_tree = &inode->io_tree;
1767	u64 range_start = start;
1768	u64 count;
1769	int ret;
1770
1771	/*
1772	* If EXTENT_NORESERVE is set it means that when the buffered write was
1773	* made we had not enough available data space and therefore we did not
1774	* reserve data space for it, since we though we could do NOCOW for the
1775	* respective file range (either there is prealloc extent or the inode
1776	* has the NOCOW bit set).
1777	*
1778	* However when we need to fallback to COW mode (because for example the
1779	* block group for the corresponding extent was turned to RO mode by a
1780	* scrub or relocation) we need to do the following:
1781	*
1782	* 1) We increment the bytes_may_use counter of the data space info.
1783	* If COW succeeds, it allocates a new data extent and after doing
1784	* that it decrements the space info's bytes_may_use counter and
1785	* increments its bytes_reserved counter by the same amount (we do
1786	* this at btrfs_add_reserved_bytes()). So we need to increment the
1787	* bytes_may_use counter to compensate (when space is reserved at
1788	* buffered write time, the bytes_may_use counter is incremented);
1789	*
1790	* 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1791	* that if the COW path fails for any reason, it decrements (through
1792	* extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1793	* data space info, which we incremented in the step above.
1794	*
1795	* If we need to fallback to cow and the inode corresponds to a free
1796	* space cache inode or an inode of the data relocation tree, we must
1797	* also increment bytes_may_use of the data space_info for the same
1798	* reason. Space caches and relocated data extents always get a prealloc
1799	* extent for them, however scrub or balance may have set the block
1800	* group that contains that extent to RO mode and therefore force COW
1801	* when starting writeback.
1802	*/
1803	count = count_range_bits(tree: io_tree, start: &range_start, search_end: end, max_bytes: range_bytes,
1804	bits: EXTENT_NORESERVE, contig: 0, NULL);
1805	if (count > 0 || is_space_ino || is_reloc_ino) {
1806	u64 bytes = count;
1807	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1808	struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1809
1810	if (is_space_ino || is_reloc_ino)
1811	bytes = range_bytes;
1812
1813	spin_lock(lock: &sinfo->lock);
1814	btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1815	spin_unlock(lock: &sinfo->lock);
1816
1817	if (count > 0)
1818	clear_extent_bit(tree: io_tree, start, end, bits: EXTENT_NORESERVE,
1819	NULL);
1820	}
1821
1822	/*
1823	* Don't try to create inline extents, as a mix of inline extent that
1824	* is written out and unlocked directly and a normal NOCOW extent
1825	* doesn't work.
1826	*/
1827	ret = cow_file_range(inode, locked_page, start, end, NULL, keep_locked: false, no_inline: true);
1828	ASSERT(ret != 1);
1829	return ret;
1830	}
1831
1832	struct can_nocow_file_extent_args {
1833	/* Input fields. */
1834
1835	/* Start file offset of the range we want to NOCOW. */
1836	u64 start;
1837	/* End file offset (inclusive) of the range we want to NOCOW. */
1838	u64 end;
1839	bool writeback_path;
1840	bool strict;
1841	/*
1842	* Free the path passed to can_nocow_file_extent() once it's not needed
1843	* anymore.
1844	*/
1845	bool free_path;
1846
1847	/* Output fields. Only set when can_nocow_file_extent() returns 1. */
1848
1849	u64 disk_bytenr;
1850	u64 disk_num_bytes;
1851	u64 extent_offset;
1852	/* Number of bytes that can be written to in NOCOW mode. */
1853	u64 num_bytes;
1854	};
1855
1856	/*
1857	* Check if we can NOCOW the file extent that the path points to.
1858	* This function may return with the path released, so the caller should check
1859	* if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1860	*
1861	* Returns: < 0 on error
1862	* 0 if we can not NOCOW
1863	* 1 if we can NOCOW
1864	*/
1865	static int can_nocow_file_extent(struct btrfs_path *path,
1866	struct btrfs_key *key,
1867	struct btrfs_inode *inode,
1868	struct can_nocow_file_extent_args *args)
1869	{
1870	const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1871	struct extent_buffer *leaf = path->nodes[0];
1872	struct btrfs_root *root = inode->root;
1873	struct btrfs_file_extent_item *fi;
1874	u64 extent_end;
1875	u8 extent_type;
1876	int can_nocow = 0;
1877	int ret = 0;
1878	bool nowait = path->nowait;
1879
1880	fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1881	extent_type = btrfs_file_extent_type(eb: leaf, s: fi);
1882
1883	if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1884	goto out;
1885
1886	/* Can't access these fields unless we know it's not an inline extent. */
1887	args->disk_bytenr = btrfs_file_extent_disk_bytenr(eb: leaf, s: fi);
1888	args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(eb: leaf, s: fi);
1889	args->extent_offset = btrfs_file_extent_offset(eb: leaf, s: fi);
1890
1891	if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1892	extent_type == BTRFS_FILE_EXTENT_REG)
1893	goto out;
1894
1895	/*
1896	* If the extent was created before the generation where the last snapshot
1897	* for its subvolume was created, then this implies the extent is shared,
1898	* hence we must COW.
1899	*/
1900	if (!args->strict &&
1901	btrfs_file_extent_generation(eb: leaf, s: fi) <=
1902	btrfs_root_last_snapshot(s: &root->root_item))
1903	goto out;
1904
1905	/* An explicit hole, must COW. */
1906	if (args->disk_bytenr == 0)
1907	goto out;
1908
1909	/* Compressed/encrypted/encoded extents must be COWed. */
1910	if (btrfs_file_extent_compression(eb: leaf, s: fi) ||
1911	btrfs_file_extent_encryption(eb: leaf, s: fi) ||
1912	btrfs_file_extent_other_encoding(eb: leaf, s: fi))
1913	goto out;
1914
1915	extent_end = btrfs_file_extent_end(path);
1916
1917	/*
1918	* The following checks can be expensive, as they need to take other
1919	* locks and do btree or rbtree searches, so release the path to avoid
1920	* blocking other tasks for too long.
1921	*/
1922	btrfs_release_path(p: path);
1923
1924	ret = btrfs_cross_ref_exist(root, objectid: btrfs_ino(inode),
1925	offset: key->offset - args->extent_offset,
1926	bytenr: args->disk_bytenr, strict: args->strict, path);
1927	WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1928	if (ret != 0)
1929	goto out;
1930
1931	if (args->free_path) {
1932	/*
1933	* We don't need the path anymore, plus through the
1934	* csum_exist_in_range() call below we will end up allocating
1935	* another path. So free the path to avoid unnecessary extra
1936	* memory usage.
1937	*/
1938	btrfs_free_path(p: path);
1939	path = NULL;
1940	}
1941
1942	/* If there are pending snapshots for this root, we must COW. */
1943	if (args->writeback_path && !is_freespace_inode &&
1944	atomic_read(v: &root->snapshot_force_cow))
1945	goto out;
1946
1947	args->disk_bytenr += args->extent_offset;
1948	args->disk_bytenr += args->start - key->offset;
1949	args->num_bytes = min(args->end + 1, extent_end) - args->start;
1950
1951	/*
1952	* Force COW if csums exist in the range. This ensures that csums for a
1953	* given extent are either valid or do not exist.
1954	*/
1955	ret = csum_exist_in_range(fs_info: root->fs_info, bytenr: args->disk_bytenr, num_bytes: args->num_bytes,
1956	nowait);
1957	WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1958	if (ret != 0)
1959	goto out;
1960
1961	can_nocow = 1;
1962	out:
1963	if (args->free_path && path)
1964	btrfs_free_path(p: path);
1965
1966	return ret < 0 ? ret : can_nocow;
1967	}
1968
1969	/*
1970	* when nowcow writeback call back. This checks for snapshots or COW copies
1971	* of the extents that exist in the file, and COWs the file as required.
1972	*
1973	* If no cow copies or snapshots exist, we write directly to the existing
1974	* blocks on disk
1975	*/
1976	static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1977	struct page *locked_page,
1978	const u64 start, const u64 end)
1979	{
1980	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1981	struct btrfs_root *root = inode->root;
1982	struct btrfs_path *path;
1983	u64 cow_start = (u64)-1;
1984	u64 cur_offset = start;
1985	int ret;
1986	bool check_prev = true;
1987	u64 ino = btrfs_ino(inode);
1988	struct can_nocow_file_extent_args nocow_args = { 0 };
1989
1990	/*
1991	* Normally on a zoned device we're only doing COW writes, but in case
1992	* of relocation on a zoned filesystem serializes I/O so that we're only
1993	* writing sequentially and can end up here as well.
1994	*/
1995	ASSERT(!btrfs_is_zoned(fs_info) || btrfs_is_data_reloc_root(root));
1996
1997	path = btrfs_alloc_path();
1998	if (!path) {
1999	ret = -ENOMEM;
2000	goto error;
2001	}
2002
2003	nocow_args.end = end;
2004	nocow_args.writeback_path = true;
2005
2006	while (1) {
2007	struct btrfs_block_group *nocow_bg = NULL;
2008	struct btrfs_ordered_extent *ordered;
2009	struct btrfs_key found_key;
2010	struct btrfs_file_extent_item *fi;
2011	struct extent_buffer *leaf;
2012	u64 extent_end;
2013	u64 ram_bytes;
2014	u64 nocow_end;
2015	int extent_type;
2016	bool is_prealloc;
2017
2018	ret = btrfs_lookup_file_extent(NULL, root, path, objectid: ino,
2019	bytenr: cur_offset, mod: 0);
2020	if (ret < 0)
2021	goto error;
2022
2023	/*
2024	* If there is no extent for our range when doing the initial
2025	* search, then go back to the previous slot as it will be the
2026	* one containing the search offset
2027	*/
2028	if (ret > 0 && path->slots[0] > 0 && check_prev) {
2029	leaf = path->nodes[0];
2030	btrfs_item_key_to_cpu(eb: leaf, cpu_key: &found_key,
2031	nr: path->slots[0] - 1);
2032	if (found_key.objectid == ino &&
2033	found_key.type == BTRFS_EXTENT_DATA_KEY)
2034	path->slots[0]--;
2035	}
2036	check_prev = false;
2037	next_slot:
2038	/* Go to next leaf if we have exhausted the current one */
2039	leaf = path->nodes[0];
2040	if (path->slots[0] >= btrfs_header_nritems(eb: leaf)) {
2041	ret = btrfs_next_leaf(root, path);
2042	if (ret < 0)
2043	goto error;
2044	if (ret > 0)
2045	break;
2046	leaf = path->nodes[0];
2047	}
2048
2049	btrfs_item_key_to_cpu(eb: leaf, cpu_key: &found_key, nr: path->slots[0]);
2050
2051	/* Didn't find anything for our INO */
2052	if (found_key.objectid > ino)
2053	break;
2054	/*
2055	* Keep searching until we find an EXTENT_ITEM or there are no
2056	* more extents for this inode
2057	*/
2058	if (WARN_ON_ONCE(found_key.objectid < ino) ||
2059	found_key.type < BTRFS_EXTENT_DATA_KEY) {
2060	path->slots[0]++;
2061	goto next_slot;
2062	}
2063
2064	/* Found key is not EXTENT_DATA_KEY or starts after req range */
2065	if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2066	found_key.offset > end)
2067	break;
2068
2069	/*
2070	* If the found extent starts after requested offset, then
2071	* adjust extent_end to be right before this extent begins
2072	*/
2073	if (found_key.offset > cur_offset) {
2074	extent_end = found_key.offset;
2075	extent_type = 0;
2076	goto must_cow;
2077	}
2078
2079	/*
2080	* Found extent which begins before our range and potentially
2081	* intersect it
2082	*/
2083	fi = btrfs_item_ptr(leaf, path->slots[0],
2084	struct btrfs_file_extent_item);
2085	extent_type = btrfs_file_extent_type(eb: leaf, s: fi);
2086	/* If this is triggered then we have a memory corruption. */
2087	ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2088	if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2089	ret = -EUCLEAN;
2090	goto error;
2091	}
2092	ram_bytes = btrfs_file_extent_ram_bytes(eb: leaf, s: fi);
2093	extent_end = btrfs_file_extent_end(path);
2094
2095	/*
2096	* If the extent we got ends before our current offset, skip to
2097	* the next extent.
2098	*/
2099	if (extent_end <= cur_offset) {
2100	path->slots[0]++;
2101	goto next_slot;
2102	}
2103
2104	nocow_args.start = cur_offset;
2105	ret = can_nocow_file_extent(path, key: &found_key, inode, args: &nocow_args);
2106	if (ret < 0)
2107	goto error;
2108	if (ret == 0)
2109	goto must_cow;
2110
2111	ret = 0;
2112	nocow_bg = btrfs_inc_nocow_writers(fs_info, bytenr: nocow_args.disk_bytenr);
2113	if (!nocow_bg) {
2114	must_cow:
2115	/*
2116	* If we can't perform NOCOW writeback for the range,
2117	* then record the beginning of the range that needs to
2118	* be COWed. It will be written out before the next
2119	* NOCOW range if we find one, or when exiting this
2120	* loop.
2121	*/
2122	if (cow_start == (u64)-1)
2123	cow_start = cur_offset;
2124	cur_offset = extent_end;
2125	if (cur_offset > end)
2126	break;
2127	if (!path->nodes[0])
2128	continue;
2129	path->slots[0]++;
2130	goto next_slot;
2131	}
2132
2133	/*
2134	* COW range from cow_start to found_key.offset - 1. As the key
2135	* will contain the beginning of the first extent that can be
2136	* NOCOW, following one which needs to be COW'ed
2137	*/
2138	if (cow_start != (u64)-1) {
2139	ret = fallback_to_cow(inode, locked_page,
2140	start: cow_start, end: found_key.offset - 1);
2141	cow_start = (u64)-1;
2142	if (ret) {
2143	btrfs_dec_nocow_writers(bg: nocow_bg);
2144	goto error;
2145	}
2146	}
2147
2148	nocow_end = cur_offset + nocow_args.num_bytes - 1;
2149	is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2150	if (is_prealloc) {
2151	u64 orig_start = found_key.offset - nocow_args.extent_offset;
2152	struct extent_map *em;
2153
2154	em = create_io_em(inode, start: cur_offset, len: nocow_args.num_bytes,
2155	orig_start,
2156	block_start: nocow_args.disk_bytenr, /* block_start */
2157	block_len: nocow_args.num_bytes, /* block_len */
2158	orig_block_len: nocow_args.disk_num_bytes, /* orig_block_len */
2159	ram_bytes, compress_type: BTRFS_COMPRESS_NONE,
2160	type: BTRFS_ORDERED_PREALLOC);
2161	if (IS_ERR(ptr: em)) {
2162	btrfs_dec_nocow_writers(bg: nocow_bg);
2163	ret = PTR_ERR(ptr: em);
2164	goto error;
2165	}
2166	free_extent_map(em);
2167	}
2168
2169	ordered = btrfs_alloc_ordered_extent(inode, file_offset: cur_offset,
2170	num_bytes: nocow_args.num_bytes, ram_bytes: nocow_args.num_bytes,
2171	disk_bytenr: nocow_args.disk_bytenr, disk_num_bytes: nocow_args.num_bytes, offset: 0,
2172	flags: is_prealloc
2173	? (1 << BTRFS_ORDERED_PREALLOC)
2174	: (1 << BTRFS_ORDERED_NOCOW),
2175	compress_type: BTRFS_COMPRESS_NONE);
2176	btrfs_dec_nocow_writers(bg: nocow_bg);
2177	if (IS_ERR(ptr: ordered)) {
2178	if (is_prealloc) {
2179	btrfs_drop_extent_map_range(inode, start: cur_offset,
2180	end: nocow_end, skip_pinned: false);
2181	}
2182	ret = PTR_ERR(ptr: ordered);
2183	goto error;
2184	}
2185
2186	if (btrfs_is_data_reloc_root(root))
2187	/*
2188	* Error handled later, as we must prevent
2189	* extent_clear_unlock_delalloc() in error handler
2190	* from freeing metadata of created ordered extent.
2191	*/
2192	ret = btrfs_reloc_clone_csums(ordered);
2193	btrfs_put_ordered_extent(entry: ordered);
2194
2195	extent_clear_unlock_delalloc(inode, start: cur_offset, end: nocow_end,
2196	locked_page, bits_to_clear: EXTENT_LOCKED |
2197	EXTENT_DELALLOC |
2198	EXTENT_CLEAR_DATA_RESV,
2199	page_ops: PAGE_UNLOCK | PAGE_SET_ORDERED);
2200
2201	cur_offset = extent_end;
2202
2203	/*
2204	* btrfs_reloc_clone_csums() error, now we're OK to call error
2205	* handler, as metadata for created ordered extent will only
2206	* be freed by btrfs_finish_ordered_io().
2207	*/
2208	if (ret)
2209	goto error;
2210	if (cur_offset > end)
2211	break;
2212	}
2213	btrfs_release_path(p: path);
2214
2215	if (cur_offset <= end && cow_start == (u64)-1)
2216	cow_start = cur_offset;
2217
2218	if (cow_start != (u64)-1) {
2219	cur_offset = end;
2220	ret = fallback_to_cow(inode, locked_page, start: cow_start, end);
2221	cow_start = (u64)-1;
2222	if (ret)
2223	goto error;
2224	}
2225
2226	btrfs_free_path(p: path);
2227	return 0;
2228
2229	error:
2230	/*
2231	* If an error happened while a COW region is outstanding, cur_offset
2232	* needs to be reset to cow_start to ensure the COW region is unlocked
2233	* as well.
2234	*/
2235	if (cow_start != (u64)-1)
2236	cur_offset = cow_start;
2237	if (cur_offset < end)
2238	extent_clear_unlock_delalloc(inode, start: cur_offset, end,
2239	locked_page, bits_to_clear: EXTENT_LOCKED |
2240	EXTENT_DELALLOC | EXTENT_DEFRAG |
2241	EXTENT_DO_ACCOUNTING, page_ops: PAGE_UNLOCK |
2242	PAGE_START_WRITEBACK |
2243	PAGE_END_WRITEBACK);
2244	btrfs_free_path(p: path);
2245	return ret;
2246	}
2247
2248	static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2249	{
2250	if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2251	if (inode->defrag_bytes &&
2252	test_range_bit_exists(tree: &inode->io_tree, start, end, bit: EXTENT_DEFRAG))
2253	return false;
2254	return true;
2255	}
2256	return false;
2257	}
2258
2259	/*
2260	* Function to process delayed allocation (create CoW) for ranges which are
2261	* being touched for the first time.
2262	*/
2263	int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2264	u64 start, u64 end, struct writeback_control *wbc)
2265	{
2266	const bool zoned = btrfs_is_zoned(fs_info: inode->root->fs_info);
2267	int ret;
2268
2269	/*
2270	* The range must cover part of the @locked_page, or a return of 1
2271	* can confuse the caller.
2272	*/
2273	ASSERT(!(end <= page_offset(locked_page) ||
2274	start >= page_offset(locked_page) + PAGE_SIZE));
2275
2276	if (should_nocow(inode, start, end)) {
2277	ret = run_delalloc_nocow(inode, locked_page, start, end);
2278	goto out;
2279	}
2280
2281	if (btrfs_inode_can_compress(inode) &&
2282	inode_need_compress(inode, start, end) &&
2283	run_delalloc_compressed(inode, locked_page, start, end, wbc))
2284	return 1;
2285
2286	if (zoned)
2287	ret = run_delalloc_cow(inode, locked_page, start, end, wbc,
2288	pages_dirty: true);
2289	else
2290	ret = cow_file_range(inode, locked_page, start, end, NULL,
2291	keep_locked: false, no_inline: false);
2292
2293	out:
2294	if (ret < 0)
2295	btrfs_cleanup_ordered_extents(inode, locked_page, offset: start,
2296	bytes: end - start + 1);
2297	return ret;
2298	}
2299
2300	void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2301	struct extent_state *orig, u64 split)
2302	{
2303	struct btrfs_fs_info *fs_info = inode->root->fs_info;
2304	u64 size;
2305
2306	lockdep_assert_held(&inode->io_tree.lock);
2307
2308	/* not delalloc, ignore it */
2309	if (!(orig->state & EXTENT_DELALLOC))
2310	return;
2311
2312	size = orig->end - orig->start + 1;
2313	if (size > fs_info->max_extent_size) {
2314	u32 num_extents;
2315	u64 new_size;
2316
2317	/*
2318	* See the explanation in btrfs_merge_delalloc_extent, the same
2319	* applies here, just in reverse.
2320	*/
2321	new_size = orig->end - split + 1;
2322	num_extents = count_max_extents(fs_info, size: new_size);
2323	new_size = split - orig->start;
2324	num_extents += count_max_extents(fs_info, size: new_size);
2325	if (count_max_extents(fs_info, size) >= num_extents)
2326	return;
2327	}
2328
2329	spin_lock(lock: &inode->lock);
2330	btrfs_mod_outstanding_extents(inode, mod: 1);
2331	spin_unlock(lock: &inode->lock);
2332	}
2333
2334	/*
2335	* Handle merged delayed allocation extents so we can keep track of new extents
2336	* that are just merged onto old extents, such as when we are doing sequential
2337	* writes, so we can properly account for the metadata space we'll need.
2338	*/
2339	void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2340	struct extent_state *other)
2341	{
2342	struct btrfs_fs_info *fs_info = inode->root->fs_info;
2343	u64 new_size, old_size;
2344	u32 num_extents;
2345
2346	lockdep_assert_held(&inode->io_tree.lock);
2347
2348	/* not delalloc, ignore it */
2349	if (!(other->state & EXTENT_DELALLOC))
2350	return;
2351
2352	if (new->start > other->start)
2353	new_size = new->end - other->start + 1;
2354	else
2355	new_size = other->end - new->start + 1;
2356
2357	/* we're not bigger than the max, unreserve the space and go */
2358	if (new_size <= fs_info->max_extent_size) {
2359	spin_lock(lock: &inode->lock);
2360	btrfs_mod_outstanding_extents(inode, mod: -1);
2361	spin_unlock(lock: &inode->lock);
2362	return;
2363	}
2364
2365	/*
2366	* We have to add up either side to figure out how many extents were
2367	* accounted for before we merged into one big extent. If the number of
2368	* extents we accounted for is <= the amount we need for the new range
2369	* then we can return, otherwise drop. Think of it like this
2370	*
2371	* [ 4k][MAX_SIZE]
2372	*
2373	* So we've grown the extent by a MAX_SIZE extent, this would mean we
2374	* need 2 outstanding extents, on one side we have 1 and the other side
2375	* we have 1 so they are == and we can return. But in this case
2376	*
2377	* [MAX_SIZE+4k][MAX_SIZE+4k]
2378	*
2379	* Each range on their own accounts for 2 extents, but merged together
2380	* they are only 3 extents worth of accounting, so we need to drop in
2381	* this case.
2382	*/
2383	old_size = other->end - other->start + 1;
2384	num_extents = count_max_extents(fs_info, size: old_size);
2385	old_size = new->end - new->start + 1;
2386	num_extents += count_max_extents(fs_info, size: old_size);
2387	if (count_max_extents(fs_info, size: new_size) >= num_extents)
2388	return;
2389
2390	spin_lock(lock: &inode->lock);
2391	btrfs_mod_outstanding_extents(inode, mod: -1);
2392	spin_unlock(lock: &inode->lock);
2393	}
2394
2395	static void btrfs_add_delalloc_inode(struct btrfs_inode *inode)
2396	{
2397	struct btrfs_root *root = inode->root;
2398	struct btrfs_fs_info *fs_info = root->fs_info;
2399
2400	spin_lock(lock: &root->delalloc_lock);
2401	ASSERT(list_empty(&inode->delalloc_inodes));
2402	list_add_tail(new: &inode->delalloc_inodes, head: &root->delalloc_inodes);
2403	root->nr_delalloc_inodes++;
2404	if (root->nr_delalloc_inodes == 1) {
2405	spin_lock(lock: &fs_info->delalloc_root_lock);
2406	ASSERT(list_empty(&root->delalloc_root));
2407	list_add_tail(new: &root->delalloc_root, head: &fs_info->delalloc_roots);
2408	spin_unlock(lock: &fs_info->delalloc_root_lock);
2409	}
2410	spin_unlock(lock: &root->delalloc_lock);
2411	}
2412
2413	void btrfs_del_delalloc_inode(struct btrfs_inode *inode)
2414	{
2415	struct btrfs_root *root = inode->root;
2416	struct btrfs_fs_info *fs_info = root->fs_info;
2417
2418	lockdep_assert_held(&root->delalloc_lock);
2419
2420	/*
2421	* We may be called after the inode was already deleted from the list,
2422	* namely in the transaction abort path btrfs_destroy_delalloc_inodes(),
2423	* and then later through btrfs_clear_delalloc_extent() while the inode
2424	* still has ->delalloc_bytes > 0.
2425	*/
2426	if (!list_empty(head: &inode->delalloc_inodes)) {
2427	list_del_init(entry: &inode->delalloc_inodes);
2428	root->nr_delalloc_inodes--;
2429	if (!root->nr_delalloc_inodes) {
2430	ASSERT(list_empty(&root->delalloc_inodes));
2431	spin_lock(lock: &fs_info->delalloc_root_lock);
2432	ASSERT(!list_empty(&root->delalloc_root));
2433	list_del_init(entry: &root->delalloc_root);
2434	spin_unlock(lock: &fs_info->delalloc_root_lock);
2435	}
2436	}
2437	}
2438
2439	/*
2440	* Properly track delayed allocation bytes in the inode and to maintain the
2441	* list of inodes that have pending delalloc work to be done.
2442	*/
2443	void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2444	u32 bits)
2445	{
2446	struct btrfs_fs_info *fs_info = inode->root->fs_info;
2447
2448	lockdep_assert_held(&inode->io_tree.lock);
2449
2450	if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2451	WARN_ON(1);
2452	/*
2453	* set_bit and clear bit hooks normally require _irqsave/restore
2454	* but in this case, we are only testing for the DELALLOC
2455	* bit, which is only set or cleared with irqs on
2456	*/
2457	if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2458	u64 len = state->end + 1 - state->start;
2459	u64 prev_delalloc_bytes;
2460	u32 num_extents = count_max_extents(fs_info, size: len);
2461
2462	spin_lock(lock: &inode->lock);
2463	btrfs_mod_outstanding_extents(inode, mod: num_extents);
2464	spin_unlock(lock: &inode->lock);
2465
2466	/* For sanity tests */
2467	if (btrfs_is_testing(fs_info))
2468	return;
2469
2470	percpu_counter_add_batch(fbc: &fs_info->delalloc_bytes, amount: len,
2471	batch: fs_info->delalloc_batch);
2472	spin_lock(lock: &inode->lock);
2473	prev_delalloc_bytes = inode->delalloc_bytes;
2474	inode->delalloc_bytes += len;
2475	if (bits & EXTENT_DEFRAG)
2476	inode->defrag_bytes += len;
2477	spin_unlock(lock: &inode->lock);
2478
2479	/*
2480	* We don't need to be under the protection of the inode's lock,
2481	* because we are called while holding the inode's io_tree lock
2482	* and are therefore protected against concurrent calls of this
2483	* function and btrfs_clear_delalloc_extent().
2484	*/
2485	if (!btrfs_is_free_space_inode(inode) && prev_delalloc_bytes == 0)
2486	btrfs_add_delalloc_inode(inode);
2487	}
2488
2489	if (!(state->state & EXTENT_DELALLOC_NEW) &&
2490	(bits & EXTENT_DELALLOC_NEW)) {
2491	spin_lock(lock: &inode->lock);
2492	inode->new_delalloc_bytes += state->end + 1 - state->start;
2493	spin_unlock(lock: &inode->lock);
2494	}
2495	}
2496
2497	/*
2498	* Once a range is no longer delalloc this function ensures that proper
2499	* accounting happens.
2500	*/
2501	void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2502	struct extent_state *state, u32 bits)
2503	{
2504	struct btrfs_fs_info *fs_info = inode->root->fs_info;
2505	u64 len = state->end + 1 - state->start;
2506	u32 num_extents = count_max_extents(fs_info, size: len);
2507
2508	lockdep_assert_held(&inode->io_tree.lock);
2509
2510	if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2511	spin_lock(lock: &inode->lock);
2512	inode->defrag_bytes -= len;
2513	spin_unlock(lock: &inode->lock);
2514	}
2515
2516	/*
2517	* set_bit and clear bit hooks normally require _irqsave/restore
2518	* but in this case, we are only testing for the DELALLOC
2519	* bit, which is only set or cleared with irqs on
2520	*/
2521	if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2522	struct btrfs_root *root = inode->root;
2523	u64 new_delalloc_bytes;
2524
2525	spin_lock(lock: &inode->lock);
2526	btrfs_mod_outstanding_extents(inode, mod: -num_extents);
2527	spin_unlock(lock: &inode->lock);
2528
2529	/*
2530	* We don't reserve metadata space for space cache inodes so we
2531	* don't need to call delalloc_release_metadata if there is an
2532	* error.
2533	*/
2534	if (bits & EXTENT_CLEAR_META_RESV &&
2535	root != fs_info->tree_root)
2536	btrfs_delalloc_release_metadata(inode, num_bytes: len, qgroup_free: true);
2537
2538	/* For sanity tests. */
2539	if (btrfs_is_testing(fs_info))
2540	return;
2541
2542	if (!btrfs_is_data_reloc_root(root) &&
2543	!btrfs_is_free_space_inode(inode) &&
2544	!(state->state & EXTENT_NORESERVE) &&
2545	(bits & EXTENT_CLEAR_DATA_RESV))
2546	btrfs_free_reserved_data_space_noquota(fs_info, len);
2547
2548	percpu_counter_add_batch(fbc: &fs_info->delalloc_bytes, amount: -len,
2549	batch: fs_info->delalloc_batch);
2550	spin_lock(lock: &inode->lock);
2551	inode->delalloc_bytes -= len;
2552	new_delalloc_bytes = inode->delalloc_bytes;
2553	spin_unlock(lock: &inode->lock);
2554
2555	/*
2556	* We don't need to be under the protection of the inode's lock,
2557	* because we are called while holding the inode's io_tree lock
2558	* and are therefore protected against concurrent calls of this
2559	* function and btrfs_set_delalloc_extent().
2560	*/
2561	if (!btrfs_is_free_space_inode(inode) && new_delalloc_bytes == 0) {
2562	spin_lock(lock: &root->delalloc_lock);
2563	btrfs_del_delalloc_inode(inode);
2564	spin_unlock(lock: &root->delalloc_lock);
2565	}
2566	}
2567
2568	if ((state->state & EXTENT_DELALLOC_NEW) &&
2569	(bits & EXTENT_DELALLOC_NEW)) {
2570	spin_lock(lock: &inode->lock);
2571	ASSERT(inode->new_delalloc_bytes >= len);
2572	inode->new_delalloc_bytes -= len;
2573	if (bits & EXTENT_ADD_INODE_BYTES)
2574	inode_add_bytes(inode: &inode->vfs_inode, bytes: len);
2575	spin_unlock(lock: &inode->lock);
2576	}
2577	}
2578
2579	static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2580	struct btrfs_ordered_extent *ordered)
2581	{
2582	u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2583	u64 len = bbio->bio.bi_iter.bi_size;
2584	struct btrfs_ordered_extent *new;
2585	int ret;
2586
2587	/* Must always be called for the beginning of an ordered extent. */
2588	if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2589	return -EINVAL;
2590
2591	/* No need to split if the ordered extent covers the entire bio. */
2592	if (ordered->disk_num_bytes == len) {
2593	refcount_inc(r: &ordered->refs);
2594	bbio->ordered = ordered;
2595	return 0;
2596	}
2597
2598	/*
2599	* Don't split the extent_map for NOCOW extents, as we're writing into
2600	* a pre-existing one.
2601	*/
2602	if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2603	ret = split_extent_map(inode: bbio->inode, start: bbio->file_offset,
2604	len: ordered->num_bytes, pre: len,
2605	new_logical: ordered->disk_bytenr);
2606	if (ret)
2607	return ret;
2608	}
2609
2610	new = btrfs_split_ordered_extent(ordered, len);
2611	if (IS_ERR(ptr: new))
2612	return PTR_ERR(ptr: new);
2613	bbio->ordered = new;
2614	return 0;
2615	}
2616
2617	/*
2618	* given a list of ordered sums record them in the inode. This happens
2619	* at IO completion time based on sums calculated at bio submission time.
2620	*/
2621	static int add_pending_csums(struct btrfs_trans_handle *trans,
2622	struct list_head *list)
2623	{
2624	struct btrfs_ordered_sum *sum;
2625	struct btrfs_root *csum_root = NULL;
2626	int ret;
2627
2628	list_for_each_entry(sum, list, list) {
2629	trans->adding_csums = true;
2630	if (!csum_root)
2631	csum_root = btrfs_csum_root(fs_info: trans->fs_info,
2632	bytenr: sum->logical);
2633	ret = btrfs_csum_file_blocks(trans, root: csum_root, sums: sum);
2634	trans->adding_csums = false;
2635	if (ret)
2636	return ret;
2637	}
2638	return 0;
2639	}
2640
2641	static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2642	const u64 start,
2643	const u64 len,
2644	struct extent_state **cached_state)
2645	{
2646	u64 search_start = start;
2647	const u64 end = start + len - 1;
2648
2649	while (search_start < end) {
2650	const u64 search_len = end - search_start + 1;
2651	struct extent_map *em;
2652	u64 em_len;
2653	int ret = 0;
2654
2655	em = btrfs_get_extent(inode, NULL, start: search_start, len: search_len);
2656	if (IS_ERR(ptr: em))
2657	return PTR_ERR(ptr: em);
2658
2659	if (em->block_start != EXTENT_MAP_HOLE)
2660	goto next;
2661
2662	em_len = em->len;
2663	if (em->start < search_start)
2664	em_len -= search_start - em->start;
2665	if (em_len > search_len)
2666	em_len = search_len;
2667
2668	ret = set_extent_bit(tree: &inode->io_tree, start: search_start,
2669	end: search_start + em_len - 1,
2670	bits: EXTENT_DELALLOC_NEW, cached_state);
2671	next:
2672	search_start = extent_map_end(em);
2673	free_extent_map(em);
2674	if (ret)
2675	return ret;
2676	}
2677	return 0;
2678	}
2679
2680	int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2681	unsigned int extra_bits,
2682	struct extent_state **cached_state)
2683	{
2684	WARN_ON(PAGE_ALIGNED(end));
2685
2686	if (start >= i_size_read(inode: &inode->vfs_inode) &&
2687	!(inode->flags & BTRFS_INODE_PREALLOC)) {
2688	/*
2689	* There can't be any extents following eof in this case so just
2690	* set the delalloc new bit for the range directly.
2691	*/
2692	extra_bits |= EXTENT_DELALLOC_NEW;
2693	} else {
2694	int ret;
2695
2696	ret = btrfs_find_new_delalloc_bytes(inode, start,
2697	len: end + 1 - start,
2698	cached_state);
2699	if (ret)
2700	return ret;
2701	}
2702
2703	return set_extent_bit(tree: &inode->io_tree, start, end,
2704	bits: EXTENT_DELALLOC | extra_bits, cached_state);
2705	}
2706
2707	/* see btrfs_writepage_start_hook for details on why this is required */
2708	struct btrfs_writepage_fixup {
2709	struct page *page;
2710	struct btrfs_inode *inode;
2711	struct btrfs_work work;
2712	};
2713
2714	static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2715	{
2716	struct btrfs_writepage_fixup *fixup =
2717	container_of(work, struct btrfs_writepage_fixup, work);
2718	struct btrfs_ordered_extent *ordered;
2719	struct extent_state *cached_state = NULL;
2720	struct extent_changeset *data_reserved = NULL;
2721	struct page *page = fixup->page;
2722	struct btrfs_inode *inode = fixup->inode;
2723	struct btrfs_fs_info *fs_info = inode->root->fs_info;
2724	u64 page_start = page_offset(page);
2725	u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2726	int ret = 0;
2727	bool free_delalloc_space = true;
2728
2729	/*
2730	* This is similar to page_mkwrite, we need to reserve the space before
2731	* we take the page lock.
2732	*/
2733	ret = btrfs_delalloc_reserve_space(inode, reserved: &data_reserved, start: page_start,
2734	PAGE_SIZE);
2735	again:
2736	lock_page(page);
2737
2738	/*
2739	* Before we queued this fixup, we took a reference on the page.
2740	* page->mapping may go NULL, but it shouldn't be moved to a different
2741	* address space.
2742	*/
2743	if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2744	/*
2745	* Unfortunately this is a little tricky, either
2746	*
2747	* 1) We got here and our page had already been dealt with and
2748	* we reserved our space, thus ret == 0, so we need to just
2749	* drop our space reservation and bail. This can happen the
2750	* first time we come into the fixup worker, or could happen
2751	* while waiting for the ordered extent.
2752	* 2) Our page was already dealt with, but we happened to get an
2753	* ENOSPC above from the btrfs_delalloc_reserve_space. In
2754	* this case we obviously don't have anything to release, but
2755	* because the page was already dealt with we don't want to
2756	* mark the page with an error, so make sure we're resetting
2757	* ret to 0. This is why we have this check _before_ the ret
2758	* check, because we do not want to have a surprise ENOSPC
2759	* when the page was already properly dealt with.
2760	*/
2761	if (!ret) {
2762	btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2763	btrfs_delalloc_release_space(inode, reserved: data_reserved,
2764	start: page_start, PAGE_SIZE,
2765	qgroup_free: true);
2766	}
2767	ret = 0;
2768	goto out_page;
2769	}
2770
2771	/*
2772	* We can't mess with the page state unless it is locked, so now that
2773	* it is locked bail if we failed to make our space reservation.
2774	*/
2775	if (ret)
2776	goto out_page;
2777
2778	lock_extent(tree: &inode->io_tree, start: page_start, end: page_end, cached: &cached_state);
2779
2780	/* already ordered? We're done */
2781	if (PageOrdered(page))
2782	goto out_reserved;
2783
2784	ordered = btrfs_lookup_ordered_range(inode, file_offset: page_start, PAGE_SIZE);
2785	if (ordered) {
2786	unlock_extent(tree: &inode->io_tree, start: page_start, end: page_end,
2787	cached: &cached_state);
2788	unlock_page(page);
2789	btrfs_start_ordered_extent(entry: ordered);
2790	btrfs_put_ordered_extent(entry: ordered);
2791	goto again;
2792	}
2793
2794	ret = btrfs_set_extent_delalloc(inode, start: page_start, end: page_end, extra_bits: 0,
2795	cached_state: &cached_state);
2796	if (ret)
2797	goto out_reserved;
2798
2799	/*
2800	* Everything went as planned, we're now the owner of a dirty page with
2801	* delayed allocation bits set and space reserved for our COW
2802	* destination.
2803	*
2804	* The page was dirty when we started, nothing should have cleaned it.
2805	*/
2806	BUG_ON(!PageDirty(page));
2807	free_delalloc_space = false;
2808	out_reserved:
2809	btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2810	if (free_delalloc_space)
2811	btrfs_delalloc_release_space(inode, reserved: data_reserved, start: page_start,
2812	PAGE_SIZE, qgroup_free: true);
2813	unlock_extent(tree: &inode->io_tree, start: page_start, end: page_end, cached: &cached_state);
2814	out_page:
2815	if (ret) {
2816	/*
2817	* We hit ENOSPC or other errors. Update the mapping and page
2818	* to reflect the errors and clean the page.
2819	*/
2820	mapping_set_error(mapping: page->mapping, error: ret);
2821	btrfs_mark_ordered_io_finished(inode, page, file_offset: page_start,
2822	PAGE_SIZE, uptodate: !ret);
2823	clear_page_dirty_for_io(page);
2824	}
2825	btrfs_folio_clear_checked(fs_info, page_folio(page), start: page_start, PAGE_SIZE);
2826	unlock_page(page);
2827	put_page(page);
2828	kfree(objp: fixup);
2829	extent_changeset_free(changeset: data_reserved);
2830	/*
2831	* As a precaution, do a delayed iput in case it would be the last iput
2832	* that could need flushing space. Recursing back to fixup worker would
2833	* deadlock.
2834	*/
2835	btrfs_add_delayed_iput(inode);
2836	}
2837
2838	/*
2839	* There are a few paths in the higher layers of the kernel that directly
2840	* set the page dirty bit without asking the filesystem if it is a
2841	* good idea. This causes problems because we want to make sure COW
2842	* properly happens and the data=ordered rules are followed.
2843	*
2844	* In our case any range that doesn't have the ORDERED bit set
2845	* hasn't been properly setup for IO. We kick off an async process
2846	* to fix it up. The async helper will wait for ordered extents, set
2847	* the delalloc bit and make it safe to write the page.
2848	*/
2849	int btrfs_writepage_cow_fixup(struct page *page)
2850	{
2851	struct inode *inode = page->mapping->host;
2852	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2853	struct btrfs_writepage_fixup *fixup;
2854
2855	/* This page has ordered extent covering it already */
2856	if (PageOrdered(page))
2857	return 0;
2858
2859	/*
2860	* PageChecked is set below when we create a fixup worker for this page,
2861	* don't try to create another one if we're already PageChecked()
2862	*
2863	* The extent_io writepage code will redirty the page if we send back
2864	* EAGAIN.
2865	*/
2866	if (PageChecked(page))
2867	return -EAGAIN;
2868
2869	fixup = kzalloc(size: sizeof(*fixup), GFP_NOFS);
2870	if (!fixup)
2871	return -EAGAIN;
2872
2873	/*
2874	* We are already holding a reference to this inode from
2875	* write_cache_pages. We need to hold it because the space reservation
2876	* takes place outside of the page lock, and we can't trust
2877	* page->mapping outside of the page lock.
2878	*/
2879	ihold(inode);
2880	btrfs_folio_set_checked(fs_info, page_folio(page), start: page_offset(page), PAGE_SIZE);
2881	get_page(page);
2882	btrfs_init_work(work: &fixup->work, func: btrfs_writepage_fixup_worker, NULL);
2883	fixup->page = page;
2884	fixup->inode = BTRFS_I(inode);
2885	btrfs_queue_work(wq: fs_info->fixup_workers, work: &fixup->work);
2886
2887	return -EAGAIN;
2888	}
2889
2890	static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2891	struct btrfs_inode *inode, u64 file_pos,
2892	struct btrfs_file_extent_item *stack_fi,
2893	const bool update_inode_bytes,
2894	u64 qgroup_reserved)
2895	{
2896	struct btrfs_root *root = inode->root;
2897	const u64 sectorsize = root->fs_info->sectorsize;
2898	struct btrfs_path *path;
2899	struct extent_buffer *leaf;
2900	struct btrfs_key ins;
2901	u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(s: stack_fi);
2902	u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(s: stack_fi);
2903	u64 offset = btrfs_stack_file_extent_offset(s: stack_fi);
2904	u64 num_bytes = btrfs_stack_file_extent_num_bytes(s: stack_fi);
2905	u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(s: stack_fi);
2906	struct btrfs_drop_extents_args drop_args = { 0 };
2907	int ret;
2908
2909	path = btrfs_alloc_path();
2910	if (!path)
2911	return -ENOMEM;
2912
2913	/*
2914	* we may be replacing one extent in the tree with another.
2915	* The new extent is pinned in the extent map, and we don't want
2916	* to drop it from the cache until it is completely in the btree.
2917	*
2918	* So, tell btrfs_drop_extents to leave this extent in the cache.
2919	* the caller is expected to unpin it and allow it to be merged
2920	* with the others.
