1	// SPDX-License-Identifier: GPL-2.0
2	/*
3	* linux/fs/ext4/inode.c
4	*
5	* Copyright (C) 1992, 1993, 1994, 1995
6	* Remy Card (card@masi.ibp.fr)
7	* Laboratoire MASI - Institut Blaise Pascal
8	* Universite Pierre et Marie Curie (Paris VI)
9	*
10	* from
11	*
12	* linux/fs/minix/inode.c
13	*
14	* Copyright (C) 1991, 1992 Linus Torvalds
15	*
16	* 64-bit file support on 64-bit platforms by Jakub Jelinek
17	* (jj@sunsite.ms.mff.cuni.cz)
18	*
19	* Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
20	*/
21
22	#include <linux/fs.h>
23	#include <linux/mount.h>
24	#include <linux/time.h>
25	#include <linux/highuid.h>
26	#include <linux/pagemap.h>
27	#include <linux/dax.h>
28	#include <linux/quotaops.h>
29	#include <linux/string.h>
30	#include <linux/buffer_head.h>
31	#include <linux/writeback.h>
32	#include <linux/pagevec.h>
33	#include <linux/mpage.h>
34	#include <linux/rmap.h>
35	#include <linux/namei.h>
36	#include <linux/uio.h>
37	#include <linux/bio.h>
38	#include <linux/workqueue.h>
39	#include <linux/kernel.h>
40	#include <linux/printk.h>
41	#include <linux/slab.h>
42	#include <linux/bitops.h>
43	#include <linux/iomap.h>
44	#include <linux/iversion.h>
45
46	#include "ext4_jbd2.h"
47	#include "xattr.h"
48	#include "acl.h"
49	#include "truncate.h"
50
51	#include <trace/events/ext4.h>
52
53	static void ext4_journalled_zero_new_buffers(handle_t *handle,
54	struct inode *inode,
55	struct folio *folio,
56	unsigned from, unsigned to);
57
58	static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
59	struct ext4_inode_info *ei)
60	{
61	__u32 csum;
62	__u16 dummy_csum = 0;
63	int offset = offsetof(struct ext4_inode, i_checksum_lo);
64	unsigned int csum_size = sizeof(dummy_csum);
65
66	csum = ext4_chksum(crc: ei->i_csum_seed, address: (__u8 *)raw, length: offset);
67	csum = ext4_chksum(crc: csum, address: (__u8 *)&dummy_csum, length: csum_size);
68	offset += csum_size;
69	csum = ext4_chksum(crc: csum, address: (__u8 *)raw + offset,
70	EXT4_GOOD_OLD_INODE_SIZE - offset);
71
72	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
73	offset = offsetof(struct ext4_inode, i_checksum_hi);
74	csum = ext4_chksum(crc: csum, address: (__u8 *)raw + EXT4_GOOD_OLD_INODE_SIZE,
75	length: offset - EXT4_GOOD_OLD_INODE_SIZE);
76	if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
77	csum = ext4_chksum(crc: csum, address: (__u8 *)&dummy_csum,
78	length: csum_size);
79	offset += csum_size;
80	}
81	csum = ext4_chksum(crc: csum, address: (__u8 *)raw + offset,
82	EXT4_INODE_SIZE(inode->i_sb) - offset);
83	}
84
85	return csum;
86	}
87
88	static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
89	struct ext4_inode_info *ei)
90	{
91	__u32 provided, calculated;
92
93	if (EXT4_SB(sb: inode->i_sb)->s_es->s_creator_os !=
94	cpu_to_le32(EXT4_OS_LINUX) ||
95	!ext4_has_feature_metadata_csum(sb: inode->i_sb))
96	return 1;
97
98	provided = le16_to_cpu(raw->i_checksum_lo);
99	calculated = ext4_inode_csum(inode, raw, ei);
100	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
101	EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
102	provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
103	else
104	calculated &= 0xFFFF;
105
106	return provided == calculated;
107	}
108
109	void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
110	struct ext4_inode_info *ei)
111	{
112	__u32 csum;
113
114	if (EXT4_SB(sb: inode->i_sb)->s_es->s_creator_os !=
115	cpu_to_le32(EXT4_OS_LINUX) ||
116	!ext4_has_feature_metadata_csum(sb: inode->i_sb))
117	return;
118
119	csum = ext4_inode_csum(inode, raw, ei);
120	raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
121	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
122	EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
123	raw->i_checksum_hi = cpu_to_le16(csum >> 16);
124	}
125
126	static inline int ext4_begin_ordered_truncate(struct inode *inode,
127	loff_t new_size)
128	{
129	trace_ext4_begin_ordered_truncate(inode, new_size);
130	/*
131	* If jinode is zero, then we never opened the file for
132	* writing, so there's no need to call
133	* jbd2_journal_begin_ordered_truncate() since there's no
134	* outstanding writes we need to flush.
135	*/
136	if (!EXT4_I(inode)->jinode)
137	return 0;
138	return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
139	EXT4_I(inode)->jinode,
140	new_size);
141	}
142
143	/*
144	* Test whether an inode is a fast symlink.