2921	*/
2922	drop_args.path = path;
2923	drop_args.start = file_pos;
2924	drop_args.end = file_pos + num_bytes;
2925	drop_args.replace_extent = true;
2926	drop_args.extent_item_size = sizeof(*stack_fi);
2927	ret = btrfs_drop_extents(trans, root, inode, args: &drop_args);
2928	if (ret)
2929	goto out;
2930
2931	if (!drop_args.extent_inserted) {
2932	ins.objectid = btrfs_ino(inode);
2933	ins.offset = file_pos;
2934	ins.type = BTRFS_EXTENT_DATA_KEY;
2935
2936	ret = btrfs_insert_empty_item(trans, root, path, key: &ins,
2937	data_size: sizeof(*stack_fi));
2938	if (ret)
2939	goto out;
2940	}
2941	leaf = path->nodes[0];
2942	btrfs_set_stack_file_extent_generation(s: stack_fi, val: trans->transid);
2943	write_extent_buffer(eb: leaf, src: stack_fi,
2944	btrfs_item_ptr_offset(leaf, path->slots[0]),
2945	len: sizeof(struct btrfs_file_extent_item));
2946
2947	btrfs_mark_buffer_dirty(trans, buf: leaf);
2948	btrfs_release_path(p: path);
2949
2950	/*
2951	* If we dropped an inline extent here, we know the range where it is
2952	* was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2953	* number of bytes only for that range containing the inline extent.
2954	* The remaining of the range will be processed when clearning the
2955	* EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2956	*/
2957	if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2958	u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2959
2960	inline_size = drop_args.bytes_found - inline_size;
2961	btrfs_update_inode_bytes(inode, add_bytes: sectorsize, del_bytes: inline_size);
2962	drop_args.bytes_found -= inline_size;
2963	num_bytes -= sectorsize;
2964	}
2965
2966	if (update_inode_bytes)
2967	btrfs_update_inode_bytes(inode, add_bytes: num_bytes, del_bytes: drop_args.bytes_found);
2968
2969	ins.objectid = disk_bytenr;
2970	ins.offset = disk_num_bytes;
2971	ins.type = BTRFS_EXTENT_ITEM_KEY;
2972
2973	ret = btrfs_inode_set_file_extent_range(inode, start: file_pos, len: ram_bytes);
2974	if (ret)
2975	goto out;
2976
2977	ret = btrfs_alloc_reserved_file_extent(trans, root, owner: btrfs_ino(inode),
2978	offset: file_pos - offset,
2979	ram_bytes: qgroup_reserved, ins: &ins);
2980	out:
2981	btrfs_free_path(p: path);
2982
2983	return ret;
2984	}
2985
2986	static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2987	u64 start, u64 len)
2988	{
2989	struct btrfs_block_group *cache;
2990
2991	cache = btrfs_lookup_block_group(info: fs_info, bytenr: start);
2992	ASSERT(cache);
2993
2994	spin_lock(lock: &cache->lock);
2995	cache->delalloc_bytes -= len;
2996	spin_unlock(lock: &cache->lock);
2997
2998	btrfs_put_block_group(cache);
2999	}
3000
3001	static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3002	struct btrfs_ordered_extent *oe)
3003	{
3004	struct btrfs_file_extent_item stack_fi;
3005	bool update_inode_bytes;
3006	u64 num_bytes = oe->num_bytes;
3007	u64 ram_bytes = oe->ram_bytes;
3008
3009	memset(&stack_fi, 0, sizeof(stack_fi));
3010	btrfs_set_stack_file_extent_type(s: &stack_fi, val: BTRFS_FILE_EXTENT_REG);
3011	btrfs_set_stack_file_extent_disk_bytenr(s: &stack_fi, val: oe->disk_bytenr);
3012	btrfs_set_stack_file_extent_disk_num_bytes(s: &stack_fi,
3013	val: oe->disk_num_bytes);
3014	btrfs_set_stack_file_extent_offset(s: &stack_fi, val: oe->offset);
3015	if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3016	num_bytes = oe->truncated_len;
3017	ram_bytes = num_bytes;
3018	}
3019	btrfs_set_stack_file_extent_num_bytes(s: &stack_fi, val: num_bytes);
3020	btrfs_set_stack_file_extent_ram_bytes(s: &stack_fi, val: ram_bytes);
3021	btrfs_set_stack_file_extent_compression(s: &stack_fi, val: oe->compress_type);
3022	/* Encryption and other encoding is reserved and all 0 */
3023
3024	/*
3025	* For delalloc, when completing an ordered extent we update the inode's
3026	* bytes when clearing the range in the inode's io tree, so pass false
3027	* as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3028	* except if the ordered extent was truncated.
3029	*/
3030	update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3031	test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3032	test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3033
3034	return insert_reserved_file_extent(trans, inode: BTRFS_I(inode: oe->inode),
3035	file_pos: oe->file_offset, stack_fi: &stack_fi,
3036	update_inode_bytes, qgroup_reserved: oe->qgroup_rsv);
3037	}
3038
3039	/*
3040	* As ordered data IO finishes, this gets called so we can finish
3041	* an ordered extent if the range of bytes in the file it covers are
3042	* fully written.
3043	*/
3044	int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3045	{
3046	struct btrfs_inode *inode = BTRFS_I(inode: ordered_extent->inode);
3047	struct btrfs_root *root = inode->root;
3048	struct btrfs_fs_info *fs_info = root->fs_info;
3049	struct btrfs_trans_handle *trans = NULL;
3050	struct extent_io_tree *io_tree = &inode->io_tree;
3051	struct extent_state *cached_state = NULL;
3052	u64 start, end;
3053	int compress_type = 0;
3054	int ret = 0;
3055	u64 logical_len = ordered_extent->num_bytes;
3056	bool freespace_inode;
3057	bool truncated = false;
3058	bool clear_reserved_extent = true;
3059	unsigned int clear_bits = EXTENT_DEFRAG;
3060
3061	start = ordered_extent->file_offset;
3062	end = start + ordered_extent->num_bytes - 1;
3063
3064	if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3065	!test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3066	!test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3067	!test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3068	clear_bits |= EXTENT_DELALLOC_NEW;
3069
3070	freespace_inode = btrfs_is_free_space_inode(inode);
3071	if (!freespace_inode)
3072	btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3073
3074	if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3075	ret = -EIO;
3076	goto out;
3077	}
3078
3079	if (btrfs_is_zoned(fs_info))
3080	btrfs_zone_finish_endio(fs_info, logical: ordered_extent->disk_bytenr,
3081	length: ordered_extent->disk_num_bytes);
3082
3083	if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3084	truncated = true;
3085	logical_len = ordered_extent->truncated_len;
3086	/* Truncated the entire extent, don't bother adding */
3087	if (!logical_len)
3088	goto out;
3089	}
3090
3091	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3092	BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3093
3094	btrfs_inode_safe_disk_i_size_write(inode, new_i_size: 0);
3095	if (freespace_inode)
3096	trans = btrfs_join_transaction_spacecache(root);
3097	else
3098	trans = btrfs_join_transaction(root);
3099	if (IS_ERR(ptr: trans)) {
3100	ret = PTR_ERR(ptr: trans);
3101	trans = NULL;
3102	goto out;
3103	}
3104	trans->block_rsv = &inode->block_rsv;
3105	ret = btrfs_update_inode_fallback(trans, inode);
3106	if (ret) /* -ENOMEM or corruption */
3107	btrfs_abort_transaction(trans, ret);
3108	goto out;
3109	}
3110
3111	clear_bits |= EXTENT_LOCKED;
3112	lock_extent(tree: io_tree, start, end, cached: &cached_state);
3113
3114	if (freespace_inode)
3115	trans = btrfs_join_transaction_spacecache(root);
3116	else
3117	trans = btrfs_join_transaction(root);
3118	if (IS_ERR(ptr: trans)) {
3119	ret = PTR_ERR(ptr: trans);
3120	trans = NULL;
3121	goto out;
3122	}
3123
3124	trans->block_rsv = &inode->block_rsv;
3125
3126	ret = btrfs_insert_raid_extent(trans, ordered_extent);
3127	if (ret)
3128	goto out;
3129
3130	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3131	compress_type = ordered_extent->compress_type;
3132	if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3133	BUG_ON(compress_type);
3134	ret = btrfs_mark_extent_written(trans, inode,
3135	start: ordered_extent->file_offset,
3136	end: ordered_extent->file_offset +
3137	logical_len);
3138	btrfs_zoned_release_data_reloc_bg(fs_info, logical: ordered_extent->disk_bytenr,
3139	length: ordered_extent->disk_num_bytes);
3140	} else {
3141	BUG_ON(root == fs_info->tree_root);
3142	ret = insert_ordered_extent_file_extent(trans, oe: ordered_extent);
3143	if (!ret) {
3144	clear_reserved_extent = false;
3145	btrfs_release_delalloc_bytes(fs_info,
3146	start: ordered_extent->disk_bytenr,
3147	len: ordered_extent->disk_num_bytes);
3148	}
3149	}
3150	if (ret < 0) {
3151	btrfs_abort_transaction(trans, ret);
3152	goto out;
3153	}
3154
3155	ret = unpin_extent_cache(inode, start: ordered_extent->file_offset,
3156	len: ordered_extent->num_bytes, gen: trans->transid);
3157	if (ret < 0) {
3158	btrfs_abort_transaction(trans, ret);
3159	goto out;
3160	}
3161
3162	ret = add_pending_csums(trans, list: &ordered_extent->list);
3163	if (ret) {
3164	btrfs_abort_transaction(trans, ret);
3165	goto out;
3166	}
3167
3168	/*
3169	* If this is a new delalloc range, clear its new delalloc flag to
3170	* update the inode's number of bytes. This needs to be done first
3171	* before updating the inode item.
3172	*/
3173	if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3174	!test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3175	clear_extent_bit(tree: &inode->io_tree, start, end,
3176	bits: EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3177	cached: &cached_state);
3178
3179	btrfs_inode_safe_disk_i_size_write(inode, new_i_size: 0);
3180	ret = btrfs_update_inode_fallback(trans, inode);
3181	if (ret) { /* -ENOMEM or corruption */
3182	btrfs_abort_transaction(trans, ret);
3183	goto out;
3184	}
3185	ret = 0;
3186	out:
3187	clear_extent_bit(tree: &inode->io_tree, start, end, bits: clear_bits,
3188	cached: &cached_state);
3189
3190	if (trans)
3191	btrfs_end_transaction(trans);
3192
3193	if (ret || truncated) {
3194	u64 unwritten_start = start;
3195
3196	/*
3197	* If we failed to finish this ordered extent for any reason we
3198	* need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3199	* extent, and mark the inode with the error if it wasn't
3200	* already set. Any error during writeback would have already
3201	* set the mapping error, so we need to set it if we're the ones
3202	* marking this ordered extent as failed.
3203	*/
3204	if (ret && !test_and_set_bit(nr: BTRFS_ORDERED_IOERR,
3205	addr: &ordered_extent->flags))
3206	mapping_set_error(mapping: ordered_extent->inode->i_mapping, error: -EIO);
3207
3208	if (truncated)
3209	unwritten_start += logical_len;
3210	clear_extent_uptodate(tree: io_tree, start: unwritten_start, end, NULL);
3211
3212	/*
3213	* Drop extent maps for the part of the extent we didn't write.
3214	*
3215	* We have an exception here for the free_space_inode, this is
3216	* because when we do btrfs_get_extent() on the free space inode
3217	* we will search the commit root. If this is a new block group
3218	* we won't find anything, and we will trip over the assert in
3219	* writepage where we do ASSERT(em->block_start !=
3220	* EXTENT_MAP_HOLE).
3221	*
3222	* Theoretically we could also skip this for any NOCOW extent as
3223	* we don't mess with the extent map tree in the NOCOW case, but
3224	* for now simply skip this if we are the free space inode.
3225	*/
3226	if (!btrfs_is_free_space_inode(inode))
3227	btrfs_drop_extent_map_range(inode, start: unwritten_start,
3228	end, skip_pinned: false);
3229
3230	/*
3231	* If the ordered extent had an IOERR or something else went
3232	* wrong we need to return the space for this ordered extent
3233	* back to the allocator. We only free the extent in the
3234	* truncated case if we didn't write out the extent at all.
3235	*
3236	* If we made it past insert_reserved_file_extent before we
3237	* errored out then we don't need to do this as the accounting
3238	* has already been done.
3239	*/
3240	if ((ret || !logical_len) &&
3241	clear_reserved_extent &&
3242	!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3243	!test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3244	/*
3245	* Discard the range before returning it back to the
3246	* free space pool
3247	*/
3248	if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3249	btrfs_discard_extent(fs_info,
3250	bytenr: ordered_extent->disk_bytenr,
3251	num_bytes: ordered_extent->disk_num_bytes,
3252	NULL);
3253	btrfs_free_reserved_extent(fs_info,
3254	start: ordered_extent->disk_bytenr,
3255	len: ordered_extent->disk_num_bytes, delalloc: 1);
3256	/*
3257	* Actually free the qgroup rsv which was released when
3258	* the ordered extent was created.
3259	*/
3260	btrfs_qgroup_free_refroot(fs_info, ref_root: inode->root->root_key.objectid,
3261	num_bytes: ordered_extent->qgroup_rsv,
3262	type: BTRFS_QGROUP_RSV_DATA);
3263	}
3264	}
3265
3266	/*
3267	* This needs to be done to make sure anybody waiting knows we are done
3268	* updating everything for this ordered extent.
3269	*/
3270	btrfs_remove_ordered_extent(btrfs_inode: inode, entry: ordered_extent);
3271
3272	/* once for us */
3273	btrfs_put_ordered_extent(entry: ordered_extent);
3274	/* once for the tree */
3275	btrfs_put_ordered_extent(entry: ordered_extent);
3276
3277	return ret;
3278	}
3279
3280	int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3281	{
3282	if (btrfs_is_zoned(inode_to_fs_info(ordered->inode)) &&
3283	!test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3284	list_empty(head: &ordered->bioc_list))
3285	btrfs_finish_ordered_zoned(ordered);
3286	return btrfs_finish_one_ordered(ordered_extent: ordered);
3287	}
3288
3289	/*
3290	* Verify the checksum for a single sector without any extra action that depend
3291	* on the type of I/O.
3292	*/
3293	int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3294	u32 pgoff, u8 *csum, const u8 * const csum_expected)
3295	{
3296	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3297	char *kaddr;
3298
3299	ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3300
3301	shash->tfm = fs_info->csum_shash;
3302
3303	kaddr = kmap_local_page(page) + pgoff;
3304	crypto_shash_digest(desc: shash, data: kaddr, len: fs_info->sectorsize, out: csum);
3305	kunmap_local(kaddr);
3306
3307	if (memcmp(p: csum, q: csum_expected, size: fs_info->csum_size))
3308	return -EIO;
3309	return 0;
3310	}
3311
3312	/*
3313	* Verify the checksum of a single data sector.
3314	*
3315	* @bbio: btrfs_io_bio which contains the csum
3316	* @dev: device the sector is on
3317	* @bio_offset: offset to the beginning of the bio (in bytes)
3318	* @bv: bio_vec to check
3319	*
3320	* Check if the checksum on a data block is valid. When a checksum mismatch is
3321	* detected, report the error and fill the corrupted range with zero.
3322	*
3323	* Return %true if the sector is ok or had no checksum to start with, else %false.
3324	*/
3325	bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3326	u32 bio_offset, struct bio_vec *bv)
3327	{
3328	struct btrfs_inode *inode = bbio->inode;
3329	struct btrfs_fs_info *fs_info = inode->root->fs_info;
3330	u64 file_offset = bbio->file_offset + bio_offset;
3331	u64 end = file_offset + bv->bv_len - 1;
3332	u8 *csum_expected;
3333	u8 csum[BTRFS_CSUM_SIZE];
3334
3335	ASSERT(bv->bv_len == fs_info->sectorsize);
3336
3337	if (!bbio->csum)
3338	return true;
3339
3340	if (btrfs_is_data_reloc_root(root: inode->root) &&
3341	test_range_bit(tree: &inode->io_tree, start: file_offset, end, bit: EXTENT_NODATASUM,
3342	NULL)) {
3343	/* Skip the range without csum for data reloc inode */
3344	clear_extent_bits(tree: &inode->io_tree, start: file_offset, end,
3345	bits: EXTENT_NODATASUM);
3346	return true;
3347	}
3348
3349	csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3350	fs_info->csum_size;
3351	if (btrfs_check_sector_csum(fs_info, page: bv->bv_page, pgoff: bv->bv_offset, csum,
3352	csum_expected))
3353	goto zeroit;
3354	return true;
3355
3356	zeroit:
3357	btrfs_print_data_csum_error(inode, logical_start: file_offset, csum, csum_expected,
3358	mirror_num: bbio->mirror_num);
3359	if (dev)
3360	btrfs_dev_stat_inc_and_print(dev, index: BTRFS_DEV_STAT_CORRUPTION_ERRS);
3361	memzero_bvec(bvec: bv);
3362	return false;
3363	}
3364
3365	/*
3366	* Perform a delayed iput on @inode.
3367	*
3368	* @inode: The inode we want to perform iput on
3369	*
3370	* This function uses the generic vfs_inode::i_count to track whether we should
3371	* just decrement it (in case it's > 1) or if this is the last iput then link
3372	* the inode to the delayed iput machinery. Delayed iputs are processed at
3373	* transaction commit time/superblock commit/cleaner kthread.
3374	*/
3375	void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3376	{
3377	struct btrfs_fs_info *fs_info = inode->root->fs_info;
3378	unsigned long flags;
3379
3380	if (atomic_add_unless(v: &inode->vfs_inode.i_count, a: -1, u: 1))
3381	return;
3382
3383	atomic_inc(v: &fs_info->nr_delayed_iputs);
3384	/*
3385	* Need to be irq safe here because we can be called from either an irq
3386	* context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3387	* context.
3388	*/
3389	spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3390	ASSERT(list_empty(&inode->delayed_iput));
3391	list_add_tail(new: &inode->delayed_iput, head: &fs_info->delayed_iputs);
3392	spin_unlock_irqrestore(lock: &fs_info->delayed_iput_lock, flags);
3393	if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3394	wake_up_process(tsk: fs_info->cleaner_kthread);
3395	}
3396
3397	static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3398	struct btrfs_inode *inode)
3399	{
3400	list_del_init(entry: &inode->delayed_iput);
3401	spin_unlock_irq(lock: &fs_info->delayed_iput_lock);
3402	iput(&inode->vfs_inode);
3403	if (atomic_dec_and_test(v: &fs_info->nr_delayed_iputs))
3404	wake_up(&fs_info->delayed_iputs_wait);
3405	spin_lock_irq(lock: &fs_info->delayed_iput_lock);
3406	}
3407
3408	static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3409	struct btrfs_inode *inode)
3410	{
3411	if (!list_empty(head: &inode->delayed_iput)) {
3412	spin_lock_irq(lock: &fs_info->delayed_iput_lock);
3413	if (!list_empty(head: &inode->delayed_iput))
3414	run_delayed_iput_locked(fs_info, inode);
3415	spin_unlock_irq(lock: &fs_info->delayed_iput_lock);
3416	}
3417	}
3418
3419	void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3420	{
3421	/*
3422	* btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3423	* calls btrfs_add_delayed_iput() and that needs to lock
3424	* fs_info->delayed_iput_lock. So we need to disable irqs here to
3425	* prevent a deadlock.
3426	*/
3427	spin_lock_irq(lock: &fs_info->delayed_iput_lock);
3428	while (!list_empty(head: &fs_info->delayed_iputs)) {
3429	struct btrfs_inode *inode;
3430
3431	inode = list_first_entry(&fs_info->delayed_iputs,
3432	struct btrfs_inode, delayed_iput);
3433	run_delayed_iput_locked(fs_info, inode);
3434	if (need_resched()) {
3435	spin_unlock_irq(lock: &fs_info->delayed_iput_lock);
3436	cond_resched();
3437	spin_lock_irq(lock: &fs_info->delayed_iput_lock);
3438	}
3439	}
3440	spin_unlock_irq(lock: &fs_info->delayed_iput_lock);
3441	}
3442
3443	/*
3444	* Wait for flushing all delayed iputs
3445	*
3446	* @fs_info: the filesystem
3447	*
3448	* This will wait on any delayed iputs that are currently running with KILLABLE
3449	* set. Once they are all done running we will return, unless we are killed in
3450	* which case we return EINTR. This helps in user operations like fallocate etc
3451	* that might get blocked on the iputs.
3452	*
3453	* Return EINTR if we were killed, 0 if nothing's pending
3454	*/
3455	int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3456	{
3457	int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3458	atomic_read(&fs_info->nr_delayed_iputs) == 0);
3459	if (ret)
3460	return -EINTR;
3461	return 0;
3462	}
3463
3464	/*
3465	* This creates an orphan entry for the given inode in case something goes wrong
3466	* in the middle of an unlink.
3467	*/
3468	int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3469	struct btrfs_inode *inode)
3470	{
3471	int ret;
3472
3473	ret = btrfs_insert_orphan_item(trans, root: inode->root, offset: btrfs_ino(inode));
3474	if (ret && ret != -EEXIST) {
3475	btrfs_abort_transaction(trans, ret);
3476	return ret;
3477	}
3478
3479	return 0;
3480	}
3481
3482	/*
3483	* We have done the delete so we can go ahead and remove the orphan item for
3484	* this particular inode.
3485	*/
3486	static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3487	struct btrfs_inode *inode)
3488	{
3489	return btrfs_del_orphan_item(trans, root: inode->root, offset: btrfs_ino(inode));
3490	}
3491
3492	/*
3493	* this cleans up any orphans that may be left on the list from the last use
3494	* of this root.
3495	*/
3496	int btrfs_orphan_cleanup(struct btrfs_root *root)
3497	{
3498	struct btrfs_fs_info *fs_info = root->fs_info;
3499	struct btrfs_path *path;
3500	struct extent_buffer *leaf;
3501	struct btrfs_key key, found_key;
3502	struct btrfs_trans_handle *trans;
3503	struct inode *inode;
3504	u64 last_objectid = 0;
3505	int ret = 0, nr_unlink = 0;
3506
3507	if (test_and_set_bit(nr: BTRFS_ROOT_ORPHAN_CLEANUP, addr: &root->state))
3508	return 0;
3509
3510	path = btrfs_alloc_path();
3511	if (!path) {
3512	ret = -ENOMEM;
3513	goto out;
3514	}
3515	path->reada = READA_BACK;
3516
3517	key.objectid = BTRFS_ORPHAN_OBJECTID;
3518	key.type = BTRFS_ORPHAN_ITEM_KEY;
3519	key.offset = (u64)-1;
3520
3521	while (1) {
3522	ret = btrfs_search_slot(NULL, root, key: &key, p: path, ins_len: 0, cow: 0);
3523	if (ret < 0)
3524	goto out;
3525
3526	/*
3527	* if ret == 0 means we found what we were searching for, which
3528	* is weird, but possible, so only screw with path if we didn't
3529	* find the key and see if we have stuff that matches
3530	*/
3531	if (ret > 0) {
3532	ret = 0;
3533	if (path->slots[0] == 0)
3534	break;
3535	path->slots[0]--;
3536	}
3537
3538	/* pull out the item */
3539	leaf = path->nodes[0];
3540	btrfs_item_key_to_cpu(eb: leaf, cpu_key: &found_key, nr: path->slots[0]);
3541
3542	/* make sure the item matches what we want */
3543	if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3544	break;
3545	if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3546	break;
3547
3548	/* release the path since we're done with it */
3549	btrfs_release_path(p: path);
3550
3551	/*
3552	* this is where we are basically btrfs_lookup, without the
3553	* crossing root thing. we store the inode number in the
3554	* offset of the orphan item.
3555	*/
3556
3557	if (found_key.offset == last_objectid) {
3558	/*
3559	* We found the same inode as before. This means we were
3560	* not able to remove its items via eviction triggered
3561	* by an iput(). A transaction abort may have happened,
3562	* due to -ENOSPC for example, so try to grab the error
3563	* that lead to a transaction abort, if any.
3564	*/
3565	btrfs_err(fs_info,
3566	"Error removing orphan entry, stopping orphan cleanup");
3567	ret = BTRFS_FS_ERROR(fs_info) ?: -EINVAL;
3568	goto out;
3569	}
3570
3571	last_objectid = found_key.offset;
3572
3573	found_key.objectid = found_key.offset;
3574	found_key.type = BTRFS_INODE_ITEM_KEY;
3575	found_key.offset = 0;
3576	inode = btrfs_iget(s: fs_info->sb, ino: last_objectid, root);
3577	if (IS_ERR(ptr: inode)) {
3578	ret = PTR_ERR(ptr: inode);
3579	inode = NULL;
3580	if (ret != -ENOENT)
3581	goto out;
3582	}
3583
3584	if (!inode && root == fs_info->tree_root) {
3585	struct btrfs_root *dead_root;
3586	int is_dead_root = 0;
3587
3588	/*
3589	* This is an orphan in the tree root. Currently these
3590	* could come from 2 sources:
3591	* a) a root (snapshot/subvolume) deletion in progress
3592	* b) a free space cache inode
3593	* We need to distinguish those two, as the orphan item
3594	* for a root must not get deleted before the deletion
3595	* of the snapshot/subvolume's tree completes.
3596	*
3597	* btrfs_find_orphan_roots() ran before us, which has
3598	* found all deleted roots and loaded them into
3599	* fs_info->fs_roots_radix. So here we can find if an
3600	* orphan item corresponds to a deleted root by looking
3601	* up the root from that radix tree.
3602	*/
3603
3604	spin_lock(lock: &fs_info->fs_roots_radix_lock);
3605	dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3606	(unsigned long)found_key.objectid);
3607	if (dead_root && btrfs_root_refs(s: &dead_root->root_item) == 0)
3608	is_dead_root = 1;
3609	spin_unlock(lock: &fs_info->fs_roots_radix_lock);
3610
3611	if (is_dead_root) {
3612	/* prevent this orphan from being found again */
3613	key.offset = found_key.objectid - 1;
3614	continue;
3615	}
3616
3617	}
3618
3619	/*
3620	* If we have an inode with links, there are a couple of
3621	* possibilities:
3622	*
3623	* 1. We were halfway through creating fsverity metadata for the
3624	* file. In that case, the orphan item represents incomplete
3625	* fsverity metadata which must be cleaned up with
3626	* btrfs_drop_verity_items and deleting the orphan item.
3627
3628	* 2. Old kernels (before v3.12) used to create an
3629	* orphan item for truncate indicating that there were possibly
3630	* extent items past i_size that needed to be deleted. In v3.12,
3631	* truncate was changed to update i_size in sync with the extent
3632	* items, but the (useless) orphan item was still created. Since
3633	* v4.18, we don't create the orphan item for truncate at all.
3634	*
3635	* So, this item could mean that we need to do a truncate, but
3636	* only if this filesystem was last used on a pre-v3.12 kernel
3637	* and was not cleanly unmounted. The odds of that are quite
3638	* slim, and it's a pain to do the truncate now, so just delete
3639	* the orphan item.
3640	*
3641	* It's also possible that this orphan item was supposed to be
3642	* deleted but wasn't. The inode number may have been reused,
3643	* but either way, we can delete the orphan item.
3644	*/
3645	if (!inode || inode->i_nlink) {
3646	if (inode) {
3647	ret = btrfs_drop_verity_items(inode: BTRFS_I(inode));
3648	iput(inode);
3649	inode = NULL;
3650	if (ret)
3651	goto out;
3652	}
3653	trans = btrfs_start_transaction(root, num_items: 1);
3654	if (IS_ERR(ptr: trans)) {
3655	ret = PTR_ERR(ptr: trans);
3656	goto out;
3657	}
3658	btrfs_debug(fs_info, "auto deleting %Lu",
3659	found_key.objectid);
3660	ret = btrfs_del_orphan_item(trans, root,
3661	offset: found_key.objectid);
3662	btrfs_end_transaction(trans);
3663	if (ret)
3664	goto out;
3665	continue;
3666	}
3667
3668	nr_unlink++;
3669
3670	/* this will do delete_inode and everything for us */
3671	iput(inode);
3672	}
3673	/* release the path since we're done with it */
3674	btrfs_release_path(p: path);
3675
3676	if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3677	trans = btrfs_join_transaction(root);
3678	if (!IS_ERR(ptr: trans))
3679	btrfs_end_transaction(trans);
3680	}
3681
3682	if (nr_unlink)
3683	btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3684
3685	out:
3686	if (ret)
3687	btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3688	btrfs_free_path(p: path);
3689	return ret;
3690	}
3691
3692	/*
3693	* very simple check to peek ahead in the leaf looking for xattrs. If we
3694	* don't find any xattrs, we know there can't be any acls.
3695	*
3696	* slot is the slot the inode is in, objectid is the objectid of the inode
3697	*/
3698	static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3699	int slot, u64 objectid,
3700	int *first_xattr_slot)
3701	{
3702	u32 nritems = btrfs_header_nritems(eb: leaf);
3703	struct btrfs_key found_key;
3704	static u64 xattr_access = 0;
3705	static u64 xattr_default = 0;
3706	int scanned = 0;
3707
3708	if (!xattr_access) {
3709	xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3710	strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3711	xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3712	strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3713	}
3714
3715	slot++;
3716	*first_xattr_slot = -1;
3717	while (slot < nritems) {
3718	btrfs_item_key_to_cpu(eb: leaf, cpu_key: &found_key, nr: slot);
3719
3720	/* we found a different objectid, there must not be acls */
3721	if (found_key.objectid != objectid)
3722	return 0;
3723
3724	/* we found an xattr, assume we've got an acl */
3725	if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3726	if (*first_xattr_slot == -1)
3727	*first_xattr_slot = slot;
3728	if (found_key.offset == xattr_access ||
3729	found_key.offset == xattr_default)
3730	return 1;
3731	}
3732
3733	/*
3734	* we found a key greater than an xattr key, there can't
3735	* be any acls later on
3736	*/
3737	if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3738	return 0;
3739
3740	slot++;
3741	scanned++;
3742
3743	/*
3744	* it goes inode, inode backrefs, xattrs, extents,
3745	* so if there are a ton of hard links to an inode there can
3746	* be a lot of backrefs. Don't waste time searching too hard,
3747	* this is just an optimization
3748	*/
3749	if (scanned >= 8)
3750	break;
3751	}
3752	/* we hit the end of the leaf before we found an xattr or
3753	* something larger than an xattr. We have to assume the inode
3754	* has acls
3755	*/
3756	if (*first_xattr_slot == -1)
3757	*first_xattr_slot = slot;
3758	return 1;
3759	}
3760
3761	/*
3762	* read an inode from the btree into the in-memory inode
3763	*/
3764	static int btrfs_read_locked_inode(struct inode *inode,
3765	struct btrfs_path *in_path)
3766	{
3767	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
3768	struct btrfs_path *path = in_path;
3769	struct extent_buffer *leaf;
3770	struct btrfs_inode_item *inode_item;
3771	struct btrfs_root *root = BTRFS_I(inode)->root;
3772	struct btrfs_key location;
3773	unsigned long ptr;
3774	int maybe_acls;
3775	u32 rdev;
3776	int ret;
3777	bool filled = false;
3778	int first_xattr_slot;
3779
3780	ret = btrfs_fill_inode(inode, rdev: &rdev);
3781	if (!ret)
3782	filled = true;
3783
3784	if (!path) {
3785	path = btrfs_alloc_path();
3786	if (!path)
3787	return -ENOMEM;
3788	}
3789
3790	memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3791
3792	ret = btrfs_lookup_inode(NULL, root, path, location: &location, mod: 0);
3793	if (ret) {
3794	if (path != in_path)
3795	btrfs_free_path(p: path);
3796	return ret;
3797	}
3798
3799	leaf = path->nodes[0];
3800
3801	if (filled)
3802	goto cache_index;
3803
3804	inode_item = btrfs_item_ptr(leaf, path->slots[0],
3805	struct btrfs_inode_item);
3806	inode->i_mode = btrfs_inode_mode(eb: leaf, s: inode_item);
3807	set_nlink(inode, nlink: btrfs_inode_nlink(eb: leaf, s: inode_item));
3808	i_uid_write(inode, uid: btrfs_inode_uid(eb: leaf, s: inode_item));
3809	i_gid_write(inode, gid: btrfs_inode_gid(eb: leaf, s: inode_item));
3810	btrfs_i_size_write(inode: BTRFS_I(inode), size: btrfs_inode_size(eb: leaf, s: inode_item));
3811	btrfs_inode_set_file_extent_range(inode: BTRFS_I(inode), start: 0,
3812	round_up(i_size_read(inode), fs_info->sectorsize));
3813
3814	inode_set_atime(inode, sec: btrfs_timespec_sec(eb: leaf, s: &inode_item->atime),
3815	nsec: btrfs_timespec_nsec(eb: leaf, s: &inode_item->atime));
3816
3817	inode_set_mtime(inode, sec: btrfs_timespec_sec(eb: leaf, s: &inode_item->mtime),
3818	nsec: btrfs_timespec_nsec(eb: leaf, s: &inode_item->mtime));
3819
3820	inode_set_ctime(inode, sec: btrfs_timespec_sec(eb: leaf, s: &inode_item->ctime),
3821	nsec: btrfs_timespec_nsec(eb: leaf, s: &inode_item->ctime));
3822
3823	BTRFS_I(inode)->i_otime_sec = btrfs_timespec_sec(eb: leaf, s: &inode_item->otime);
3824	BTRFS_I(inode)->i_otime_nsec = btrfs_timespec_nsec(eb: leaf, s: &inode_item->otime);
3825
3826	inode_set_bytes(inode, bytes: btrfs_inode_nbytes(eb: leaf, s: inode_item));
3827	BTRFS_I(inode)->generation = btrfs_inode_generation(eb: leaf, s: inode_item);
3828	BTRFS_I(inode)->last_trans = btrfs_inode_transid(eb: leaf, s: inode_item);
3829
3830	inode_set_iversion_queried(inode,
3831	val: btrfs_inode_sequence(eb: leaf, s: inode_item));
3832	inode->i_generation = BTRFS_I(inode)->generation;
3833	inode->i_rdev = 0;
3834	rdev = btrfs_inode_rdev(eb: leaf, s: inode_item);
3835
3836	BTRFS_I(inode)->index_cnt = (u64)-1;
3837	btrfs_inode_split_flags(inode_item_flags: btrfs_inode_flags(eb: leaf, s: inode_item),
3838	flags: &BTRFS_I(inode)->flags, ro_flags: &BTRFS_I(inode)->ro_flags);
3839
3840	cache_index:
3841	/*
3842	* If we were modified in the current generation and evicted from memory
3843	* and then re-read we need to do a full sync since we don't have any
3844	* idea about which extents were modified before we were evicted from
3845	* cache.
3846	*
3847	* This is required for both inode re-read from disk and delayed inode
3848	* in the delayed_nodes xarray.
3849	*/
3850	if (BTRFS_I(inode)->last_trans == btrfs_get_fs_generation(fs_info))
3851	set_bit(nr: BTRFS_INODE_NEEDS_FULL_SYNC,
3852	addr: &BTRFS_I(inode)->runtime_flags);
3853
3854	/*
3855	* We don't persist the id of the transaction where an unlink operation
3856	* against the inode was last made. So here we assume the inode might
3857	* have been evicted, and therefore the exact value of last_unlink_trans
3858	* lost, and set it to last_trans to avoid metadata inconsistencies
3859	* between the inode and its parent if the inode is fsync'ed and the log
3860	* replayed. For example, in the scenario:
3861	*
3862	* touch mydir/foo
3863	* ln mydir/foo mydir/bar
3864	* sync
3865	* unlink mydir/bar
3866	* echo 2 > /proc/sys/vm/drop_caches # evicts inode
3867	* xfs_io -c fsync mydir/foo
3868	* <power failure>
3869	* mount fs, triggers fsync log replay
3870	*
3871	* We must make sure that when we fsync our inode foo we also log its
3872	* parent inode, otherwise after log replay the parent still has the
3873	* dentry with the "bar" name but our inode foo has a link count of 1
3874	* and doesn't have an inode ref with the name "bar" anymore.
3875	*
3876	* Setting last_unlink_trans to last_trans is a pessimistic approach,
3877	* but it guarantees correctness at the expense of occasional full
3878	* transaction commits on fsync if our inode is a directory, or if our
3879	* inode is not a directory, logging its parent unnecessarily.
3880	*/
3881	BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3882
3883	/*
3884	* Same logic as for last_unlink_trans. We don't persist the generation
3885	* of the last transaction where this inode was used for a reflink
3886	* operation, so after eviction and reloading the inode we must be
3887	* pessimistic and assume the last transaction that modified the inode.
3888	*/
3889	BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3890
3891	path->slots[0]++;
3892	if (inode->i_nlink != 1 ||
3893	path->slots[0] >= btrfs_header_nritems(eb: leaf))
3894	goto cache_acl;
3895
3896	btrfs_item_key_to_cpu(eb: leaf, cpu_key: &location, nr: path->slots[0]);
3897	if (location.objectid != btrfs_ino(inode: BTRFS_I(inode)))
3898	goto cache_acl;
3899
3900	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3901	if (location.type == BTRFS_INODE_REF_KEY) {
3902	struct btrfs_inode_ref *ref;
3903
3904	ref = (struct btrfs_inode_ref *)ptr;
3905	BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(eb: leaf, s: ref);
3906	} else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3907	struct btrfs_inode_extref *extref;
3908
3909	extref = (struct btrfs_inode_extref *)ptr;
3910	BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(eb: leaf,
3911	s: extref);
3912	}
3913	cache_acl:
3914	/*
3915	* try to precache a NULL acl entry for files that don't have
3916	* any xattrs or acls
3917	*/
3918	maybe_acls = acls_after_inode_item(leaf, slot: path->slots[0],
3919	objectid: btrfs_ino(inode: BTRFS_I(inode)), first_xattr_slot: &first_xattr_slot);
3920	if (first_xattr_slot != -1) {
3921	path->slots[0] = first_xattr_slot;
3922	ret = btrfs_load_inode_props(inode, path);
3923	if (ret)
3924	btrfs_err(fs_info,
3925	"error loading props for ino %llu (root %llu): %d",
3926	btrfs_ino(BTRFS_I(inode)),
3927	root->root_key.objectid, ret);
3928	}
3929	if (path != in_path)
3930	btrfs_free_path(p: path);
3931
3932	if (!maybe_acls)
3933	cache_no_acl(inode);
3934
3935	switch (inode->i_mode & S_IFMT) {
3936	case S_IFREG:
3937	inode->i_mapping->a_ops = &btrfs_aops;
3938	inode->i_fop = &btrfs_file_operations;
3939	inode->i_op = &btrfs_file_inode_operations;
3940	break;
3941	case S_IFDIR:
3942	inode->i_fop = &btrfs_dir_file_operations;
3943	inode->i_op = &btrfs_dir_inode_operations;
3944	break;
3945	case S_IFLNK:
3946	inode->i_op = &btrfs_symlink_inode_operations;
3947	inode_nohighmem(inode);
3948	inode->i_mapping->a_ops = &btrfs_aops;
3949	break;
3950	default:
3951	inode->i_op = &btrfs_special_inode_operations;
3952	init_special_inode(inode, inode->i_mode, rdev);
3953	break;
3954	}
3955
3956	btrfs_sync_inode_flags_to_i_flags(inode);
3957	return 0;
3958	}
3959
3960	/*
3961	* given a leaf and an inode, copy the inode fields into the leaf
3962	*/
3963	static void fill_inode_item(struct btrfs_trans_handle *trans,
3964	struct extent_buffer *leaf,
3965	struct btrfs_inode_item *item,
3966	struct inode *inode)
3967	{
3968	struct btrfs_map_token token;
3969	u64 flags;
3970
3971	btrfs_init_map_token(token: &token, eb: leaf);
3972
3973	btrfs_set_token_inode_uid(token: &token, s: item, val: i_uid_read(inode));
3974	btrfs_set_token_inode_gid(token: &token, s: item, val: i_gid_read(inode));
3975	btrfs_set_token_inode_size(token: &token, s: item, val: BTRFS_I(inode)->disk_i_size);
3976	btrfs_set_token_inode_mode(token: &token, s: item, val: inode->i_mode);
3977	btrfs_set_token_inode_nlink(token: &token, s: item, val: inode->i_nlink);
3978
3979	btrfs_set_token_timespec_sec(token: &token, s: &item->atime,
3980	val: inode_get_atime_sec(inode));
3981	btrfs_set_token_timespec_nsec(token: &token, s: &item->atime,
3982	val: inode_get_atime_nsec(inode));
3983
3984	btrfs_set_token_timespec_sec(token: &token, s: &item->mtime,
3985	val: inode_get_mtime_sec(inode));
3986	btrfs_set_token_timespec_nsec(token: &token, s: &item->mtime,
3987	val: inode_get_mtime_nsec(inode));
3988
3989	btrfs_set_token_timespec_sec(token: &token, s: &item->ctime,
3990	val: inode_get_ctime_sec(inode));
3991	btrfs_set_token_timespec_nsec(token: &token, s: &item->ctime,
3992	val: inode_get_ctime_nsec(inode));
3993
3994	btrfs_set_token_timespec_sec(token: &token, s: &item->otime, val: BTRFS_I(inode)->i_otime_sec);
3995	btrfs_set_token_timespec_nsec(token: &token, s: &item->otime, val: BTRFS_I(inode)->i_otime_nsec);
3996
3997	btrfs_set_token_inode_nbytes(token: &token, s: item, val: inode_get_bytes(inode));
3998	btrfs_set_token_inode_generation(token: &token, s: item,
3999	val: BTRFS_I(inode)->generation);
4000	btrfs_set_token_inode_sequence(token: &token, s: item, val: inode_peek_iversion(inode));
4001	btrfs_set_token_inode_transid(token: &token, s: item, val: trans->transid);
4002	btrfs_set_token_inode_rdev(token: &token, s: item, val: inode->i_rdev);
4003	flags = btrfs_inode_combine_flags(flags: BTRFS_I(inode)->flags,
4004	ro_flags: BTRFS_I(inode)->ro_flags);
4005	btrfs_set_token_inode_flags(token: &token, s: item, val: flags);
4006	btrfs_set_token_inode_block_group(token: &token, s: item, val: 0);
4007	}
4008
4009	/*
4010	* copy everything in the in-memory inode into the btree.
4011	*/
4012	static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4013	struct btrfs_inode *inode)
4014	{
4015	struct btrfs_inode_item *inode_item;
4016	struct btrfs_path *path;
4017	struct extent_buffer *leaf;
4018	int ret;
4019
4020	path = btrfs_alloc_path();
4021	if (!path)
4022	return -ENOMEM;
4023
4024	ret = btrfs_lookup_inode(trans, root: inode->root, path, location: &inode->location, mod: 1);
4025	if (ret) {
4026	if (ret > 0)
4027	ret = -ENOENT;
4028	goto failed;
4029	}
4030
4031	leaf = path->nodes[0];
4032	inode_item = btrfs_item_ptr(leaf, path->slots[0],
4033	struct btrfs_inode_item);
4034
4035	fill_inode_item(trans, leaf, item: inode_item, inode: &inode->vfs_inode);
4036	btrfs_mark_buffer_dirty(trans, buf: leaf);
4037	btrfs_set_inode_last_trans(trans, inode);
4038	ret = 0;
4039	failed:
4040	btrfs_free_path(p: path);
4041	return ret;
4042	}
4043
4044	/*
4045	* copy everything in the in-memory inode into the btree.
4046	*/
4047	int btrfs_update_inode(struct btrfs_trans_handle *trans,
4048	struct btrfs_inode *inode)
4049	{
4050	struct btrfs_root *root = inode->root;
4051	struct btrfs_fs_info *fs_info = root->fs_info;
4052	int ret;
4053
4054	/*
4055	* If the inode is a free space inode, we can deadlock during commit
4056	* if we put it into the delayed code.
4057	*
4058	* The data relocation inode should also be directly updated
4059	* without delay
4060	*/
4061	if (!btrfs_is_free_space_inode(inode)
4062	&& !btrfs_is_data_reloc_root(root)
4063	&& !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4064	btrfs_update_root_times(trans, root);
4065
4066	ret = btrfs_delayed_update_inode(trans, inode);
4067	if (!ret)
4068	btrfs_set_inode_last_trans(trans, inode);
4069	return ret;
4070	}
4071
4072	return btrfs_update_inode_item(trans, inode);
4073	}
4074
4075	int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4076	struct btrfs_inode *inode)
4077	{
4078	int ret;
4079
4080	ret = btrfs_update_inode(trans, inode);
4081	if (ret == -ENOSPC)
4082	return btrfs_update_inode_item(trans, inode);
4083	return ret;
4084	}
4085
4086	/*
4087	* unlink helper that gets used here in inode.c and in the tree logging
4088	* recovery code. It remove a link in a directory with a given name, and
4089	* also drops the back refs in the inode to the directory
4090	*/
4091	static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4092	struct btrfs_inode *dir,
4093	struct btrfs_inode *inode,
4094	const struct fscrypt_str *name,
4095	struct btrfs_rename_ctx *rename_ctx)
4096	{
4097	struct btrfs_root *root = dir->root;
4098	struct btrfs_fs_info *fs_info = root->fs_info;
4099	struct btrfs_path *path;
4100	int ret = 0;
4101	struct btrfs_dir_item *di;
4102	u64 index;
4103	u64 ino = btrfs_ino(inode);
4104	u64 dir_ino = btrfs_ino(inode: dir);
4105
4106	path = btrfs_alloc_path();
4107	if (!path) {
4108	ret = -ENOMEM;
4109	goto out;
4110	}
4111
4112	di = btrfs_lookup_dir_item(trans, root, path, dir: dir_ino, name, mod: -1);
4113	if (IS_ERR_OR_NULL(ptr: di)) {
4114	ret = di ? PTR_ERR(ptr: di) : -ENOENT;
4115	goto err;
4116	}
4117	ret = btrfs_delete_one_dir_name(trans, root, path, di);
4118	if (ret)
4119	goto err;
4120	btrfs_release_path(p: path);
4121
4122	/*
4123	* If we don't have dir index, we have to get it by looking up
4124	* the inode ref, since we get the inode ref, remove it directly,
4125	* it is unnecessary to do delayed deletion.
4126	*
4127	* But if we have dir index, needn't search inode ref to get it.
4128	* Since the inode ref is close to the inode item, it is better
4129	* that we delay to delete it, and just do this deletion when
4130	* we update the inode item.
4131	*/
4132	if (inode->dir_index) {
4133	ret = btrfs_delayed_delete_inode_ref(inode);
4134	if (!ret) {
4135	index = inode->dir_index;
4136	goto skip_backref;
4137	}
4138	}
4139
4140	ret = btrfs_del_inode_ref(trans, root, name, inode_objectid: ino, ref_objectid: dir_ino, index: &index);
4141	if (ret) {
4142	btrfs_info(fs_info,
4143	"failed to delete reference to %.*s, inode %llu parent %llu",
4144	name->len, name->name, ino, dir_ino);
4145	btrfs_abort_transaction(trans, ret);
4146	goto err;
4147	}
4148	skip_backref:
4149	if (rename_ctx)
4150	rename_ctx->index = index;
4151
4152	ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4153	if (ret) {
4154	btrfs_abort_transaction(trans, ret);
4155	goto err;
4156	}
4157
4158	/*
4159	* If we are in a rename context, we don't need to update anything in the
4160	* log. That will be done later during the rename by btrfs_log_new_name().
4161	* Besides that, doing it here would only cause extra unnecessary btree
4162	* operations on the log tree, increasing latency for applications.
4163	*/
4164	if (!rename_ctx) {
4165	btrfs_del_inode_ref_in_log(trans, root, name, inode, dirid: dir_ino);
4166	btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4167	}
4168
4169	/*
4170	* If we have a pending delayed iput we could end up with the final iput
4171	* being run in btrfs-cleaner context. If we have enough of these built
4172	* up we can end up burning a lot of time in btrfs-cleaner without any
4173	* way to throttle the unlinks. Since we're currently holding a ref on
4174	* the inode we can run the delayed iput here without any issues as the
4175	* final iput won't be done until after we drop the ref we're currently
4176	* holding.
4177	*/
4178	btrfs_run_delayed_iput(fs_info, inode);
4179	err:
4180	btrfs_free_path(p: path);
4181	if (ret)
4182	goto out;
4183
4184	btrfs_i_size_write(inode: dir, size: dir->vfs_inode.i_size - name->len * 2);
4185	inode_inc_iversion(inode: &inode->vfs_inode);
4186	inode_inc_iversion(inode: &dir->vfs_inode);
4187	inode_set_mtime_to_ts(inode: &dir->vfs_inode, ts: inode_set_ctime_current(inode: &dir->vfs_inode));
4188	ret = btrfs_update_inode(trans, inode: dir);
4189	out:
4190	return ret;
4191	}
4192
4193	int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4194	struct btrfs_inode *dir, struct btrfs_inode *inode,
4195	const struct fscrypt_str *name)
4196	{
4197	int ret;
4198
4199	ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4200	if (!ret) {
4201	drop_nlink(inode: &inode->vfs_inode);
4202	ret = btrfs_update_inode(trans, inode);
4203	}
4204	return ret;
4205	}
4206
4207	/*
4208	* helper to start transaction for unlink and rmdir.
4209	*
4210	* unlink and rmdir are special in btrfs, they do not always free space, so
4211	* if we cannot make our reservations the normal way try and see if there is
4212	* plenty of slack room in the global reserve to migrate, otherwise we cannot
4213	* allow the unlink to occur.
4214	*/
4215	static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4216	{
4217	struct btrfs_root *root = dir->root;
4218
4219	return btrfs_start_transaction_fallback_global_rsv(root,
4220	BTRFS_UNLINK_METADATA_UNITS);
4221	}
4222
4223	static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4224	{
4225	struct btrfs_trans_handle *trans;
4226	struct inode *inode = d_inode(dentry);
4227	int ret;
4228	struct fscrypt_name fname;
4229
4230	ret = fscrypt_setup_filename(inode: dir, iname: &dentry->d_name, lookup: 1, fname: &fname);
4231	if (ret)
4232	return ret;
4233
4234	/* This needs to handle no-key deletions later on */
4235
4236	trans = __unlink_start_trans(dir: BTRFS_I(inode: dir));
4237	if (IS_ERR(ptr: trans)) {
4238	ret = PTR_ERR(ptr: trans);
4239	goto fscrypt_free;
4240	}
4241
4242	btrfs_record_unlink_dir(trans, dir: BTRFS_I(inode: dir), inode: BTRFS_I(inode: d_inode(dentry)),
4243	for_rename: false);
4244
4245	ret = btrfs_unlink_inode(trans, dir: BTRFS_I(inode: dir), inode: BTRFS_I(inode: d_inode(dentry)),
4246	name: &fname.disk_name);
4247	if (ret)
4248	goto end_trans;
4249
4250	if (inode->i_nlink == 0) {
4251	ret = btrfs_orphan_add(trans, inode: BTRFS_I(inode));
4252	if (ret)
4253	goto end_trans;
4254	}
4255
4256	end_trans:
4257	btrfs_end_transaction(trans);
4258	btrfs_btree_balance_dirty(fs_info: BTRFS_I(inode: dir)->root->fs_info);
4259	fscrypt_free:
4260	fscrypt_free_filename(fname: &fname);
4261	return ret;
4262	}
4263
4264	static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4265	struct btrfs_inode *dir, struct dentry *dentry)
4266	{
4267	struct btrfs_root *root = dir->root;
4268	struct btrfs_inode *inode = BTRFS_I(inode: d_inode(dentry));
4269	struct btrfs_path *path;
4270	struct extent_buffer *leaf;
4271	struct btrfs_dir_item *di;
4272	struct btrfs_key key;
4273	u64 index;
4274	int ret;
4275	u64 objectid;
4276	u64 dir_ino = btrfs_ino(inode: dir);
4277	struct fscrypt_name fname;
4278
4279	ret = fscrypt_setup_filename(inode: &dir->vfs_inode, iname: &dentry->d_name, lookup: 1, fname: &fname);
4280	if (ret)
4281	return ret;
4282
4283	/* This needs to handle no-key deletions later on */
4284
4285	if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4286	objectid = inode->root->root_key.objectid;
4287	} else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4288	objectid = inode->location.objectid;
4289	} else {
4290	WARN_ON(1);
4291	fscrypt_free_filename(fname: &fname);
4292	return -EINVAL;
4293	}
4294
4295	path = btrfs_alloc_path();
4296	if (!path) {
4297	ret = -ENOMEM;
4298	goto out;
4299	}
4300
4301	di = btrfs_lookup_dir_item(trans, root, path, dir: dir_ino,
4302	name: &fname.disk_name, mod: -1);
4303	if (IS_ERR_OR_NULL(ptr: di)) {
4304	ret = di ? PTR_ERR(ptr: di) : -ENOENT;
4305	goto out;
4306	}
4307
4308	leaf = path->nodes[0];
4309	btrfs_dir_item_key_to_cpu(eb: leaf, item: di, cpu_key: &key);
4310	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4311	ret = btrfs_delete_one_dir_name(trans, root, path, di);
4312	if (ret) {
4313	btrfs_abort_transaction(trans, ret);
4314	goto out;
4315	}
4316	btrfs_release_path(p: path);
4317
4318	/*
4319	* This is a placeholder inode for a subvolume we didn't have a
4320	* reference to at the time of the snapshot creation. In the meantime
4321	* we could have renamed the real subvol link into our snapshot, so
4322	* depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4323	* Instead simply lookup the dir_index_item for this entry so we can
4324	* remove it. Otherwise we know we have a ref to the root and we can
4325	* call btrfs_del_root_ref, and it _shouldn't_ fail.
4326	*/
4327	if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4328	di = btrfs_search_dir_index_item(root, path, dirid: dir_ino, name: &fname.disk_name);
4329	if (IS_ERR_OR_NULL(ptr: di)) {
4330	if (!di)
4331	ret = -ENOENT;
4332	else
4333	ret = PTR_ERR(ptr: di);
4334	btrfs_abort_transaction(trans, ret);
4335	goto out;
4336	}
4337
4338	leaf = path->nodes[0];
4339	btrfs_item_key_to_cpu(eb: leaf, cpu_key: &key, nr: path->slots[0]);
4340	index = key.offset;
4341	btrfs_release_path(p: path);
4342	} else {
4343	ret = btrfs_del_root_ref(trans, root_id: objectid,
4344	ref_id: root->root_key.objectid, dirid: dir_ino,
4345	sequence: &index, name: &fname.disk_name);
4346	if (ret) {
4347	btrfs_abort_transaction(trans, ret);
4348	goto out;
4349	}
4350	}
4351
4352	ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4353	if (ret) {
4354	btrfs_abort_transaction(trans, ret);
4355	goto out;
4356	}
4357
4358	btrfs_i_size_write(inode: dir, size: dir->vfs_inode.i_size - fname.disk_name.len * 2);
4359	inode_inc_iversion(inode: &dir->vfs_inode);
4360	inode_set_mtime_to_ts(inode: &dir->vfs_inode, ts: inode_set_ctime_current(inode: &dir->vfs_inode));
4361	ret = btrfs_update_inode_fallback(trans, inode: dir);
4362	if (ret)
4363	btrfs_abort_transaction(trans, ret);
4364	out:
4365	btrfs_free_path(p: path);
4366	fscrypt_free_filename(fname: &fname);
4367	return ret;
4368	}
4369
4370	/*
4371	* Helper to check if the subvolume references other subvolumes or if it's
4372	* default.
4373	*/
4374	static noinline int may_destroy_subvol(struct btrfs_root *root)
4375	{
4376	struct btrfs_fs_info *fs_info = root->fs_info;
4377	struct btrfs_path *path;
4378	struct btrfs_dir_item *di;
4379	struct btrfs_key key;
4380	struct fscrypt_str name = FSTR_INIT("default", 7);
4381	u64 dir_id;
4382	int ret;
4383
4384	path = btrfs_alloc_path();
4385	if (!path)
4386	return -ENOMEM;
4387
4388	/* Make sure this root isn't set as the default subvol */
4389	dir_id = btrfs_super_root_dir(s: fs_info->super_copy);
4390	di = btrfs_lookup_dir_item(NULL, root: fs_info->tree_root, path,
4391	dir: dir_id, name: &name, mod: 0);
4392	if (di && !IS_ERR(ptr: di)) {
4393	btrfs_dir_item_key_to_cpu(eb: path->nodes[0], item: di, cpu_key: &key);
4394	if (key.objectid == root->root_key.objectid) {
4395	ret = -EPERM;
4396	btrfs_err(fs_info,
4397	"deleting default subvolume %llu is not allowed",
4398	key.objectid);
4399	goto out;
4400	}
4401	btrfs_release_path(p: path);
4402	}
4403
4404	key.objectid = root->root_key.objectid;
4405	key.type = BTRFS_ROOT_REF_KEY;
4406	key.offset = (u64)-1;
4407
4408	ret = btrfs_search_slot(NULL, root: fs_info->tree_root, key: &key, p: path, ins_len: 0, cow: 0);
4409	if (ret < 0)
4410	goto out;
4411	if (ret == 0) {
4412	/*
4413	* Key with offset -1 found, there would have to exist a root
4414	* with such id, but this is out of valid range.
4415	*/
4416	ret = -EUCLEAN;
4417	goto out;
4418	}
4419
4420	ret = 0;
4421	if (path->slots[0] > 0) {
4422	path->slots[0]--;
4423	btrfs_item_key_to_cpu(eb: path->nodes[0], cpu_key: &key, nr: path->slots[0]);
4424	if (key.objectid == root->root_key.objectid &&
4425	key.type == BTRFS_ROOT_REF_KEY)
4426	ret = -ENOTEMPTY;
4427	}
4428	out:
4429	btrfs_free_path(p: path);
4430	return ret;
4431	}
4432
4433	/* Delete all dentries for inodes belonging to the root */
4434	static void btrfs_prune_dentries(struct btrfs_root *root)
4435	{
4436	struct btrfs_fs_info *fs_info = root->fs_info;
4437	struct rb_node *node;
4438	struct rb_node *prev;
4439	struct btrfs_inode *entry;
4440	struct inode *inode;
4441	u64 objectid = 0;
4442
4443	if (!BTRFS_FS_ERROR(fs_info))
4444	WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4445
4446	spin_lock(lock: &root->inode_lock);
4447	again:
4448	node = root->inode_tree.rb_node;
4449	prev = NULL;
4450	while (node) {
4451	prev = node;
4452	entry = rb_entry(node, struct btrfs_inode, rb_node);
4453
4454	if (objectid < btrfs_ino(inode: entry))
4455	node = node->rb_left;
4456	else if (objectid > btrfs_ino(inode: entry))
4457	node = node->rb_right;
4458	else
4459	break;
4460	}
4461	if (!node) {
4462	while (prev) {
4463	entry = rb_entry(prev, struct btrfs_inode, rb_node);
4464	if (objectid <= btrfs_ino(inode: entry)) {
4465	node = prev;
4466	break;
4467	}
4468	prev = rb_next(prev);
4469	}
4470	}
4471	while (node) {
4472	entry = rb_entry(node, struct btrfs_inode, rb_node);
4473	objectid = btrfs_ino(inode: entry) + 1;
4474	inode = igrab(&entry->vfs_inode);
4475	if (inode) {
4476	spin_unlock(lock: &root->inode_lock);
4477	if (atomic_read(v: &inode->i_count) > 1)
4478	d_prune_aliases(inode);
4479	/*
4480	* btrfs_drop_inode will have it removed from the inode
4481	* cache when its usage count hits zero.
4482	*/
4483	iput(inode);
4484	cond_resched();
4485	spin_lock(lock: &root->inode_lock);
4486	goto again;
4487	}
4488
4489	if (cond_resched_lock(&root->inode_lock))
4490	goto again;
4491
4492	node = rb_next(node);
4493	}
4494	spin_unlock(lock: &root->inode_lock);
4495	}
4496
4497	int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4498	{
4499	struct btrfs_root *root = dir->root;
4500	struct btrfs_fs_info *fs_info = root->fs_info;
4501	struct inode *inode = d_inode(dentry);
4502	struct btrfs_root *dest = BTRFS_I(inode)->root;
4503	struct btrfs_trans_handle *trans;
4504	struct btrfs_block_rsv block_rsv;
4505	u64 root_flags;
4506	u64 qgroup_reserved = 0;
4507	int ret;
4508
4509	down_write(sem: &fs_info->subvol_sem);
4510
4511	/*
4512	* Don't allow to delete a subvolume with send in progress. This is
4513	* inside the inode lock so the error handling that has to drop the bit
4514	* again is not run concurrently.
4515	*/
4516	spin_lock(lock: &dest->root_item_lock);
4517	if (dest->send_in_progress) {
4518	spin_unlock(lock: &dest->root_item_lock);
4519	btrfs_warn(fs_info,
4520	"attempt to delete subvolume %llu during send",
4521	dest->root_key.objectid);
4522	ret = -EPERM;
4523	goto out_up_write;
4524	}
4525	if (atomic_read(v: &dest->nr_swapfiles)) {
4526	spin_unlock(lock: &dest->root_item_lock);
4527	btrfs_warn(fs_info,
4528	"attempt to delete subvolume %llu with active swapfile",
4529	root->root_key.objectid);
4530	ret = -EPERM;
4531	goto out_up_write;
4532	}
4533	root_flags = btrfs_root_flags(s: &dest->root_item);
4534	btrfs_set_root_flags(s: &dest->root_item,
4535	val: root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4536	spin_unlock(lock: &dest->root_item_lock);
4537
4538	ret = may_destroy_subvol(root: dest);
4539	if (ret)
4540	goto out_undead;
4541
4542	btrfs_init_block_rsv(rsv: &block_rsv, type: BTRFS_BLOCK_RSV_TEMP);
4543	/*
4544	* One for dir inode,
4545	* two for dir entries,
4546	* two for root ref/backref.
4547	*/
4548	ret = btrfs_subvolume_reserve_metadata(root, rsv: &block_rsv, nitems: 5, use_global_rsv: true);
4549	if (ret)
4550	goto out_undead;
4551	qgroup_reserved = block_rsv.qgroup_rsv_reserved;
4552
4553	trans = btrfs_start_transaction(root, num_items: 0);
4554	if (IS_ERR(ptr: trans)) {
4555	ret = PTR_ERR(ptr: trans);
4556	goto out_release;
4557	}
4558	ret = btrfs_record_root_in_trans(trans, root);
4559	if (ret) {
4560	btrfs_abort_transaction(trans, ret);
4561	goto out_end_trans;
4562	}
4563	btrfs_qgroup_convert_reserved_meta(root, num_bytes: qgroup_reserved);
4564	qgroup_reserved = 0;
4565	trans->block_rsv = &block_rsv;
4566	trans->bytes_reserved = block_rsv.size;
4567
4568	btrfs_record_snapshot_destroy(trans, dir);
4569
4570	ret = btrfs_unlink_subvol(trans, dir, dentry);
4571	if (ret) {
4572	btrfs_abort_transaction(trans, ret);
4573	goto out_end_trans;
4574	}
4575
4576	ret = btrfs_record_root_in_trans(trans, root: dest);
4577	if (ret) {
4578	btrfs_abort_transaction(trans, ret);
4579	goto out_end_trans;
4580	}
4581
4582	memset(&dest->root_item.drop_progress, 0,
4583	sizeof(dest->root_item.drop_progress));
4584	btrfs_set_root_drop_level(s: &dest->root_item, val: 0);
4585	btrfs_set_root_refs(s: &dest->root_item, val: 0);
4586
4587	if (!test_and_set_bit(nr: BTRFS_ROOT_ORPHAN_ITEM_INSERTED, addr: &dest->state)) {
4588	ret = btrfs_insert_orphan_item(trans,
4589	root: fs_info->tree_root,
4590	offset: dest->root_key.objectid);
4591	if (ret) {
4592	btrfs_abort_transaction(trans, ret);
4593	goto out_end_trans;
4594	}
4595	}
4596
4597	ret = btrfs_uuid_tree_remove(trans, uuid: dest->root_item.uuid,
4598	BTRFS_UUID_KEY_SUBVOL,
4599	subid: dest->root_key.objectid);
4600	if (ret && ret != -ENOENT) {
4601	btrfs_abort_transaction(trans, ret);
4602	goto out_end_trans;
4603	}
4604	if (!btrfs_is_empty_uuid(uuid: dest->root_item.received_uuid)) {
4605	ret = btrfs_uuid_tree_remove(trans,
4606	uuid: dest->root_item.received_uuid,
4607	BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4608	subid: dest->root_key.objectid);
4609	if (ret && ret != -ENOENT) {
4610	btrfs_abort_transaction(trans, ret);
4611	goto out_end_trans;
4612	}
4613	}
4614
4615	free_anon_bdev(dest->anon_dev);
4616	dest->anon_dev = 0;
4617	out_end_trans:
4618	trans->block_rsv = NULL;
4619	trans->bytes_reserved = 0;
4620	ret = btrfs_end_transaction(trans);
4621	inode->i_flags |= S_DEAD;
4622	out_release:
4623	btrfs_block_rsv_release(fs_info, block_rsv: &block_rsv, num_bytes: (u64)-1, NULL);
4624	if (qgroup_reserved)
4625	btrfs_qgroup_free_meta_prealloc(root, num_bytes: qgroup_reserved);
4626	out_undead:
4627	if (ret) {
4628	spin_lock(lock: &dest->root_item_lock);
4629	root_flags = btrfs_root_flags(s: &dest->root_item);
4630	btrfs_set_root_flags(s: &dest->root_item,
4631	val: root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4632	spin_unlock(lock: &dest->root_item_lock);
4633	}
4634	out_up_write:
4635	up_write(sem: &fs_info->subvol_sem);
4636	if (!ret) {
4637	d_invalidate(dentry);
4638	btrfs_prune_dentries(root: dest);
4639	ASSERT(dest->send_in_progress == 0);
4640	}
4641
4642	return ret;
4643	}
4644
4645	static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4646	{
4647	struct inode *inode = d_inode(dentry);
4648	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4649	int err = 0;
4650	struct btrfs_trans_handle *trans;
4651	u64 last_unlink_trans;
4652	struct fscrypt_name fname;
4653
4654	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4655	return -ENOTEMPTY;
4656	if (btrfs_ino(inode: BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4657	if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4658	btrfs_err(fs_info,
4659	"extent tree v2 doesn't support snapshot deletion yet");
4660	return -EOPNOTSUPP;
4661	}
4662	return btrfs_delete_subvolume(dir: BTRFS_I(inode: dir), dentry);
4663	}
4664
4665	err = fscrypt_setup_filename(inode: dir, iname: &dentry->d_name, lookup: 1, fname: &fname);
4666	if (err)
4667	return err;
4668
4669	/* This needs to handle no-key deletions later on */
4670
4671	trans = __unlink_start_trans(dir: BTRFS_I(inode: dir));
4672	if (IS_ERR(ptr: trans)) {
4673	err = PTR_ERR(ptr: trans);
4674	goto out_notrans;
4675	}
4676
4677	if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4678	err = btrfs_unlink_subvol(trans, dir: BTRFS_I(inode: dir), dentry);
4679	goto out;
4680	}
4681
4682	err = btrfs_orphan_add(trans, inode: BTRFS_I(inode));
4683	if (err)
4684	goto out;
4685
4686	last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4687
4688	/* now the directory is empty */
4689	err = btrfs_unlink_inode(trans, dir: BTRFS_I(inode: dir), inode: BTRFS_I(inode: d_inode(dentry)),
4690	name: &fname.disk_name);
4691	if (!err) {
4692	btrfs_i_size_write(inode: BTRFS_I(inode), size: 0);
4693	/*
4694	* Propagate the last_unlink_trans value of the deleted dir to
4695	* its parent directory. This is to prevent an unrecoverable
4696	* log tree in the case we do something like this:
4697	* 1) create dir foo
4698	* 2) create snapshot under dir foo
4699	* 3) delete the snapshot
4700	* 4) rmdir foo
4701	* 5) mkdir foo
4702	* 6) fsync foo or some file inside foo
4703	*/
4704	if (last_unlink_trans >= trans->transid)
4705	BTRFS_I(inode: dir)->last_unlink_trans = last_unlink_trans;
4706	}
4707	out:
4708	btrfs_end_transaction(trans);
4709	out_notrans:
4710	btrfs_btree_balance_dirty(fs_info);
4711	fscrypt_free_filename(fname: &fname);
4712
4713	return err;
4714	}
4715
4716	/*
4717	* Read, zero a chunk and write a block.
4718	*
4719	* @inode - inode that we're zeroing
4720	* @from - the offset to start zeroing
4721	* @len - the length to zero, 0 to zero the entire range respective to the
4722	* offset
4723	* @front - zero up to the offset instead of from the offset on
4724	*
4725	* This will find the block for the "from" offset and cow the block and zero the
4726	* part we want to zero. This is used with truncate and hole punching.
4727	*/
4728	int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4729	int front)
4730	{
4731	struct btrfs_fs_info *fs_info = inode->root->fs_info;
4732	struct address_space *mapping = inode->vfs_inode.i_mapping;
4733	struct extent_io_tree *io_tree = &inode->io_tree;
4734	struct btrfs_ordered_extent *ordered;
4735	struct extent_state *cached_state = NULL;
4736	struct extent_changeset *data_reserved = NULL;
4737	bool only_release_metadata = false;
4738	u32 blocksize = fs_info->sectorsize;
4739	pgoff_t index = from >> PAGE_SHIFT;
4740	unsigned offset = from & (blocksize - 1);
4741	struct folio *folio;
4742	gfp_t mask = btrfs_alloc_write_mask(mapping);
4743	size_t write_bytes = blocksize;
4744	int ret = 0;
4745	u64 block_start;
4746	u64 block_end;
4747
4748	if (IS_ALIGNED(offset, blocksize) &&
4749	(!len || IS_ALIGNED(len, blocksize)))
4750	goto out;
4751
4752	block_start = round_down(from, blocksize);
4753	block_end = block_start + blocksize - 1;
4754
4755	ret = btrfs_check_data_free_space(inode, reserved: &data_reserved, start: block_start,
4756	len: blocksize, noflush: false);
4757	if (ret < 0) {
4758	if (btrfs_check_nocow_lock(inode, pos: block_start, write_bytes: &write_bytes, nowait: false) > 0) {
4759	/* For nocow case, no need to reserve data space */
4760	only_release_metadata = true;
4761	} else {
4762	goto out;
4763	}
4764	}
4765	ret = btrfs_delalloc_reserve_metadata(inode, num_bytes: blocksize, disk_num_bytes: blocksize, noflush: false);
4766	if (ret < 0) {
4767	if (!only_release_metadata)
4768	btrfs_free_reserved_data_space(inode, reserved: data_reserved,
4769	start: block_start, len: blocksize);
4770	goto out;
4771	}
4772	again:
4773	folio = __filemap_get_folio(mapping, index,
4774	FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp: mask);
4775	if (IS_ERR(ptr: folio)) {
4776	btrfs_delalloc_release_space(inode, reserved: data_reserved, start: block_start,
4777	len: blocksize, qgroup_free: true);
4778	btrfs_delalloc_release_extents(inode, num_bytes: blocksize);
4779	ret = -ENOMEM;
4780	goto out;
4781	}
4782
4783	if (!folio_test_uptodate(folio)) {
4784	ret = btrfs_read_folio(NULL, folio);
4785	folio_lock(folio);
4786	if (folio->mapping != mapping) {
4787	folio_unlock(folio);
4788	folio_put(folio);
4789	goto again;
4790	}
4791	if (!folio_test_uptodate(folio)) {
4792	ret = -EIO;
4793	goto out_unlock;
4794	}
4795	}
4796
4797	/*
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	* folio private, but left the page in the mapping. Set the page mapped
4801	* here to make sure it's properly set for the subpage stuff.
4802	*/
4803	ret = set_folio_extent_mapped(folio);
4804	if (ret < 0)
4805	goto out_unlock;
4806
4807	folio_wait_writeback(folio);
4808
4809	lock_extent(tree: io_tree, start: block_start, end: block_end, cached: &cached_state);
4810
4811	ordered = btrfs_lookup_ordered_extent(inode, file_offset: block_start);
4812	if (ordered) {
4813	unlock_extent(tree: io_tree, start: block_start, end: block_end, cached: &cached_state);
4814	folio_unlock(folio);
4815	folio_put(folio);
4816	btrfs_start_ordered_extent(entry: ordered);
4817	btrfs_put_ordered_extent(entry: ordered);
4818	goto again;
4819	}
4820
4821	clear_extent_bit(tree: &inode->io_tree, start: block_start, end: block_end,
4822	bits: EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4823	cached: &cached_state);
4824
4825	ret = btrfs_set_extent_delalloc(inode, start: block_start, end: block_end, extra_bits: 0,
4826	cached_state: &cached_state);
4827	if (ret) {
4828	unlock_extent(tree: io_tree, start: block_start, end: block_end, cached: &cached_state);
4829	goto out_unlock;
4830	}
4831
4832	if (offset != blocksize) {
4833	if (!len)
4834	len = blocksize - offset;
4835	if (front)
4836	folio_zero_range(folio, start: block_start - folio_pos(folio),
4837	length: offset);
4838	else
4839	folio_zero_range(folio,
4840	start: (block_start - folio_pos(folio)) + offset,
4841	length: len);
4842	}
4843	btrfs_folio_clear_checked(fs_info, folio, start: block_start,
4844	len: block_end + 1 - block_start);
4845	btrfs_folio_set_dirty(fs_info, folio, start: block_start,
4846	len: block_end + 1 - block_start);
4847	unlock_extent(tree: io_tree, start: block_start, end: block_end, cached: &cached_state);
4848
4849	if (only_release_metadata)
4850	set_extent_bit(tree: &inode->io_tree, start: block_start, end: block_end,
4851	bits: EXTENT_NORESERVE, NULL);
4852
4853	out_unlock:
4854	if (ret) {
4855	if (only_release_metadata)
4856	btrfs_delalloc_release_metadata(inode, num_bytes: blocksize, qgroup_free: true);
4857	else
4858	btrfs_delalloc_release_space(inode, reserved: data_reserved,
4859	start: block_start, len: blocksize, qgroup_free: true);
4860	}
4861	btrfs_delalloc_release_extents(inode, num_bytes: blocksize);
4862	folio_unlock(folio);
4863	folio_put(folio);
4864	out:
4865	if (only_release_metadata)
4866	btrfs_check_nocow_unlock(inode);
4867	extent_changeset_free(changeset: data_reserved);
4868	return ret;
4869	}
4870
4871	static int maybe_insert_hole(struct btrfs_inode *inode, u64 offset, u64 len)
4872	{
4873	struct btrfs_root *root = inode->root;
4874	struct btrfs_fs_info *fs_info = root->fs_info;
4875	struct btrfs_trans_handle *trans;
4876	struct btrfs_drop_extents_args drop_args = { 0 };
4877	int ret;
4878
4879	/*
4880	* If NO_HOLES is enabled, we don't need to do anything.
4881	* Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4882	* or btrfs_update_inode() will be called, which guarantee that the next
4883	* fsync will know this inode was changed and needs to be logged.
4884	*/
4885	if (btrfs_fs_incompat(fs_info, NO_HOLES))
4886	return 0;
4887
4888	/*
4889	* 1 - for the one we're dropping
4890	* 1 - for the one we're adding
4891	* 1 - for updating the inode.
4892	*/
4893	trans = btrfs_start_transaction(root, num_items: 3);
4894	if (IS_ERR(ptr: trans))
4895	return PTR_ERR(ptr: trans);
4896
4897	drop_args.start = offset;
4898	drop_args.end = offset + len;
4899	drop_args.drop_cache = true;
4900
4901	ret = btrfs_drop_extents(trans, root, inode, args: &drop_args);
4902	if (ret) {
4903	btrfs_abort_transaction(trans, ret);
4904	btrfs_end_transaction(trans);
4905	return ret;
4906	}
4907
4908	ret = btrfs_insert_hole_extent(trans, root, objectid: btrfs_ino(inode), pos: offset, num_bytes: len);
4909	if (ret) {
4910	btrfs_abort_transaction(trans, ret);
4911	} else {
4912	btrfs_update_inode_bytes(inode, add_bytes: 0, del_bytes: drop_args.bytes_found);
4913	btrfs_update_inode(trans, inode);
4914	}
4915	btrfs_end_transaction(trans);
4916	return ret;
4917	}
4918
4919	/*
4920	* This function puts in dummy file extents for the area we're creating a hole
4921	* for. So if we are truncating this file to a larger size we need to insert
4922	* these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4923	* the range between oldsize and size
4924	*/
4925	int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4926	{
4927	struct btrfs_root *root = inode->root;
4928	struct btrfs_fs_info *fs_info = root->fs_info;
4929	struct extent_io_tree *io_tree = &inode->io_tree;
4930	struct extent_map *em = NULL;
4931	struct extent_state *cached_state = NULL;
4932	u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4933	u64 block_end = ALIGN(size, fs_info->sectorsize);
4934	u64 last_byte;
4935	u64 cur_offset;
4936	u64 hole_size;
4937	int err = 0;
4938
4939	/*
4940	* If our size started in the middle of a block we need to zero out the
4941	* rest of the block before we expand the i_size, otherwise we could
4942	* expose stale data.
4943	*/
4944	err = btrfs_truncate_block(inode, from: oldsize, len: 0, front: 0);
4945	if (err)
4946	return err;
4947
4948	if (size <= hole_start)
4949	return 0;
4950
4951	btrfs_lock_and_flush_ordered_range(inode, start: hole_start, end: block_end - 1,
4952	cached_state: &cached_state);
4953	cur_offset = hole_start;
4954	while (1) {
4955	em = btrfs_get_extent(inode, NULL, start: cur_offset, len: block_end - cur_offset);
4956	if (IS_ERR(ptr: em)) {
4957	err = PTR_ERR(ptr: em);
4958	em = NULL;
4959	break;
4960	}
4961	last_byte = min(extent_map_end(em), block_end);
4962	last_byte = ALIGN(last_byte, fs_info->sectorsize);
4963	hole_size = last_byte - cur_offset;
4964
4965	if (!(em->flags & EXTENT_FLAG_PREALLOC)) {
4966	struct extent_map *hole_em;
4967
4968	err = maybe_insert_hole(inode, offset: cur_offset, len: hole_size);
4969	if (err)
4970	break;
4971
4972	err = btrfs_inode_set_file_extent_range(inode,
4973	start: cur_offset, len: hole_size);
4974	if (err)
4975	break;
4976
4977	hole_em = alloc_extent_map();
4978	if (!hole_em) {
4979	btrfs_drop_extent_map_range(inode, start: cur_offset,
4980	end: cur_offset + hole_size - 1,
4981	skip_pinned: false);
4982	btrfs_set_inode_full_sync(inode);
4983	goto next;
4984	}
4985	hole_em->start = cur_offset;
4986	hole_em->len = hole_size;
4987	hole_em->orig_start = cur_offset;
4988
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->generation = btrfs_get_fs_generation(fs_info);
4994
4995	err = btrfs_replace_extent_map_range(inode, new_em: hole_em, modified: true);
4996	free_extent_map(em: hole_em);
4997	} else {
4998	err = btrfs_inode_set_file_extent_range(inode,
4999	start: cur_offset, len: hole_size);
5000	if (err)
5001	break;
5002	}
5003	next:
5004	free_extent_map(em);
5005	em = NULL;
5006	cur_offset = last_byte;
5007	if (cur_offset >= block_end)
5008	break;
5009	}
5010	free_extent_map(em);
5011	unlock_extent(tree: io_tree, start: hole_start, end: block_end - 1, cached: &cached_state);
5012	return err;
5013	}
5014
5015	static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5016	{
5017	struct btrfs_root *root = BTRFS_I(inode)->root;
5018	struct btrfs_trans_handle *trans;
5019	loff_t oldsize = i_size_read(inode);
5020	loff_t newsize = attr->ia_size;
5021	int mask = attr->ia_valid;
5022	int ret;
5023
5024	/*
5025	* The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5026	* special case where we need to update the times despite not having
5027	* these flags set. For all other operations the VFS set these flags
5028	* explicitly if it wants a timestamp update.
5029	*/
5030	if (newsize != oldsize) {
5031	inode_inc_iversion(inode);
5032	if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5033	inode_set_mtime_to_ts(inode,
5034	ts: inode_set_ctime_current(inode));
5035	}
5036	}
5037
5038	if (newsize > oldsize) {
5039	/*
5040	* Don't do an expanding truncate while snapshotting is ongoing.
5041	* This is to ensure the snapshot captures a fully consistent
5042	* state of this file - if the snapshot captures this expanding
5043	* truncation, it must capture all writes that happened before
5044	* this truncation.
5045	*/
5046	btrfs_drew_write_lock(lock: &root->snapshot_lock);
5047	ret = btrfs_cont_expand(inode: BTRFS_I(inode), oldsize, size: newsize);
5048	if (ret) {
5049	btrfs_drew_write_unlock(lock: &root->snapshot_lock);
5050	return ret;
5051	}
5052
5053	trans = btrfs_start_transaction(root, num_items: 1);
5054	if (IS_ERR(ptr: trans)) {
5055	btrfs_drew_write_unlock(lock: &root->snapshot_lock);
5056	return PTR_ERR(ptr: trans);
5057	}
5058
5059	i_size_write(inode, i_size: newsize);
5060	btrfs_inode_safe_disk_i_size_write(inode: BTRFS_I(inode), new_i_size: 0);
5061	pagecache_isize_extended(inode, from: oldsize, to: newsize);
5062	ret = btrfs_update_inode(trans, inode: BTRFS_I(inode));
5063	btrfs_drew_write_unlock(lock: &root->snapshot_lock);
5064	btrfs_end_transaction(trans);
5065	} else {
5066	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
5067
5068	if (btrfs_is_zoned(fs_info)) {
5069	ret = btrfs_wait_ordered_range(inode,
5070	ALIGN(newsize, fs_info->sectorsize),
5071	len: (u64)-1);
5072	if (ret)
5073	return ret;
5074	}
5075
5076	/*
5077	* We're truncating a file that used to have good data down to
5078	* zero. Make sure any new writes to the file get on disk
5079	* on close.
5080	*/
5081	if (newsize == 0)
5082	set_bit(nr: BTRFS_INODE_FLUSH_ON_CLOSE,
5083	addr: &BTRFS_I(inode)->runtime_flags);
5084
5085	truncate_setsize(inode, newsize);
5086
5087	inode_dio_wait(inode);
5088
5089	ret = btrfs_truncate(inode: BTRFS_I(inode), skip_writeback: newsize == oldsize);
5090	if (ret && inode->i_nlink) {
5091	int err;
5092
5093	/*
5094	* Truncate failed, so fix up the in-memory size. We
5095	* adjusted disk_i_size down as we removed extents, so
5096	* wait for disk_i_size to be stable and then update the
5097	* in-memory size to match.
5098	*/
5099	err = btrfs_wait_ordered_range(inode, start: 0, len: (u64)-1);
5100	if (err)
5101	return err;
5102	i_size_write(inode, i_size: BTRFS_I(inode)->disk_i_size);
5103	}
5104	}
5105
5106	return ret;
5107	}
5108
5109	static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5110	struct iattr *attr)
5111	{
5112	struct inode *inode = d_inode(dentry);
5113	struct btrfs_root *root = BTRFS_I(inode)->root;
5114	int err;
5115
5116	if (btrfs_root_readonly(root))
5117	return -EROFS;
5118
5119	err = setattr_prepare(idmap, dentry, attr);
5120	if (err)
5121	return err;
5122
5123	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5124	err = btrfs_setsize(inode, attr);
5125	if (err)
5126	return err;
5127	}
5128
5129	if (attr->ia_valid) {
5130	setattr_copy(idmap, inode, attr);
5131	inode_inc_iversion(inode);
5132	err = btrfs_dirty_inode(inode: BTRFS_I(inode));
5133
5134	if (!err && attr->ia_valid & ATTR_MODE)
5135	err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5136	}
5137
5138	return err;
5139	}
5140
5141	/*
5142	* While truncating the inode pages during eviction, we get the VFS
5143	* calling btrfs_invalidate_folio() against each folio of the inode. This
5144	* is slow because the calls to btrfs_invalidate_folio() result in a
5145	* huge amount of calls to lock_extent() and clear_extent_bit(),
5146	* which keep merging and splitting extent_state structures over and over,
5147	* wasting lots of time.
5148	*
5149	* Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5150	* skip all those expensive operations on a per folio basis and do only
5151	* the ordered io finishing, while we release here the extent_map and
5152	* extent_state structures, without the excessive merging and splitting.
5153	*/
5154	static void evict_inode_truncate_pages(struct inode *inode)
5155	{
5156	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5157	struct rb_node *node;
5158
5159	ASSERT(inode->i_state & I_FREEING);
5160	truncate_inode_pages_final(&inode->i_data);
5161
5162	btrfs_drop_extent_map_range(inode: BTRFS_I(inode), start: 0, end: (u64)-1, skip_pinned: false);
5163
5164	/*
5165	* Keep looping until we have no more ranges in the io tree.
5166	* We can have ongoing bios started by readahead that have
5167	* their endio callback (extent_io.c:end_bio_extent_readpage)
5168	* still in progress (unlocked the pages in the bio but did not yet
5169	* unlocked the ranges in the io tree). Therefore this means some
5170	* ranges can still be locked and eviction started because before
5171	* submitting those bios, which are executed by a separate task (work
5172	* queue kthread), inode references (inode->i_count) were not taken
5173	* (which would be dropped in the end io callback of each bio).
5174	* Therefore here we effectively end up waiting for those bios and
5175	* anyone else holding locked ranges without having bumped the inode's
5176	* reference count - if we don't do it, when they access the inode's
5177	* io_tree to unlock a range it may be too late, leading to an
5178	* use-after-free issue.
5179	*/
5180	spin_lock(lock: &io_tree->lock);
5181	while (!RB_EMPTY_ROOT(&io_tree->state)) {
5182	struct extent_state *state;
5183	struct extent_state *cached_state = NULL;
5184	u64 start;
5185	u64 end;
5186	unsigned state_flags;
5187
5188	node = rb_first(&io_tree->state);
5189	state = rb_entry(node, struct extent_state, rb_node);
5190	start = state->start;
5191	end = state->end;
5192	state_flags = state->state;
5193	spin_unlock(lock: &io_tree->lock);
5194
5195	lock_extent(tree: io_tree, start, end, cached: &cached_state);
5196
5197	/*
5198	* If still has DELALLOC flag, the extent didn't reach disk,
5199	* and its reserved space won't be freed by delayed_ref.
5200	* So we need to free its reserved space here.
5201	* (Refer to comment in btrfs_invalidate_folio, case 2)
5202	*
5203	* Note, end is the bytenr of last byte, so we need + 1 here.
5204	*/
5205	if (state_flags & EXTENT_DELALLOC)
5206	btrfs_qgroup_free_data(inode: BTRFS_I(inode), NULL, start,
5207	len: end - start + 1, NULL);
5208
5209	clear_extent_bit(tree: io_tree, start, end,
5210	bits: EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5211	cached: &cached_state);
5212
5213	cond_resched();
5214	spin_lock(lock: &io_tree->lock);
5215	}
5216	spin_unlock(lock: &io_tree->lock);
5217	}
5218
5219	static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5220	struct btrfs_block_rsv *rsv)
5221	{
5222	struct btrfs_fs_info *fs_info = root->fs_info;
5223	struct btrfs_trans_handle *trans;
5224	u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, num_delayed_refs: 1);
5225	int ret;
5226
5227	/*
5228	* Eviction should be taking place at some place safe because of our
5229	* delayed iputs. However the normal flushing code will run delayed
5230	* iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5231	*
5232	* We reserve the delayed_refs_extra here again because we can't use
5233	* btrfs_start_transaction(root, 0) for the same deadlocky reason as
5234	* above. We reserve our extra bit here because we generate a ton of
5235	* delayed refs activity by truncating.