145	* A fast symlink has its symlink data stored in ext4_inode_info->i_data.
146	*/
147	int ext4_inode_is_fast_symlink(struct inode *inode)
148	{
149	if (!ext4_has_feature_ea_inode(sb: inode->i_sb)) {
150	int ea_blocks = EXT4_I(inode)->i_file_acl ?
151	EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
152
153	if (ext4_has_inline_data(inode))
154	return 0;
155
156	return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
157	}
158	return S_ISLNK(inode->i_mode) && inode->i_size &&
159	(inode->i_size < EXT4_N_BLOCKS * 4);
160	}
161
162	/*
163	* Called at the last iput() if i_nlink is zero.
164	*/
165	void ext4_evict_inode(struct inode *inode)
166	{
167	handle_t *handle;
168	int err;
169	/*
170	* Credits for final inode cleanup and freeing:
171	* sb + inode (ext4_orphan_del()), block bitmap, group descriptor
172	* (xattr block freeing), bitmap, group descriptor (inode freeing)
173	*/
174	int extra_credits = 6;
175	struct ext4_xattr_inode_array *ea_inode_array = NULL;
176	bool freeze_protected = false;
177
178	trace_ext4_evict_inode(inode);
179
180	dax_break_layout_final(inode);
181
182	if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)
183	ext4_evict_ea_inode(inode);
184	if (inode->i_nlink) {
185	truncate_inode_pages_final(mapping: &inode->i_data);
186
187	goto no_delete;
188	}
189
190	if (is_bad_inode(inode))
191	goto no_delete;
192	dquot_initialize(inode);
193
194	if (ext4_should_order_data(inode))
195	ext4_begin_ordered_truncate(inode, new_size: 0);
196	truncate_inode_pages_final(mapping: &inode->i_data);
197
198	/*
199	* For inodes with journalled data, transaction commit could have
200	* dirtied the inode. And for inodes with dioread_nolock, unwritten
201	* extents converting worker could merge extents and also have dirtied
202	* the inode. Flush worker is ignoring it because of I_FREEING flag but
203	* we still need to remove the inode from the writeback lists.
204	*/
205	inode_io_list_del(inode);
206
207	/*
208	* Protect us against freezing - iput() caller didn't have to have any
209	* protection against it. When we are in a running transaction though,
210	* we are already protected against freezing and we cannot grab further
211	* protection due to lock ordering constraints.
212	*/
213	if (!ext4_journal_current_handle()) {
214	sb_start_intwrite(sb: inode->i_sb);
215	freeze_protected = true;
216	}
217
218	if (!IS_NOQUOTA(inode))
219	extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
220
221	/*
222	* Block bitmap, group descriptor, and inode are accounted in both
223	* ext4_blocks_for_truncate() and extra_credits. So subtract 3.
224	*/
225	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
226	ext4_blocks_for_truncate(inode) + extra_credits - 3);
227	if (IS_ERR(ptr: handle)) {
228	ext4_std_error(inode->i_sb, PTR_ERR(handle));
229	/*
230	* If we're going to skip the normal cleanup, we still need to
231	* make sure that the in-core orphan linked list is properly
232	* cleaned up.
233	*/
234	ext4_orphan_del(NULL, inode);
235	if (freeze_protected)
236	sb_end_intwrite(sb: inode->i_sb);
237	goto no_delete;
238	}
239
240	if (IS_SYNC(inode))
241	ext4_handle_sync(handle);
242
243	/*
244	* Set inode->i_size to 0 before calling ext4_truncate(). We need
245	* special handling of symlinks here because i_size is used to
246	* determine whether ext4_inode_info->i_data contains symlink data or
247	* block mappings. Setting i_size to 0 will remove its fast symlink
248	* status. Erase i_data so that it becomes a valid empty block map.
249	*/
250	if (ext4_inode_is_fast_symlink(inode))
251	memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
252	inode->i_size = 0;
253	err = ext4_mark_inode_dirty(handle, inode);
254	if (err) {
255	ext4_warning(inode->i_sb,
256	"couldn't mark inode dirty (err %d)", err);
257	goto stop_handle;
258	}
259	if (inode->i_blocks) {
260	err = ext4_truncate(inode);
261	if (err) {
262	ext4_error_err(inode->i_sb, -err,
263	"couldn't truncate inode %lu (err %d)",
264	inode->i_ino, err);
265	goto stop_handle;
266	}
267	}
268
269	/* Remove xattr references. */
270	err = ext4_xattr_delete_inode(handle, inode, array: &ea_inode_array,
271	extra_credits);
272	if (err) {
273	ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
274	stop_handle:
275	ext4_journal_stop(handle);
276	ext4_orphan_del(NULL, inode);
277	if (freeze_protected)
278	sb_end_intwrite(sb: inode->i_sb);
279	ext4_xattr_inode_array_free(array: ea_inode_array);
280	goto no_delete;
281	}
282
283	/*
284	* Kill off the orphan record which ext4_truncate created.