5236	*
5237	* BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5238	* if we fail to make this reservation we can re-try without the
5239	* delayed_refs_extra so we can make some forward progress.
5240	*/
5241	ret = btrfs_block_rsv_refill(fs_info, block_rsv: rsv, num_bytes: rsv->size + delayed_refs_extra,
5242	flush: BTRFS_RESERVE_FLUSH_EVICT);
5243	if (ret) {
5244	ret = btrfs_block_rsv_refill(fs_info, block_rsv: rsv, num_bytes: rsv->size,
5245	flush: BTRFS_RESERVE_FLUSH_EVICT);
5246	if (ret) {
5247	btrfs_warn(fs_info,
5248	"could not allocate space for delete; will truncate on mount");
5249	return ERR_PTR(error: -ENOSPC);
5250	}
5251	delayed_refs_extra = 0;
5252	}
5253
5254	trans = btrfs_join_transaction(root);
5255	if (IS_ERR(ptr: trans))
5256	return trans;
5257
5258	if (delayed_refs_extra) {
5259	trans->block_rsv = &fs_info->trans_block_rsv;
5260	trans->bytes_reserved = delayed_refs_extra;
5261	btrfs_block_rsv_migrate(src_rsv: rsv, dst_rsv: trans->block_rsv,
5262	num_bytes: delayed_refs_extra, update_size: true);
5263	}
5264	return trans;
5265	}
5266
5267	void btrfs_evict_inode(struct inode *inode)
5268	{
5269	struct btrfs_fs_info *fs_info;
5270	struct btrfs_trans_handle *trans;
5271	struct btrfs_root *root = BTRFS_I(inode)->root;
5272	struct btrfs_block_rsv *rsv = NULL;
5273	int ret;
5274
5275	trace_btrfs_inode_evict(inode);
5276
5277	if (!root) {
5278	fsverity_cleanup_inode(inode);
5279	clear_inode(inode);
5280	return;
5281	}
5282
5283	fs_info = inode_to_fs_info(inode);
5284	evict_inode_truncate_pages(inode);
5285
5286	if (inode->i_nlink &&
5287	((btrfs_root_refs(s: &root->root_item) != 0 &&
5288	root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5289	btrfs_is_free_space_inode(inode: BTRFS_I(inode))))
5290	goto out;
5291
5292	if (is_bad_inode(inode))
5293	goto out;
5294
5295	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5296	goto out;
5297
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);
5301	goto out;
5302	}
5303
5304	/*
5305	* This makes sure the inode item in tree is uptodate and the space for
5306	* the inode update is released.
5307	*/
5308	ret = btrfs_commit_inode_delayed_inode(inode: BTRFS_I(inode));
5309	if (ret)
5310	goto out;
5311
5312	/*
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
5316	* the truncate.
5317	*/
5318	btrfs_kill_delayed_inode_items(inode: BTRFS_I(inode));
5319
5320	rsv = btrfs_alloc_block_rsv(fs_info, type: BTRFS_BLOCK_RSV_TEMP);
5321	if (!rsv)
5322	goto out;
5323	rsv->size = btrfs_calc_metadata_size(fs_info, num_items: 1);
5324	rsv->failfast = true;
5325
5326	btrfs_i_size_write(inode: BTRFS_I(inode), size: 0);
5327
5328	while (1) {
5329	struct btrfs_truncate_control control = {
5330	.inode = BTRFS_I(inode),
5331	.ino = btrfs_ino(inode: BTRFS_I(inode)),
5332	.new_size = 0,
5333	.min_type = 0,
5334	};
5335
5336	trans = evict_refill_and_join(root, rsv);
5337	if (IS_ERR(ptr: trans))
5338	goto out;
5339
5340	trans->block_rsv = rsv;
5341
5342	ret = btrfs_truncate_inode_items(trans, root, control: &control);
5343	trans->block_rsv = &fs_info->trans_block_rsv;
5344	btrfs_end_transaction(trans);
5345	/*
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.
5349	*/
5350	btrfs_btree_balance_dirty_nodelay(fs_info);
5351	if (ret && ret != -ENOSPC && ret != -EAGAIN)
5352	goto out;
5353	else if (!ret)
5354	break;
5355	}
5356
5357	/*
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.
5362	*
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.
5365	*/
5366	trans = evict_refill_and_join(root, rsv);
5367	if (!IS_ERR(ptr: trans)) {
5368	trans->block_rsv = rsv;
5369	btrfs_orphan_del(trans, inode: BTRFS_I(inode));
5370	trans->block_rsv = &fs_info->trans_block_rsv;
5371	btrfs_end_transaction(trans);
5372	}
5373
5374	out:
5375	btrfs_free_block_rsv(fs_info, rsv);
5376	/*
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.
5380	*/
5381	btrfs_remove_delayed_node(inode: BTRFS_I(inode));
5382	fsverity_cleanup_inode(inode);
5383	clear_inode(inode);
5384	}
5385
5386	/*
5387	* Return the key found in the dir entry in the location pointer, fill @type
5388	* with BTRFS_FT_*, and return 0.
5389	*
5390	* If no dir entries were found, returns -ENOENT.
5391	* If found a corrupted location in dir entry, returns -EUCLEAN.
5392	*/
5393	static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5394	struct btrfs_key *location, u8 *type)
5395	{
5396	struct btrfs_dir_item *di;
5397	struct btrfs_path *path;
5398	struct btrfs_root *root = dir->root;
5399	int ret = 0;
5400	struct fscrypt_name fname;
5401
5402	path = btrfs_alloc_path();
5403	if (!path)
5404	return -ENOMEM;
5405
5406	ret = fscrypt_setup_filename(inode: &dir->vfs_inode, iname: &dentry->d_name, lookup: 1, fname: &fname);
5407	if (ret < 0)
5408	goto out;
5409	/*
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.
5412	*/
5413	ASSERT(ret == 0);
5414
5415	/* This needs to handle no-key deletions later on */
5416
5417	di = btrfs_lookup_dir_item(NULL, root, path, dir: btrfs_ino(inode: dir),
5418	name: &fname.disk_name, mod: 0);
5419	if (IS_ERR_OR_NULL(ptr: di)) {
5420	ret = di ? PTR_ERR(ptr: di) : -ENOENT;
5421	goto out;
5422	}
5423
5424	btrfs_dir_item_key_to_cpu(eb: path->nodes[0], item: di, cpu_key: location);
5425	if (location->type != BTRFS_INODE_ITEM_KEY &&
5426	location->type != BTRFS_ROOT_ITEM_KEY) {
5427	ret = -EUCLEAN;
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);
5432	}
5433	if (!ret)
5434	*type = btrfs_dir_ftype(eb: path->nodes[0], item: di);
5435	out:
5436	fscrypt_free_filename(fname: &fname);
5437	btrfs_free_path(p: path);
5438	return ret;
5439	}
5440
5441	/*
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.
5445	*/
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)
5451	{
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;
5457	int ret;
5458	int err = 0;
5459	struct fscrypt_name fname;
5460
5461	ret = fscrypt_setup_filename(inode: &dir->vfs_inode, iname: &dentry->d_name, lookup: 0, fname: &fname);
5462	if (ret)
5463	return ret;
5464
5465	path = btrfs_alloc_path();
5466	if (!path) {
5467	err = -ENOMEM;
5468	goto out;
5469	}
5470
5471	err = -ENOENT;
5472	key.objectid = dir->root->root_key.objectid;
5473	key.type = BTRFS_ROOT_REF_KEY;
5474	key.offset = location->objectid;
5475
5476	ret = btrfs_search_slot(NULL, root: fs_info->tree_root, key: &key, p: path, ins_len: 0, cow: 0);
5477	if (ret) {
5478	if (ret < 0)
5479	err = ret;
5480	goto out;
5481	}
5482
5483	leaf = path->nodes[0];
5484	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5485	if (btrfs_root_ref_dirid(eb: leaf, s: ref) != btrfs_ino(inode: dir) ||
5486	btrfs_root_ref_name_len(eb: leaf, s: ref) != fname.disk_name.len)
5487	goto out;
5488
5489	ret = memcmp_extent_buffer(eb: leaf, ptrv: fname.disk_name.name,
5490	start: (unsigned long)(ref + 1), len: fname.disk_name.len);
5491	if (ret)
5492	goto out;
5493
5494	btrfs_release_path(p: path);
5495
5496	new_root = btrfs_get_fs_root(fs_info, objectid: location->objectid, check_ref: true);
5497	if (IS_ERR(ptr: new_root)) {
5498	err = PTR_ERR(ptr: new_root);
5499	goto out;
5500	}
5501
5502	*sub_root = new_root;
5503	location->objectid = btrfs_root_dirid(s: &new_root->root_item);
5504	location->type = BTRFS_INODE_ITEM_KEY;
5505	location->offset = 0;
5506	err = 0;
5507	out:
5508	btrfs_free_path(p: path);
5509	fscrypt_free_filename(fname: &fname);
5510	return err;
5511	}
5512
5513	static void inode_tree_add(struct btrfs_inode *inode)
5514	{
5515	struct btrfs_root *root = inode->root;
5516	struct btrfs_inode *entry;
5517	struct rb_node **p;
5518	struct rb_node *parent;
5519	struct rb_node *new = &inode->rb_node;
5520	u64 ino = btrfs_ino(inode);
5521
5522	if (inode_unhashed(inode: &inode->vfs_inode))
5523	return;
5524	parent = NULL;
5525	spin_lock(lock: &root->inode_lock);
5526	p = &root->inode_tree.rb_node;
5527	while (*p) {
5528	parent = *p;
5529	entry = rb_entry(parent, struct btrfs_inode, rb_node);
5530
5531	if (ino < btrfs_ino(inode: entry))
5532	p = &parent->rb_left;
5533	else if (ino > btrfs_ino(inode: entry))
5534	p = &parent->rb_right;
5535	else {
5536	WARN_ON(!(entry->vfs_inode.i_state &
5537	(I_WILL_FREE | I_FREEING)));
5538	rb_replace_node(victim: parent, new, root: &root->inode_tree);
5539	RB_CLEAR_NODE(parent);
5540	spin_unlock(lock: &root->inode_lock);
5541	return;
5542	}
5543	}
5544	rb_link_node(node: new, parent, rb_link: p);
5545	rb_insert_color(new, &root->inode_tree);
5546	spin_unlock(lock: &root->inode_lock);
5547	}
5548
5549	static void inode_tree_del(struct btrfs_inode *inode)
5550	{
5551	struct btrfs_root *root = inode->root;
5552	int empty = 0;
5553
5554	spin_lock(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);
5559	}
5560	spin_unlock(lock: &root->inode_lock);
5561
5562	if (empty && btrfs_root_refs(s: &root->root_item) == 0) {
5563	spin_lock(lock: &root->inode_lock);
5564	empty = RB_EMPTY_ROOT(&root->inode_tree);
5565	spin_unlock(lock: &root->inode_lock);
5566	if (empty)
5567	btrfs_add_dead_root(root);
5568	}
5569	}
5570
5571
5572	static int btrfs_init_locked_inode(struct inode *inode, void *p)
5573	{
5574	struct btrfs_iget_args *args = p;
5575
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(root: args->root);
5581
5582	if (args->root && args->root == args->root->fs_info->tree_root &&
5583	args->ino != BTRFS_BTREE_INODE_OBJECTID)
5584	set_bit(nr: BTRFS_INODE_FREE_SPACE_INODE,
5585	addr: &BTRFS_I(inode)->runtime_flags);
5586	return 0;
5587	}
5588
5589	static int btrfs_find_actor(struct inode *inode, void *opaque)
5590	{
5591	struct btrfs_iget_args *args = opaque;
5592
5593	return args->ino == BTRFS_I(inode)->location.objectid &&
5594	args->root == BTRFS_I(inode)->root;
5595	}
5596
5597	static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5598	struct btrfs_root *root)
5599	{
5600	struct inode *inode;
5601	struct btrfs_iget_args args;
5602	unsigned long hashval = btrfs_inode_hash(objectid: ino, root);
5603
5604	args.ino = ino;
5605	args.root = root;
5606
5607	inode = iget5_locked(s, hashval, test: btrfs_find_actor,
5608	set: btrfs_init_locked_inode,
5609	(void *)&args);
5610	return inode;
5611	}
5612
5613	/*
5614	* Get an inode object given its inode number and corresponding root.
5615	* Path can be preallocated to prevent recursing back to iget through
5616	* allocator. NULL is also valid but may require an additional allocation
5617	* later.
5618	*/
5619	struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5620	struct btrfs_root *root, struct btrfs_path *path)
5621	{
5622	struct inode *inode;
5623
5624	inode = btrfs_iget_locked(s, ino, root);
5625	if (!inode)
5626	return ERR_PTR(error: -ENOMEM);
5627
5628	if (inode->i_state & I_NEW) {
5629	int ret;
5630
5631	ret = btrfs_read_locked_inode(inode, in_path: path);
5632	if (!ret) {
5633	inode_tree_add(inode: BTRFS_I(inode));
5634	unlock_new_inode(inode);
5635	} else {
5636	iget_failed(inode);
5637	/*
5638	* ret > 0 can come from btrfs_search_slot called by
5639	* btrfs_read_locked_inode, this means the inode item
5640	* was not found.
5641	*/
5642	if (ret > 0)
5643	ret = -ENOENT;
5644	inode = ERR_PTR(error: ret);
5645	}
5646	}
5647
5648	return inode;
5649	}
5650
5651	struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5652	{
5653	return btrfs_iget_path(s, ino, root, NULL);
5654	}
5655
5656	static struct inode *new_simple_dir(struct inode *dir,
5657	struct btrfs_key *key,
5658	struct btrfs_root *root)
5659	{
5660	struct timespec64 ts;
5661	struct inode *inode = new_inode(sb: dir->i_sb);
5662
5663	if (!inode)
5664	return ERR_PTR(error: -ENOMEM);
5665
5666	BTRFS_I(inode)->root = btrfs_grab_root(root);
5667	memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5668	set_bit(nr: BTRFS_INODE_DUMMY, addr: &BTRFS_I(inode)->runtime_flags);
5669
5670	inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5671	/*
5672	* We only need lookup, the rest is read-only and there's no inode
5673	* associated with the dentry
5674	*/
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
5680	ts = inode_set_ctime_current(inode);
5681	inode_set_mtime_to_ts(inode, ts);
5682	inode_set_atime_to_ts(inode, ts: inode_get_atime(inode: dir));
5683	BTRFS_I(inode)->i_otime_sec = ts.tv_sec;
5684	BTRFS_I(inode)->i_otime_nsec = ts.tv_nsec;
5685
5686	inode->i_uid = dir->i_uid;
5687	inode->i_gid = dir->i_gid;
5688
5689	return inode;
5690	}
5691
5692	static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5693	static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5694	static_assert(BTRFS_FT_DIR == FT_DIR);
5695	static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5696	static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5697	static_assert(BTRFS_FT_FIFO == FT_FIFO);
5698	static_assert(BTRFS_FT_SOCK == FT_SOCK);
5699	static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5700
5701	static inline u8 btrfs_inode_type(struct inode *inode)
5702	{
5703	return fs_umode_to_ftype(mode: inode->i_mode);
5704	}
5705
5706	struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5707	{
5708	struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
5709	struct inode *inode;
5710	struct btrfs_root *root = BTRFS_I(inode: dir)->root;
5711	struct btrfs_root *sub_root = root;
5712	struct btrfs_key location;
5713	u8 di_type = 0;
5714	int ret = 0;
5715
5716	if (dentry->d_name.len > BTRFS_NAME_LEN)
5717	return ERR_PTR(error: -ENAMETOOLONG);
5718
5719	ret = btrfs_inode_by_name(dir: BTRFS_I(inode: dir), dentry, location: &location, type: &di_type);
5720	if (ret < 0)
5721	return ERR_PTR(error: ret);
5722
5723	if (location.type == BTRFS_INODE_ITEM_KEY) {
5724	inode = btrfs_iget(s: dir->i_sb, ino: location.objectid, root);
5725	if (IS_ERR(ptr: inode))
5726	return inode;
5727
5728	/* Do extra check against inode mode with di_type */
5729	if (btrfs_inode_type(inode) != di_type) {
5730	btrfs_crit(fs_info,
5731	"inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5732	inode->i_mode, btrfs_inode_type(inode),
5733	di_type);
5734	iput(inode);
5735	return ERR_PTR(error: -EUCLEAN);
5736	}
5737	return inode;
5738	}
5739
5740	ret = fixup_tree_root_location(fs_info, dir: BTRFS_I(inode: dir), dentry,
5741	location: &location, sub_root: &sub_root);
5742	if (ret < 0) {
5743	if (ret != -ENOENT)
5744	inode = ERR_PTR(error: ret);
5745	else
5746	inode = new_simple_dir(dir, key: &location, root);
5747	} else {
5748	inode = btrfs_iget(s: dir->i_sb, ino: location.objectid, root: sub_root);
5749	btrfs_put_root(root: sub_root);
5750
5751	if (IS_ERR(ptr: inode))
5752	return inode;
5753
5754	down_read(sem: &fs_info->cleanup_work_sem);
5755	if (!sb_rdonly(sb: inode->i_sb))
5756	ret = btrfs_orphan_cleanup(root: sub_root);
5757	up_read(sem: &fs_info->cleanup_work_sem);
5758	if (ret) {
5759	iput(inode);
5760	inode = ERR_PTR(error: ret);
5761	}
5762	}
5763
5764	return inode;
5765	}
5766
5767	static int btrfs_dentry_delete(const struct dentry *dentry)
5768	{
5769	struct btrfs_root *root;
5770	struct inode *inode = d_inode(dentry);
5771
5772	if (!inode && !IS_ROOT(dentry))
5773	inode = d_inode(dentry: dentry->d_parent);
5774
5775	if (inode) {
5776	root = BTRFS_I(inode)->root;
5777	if (btrfs_root_refs(s: &root->root_item) == 0)
5778	return 1;
5779
5780	if (btrfs_ino(inode: BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5781	return 1;
5782	}
5783	return 0;
5784	}
5785
5786	static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5787	unsigned int flags)
5788	{
5789	struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5790
5791	if (inode == ERR_PTR(error: -ENOENT))
5792	inode = NULL;
5793	return d_splice_alias(inode, dentry);
5794	}
5795
5796	/*
5797	* Find the highest existing sequence number in a directory and then set the
5798	* in-memory index_cnt variable to the first free sequence number.
5799	*/
5800	static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5801	{
5802	struct btrfs_root *root = inode->root;
5803	struct btrfs_key key, found_key;
5804	struct btrfs_path *path;
5805	struct extent_buffer *leaf;
5806	int ret;
5807
5808	key.objectid = btrfs_ino(inode);
5809	key.type = BTRFS_DIR_INDEX_KEY;
5810	key.offset = (u64)-1;
5811
5812	path = btrfs_alloc_path();
5813	if (!path)
5814	return -ENOMEM;
5815
5816	ret = btrfs_search_slot(NULL, root, key: &key, p: path, ins_len: 0, cow: 0);
5817	if (ret < 0)
5818	goto out;
5819	/* FIXME: we should be able to handle this */
5820	if (ret == 0)
5821	goto out;
5822	ret = 0;
5823
5824	if (path->slots[0] == 0) {
5825	inode->index_cnt = BTRFS_DIR_START_INDEX;
5826	goto out;
5827	}
5828
5829	path->slots[0]--;
5830
5831	leaf = path->nodes[0];
5832	btrfs_item_key_to_cpu(eb: leaf, cpu_key: &found_key, nr: path->slots[0]);
5833
5834	if (found_key.objectid != btrfs_ino(inode) ||
5835	found_key.type != BTRFS_DIR_INDEX_KEY) {
5836	inode->index_cnt = BTRFS_DIR_START_INDEX;
5837	goto out;
5838	}
5839
5840	inode->index_cnt = found_key.offset + 1;
5841	out:
5842	btrfs_free_path(p: path);
5843	return ret;
5844	}
5845
5846	static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5847	{
5848	int ret = 0;
5849
5850	btrfs_inode_lock(inode: dir, ilock_flags: 0);
5851	if (dir->index_cnt == (u64)-1) {
5852	ret = btrfs_inode_delayed_dir_index_count(inode: dir);
5853	if (ret) {
5854	ret = btrfs_set_inode_index_count(inode: dir);
5855	if (ret)
5856	goto out;
5857	}
5858	}
5859
5860	/* index_cnt is the index number of next new entry, so decrement it. */
5861	*index = dir->index_cnt - 1;
5862	out:
5863	btrfs_inode_unlock(inode: dir, ilock_flags: 0);
5864
5865	return ret;
5866	}
5867
5868	/*
5869	* All this infrastructure exists because dir_emit can fault, and we are holding
5870	* the tree lock when doing readdir. For now just allocate a buffer and copy
5871	* our information into that, and then dir_emit from the buffer. This is
5872	* similar to what NFS does, only we don't keep the buffer around in pagecache
5873	* because I'm afraid I'll mess that up. Long term we need to make filldir do
5874	* copy_to_user_inatomic so we don't have to worry about page faulting under the
5875	* tree lock.
5876	*/
5877	static int btrfs_opendir(struct inode *inode, struct file *file)
5878	{
5879	struct btrfs_file_private *private;
5880	u64 last_index;
5881	int ret;
5882
5883	ret = btrfs_get_dir_last_index(dir: BTRFS_I(inode), index: &last_index);
5884	if (ret)
5885	return ret;
5886
5887	private = kzalloc(size: sizeof(struct btrfs_file_private), GFP_KERNEL);
5888	if (!private)
5889	return -ENOMEM;
5890	private->last_index = last_index;
5891	private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5892	if (!private->filldir_buf) {
5893	kfree(objp: private);
5894	return -ENOMEM;
5895	}
5896	file->private_data = private;
5897	return 0;
5898	}
5899
5900	static loff_t btrfs_dir_llseek(struct file *file, loff_t offset, int whence)
5901	{
5902	struct btrfs_file_private *private = file->private_data;
5903	int ret;
5904
5905	ret = btrfs_get_dir_last_index(dir: BTRFS_I(inode: file_inode(f: file)),
5906	index: &private->last_index);
5907	if (ret)
5908	return ret;
5909
5910	return generic_file_llseek(file, offset, whence);
5911	}
5912
5913	struct dir_entry {
5914	u64 ino;
5915	u64 offset;
5916	unsigned type;
5917	int name_len;
5918	};
5919
5920	static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5921	{
5922	while (entries--) {
5923	struct dir_entry *entry = addr;
5924	char *name = (char *)(entry + 1);
5925
5926	ctx->pos = get_unaligned(&entry->offset);
5927	if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5928	get_unaligned(&entry->ino),
5929	get_unaligned(&entry->type)))
5930	return 1;
5931	addr += sizeof(struct dir_entry) +
5932	get_unaligned(&entry->name_len);
5933	ctx->pos++;
5934	}
5935	return 0;
5936	}
5937
5938	static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5939	{
5940	struct inode *inode = file_inode(f: file);
5941	struct btrfs_root *root = BTRFS_I(inode)->root;
5942	struct btrfs_file_private *private = file->private_data;
5943	struct btrfs_dir_item *di;
5944	struct btrfs_key key;
5945	struct btrfs_key found_key;
5946	struct btrfs_path *path;
5947	void *addr;
5948	LIST_HEAD(ins_list);
5949	LIST_HEAD(del_list);
5950	int ret;
5951	char *name_ptr;
5952	int name_len;
5953	int entries = 0;
5954	int total_len = 0;
5955	bool put = false;
5956	struct btrfs_key location;
5957
5958	if (!dir_emit_dots(file, ctx))
5959	return 0;
5960
5961	path = btrfs_alloc_path();
5962	if (!path)
5963	return -ENOMEM;
5964
5965	addr = private->filldir_buf;
5966	path->reada = READA_FORWARD;
5967
5968	put = btrfs_readdir_get_delayed_items(inode, last_index: private->last_index,
5969	ins_list: &ins_list, del_list: &del_list);
5970
5971	again:
5972	key.type = BTRFS_DIR_INDEX_KEY;
5973	key.offset = ctx->pos;
5974	key.objectid = btrfs_ino(inode: BTRFS_I(inode));
5975
5976	btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5977	struct dir_entry *entry;
5978	struct extent_buffer *leaf = path->nodes[0];
5979	u8 ftype;
5980
5981	if (found_key.objectid != key.objectid)
5982	break;
5983	if (found_key.type != BTRFS_DIR_INDEX_KEY)
5984	break;
5985	if (found_key.offset < ctx->pos)
5986	continue;
5987	if (found_key.offset > private->last_index)
5988	break;
5989	if (btrfs_should_delete_dir_index(del_list: &del_list, index: found_key.offset))
5990	continue;
5991	di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5992	name_len = btrfs_dir_name_len(eb: leaf, s: di);
5993	if ((total_len + sizeof(struct dir_entry) + name_len) >=
5994	PAGE_SIZE) {
5995	btrfs_release_path(p: path);
5996	ret = btrfs_filldir(addr: private->filldir_buf, entries, ctx);
5997	if (ret)
5998	goto nopos;
5999	addr = private->filldir_buf;
6000	entries = 0;
6001	total_len = 0;
6002	goto again;
6003	}
6004
6005	ftype = btrfs_dir_flags_to_ftype(flags: btrfs_dir_flags(eb: leaf, s: di));
6006	entry = addr;
6007	name_ptr = (char *)(entry + 1);
6008	read_extent_buffer(eb: leaf, dst: name_ptr,
6009	start: (unsigned long)(di + 1), len: name_len);
6010	put_unaligned(name_len, &entry->name_len);
6011	put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
6012	btrfs_dir_item_key_to_cpu(eb: leaf, item: di, cpu_key: &location);
6013	put_unaligned(location.objectid, &entry->ino);
6014	put_unaligned(found_key.offset, &entry->offset);
6015	entries++;
6016	addr += sizeof(struct dir_entry) + name_len;
6017	total_len += sizeof(struct dir_entry) + name_len;
6018	}
6019	/* Catch error encountered during iteration */
6020	if (ret < 0)
6021	goto err;
6022
6023	btrfs_release_path(p: path);
6024
6025	ret = btrfs_filldir(addr: private->filldir_buf, entries, ctx);
6026	if (ret)
6027	goto nopos;
6028
6029	ret = btrfs_readdir_delayed_dir_index(ctx, ins_list: &ins_list);
6030	if (ret)
6031	goto nopos;
6032
6033	/*
6034	* Stop new entries from being returned after we return the last
6035	* entry.
6036	*
6037	* New directory entries are assigned a strictly increasing
6038	* offset. This means that new entries created during readdir
6039	* are *guaranteed* to be seen in the future by that readdir.
6040	* This has broken buggy programs which operate on names as
6041	* they're returned by readdir. Until we re-use freed offsets
6042	* we have this hack to stop new entries from being returned
6043	* under the assumption that they'll never reach this huge
6044	* offset.
6045	*
6046	* This is being careful not to overflow 32bit loff_t unless the
6047	* last entry requires it because doing so has broken 32bit apps
6048	* in the past.
6049	*/
6050	if (ctx->pos >= INT_MAX)
6051	ctx->pos = LLONG_MAX;
6052	else
6053	ctx->pos = INT_MAX;
6054	nopos:
6055	ret = 0;
6056	err:
6057	if (put)
6058	btrfs_readdir_put_delayed_items(inode, ins_list: &ins_list, del_list: &del_list);
6059	btrfs_free_path(p: path);
6060	return ret;
6061	}
6062
6063	/*
6064	* This is somewhat expensive, updating the tree every time the
6065	* inode changes. But, it is most likely to find the inode in cache.
6066	* FIXME, needs more benchmarking...there are no reasons other than performance
6067	* to keep or drop this code.
6068	*/
6069	static int btrfs_dirty_inode(struct btrfs_inode *inode)
6070	{
6071	struct btrfs_root *root = inode->root;
6072	struct btrfs_fs_info *fs_info = root->fs_info;
6073	struct btrfs_trans_handle *trans;
6074	int ret;
6075
6076	if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6077	return 0;
6078
6079	trans = btrfs_join_transaction(root);
6080	if (IS_ERR(ptr: trans))
6081	return PTR_ERR(ptr: trans);
6082
6083	ret = btrfs_update_inode(trans, inode);
6084	if (ret == -ENOSPC || ret == -EDQUOT) {
6085	/* whoops, lets try again with the full transaction */
6086	btrfs_end_transaction(trans);
6087	trans = btrfs_start_transaction(root, num_items: 1);
6088	if (IS_ERR(ptr: trans))
6089	return PTR_ERR(ptr: trans);
6090
6091	ret = btrfs_update_inode(trans, inode);
6092	}
6093	btrfs_end_transaction(trans);
6094	if (inode->delayed_node)
6095	btrfs_balance_delayed_items(fs_info);
6096
6097	return ret;
6098	}
6099
6100	/*
6101	* This is a copy of file_update_time. We need this so we can return error on
6102	* ENOSPC for updating the inode in the case of file write and mmap writes.
6103	*/
6104	static int btrfs_update_time(struct inode *inode, int flags)
6105	{
6106	struct btrfs_root *root = BTRFS_I(inode)->root;
6107	bool dirty;
6108
6109	if (btrfs_root_readonly(root))
6110	return -EROFS;
6111
6112	dirty = inode_update_timestamps(inode, flags);
6113	return dirty ? btrfs_dirty_inode(inode: BTRFS_I(inode)) : 0;
6114	}
6115
6116	/*
6117	* helper to find a free sequence number in a given directory. This current
6118	* code is very simple, later versions will do smarter things in the btree
6119	*/
6120	int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6121	{
6122	int ret = 0;
6123
6124	if (dir->index_cnt == (u64)-1) {
6125	ret = btrfs_inode_delayed_dir_index_count(inode: dir);
6126	if (ret) {
6127	ret = btrfs_set_inode_index_count(inode: dir);
6128	if (ret)
6129	return ret;
6130	}
6131	}
6132
6133	*index = dir->index_cnt;
6134	dir->index_cnt++;
6135
6136	return ret;
6137	}
6138
6139	static int btrfs_insert_inode_locked(struct inode *inode)
6140	{
6141	struct btrfs_iget_args args;
6142
6143	args.ino = BTRFS_I(inode)->location.objectid;
6144	args.root = BTRFS_I(inode)->root;
6145
6146	return insert_inode_locked4(inode,
6147	btrfs_inode_hash(objectid: inode->i_ino, root: BTRFS_I(inode)->root),
6148	test: btrfs_find_actor, &args);
6149	}
6150
6151	int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6152	unsigned int *trans_num_items)
6153	{
6154	struct inode *dir = args->dir;
6155	struct inode *inode = args->inode;
6156	int ret;
6157
6158	if (!args->orphan) {
6159	ret = fscrypt_setup_filename(inode: dir, iname: &args->dentry->d_name, lookup: 0,
6160	fname: &args->fname);
6161	if (ret)
6162	return ret;
6163	}
6164
6165	ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6166	if (ret) {
6167	fscrypt_free_filename(fname: &args->fname);
6168	return ret;
6169	}
6170
6171	/* 1 to add inode item */
6172	*trans_num_items = 1;
6173	/* 1 to add compression property */
6174	if (BTRFS_I(inode: dir)->prop_compress)
6175	(*trans_num_items)++;
6176	/* 1 to add default ACL xattr */
6177	if (args->default_acl)
6178	(*trans_num_items)++;
6179	/* 1 to add access ACL xattr */
6180	if (args->acl)
6181	(*trans_num_items)++;
6182	#ifdef CONFIG_SECURITY
6183	/* 1 to add LSM xattr */
6184	if (dir->i_security)
6185	(*trans_num_items)++;
6186	#endif
6187	if (args->orphan) {
6188	/* 1 to add orphan item */
6189	(*trans_num_items)++;
6190	} else {
6191	/*
6192	* 1 to add dir item
6193	* 1 to add dir index
6194	* 1 to update parent inode item
6195	*
6196	* No need for 1 unit for the inode ref item because it is
6197	* inserted in a batch together with the inode item at
6198	* btrfs_create_new_inode().
6199	*/
6200	*trans_num_items += 3;
6201	}
6202	return 0;
6203	}
6204
6205	void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6206	{
6207	posix_acl_release(acl: args->acl);
6208	posix_acl_release(acl: args->default_acl);
6209	fscrypt_free_filename(fname: &args->fname);
6210	}
6211
6212	/*
6213	* Inherit flags from the parent inode.
6214	*
6215	* Currently only the compression flags and the cow flags are inherited.
6216	*/
6217	static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6218	{
6219	unsigned int flags;
6220
6221	flags = dir->flags;
6222
6223	if (flags & BTRFS_INODE_NOCOMPRESS) {
6224	inode->flags &= ~BTRFS_INODE_COMPRESS;
6225	inode->flags |= BTRFS_INODE_NOCOMPRESS;
6226	} else if (flags & BTRFS_INODE_COMPRESS) {
6227	inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6228	inode->flags |= BTRFS_INODE_COMPRESS;
6229	}
6230
6231	if (flags & BTRFS_INODE_NODATACOW) {
6232	inode->flags |= BTRFS_INODE_NODATACOW;
6233	if (S_ISREG(inode->vfs_inode.i_mode))
6234	inode->flags |= BTRFS_INODE_NODATASUM;
6235	}
6236
6237	btrfs_sync_inode_flags_to_i_flags(inode: &inode->vfs_inode);
6238	}
6239
6240	int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6241	struct btrfs_new_inode_args *args)
6242	{
6243	struct timespec64 ts;
6244	struct inode *dir = args->dir;
6245	struct inode *inode = args->inode;
6246	const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6247	struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
6248	struct btrfs_root *root;
6249	struct btrfs_inode_item *inode_item;
6250	struct btrfs_key *location;
6251	struct btrfs_path *path;
6252	u64 objectid;
6253	struct btrfs_inode_ref *ref;
6254	struct btrfs_key key[2];
6255	u32 sizes[2];
6256	struct btrfs_item_batch batch;
6257	unsigned long ptr;
6258	int ret;
6259
6260	path = btrfs_alloc_path();
6261	if (!path)
6262	return -ENOMEM;
6263
6264	if (!args->subvol)
6265	BTRFS_I(inode)->root = btrfs_grab_root(root: BTRFS_I(inode: dir)->root);
6266	root = BTRFS_I(inode)->root;
6267
6268	ret = btrfs_get_free_objectid(root, objectid: &objectid);
6269	if (ret)
6270	goto out;
6271	inode->i_ino = objectid;
6272
6273	if (args->orphan) {
6274	/*
6275	* O_TMPFILE, set link count to 0, so that after this point, we
6276	* fill in an inode item with the correct link count.
6277	*/
6278	set_nlink(inode, nlink: 0);
6279	} else {
6280	trace_btrfs_inode_request(inode: dir);
6281
6282	ret = btrfs_set_inode_index(dir: BTRFS_I(inode: dir), index: &BTRFS_I(inode)->dir_index);
6283	if (ret)
6284	goto out;
6285	}
6286	/* index_cnt is ignored for everything but a dir. */
6287	BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6288	BTRFS_I(inode)->generation = trans->transid;
6289	inode->i_generation = BTRFS_I(inode)->generation;
6290
6291	/*
6292	* We don't have any capability xattrs set here yet, shortcut any
6293	* queries for the xattrs here. If we add them later via the inode
6294	* security init path or any other path this flag will be cleared.
6295	*/
6296	set_bit(nr: BTRFS_INODE_NO_CAP_XATTR, addr: &BTRFS_I(inode)->runtime_flags);
6297
6298	/*
6299	* Subvolumes don't inherit flags from their parent directory.
6300	* Originally this was probably by accident, but we probably can't
6301	* change it now without compatibility issues.
6302	*/
6303	if (!args->subvol)
6304	btrfs_inherit_iflags(inode: BTRFS_I(inode), dir: BTRFS_I(inode: dir));
6305
6306	if (S_ISREG(inode->i_mode)) {
6307	if (btrfs_test_opt(fs_info, NODATASUM))
6308	BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6309	if (btrfs_test_opt(fs_info, NODATACOW))
6310	BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6311	BTRFS_INODE_NODATASUM;
6312	}
6313
6314	location = &BTRFS_I(inode)->location;
6315	location->objectid = objectid;
6316	location->offset = 0;
6317	location->type = BTRFS_INODE_ITEM_KEY;
6318
6319	ret = btrfs_insert_inode_locked(inode);
6320	if (ret < 0) {
6321	if (!args->orphan)
6322	BTRFS_I(inode: dir)->index_cnt--;
6323	goto out;
6324	}
6325
6326	/*
6327	* We could have gotten an inode number from somebody who was fsynced
6328	* and then removed in this same transaction, so let's just set full
6329	* sync since it will be a full sync anyway and this will blow away the
6330	* old info in the log.
6331	*/
6332	btrfs_set_inode_full_sync(inode: BTRFS_I(inode));
6333
6334	key[0].objectid = objectid;
6335	key[0].type = BTRFS_INODE_ITEM_KEY;
6336	key[0].offset = 0;
6337
6338	sizes[0] = sizeof(struct btrfs_inode_item);
6339
6340	if (!args->orphan) {
6341	/*
6342	* Start new inodes with an inode_ref. This is slightly more
6343	* efficient for small numbers of hard links since they will
6344	* be packed into one item. Extended refs will kick in if we
6345	* add more hard links than can fit in the ref item.
6346	*/
6347	key[1].objectid = objectid;
6348	key[1].type = BTRFS_INODE_REF_KEY;
6349	if (args->subvol) {
6350	key[1].offset = objectid;
6351	sizes[1] = 2 + sizeof(*ref);
6352	} else {
6353	key[1].offset = btrfs_ino(inode: BTRFS_I(inode: dir));
6354	sizes[1] = name->len + sizeof(*ref);
6355	}
6356	}
6357
6358	batch.keys = &key[0];
6359	batch.data_sizes = &sizes[0];
6360	batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6361	batch.nr = args->orphan ? 1 : 2;
6362	ret = btrfs_insert_empty_items(trans, root, path, batch: &batch);
6363	if (ret != 0) {
6364	btrfs_abort_transaction(trans, ret);
6365	goto discard;
6366	}
6367
6368	ts = simple_inode_init_ts(inode);
6369	BTRFS_I(inode)->i_otime_sec = ts.tv_sec;
6370	BTRFS_I(inode)->i_otime_nsec = ts.tv_nsec;
6371
6372	/*
6373	* We're going to fill the inode item now, so at this point the inode
6374	* must be fully initialized.
6375	*/
6376
6377	inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6378	struct btrfs_inode_item);
6379	memzero_extent_buffer(eb: path->nodes[0], start: (unsigned long)inode_item,
6380	len: sizeof(*inode_item));
6381	fill_inode_item(trans, leaf: path->nodes[0], item: inode_item, inode);
6382
6383	if (!args->orphan) {
6384	ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6385	struct btrfs_inode_ref);
6386	ptr = (unsigned long)(ref + 1);
6387	if (args->subvol) {
6388	btrfs_set_inode_ref_name_len(eb: path->nodes[0], s: ref, val: 2);
6389	btrfs_set_inode_ref_index(eb: path->nodes[0], s: ref, val: 0);
6390	write_extent_buffer(eb: path->nodes[0], src: "..", start: ptr, len: 2);
6391	} else {
6392	btrfs_set_inode_ref_name_len(eb: path->nodes[0], s: ref,
6393	val: name->len);
6394	btrfs_set_inode_ref_index(eb: path->nodes[0], s: ref,
6395	val: BTRFS_I(inode)->dir_index);
6396	write_extent_buffer(eb: path->nodes[0], src: name->name, start: ptr,
6397	len: name->len);
6398	}
6399	}
6400
6401	btrfs_mark_buffer_dirty(trans, buf: path->nodes[0]);
6402	/*
6403	* We don't need the path anymore, plus inheriting properties, adding
6404	* ACLs, security xattrs, orphan item or adding the link, will result in
6405	* allocating yet another path. So just free our path.
6406	*/
6407	btrfs_free_path(p: path);
6408	path = NULL;
6409
6410	if (args->subvol) {
6411	struct inode *parent;
6412
6413	/*
6414	* Subvolumes inherit properties from their parent subvolume,
6415	* not the directory they were created in.
6416	*/
6417	parent = btrfs_iget(s: fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6418	root: BTRFS_I(inode: dir)->root);
6419	if (IS_ERR(ptr: parent)) {
6420	ret = PTR_ERR(ptr: parent);
6421	} else {
6422	ret = btrfs_inode_inherit_props(trans, inode, dir: parent);
6423	iput(parent);
6424	}
6425	} else {
6426	ret = btrfs_inode_inherit_props(trans, inode, dir);
6427	}
6428	if (ret) {
6429	btrfs_err(fs_info,
6430	"error inheriting props for ino %llu (root %llu): %d",
6431	btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6432	ret);
6433	}
6434
6435	/*
6436	* Subvolumes don't inherit ACLs or get passed to the LSM. This is
6437	* probably a bug.
6438	*/
6439	if (!args->subvol) {
6440	ret = btrfs_init_inode_security(trans, args);
6441	if (ret) {
6442	btrfs_abort_transaction(trans, ret);
6443	goto discard;
6444	}
6445	}
6446
6447	inode_tree_add(inode: BTRFS_I(inode));
6448
6449	trace_btrfs_inode_new(inode);
6450	btrfs_set_inode_last_trans(trans, inode: BTRFS_I(inode));
6451
6452	btrfs_update_root_times(trans, root);
6453
6454	if (args->orphan) {
6455	ret = btrfs_orphan_add(trans, inode: BTRFS_I(inode));
6456	} else {
6457	ret = btrfs_add_link(trans, parent_inode: BTRFS_I(inode: dir), inode: BTRFS_I(inode), name,
6458	add_backref: 0, index: BTRFS_I(inode)->dir_index);
6459	}
6460	if (ret) {
6461	btrfs_abort_transaction(trans, ret);
6462	goto discard;
6463	}
6464
6465	return 0;
6466
6467	discard:
6468	/*
6469	* discard_new_inode() calls iput(), but the caller owns the reference
6470	* to the inode.
6471	*/
6472	ihold(inode);
6473	discard_new_inode(inode);
6474	out:
6475	btrfs_free_path(p: path);
6476	return ret;
6477	}
6478
6479	/*
6480	* utility function to add 'inode' into 'parent_inode' with
6481	* a give name and a given sequence number.
6482	* if 'add_backref' is true, also insert a backref from the
6483	* inode to the parent directory.
6484	*/
6485	int btrfs_add_link(struct btrfs_trans_handle *trans,
6486	struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6487	const struct fscrypt_str *name, int add_backref, u64 index)
6488	{
6489	int ret = 0;
6490	struct btrfs_key key;
6491	struct btrfs_root *root = parent_inode->root;
6492	u64 ino = btrfs_ino(inode);
6493	u64 parent_ino = btrfs_ino(inode: parent_inode);
6494
6495	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6496	memcpy(&key, &inode->root->root_key, sizeof(key));
6497	} else {
6498	key.objectid = ino;
6499	key.type = BTRFS_INODE_ITEM_KEY;
6500	key.offset = 0;
6501	}
6502
6503	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6504	ret = btrfs_add_root_ref(trans, root_id: key.objectid,
6505	ref_id: root->root_key.objectid, dirid: parent_ino,
6506	sequence: index, name);
6507	} else if (add_backref) {
6508	ret = btrfs_insert_inode_ref(trans, root, name,
6509	inode_objectid: ino, ref_objectid: parent_ino, index);
6510	}
6511
6512	/* Nothing to clean up yet */
6513	if (ret)
6514	return ret;
6515
6516	ret = btrfs_insert_dir_item(trans, name, dir: parent_inode, location: &key,
6517	type: btrfs_inode_type(inode: &inode->vfs_inode), index);
6518	if (ret == -EEXIST || ret == -EOVERFLOW)
6519	goto fail_dir_item;
6520	else if (ret) {
6521	btrfs_abort_transaction(trans, ret);
6522	return ret;
6523	}
6524
6525	btrfs_i_size_write(inode: parent_inode, size: parent_inode->vfs_inode.i_size +
6526	name->len * 2);
6527	inode_inc_iversion(inode: &parent_inode->vfs_inode);
6528	/*
6529	* If we are replaying a log tree, we do not want to update the mtime
6530	* and ctime of the parent directory with the current time, since the
6531	* log replay procedure is responsible for setting them to their correct
6532	* values (the ones it had when the fsync was done).