285	* AKPM: I think this can be inside the above `if'.
286	* Note that ext4_orphan_del() has to be able to cope with the
287	* deletion of a non-existent orphan - this is because we don't
288	* know if ext4_truncate() actually created an orphan record.
289	* (Well, we could do this if we need to, but heck - it works)
290	*/
291	ext4_orphan_del(handle, inode);
292	EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
293
294	/*
295	* One subtle ordering requirement: if anything has gone wrong
296	* (transaction abort, IO errors, whatever), then we can still
297	* do these next steps (the fs will already have been marked as
298	* having errors), but we can't free the inode if the mark_dirty
299	* fails.
300	*/
301	if (ext4_mark_inode_dirty(handle, inode))
302	/* If that failed, just do the required in-core inode clear. */
303	ext4_clear_inode(inode);
304	else
305	ext4_free_inode(handle, inode);
306	ext4_journal_stop(handle);
307	if (freeze_protected)
308	sb_end_intwrite(sb: inode->i_sb);
309	ext4_xattr_inode_array_free(array: ea_inode_array);
310	return;
311	no_delete:
312	/*
313	* Check out some where else accidentally dirty the evicting inode,
314	* which may probably cause inode use-after-free issues later.
315	*/
316	WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list));
317
318	if (!list_empty(head: &EXT4_I(inode)->i_fc_list))
319	ext4_fc_mark_ineligible(sb: inode->i_sb, reason: EXT4_FC_REASON_NOMEM, NULL);
320	ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
321	}
322
323	#ifdef CONFIG_QUOTA
324	qsize_t *ext4_get_reserved_space(struct inode *inode)
325	{
326	return &EXT4_I(inode)->i_reserved_quota;
327	}
328	#endif
329
330	/*
331	* Called with i_data_sem down, which is important since we can call
332	* ext4_discard_preallocations() from here.
333	*/
334	void ext4_da_update_reserve_space(struct inode *inode,
335	int used, int quota_claim)
336	{
337	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
338	struct ext4_inode_info *ei = EXT4_I(inode);
339
340	spin_lock(lock: &ei->i_block_reservation_lock);
341	trace_ext4_da_update_reserve_space(inode, used_blocks: used, quota_claim);
342	if (unlikely(used > ei->i_reserved_data_blocks)) {
343	ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
344	"with only %d reserved data blocks",
345	__func__, inode->i_ino, used,
346	ei->i_reserved_data_blocks);
347	WARN_ON(1);
348	used = ei->i_reserved_data_blocks;
349	}
350
351	/* Update per-inode reservations */
352	ei->i_reserved_data_blocks -= used;
353	percpu_counter_sub(fbc: &sbi->s_dirtyclusters_counter, amount: used);
354
355	spin_unlock(lock: &ei->i_block_reservation_lock);
356
357	/* Update quota subsystem for data blocks */
358	if (quota_claim)
359	dquot_claim_block(inode, EXT4_C2B(sbi, used));
360	else {
361	/*
362	* We did fallocate with an offset that is already delayed
363	* allocated. So on delayed allocated writeback we should
364	* not re-claim the quota for fallocated blocks.
365	*/
366	dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
367	}
368
369	/*
370	* If we have done all the pending block allocations and if
371	* there aren't any writers on the inode, we can discard the
372	* inode's preallocations.
373	*/
374	if ((ei->i_reserved_data_blocks == 0) &&
375	!inode_is_open_for_write(inode))
376	ext4_discard_preallocations(inode);
377	}
378
379	static int __check_block_validity(struct inode *inode, const char *func,
380	unsigned int line,
381	struct ext4_map_blocks *map)
382	{
383	journal_t *journal = EXT4_SB(sb: inode->i_sb)->s_journal;
384
385	if (journal && inode == journal->j_inode)
386	return 0;
387
388	if (!ext4_inode_block_valid(inode, start_blk: map->m_pblk, count: map->m_len)) {
389	ext4_error_inode(inode, func, line, map->m_pblk,
390	"lblock %lu mapped to illegal pblock %llu "
391	"(length %d)", (unsigned long) map->m_lblk,
392	map->m_pblk, map->m_len);
393	return -EFSCORRUPTED;
394	}
395	return 0;
396	}
397
398	int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
399	ext4_lblk_t len)
400	{
401	int ret;
402
403	if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
404	return fscrypt_zeroout_range(inode, lblk, pblk, len);
405
406	ret = sb_issue_zeroout(sb: inode->i_sb, block: pblk, nr_blocks: len, GFP_NOFS);
407	if (ret > 0)
408	ret = 0;
409
410	return ret;
411	}
412
413	/*
414	* For generic regular files, when updating the extent tree, Ext4 should
415	* hold the i_rwsem and invalidate_lock exclusively. This ensures
416	* exclusion against concurrent page faults, as well as reads and writes.