6533	*/
6534	if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
6535	inode_set_mtime_to_ts(inode: &parent_inode->vfs_inode,
6536	ts: inode_set_ctime_current(inode: &parent_inode->vfs_inode));
6537
6538	ret = btrfs_update_inode(trans, inode: parent_inode);
6539	if (ret)
6540	btrfs_abort_transaction(trans, ret);
6541	return ret;
6542
6543	fail_dir_item:
6544	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6545	u64 local_index;
6546	int err;
6547	err = btrfs_del_root_ref(trans, root_id: key.objectid,
6548	ref_id: root->root_key.objectid, dirid: parent_ino,
6549	sequence: &local_index, name);
6550	if (err)
6551	btrfs_abort_transaction(trans, err);
6552	} else if (add_backref) {
6553	u64 local_index;
6554	int err;
6555
6556	err = btrfs_del_inode_ref(trans, root, name, inode_objectid: ino, ref_objectid: parent_ino,
6557	index: &local_index);
6558	if (err)
6559	btrfs_abort_transaction(trans, err);
6560	}
6561
6562	/* Return the original error code */
6563	return ret;
6564	}
6565
6566	static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6567	struct inode *inode)
6568	{
6569	struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
6570	struct btrfs_root *root = BTRFS_I(inode: dir)->root;
6571	struct btrfs_new_inode_args new_inode_args = {
6572	.dir = dir,
6573	.dentry = dentry,
6574	.inode = inode,
6575	};
6576	unsigned int trans_num_items;
6577	struct btrfs_trans_handle *trans;
6578	int err;
6579
6580	err = btrfs_new_inode_prepare(args: &new_inode_args, trans_num_items: &trans_num_items);
6581	if (err)
6582	goto out_inode;
6583
6584	trans = btrfs_start_transaction(root, num_items: trans_num_items);
6585	if (IS_ERR(ptr: trans)) {
6586	err = PTR_ERR(ptr: trans);
6587	goto out_new_inode_args;
6588	}
6589
6590	err = btrfs_create_new_inode(trans, args: &new_inode_args);
6591	if (!err)
6592	d_instantiate_new(dentry, inode);
6593
6594	btrfs_end_transaction(trans);
6595	btrfs_btree_balance_dirty(fs_info);
6596	out_new_inode_args:
6597	btrfs_new_inode_args_destroy(args: &new_inode_args);
6598	out_inode:
6599	if (err)
6600	iput(inode);
6601	return err;
6602	}
6603
6604	static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6605	struct dentry *dentry, umode_t mode, dev_t rdev)
6606	{
6607	struct inode *inode;
6608
6609	inode = new_inode(sb: dir->i_sb);
6610	if (!inode)
6611	return -ENOMEM;
6612	inode_init_owner(idmap, inode, dir, mode);
6613	inode->i_op = &btrfs_special_inode_operations;
6614	init_special_inode(inode, inode->i_mode, rdev);
6615	return btrfs_create_common(dir, dentry, inode);
6616	}
6617
6618	static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6619	struct dentry *dentry, umode_t mode, bool excl)
6620	{
6621	struct inode *inode;
6622
6623	inode = new_inode(sb: dir->i_sb);
6624	if (!inode)
6625	return -ENOMEM;
6626	inode_init_owner(idmap, inode, dir, mode);
6627	inode->i_fop = &btrfs_file_operations;
6628	inode->i_op = &btrfs_file_inode_operations;
6629	inode->i_mapping->a_ops = &btrfs_aops;
6630	return btrfs_create_common(dir, dentry, inode);
6631	}
6632
6633	static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6634	struct dentry *dentry)
6635	{
6636	struct btrfs_trans_handle *trans = NULL;
6637	struct btrfs_root *root = BTRFS_I(inode: dir)->root;
6638	struct inode *inode = d_inode(dentry: old_dentry);
6639	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
6640	struct fscrypt_name fname;
6641	u64 index;
6642	int err;
6643	int drop_inode = 0;
6644
6645	/* do not allow sys_link's with other subvols of the same device */
6646	if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6647	return -EXDEV;
6648
6649	if (inode->i_nlink >= BTRFS_LINK_MAX)
6650	return -EMLINK;
6651
6652	err = fscrypt_setup_filename(inode: dir, iname: &dentry->d_name, lookup: 0, fname: &fname);
6653	if (err)
6654	goto fail;
6655
6656	err = btrfs_set_inode_index(dir: BTRFS_I(inode: dir), index: &index);
6657	if (err)
6658	goto fail;
6659
6660	/*
6661	* 2 items for inode and inode ref
6662	* 2 items for dir items
6663	* 1 item for parent inode
6664	* 1 item for orphan item deletion if O_TMPFILE
6665	*/
6666	trans = btrfs_start_transaction(root, num_items: inode->i_nlink ? 5 : 6);
6667	if (IS_ERR(ptr: trans)) {
6668	err = PTR_ERR(ptr: trans);
6669	trans = NULL;
6670	goto fail;
6671	}
6672
6673	/* There are several dir indexes for this inode, clear the cache. */
6674	BTRFS_I(inode)->dir_index = 0ULL;
6675	inc_nlink(inode);
6676	inode_inc_iversion(inode);
6677	inode_set_ctime_current(inode);
6678	ihold(inode);
6679	set_bit(nr: BTRFS_INODE_COPY_EVERYTHING, addr: &BTRFS_I(inode)->runtime_flags);
6680
6681	err = btrfs_add_link(trans, parent_inode: BTRFS_I(inode: dir), inode: BTRFS_I(inode),
6682	name: &fname.disk_name, add_backref: 1, index);
6683
6684	if (err) {
6685	drop_inode = 1;
6686	} else {
6687	struct dentry *parent = dentry->d_parent;
6688
6689	err = btrfs_update_inode(trans, inode: BTRFS_I(inode));
6690	if (err)
6691	goto fail;
6692	if (inode->i_nlink == 1) {
6693	/*
6694	* If new hard link count is 1, it's a file created
6695	* with open(2) O_TMPFILE flag.
6696	*/
6697	err = btrfs_orphan_del(trans, inode: BTRFS_I(inode));
6698	if (err)
6699	goto fail;
6700	}
6701	d_instantiate(dentry, inode);
6702	btrfs_log_new_name(trans, old_dentry, NULL, old_dir_index: 0, parent);
6703	}
6704
6705	fail:
6706	fscrypt_free_filename(fname: &fname);
6707	if (trans)
6708	btrfs_end_transaction(trans);
6709	if (drop_inode) {
6710	inode_dec_link_count(inode);
6711	iput(inode);
6712	}
6713	btrfs_btree_balance_dirty(fs_info);
6714	return err;
6715	}
6716
6717	static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6718	struct dentry *dentry, umode_t mode)
6719	{
6720	struct inode *inode;
6721
6722	inode = new_inode(sb: dir->i_sb);
6723	if (!inode)
6724	return -ENOMEM;
6725	inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6726	inode->i_op = &btrfs_dir_inode_operations;
6727	inode->i_fop = &btrfs_dir_file_operations;
6728	return btrfs_create_common(dir, dentry, inode);
6729	}
6730
6731	static noinline int uncompress_inline(struct btrfs_path *path,
6732	struct page *page,
6733	struct btrfs_file_extent_item *item)
6734	{
6735	int ret;
6736	struct extent_buffer *leaf = path->nodes[0];
6737	char *tmp;
6738	size_t max_size;
6739	unsigned long inline_size;
6740	unsigned long ptr;
6741	int compress_type;
6742
6743	compress_type = btrfs_file_extent_compression(eb: leaf, s: item);
6744	max_size = btrfs_file_extent_ram_bytes(eb: leaf, s: item);
6745	inline_size = btrfs_file_extent_inline_item_len(eb: leaf, nr: path->slots[0]);
6746	tmp = kmalloc(size: inline_size, GFP_NOFS);
6747	if (!tmp)
6748	return -ENOMEM;
6749	ptr = btrfs_file_extent_inline_start(e: item);
6750
6751	read_extent_buffer(eb: leaf, dst: tmp, start: ptr, len: inline_size);
6752
6753	max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6754	ret = btrfs_decompress(type: compress_type, data_in: tmp, dest_page: page, start_byte: 0, srclen: inline_size, destlen: max_size);
6755
6756	/*
6757	* decompression code contains a memset to fill in any space between the end
6758	* of the uncompressed data and the end of max_size in case the decompressed
6759	* data ends up shorter than ram_bytes. That doesn't cover the hole between
6760	* the end of an inline extent and the beginning of the next block, so we
6761	* cover that region here.
6762	*/
6763
6764	if (max_size < PAGE_SIZE)
6765	memzero_page(page, offset: max_size, PAGE_SIZE - max_size);
6766	kfree(objp: tmp);
6767	return ret;
6768	}
6769
6770	static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6771	struct page *page)
6772	{
6773	struct btrfs_file_extent_item *fi;
6774	void *kaddr;
6775	size_t copy_size;
6776
6777	if (!page || PageUptodate(page))
6778	return 0;
6779
6780	ASSERT(page_offset(page) == 0);
6781
6782	fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6783	struct btrfs_file_extent_item);
6784	if (btrfs_file_extent_compression(eb: path->nodes[0], s: fi) != BTRFS_COMPRESS_NONE)
6785	return uncompress_inline(path, page, item: fi);
6786
6787	copy_size = min_t(u64, PAGE_SIZE,
6788	btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6789	kaddr = kmap_local_page(page);
6790	read_extent_buffer(eb: path->nodes[0], dst: kaddr,
6791	start: btrfs_file_extent_inline_start(e: fi), len: copy_size);
6792	kunmap_local(kaddr);
6793	if (copy_size < PAGE_SIZE)
6794	memzero_page(page, offset: copy_size, PAGE_SIZE - copy_size);
6795	return 0;
6796	}
6797
6798	/*
6799	* Lookup the first extent overlapping a range in a file.
6800	*
6801	* @inode: file to search in
6802	* @page: page to read extent data into if the extent is inline
6803	* @start: file offset
6804	* @len: length of range starting at @start
6805	*
6806	* Return the first &struct extent_map which overlaps the given range, reading
6807	* it from the B-tree and caching it if necessary. Note that there may be more
6808	* extents which overlap the given range after the returned extent_map.
6809	*
6810	* If @page is not NULL and the extent is inline, this also reads the extent
6811	* data directly into the page and marks the extent up to date in the io_tree.
6812	*
6813	* Return: ERR_PTR on error, non-NULL extent_map on success.
6814	*/
6815	struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6816	struct page *page, u64 start, u64 len)
6817	{
6818	struct btrfs_fs_info *fs_info = inode->root->fs_info;
6819	int ret = 0;
6820	u64 extent_start = 0;
6821	u64 extent_end = 0;
6822	u64 objectid = btrfs_ino(inode);
6823	int extent_type = -1;
6824	struct btrfs_path *path = NULL;
6825	struct btrfs_root *root = inode->root;
6826	struct btrfs_file_extent_item *item;
6827	struct extent_buffer *leaf;
6828	struct btrfs_key found_key;
6829	struct extent_map *em = NULL;
6830	struct extent_map_tree *em_tree = &inode->extent_tree;
6831
6832	read_lock(&em_tree->lock);
6833	em = lookup_extent_mapping(tree: em_tree, start, len);
6834	read_unlock(&em_tree->lock);
6835
6836	if (em) {
6837	if (em->start > start || em->start + em->len <= start)
6838	free_extent_map(em);
6839	else if (em->block_start == EXTENT_MAP_INLINE && page)
6840	free_extent_map(em);
6841	else
6842	goto out;
6843	}
6844	em = alloc_extent_map();
6845	if (!em) {
6846	ret = -ENOMEM;
6847	goto out;
6848	}
6849	em->start = EXTENT_MAP_HOLE;
6850	em->orig_start = EXTENT_MAP_HOLE;
6851	em->len = (u64)-1;
6852	em->block_len = (u64)-1;
6853
6854	path = btrfs_alloc_path();
6855	if (!path) {
6856	ret = -ENOMEM;
6857	goto out;
6858	}
6859
6860	/* Chances are we'll be called again, so go ahead and do readahead */
6861	path->reada = READA_FORWARD;
6862
6863	/*
6864	* The same explanation in load_free_space_cache applies here as well,
6865	* we only read when we're loading the free space cache, and at that
6866	* point the commit_root has everything we need.
6867	*/
6868	if (btrfs_is_free_space_inode(inode)) {
6869	path->search_commit_root = 1;
6870	path->skip_locking = 1;
6871	}
6872
6873	ret = btrfs_lookup_file_extent(NULL, root, path, objectid, bytenr: start, mod: 0);
6874	if (ret < 0) {
6875	goto out;
6876	} else if (ret > 0) {
6877	if (path->slots[0] == 0)
6878	goto not_found;
6879	path->slots[0]--;
6880	ret = 0;
6881	}
6882
6883	leaf = path->nodes[0];
6884	item = btrfs_item_ptr(leaf, path->slots[0],
6885	struct btrfs_file_extent_item);
6886	btrfs_item_key_to_cpu(eb: leaf, cpu_key: &found_key, nr: path->slots[0]);
6887	if (found_key.objectid != objectid ||
6888	found_key.type != BTRFS_EXTENT_DATA_KEY) {
6889	/*
6890	* If we backup past the first extent we want to move forward
6891	* and see if there is an extent in front of us, otherwise we'll
6892	* say there is a hole for our whole search range which can
6893	* cause problems.
6894	*/
6895	extent_end = start;
6896	goto next;
6897	}
6898
6899	extent_type = btrfs_file_extent_type(eb: leaf, s: item);
6900	extent_start = found_key.offset;
6901	extent_end = btrfs_file_extent_end(path);
6902	if (extent_type == BTRFS_FILE_EXTENT_REG ||
6903	extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6904	/* Only regular file could have regular/prealloc extent */
6905	if (!S_ISREG(inode->vfs_inode.i_mode)) {
6906	ret = -EUCLEAN;
6907	btrfs_crit(fs_info,
6908	"regular/prealloc extent found for non-regular inode %llu",
6909	btrfs_ino(inode));
6910	goto out;
6911	}
6912	trace_btrfs_get_extent_show_fi_regular(bi: inode, l: leaf, fi: item,
6913	start: extent_start);
6914	} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6915	trace_btrfs_get_extent_show_fi_inline(bi: inode, l: leaf, fi: item,
6916	slot: path->slots[0],
6917	start: extent_start);
6918	}
6919	next:
6920	if (start >= extent_end) {
6921	path->slots[0]++;
6922	if (path->slots[0] >= btrfs_header_nritems(eb: leaf)) {
6923	ret = btrfs_next_leaf(root, path);
6924	if (ret < 0)
6925	goto out;
6926	else if (ret > 0)
6927	goto not_found;
6928
6929	leaf = path->nodes[0];
6930	}
6931	btrfs_item_key_to_cpu(eb: leaf, cpu_key: &found_key, nr: path->slots[0]);
6932	if (found_key.objectid != objectid ||
6933	found_key.type != BTRFS_EXTENT_DATA_KEY)
6934	goto not_found;
6935	if (start + len <= found_key.offset)
6936	goto not_found;
6937	if (start > found_key.offset)
6938	goto next;
6939
6940	/* New extent overlaps with existing one */
6941	em->start = start;
6942	em->orig_start = start;
6943	em->len = found_key.offset - start;
6944	em->block_start = EXTENT_MAP_HOLE;
6945	goto insert;
6946	}
6947
6948	btrfs_extent_item_to_extent_map(inode, path, fi: item, em);
6949
6950	if (extent_type == BTRFS_FILE_EXTENT_REG ||
6951	extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6952	goto insert;
6953	} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6954	/*
6955	* Inline extent can only exist at file offset 0. This is
6956	* ensured by tree-checker and inline extent creation path.
6957	* Thus all members representing file offsets should be zero.
6958	*/
6959	ASSERT(extent_start == 0);
6960	ASSERT(em->start == 0);
6961
6962	/*
6963	* btrfs_extent_item_to_extent_map() should have properly
6964	* initialized em members already.
6965	*
6966	* Other members are not utilized for inline extents.
6967	*/
6968	ASSERT(em->block_start == EXTENT_MAP_INLINE);
6969	ASSERT(em->len == fs_info->sectorsize);
6970
6971	ret = read_inline_extent(inode, path, page);
6972	if (ret < 0)
6973	goto out;
6974	goto insert;
6975	}
6976	not_found:
6977	em->start = start;
6978	em->orig_start = start;
6979	em->len = len;
6980	em->block_start = EXTENT_MAP_HOLE;
6981	insert:
6982	ret = 0;
6983	btrfs_release_path(p: path);
6984	if (em->start > start || extent_map_end(em) <= start) {
6985	btrfs_err(fs_info,
6986	"bad extent! em: [%llu %llu] passed [%llu %llu]",
6987	em->start, em->len, start, len);
6988	ret = -EIO;
6989	goto out;
6990	}
6991
6992	write_lock(&em_tree->lock);
6993	ret = btrfs_add_extent_mapping(fs_info, em_tree, em_in: &em, start, len);
6994	write_unlock(&em_tree->lock);
6995	out:
6996	btrfs_free_path(p: path);
6997
6998	trace_btrfs_get_extent(root, inode, map: em);
6999
7000	if (ret) {
7001	free_extent_map(em);
7002	return ERR_PTR(error: ret);
7003	}
7004	return em;
7005	}
7006
7007	static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7008	struct btrfs_dio_data *dio_data,
7009	const u64 start,
7010	const u64 len,
7011	const u64 orig_start,
7012	const u64 block_start,
7013	const u64 block_len,
7014	const u64 orig_block_len,
7015	const u64 ram_bytes,
7016	const int type)
7017	{
7018	struct extent_map *em = NULL;
7019	struct btrfs_ordered_extent *ordered;
7020
7021	if (type != BTRFS_ORDERED_NOCOW) {
7022	em = create_io_em(inode, start, len, orig_start, block_start,
7023	block_len, orig_block_len, ram_bytes,
7024	compress_type: BTRFS_COMPRESS_NONE, /* compress_type */
7025	type);
7026	if (IS_ERR(ptr: em))
7027	goto out;
7028	}
7029	ordered = btrfs_alloc_ordered_extent(inode, file_offset: start, num_bytes: len, ram_bytes: len,
7030	disk_bytenr: block_start, disk_num_bytes: block_len, offset: 0,
7031	flags: (1 << type) |
7032	(1 << BTRFS_ORDERED_DIRECT),
7033	compress_type: BTRFS_COMPRESS_NONE);
7034	if (IS_ERR(ptr: ordered)) {
7035	if (em) {
7036	free_extent_map(em);
7037	btrfs_drop_extent_map_range(inode, start,
7038	end: start + len - 1, skip_pinned: false);
7039	}
7040	em = ERR_CAST(ptr: ordered);
7041	} else {
7042	ASSERT(!dio_data->ordered);
7043	dio_data->ordered = ordered;
7044	}
7045	out:
7046
7047	return em;
7048	}
7049
7050	static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7051	struct btrfs_dio_data *dio_data,
7052	u64 start, u64 len)
7053	{
7054	struct btrfs_root *root = inode->root;
7055	struct btrfs_fs_info *fs_info = root->fs_info;
7056	struct extent_map *em;
7057	struct btrfs_key ins;
7058	u64 alloc_hint;
7059	int ret;
7060
7061	alloc_hint = get_extent_allocation_hint(inode, start, num_bytes: len);
7062	again:
7063	ret = btrfs_reserve_extent(root, ram_bytes: len, num_bytes: len, min_alloc_size: fs_info->sectorsize,
7064	empty_size: 0, hint_byte: alloc_hint, ins: &ins, is_data: 1, delalloc: 1);
7065	if (ret == -EAGAIN) {
7066	ASSERT(btrfs_is_zoned(fs_info));
7067	wait_on_bit_io(word: &inode->root->fs_info->flags, bit: BTRFS_FS_NEED_ZONE_FINISH,
7068	TASK_UNINTERRUPTIBLE);
7069	goto again;
7070	}
7071	if (ret)
7072	return ERR_PTR(error: ret);
7073
7074	em = btrfs_create_dio_extent(inode, dio_data, start, len: ins.offset, orig_start: start,
7075	block_start: ins.objectid, block_len: ins.offset, orig_block_len: ins.offset,
7076	ram_bytes: ins.offset, type: BTRFS_ORDERED_REGULAR);
7077	btrfs_dec_block_group_reservations(fs_info, start: ins.objectid);
7078	if (IS_ERR(ptr: em))
7079	btrfs_free_reserved_extent(fs_info, start: ins.objectid, len: ins.offset,
7080	delalloc: 1);
7081
7082	return em;
7083	}
7084
7085	static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7086	{
7087	struct btrfs_block_group *block_group;
7088	bool readonly = false;
7089
7090	block_group = btrfs_lookup_block_group(info: fs_info, bytenr);
7091	if (!block_group || block_group->ro)
7092	readonly = true;
7093	if (block_group)
7094	btrfs_put_block_group(cache: block_group);
7095	return readonly;
7096	}
7097
7098	/*
7099	* Check if we can do nocow write into the range [@offset, @offset + @len)
7100	*
7101	* @offset: File offset
7102	* @len: The length to write, will be updated to the nocow writeable
7103	* range
7104	* @orig_start: (optional) Return the original file offset of the file extent
7105	* @orig_len: (optional) Return the original on-disk length of the file extent
7106	* @ram_bytes: (optional) Return the ram_bytes of the file extent
7107	* @strict: if true, omit optimizations that might force us into unnecessary
7108	* cow. e.g., don't trust generation number.
7109	*
7110	* Return:
7111	* >0 and update @len if we can do nocow write
7112	* 0 if we can't do nocow write
7113	* <0 if error happened
7114	*
7115	* NOTE: This only checks the file extents, caller is responsible to wait for
7116	* any ordered extents.
7117	*/
7118	noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7119	u64 *orig_start, u64 *orig_block_len,
7120	u64 *ram_bytes, bool nowait, bool strict)
7121	{
7122	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
7123	struct can_nocow_file_extent_args nocow_args = { 0 };
7124	struct btrfs_path *path;
7125	int ret;
7126	struct extent_buffer *leaf;
7127	struct btrfs_root *root = BTRFS_I(inode)->root;
7128	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7129	struct btrfs_file_extent_item *fi;
7130	struct btrfs_key key;
7131	int found_type;
7132
7133	path = btrfs_alloc_path();
7134	if (!path)
7135	return -ENOMEM;
7136	path->nowait = nowait;
7137
7138	ret = btrfs_lookup_file_extent(NULL, root, path,
7139	objectid: btrfs_ino(inode: BTRFS_I(inode)), bytenr: offset, mod: 0);
7140	if (ret < 0)
7141	goto out;
7142
7143	if (ret == 1) {
7144	if (path->slots[0] == 0) {
7145	/* can't find the item, must cow */
7146	ret = 0;
7147	goto out;
7148	}
7149	path->slots[0]--;
7150	}
7151	ret = 0;
7152	leaf = path->nodes[0];
7153	btrfs_item_key_to_cpu(eb: leaf, cpu_key: &key, nr: path->slots[0]);
7154	if (key.objectid != btrfs_ino(inode: BTRFS_I(inode)) ||
7155	key.type != BTRFS_EXTENT_DATA_KEY) {
7156	/* not our file or wrong item type, must cow */
7157	goto out;
7158	}
7159
7160	if (key.offset > offset) {
7161	/* Wrong offset, must cow */
7162	goto out;
7163	}
7164
7165	if (btrfs_file_extent_end(path) <= offset)
7166	goto out;
7167
7168	fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7169	found_type = btrfs_file_extent_type(eb: leaf, s: fi);
7170	if (ram_bytes)
7171	*ram_bytes = btrfs_file_extent_ram_bytes(eb: leaf, s: fi);
7172
7173	nocow_args.start = offset;
7174	nocow_args.end = offset + *len - 1;
7175	nocow_args.strict = strict;
7176	nocow_args.free_path = true;
7177
7178	ret = can_nocow_file_extent(path, key: &key, inode: BTRFS_I(inode), args: &nocow_args);
7179	/* can_nocow_file_extent() has freed the path. */
7180	path = NULL;
7181
7182	if (ret != 1) {
7183	/* Treat errors as not being able to NOCOW. */
7184	ret = 0;
7185	goto out;
7186	}
7187
7188	ret = 0;
7189	if (btrfs_extent_readonly(fs_info, bytenr: nocow_args.disk_bytenr))
7190	goto out;
7191
7192	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7193	found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7194	u64 range_end;
7195
7196	range_end = round_up(offset + nocow_args.num_bytes,
7197	root->fs_info->sectorsize) - 1;
7198	ret = test_range_bit_exists(tree: io_tree, start: offset, end: range_end, bit: EXTENT_DELALLOC);
7199	if (ret) {
7200	ret = -EAGAIN;
7201	goto out;
7202	}
7203	}
7204
7205	if (orig_start)
7206	*orig_start = key.offset - nocow_args.extent_offset;
7207	if (orig_block_len)
7208	*orig_block_len = nocow_args.disk_num_bytes;
7209
7210	*len = nocow_args.num_bytes;
7211	ret = 1;
7212	out:
7213	btrfs_free_path(p: path);
7214	return ret;
7215	}
7216
7217	static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7218	struct extent_state **cached_state,
7219	unsigned int iomap_flags)
7220	{
7221	const bool writing = (iomap_flags & IOMAP_WRITE);
7222	const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7223	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7224	struct btrfs_ordered_extent *ordered;
7225	int ret = 0;
7226
7227	while (1) {
7228	if (nowait) {
7229	if (!try_lock_extent(tree: io_tree, start: lockstart, end: lockend,
7230	cached: cached_state))
7231	return -EAGAIN;
7232	} else {
7233	lock_extent(tree: io_tree, start: lockstart, end: lockend, cached: cached_state);
7234	}
7235	/*
7236	* We're concerned with the entire range that we're going to be
7237	* doing DIO to, so we need to make sure there's no ordered
7238	* extents in this range.
7239	*/
7240	ordered = btrfs_lookup_ordered_range(inode: BTRFS_I(inode), file_offset: lockstart,
7241	len: lockend - lockstart + 1);
7242
7243	/*
7244	* We need to make sure there are no buffered pages in this
7245	* range either, we could have raced between the invalidate in
7246	* generic_file_direct_write and locking the extent. The
7247	* invalidate needs to happen so that reads after a write do not
7248	* get stale data.
7249	*/
7250	if (!ordered &&
7251	(!writing || !filemap_range_has_page(inode->i_mapping,
7252	lstart: lockstart, lend: lockend)))
7253	break;
7254
7255	unlock_extent(tree: io_tree, start: lockstart, end: lockend, cached: cached_state);
7256
7257	if (ordered) {
7258	if (nowait) {
7259	btrfs_put_ordered_extent(entry: ordered);
7260	ret = -EAGAIN;
7261	break;
7262	}
7263	/*
7264	* If we are doing a DIO read and the ordered extent we
7265	* found is for a buffered write, we can not wait for it
7266	* to complete and retry, because if we do so we can
7267	* deadlock with concurrent buffered writes on page
7268	* locks. This happens only if our DIO read covers more
7269	* than one extent map, if at this point has already
7270	* created an ordered extent for a previous extent map
7271	* and locked its range in the inode's io tree, and a
7272	* concurrent write against that previous extent map's
7273	* range and this range started (we unlock the ranges
7274	* in the io tree only when the bios complete and
7275	* buffered writes always lock pages before attempting
7276	* to lock range in the io tree).
7277	*/
7278	if (writing ||
7279	test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7280	btrfs_start_ordered_extent(entry: ordered);
7281	else
7282	ret = nowait ? -EAGAIN : -ENOTBLK;
7283	btrfs_put_ordered_extent(entry: ordered);
7284	} else {
7285	/*
7286	* We could trigger writeback for this range (and wait
7287	* for it to complete) and then invalidate the pages for
7288	* this range (through invalidate_inode_pages2_range()),
7289	* but that can lead us to a deadlock with a concurrent
7290	* call to readahead (a buffered read or a defrag call
7291	* triggered a readahead) on a page lock due to an
7292	* ordered dio extent we created before but did not have
7293	* yet a corresponding bio submitted (whence it can not
7294	* complete), which makes readahead wait for that
7295	* ordered extent to complete while holding a lock on
7296	* that page.
7297	*/
7298	ret = nowait ? -EAGAIN : -ENOTBLK;
7299	}
7300
7301	if (ret)
7302	break;
7303
7304	cond_resched();
7305	}
7306
7307	return ret;
7308	}
7309
7310	/* The callers of this must take lock_extent() */
7311	static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7312	u64 len, u64 orig_start, u64 block_start,
7313	u64 block_len, u64 orig_block_len,
7314	u64 ram_bytes, int compress_type,
7315	int type)
7316	{
7317	struct extent_map *em;
7318	int ret;
7319
7320	ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7321	type == BTRFS_ORDERED_COMPRESSED ||
7322	type == BTRFS_ORDERED_NOCOW ||
7323	type == BTRFS_ORDERED_REGULAR);
7324
7325	em = alloc_extent_map();
7326	if (!em)
7327	return ERR_PTR(error: -ENOMEM);
7328
7329	em->start = start;
7330	em->orig_start = orig_start;
7331	em->len = len;
7332	em->block_len = block_len;
7333	em->block_start = block_start;
7334	em->orig_block_len = orig_block_len;
7335	em->ram_bytes = ram_bytes;
7336	em->generation = -1;
7337	em->flags |= EXTENT_FLAG_PINNED;
7338	if (type == BTRFS_ORDERED_PREALLOC)
7339	em->flags |= EXTENT_FLAG_FILLING;
7340	else if (type == BTRFS_ORDERED_COMPRESSED)
7341	extent_map_set_compression(em, type: compress_type);
7342
7343	ret = btrfs_replace_extent_map_range(inode, new_em: em, modified: true);
7344	if (ret) {
7345	free_extent_map(em);
7346	return ERR_PTR(error: ret);
7347	}
7348
7349	/* em got 2 refs now, callers needs to do free_extent_map once. */
7350	return em;
7351	}
7352
7353
7354	static int btrfs_get_blocks_direct_write(struct extent_map **map,
7355	struct inode *inode,
7356	struct btrfs_dio_data *dio_data,
7357	u64 start, u64 *lenp,
7358	unsigned int iomap_flags)
7359	{
7360	const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7361	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
7362	struct extent_map *em = *map;
7363	int type;
7364	u64 block_start, orig_start, orig_block_len, ram_bytes;
7365	struct btrfs_block_group *bg;
7366	bool can_nocow = false;
7367	bool space_reserved = false;
7368	u64 len = *lenp;
7369	u64 prev_len;
7370	int ret = 0;
7371
7372	/*
7373	* We don't allocate a new extent in the following cases
7374	*
7375	* 1) The inode is marked as NODATACOW. In this case we'll just use the
7376	* existing extent.
7377	* 2) The extent is marked as PREALLOC. We're good to go here and can
7378	* just use the extent.
7379	*
7380	*/
7381	if ((em->flags & EXTENT_FLAG_PREALLOC) ||
7382	((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7383	em->block_start != EXTENT_MAP_HOLE)) {
7384	if (em->flags & EXTENT_FLAG_PREALLOC)
7385	type = BTRFS_ORDERED_PREALLOC;
7386	else
7387	type = BTRFS_ORDERED_NOCOW;
7388	len = min(len, em->len - (start - em->start));
7389	block_start = em->block_start + (start - em->start);
7390
7391	if (can_nocow_extent(inode, offset: start, len: &len, orig_start: &orig_start,
7392	orig_block_len: &orig_block_len, ram_bytes: &ram_bytes, nowait: false, strict: false) == 1) {
7393	bg = btrfs_inc_nocow_writers(fs_info, bytenr: block_start);
7394	if (bg)
7395	can_nocow = true;
7396	}
7397	}
7398
7399	prev_len = len;
7400	if (can_nocow) {
7401	struct extent_map *em2;
7402
7403	/* We can NOCOW, so only need to reserve metadata space. */
7404	ret = btrfs_delalloc_reserve_metadata(inode: BTRFS_I(inode), num_bytes: len, disk_num_bytes: len,
7405	noflush: nowait);
7406	if (ret < 0) {
7407	/* Our caller expects us to free the input extent map. */
7408	free_extent_map(em);
7409	*map = NULL;
7410	btrfs_dec_nocow_writers(bg);
7411	if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7412	ret = -EAGAIN;
7413	goto out;
7414	}
7415	space_reserved = true;
7416
7417	em2 = btrfs_create_dio_extent(inode: BTRFS_I(inode), dio_data, start, len,
7418	orig_start, block_start,
7419	block_len: len, orig_block_len,
7420	ram_bytes, type);
7421	btrfs_dec_nocow_writers(bg);
7422	if (type == BTRFS_ORDERED_PREALLOC) {
7423	free_extent_map(em);
7424	*map = em2;
7425	em = em2;
7426	}
7427
7428	if (IS_ERR(ptr: em2)) {
7429	ret = PTR_ERR(ptr: em2);
7430	goto out;
7431	}
7432
7433	dio_data->nocow_done = true;
7434	} else {
7435	/* Our caller expects us to free the input extent map. */
7436	free_extent_map(em);
7437	*map = NULL;
7438
7439	if (nowait) {
7440	ret = -EAGAIN;
7441	goto out;
7442	}
7443
7444	/*
7445	* If we could not allocate data space before locking the file
7446	* range and we can't do a NOCOW write, then we have to fail.
7447	*/
7448	if (!dio_data->data_space_reserved) {
7449	ret = -ENOSPC;
7450	goto out;
7451	}
7452
7453	/*
7454	* We have to COW and we have already reserved data space before,
7455	* so now we reserve only metadata.
7456	*/
7457	ret = btrfs_delalloc_reserve_metadata(inode: BTRFS_I(inode), num_bytes: len, disk_num_bytes: len,
7458	noflush: false);
7459	if (ret < 0)
7460	goto out;
7461	space_reserved = true;
7462
7463	em = btrfs_new_extent_direct(inode: BTRFS_I(inode), dio_data, start, len);
7464	if (IS_ERR(ptr: em)) {
7465	ret = PTR_ERR(ptr: em);
7466	goto out;
7467	}
7468	*map = em;
7469	len = min(len, em->len - (start - em->start));
7470	if (len < prev_len)
7471	btrfs_delalloc_release_metadata(inode: BTRFS_I(inode),
7472	num_bytes: prev_len - len, qgroup_free: true);
7473	}
7474
7475	/*
7476	* We have created our ordered extent, so we can now release our reservation
7477	* for an outstanding extent.
7478	*/
7479	btrfs_delalloc_release_extents(inode: BTRFS_I(inode), num_bytes: prev_len);
7480
7481	/*
7482	* Need to update the i_size under the extent lock so buffered
7483	* readers will get the updated i_size when we unlock.
7484	*/
7485	if (start + len > i_size_read(inode))
7486	i_size_write(inode, i_size: start + len);
7487	out:
7488	if (ret && space_reserved) {
7489	btrfs_delalloc_release_extents(inode: BTRFS_I(inode), num_bytes: len);
7490	btrfs_delalloc_release_metadata(inode: BTRFS_I(inode), num_bytes: len, qgroup_free: true);
7491	}
7492	*lenp = len;
7493	return ret;
7494	}
7495
7496	static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7497	loff_t length, unsigned int flags, struct iomap *iomap,
7498	struct iomap *srcmap)
7499	{
7500	struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7501	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
7502	struct extent_map *em;
7503	struct extent_state *cached_state = NULL;
7504	struct btrfs_dio_data *dio_data = iter->private;
7505	u64 lockstart, lockend;
7506	const bool write = !!(flags & IOMAP_WRITE);
7507	int ret = 0;
7508	u64 len = length;
7509	const u64 data_alloc_len = length;
7510	bool unlock_extents = false;
7511
7512	/*
7513	* We could potentially fault if we have a buffer > PAGE_SIZE, and if
7514	* we're NOWAIT we may submit a bio for a partial range and return
7515	* EIOCBQUEUED, which would result in an errant short read.
7516	*
7517	* The best way to handle this would be to allow for partial completions
7518	* of iocb's, so we could submit the partial bio, return and fault in
7519	* the rest of the pages, and then submit the io for the rest of the
7520	* range. However we don't have that currently, so simply return
7521	* -EAGAIN at this point so that the normal path is used.
7522	*/
7523	if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7524	return -EAGAIN;
7525
7526	/*
7527	* Cap the size of reads to that usually seen in buffered I/O as we need
7528	* to allocate a contiguous array for the checksums.
7529	*/
7530	if (!write)
7531	len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7532
7533	lockstart = start;
7534	lockend = start + len - 1;
7535
7536	/*
7537	* iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7538	* enough if we've written compressed pages to this area, so we need to
7539	* flush the dirty pages again to make absolutely sure that any
7540	* outstanding dirty pages are on disk - the first flush only starts
7541	* compression on the data, while keeping the pages locked, so by the
7542	* time the second flush returns we know bios for the compressed pages
7543	* were submitted and finished, and the pages no longer under writeback.
7544	*
7545	* If we have a NOWAIT request and we have any pages in the range that
7546	* are locked, likely due to compression still in progress, we don't want
7547	* to block on page locks. We also don't want to block on pages marked as
7548	* dirty or under writeback (same as for the non-compression case).
7549	* iomap_dio_rw() did the same check, but after that and before we got
7550	* here, mmap'ed writes may have happened or buffered reads started
7551	* (readpage() and readahead(), which lock pages), as we haven't locked
7552	* the file range yet.
7553	*/
7554	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7555	&BTRFS_I(inode)->runtime_flags)) {
7556	if (flags & IOMAP_NOWAIT) {
7557	if (filemap_range_needs_writeback(mapping: inode->i_mapping,
7558	start_byte: lockstart, end_byte: lockend))
7559	return -EAGAIN;
7560	} else {
7561	ret = filemap_fdatawrite_range(mapping: inode->i_mapping, start,
7562	end: start + length - 1);
7563	if (ret)
7564	return ret;
7565	}
7566	}
7567
7568	memset(dio_data, 0, sizeof(*dio_data));
7569
7570	/*
7571	* We always try to allocate data space and must do it before locking
7572	* the file range, to avoid deadlocks with concurrent writes to the same
7573	* range if the range has several extents and the writes don't expand the
7574	* current i_size (the inode lock is taken in shared mode). If we fail to
7575	* allocate data space here we continue and later, after locking the
7576	* file range, we fail with ENOSPC only if we figure out we can not do a
7577	* NOCOW write.
7578	*/
7579	if (write && !(flags & IOMAP_NOWAIT)) {
7580	ret = btrfs_check_data_free_space(inode: BTRFS_I(inode),
7581	reserved: &dio_data->data_reserved,
7582	start, len: data_alloc_len, noflush: false);
7583	if (!ret)
7584	dio_data->data_space_reserved = true;
7585	else if (ret && !(BTRFS_I(inode)->flags &
7586	(BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7587	goto err;
7588	}
7589
7590	/*
7591	* If this errors out it's because we couldn't invalidate pagecache for
7592	* this range and we need to fallback to buffered IO, or we are doing a
7593	* NOWAIT read/write and we need to block.
7594	*/
7595	ret = lock_extent_direct(inode, lockstart, lockend, cached_state: &cached_state, iomap_flags: flags);
7596	if (ret < 0)
7597	goto err;
7598
7599	em = btrfs_get_extent(inode: BTRFS_I(inode), NULL, start, len);
7600	if (IS_ERR(ptr: em)) {
7601	ret = PTR_ERR(ptr: em);
7602	goto unlock_err;
7603	}
7604
7605	/*
7606	* Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7607	* io. INLINE is special, and we could probably kludge it in here, but
7608	* it's still buffered so for safety lets just fall back to the generic
7609	* buffered path.
7610	*
7611	* For COMPRESSED we _have_ to read the entire extent in so we can
7612	* decompress it, so there will be buffering required no matter what we
7613	* do, so go ahead and fallback to buffered.
7614	*
7615	* We return -ENOTBLK because that's what makes DIO go ahead and go back
7616	* to buffered IO. Don't blame me, this is the price we pay for using
7617	* the generic code.
7618	*/
7619	if (extent_map_is_compressed(em) ||
7620	em->block_start == EXTENT_MAP_INLINE) {
7621	free_extent_map(em);
7622	/*
7623	* If we are in a NOWAIT context, return -EAGAIN in order to
7624	* fallback to buffered IO. This is not only because we can
7625	* block with buffered IO (no support for NOWAIT semantics at
7626	* the moment) but also to avoid returning short reads to user
7627	* space - this happens if we were able to read some data from
7628	* previous non-compressed extents and then when we fallback to
7629	* buffered IO, at btrfs_file_read_iter() by calling
7630	* filemap_read(), we fail to fault in pages for the read buffer,
7631	* in which case filemap_read() returns a short read (the number
7632	* of bytes previously read is > 0, so it does not return -EFAULT).
7633	*/
7634	ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7635	goto unlock_err;
7636	}
7637
7638	len = min(len, em->len - (start - em->start));
7639
7640	/*
7641	* If we have a NOWAIT request and the range contains multiple extents
7642	* (or a mix of extents and holes), then we return -EAGAIN to make the
7643	* caller fallback to a context where it can do a blocking (without
7644	* NOWAIT) request. This way we avoid doing partial IO and returning
7645	* success to the caller, which is not optimal for writes and for reads
7646	* it can result in unexpected behaviour for an application.
7647	*
7648	* When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7649	* iomap_dio_rw(), we can end up returning less data then what the caller
7650	* asked for, resulting in an unexpected, and incorrect, short read.
7651	* That is, the caller asked to read N bytes and we return less than that,
7652	* which is wrong unless we are crossing EOF. This happens if we get a
7653	* page fault error when trying to fault in pages for the buffer that is
7654	* associated to the struct iov_iter passed to iomap_dio_rw(), and we
7655	* have previously submitted bios for other extents in the range, in
7656	* which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7657	* those bios have completed by the time we get the page fault error,
7658	* which we return back to our caller - we should only return EIOCBQUEUED
7659	* after we have submitted bios for all the extents in the range.
7660	*/
7661	if ((flags & IOMAP_NOWAIT) && len < length) {
7662	free_extent_map(em);
7663	ret = -EAGAIN;
7664	goto unlock_err;
7665	}
7666
7667	if (write) {
7668	ret = btrfs_get_blocks_direct_write(map: &em, inode, dio_data,
7669	start, lenp: &len, iomap_flags: flags);
7670	if (ret < 0)
7671	goto unlock_err;
7672	unlock_extents = true;
7673	/* Recalc len in case the new em is smaller than requested */
7674	len = min(len, em->len - (start - em->start));
7675	if (dio_data->data_space_reserved) {
7676	u64 release_offset;
7677	u64 release_len = 0;
7678
7679	if (dio_data->nocow_done) {
7680	release_offset = start;
7681	release_len = data_alloc_len;
7682	} else if (len < data_alloc_len) {
7683	release_offset = start + len;
7684	release_len = data_alloc_len - len;
7685	}
7686
7687	if (release_len > 0)
7688	btrfs_free_reserved_data_space(inode: BTRFS_I(inode),
7689	reserved: dio_data->data_reserved,
7690	start: release_offset,
7691	len: release_len);
7692	}
7693	} else {
7694	/*
7695	* We need to unlock only the end area that we aren't using.