417	*/
418	#ifdef CONFIG_EXT4_DEBUG
419	void ext4_check_map_extents_env(struct inode *inode)
420	{
421	if (EXT4_SB(sb: inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
422	return;
423
424	if (!S_ISREG(inode->i_mode) ||
425	IS_NOQUOTA(inode) || IS_VERITY(inode) ||
426	is_special_ino(sb: inode->i_sb, ino: inode->i_ino) ||
427	(inode_state_read_once(inode) & (I_FREEING | I_WILL_FREE | I_NEW)) ||
428	ext4_test_inode_flag(inode, bit: EXT4_INODE_EA_INODE) ||
429	ext4_verity_in_progress(inode))
430	return;
431
432	WARN_ON_ONCE(!inode_is_locked(inode) &&
433	!rwsem_is_locked(&inode->i_mapping->invalidate_lock));
434	}
435	#else
436	void ext4_check_map_extents_env(struct inode *inode) {}
437	#endif
438
439	#define check_block_validity(inode, map) \
440	__check_block_validity((inode), __func__, __LINE__, (map))
441
442	#ifdef ES_AGGRESSIVE_TEST
443	static void ext4_map_blocks_es_recheck(handle_t *handle,
444	struct inode *inode,
445	struct ext4_map_blocks *es_map,
446	struct ext4_map_blocks *map,
447	int flags)
448	{
449	int retval;
450
451	map->m_flags = 0;
452	/*
453	* There is a race window that the result is not the same.
454	* e.g. xfstests #223 when dioread_nolock enables. The reason
455	* is that we lookup a block mapping in extent status tree with
456	* out taking i_data_sem. So at the time the unwritten extent
457	* could be converted.
458	*/
459	down_read(&EXT4_I(inode)->i_data_sem);
460	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
461	retval = ext4_ext_map_blocks(handle, inode, map, 0);
462	} else {
463	retval = ext4_ind_map_blocks(handle, inode, map, 0);
464	}
465	up_read((&EXT4_I(inode)->i_data_sem));
466
467	/*
468	* We don't check m_len because extent will be collpased in status
469	* tree. So the m_len might not equal.
470	*/
471	if (es_map->m_lblk != map->m_lblk ||
472	es_map->m_flags != map->m_flags ||
473	es_map->m_pblk != map->m_pblk) {
474	printk("ES cache assertion failed for inode: %lu "
475	"es_cached ex [%d/%d/%llu/%x] != "
476	"found ex [%d/%d/%llu/%x] retval %d flags %x\n",
477	inode->i_ino, es_map->m_lblk, es_map->m_len,
478	es_map->m_pblk, es_map->m_flags, map->m_lblk,
479	map->m_len, map->m_pblk, map->m_flags,
480	retval, flags);
481	}
482	}
483	#endif /* ES_AGGRESSIVE_TEST */
484
485	static int ext4_map_query_blocks_next_in_leaf(handle_t *handle,
486	struct inode *inode, struct ext4_map_blocks *map,
487	unsigned int orig_mlen)
488	{
489	struct ext4_map_blocks map2;
490	unsigned int status, status2;
491	int retval;
492
493	status = map->m_flags & EXT4_MAP_UNWRITTEN ?
494	EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
495
496	WARN_ON_ONCE(!(map->m_flags & EXT4_MAP_QUERY_LAST_IN_LEAF));
497	WARN_ON_ONCE(orig_mlen <= map->m_len);
498
499	/* Prepare map2 for lookup in next leaf block */
500	map2.m_lblk = map->m_lblk + map->m_len;
501	map2.m_len = orig_mlen - map->m_len;
502	map2.m_flags = 0;
503	retval = ext4_ext_map_blocks(handle, inode, map: &map2, flags: 0);
504
505	if (retval <= 0) {
506	ext4_es_insert_extent(inode, lblk: map->m_lblk, len: map->m_len,
507	pblk: map->m_pblk, status, delalloc_reserve_used: false);
508	return map->m_len;
509	}
510
511	if (unlikely(retval != map2.m_len)) {
512	ext4_warning(inode->i_sb,
513	"ES len assertion failed for inode "
514	"%lu: retval %d != map->m_len %d",
515	inode->i_ino, retval, map2.m_len);
516	WARN_ON(1);
517	}
518
519	status2 = map2.m_flags & EXT4_MAP_UNWRITTEN ?
520	EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
521
522	/*
523	* If map2 is contiguous with map, then let's insert it as a single
524	* extent in es cache and return the combined length of both the maps.
525	*/
526	if (map->m_pblk + map->m_len == map2.m_pblk &&
527	status == status2) {
528	ext4_es_insert_extent(inode, lblk: map->m_lblk,
529	len: map->m_len + map2.m_len, pblk: map->m_pblk,
530	status, delalloc_reserve_used: false);
531	map->m_len += map2.m_len;
532	} else {
533	ext4_es_insert_extent(inode, lblk: map->m_lblk, len: map->m_len,
534	pblk: map->m_pblk, status, delalloc_reserve_used: false);
535	}
536
537	return map->m_len;
538	}
539
540	static int ext4_map_query_blocks(handle_t *handle, struct inode *inode,
541	struct ext4_map_blocks *map, int flags)
542	{
543	unsigned int status;
544	int retval;
545	unsigned int orig_mlen = map->m_len;
546
547	flags &= EXT4_EX_QUERY_FILTER;
548	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
549	retval = ext4_ext_map_blocks(handle, inode, map, flags);
550	else
551	retval = ext4_ind_map_blocks(handle, inode, map, flags);
552	if (retval < 0)
553	return retval;
554
555	/* A hole? */
556	if (retval == 0)
557	goto out;
558
559	if (unlikely(retval != map->m_len)) {
560	ext4_warning(inode->i_sb,
561	"ES len assertion failed for inode "
562	"%lu: retval %d != map->m_len %d",
563	inode->i_ino, retval, map->m_len);
564	WARN_ON(1);
565	}
566
567	/*
568	* No need to query next in leaf:
569	* - if returned extent is not last in leaf or
570	* - if the last in leaf is the full requested range
571	*/
572	if (!(map->m_flags & EXT4_MAP_QUERY_LAST_IN_LEAF) ||
573	map->m_len == orig_mlen) {
574	status = map->m_flags & EXT4_MAP_UNWRITTEN ?
575	EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
576	ext4_es_insert_extent(inode, lblk: map->m_lblk, len: map->m_len,
577	pblk: map->m_pblk, status, delalloc_reserve_used: false);
578	} else {
579	retval = ext4_map_query_blocks_next_in_leaf(handle, inode, map,
580	orig_mlen);
581	}
582	out:
583	map->m_seq = READ_ONCE(EXT4_I(inode)->i_es_seq);
584	return retval;
585	}
586
587	static int ext4_map_create_blocks(handle_t *handle, struct inode *inode,
588	struct ext4_map_blocks *map, int flags)
589	{
590	struct extent_status es;
591	unsigned int status;
592	int err, retval = 0;
593
594	/*
595	* We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE
596	* indicates that the blocks and quotas has already been
597	* checked when the data was copied into the page cache.
598	*/
599	if (map->m_flags & EXT4_MAP_DELAYED)
600	flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
601
602	/*
603	* Here we clear m_flags because after allocating an new extent,
604	* it will be set again.
605	*/
606	map->m_flags &= ~EXT4_MAP_FLAGS;
607
608	/*
609	* We need to check for EXT4 here because migrate could have
610	* changed the inode type in between.
611	*/
612	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS)) {
613	retval = ext4_ext_map_blocks(handle, inode, map, flags);
614	} else {
615	retval = ext4_ind_map_blocks(handle, inode, map, flags);
616
617	/*
618	* We allocated new blocks which will result in i_data's
619	* format changing. Force the migrate to fail by clearing
620	* migrate flags.
621	*/
622	if (retval > 0 && map->m_flags & EXT4_MAP_NEW)
623	ext4_clear_inode_state(inode, bit: EXT4_STATE_EXT_MIGRATE);
624	}
625	if (retval <= 0)
626	return retval;
627
628	if (unlikely(retval != map->m_len)) {
629	ext4_warning(inode->i_sb,
630	"ES len assertion failed for inode %lu: "
631	"retval %d != map->m_len %d",
632	inode->i_ino, retval, map->m_len);
633	WARN_ON(1);
634	}
635
636	/*
637	* We have to zeroout blocks before inserting them into extent
638	* status tree. Otherwise someone could look them up there and
639	* use them before they are really zeroed. We also have to
640	* unmap metadata before zeroing as otherwise writeback can
641	* overwrite zeros with stale data from block device.
642	*/
643	if (flags & EXT4_GET_BLOCKS_ZERO &&
644	map->m_flags & EXT4_MAP_MAPPED && map->m_flags & EXT4_MAP_NEW) {
645	err = ext4_issue_zeroout(inode, lblk: map->m_lblk, pblk: map->m_pblk,
646	len: map->m_len);
647	if (err)
648	return err;
649	}
650
651	/*
652	* If the extent has been zeroed out, we don't need to update
653	* extent status tree.
654	*/
655	if (flags & EXT4_GET_BLOCKS_SPLIT_NOMERGE &&
656	ext4_es_lookup_extent(inode, lblk: map->m_lblk, NULL, es: &es, pseq: &map->m_seq)) {
657	if (ext4_es_is_written(es: &es))
658	return retval;
659	}
660
661	status = map->m_flags & EXT4_MAP_UNWRITTEN ?
662	EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
663	ext4_es_insert_extent(inode, lblk: map->m_lblk, len: map->m_len, pblk: map->m_pblk,
664	status, delalloc_reserve_used: flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE);
665	map->m_seq = READ_ONCE(EXT4_I(inode)->i_es_seq);
666
667	return retval;
668	}
669
670	/*
671	* The ext4_map_blocks() function tries to look up the requested blocks,
672	* and returns if the blocks are already mapped.
673	*
674	* Otherwise it takes the write lock of the i_data_sem and allocate blocks
675	* and store the allocated blocks in the result buffer head and mark it
676	* mapped.