7696	* The rest is going to be unlocked by the endio routine.
7697	*/
7698	lockstart = start + len;
7699	if (lockstart < lockend)
7700	unlock_extents = true;
7701	}
7702
7703	if (unlock_extents)
7704	unlock_extent(tree: &BTRFS_I(inode)->io_tree, start: lockstart, end: lockend,
7705	cached: &cached_state);
7706	else
7707	free_extent_state(state: cached_state);
7708
7709	/*
7710	* Translate extent map information to iomap.
7711	* We trim the extents (and move the addr) even though iomap code does
7712	* that, since we have locked only the parts we are performing I/O in.
7713	*/
7714	if ((em->block_start == EXTENT_MAP_HOLE) ||
7715	((em->flags & EXTENT_FLAG_PREALLOC) && !write)) {
7716	iomap->addr = IOMAP_NULL_ADDR;
7717	iomap->type = IOMAP_HOLE;
7718	} else {
7719	iomap->addr = em->block_start + (start - em->start);
7720	iomap->type = IOMAP_MAPPED;
7721	}
7722	iomap->offset = start;
7723	iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7724	iomap->length = len;
7725	free_extent_map(em);
7726
7727	return 0;
7728
7729	unlock_err:
7730	unlock_extent(tree: &BTRFS_I(inode)->io_tree, start: lockstart, end: lockend,
7731	cached: &cached_state);
7732	err:
7733	if (dio_data->data_space_reserved) {
7734	btrfs_free_reserved_data_space(inode: BTRFS_I(inode),
7735	reserved: dio_data->data_reserved,
7736	start, len: data_alloc_len);
7737	extent_changeset_free(changeset: dio_data->data_reserved);
7738	}
7739
7740	return ret;
7741	}
7742
7743	static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7744	ssize_t written, unsigned int flags, struct iomap *iomap)
7745	{
7746	struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7747	struct btrfs_dio_data *dio_data = iter->private;
7748	size_t submitted = dio_data->submitted;
7749	const bool write = !!(flags & IOMAP_WRITE);
7750	int ret = 0;
7751
7752	if (!write && (iomap->type == IOMAP_HOLE)) {
7753	/* If reading from a hole, unlock and return */
7754	unlock_extent(tree: &BTRFS_I(inode)->io_tree, start: pos, end: pos + length - 1,
7755	NULL);
7756	return 0;
7757	}
7758
7759	if (submitted < length) {
7760	pos += submitted;
7761	length -= submitted;
7762	if (write)
7763	btrfs_finish_ordered_extent(ordered: dio_data->ordered, NULL,
7764	file_offset: pos, len: length, uptodate: false);
7765	else
7766	unlock_extent(tree: &BTRFS_I(inode)->io_tree, start: pos,
7767	end: pos + length - 1, NULL);
7768	ret = -ENOTBLK;
7769	}
7770	if (write) {
7771	btrfs_put_ordered_extent(entry: dio_data->ordered);
7772	dio_data->ordered = NULL;
7773	}
7774
7775	if (write)
7776	extent_changeset_free(changeset: dio_data->data_reserved);
7777	return ret;
7778	}
7779
7780	static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7781	{
7782	struct btrfs_dio_private *dip =
7783	container_of(bbio, struct btrfs_dio_private, bbio);
7784	struct btrfs_inode *inode = bbio->inode;
7785	struct bio *bio = &bbio->bio;
7786
7787	if (bio->bi_status) {
7788	btrfs_warn(inode->root->fs_info,
7789	"direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7790	btrfs_ino(inode), bio->bi_opf,
7791	dip->file_offset, dip->bytes, bio->bi_status);
7792	}
7793
7794	if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7795	btrfs_finish_ordered_extent(ordered: bbio->ordered, NULL,
7796	file_offset: dip->file_offset, len: dip->bytes,
7797	uptodate: !bio->bi_status);
7798	} else {
7799	unlock_extent(tree: &inode->io_tree, start: dip->file_offset,
7800	end: dip->file_offset + dip->bytes - 1, NULL);
7801	}
7802
7803	bbio->bio.bi_private = bbio->private;
7804	iomap_dio_bio_end_io(bio);
7805	}
7806
7807	static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7808	loff_t file_offset)
7809	{
7810	struct btrfs_bio *bbio = btrfs_bio(bio);
7811	struct btrfs_dio_private *dip =
7812	container_of(bbio, struct btrfs_dio_private, bbio);
7813	struct btrfs_dio_data *dio_data = iter->private;
7814
7815	btrfs_bio_init(bbio, fs_info: BTRFS_I(inode: iter->inode)->root->fs_info,
7816	end_io: btrfs_dio_end_io, private: bio->bi_private);
7817	bbio->inode = BTRFS_I(inode: iter->inode);
7818	bbio->file_offset = file_offset;
7819
7820	dip->file_offset = file_offset;
7821	dip->bytes = bio->bi_iter.bi_size;
7822
7823	dio_data->submitted += bio->bi_iter.bi_size;
7824
7825	/*
7826	* Check if we are doing a partial write. If we are, we need to split
7827	* the ordered extent to match the submitted bio. Hang on to the
7828	* remaining unfinishable ordered_extent in dio_data so that it can be
7829	* cancelled in iomap_end to avoid a deadlock wherein faulting the
7830	* remaining pages is blocked on the outstanding ordered extent.
7831	*/
7832	if (iter->flags & IOMAP_WRITE) {
7833	int ret;
7834
7835	ret = btrfs_extract_ordered_extent(bbio, ordered: dio_data->ordered);
7836	if (ret) {
7837	btrfs_finish_ordered_extent(ordered: dio_data->ordered, NULL,
7838	file_offset, len: dip->bytes,
7839	uptodate: !ret);
7840	bio->bi_status = errno_to_blk_status(errno: ret);
7841	iomap_dio_bio_end_io(bio);
7842	return;
7843	}
7844	}
7845
7846	btrfs_submit_bio(bbio, mirror_num: 0);
7847	}
7848
7849	static const struct iomap_ops btrfs_dio_iomap_ops = {
7850	.iomap_begin = btrfs_dio_iomap_begin,
7851	.iomap_end = btrfs_dio_iomap_end,
7852	};
7853
7854	static const struct iomap_dio_ops btrfs_dio_ops = {
7855	.submit_io = btrfs_dio_submit_io,
7856	.bio_set = &btrfs_dio_bioset,
7857	};
7858
7859	ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7860	{
7861	struct btrfs_dio_data data = { 0 };
7862
7863	return iomap_dio_rw(iocb, iter, ops: &btrfs_dio_iomap_ops, dops: &btrfs_dio_ops,
7864	IOMAP_DIO_PARTIAL, private: &data, done_before);
7865	}
7866
7867	struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7868	size_t done_before)
7869	{
7870	struct btrfs_dio_data data = { 0 };
7871
7872	return __iomap_dio_rw(iocb, iter, ops: &btrfs_dio_iomap_ops, dops: &btrfs_dio_ops,
7873	IOMAP_DIO_PARTIAL, private: &data, done_before);
7874	}
7875
7876	static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7877	u64 start, u64 len)
7878	{
7879	struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
7880	int ret;
7881
7882	ret = fiemap_prep(inode, fieinfo, start, len: &len, supported_flags: 0);
7883	if (ret)
7884	return ret;
7885
7886	/*
7887	* fiemap_prep() called filemap_write_and_wait() for the whole possible
7888	* file range (0 to LLONG_MAX), but that is not enough if we have
7889	* compression enabled. The first filemap_fdatawrite_range() only kicks
7890	* in the compression of data (in an async thread) and will return
7891	* before the compression is done and writeback is started. A second
7892	* filemap_fdatawrite_range() is needed to wait for the compression to
7893	* complete and writeback to start. We also need to wait for ordered
7894	* extents to complete, because our fiemap implementation uses mainly
7895	* file extent items to list the extents, searching for extent maps
7896	* only for file ranges with holes or prealloc extents to figure out
7897	* if we have delalloc in those ranges.
7898	*/
7899	if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7900	ret = btrfs_wait_ordered_range(inode, start: 0, LLONG_MAX);
7901	if (ret)
7902	return ret;
7903	}
7904
7905	btrfs_inode_lock(inode: btrfs_inode, ilock_flags: BTRFS_ILOCK_SHARED);
7906
7907	/*
7908	* We did an initial flush to avoid holding the inode's lock while
7909	* triggering writeback and waiting for the completion of IO and ordered
7910	* extents. Now after we locked the inode we do it again, because it's
7911	* possible a new write may have happened in between those two steps.
7912	*/
7913	if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7914	ret = btrfs_wait_ordered_range(inode, start: 0, LLONG_MAX);
7915	if (ret) {
7916	btrfs_inode_unlock(inode: btrfs_inode, ilock_flags: BTRFS_ILOCK_SHARED);
7917	return ret;
7918	}
7919	}
7920
7921	ret = extent_fiemap(inode: btrfs_inode, fieinfo, start, len);
7922	btrfs_inode_unlock(inode: btrfs_inode, ilock_flags: BTRFS_ILOCK_SHARED);
7923
7924	return ret;
7925	}
7926
7927	static int btrfs_writepages(struct address_space *mapping,
7928	struct writeback_control *wbc)
7929	{
7930	return extent_writepages(mapping, wbc);
7931	}
7932
7933	static void btrfs_readahead(struct readahead_control *rac)
7934	{
7935	extent_readahead(rac);
7936	}
7937
7938	/*
7939	* For release_folio() and invalidate_folio() we have a race window where
7940	* folio_end_writeback() is called but the subpage spinlock is not yet released.
7941	* If we continue to release/invalidate the page, we could cause use-after-free
7942	* for subpage spinlock. So this function is to spin and wait for subpage
7943	* spinlock.
7944	*/
7945	static void wait_subpage_spinlock(struct page *page)
7946	{
7947	struct btrfs_fs_info *fs_info = page_to_fs_info(page);
7948	struct folio *folio = page_folio(page);
7949	struct btrfs_subpage *subpage;
7950
7951	if (!btrfs_is_subpage(fs_info, mapping: page->mapping))
7952	return;
7953
7954	ASSERT(folio_test_private(folio) && folio_get_private(folio));
7955	subpage = folio_get_private(folio);
7956
7957	/*
7958	* This may look insane as we just acquire the spinlock and release it,
7959	* without doing anything. But we just want to make sure no one is
7960	* still holding the subpage spinlock.
7961	* And since the page is not dirty nor writeback, and we have page
7962	* locked, the only possible way to hold a spinlock is from the endio
7963	* function to clear page writeback.
7964	*
7965	* Here we just acquire the spinlock so that all existing callers
7966	* should exit and we're safe to release/invalidate the page.
7967	*/
7968	spin_lock_irq(lock: &subpage->lock);
7969	spin_unlock_irq(lock: &subpage->lock);
7970	}
7971
7972	static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7973	{
7974	int ret = try_release_extent_mapping(page: &folio->page, mask: gfp_flags);
7975
7976	if (ret == 1) {
7977	wait_subpage_spinlock(page: &folio->page);
7978	clear_page_extent_mapped(page: &folio->page);
7979	}
7980	return ret;
7981	}
7982
7983	static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7984	{
7985	if (folio_test_writeback(folio) || folio_test_dirty(folio))
7986	return false;
7987	return __btrfs_release_folio(folio, gfp_flags);
7988	}
7989
7990	#ifdef CONFIG_MIGRATION
7991	static int btrfs_migrate_folio(struct address_space *mapping,
7992	struct folio *dst, struct folio *src,
7993	enum migrate_mode mode)
7994	{
7995	int ret = filemap_migrate_folio(mapping, dst, src, mode);
7996
7997	if (ret != MIGRATEPAGE_SUCCESS)
7998	return ret;
7999
8000	if (folio_test_ordered(src)) {
8001	folio_clear_ordered(src);
8002	folio_set_ordered(dst);
8003	}
8004
8005	return MIGRATEPAGE_SUCCESS;
8006	}
8007	#else
8008	#define btrfs_migrate_folio NULL
8009	#endif
8010
8011	static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8012	size_t length)
8013	{
8014	struct btrfs_inode *inode = folio_to_inode(folio);
8015	struct btrfs_fs_info *fs_info = inode->root->fs_info;
8016	struct extent_io_tree *tree = &inode->io_tree;
8017	struct extent_state *cached_state = NULL;
8018	u64 page_start = folio_pos(folio);
8019	u64 page_end = page_start + folio_size(folio) - 1;
8020	u64 cur;
8021	int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8022
8023	/*
8024	* We have folio locked so no new ordered extent can be created on this
8025	* page, nor bio can be submitted for this folio.
8026	*
8027	* But already submitted bio can still be finished on this folio.
8028	* Furthermore, endio function won't skip folio which has Ordered
8029	* (Private2) already cleared, so it's possible for endio and
8030	* invalidate_folio to do the same ordered extent accounting twice
8031	* on one folio.
8032	*
8033	* So here we wait for any submitted bios to finish, so that we won't
8034	* do double ordered extent accounting on the same folio.
8035	*/
8036	folio_wait_writeback(folio);
8037	wait_subpage_spinlock(page: &folio->page);
8038
8039	/*
8040	* For subpage case, we have call sites like
8041	* btrfs_punch_hole_lock_range() which passes range not aligned to
8042	* sectorsize.
8043	* If the range doesn't cover the full folio, we don't need to and
8044	* shouldn't clear page extent mapped, as folio->private can still
8045	* record subpage dirty bits for other part of the range.
8046	*
8047	* For cases that invalidate the full folio even the range doesn't
8048	* cover the full folio, like invalidating the last folio, we're
8049	* still safe to wait for ordered extent to finish.
8050	*/
8051	if (!(offset == 0 && length == folio_size(folio))) {
8052	btrfs_release_folio(folio, GFP_NOFS);
8053	return;
8054	}
8055
8056	if (!inode_evicting)
8057	lock_extent(tree, start: page_start, end: page_end, cached: &cached_state);
8058
8059	cur = page_start;
8060	while (cur < page_end) {
8061	struct btrfs_ordered_extent *ordered;
8062	u64 range_end;
8063	u32 range_len;
8064	u32 extra_flags = 0;
8065
8066	ordered = btrfs_lookup_first_ordered_range(inode, file_offset: cur,
8067	len: page_end + 1 - cur);
8068	if (!ordered) {
8069	range_end = page_end;
8070	/*
8071	* No ordered extent covering this range, we are safe
8072	* to delete all extent states in the range.
8073	*/
8074	extra_flags = EXTENT_CLEAR_ALL_BITS;
8075	goto next;
8076	}
8077	if (ordered->file_offset > cur) {
8078	/*
8079	* There is a range between [cur, oe->file_offset) not
8080	* covered by any ordered extent.
8081	* We are safe to delete all extent states, and handle
8082	* the ordered extent in the next iteration.
8083	*/
8084	range_end = ordered->file_offset - 1;
8085	extra_flags = EXTENT_CLEAR_ALL_BITS;
8086	goto next;
8087	}
8088
8089	range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8090	page_end);
8091	ASSERT(range_end + 1 - cur < U32_MAX);
8092	range_len = range_end + 1 - cur;
8093	if (!btrfs_folio_test_ordered(fs_info, folio, start: cur, len: range_len)) {
8094	/*
8095	* If Ordered (Private2) is cleared, it means endio has
8096	* already been executed for the range.
8097	* We can't delete the extent states as
8098	* btrfs_finish_ordered_io() may still use some of them.
8099	*/
8100	goto next;
8101	}
8102	btrfs_folio_clear_ordered(fs_info, folio, start: cur, len: range_len);
8103
8104	/*
8105	* IO on this page will never be started, so we need to account
8106	* for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8107	* here, must leave that up for the ordered extent completion.
8108	*
8109	* This will also unlock the range for incoming
8110	* btrfs_finish_ordered_io().
8111	*/
8112	if (!inode_evicting)
8113	clear_extent_bit(tree, start: cur, end: range_end,
8114	bits: EXTENT_DELALLOC |
8115	EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8116	EXTENT_DEFRAG, cached: &cached_state);
8117
8118	spin_lock_irq(lock: &inode->ordered_tree_lock);
8119	set_bit(nr: BTRFS_ORDERED_TRUNCATED, addr: &ordered->flags);
8120	ordered->truncated_len = min(ordered->truncated_len,
8121	cur - ordered->file_offset);
8122	spin_unlock_irq(lock: &inode->ordered_tree_lock);
8123
8124	/*
8125	* If the ordered extent has finished, we're safe to delete all
8126	* the extent states of the range, otherwise
8127	* btrfs_finish_ordered_io() will get executed by endio for
8128	* other pages, so we can't delete extent states.
8129	*/
8130	if (btrfs_dec_test_ordered_pending(inode, cached: &ordered,
8131	file_offset: cur, io_size: range_end + 1 - cur)) {
8132	btrfs_finish_ordered_io(ordered);
8133	/*
8134	* The ordered extent has finished, now we're again
8135	* safe to delete all extent states of the range.
8136	*/
8137	extra_flags = EXTENT_CLEAR_ALL_BITS;
8138	}
8139	next:
8140	if (ordered)
8141	btrfs_put_ordered_extent(entry: ordered);
8142	/*
8143	* Qgroup reserved space handler
8144	* Sector(s) here will be either:
8145	*
8146	* 1) Already written to disk or bio already finished
8147	* Then its QGROUP_RESERVED bit in io_tree is already cleared.
8148	* Qgroup will be handled by its qgroup_record then.
8149	* btrfs_qgroup_free_data() call will do nothing here.
8150	*
8151	* 2) Not written to disk yet
8152	* Then btrfs_qgroup_free_data() call will clear the
8153	* QGROUP_RESERVED bit of its io_tree, and free the qgroup
8154	* reserved data space.
8155	* Since the IO will never happen for this page.
8156	*/
8157	btrfs_qgroup_free_data(inode, NULL, start: cur, len: range_end + 1 - cur, NULL);
8158	if (!inode_evicting) {
8159	clear_extent_bit(tree, start: cur, end: range_end, bits: EXTENT_LOCKED |
8160	EXTENT_DELALLOC | EXTENT_UPTODATE |
8161	EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8162	extra_flags, cached: &cached_state);
8163	}
8164	cur = range_end + 1;
8165	}
8166	/*
8167	* We have iterated through all ordered extents of the page, the page
8168	* should not have Ordered (Private2) anymore, or the above iteration
8169	* did something wrong.
8170	*/
8171	ASSERT(!folio_test_ordered(folio));
8172	btrfs_folio_clear_checked(fs_info, folio, start: folio_pos(folio), len: folio_size(folio));
8173	if (!inode_evicting)
8174	__btrfs_release_folio(folio, GFP_NOFS);
8175	clear_page_extent_mapped(page: &folio->page);
8176	}
8177
8178	/*
8179	* btrfs_page_mkwrite() is not allowed to change the file size as it gets
8180	* called from a page fault handler when a page is first dirtied. Hence we must
8181	* be careful to check for EOF conditions here. We set the page up correctly
8182	* for a written page which means we get ENOSPC checking when writing into
8183	* holes and correct delalloc and unwritten extent mapping on filesystems that
8184	* support these features.
8185	*
8186	* We are not allowed to take the i_mutex here so we have to play games to
8187	* protect against truncate races as the page could now be beyond EOF. Because
8188	* truncate_setsize() writes the inode size before removing pages, once we have
8189	* the page lock we can determine safely if the page is beyond EOF. If it is not
8190	* beyond EOF, then the page is guaranteed safe against truncation until we
8191	* unlock the page.
8192	*/
8193	vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8194	{
8195	struct page *page = vmf->page;
8196	struct folio *folio = page_folio(page);
8197	struct inode *inode = file_inode(f: vmf->vma->vm_file);
8198	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
8199	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8200	struct btrfs_ordered_extent *ordered;
8201	struct extent_state *cached_state = NULL;
8202	struct extent_changeset *data_reserved = NULL;
8203	unsigned long zero_start;
8204	loff_t size;
8205	vm_fault_t ret;
8206	int ret2;
8207	int reserved = 0;
8208	u64 reserved_space;
8209	u64 page_start;
8210	u64 page_end;
8211	u64 end;
8212
8213	ASSERT(folio_order(folio) == 0);
8214
8215	reserved_space = PAGE_SIZE;
8216
8217	sb_start_pagefault(sb: inode->i_sb);
8218	page_start = page_offset(page);
8219	page_end = page_start + PAGE_SIZE - 1;
8220	end = page_end;
8221
8222	/*
8223	* Reserving delalloc space after obtaining the page lock can lead to
8224	* deadlock. For example, if a dirty page is locked by this function
8225	* and the call to btrfs_delalloc_reserve_space() ends up triggering
8226	* dirty page write out, then the btrfs_writepages() function could
8227	* end up waiting indefinitely to get a lock on the page currently
8228	* being processed by btrfs_page_mkwrite() function.
8229	*/
8230	ret2 = btrfs_delalloc_reserve_space(inode: BTRFS_I(inode), reserved: &data_reserved,
8231	start: page_start, len: reserved_space);
8232	if (!ret2) {
8233	ret2 = file_update_time(file: vmf->vma->vm_file);
8234	reserved = 1;
8235	}
8236	if (ret2) {
8237	ret = vmf_error(err: ret2);
8238	if (reserved)
8239	goto out;
8240	goto out_noreserve;
8241	}
8242
8243	ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8244	again:
8245	down_read(sem: &BTRFS_I(inode)->i_mmap_lock);
8246	lock_page(page);
8247	size = i_size_read(inode);
8248
8249	if ((page->mapping != inode->i_mapping) ||
8250	(page_start >= size)) {
8251	/* page got truncated out from underneath us */
8252	goto out_unlock;
8253	}
8254	wait_on_page_writeback(page);
8255
8256	lock_extent(tree: io_tree, start: page_start, end: page_end, cached: &cached_state);
8257	ret2 = set_page_extent_mapped(page);
8258	if (ret2 < 0) {
8259	ret = vmf_error(err: ret2);
8260	unlock_extent(tree: io_tree, start: page_start, end: page_end, cached: &cached_state);
8261	goto out_unlock;
8262	}
8263
8264	/*
8265	* we can't set the delalloc bits if there are pending ordered
8266	* extents. Drop our locks and wait for them to finish
8267	*/
8268	ordered = btrfs_lookup_ordered_range(inode: BTRFS_I(inode), file_offset: page_start,
8269	PAGE_SIZE);
8270	if (ordered) {
8271	unlock_extent(tree: io_tree, start: page_start, end: page_end, cached: &cached_state);
8272	unlock_page(page);
8273	up_read(sem: &BTRFS_I(inode)->i_mmap_lock);
8274	btrfs_start_ordered_extent(entry: ordered);
8275	btrfs_put_ordered_extent(entry: ordered);
8276	goto again;
8277	}
8278
8279	if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8280	reserved_space = round_up(size - page_start,
8281	fs_info->sectorsize);
8282	if (reserved_space < PAGE_SIZE) {
8283	end = page_start + reserved_space - 1;
8284	btrfs_delalloc_release_space(inode: BTRFS_I(inode),
8285	reserved: data_reserved, start: page_start,
8286	PAGE_SIZE - reserved_space, qgroup_free: true);
8287	}
8288	}
8289
8290	/*
8291	* page_mkwrite gets called when the page is firstly dirtied after it's
8292	* faulted in, but write(2) could also dirty a page and set delalloc
8293	* bits, thus in this case for space account reason, we still need to
8294	* clear any delalloc bits within this page range since we have to
8295	* reserve data&meta space before lock_page() (see above comments).
8296	*/
8297	clear_extent_bit(tree: &BTRFS_I(inode)->io_tree, start: page_start, end,
8298	bits: EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8299	EXTENT_DEFRAG, cached: &cached_state);
8300
8301	ret2 = btrfs_set_extent_delalloc(inode: BTRFS_I(inode), start: page_start, end, extra_bits: 0,
8302	cached_state: &cached_state);
8303	if (ret2) {
8304	unlock_extent(tree: io_tree, start: page_start, end: page_end, cached: &cached_state);
8305	ret = VM_FAULT_SIGBUS;
8306	goto out_unlock;
8307	}
8308
8309	/* page is wholly or partially inside EOF */
8310	if (page_start + PAGE_SIZE > size)
8311	zero_start = offset_in_page(size);
8312	else
8313	zero_start = PAGE_SIZE;
8314
8315	if (zero_start != PAGE_SIZE)
8316	memzero_page(page, offset: zero_start, PAGE_SIZE - zero_start);
8317
8318	btrfs_folio_clear_checked(fs_info, folio, start: page_start, PAGE_SIZE);
8319	btrfs_folio_set_dirty(fs_info, folio, start: page_start, len: end + 1 - page_start);
8320	btrfs_folio_set_uptodate(fs_info, folio, start: page_start, len: end + 1 - page_start);
8321
8322	btrfs_set_inode_last_sub_trans(inode: BTRFS_I(inode));
8323
8324	unlock_extent(tree: io_tree, start: page_start, end: page_end, cached: &cached_state);
8325	up_read(sem: &BTRFS_I(inode)->i_mmap_lock);
8326
8327	btrfs_delalloc_release_extents(inode: BTRFS_I(inode), PAGE_SIZE);
8328	sb_end_pagefault(sb: inode->i_sb);
8329	extent_changeset_free(changeset: data_reserved);
8330	return VM_FAULT_LOCKED;
8331
8332	out_unlock:
8333	unlock_page(page);
8334	up_read(sem: &BTRFS_I(inode)->i_mmap_lock);
8335	out:
8336	btrfs_delalloc_release_extents(inode: BTRFS_I(inode), PAGE_SIZE);
8337	btrfs_delalloc_release_space(inode: BTRFS_I(inode), reserved: data_reserved, start: page_start,
8338	len: reserved_space, qgroup_free: (ret != 0));
8339	out_noreserve:
8340	sb_end_pagefault(sb: inode->i_sb);
8341	extent_changeset_free(changeset: data_reserved);
8342	return ret;
8343	}
8344
8345	static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8346	{
8347	struct btrfs_truncate_control control = {
8348	.inode = inode,
8349	.ino = btrfs_ino(inode),
8350	.min_type = BTRFS_EXTENT_DATA_KEY,
8351	.clear_extent_range = true,
8352	};
8353	struct btrfs_root *root = inode->root;
8354	struct btrfs_fs_info *fs_info = root->fs_info;
8355	struct btrfs_block_rsv *rsv;
8356	int ret;
8357	struct btrfs_trans_handle *trans;
8358	u64 mask = fs_info->sectorsize - 1;
8359	const u64 min_size = btrfs_calc_metadata_size(fs_info, num_items: 1);
8360
8361	if (!skip_writeback) {
8362	ret = btrfs_wait_ordered_range(inode: &inode->vfs_inode,
8363	start: inode->vfs_inode.i_size & (~mask),
8364	len: (u64)-1);
8365	if (ret)
8366	return ret;
8367	}
8368
8369	/*
8370	* Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8371	* things going on here:
8372	*
8373	* 1) We need to reserve space to update our inode.
8374	*
8375	* 2) We need to have something to cache all the space that is going to
8376	* be free'd up by the truncate operation, but also have some slack
8377	* space reserved in case it uses space during the truncate (thank you
8378	* very much snapshotting).
8379	*
8380	* And we need these to be separate. The fact is we can use a lot of
8381	* space doing the truncate, and we have no earthly idea how much space
8382	* we will use, so we need the truncate reservation to be separate so it
8383	* doesn't end up using space reserved for updating the inode. We also
8384	* need to be able to stop the transaction and start a new one, which
8385	* means we need to be able to update the inode several times, and we
8386	* have no idea of knowing how many times that will be, so we can't just
8387	* reserve 1 item for the entirety of the operation, so that has to be
8388	* done separately as well.
8389	*
8390	* So that leaves us with
8391	*
8392	* 1) rsv - for the truncate reservation, which we will steal from the
8393	* transaction reservation.
8394	* 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8395	* updating the inode.
8396	*/
8397	rsv = btrfs_alloc_block_rsv(fs_info, type: BTRFS_BLOCK_RSV_TEMP);
8398	if (!rsv)
8399	return -ENOMEM;
8400	rsv->size = min_size;
8401	rsv->failfast = true;
8402
8403	/*
8404	* 1 for the truncate slack space
8405	* 1 for updating the inode.
8406	*/
8407	trans = btrfs_start_transaction(root, num_items: 2);
8408	if (IS_ERR(ptr: trans)) {
8409	ret = PTR_ERR(ptr: trans);
8410	goto out;
8411	}
8412
8413	/* Migrate the slack space for the truncate to our reserve */
8414	ret = btrfs_block_rsv_migrate(src_rsv: &fs_info->trans_block_rsv, dst_rsv: rsv,
8415	num_bytes: min_size, update_size: false);
8416	/*
8417	* We have reserved 2 metadata units when we started the transaction and
8418	* min_size matches 1 unit, so this should never fail, but if it does,
8419	* it's not critical we just fail truncation.
8420	*/
8421	if (WARN_ON(ret)) {
8422	btrfs_end_transaction(trans);
8423	goto out;
8424	}
8425
8426	trans->block_rsv = rsv;
8427
8428	while (1) {
8429	struct extent_state *cached_state = NULL;
8430	const u64 new_size = inode->vfs_inode.i_size;
8431	const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8432
8433	control.new_size = new_size;
8434	lock_extent(tree: &inode->io_tree, start: lock_start, end: (u64)-1, cached: &cached_state);
8435	/*
8436	* We want to drop from the next block forward in case this new
8437	* size is not block aligned since we will be keeping the last
8438	* block of the extent just the way it is.
8439	*/
8440	btrfs_drop_extent_map_range(inode,
8441	ALIGN(new_size, fs_info->sectorsize),
8442	end: (u64)-1, skip_pinned: false);
8443
8444	ret = btrfs_truncate_inode_items(trans, root, control: &control);
8445
8446	inode_sub_bytes(inode: &inode->vfs_inode, bytes: control.sub_bytes);
8447	btrfs_inode_safe_disk_i_size_write(inode, new_i_size: control.last_size);
8448
8449	unlock_extent(tree: &inode->io_tree, start: lock_start, end: (u64)-1, cached: &cached_state);
8450
8451	trans->block_rsv = &fs_info->trans_block_rsv;
8452	if (ret != -ENOSPC && ret != -EAGAIN)
8453	break;
8454
8455	ret = btrfs_update_inode(trans, inode);
8456	if (ret)
8457	break;
8458
8459	btrfs_end_transaction(trans);
8460	btrfs_btree_balance_dirty(fs_info);
8461
8462	trans = btrfs_start_transaction(root, num_items: 2);
8463	if (IS_ERR(ptr: trans)) {
8464	ret = PTR_ERR(ptr: trans);
8465	trans = NULL;
8466	break;
8467	}
8468
8469	btrfs_block_rsv_release(fs_info, block_rsv: rsv, num_bytes: -1, NULL);
8470	ret = btrfs_block_rsv_migrate(src_rsv: &fs_info->trans_block_rsv,
8471	dst_rsv: rsv, num_bytes: min_size, update_size: false);
8472	/*
8473	* We have reserved 2 metadata units when we started the
8474	* transaction and min_size matches 1 unit, so this should never
8475	* fail, but if it does, it's not critical we just fail truncation.
8476	*/
8477	if (WARN_ON(ret))
8478	break;
8479
8480	trans->block_rsv = rsv;
8481	}
8482
8483	/*
8484	* We can't call btrfs_truncate_block inside a trans handle as we could
8485	* deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8486	* know we've truncated everything except the last little bit, and can
8487	* do btrfs_truncate_block and then update the disk_i_size.
8488	*/
8489	if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8490	btrfs_end_transaction(trans);
8491	btrfs_btree_balance_dirty(fs_info);
8492
8493	ret = btrfs_truncate_block(inode, from: inode->vfs_inode.i_size, len: 0, front: 0);
8494	if (ret)
8495	goto out;
8496	trans = btrfs_start_transaction(root, num_items: 1);
8497	if (IS_ERR(ptr: trans)) {
8498	ret = PTR_ERR(ptr: trans);
8499	goto out;
8500	}
8501	btrfs_inode_safe_disk_i_size_write(inode, new_i_size: 0);
8502	}
8503
8504	if (trans) {
8505	int ret2;
8506
8507	trans->block_rsv = &fs_info->trans_block_rsv;
8508	ret2 = btrfs_update_inode(trans, inode);
8509	if (ret2 && !ret)
8510	ret = ret2;
8511
8512	ret2 = btrfs_end_transaction(trans);
8513	if (ret2 && !ret)
8514	ret = ret2;
8515	btrfs_btree_balance_dirty(fs_info);
8516	}
8517	out:
8518	btrfs_free_block_rsv(fs_info, rsv);
8519	/*
8520	* So if we truncate and then write and fsync we normally would just
8521	* write the extents that changed, which is a problem if we need to
8522	* first truncate that entire inode. So set this flag so we write out
8523	* all of the extents in the inode to the sync log so we're completely
8524	* safe.
8525	*
8526	* If no extents were dropped or trimmed we don't need to force the next
8527	* fsync to truncate all the inode's items from the log and re-log them
8528	* all. This means the truncate operation did not change the file size,
8529	* or changed it to a smaller size but there was only an implicit hole
8530	* between the old i_size and the new i_size, and there were no prealloc
8531	* extents beyond i_size to drop.
8532	*/
8533	if (control.extents_found > 0)
8534	btrfs_set_inode_full_sync(inode);
8535
8536	return ret;
8537	}
8538
8539	struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8540	struct inode *dir)
8541	{
8542	struct inode *inode;
8543
8544	inode = new_inode(sb: dir->i_sb);
8545	if (inode) {
8546	/*
8547	* Subvolumes don't inherit the sgid bit or the parent's gid if
8548	* the parent's sgid bit is set. This is probably a bug.
8549	*/
8550	inode_init_owner(idmap, inode, NULL,
8551	S_IFDIR | (~current_umask() & S_IRWXUGO));
8552	inode->i_op = &btrfs_dir_inode_operations;
8553	inode->i_fop = &btrfs_dir_file_operations;
8554	}
8555	return inode;
8556	}
8557
8558	struct inode *btrfs_alloc_inode(struct super_block *sb)
8559	{
8560	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8561	struct btrfs_inode *ei;
8562	struct inode *inode;
8563	struct extent_io_tree *file_extent_tree = NULL;
8564
8565	/* Self tests may pass a NULL fs_info. */
8566	if (fs_info && !btrfs_fs_incompat(fs_info, NO_HOLES)) {
8567	file_extent_tree = kmalloc(size: sizeof(struct extent_io_tree), GFP_KERNEL);
8568	if (!file_extent_tree)
8569	return NULL;
8570	}
8571
8572	ei = alloc_inode_sb(sb, cache: btrfs_inode_cachep, GFP_KERNEL);
8573	if (!ei) {
8574	kfree(objp: file_extent_tree);
8575	return NULL;
8576	}
8577
8578	ei->root = NULL;
8579	ei->generation = 0;
8580	ei->last_trans = 0;
8581	ei->last_sub_trans = 0;
8582	ei->logged_trans = 0;
8583	ei->delalloc_bytes = 0;
8584	ei->new_delalloc_bytes = 0;
8585	ei->defrag_bytes = 0;
8586	ei->disk_i_size = 0;
8587	ei->flags = 0;
8588	ei->ro_flags = 0;
8589	ei->csum_bytes = 0;
8590	ei->index_cnt = (u64)-1;
8591	ei->dir_index = 0;
8592	ei->last_unlink_trans = 0;
8593	ei->last_reflink_trans = 0;
8594	ei->last_log_commit = 0;
8595
8596	spin_lock_init(&ei->lock);
8597	ei->outstanding_extents = 0;
8598	if (sb->s_magic != BTRFS_TEST_MAGIC)
8599	btrfs_init_metadata_block_rsv(fs_info, rsv: &ei->block_rsv,
8600	type: BTRFS_BLOCK_RSV_DELALLOC);
8601	ei->runtime_flags = 0;
8602	ei->prop_compress = BTRFS_COMPRESS_NONE;
8603	ei->defrag_compress = BTRFS_COMPRESS_NONE;
8604
8605	ei->delayed_node = NULL;
8606
8607	ei->i_otime_sec = 0;
8608	ei->i_otime_nsec = 0;
8609
8610	inode = &ei->vfs_inode;
8611	extent_map_tree_init(tree: &ei->extent_tree);
8612
8613	/* This io tree sets the valid inode. */
8614	extent_io_tree_init(fs_info, tree: &ei->io_tree, owner: IO_TREE_INODE_IO);
8615	ei->io_tree.inode = ei;
8616
8617	ei->file_extent_tree = file_extent_tree;
8618	if (file_extent_tree) {
8619	extent_io_tree_init(fs_info, tree: ei->file_extent_tree,
8620	owner: IO_TREE_INODE_FILE_EXTENT);
8621	/* Lockdep class is set only for the file extent tree. */
8622	lockdep_set_class(&ei->file_extent_tree->lock, &file_extent_tree_class);
8623	}
8624	mutex_init(&ei->log_mutex);
8625	spin_lock_init(&ei->ordered_tree_lock);
8626	ei->ordered_tree = RB_ROOT;
8627	ei->ordered_tree_last = NULL;
8628	INIT_LIST_HEAD(list: &ei->delalloc_inodes);
8629	INIT_LIST_HEAD(list: &ei->delayed_iput);
8630	RB_CLEAR_NODE(&ei->rb_node);
8631	init_rwsem(&ei->i_mmap_lock);
8632
8633	return inode;
8634	}
8635
8636	#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8637	void btrfs_test_destroy_inode(struct inode *inode)
8638	{
8639	btrfs_drop_extent_map_range(inode: BTRFS_I(inode), start: 0, end: (u64)-1, skip_pinned: false);
8640	kfree(objp: BTRFS_I(inode)->file_extent_tree);
8641	kmem_cache_free(s: btrfs_inode_cachep, objp: BTRFS_I(inode));
8642	}
8643	#endif
8644
8645	void btrfs_free_inode(struct inode *inode)
8646	{
8647	kfree(objp: BTRFS_I(inode)->file_extent_tree);
8648	kmem_cache_free(s: btrfs_inode_cachep, objp: BTRFS_I(inode));
8649	}
8650
8651	void btrfs_destroy_inode(struct inode *vfs_inode)
8652	{
8653	struct btrfs_ordered_extent *ordered;
8654	struct btrfs_inode *inode = BTRFS_I(inode: vfs_inode);
8655	struct btrfs_root *root = inode->root;
8656	bool freespace_inode;
8657
8658	WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8659	WARN_ON(vfs_inode->i_data.nrpages);
8660	WARN_ON(inode->block_rsv.reserved);
8661	WARN_ON(inode->block_rsv.size);
8662	WARN_ON(inode->outstanding_extents);
8663	if (!S_ISDIR(vfs_inode->i_mode)) {
8664	WARN_ON(inode->delalloc_bytes);
8665	WARN_ON(inode->new_delalloc_bytes);
8666	}
8667	WARN_ON(inode->csum_bytes);
8668	WARN_ON(inode->defrag_bytes);
8669
8670	/*
8671	* This can happen where we create an inode, but somebody else also
8672	* created the same inode and we need to destroy the one we already
8673	* created.
8674	*/
8675	if (!root)
8676	return;
8677
8678	/*
8679	* If this is a free space inode do not take the ordered extents lockdep
8680	* map.
8681	*/
8682	freespace_inode = btrfs_is_free_space_inode(inode);
8683
8684	while (1) {
8685	ordered = btrfs_lookup_first_ordered_extent(inode, file_offset: (u64)-1);
8686	if (!ordered)
8687	break;
8688	else {
8689	btrfs_err(root->fs_info,
8690	"found ordered extent %llu %llu on inode cleanup",
8691	ordered->file_offset, ordered->num_bytes);
8692
8693	if (!freespace_inode)
8694	btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8695
8696	btrfs_remove_ordered_extent(btrfs_inode: inode, entry: ordered);
8697	btrfs_put_ordered_extent(entry: ordered);
8698	btrfs_put_ordered_extent(entry: ordered);
8699	}
8700	}
8701	btrfs_qgroup_check_reserved_leak(inode);
8702	inode_tree_del(inode);
8703	btrfs_drop_extent_map_range(inode, start: 0, end: (u64)-1, skip_pinned: false);
8704	btrfs_inode_clear_file_extent_range(inode, start: 0, len: (u64)-1);
8705	btrfs_put_root(root: inode->root);
8706	}
8707
8708	int btrfs_drop_inode(struct inode *inode)
8709	{
8710	struct btrfs_root *root = BTRFS_I(inode)->root;
8711
8712	if (root == NULL)
8713	return 1;
8714
8715	/* the snap/subvol tree is on deleting */
8716	if (btrfs_root_refs(s: &root->root_item) == 0)
8717	return 1;
8718	else
8719	return generic_drop_inode(inode);
8720	}
8721
8722	static void init_once(void *foo)
8723	{
8724	struct btrfs_inode *ei = foo;
8725
8726	inode_init_once(&ei->vfs_inode);
8727	}
8728
8729	void __cold btrfs_destroy_cachep(void)
8730	{
8731	/*
8732	* Make sure all delayed rcu free inodes are flushed before we
8733	* destroy cache.
8734	*/
8735	rcu_barrier();
8736	bioset_exit(&btrfs_dio_bioset);
8737	kmem_cache_destroy(s: btrfs_inode_cachep);
8738	}
8739
8740	int __init btrfs_init_cachep(void)
8741	{
8742	btrfs_inode_cachep = kmem_cache_create(name: "btrfs_inode",
8743	size: sizeof(struct btrfs_inode), align: 0,
8744	SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
8745	ctor: init_once);
8746	if (!btrfs_inode_cachep)
8747	goto fail;
8748
8749	if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8750	offsetof(struct btrfs_dio_private, bbio.bio),
8751	flags: BIOSET_NEED_BVECS))
8752	goto fail;
8753
8754	return 0;
8755	fail:
8756	btrfs_destroy_cachep();
8757	return -ENOMEM;
8758	}
8759
8760	static int btrfs_getattr(struct mnt_idmap *idmap,
8761	const struct path *path, struct kstat *stat,
8762	u32 request_mask, unsigned int flags)
8763	{
8764	u64 delalloc_bytes;
8765	u64 inode_bytes;
8766	struct inode *inode = d_inode(dentry: path->dentry);
8767	u32 blocksize = btrfs_sb(sb: inode->i_sb)->sectorsize;
8768	u32 bi_flags = BTRFS_I(inode)->flags;
8769	u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8770
8771	stat->result_mask |= STATX_BTIME;
8772	stat->btime.tv_sec = BTRFS_I(inode)->i_otime_sec;
8773	stat->btime.tv_nsec = BTRFS_I(inode)->i_otime_nsec;
8774	if (bi_flags & BTRFS_INODE_APPEND)
8775	stat->attributes |= STATX_ATTR_APPEND;
8776	if (bi_flags & BTRFS_INODE_COMPRESS)
8777	stat->attributes |= STATX_ATTR_COMPRESSED;
8778	if (bi_flags & BTRFS_INODE_IMMUTABLE)
8779	stat->attributes |= STATX_ATTR_IMMUTABLE;
8780	if (bi_flags & BTRFS_INODE_NODUMP)
8781	stat->attributes |= STATX_ATTR_NODUMP;
8782	if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8783	stat->attributes |= STATX_ATTR_VERITY;
8784
8785	stat->attributes_mask |= (STATX_ATTR_APPEND |
8786	STATX_ATTR_COMPRESSED |
8787	STATX_ATTR_IMMUTABLE |
8788	STATX_ATTR_NODUMP);
8789
8790	generic_fillattr(idmap, request_mask, inode, stat);
8791	stat->dev = BTRFS_I(inode)->root->anon_dev;
8792
8793	spin_lock(lock: &BTRFS_I(inode)->lock);
8794	delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8795	inode_bytes = inode_get_bytes(inode);
8796	spin_unlock(lock: &BTRFS_I(inode)->lock);
8797	stat->blocks = (ALIGN(inode_bytes, blocksize) +
8798	ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8799	return 0;
8800	}
8801
8802	static int btrfs_rename_exchange(struct inode *old_dir,
8803	struct dentry *old_dentry,
8804	struct inode *new_dir,
8805	struct dentry *new_dentry)
8806	{
8807	struct btrfs_fs_info *fs_info = inode_to_fs_info(old_dir);
8808	struct btrfs_trans_handle *trans;
8809	unsigned int trans_num_items;
8810	struct btrfs_root *root = BTRFS_I(inode: old_dir)->root;
8811	struct btrfs_root *dest = BTRFS_I(inode: new_dir)->root;
8812	struct inode *new_inode = new_dentry->d_inode;
8813	struct inode *old_inode = old_dentry->d_inode;
8814	struct btrfs_rename_ctx old_rename_ctx;
8815	struct btrfs_rename_ctx new_rename_ctx;
8816	u64 old_ino = btrfs_ino(inode: BTRFS_I(inode: old_inode));
8817	u64 new_ino = btrfs_ino(inode: BTRFS_I(inode: new_inode));
8818	u64 old_idx = 0;
8819	u64 new_idx = 0;
8820	int ret;
8821	int ret2;
8822	bool need_abort = false;
8823	struct fscrypt_name old_fname, new_fname;
8824	struct fscrypt_str *old_name, *new_name;
8825
8826	/*
8827	* For non-subvolumes allow exchange only within one subvolume, in the
8828	* same inode namespace. Two subvolumes (represented as directory) can
8829	* be exchanged as they're a logical link and have a fixed inode number.
8830	*/
8831	if (root != dest &&
8832	(old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8833	new_ino != BTRFS_FIRST_FREE_OBJECTID))
8834	return -EXDEV;
8835
8836	ret = fscrypt_setup_filename(inode: old_dir, iname: &old_dentry->d_name, lookup: 0, fname: &old_fname);
8837	if (ret)
8838	return ret;
8839
8840	ret = fscrypt_setup_filename(inode: new_dir, iname: &new_dentry->d_name, lookup: 0, fname: &new_fname);
8841	if (ret) {
8842	fscrypt_free_filename(fname: &old_fname);
8843	return ret;
8844	}
8845
8846	old_name = &old_fname.disk_name;
8847	new_name = &new_fname.disk_name;
8848
8849	/* close the race window with snapshot create/destroy ioctl */
8850	if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8851	new_ino == BTRFS_FIRST_FREE_OBJECTID)
8852	down_read(sem: &fs_info->subvol_sem);
8853
8854	/*
8855	* For each inode:
8856	* 1 to remove old dir item
8857	* 1 to remove old dir index
8858	* 1 to add new dir item
8859	* 1 to add new dir index
8860	* 1 to update parent inode
8861	*
8862	* If the parents are the same, we only need to account for one
8863	*/
8864	trans_num_items = (old_dir == new_dir ? 9 : 10);
8865	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8866	/*
8867	* 1 to remove old root ref
8868	* 1 to remove old root backref
8869	* 1 to add new root ref
8870	* 1 to add new root backref
8871	*/
8872	trans_num_items += 4;
8873	} else {
8874	/*
8875	* 1 to update inode item
8876	* 1 to remove old inode ref
8877	* 1 to add new inode ref
8878	*/
8879	trans_num_items += 3;
8880	}
8881	if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8882	trans_num_items += 4;
8883	else
8884	trans_num_items += 3;
8885	trans = btrfs_start_transaction(root, num_items: trans_num_items);
8886	if (IS_ERR(ptr: trans)) {
8887	ret = PTR_ERR(ptr: trans);
8888	goto out_notrans;
8889	}
8890
8891	if (dest != root) {
8892	ret = btrfs_record_root_in_trans(trans, root: dest);
8893	if (ret)
8894	goto out_fail;
8895	}
8896
8897	/*
8898	* We need to find a free sequence number both in the source and
8899	* in the destination directory for the exchange.
8900	*/
8901	ret = btrfs_set_inode_index(dir: BTRFS_I(inode: new_dir), index: &old_idx);
8902	if (ret)
8903	goto out_fail;
8904	ret = btrfs_set_inode_index(dir: BTRFS_I(inode: old_dir), index: &new_idx);
8905	if (ret)
8906	goto out_fail;
8907
8908	BTRFS_I(inode: old_inode)->dir_index = 0ULL;
8909	BTRFS_I(inode: new_inode)->dir_index = 0ULL;
8910
8911	/* Reference for the source. */
8912	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8913	/* force full log commit if subvolume involved. */
8914	btrfs_set_log_full_commit(trans);
8915	} else {
8916	ret = btrfs_insert_inode_ref(trans, root: dest, name: new_name, inode_objectid: old_ino,
8917	ref_objectid: btrfs_ino(inode: BTRFS_I(inode: new_dir)),
8918	index: old_idx);
8919	if (ret)
8920	goto out_fail;
8921	need_abort = true;
8922	}
8923
8924	/* And now for the dest. */
8925	if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8926	/* force full log commit if subvolume involved. */
8927	btrfs_set_log_full_commit(trans);
8928	} else {
8929	ret = btrfs_insert_inode_ref(trans, root, name: old_name, inode_objectid: new_ino,
8930	ref_objectid: btrfs_ino(inode: BTRFS_I(inode: old_dir)),
8931	index: new_idx);
8932	if (ret) {
8933	if (need_abort)
8934	btrfs_abort_transaction(trans, ret);
8935	goto out_fail;
8936	}
8937	}
8938
8939	/* Update inode version and ctime/mtime. */
8940	inode_inc_iversion(inode: old_dir);
8941	inode_inc_iversion(inode: new_dir);
8942	inode_inc_iversion(inode: old_inode);
8943	inode_inc_iversion(inode: new_inode);
8944	simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
8945
8946	if (old_dentry->d_parent != new_dentry->d_parent) {
8947	btrfs_record_unlink_dir(trans, dir: BTRFS_I(inode: old_dir),
8948	inode: BTRFS_I(inode: old_inode), for_rename: true);
8949	btrfs_record_unlink_dir(trans, dir: BTRFS_I(inode: new_dir),
8950	inode: BTRFS_I(inode: new_inode), for_rename: true);
8951	}
8952
8953	/* src is a subvolume */
8954	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8955	ret = btrfs_unlink_subvol(trans, dir: BTRFS_I(inode: old_dir), dentry: old_dentry);
8956	} else { /* src is an inode */
8957	ret = __btrfs_unlink_inode(trans, dir: BTRFS_I(inode: old_dir),
8958	inode: BTRFS_I(inode: old_dentry->d_inode),
8959	name: old_name, rename_ctx: &old_rename_ctx);
8960	if (!ret)
8961	ret = btrfs_update_inode(trans, inode: BTRFS_I(inode: old_inode));
8962	}
8963	if (ret) {
8964	btrfs_abort_transaction(trans, ret);
8965	goto out_fail;
8966	}
8967
8968	/* dest is a subvolume */
8969	if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8970	ret = btrfs_unlink_subvol(trans, dir: BTRFS_I(inode: new_dir), dentry: new_dentry);
8971	} else { /* dest is an inode */
8972	ret = __btrfs_unlink_inode(trans, dir: BTRFS_I(inode: new_dir),
8973	inode: BTRFS_I(inode: new_dentry->d_inode),
8974	name: new_name, rename_ctx: &new_rename_ctx);
8975	if (!ret)
8976	ret = btrfs_update_inode(trans, inode: BTRFS_I(inode: new_inode));
8977	}
8978	if (ret) {
8979	btrfs_abort_transaction(trans, ret);
8980	goto out_fail;
8981	}
8982
8983	ret = btrfs_add_link(trans, parent_inode: BTRFS_I(inode: new_dir), inode: BTRFS_I(inode: old_inode),
8984	name: new_name, add_backref: 0, index: old_idx);
8985	if (ret) {
8986	btrfs_abort_transaction(trans, ret);
8987	goto out_fail;
8988	}
8989
8990	ret = btrfs_add_link(trans, parent_inode: BTRFS_I(inode: old_dir), inode: BTRFS_I(inode: new_inode),
8991	name: old_name, add_backref: 0, index: new_idx);
8992	if (ret) {
8993	btrfs_abort_transaction(trans, ret);
8994	goto out_fail;
8995	}
8996
8997	if (old_inode->i_nlink == 1)
8998	BTRFS_I(inode: old_inode)->dir_index = old_idx;
8999	if (new_inode->i_nlink == 1)
9000	BTRFS_I(inode: new_inode)->dir_index = new_idx;
9001
9002	/*
9003	* Now pin the logs of the roots. We do it to ensure that no other task
9004	* can sync the logs while we are in progress with the rename, because
9005	* that could result in an inconsistency in case any of the inodes that
9006	* are part of this rename operation were logged before.
9007	*/
9008	if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9009	btrfs_pin_log_trans(root);
9010	if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9011	btrfs_pin_log_trans(root: dest);
9012
9013	/* Do the log updates for all inodes. */
9014	if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9015	btrfs_log_new_name(trans, old_dentry, old_dir: BTRFS_I(inode: old_dir),
9016	old_dir_index: old_rename_ctx.index, parent: new_dentry->d_parent);
9017	if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9018	btrfs_log_new_name(trans, old_dentry: new_dentry, old_dir: BTRFS_I(inode: new_dir),
9019	old_dir_index: new_rename_ctx.index, parent: old_dentry->d_parent);
9020
9021	/* Now unpin the logs. */
9022	if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9023	btrfs_end_log_trans(root);
9024	if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9025	btrfs_end_log_trans(root: dest);
9026	out_fail:
9027	ret2 = btrfs_end_transaction(trans);
9028	ret = ret ? ret : ret2;
9029	out_notrans:
9030	if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9031	old_ino == BTRFS_FIRST_FREE_OBJECTID)
9032	up_read(sem: &fs_info->subvol_sem);
9033
9034	fscrypt_free_filename(fname: &new_fname);
9035	fscrypt_free_filename(fname: &old_fname);
9036	return ret;
9037	}
9038
9039	static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
9040	struct inode *dir)
9041	{
9042	struct inode *inode;
9043
9044	inode = new_inode(sb: dir->i_sb);
9045	if (inode) {
9046	inode_init_owner(idmap, inode, dir,
9047	S_IFCHR | WHITEOUT_MODE);
9048	inode->i_op = &btrfs_special_inode_operations;
9049	init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9050	}
9051	return inode;
9052	}
9053
9054	static int btrfs_rename(struct mnt_idmap *idmap,
9055	struct inode *old_dir, struct dentry *old_dentry,
9056	struct inode *new_dir, struct dentry *new_dentry,
9057	unsigned int flags)
9058	{
9059	struct btrfs_fs_info *fs_info = inode_to_fs_info(old_dir);
9060	struct btrfs_new_inode_args whiteout_args = {
9061	.dir = old_dir,
9062	.dentry = old_dentry,
9063	};
9064	struct btrfs_trans_handle *trans;
9065	unsigned int trans_num_items;
9066	struct btrfs_root *root = BTRFS_I(inode: old_dir)->root;
9067	struct btrfs_root *dest = BTRFS_I(inode: new_dir)->root;
9068	struct inode *new_inode = d_inode(dentry: new_dentry);
9069	struct inode *old_inode = d_inode(dentry: old_dentry);
9070	struct btrfs_rename_ctx rename_ctx;
9071	u64 index = 0;
9072	int ret;
9073	int ret2;
9074	u64 old_ino = btrfs_ino(inode: BTRFS_I(inode: old_inode));
9075	struct fscrypt_name old_fname, new_fname;
9076
9077	if (btrfs_ino(inode: BTRFS_I(inode: new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9078	return -EPERM;
9079
9080	/* we only allow rename subvolume link between subvolumes */
9081	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9082	return -EXDEV;
9083
9084	if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9085	(new_inode && btrfs_ino(inode: BTRFS_I(inode: new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9086	return -ENOTEMPTY;
9087
9088	if (S_ISDIR(old_inode->i_mode) && new_inode &&
9089	new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9090	return -ENOTEMPTY;
9091
9092	ret = fscrypt_setup_filename(inode: old_dir, iname: &old_dentry->d_name, lookup: 0, fname: &old_fname);
9093	if (ret)
9094	return ret;
9095
9096	ret = fscrypt_setup_filename(inode: new_dir, iname: &new_dentry->d_name, lookup: 0, fname: &new_fname);
9097	if (ret) {
9098	fscrypt_free_filename(fname: &old_fname);
9099	return ret;
9100	}
9101
9102	/* check for collisions, even if the name isn't there */
9103	ret = btrfs_check_dir_item_collision(root: dest, dir: new_dir->i_ino, name: &new_fname.disk_name);
9104	if (ret) {
9105	if (ret == -EEXIST) {
9106	/* we shouldn't get
9107	* eexist without a new_inode */
9108	if (WARN_ON(!new_inode)) {
9109	goto out_fscrypt_names;
9110	}
9111	} else {
9112	/* maybe -EOVERFLOW */
9113	goto out_fscrypt_names;
9114	}
9115	}
9116	ret = 0;
9117
9118	/*
9119	* we're using rename to replace one file with another. Start IO on it
9120	* now so we don't add too much work to the end of the transaction
9121	*/
9122	if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9123	filemap_flush(old_inode->i_mapping);
9124
9125	if (flags & RENAME_WHITEOUT) {
9126	whiteout_args.inode = new_whiteout_inode(idmap, dir: old_dir);
9127	if (!whiteout_args.inode) {
9128	ret = -ENOMEM;
9129	goto out_fscrypt_names;
9130	}
9131	ret = btrfs_new_inode_prepare(args: &whiteout_args, trans_num_items: &trans_num_items);
9132	if (ret)
9133	goto out_whiteout_inode;
9134	} else {
9135	/* 1 to update the old parent inode. */
9136	trans_num_items = 1;
9137	}
9138
9139	if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9140	/* Close the race window with snapshot create/destroy ioctl */
9141	down_read(sem: &fs_info->subvol_sem);
9142	/*
9143	* 1 to remove old root ref
9144	* 1 to remove old root backref
9145	* 1 to add new root ref
9146	* 1 to add new root backref
9147	*/
9148	trans_num_items += 4;
9149	} else {
9150	/*
9151	* 1 to update inode
9152	* 1 to remove old inode ref
9153	* 1 to add new inode ref
9154	*/
9155	trans_num_items += 3;
9156	}
9157	/*
9158	* 1 to remove old dir item
9159	* 1 to remove old dir index
9160	* 1 to add new dir item
9161	* 1 to add new dir index
9162	*/
9163	trans_num_items += 4;
9164	/* 1 to update new parent inode if it's not the same as the old parent */
9165	if (new_dir != old_dir)
9166	trans_num_items++;
9167	if (new_inode) {
9168	/*
9169	* 1 to update inode
9170	* 1 to remove inode ref
9171	* 1 to remove dir item
9172	* 1 to remove dir index
9173	* 1 to possibly add orphan item
9174	*/
9175	trans_num_items += 5;
9176	}
9177	trans = btrfs_start_transaction(root, num_items: trans_num_items);
9178	if (IS_ERR(ptr: trans)) {
9179	ret = PTR_ERR(ptr: trans);
9180	goto out_notrans;
9181	}
9182
9183	if (dest != root) {
9184	ret = btrfs_record_root_in_trans(trans, root: dest);
9185	if (ret)
9186	goto out_fail;
9187	}
9188
9189	ret = btrfs_set_inode_index(dir: BTRFS_I(inode: new_dir), index: &index);
9190	if (ret)
9191	goto out_fail;
9192
9193	BTRFS_I(inode: old_inode)->dir_index = 0ULL;
9194	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9195	/* force full log commit if subvolume involved. */
9196	btrfs_set_log_full_commit(trans);
9197	} else {
9198	ret = btrfs_insert_inode_ref(trans, root: dest, name: &new_fname.disk_name,
9199	inode_objectid: old_ino, ref_objectid: btrfs_ino(inode: BTRFS_I(inode: new_dir)),
9200	index);
9201	if (ret)
9202	goto out_fail;
9203	}
9204
9205	inode_inc_iversion(inode: old_dir);
9206	inode_inc_iversion(inode: new_dir);
9207	inode_inc_iversion(inode: old_inode);
9208	simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
9209
9210	if (old_dentry->d_parent != new_dentry->d_parent)
9211	btrfs_record_unlink_dir(trans, dir: BTRFS_I(inode: old_dir),
9212	inode: BTRFS_I(inode: old_inode), for_rename: true);
9213
9214	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9215	ret = btrfs_unlink_subvol(trans, dir: BTRFS_I(inode: old_dir), dentry: old_dentry);
9216	} else {
9217	ret = __btrfs_unlink_inode(trans, dir: BTRFS_I(inode: old_dir),
9218	inode: BTRFS_I(inode: d_inode(dentry: old_dentry)),
9219	name: &old_fname.disk_name, rename_ctx: &rename_ctx);
9220	if (!ret)
9221	ret = btrfs_update_inode(trans, inode: BTRFS_I(inode: old_inode));
9222	}
9223	if (ret) {
9224	btrfs_abort_transaction(trans, ret);
9225	goto out_fail;
9226	}
9227
9228	if (new_inode) {
9229	inode_inc_iversion(inode: new_inode);
9230	if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9231	BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9232	ret = btrfs_unlink_subvol(trans, dir: BTRFS_I(inode: new_dir), dentry: new_dentry);
9233	BUG_ON(new_inode->i_nlink == 0);
9234	} else {
9235	ret = btrfs_unlink_inode(trans, dir: BTRFS_I(inode: new_dir),
9236	inode: BTRFS_I(inode: d_inode(dentry: new_dentry)),
9237	name: &new_fname.disk_name);
9238	}
9239	if (!ret && new_inode->i_nlink == 0)
9240	ret = btrfs_orphan_add(trans,
9241	inode: BTRFS_I(inode: d_inode(dentry: new_dentry)));
9242	if (ret) {
9243	btrfs_abort_transaction(trans, ret);
9244	goto out_fail;
9245	}
9246	}
9247
9248	ret = btrfs_add_link(trans, parent_inode: BTRFS_I(inode: new_dir), inode: BTRFS_I(inode: old_inode),
9249	name: &new_fname.disk_name, add_backref: 0, index);
9250	if (ret) {
9251	btrfs_abort_transaction(trans, ret);
9252	goto out_fail;
9253	}
9254
9255	if (old_inode->i_nlink == 1)
9256	BTRFS_I(inode: old_inode)->dir_index = index;
9257
9258	if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9259	btrfs_log_new_name(trans, old_dentry, old_dir: BTRFS_I(inode: old_dir),
9260	old_dir_index: rename_ctx.index, parent: new_dentry->d_parent);
9261
9262	if (flags & RENAME_WHITEOUT) {
9263	ret = btrfs_create_new_inode(trans, args: &whiteout_args);
9264	if (ret) {
9265	btrfs_abort_transaction(trans, ret);
9266	goto out_fail;
9267	} else {
9268	unlock_new_inode(whiteout_args.inode);
9269	iput(whiteout_args.inode);
9270	whiteout_args.inode = NULL;
9271	}
9272	}
9273	out_fail:
9274	ret2 = btrfs_end_transaction(trans);
9275	ret = ret ? ret : ret2;
9276	out_notrans:
9277	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9278	up_read(sem: &fs_info->subvol_sem);
9279	if (flags & RENAME_WHITEOUT)
9280	btrfs_new_inode_args_destroy(args: &whiteout_args);
9281	out_whiteout_inode:
9282	if (flags & RENAME_WHITEOUT)
9283	iput(whiteout_args.inode);
9284	out_fscrypt_names:
9285	fscrypt_free_filename(fname: &old_fname);
9286	fscrypt_free_filename(fname: &new_fname);
9287	return ret;
9288	}
9289
9290	static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9291	struct dentry *old_dentry, struct inode *new_dir,
9292	struct dentry *new_dentry, unsigned int flags)
9293	{
9294	int ret;
9295
9296	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9297	return -EINVAL;
9298
9299	if (flags & RENAME_EXCHANGE)
9300	ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9301	new_dentry);
9302	else
9303	ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9304	new_dentry, flags);
9305
9306	btrfs_btree_balance_dirty(fs_info: BTRFS_I(inode: new_dir)->root->fs_info);
9307
9308	return ret;
9309	}
9310
9311	struct btrfs_delalloc_work {
9312	struct inode *inode;
9313	struct completion completion;
9314	struct list_head list;
9315	struct btrfs_work work;
9316	};
9317
9318	static void btrfs_run_delalloc_work(struct btrfs_work *work)
9319	{
9320	struct btrfs_delalloc_work *delalloc_work;
9321	struct inode *inode;
9322
9323	delalloc_work = container_of(work, struct btrfs_delalloc_work,
9324	work);
9325	inode = delalloc_work->inode;
9326	filemap_flush(inode->i_mapping);
9327	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9328	&BTRFS_I(inode)->runtime_flags))
9329	filemap_flush(inode->i_mapping);
9330
9331	iput(inode);
9332	complete(&delalloc_work->completion);
9333	}
9334
9335	static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9336	{
9337	struct btrfs_delalloc_work *work;
9338
9339	work = kmalloc(size: sizeof(*work), GFP_NOFS);
9340	if (!work)
9341	return NULL;
9342
9343	init_completion(x: &work->completion);
9344	INIT_LIST_HEAD(list: &work->list);
9345	work->inode = inode;
9346	btrfs_init_work(work: &work->work, func: btrfs_run_delalloc_work, NULL);
9347
9348	return work;
9349	}
9350
9351	/*
9352	* some fairly slow code that needs optimization. This walks the list
9353	* of all the inodes with pending delalloc and forces them to disk.
9354	*/
9355	static int start_delalloc_inodes(struct btrfs_root *root,
9356	struct writeback_control *wbc, bool snapshot,
9357	bool in_reclaim_context)
9358	{
9359	struct btrfs_inode *binode;
9360	struct inode *inode;
9361	struct btrfs_delalloc_work *work, *next;
9362	LIST_HEAD(works);
9363	LIST_HEAD(splice);
9364	int ret = 0;
9365	bool full_flush = wbc->nr_to_write == LONG_MAX;
9366
9367	mutex_lock(&root->delalloc_mutex);
9368	spin_lock(lock: &root->delalloc_lock);
9369	list_splice_init(list: &root->delalloc_inodes, head: &splice);
9370	while (!list_empty(head: &splice)) {
9371	binode = list_entry(splice.next, struct btrfs_inode,
9372	delalloc_inodes);
9373
9374	list_move_tail(list: &binode->delalloc_inodes,
9375	head: &root->delalloc_inodes);
9376
9377	if (in_reclaim_context &&
9378	test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9379	continue;
9380
9381	inode = igrab(&binode->vfs_inode);
9382	if (!inode) {
9383	cond_resched_lock(&root->delalloc_lock);
9384	continue;
9385	}
9386	spin_unlock(lock: &root->delalloc_lock);
9387
9388	if (snapshot)
9389	set_bit(nr: BTRFS_INODE_SNAPSHOT_FLUSH,
9390	addr: &binode->runtime_flags);
9391	if (full_flush) {
9392	work = btrfs_alloc_delalloc_work(inode);
9393	if (!work) {
9394	iput(inode);
9395	ret = -ENOMEM;
9396	goto out;
9397	}
9398	list_add_tail(new: &work->list, head: &works);
9399	btrfs_queue_work(wq: root->fs_info->flush_workers,
9400	work: &work->work);
9401	} else {
9402	ret = filemap_fdatawrite_wbc(mapping: inode->i_mapping, wbc);
9403	btrfs_add_delayed_iput(inode: BTRFS_I(inode));
9404	if (ret || wbc->nr_to_write <= 0)
9405	goto out;
9406	}
9407	cond_resched();
9408	spin_lock(lock: &root->delalloc_lock);
9409	}
9410	spin_unlock(lock: &root->delalloc_lock);
9411
9412	out:
9413	list_for_each_entry_safe(work, next, &works, list) {
9414	list_del_init(entry: &work->list);
9415	wait_for_completion(&work->completion);
9416	kfree(objp: work);
9417	}
9418
9419	if (!list_empty(head: &splice)) {
9420	spin_lock(lock: &root->delalloc_lock);
9421	list_splice_tail(list: &splice, head: &root->delalloc_inodes);
9422	spin_unlock(lock: &root->delalloc_lock);
9423	}
9424	mutex_unlock(lock: &root->delalloc_mutex);
9425	return ret;
9426	}
9427
9428	int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9429	{
9430	struct writeback_control wbc = {
9431	.nr_to_write = LONG_MAX,
9432	.sync_mode = WB_SYNC_NONE,
9433	.range_start = 0,
9434	.range_end = LLONG_MAX,
9435	};
9436	struct btrfs_fs_info *fs_info = root->fs_info;
9437
9438	if (BTRFS_FS_ERROR(fs_info))
9439	return -EROFS;
9440
9441	return start_delalloc_inodes(root, wbc: &wbc, snapshot: true, in_reclaim_context);
9442	}
9443
9444	int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9445	bool in_reclaim_context)
9446	{
9447	struct writeback_control wbc = {
9448	.nr_to_write = nr,
9449	.sync_mode = WB_SYNC_NONE,
9450	.range_start = 0,
9451	.range_end = LLONG_MAX,
9452	};
9453	struct btrfs_root *root;
9454	LIST_HEAD(splice);
9455	int ret;
9456
9457	if (BTRFS_FS_ERROR(fs_info))
9458	return -EROFS;
9459
9460	mutex_lock(&fs_info->delalloc_root_mutex);
9461	spin_lock(lock: &fs_info->delalloc_root_lock);
9462	list_splice_init(list: &fs_info->delalloc_roots, head: &splice);
9463	while (!list_empty(head: &splice)) {
9464	/*
9465	* Reset nr_to_write here so we know that we're doing a full
9466	* flush.
9467	*/
9468	if (nr == LONG_MAX)
9469	wbc.nr_to_write = LONG_MAX;
9470
9471	root = list_first_entry(&splice, struct btrfs_root,
9472	delalloc_root);
9473	root = btrfs_grab_root(root);
9474	BUG_ON(!root);
9475	list_move_tail(list: &root->delalloc_root,
9476	head: &fs_info->delalloc_roots);
9477	spin_unlock(lock: &fs_info->delalloc_root_lock);
9478
9479	ret = start_delalloc_inodes(root, wbc: &wbc, snapshot: false, in_reclaim_context);
9480	btrfs_put_root(root);
9481	if (ret < 0 || wbc.nr_to_write <= 0)
9482	goto out;
9483	spin_lock(lock: &fs_info->delalloc_root_lock);
9484	}
9485	spin_unlock(lock: &fs_info->delalloc_root_lock);
9486
9487	ret = 0;
9488	out:
9489	if (!list_empty(head: &splice)) {
9490	spin_lock(lock: &fs_info->delalloc_root_lock);
9491	list_splice_tail(list: &splice, head: &fs_info->delalloc_roots);
9492	spin_unlock(lock: &fs_info->delalloc_root_lock);
9493	}
9494	mutex_unlock(lock: &fs_info->delalloc_root_mutex);
9495	return ret;
9496	}
9497
9498	static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9499	struct dentry *dentry, const char *symname)
9500	{
9501	struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
9502	struct btrfs_trans_handle *trans;
9503	struct btrfs_root *root = BTRFS_I(inode: dir)->root;
9504	struct btrfs_path *path;
9505	struct btrfs_key key;
9506	struct inode *inode;
9507	struct btrfs_new_inode_args new_inode_args = {
9508	.dir = dir,
9509	.dentry = dentry,
9510	};
9511	unsigned int trans_num_items;
9512	int err;
9513	int name_len;
9514	int datasize;
9515	unsigned long ptr;
9516	struct btrfs_file_extent_item *ei;
9517	struct extent_buffer *leaf;
9518
9519	name_len = strlen(symname);
9520	if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(info: fs_info))
9521	return -ENAMETOOLONG;
9522
9523	inode = new_inode(sb: dir->i_sb);
9524	if (!inode)
9525	return -ENOMEM;
9526	inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9527	inode->i_op = &btrfs_symlink_inode_operations;
9528	inode_nohighmem(inode);
9529	inode->i_mapping->a_ops = &btrfs_aops;
9530	btrfs_i_size_write(inode: BTRFS_I(inode), size: name_len);
9531	inode_set_bytes(inode, bytes: name_len);
9532
9533	new_inode_args.inode = inode;
9534	err = btrfs_new_inode_prepare(args: &new_inode_args, trans_num_items: &trans_num_items);
9535	if (err)
9536	goto out_inode;
9537	/* 1 additional item for the inline extent */
9538	trans_num_items++;
9539
9540	trans = btrfs_start_transaction(root, num_items: trans_num_items);
9541	if (IS_ERR(ptr: trans)) {
9542	err = PTR_ERR(ptr: trans);
9543	goto out_new_inode_args;
9544	}
9545
9546	err = btrfs_create_new_inode(trans, args: &new_inode_args);
9547	if (err)
9548	goto out;
9549
9550	path = btrfs_alloc_path();
9551	if (!path) {
9552	err = -ENOMEM;
9553	btrfs_abort_transaction(trans, err);
9554	discard_new_inode(inode);
9555	inode = NULL;
9556	goto out;
9557	}
9558	key.objectid = btrfs_ino(inode: BTRFS_I(inode));
9559	key.offset = 0;
9560	key.type = BTRFS_EXTENT_DATA_KEY;
9561	datasize = btrfs_file_extent_calc_inline_size(datasize: name_len);
9562	err = btrfs_insert_empty_item(trans, root, path, key: &key,
9563	data_size: datasize);
9564	if (err) {
9565	btrfs_abort_transaction(trans, err);
9566	btrfs_free_path(p: path);
9567	discard_new_inode(inode);
9568	inode = NULL;
9569	goto out;
9570	}
9571	leaf = path->nodes[0];
9572	ei = btrfs_item_ptr(leaf, path->slots[0],
9573	struct btrfs_file_extent_item);
9574	btrfs_set_file_extent_generation(eb: leaf, s: ei, val: trans->transid);
9575	btrfs_set_file_extent_type(eb: leaf, s: ei,
9576	val: BTRFS_FILE_EXTENT_INLINE);
9577	btrfs_set_file_extent_encryption(eb: leaf, s: ei, val: 0);
9578	btrfs_set_file_extent_compression(eb: leaf, s: ei, val: 0);
9579	btrfs_set_file_extent_other_encoding(eb: leaf, s: ei, val: 0);
9580	btrfs_set_file_extent_ram_bytes(eb: leaf, s: ei, val: name_len);
9581
9582	ptr = btrfs_file_extent_inline_start(e: ei);
9583	write_extent_buffer(eb: leaf, src: symname, start: ptr, len: name_len);
9584	btrfs_mark_buffer_dirty(trans, buf: leaf);
9585	btrfs_free_path(p: path);
9586
9587	d_instantiate_new(dentry, inode);
9588	err = 0;
9589	out:
9590	btrfs_end_transaction(trans);
9591	btrfs_btree_balance_dirty(fs_info);
9592	out_new_inode_args:
9593	btrfs_new_inode_args_destroy(args: &new_inode_args);
9594	out_inode:
9595	if (err)
9596	iput(inode);
9597	return err;
9598	}
9599
9600	static struct btrfs_trans_handle *insert_prealloc_file_extent(
9601	struct btrfs_trans_handle *trans_in,
9602	struct btrfs_inode *inode,
9603	struct btrfs_key *ins,
9604	u64 file_offset)
9605	{
9606	struct btrfs_file_extent_item stack_fi;
9607	struct btrfs_replace_extent_info extent_info;
9608	struct btrfs_trans_handle *trans = trans_in;
9609	struct btrfs_path *path;
9610	u64 start = ins->objectid;
9611	u64 len = ins->offset;
9612	u64 qgroup_released = 0;
9613	int ret;
9614
9615	memset(&stack_fi, 0, sizeof(stack_fi));
9616
9617	btrfs_set_stack_file_extent_type(s: &stack_fi, val: BTRFS_FILE_EXTENT_PREALLOC);
9618	btrfs_set_stack_file_extent_disk_bytenr(s: &stack_fi, val: start);
9619	btrfs_set_stack_file_extent_disk_num_bytes(s: &stack_fi, val: len);
9620	btrfs_set_stack_file_extent_num_bytes(s: &stack_fi, val: len);
9621	btrfs_set_stack_file_extent_ram_bytes(s: &stack_fi, val: len);
9622	btrfs_set_stack_file_extent_compression(s: &stack_fi, val: BTRFS_COMPRESS_NONE);
9623	/* Encryption and other encoding is reserved and all 0 */
9624
9625	ret = btrfs_qgroup_release_data(inode, start: file_offset, len, released: &qgroup_released);
9626	if (ret < 0)
9627	return ERR_PTR(error: ret);
9628
9629	if (trans) {
9630	ret = insert_reserved_file_extent(trans, inode,
9631	file_pos: file_offset, stack_fi: &stack_fi,
9632	update_inode_bytes: true, qgroup_reserved: qgroup_released);
9633	if (ret)
9634	goto free_qgroup;
9635	return trans;
9636	}
9637
9638	extent_info.disk_offset = start;
9639	extent_info.disk_len = len;
9640	extent_info.data_offset = 0;
9641	extent_info.data_len = len;
9642	extent_info.file_offset = file_offset;
9643	extent_info.extent_buf = (char *)&stack_fi;
9644	extent_info.is_new_extent = true;
9645	extent_info.update_times = true;
9646	extent_info.qgroup_reserved = qgroup_released;
9647	extent_info.insertions = 0;
9648
9649	path = btrfs_alloc_path();
9650	if (!path) {
9651	ret = -ENOMEM;
9652	goto free_qgroup;
9653	}
9654
9655	ret = btrfs_replace_file_extents(inode, path, start: file_offset,
9656	end: file_offset + len - 1, extent_info: &extent_info,
9657	trans_out: &trans);
9658	btrfs_free_path(p: path);
9659	if (ret)
9660	goto free_qgroup;
9661	return trans;
9662
9663	free_qgroup:
9664	/*
9665	* We have released qgroup data range at the beginning of the function,
9666	* and normally qgroup_released bytes will be freed when committing
9667	* transaction.
9668	* But if we error out early, we have to free what we have released
9669	* or we leak qgroup data reservation.
9670	*/
9671	btrfs_qgroup_free_refroot(fs_info: inode->root->fs_info,
9672	ref_root: inode->root->root_key.objectid, num_bytes: qgroup_released,
9673	type: BTRFS_QGROUP_RSV_DATA);
9674	return ERR_PTR(error: ret);
9675	}
9676
9677	static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9678	u64 start, u64 num_bytes, u64 min_size,
9679	loff_t actual_len, u64 *alloc_hint,
9680	struct btrfs_trans_handle *trans)
9681	{
9682	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
9683	struct extent_map *em;
9684	struct btrfs_root *root = BTRFS_I(inode)->root;
9685	struct btrfs_key ins;
9686	u64 cur_offset = start;
9687	u64 clear_offset = start;
9688	u64 i_size;
9689	u64 cur_bytes;
9690	u64 last_alloc = (u64)-1;
9691	int ret = 0;
9692	bool own_trans = true;
9693	u64 end = start + num_bytes - 1;
9694
9695	if (trans)
9696	own_trans = false;
9697	while (num_bytes > 0) {
9698	cur_bytes = min_t(u64, num_bytes, SZ_256M);
9699	cur_bytes = max(cur_bytes, min_size);
9700	/*
9701	* If we are severely fragmented we could end up with really
9702	* small allocations, so if the allocator is returning small
9703	* chunks lets make its job easier by only searching for those
9704	* sized chunks.
9705	*/
9706	cur_bytes = min(cur_bytes, last_alloc);
9707	ret = btrfs_reserve_extent(root, ram_bytes: cur_bytes, num_bytes: cur_bytes,
9708	min_alloc_size: min_size, empty_size: 0, hint_byte: *alloc_hint, ins: &ins, is_data: 1, delalloc: 0);
9709	if (ret)
9710	break;
9711
9712	/*
9713	* We've reserved this space, and thus converted it from
9714	* ->bytes_may_use to ->bytes_reserved. Any error that happens
9715	* from here on out we will only need to clear our reservation
9716	* for the remaining unreserved area, so advance our
9717	* clear_offset by our extent size.
9718	*/
9719	clear_offset += ins.offset;
9720
9721	last_alloc = ins.offset;
9722	trans = insert_prealloc_file_extent(trans_in: trans, inode: BTRFS_I(inode),
9723	ins: &ins, file_offset: cur_offset);
9724	/*
9725	* Now that we inserted the prealloc extent we can finally
9726	* decrement the number of reservations in the block group.
9727	* If we did it before, we could race with relocation and have
9728	* relocation miss the reserved extent, making it fail later.
9729	*/
9730	btrfs_dec_block_group_reservations(fs_info, start: ins.objectid);
9731	if (IS_ERR(ptr: trans)) {
9732	ret = PTR_ERR(ptr: trans);
9733	btrfs_free_reserved_extent(fs_info, start: ins.objectid,
9734	len: ins.offset, delalloc: 0);
9735	break;
9736	}
9737
9738	em = alloc_extent_map();
9739	if (!em) {
9740	btrfs_drop_extent_map_range(inode: BTRFS_I(inode), start: cur_offset,
9741	end: cur_offset + ins.offset - 1, skip_pinned: false);
9742	btrfs_set_inode_full_sync(inode: BTRFS_I(inode));
9743	goto next;
9744	}
9745
9746	em->start = cur_offset;
9747	em->orig_start = cur_offset;
9748	em->len = ins.offset;
9749	em->block_start = ins.objectid;
9750	em->block_len = ins.offset;
9751	em->orig_block_len = ins.offset;
9752	em->ram_bytes = ins.offset;
9753	em->flags |= EXTENT_FLAG_PREALLOC;
9754	em->generation = trans->transid;
9755
9756	ret = btrfs_replace_extent_map_range(inode: BTRFS_I(inode), new_em: em, modified: true);
9757	free_extent_map(em);
9758	next:
9759	num_bytes -= ins.offset;
9760	cur_offset += ins.offset;
9761	*alloc_hint = ins.objectid + ins.offset;
9762
9763	inode_inc_iversion(inode);
9764	inode_set_ctime_current(inode);
9765	BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9766	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9767	(actual_len > inode->i_size) &&
9768	(cur_offset > inode->i_size)) {
9769	if (cur_offset > actual_len)
9770	i_size = actual_len;
9771	else
9772	i_size = cur_offset;
9773	i_size_write(inode, i_size);
9774	btrfs_inode_safe_disk_i_size_write(inode: BTRFS_I(inode), new_i_size: 0);
9775	}
9776
9777	ret = btrfs_update_inode(trans, inode: BTRFS_I(inode));
9778
9779	if (ret) {
9780	btrfs_abort_transaction(trans, ret);
9781	if (own_trans)
9782	btrfs_end_transaction(trans);
9783	break;
9784	}
9785
9786	if (own_trans) {
9787	btrfs_end_transaction(trans);
9788	trans = NULL;
9789	}
9790	}
9791	if (clear_offset < end)
9792	btrfs_free_reserved_data_space(inode: BTRFS_I(inode), NULL, start: clear_offset,
9793	len: end - clear_offset + 1);
9794	return ret;
9795	}
9796
9797	int btrfs_prealloc_file_range(struct inode *inode, int mode,
9798	u64 start, u64 num_bytes, u64 min_size,
9799	loff_t actual_len, u64 *alloc_hint)
9800	{
9801	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9802	min_size, actual_len, alloc_hint,
9803	NULL);
9804	}
9805
9806	int btrfs_prealloc_file_range_trans(struct inode *inode,
9807	struct btrfs_trans_handle *trans, int mode,
9808	u64 start, u64 num_bytes, u64 min_size,
9809	loff_t actual_len, u64 *alloc_hint)
9810	{
9811	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9812	min_size, actual_len, alloc_hint, trans);
9813	}
9814
9815	static int btrfs_permission(struct mnt_idmap *idmap,
9816	struct inode *inode, int mask)
9817	{
9818	struct btrfs_root *root = BTRFS_I(inode)->root;
9819	umode_t mode = inode->i_mode;
9820
9821	if (mask & MAY_WRITE &&
9822	(S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9823	if (btrfs_root_readonly(root))
9824	return -EROFS;
9825	if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9826	return -EACCES;
9827	}
9828	return generic_permission(idmap, inode, mask);
9829	}
9830
9831	static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9832	struct file *file, umode_t mode)
9833	{
9834	struct btrfs_fs_info *fs_info = inode_to_fs_info(dir);
9835	struct btrfs_trans_handle *trans;
9836	struct btrfs_root *root = BTRFS_I(inode: dir)->root;
9837	struct inode *inode;
9838	struct btrfs_new_inode_args new_inode_args = {
9839	.dir = dir,
9840	.dentry = file->f_path.dentry,
9841	.orphan = true,
9842	};
9843	unsigned int trans_num_items;
9844	int ret;
9845
9846	inode = new_inode(sb: dir->i_sb);
9847	if (!inode)
9848	return -ENOMEM;
9849	inode_init_owner(idmap, inode, dir, mode);
9850	inode->i_fop = &btrfs_file_operations;
9851	inode->i_op = &btrfs_file_inode_operations;
9852	inode->i_mapping->a_ops = &btrfs_aops;
9853
9854	new_inode_args.inode = inode;
9855	ret = btrfs_new_inode_prepare(args: &new_inode_args, trans_num_items: &trans_num_items);
9856	if (ret)
9857	goto out_inode;
9858
9859	trans = btrfs_start_transaction(root, num_items: trans_num_items);
9860	if (IS_ERR(ptr: trans)) {
9861	ret = PTR_ERR(ptr: trans);
9862	goto out_new_inode_args;
9863	}
9864
9865	ret = btrfs_create_new_inode(trans, args: &new_inode_args);
9866
9867	/*
9868	* We set number of links to 0 in btrfs_create_new_inode(), and here we
9869	* set it to 1 because d_tmpfile() will issue a warning if the count is
9870	* 0, through:
9871	*
9872	* d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9873	*/
9874	set_nlink(inode, nlink: 1);
9875
9876	if (!ret) {
9877	d_tmpfile(file, inode);
9878	unlock_new_inode(inode);
9879	mark_inode_dirty(inode);
9880	}
9881
9882	btrfs_end_transaction(trans);
9883	btrfs_btree_balance_dirty(fs_info);
9884	out_new_inode_args:
9885	btrfs_new_inode_args_destroy(args: &new_inode_args);
9886	out_inode:
9887	if (ret)
9888	iput(inode);
9889	return finish_open_simple(file, error: ret);
9890	}
9891
9892	void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9893	{
9894	struct btrfs_fs_info *fs_info = inode->root->fs_info;
9895	unsigned long index = start >> PAGE_SHIFT;
9896	unsigned long end_index = end >> PAGE_SHIFT;
9897	struct page *page;
9898	u32 len;
9899
9900	ASSERT(end + 1 - start <= U32_MAX);
9901	len = end + 1 - start;
9902	while (index <= end_index) {
9903	page = find_get_page(mapping: inode->vfs_inode.i_mapping, offset: index);
9904	ASSERT(page); /* Pages should be in the extent_io_tree */
9905
9906	/* This is for data, which doesn't yet support larger folio. */
9907	ASSERT(folio_order(page_folio(page)) == 0);
9908	btrfs_folio_set_writeback(fs_info, page_folio(page), start, len);
9909	put_page(page);
9910	index++;
9911	}
9912	}
9913
9914	int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9915	int compress_type)
9916	{
9917	switch (compress_type) {
9918	case BTRFS_COMPRESS_NONE:
9919	return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9920	case BTRFS_COMPRESS_ZLIB:
9921	return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9922	case BTRFS_COMPRESS_LZO:
9923	/*
9924	* The LZO format depends on the sector size. 64K is the maximum
9925	* sector size that we support.