677	*
678	* If file type is extents based, it will call ext4_ext_map_blocks(),
679	* Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
680	* based files
681	*
682	* On success, it returns the number of blocks being mapped or allocated.
683	* If flags doesn't contain EXT4_GET_BLOCKS_CREATE the blocks are
684	* pre-allocated and unwritten, the resulting @map is marked as unwritten.
685	* If the flags contain EXT4_GET_BLOCKS_CREATE, it will mark @map as mapped.
686	*
687	* It returns 0 if plain look up failed (blocks have not been allocated), in
688	* that case, @map is returned as unmapped but we still do fill map->m_len to
689	* indicate the length of a hole starting at map->m_lblk.
690	*
691	* It returns the error in case of allocation failure.
692	*/
693	int ext4_map_blocks(handle_t *handle, struct inode *inode,
694	struct ext4_map_blocks *map, int flags)
695	{
696	struct extent_status es;
697	int retval;
698	int ret = 0;
699	unsigned int orig_mlen = map->m_len;
700	#ifdef ES_AGGRESSIVE_TEST
701	struct ext4_map_blocks orig_map;
702
703	memcpy(&orig_map, map, sizeof(*map));
704	#endif
705
706	map->m_flags = 0;
707	ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
708	flags, map->m_len, (unsigned long) map->m_lblk);
709
710	/*
711	* ext4_map_blocks returns an int, and m_len is an unsigned int
712	*/
713	if (unlikely(map->m_len > INT_MAX))
714	map->m_len = INT_MAX;
715
716	/* We can handle the block number less than EXT_MAX_BLOCKS */
717	if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
718	return -EFSCORRUPTED;
719
720	/*
721	* Callers from the context of data submission are the only exceptions
722	* for regular files that do not hold the i_rwsem or invalidate_lock.
723	* However, caching unrelated ranges is not permitted.
724	*/
725	if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
726	WARN_ON_ONCE(!(flags & EXT4_EX_NOCACHE));
727	else
728	ext4_check_map_extents_env(inode);
729
730	/* Lookup extent status tree firstly */
731	if (ext4_es_lookup_extent(inode, lblk: map->m_lblk, NULL, es: &es, pseq: &map->m_seq)) {
732	if (ext4_es_is_written(es: &es) || ext4_es_is_unwritten(es: &es)) {
733	map->m_pblk = ext4_es_pblock(es: &es) +
734	map->m_lblk - es.es_lblk;
735	map->m_flags |= ext4_es_is_written(es: &es) ?
736	EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
737	retval = es.es_len - (map->m_lblk - es.es_lblk);
738	if (retval > map->m_len)
739	retval = map->m_len;
740	map->m_len = retval;
741	} else if (ext4_es_is_delayed(es: &es) || ext4_es_is_hole(es: &es)) {
742	map->m_pblk = 0;
743	map->m_flags |= ext4_es_is_delayed(es: &es) ?
744	EXT4_MAP_DELAYED : 0;
745	retval = es.es_len - (map->m_lblk - es.es_lblk);
746	if (retval > map->m_len)
747	retval = map->m_len;
748	map->m_len = retval;
749	retval = 0;
750	} else {
751	BUG();
752	}
753
754	if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
755	return retval;
756	#ifdef ES_AGGRESSIVE_TEST
757	ext4_map_blocks_es_recheck(handle, inode, map,
758	&orig_map, flags);
759	#endif
760	if (!(flags & EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF) ||
761	orig_mlen == map->m_len)
762	goto found;
763
764	map->m_len = orig_mlen;
765	}
766	/*
767	* In the query cache no-wait mode, nothing we can do more if we
768	* cannot find extent in the cache.
769	*/
770	if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
771	return 0;
772
773	/*
774	* Try to see if we can get the block without requesting a new
775	* file system block.
776	*/
777	down_read(sem: &EXT4_I(inode)->i_data_sem);
778	retval = ext4_map_query_blocks(handle, inode, map, flags);
779	up_read(sem: (&EXT4_I(inode)->i_data_sem));
780
781	found:
782	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
783	ret = check_block_validity(inode, map);
784	if (ret != 0)
785	return ret;
786	}
787
788	/* If it is only a block(s) look up */
789	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
790	return retval;
791
792	/*
793	* Returns if the blocks have already allocated
794	*
795	* Note that if blocks have been preallocated
796	* ext4_ext_map_blocks() returns with buffer head unmapped
797	*/
798	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
799	/*
800	* If we need to convert extent to unwritten
801	* we continue and do the actual work in
802	* ext4_ext_map_blocks()
803	*/
804	if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
805	return retval;
806
807
808	ext4_fc_track_inode(handle, inode);
809	/*
810	* New blocks allocate and/or writing to unwritten extent
811	* will possibly result in updating i_data, so we take
812	* the write lock of i_data_sem, and call get_block()
813	* with create == 1 flag.
814	*/
815	down_write(sem: &EXT4_I(inode)->i_data_sem);
816	retval = ext4_map_create_blocks(handle, inode, map, flags);
817	up_write(sem: (&EXT4_I(inode)->i_data_sem));
818
819	if (retval < 0)
820	ext_debug(inode, "failed with err %d\n", retval);
821	if (retval <= 0)
822	return retval;
823
824	if (map->m_flags & EXT4_MAP_MAPPED) {
825	ret = check_block_validity(inode, map);
826	if (ret != 0)
827	return ret;
828
829	/*
830	* Inodes with freshly allocated blocks where contents will be
831	* visible after transaction commit must be on transaction's
832	* ordered data list.
833	*/
834	if (map->m_flags & EXT4_MAP_NEW &&
835	!(map->m_flags & EXT4_MAP_UNWRITTEN) &&
836	!(flags & EXT4_GET_BLOCKS_ZERO) &&
837	!ext4_is_quota_file(inode) &&
838	ext4_should_order_data(inode)) {
839	loff_t start_byte = EXT4_LBLK_TO_B(inode, map->m_lblk);
840	loff_t length = EXT4_LBLK_TO_B(inode, map->m_len);
841
842	if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
843	ret = ext4_jbd2_inode_add_wait(handle, inode,
844	start_byte, length);
845	else
846	ret = ext4_jbd2_inode_add_write(handle, inode,
847	start_byte, length);
848	if (ret)
849	return ret;
850	}
851	}
852	ext4_fc_track_range(handle, inode, start: map->m_lblk, end: map->m_lblk +
853	map->m_len - 1);
854	return retval;
855	}
856
857	/*
858	* Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
859	* we have to be careful as someone else may be manipulating b_state as well.
860	*/
861	static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
862	{
863	unsigned long old_state;
864	unsigned long new_state;
865
866	flags &= EXT4_MAP_FLAGS;
867
868	/* Dummy buffer_head? Set non-atomically. */
869	if (!bh->b_folio) {
870	bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
871	return;
872	}
873	/*
874	* Someone else may be modifying b_state. Be careful! This is ugly but
875	* once we get rid of using bh as a container for mapping information
876	* to pass to / from get_block functions, this can go away.
877	*/
878	old_state = READ_ONCE(bh->b_state);
879	do {
880	new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
881	} while (unlikely(!try_cmpxchg(&bh->b_state, &old_state, new_state)));
882	}
883
884	/*
885	* Make sure that the current journal transaction has enough credits to map
886	* one extent. Return -EAGAIN if it cannot extend the current running
887	* transaction.
888	*/
889	static inline int ext4_journal_ensure_extent_credits(handle_t *handle,
890	struct inode *inode)
891	{
892	int credits;
893	int ret;
894
895	/* Called from ext4_da_write_begin() which has no handle started? */
896	if (!handle)
897	return 0;
898
899	credits = ext4_chunk_trans_blocks(inode, nrblocks: 1);
900	ret = __ext4_journal_ensure_credits(handle, check_cred: credits, extend_cred: credits, revoke_cred: 0);
901	return ret <= 0 ? ret : -EAGAIN;
902	}
903
904	static int _ext4_get_block(struct inode *inode, sector_t iblock,
905	struct buffer_head *bh, int flags)
906	{
907	struct ext4_map_blocks map;
908	int ret = 0;
909
910	if (ext4_has_inline_data(inode))
911	return -ERANGE;
912
913	map.m_lblk = iblock;
914	map.m_len = bh->b_size >> inode->i_blkbits;
915
916	ret = ext4_map_blocks(handle: ext4_journal_current_handle(), inode, map: &map,
917	flags);
918	if (ret > 0) {
919	map_bh(bh, sb: inode->i_sb, block: map.m_pblk);
920	ext4_update_bh_state(bh, flags: map.m_flags);
921	bh->b_size = inode->i_sb->s_blocksize * map.m_len;
922	ret = 0;
923	} else if (ret == 0) {
924	/* hole case, need to fill in bh->b_size */
925	bh->b_size = inode->i_sb->s_blocksize * map.m_len;
926	}
927	return ret;
928	}
929
930	int ext4_get_block(struct inode *inode, sector_t iblock,
931	struct buffer_head *bh, int create)
932	{
933	return _ext4_get_block(inode, iblock, bh,
934	flags: create ? EXT4_GET_BLOCKS_CREATE : 0);
935	}
936
937	/*
938	* Get block function used when preparing for buffered write if we require
939	* creating an unwritten extent if blocks haven't been allocated. The extent
940	* will be converted to written after the IO is complete.
941	*/
942	int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
943	struct buffer_head *bh_result, int create)
944	{
945	int ret = 0;
946
947	ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
948	inode->i_ino, create);
949	ret = _ext4_get_block(inode, iblock, bh: bh_result,
950	EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
951
952	/*
953	* If the buffer is marked unwritten, mark it as new to make sure it is
954	* zeroed out correctly in case of partial writes. Otherwise, there is
955	* a chance of stale data getting exposed.