9926	*/
9927	if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9928	return -EINVAL;
9929	return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9930	(fs_info->sectorsize_bits - 12);
9931	case BTRFS_COMPRESS_ZSTD:
9932	return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9933	default:
9934	return -EUCLEAN;
9935	}
9936	}
9937
9938	static ssize_t btrfs_encoded_read_inline(
9939	struct kiocb *iocb,
9940	struct iov_iter *iter, u64 start,
9941	u64 lockend,
9942	struct extent_state **cached_state,
9943	u64 extent_start, size_t count,
9944	struct btrfs_ioctl_encoded_io_args *encoded,
9945	bool *unlocked)
9946	{
9947	struct btrfs_inode *inode = BTRFS_I(inode: file_inode(f: iocb->ki_filp));
9948	struct btrfs_root *root = inode->root;
9949	struct btrfs_fs_info *fs_info = root->fs_info;
9950	struct extent_io_tree *io_tree = &inode->io_tree;
9951	struct btrfs_path *path;
9952	struct extent_buffer *leaf;
9953	struct btrfs_file_extent_item *item;
9954	u64 ram_bytes;
9955	unsigned long ptr;
9956	void *tmp;
9957	ssize_t ret;
9958
9959	path = btrfs_alloc_path();
9960	if (!path) {
9961	ret = -ENOMEM;
9962	goto out;
9963	}
9964	ret = btrfs_lookup_file_extent(NULL, root, path, objectid: btrfs_ino(inode),
9965	bytenr: extent_start, mod: 0);
9966	if (ret) {
9967	if (ret > 0) {
9968	/* The extent item disappeared? */
9969	ret = -EIO;
9970	}
9971	goto out;
9972	}
9973	leaf = path->nodes[0];
9974	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9975
9976	ram_bytes = btrfs_file_extent_ram_bytes(eb: leaf, s: item);
9977	ptr = btrfs_file_extent_inline_start(e: item);
9978
9979	encoded->len = min_t(u64, extent_start + ram_bytes,
9980	inode->vfs_inode.i_size) - iocb->ki_pos;
9981	ret = btrfs_encoded_io_compression_from_extent(fs_info,
9982	compress_type: btrfs_file_extent_compression(eb: leaf, s: item));
9983	if (ret < 0)
9984	goto out;
9985	encoded->compression = ret;
9986	if (encoded->compression) {
9987	size_t inline_size;
9988
9989	inline_size = btrfs_file_extent_inline_item_len(eb: leaf,
9990	nr: path->slots[0]);
9991	if (inline_size > count) {
9992	ret = -ENOBUFS;
9993	goto out;
9994	}
9995	count = inline_size;
9996	encoded->unencoded_len = ram_bytes;
9997	encoded->unencoded_offset = iocb->ki_pos - extent_start;
9998	} else {
9999	count = min_t(u64, count, encoded->len);
10000	encoded->len = count;
10001	encoded->unencoded_len = count;
10002	ptr += iocb->ki_pos - extent_start;
10003	}
10004
10005	tmp = kmalloc(size: count, GFP_NOFS);
10006	if (!tmp) {
10007	ret = -ENOMEM;
10008	goto out;
10009	}
10010	read_extent_buffer(eb: leaf, dst: tmp, start: ptr, len: count);
10011	btrfs_release_path(p: path);
10012	unlock_extent(tree: io_tree, start, end: lockend, cached: cached_state);
10013	btrfs_inode_unlock(inode, ilock_flags: BTRFS_ILOCK_SHARED);
10014	*unlocked = true;
10015
10016	ret = copy_to_iter(addr: tmp, bytes: count, i: iter);
10017	if (ret != count)
10018	ret = -EFAULT;
10019	kfree(objp: tmp);
10020	out:
10021	btrfs_free_path(p: path);
10022	return ret;
10023	}
10024
10025	struct btrfs_encoded_read_private {
10026	wait_queue_head_t wait;
10027	atomic_t pending;
10028	blk_status_t status;
10029	};
10030
10031	static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
10032	{
10033	struct btrfs_encoded_read_private *priv = bbio->private;
10034
10035	if (bbio->bio.bi_status) {
10036	/*
10037	* The memory barrier implied by the atomic_dec_return() here
10038	* pairs with the memory barrier implied by the
10039	* atomic_dec_return() or io_wait_event() in
10040	* btrfs_encoded_read_regular_fill_pages() to ensure that this
10041	* write is observed before the load of status in
10042	* btrfs_encoded_read_regular_fill_pages().
10043	*/
10044	WRITE_ONCE(priv->status, bbio->bio.bi_status);
10045	}
10046	if (!atomic_dec_return(v: &priv->pending))
10047	wake_up(&priv->wait);
10048	bio_put(&bbio->bio);
10049	}
10050
10051	int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10052	u64 file_offset, u64 disk_bytenr,
10053	u64 disk_io_size, struct page **pages)
10054	{
10055	struct btrfs_fs_info *fs_info = inode->root->fs_info;
10056	struct btrfs_encoded_read_private priv = {
10057	.pending = ATOMIC_INIT(1),
10058	};
10059	unsigned long i = 0;
10060	struct btrfs_bio *bbio;
10061
10062	init_waitqueue_head(&priv.wait);
10063
10064	bbio = btrfs_bio_alloc(BIO_MAX_VECS, opf: REQ_OP_READ, fs_info,
10065	end_io: btrfs_encoded_read_endio, private: &priv);
10066	bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10067	bbio->inode = inode;
10068
10069	do {
10070	size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
10071
10072	if (bio_add_page(bio: &bbio->bio, page: pages[i], len: bytes, off: 0) < bytes) {
10073	atomic_inc(v: &priv.pending);
10074	btrfs_submit_bio(bbio, mirror_num: 0);
10075
10076	bbio = btrfs_bio_alloc(BIO_MAX_VECS, opf: REQ_OP_READ, fs_info,
10077	end_io: btrfs_encoded_read_endio, private: &priv);
10078	bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10079	bbio->inode = inode;
10080	continue;
10081	}
10082
10083	i++;
10084	disk_bytenr += bytes;
10085	disk_io_size -= bytes;
10086	} while (disk_io_size);
10087
10088	atomic_inc(v: &priv.pending);
10089	btrfs_submit_bio(bbio, mirror_num: 0);
10090
10091	if (atomic_dec_return(v: &priv.pending))
10092	io_wait_event(priv.wait, !atomic_read(&priv.pending));
10093	/* See btrfs_encoded_read_endio() for ordering. */
10094	return blk_status_to_errno(READ_ONCE(priv.status));
10095	}
10096
10097	static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10098	struct iov_iter *iter,
10099	u64 start, u64 lockend,
10100	struct extent_state **cached_state,
10101	u64 disk_bytenr, u64 disk_io_size,
10102	size_t count, bool compressed,
10103	bool *unlocked)
10104	{
10105	struct btrfs_inode *inode = BTRFS_I(inode: file_inode(f: iocb->ki_filp));
10106	struct extent_io_tree *io_tree = &inode->io_tree;
10107	struct page **pages;
10108	unsigned long nr_pages, i;
10109	u64 cur;
10110	size_t page_offset;
10111	ssize_t ret;
10112
10113	nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10114	pages = kcalloc(n: nr_pages, size: sizeof(struct page *), GFP_NOFS);
10115	if (!pages)
10116	return -ENOMEM;
10117	ret = btrfs_alloc_page_array(nr_pages, page_array: pages, extra_gfp: 0);
10118	if (ret) {
10119	ret = -ENOMEM;
10120	goto out;
10121	}
10122
10123	ret = btrfs_encoded_read_regular_fill_pages(inode, file_offset: start, disk_bytenr,
10124	disk_io_size, pages);
10125	if (ret)
10126	goto out;
10127
10128	unlock_extent(tree: io_tree, start, end: lockend, cached: cached_state);
10129	btrfs_inode_unlock(inode, ilock_flags: BTRFS_ILOCK_SHARED);
10130	*unlocked = true;
10131
10132	if (compressed) {
10133	i = 0;
10134	page_offset = 0;
10135	} else {
10136	i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10137	page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10138	}
10139	cur = 0;
10140	while (cur < count) {
10141	size_t bytes = min_t(size_t, count - cur,
10142	PAGE_SIZE - page_offset);
10143
10144	if (copy_page_to_iter(page: pages[i], offset: page_offset, bytes,
10145	i: iter) != bytes) {
10146	ret = -EFAULT;
10147	goto out;
10148	}
10149	i++;
10150	cur += bytes;
10151	page_offset = 0;
10152	}
10153	ret = count;
10154	out:
10155	for (i = 0; i < nr_pages; i++) {
10156	if (pages[i])
10157	__free_page(pages[i]);
10158	}
10159	kfree(objp: pages);
10160	return ret;
10161	}
10162
10163	ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10164	struct btrfs_ioctl_encoded_io_args *encoded)
10165	{
10166	struct btrfs_inode *inode = BTRFS_I(inode: file_inode(f: iocb->ki_filp));
10167	struct btrfs_fs_info *fs_info = inode->root->fs_info;
10168	struct extent_io_tree *io_tree = &inode->io_tree;
10169	ssize_t ret;
10170	size_t count = iov_iter_count(i: iter);
10171	u64 start, lockend, disk_bytenr, disk_io_size;
10172	struct extent_state *cached_state = NULL;
10173	struct extent_map *em;
10174	bool unlocked = false;
10175
10176	file_accessed(file: iocb->ki_filp);
10177
10178	btrfs_inode_lock(inode, ilock_flags: BTRFS_ILOCK_SHARED);
10179
10180	if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10181	btrfs_inode_unlock(inode, ilock_flags: BTRFS_ILOCK_SHARED);
10182	return 0;
10183	}
10184	start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10185	/*
10186	* We don't know how long the extent containing iocb->ki_pos is, but if
10187	* it's compressed we know that it won't be longer than this.
10188	*/
10189	lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10190
10191	for (;;) {
10192	struct btrfs_ordered_extent *ordered;
10193
10194	ret = btrfs_wait_ordered_range(inode: &inode->vfs_inode, start,
10195	len: lockend - start + 1);
10196	if (ret)
10197	goto out_unlock_inode;
10198	lock_extent(tree: io_tree, start, end: lockend, cached: &cached_state);
10199	ordered = btrfs_lookup_ordered_range(inode, file_offset: start,
10200	len: lockend - start + 1);
10201	if (!ordered)
10202	break;
10203	btrfs_put_ordered_extent(entry: ordered);
10204	unlock_extent(tree: io_tree, start, end: lockend, cached: &cached_state);
10205	cond_resched();
10206	}
10207
10208	em = btrfs_get_extent(inode, NULL, start, len: lockend - start + 1);
10209	if (IS_ERR(ptr: em)) {
10210	ret = PTR_ERR(ptr: em);
10211	goto out_unlock_extent;
10212	}
10213
10214	if (em->block_start == EXTENT_MAP_INLINE) {
10215	u64 extent_start = em->start;
10216
10217	/*
10218	* For inline extents we get everything we need out of the
10219	* extent item.
10220	*/
10221	free_extent_map(em);
10222	em = NULL;
10223	ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10224	cached_state: &cached_state, extent_start,
10225	count, encoded, unlocked: &unlocked);
10226	goto out;
10227	}
10228
10229	/*
10230	* We only want to return up to EOF even if the extent extends beyond
10231	* that.
10232	*/
10233	encoded->len = min_t(u64, extent_map_end(em),
10234	inode->vfs_inode.i_size) - iocb->ki_pos;
10235	if (em->block_start == EXTENT_MAP_HOLE ||
10236	(em->flags & EXTENT_FLAG_PREALLOC)) {
10237	disk_bytenr = EXTENT_MAP_HOLE;
10238	count = min_t(u64, count, encoded->len);
10239	encoded->len = count;
10240	encoded->unencoded_len = count;
10241	} else if (extent_map_is_compressed(em)) {
10242	disk_bytenr = em->block_start;
10243	/*
10244	* Bail if the buffer isn't large enough to return the whole
10245	* compressed extent.
10246	*/
10247	if (em->block_len > count) {
10248	ret = -ENOBUFS;
10249	goto out_em;
10250	}
10251	disk_io_size = em->block_len;
10252	count = em->block_len;
10253	encoded->unencoded_len = em->ram_bytes;
10254	encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10255	ret = btrfs_encoded_io_compression_from_extent(fs_info,
10256	compress_type: extent_map_compression(em));
10257	if (ret < 0)
10258	goto out_em;
10259	encoded->compression = ret;
10260	} else {
10261	disk_bytenr = em->block_start + (start - em->start);
10262	if (encoded->len > count)
10263	encoded->len = count;
10264	/*
10265	* Don't read beyond what we locked. This also limits the page
10266	* allocations that we'll do.
10267	*/
10268	disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10269	count = start + disk_io_size - iocb->ki_pos;
10270	encoded->len = count;
10271	encoded->unencoded_len = count;
10272	disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10273	}
10274	free_extent_map(em);
10275	em = NULL;
10276
10277	if (disk_bytenr == EXTENT_MAP_HOLE) {
10278	unlock_extent(tree: io_tree, start, end: lockend, cached: &cached_state);
10279	btrfs_inode_unlock(inode, ilock_flags: BTRFS_ILOCK_SHARED);
10280	unlocked = true;
10281	ret = iov_iter_zero(bytes: count, iter);
10282	if (ret != count)
10283	ret = -EFAULT;
10284	} else {
10285	ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10286	cached_state: &cached_state, disk_bytenr,
10287	disk_io_size, count,
10288	compressed: encoded->compression,
10289	unlocked: &unlocked);
10290	}
10291
10292	out:
10293	if (ret >= 0)
10294	iocb->ki_pos += encoded->len;
10295	out_em:
10296	free_extent_map(em);
10297	out_unlock_extent:
10298	if (!unlocked)
10299	unlock_extent(tree: io_tree, start, end: lockend, cached: &cached_state);
10300	out_unlock_inode:
10301	if (!unlocked)
10302	btrfs_inode_unlock(inode, ilock_flags: BTRFS_ILOCK_SHARED);
10303	return ret;
10304	}
10305
10306	ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10307	const struct btrfs_ioctl_encoded_io_args *encoded)
10308	{
10309	struct btrfs_inode *inode = BTRFS_I(inode: file_inode(f: iocb->ki_filp));
10310	struct btrfs_root *root = inode->root;
10311	struct btrfs_fs_info *fs_info = root->fs_info;
10312	struct extent_io_tree *io_tree = &inode->io_tree;
10313	struct extent_changeset *data_reserved = NULL;
10314	struct extent_state *cached_state = NULL;
10315	struct btrfs_ordered_extent *ordered;
10316	int compression;
10317	size_t orig_count;
10318	u64 start, end;
10319	u64 num_bytes, ram_bytes, disk_num_bytes;
10320	unsigned long nr_pages, i;
10321	struct page **pages;
10322	struct btrfs_key ins;
10323	bool extent_reserved = false;
10324	struct extent_map *em;
10325	ssize_t ret;
10326
10327	switch (encoded->compression) {
10328	case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10329	compression = BTRFS_COMPRESS_ZLIB;
10330	break;
10331	case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10332	compression = BTRFS_COMPRESS_ZSTD;
10333	break;
10334	case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10335	case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10336	case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10337	case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10338	case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10339	/* The sector size must match for LZO. */
10340	if (encoded->compression -
10341	BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10342	fs_info->sectorsize_bits)
10343	return -EINVAL;
10344	compression = BTRFS_COMPRESS_LZO;
10345	break;
10346	default:
10347	return -EINVAL;
10348	}
10349	if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10350	return -EINVAL;
10351
10352	/*
10353	* Compressed extents should always have checksums, so error out if we
10354	* have a NOCOW file or inode was created while mounted with NODATASUM.
10355	*/
10356	if (inode->flags & BTRFS_INODE_NODATASUM)
10357	return -EINVAL;
10358
10359	orig_count = iov_iter_count(i: from);
10360
10361	/* The extent size must be sane. */
10362	if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10363	orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10364	return -EINVAL;
10365
10366	/*
10367	* The compressed data must be smaller than the decompressed data.
10368	*
10369	* It's of course possible for data to compress to larger or the same
10370	* size, but the buffered I/O path falls back to no compression for such
10371	* data, and we don't want to break any assumptions by creating these
10372	* extents.
10373	*
10374	* Note that this is less strict than the current check we have that the
10375	* compressed data must be at least one sector smaller than the
10376	* decompressed data. We only want to enforce the weaker requirement
10377	* from old kernels that it is at least one byte smaller.
10378	*/
10379	if (orig_count >= encoded->unencoded_len)
10380	return -EINVAL;
10381
10382	/* The extent must start on a sector boundary. */
10383	start = iocb->ki_pos;
10384	if (!IS_ALIGNED(start, fs_info->sectorsize))
10385	return -EINVAL;
10386
10387	/*
10388	* The extent must end on a sector boundary. However, we allow a write
10389	* which ends at or extends i_size to have an unaligned length; we round
10390	* up the extent size and set i_size to the unaligned end.
10391	*/
10392	if (start + encoded->len < inode->vfs_inode.i_size &&
10393	!IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10394	return -EINVAL;
10395
10396	/* Finally, the offset in the unencoded data must be sector-aligned. */
10397	if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10398	return -EINVAL;
10399
10400	num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10401	ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10402	end = start + num_bytes - 1;
10403
10404	/*
10405	* If the extent cannot be inline, the compressed data on disk must be
10406	* sector-aligned. For convenience, we extend it with zeroes if it
10407	* isn't.
10408	*/
10409	disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10410	nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10411	pages = kvcalloc(n: nr_pages, size: sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10412	if (!pages)
10413	return -ENOMEM;
10414	for (i = 0; i < nr_pages; i++) {
10415	size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10416	char *kaddr;
10417
10418	pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10419	if (!pages[i]) {
10420	ret = -ENOMEM;
10421	goto out_pages;
10422	}
10423	kaddr = kmap_local_page(page: pages[i]);
10424	if (copy_from_iter(addr: kaddr, bytes, i: from) != bytes) {
10425	kunmap_local(kaddr);
10426	ret = -EFAULT;
10427	goto out_pages;
10428	}
10429	if (bytes < PAGE_SIZE)
10430	memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10431	kunmap_local(kaddr);
10432	}
10433
10434	for (;;) {
10435	struct btrfs_ordered_extent *ordered;
10436
10437	ret = btrfs_wait_ordered_range(inode: &inode->vfs_inode, start, len: num_bytes);
10438	if (ret)
10439	goto out_pages;
10440	ret = invalidate_inode_pages2_range(mapping: inode->vfs_inode.i_mapping,
10441	start: start >> PAGE_SHIFT,
10442	end: end >> PAGE_SHIFT);
10443	if (ret)
10444	goto out_pages;
10445	lock_extent(tree: io_tree, start, end, cached: &cached_state);
10446	ordered = btrfs_lookup_ordered_range(inode, file_offset: start, len: num_bytes);
10447	if (!ordered &&
10448	!filemap_range_has_page(inode->vfs_inode.i_mapping, lstart: start, lend: end))
10449	break;
10450	if (ordered)
10451	btrfs_put_ordered_extent(entry: ordered);
10452	unlock_extent(tree: io_tree, start, end, cached: &cached_state);
10453	cond_resched();
10454	}
10455
10456	/*
10457	* We don't use the higher-level delalloc space functions because our
10458	* num_bytes and disk_num_bytes are different.
10459	*/
10460	ret = btrfs_alloc_data_chunk_ondemand(inode, bytes: disk_num_bytes);
10461	if (ret)
10462	goto out_unlock;
10463	ret = btrfs_qgroup_reserve_data(inode, reserved: &data_reserved, start, len: num_bytes);
10464	if (ret)
10465	goto out_free_data_space;
10466	ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10467	noflush: false);
10468	if (ret)
10469	goto out_qgroup_free_data;
10470
10471	/* Try an inline extent first. */
10472	if (start == 0 && encoded->unencoded_len == encoded->len &&
10473	encoded->unencoded_offset == 0) {
10474	ret = cow_file_range_inline(inode, size: encoded->len, compressed_size: orig_count,
10475	compress_type: compression, compressed_pages: pages, update_i_size: true);
10476	if (ret <= 0) {
10477	if (ret == 0)
10478	ret = orig_count;
10479	goto out_delalloc_release;
10480	}
10481	}
10482
10483	ret = btrfs_reserve_extent(root, ram_bytes: disk_num_bytes, num_bytes: disk_num_bytes,
10484	min_alloc_size: disk_num_bytes, empty_size: 0, hint_byte: 0, ins: &ins, is_data: 1, delalloc: 1);
10485	if (ret)
10486	goto out_delalloc_release;
10487	extent_reserved = true;
10488
10489	em = create_io_em(inode, start, len: num_bytes,
10490	orig_start: start - encoded->unencoded_offset, block_start: ins.objectid,
10491	block_len: ins.offset, orig_block_len: ins.offset, ram_bytes, compress_type: compression,
10492	type: BTRFS_ORDERED_COMPRESSED);
10493	if (IS_ERR(ptr: em)) {
10494	ret = PTR_ERR(ptr: em);
10495	goto out_free_reserved;
10496	}
10497	free_extent_map(em);
10498
10499	ordered = btrfs_alloc_ordered_extent(inode, file_offset: start, num_bytes, ram_bytes,
10500	disk_bytenr: ins.objectid, disk_num_bytes: ins.offset,
10501	offset: encoded->unencoded_offset,
10502	flags: (1 << BTRFS_ORDERED_ENCODED) |
10503	(1 << BTRFS_ORDERED_COMPRESSED),
10504	compress_type: compression);
10505	if (IS_ERR(ptr: ordered)) {
10506	btrfs_drop_extent_map_range(inode, start, end, skip_pinned: false);
10507	ret = PTR_ERR(ptr: ordered);
10508	goto out_free_reserved;
10509	}
10510	btrfs_dec_block_group_reservations(fs_info, start: ins.objectid);
10511
10512	if (start + encoded->len > inode->vfs_inode.i_size)
10513	i_size_write(inode: &inode->vfs_inode, i_size: start + encoded->len);
10514
10515	unlock_extent(tree: io_tree, start, end, cached: &cached_state);
10516
10517	btrfs_delalloc_release_extents(inode, num_bytes);
10518
10519	btrfs_submit_compressed_write(ordered, compressed_pages: pages, nr_pages, write_flags: 0, writeback: false);
10520	ret = orig_count;
10521	goto out;
10522
10523	out_free_reserved:
10524	btrfs_dec_block_group_reservations(fs_info, start: ins.objectid);
10525	btrfs_free_reserved_extent(fs_info, start: ins.objectid, len: ins.offset, delalloc: 1);
10526	out_delalloc_release:
10527	btrfs_delalloc_release_extents(inode, num_bytes);
10528	btrfs_delalloc_release_metadata(inode, num_bytes: disk_num_bytes, qgroup_free: ret < 0);
10529	out_qgroup_free_data:
10530	if (ret < 0)
10531	btrfs_qgroup_free_data(inode, reserved: data_reserved, start, len: num_bytes, NULL);
10532	out_free_data_space:
10533	/*
10534	* If btrfs_reserve_extent() succeeded, then we already decremented
10535	* bytes_may_use.
10536	*/
10537	if (!extent_reserved)
10538	btrfs_free_reserved_data_space_noquota(fs_info, len: disk_num_bytes);
10539	out_unlock:
10540	unlock_extent(tree: io_tree, start, end, cached: &cached_state);
10541	out_pages:
10542	for (i = 0; i < nr_pages; i++) {
10543	if (pages[i])
10544	__free_page(pages[i]);
10545	}
10546	kvfree(addr: pages);
10547	out:
10548	if (ret >= 0)
10549	iocb->ki_pos += encoded->len;
10550	return ret;
10551	}
10552
10553	#ifdef CONFIG_SWAP
10554	/*
10555	* Add an entry indicating a block group or device which is pinned by a
10556	* swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10557	* negative errno on failure.
10558	*/
10559	static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10560	bool is_block_group)
10561	{
10562	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10563	struct btrfs_swapfile_pin *sp, *entry;
10564	struct rb_node **p;
10565	struct rb_node *parent = NULL;
10566
10567	sp = kmalloc(size: sizeof(*sp), GFP_NOFS);
10568	if (!sp)
10569	return -ENOMEM;
10570	sp->ptr = ptr;
10571	sp->inode = inode;
10572	sp->is_block_group = is_block_group;
10573	sp->bg_extent_count = 1;
10574
10575	spin_lock(lock: &fs_info->swapfile_pins_lock);
10576	p = &fs_info->swapfile_pins.rb_node;
10577	while (*p) {
10578	parent = *p;
10579	entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10580	if (sp->ptr < entry->ptr ||
10581	(sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10582	p = &(*p)->rb_left;
10583	} else if (sp->ptr > entry->ptr ||
10584	(sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10585	p = &(*p)->rb_right;
10586	} else {
10587	if (is_block_group)
10588	entry->bg_extent_count++;
10589	spin_unlock(lock: &fs_info->swapfile_pins_lock);
10590	kfree(objp: sp);
10591	return 1;
10592	}
10593	}
10594	rb_link_node(node: &sp->node, parent, rb_link: p);
10595	rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10596	spin_unlock(lock: &fs_info->swapfile_pins_lock);
10597	return 0;
10598	}
10599
10600	/* Free all of the entries pinned by this swapfile. */
10601	static void btrfs_free_swapfile_pins(struct inode *inode)
10602	{
10603	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10604	struct btrfs_swapfile_pin *sp;
10605	struct rb_node *node, *next;
10606
10607	spin_lock(lock: &fs_info->swapfile_pins_lock);
10608	node = rb_first(&fs_info->swapfile_pins);
10609	while (node) {
10610	next = rb_next(node);
10611	sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10612	if (sp->inode == inode) {
10613	rb_erase(&sp->node, &fs_info->swapfile_pins);
10614	if (sp->is_block_group) {
10615	btrfs_dec_block_group_swap_extents(bg: sp->ptr,
10616	amount: sp->bg_extent_count);
10617	btrfs_put_block_group(cache: sp->ptr);
10618	}
10619	kfree(objp: sp);
10620	}
10621	node = next;
10622	}
10623	spin_unlock(lock: &fs_info->swapfile_pins_lock);
10624	}
10625
10626	struct btrfs_swap_info {
10627	u64 start;
10628	u64 block_start;
10629	u64 block_len;
10630	u64 lowest_ppage;
10631	u64 highest_ppage;
10632	unsigned long nr_pages;
10633	int nr_extents;
10634	};
10635
10636	static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10637	struct btrfs_swap_info *bsi)
10638	{
10639	unsigned long nr_pages;
10640	unsigned long max_pages;
10641	u64 first_ppage, first_ppage_reported, next_ppage;
10642	int ret;
10643
10644	/*
10645	* Our swapfile may have had its size extended after the swap header was
10646	* written. In that case activating the swapfile should not go beyond
10647	* the max size set in the swap header.
10648	*/
10649	if (bsi->nr_pages >= sis->max)
10650	return 0;
10651
10652	max_pages = sis->max - bsi->nr_pages;
10653	first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10654	next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10655
10656	if (first_ppage >= next_ppage)
10657	return 0;
10658	nr_pages = next_ppage - first_ppage;
10659	nr_pages = min(nr_pages, max_pages);
10660
10661	first_ppage_reported = first_ppage;
10662	if (bsi->start == 0)
10663	first_ppage_reported++;
10664	if (bsi->lowest_ppage > first_ppage_reported)
10665	bsi->lowest_ppage = first_ppage_reported;
10666	if (bsi->highest_ppage < (next_ppage - 1))
10667	bsi->highest_ppage = next_ppage - 1;
10668
10669	ret = add_swap_extent(sis, start_page: bsi->nr_pages, nr_pages, start_block: first_ppage);
10670	if (ret < 0)
10671	return ret;
10672	bsi->nr_extents += ret;
10673	bsi->nr_pages += nr_pages;
10674	return 0;
10675	}
10676
10677	static void btrfs_swap_deactivate(struct file *file)
10678	{
10679	struct inode *inode = file_inode(f: file);
10680
10681	btrfs_free_swapfile_pins(inode);
10682	atomic_dec(v: &BTRFS_I(inode)->root->nr_swapfiles);
10683	}
10684
10685	static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10686	sector_t *span)
10687	{
10688	struct inode *inode = file_inode(f: file);
10689	struct btrfs_root *root = BTRFS_I(inode)->root;
10690	struct btrfs_fs_info *fs_info = root->fs_info;
10691	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10692	struct extent_state *cached_state = NULL;
10693	struct extent_map *em = NULL;
10694	struct btrfs_chunk_map *map = NULL;
10695	struct btrfs_device *device = NULL;
10696	struct btrfs_swap_info bsi = {
10697	.lowest_ppage = (sector_t)-1ULL,
10698	};
10699	int ret = 0;
10700	u64 isize;
10701	u64 start;
10702
10703	/*
10704	* If the swap file was just created, make sure delalloc is done. If the
10705	* file changes again after this, the user is doing something stupid and
10706	* we don't really care.
10707	*/
10708	ret = btrfs_wait_ordered_range(inode, start: 0, len: (u64)-1);
10709	if (ret)
10710	return ret;
10711
10712	/*
10713	* The inode is locked, so these flags won't change after we check them.
10714	*/
10715	if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10716	btrfs_warn(fs_info, "swapfile must not be compressed");
10717	return -EINVAL;
10718	}
10719	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10720	btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10721	return -EINVAL;
10722	}
10723	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10724	btrfs_warn(fs_info, "swapfile must not be checksummed");
10725	return -EINVAL;
10726	}
10727
10728	/*
10729	* Balance or device remove/replace/resize can move stuff around from
10730	* under us. The exclop protection makes sure they aren't running/won't
10731	* run concurrently while we are mapping the swap extents, and
10732	* fs_info->swapfile_pins prevents them from running while the swap
10733	* file is active and moving the extents. Note that this also prevents
10734	* a concurrent device add which isn't actually necessary, but it's not
10735	* really worth the trouble to allow it.
10736	*/
10737	if (!btrfs_exclop_start(fs_info, type: BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10738	btrfs_warn(fs_info,
10739	"cannot activate swapfile while exclusive operation is running");
10740	return -EBUSY;
10741	}
10742
10743	/*
10744	* Prevent snapshot creation while we are activating the swap file.
10745	* We do not want to race with snapshot creation. If snapshot creation
10746	* already started before we bumped nr_swapfiles from 0 to 1 and
10747	* completes before the first write into the swap file after it is
10748	* activated, than that write would fallback to COW.
10749	*/
10750	if (!btrfs_drew_try_write_lock(lock: &root->snapshot_lock)) {
10751	btrfs_exclop_finish(fs_info);
10752	btrfs_warn(fs_info,
10753	"cannot activate swapfile because snapshot creation is in progress");
10754	return -EINVAL;
10755	}
10756	/*
10757	* Snapshots can create extents which require COW even if NODATACOW is
10758	* set. We use this counter to prevent snapshots. We must increment it
10759	* before walking the extents because we don't want a concurrent
10760	* snapshot to run after we've already checked the extents.
10761	*
10762	* It is possible that subvolume is marked for deletion but still not
10763	* removed yet. To prevent this race, we check the root status before
10764	* activating the swapfile.
10765	*/
10766	spin_lock(lock: &root->root_item_lock);
10767	if (btrfs_root_dead(root)) {
10768	spin_unlock(lock: &root->root_item_lock);
10769
10770	btrfs_exclop_finish(fs_info);
10771	btrfs_warn(fs_info,
10772	"cannot activate swapfile because subvolume %llu is being deleted",
10773	root->root_key.objectid);
10774	return -EPERM;
10775	}
10776	atomic_inc(v: &root->nr_swapfiles);
10777	spin_unlock(lock: &root->root_item_lock);
10778
10779	isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10780
10781	lock_extent(tree: io_tree, start: 0, end: isize - 1, cached: &cached_state);
10782	start = 0;
10783	while (start < isize) {
10784	u64 logical_block_start, physical_block_start;
10785	struct btrfs_block_group *bg;
10786	u64 len = isize - start;
10787
10788	em = btrfs_get_extent(inode: BTRFS_I(inode), NULL, start, len);
10789	if (IS_ERR(ptr: em)) {
10790	ret = PTR_ERR(ptr: em);
10791	goto out;
10792	}
10793
10794	if (em->block_start == EXTENT_MAP_HOLE) {
10795	btrfs_warn(fs_info, "swapfile must not have holes");
10796	ret = -EINVAL;
10797	goto out;
10798	}
10799	if (em->block_start == EXTENT_MAP_INLINE) {
10800	/*
10801	* It's unlikely we'll ever actually find ourselves
10802	* here, as a file small enough to fit inline won't be
10803	* big enough to store more than the swap header, but in
10804	* case something changes in the future, let's catch it
10805	* here rather than later.
10806	*/
10807	btrfs_warn(fs_info, "swapfile must not be inline");
10808	ret = -EINVAL;
10809	goto out;
10810	}
10811	if (extent_map_is_compressed(em)) {
10812	btrfs_warn(fs_info, "swapfile must not be compressed");
10813	ret = -EINVAL;
10814	goto out;
10815	}
10816
10817	logical_block_start = em->block_start + (start - em->start);
10818	len = min(len, em->len - (start - em->start));
10819	free_extent_map(em);
10820	em = NULL;
10821
10822	ret = can_nocow_extent(inode, offset: start, len: &len, NULL, NULL, NULL, nowait: false, strict: true);
10823	if (ret < 0) {
10824	goto out;
10825	} else if (ret) {
10826	ret = 0;
10827	} else {
10828	btrfs_warn(fs_info,
10829	"swapfile must not be copy-on-write");
10830	ret = -EINVAL;
10831	goto out;
10832	}
10833
10834	map = btrfs_get_chunk_map(fs_info, logical: logical_block_start, length: len);
10835	if (IS_ERR(ptr: map)) {
10836	ret = PTR_ERR(ptr: map);
10837	goto out;
10838	}
10839
10840	if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10841	btrfs_warn(fs_info,
10842	"swapfile must have single data profile");
10843	ret = -EINVAL;
10844	goto out;
10845	}
10846
10847	if (device == NULL) {
10848	device = map->stripes[0].dev;
10849	ret = btrfs_add_swapfile_pin(inode, ptr: device, is_block_group: false);
10850	if (ret == 1)
10851	ret = 0;
10852	else if (ret)
10853	goto out;
10854	} else if (device != map->stripes[0].dev) {
10855	btrfs_warn(fs_info, "swapfile must be on one device");
10856	ret = -EINVAL;
10857	goto out;
10858	}
10859
10860	physical_block_start = (map->stripes[0].physical +
10861	(logical_block_start - map->start));
10862	len = min(len, map->chunk_len - (logical_block_start - map->start));
10863	btrfs_free_chunk_map(map);
10864	map = NULL;
10865
10866	bg = btrfs_lookup_block_group(info: fs_info, bytenr: logical_block_start);
10867	if (!bg) {
10868	btrfs_warn(fs_info,
10869	"could not find block group containing swapfile");
10870	ret = -EINVAL;
10871	goto out;
10872	}
10873
10874	if (!btrfs_inc_block_group_swap_extents(bg)) {
10875	btrfs_warn(fs_info,
10876	"block group for swapfile at %llu is read-only%s",
10877	bg->start,
10878	atomic_read(&fs_info->scrubs_running) ?
10879	" (scrub running)" : "");
10880	btrfs_put_block_group(cache: bg);
10881	ret = -EINVAL;
10882	goto out;
10883	}
10884
10885	ret = btrfs_add_swapfile_pin(inode, ptr: bg, is_block_group: true);
10886	if (ret) {
10887	btrfs_put_block_group(cache: bg);
10888	if (ret == 1)
10889	ret = 0;
10890	else
10891	goto out;
10892	}
10893
10894	if (bsi.block_len &&
10895	bsi.block_start + bsi.block_len == physical_block_start) {
10896	bsi.block_len += len;
10897	} else {
10898	if (bsi.block_len) {
10899	ret = btrfs_add_swap_extent(sis, bsi: &bsi);
10900	if (ret)
10901	goto out;
10902	}
10903	bsi.start = start;
10904	bsi.block_start = physical_block_start;
10905	bsi.block_len = len;
10906	}
10907
10908	start += len;
10909	}
10910
10911	if (bsi.block_len)
10912	ret = btrfs_add_swap_extent(sis, bsi: &bsi);
10913
10914	out:
10915	if (!IS_ERR_OR_NULL(ptr: em))
10916	free_extent_map(em);
10917	if (!IS_ERR_OR_NULL(ptr: map))
10918	btrfs_free_chunk_map(map);
10919
10920	unlock_extent(tree: io_tree, start: 0, end: isize - 1, cached: &cached_state);
10921
10922	if (ret)
10923	btrfs_swap_deactivate(file);
10924
10925	btrfs_drew_write_unlock(lock: &root->snapshot_lock);
10926
10927	btrfs_exclop_finish(fs_info);
10928
10929	if (ret)
10930	return ret;
10931
10932	if (device)
10933	sis->bdev = device->bdev;
10934	*span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10935	sis->max = bsi.nr_pages;
10936	sis->pages = bsi.nr_pages - 1;
10937	sis->highest_bit = bsi.nr_pages - 1;
10938	return bsi.nr_extents;
10939	}
10940	#else
10941	static void btrfs_swap_deactivate(struct file *file)
10942	{
10943	}
10944
10945	static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10946	sector_t *span)
10947	{
10948	return -EOPNOTSUPP;
10949	}
10950	#endif
10951
10952	/*
10953	* Update the number of bytes used in the VFS' inode. When we replace extents in
10954	* a range (clone, dedupe, fallocate's zero range), we must update the number of
10955	* bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10956	* always get a correct value.
10957	*/
10958	void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10959	const u64 add_bytes,
10960	const u64 del_bytes)
10961	{
10962	if (add_bytes == del_bytes)
10963	return;
10964
10965	spin_lock(lock: &inode->lock);
10966	if (del_bytes > 0)
10967	inode_sub_bytes(inode: &inode->vfs_inode, bytes: del_bytes);
10968	if (add_bytes > 0)
10969	inode_add_bytes(inode: &inode->vfs_inode, bytes: add_bytes);
10970	spin_unlock(lock: &inode->lock);
10971	}
10972
10973	/*
10974	* Verify that there are no ordered extents for a given file range.
10975	*
10976	* @inode: The target inode.
10977	* @start: Start offset of the file range, should be sector size aligned.
10978	* @end: End offset (inclusive) of the file range, its value +1 should be
10979	* sector size aligned.
10980	*
10981	* This should typically be used for cases where we locked an inode's VFS lock in
10982	* exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10983	* we have flushed all delalloc in the range, we have waited for all ordered
10984	* extents in the range to complete and finally we have locked the file range in
10985	* the inode's io_tree.
10986	*/
10987	void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10988	{
10989	struct btrfs_root *root = inode->root;
10990	struct btrfs_ordered_extent *ordered;
10991
10992	if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10993	return;
10994
10995	ordered = btrfs_lookup_first_ordered_range(inode, file_offset: start, len: end + 1 - start);
10996	if (ordered) {
10997	btrfs_err(root->fs_info,
10998	"found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10999	start, end, btrfs_ino(inode), root->root_key.objectid,
11000	ordered->file_offset,
11001	ordered->file_offset + ordered->num_bytes - 1);
11002	btrfs_put_ordered_extent(entry: ordered);
11003	}
11004
11005	ASSERT(ordered == NULL);
11006	}
11007
11008	static const struct inode_operations btrfs_dir_inode_operations = {
11009	.getattr = btrfs_getattr,
11010	.lookup = btrfs_lookup,
11011	.create = btrfs_create,
11012	.unlink = btrfs_unlink,
11013	.link = btrfs_link,
11014	.mkdir = btrfs_mkdir,
11015	.rmdir = btrfs_rmdir,
11016	.rename = btrfs_rename2,
11017	.symlink = btrfs_symlink,
11018	.setattr = btrfs_setattr,
11019	.mknod = btrfs_mknod,
11020	.listxattr = btrfs_listxattr,
11021	.permission = btrfs_permission,
11022	.get_inode_acl = btrfs_get_acl,
11023	.set_acl = btrfs_set_acl,
11024	.update_time = btrfs_update_time,
11025	.tmpfile = btrfs_tmpfile,
11026	.fileattr_get = btrfs_fileattr_get,
11027	.fileattr_set = btrfs_fileattr_set,
11028	};
11029
11030	static const struct file_operations btrfs_dir_file_operations = {
11031	.llseek = btrfs_dir_llseek,
11032	.read = generic_read_dir,
11033	.iterate_shared = btrfs_real_readdir,
11034	.open = btrfs_opendir,
11035	.unlocked_ioctl = btrfs_ioctl,
11036	#ifdef CONFIG_COMPAT
11037	.compat_ioctl = btrfs_compat_ioctl,
11038	#endif
11039	.release = btrfs_release_file,
11040	.fsync = btrfs_sync_file,
11041	};
11042
11043	/*
11044	* btrfs doesn't support the bmap operation because swapfiles
11045	* use bmap to make a mapping of extents in the file. They assume
11046	* these extents won't change over the life of the file and they
11047	* use the bmap result to do IO directly to the drive.
11048	*
11049	* the btrfs bmap call would return logical addresses that aren't
11050	* suitable for IO and they also will change frequently as COW
11051	* operations happen. So, swapfile + btrfs == corruption.
11052	*
11053	* For now we're avoiding this by dropping bmap.
11054	*/
11055	static const struct address_space_operations btrfs_aops = {
11056	.read_folio = btrfs_read_folio,
11057	.writepages = btrfs_writepages,
11058	.readahead = btrfs_readahead,
11059	.invalidate_folio = btrfs_invalidate_folio,
11060	.release_folio = btrfs_release_folio,
11061	.migrate_folio = btrfs_migrate_folio,
11062	.dirty_folio = filemap_dirty_folio,
11063	.error_remove_folio = generic_error_remove_folio,
11064	.swap_activate = btrfs_swap_activate,
11065	.swap_deactivate = btrfs_swap_deactivate,
11066	};
11067
11068	static const struct inode_operations btrfs_file_inode_operations = {
11069	.getattr = btrfs_getattr,
11070	.setattr = btrfs_setattr,
11071	.listxattr = btrfs_listxattr,
11072	.permission = btrfs_permission,
11073	.fiemap = btrfs_fiemap,
11074	.get_inode_acl = btrfs_get_acl,
11075	.set_acl = btrfs_set_acl,
11076	.update_time = btrfs_update_time,
11077	.fileattr_get = btrfs_fileattr_get,
11078	.fileattr_set = btrfs_fileattr_set,
11079	};
11080	static const struct inode_operations btrfs_special_inode_operations = {
11081	.getattr = btrfs_getattr,
11082	.setattr = btrfs_setattr,
11083	.permission = btrfs_permission,
11084	.listxattr = btrfs_listxattr,
11085	.get_inode_acl = btrfs_get_acl,
11086	.set_acl = btrfs_set_acl,
11087	.update_time = btrfs_update_time,
11088	};
11089	static const struct inode_operations btrfs_symlink_inode_operations = {
11090	.get_link = page_get_link,
11091	.getattr = btrfs_getattr,
11092	.setattr = btrfs_setattr,
11093	.permission = btrfs_permission,
11094	.listxattr = btrfs_listxattr,
11095	.update_time = btrfs_update_time,
11096	};
11097
11098	const struct dentry_operations btrfs_dentry_operations = {
11099	.d_delete = btrfs_dentry_delete,
11100	};
11101