956	*/
957	if (ret == 0 && buffer_unwritten(bh: bh_result))
958	set_buffer_new(bh_result);
959
960	return ret;
961	}
962
963	/* Maximum number of blocks we map for direct IO at once. */
964	#define DIO_MAX_BLOCKS 4096
965
966	/*
967	* `handle' can be NULL if create is zero
968	*/
969	struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
970	ext4_lblk_t block, int map_flags)
971	{
972	struct ext4_map_blocks map;
973	struct buffer_head *bh;
974	int create = map_flags & EXT4_GET_BLOCKS_CREATE;
975	bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
976	int err;
977
978	ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
979	|| handle != NULL || create == 0);
980	ASSERT(create == 0 || !nowait);
981
982	map.m_lblk = block;
983	map.m_len = 1;
984	err = ext4_map_blocks(handle, inode, map: &map, flags: map_flags);
985
986	if (err == 0)
987	return create ? ERR_PTR(error: -ENOSPC) : NULL;
988	if (err < 0)
989	return ERR_PTR(error: err);
990
991	if (nowait)
992	return sb_find_get_block(sb: inode->i_sb, block: map.m_pblk);
993
994	/*
995	* Since bh could introduce extra ref count such as referred by
996	* journal_head etc. Try to avoid using __GFP_MOVABLE here
997	* as it may fail the migration when journal_head remains.
998	*/
999	bh = getblk_unmovable(bdev: inode->i_sb->s_bdev, block: map.m_pblk,
1000	size: inode->i_sb->s_blocksize);
1001
1002	if (unlikely(!bh))
1003	return ERR_PTR(error: -ENOMEM);
1004	if (map.m_flags & EXT4_MAP_NEW) {
1005	ASSERT(create != 0);
1006	ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
1007	|| (handle != NULL));
1008
1009	/*
1010	* Now that we do not always journal data, we should
1011	* keep in mind whether this should always journal the
1012	* new buffer as metadata. For now, regular file
1013	* writes use ext4_get_block instead, so it's not a
1014	* problem.
1015	*/
1016	lock_buffer(bh);
1017	BUFFER_TRACE(bh, "call get_create_access");
1018	err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1019	EXT4_JTR_NONE);
1020	if (unlikely(err)) {
1021	unlock_buffer(bh);
1022	goto errout;
1023	}
1024	if (!buffer_uptodate(bh)) {
1025	memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1026	set_buffer_uptodate(bh);
1027	}
1028	unlock_buffer(bh);
1029	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1030	err = ext4_handle_dirty_metadata(handle, inode, bh);
1031	if (unlikely(err))
1032	goto errout;
1033	} else
1034	BUFFER_TRACE(bh, "not a new buffer");
1035	return bh;
1036	errout:
1037	brelse(bh);
1038	return ERR_PTR(error: err);
1039	}
1040
1041	struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1042	ext4_lblk_t block, int map_flags)
1043	{
1044	struct buffer_head *bh;
1045	int ret;
1046
1047	bh = ext4_getblk(handle, inode, block, map_flags);
1048	if (IS_ERR(ptr: bh))
1049	return bh;
1050	if (!bh || ext4_buffer_uptodate(bh))
1051	return bh;
1052
1053	ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, wait: true);
1054	if (ret) {
1055	put_bh(bh);
1056	return ERR_PTR(error: ret);
1057	}
1058	return bh;
1059	}
1060
1061	/* Read a contiguous batch of blocks. */
1062	int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1063	bool wait, struct buffer_head **bhs)
1064	{
1065	int i, err;
1066
1067	for (i = 0; i < bh_count; i++) {
1068	bhs[i] = ext4_getblk(NULL, inode, block: block + i, map_flags: 0 /* map_flags */);
1069	if (IS_ERR(ptr: bhs[i])) {
1070	err = PTR_ERR(ptr: bhs[i]);
1071	bh_count = i;
1072	goto out_brelse;
1073	}
1074	}
1075
1076	for (i = 0; i < bh_count; i++)
1077	/* Note that NULL bhs[i] is valid because of holes. */
1078	if (bhs[i] && !ext4_buffer_uptodate(bh: bhs[i]))
1079	ext4_read_bh_lock(bh: bhs[i], REQ_META | REQ_PRIO, wait: false);
1080
1081	if (!wait)
1082	return 0;
1083
1084	for (i = 0; i < bh_count; i++)
1085	if (bhs[i])
1086	wait_on_buffer(bh: bhs[i]);
1087
1088	for (i = 0; i < bh_count; i++) {
1089	if (bhs[i] && !buffer_uptodate(bh: bhs[i])) {
1090	err = -EIO;
1091	goto out_brelse;
1092	}
1093	}
1094	return 0;
1095
1096	out_brelse:
1097	for (i = 0; i < bh_count; i++) {
1098	brelse(bh: bhs[i]);
1099	bhs[i] = NULL;
1100	}
1101	return err;
1102	}
1103
1104	int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
1105	struct buffer_head *head,
1106	unsigned from,
1107	unsigned to,
1108	int *partial,
1109	int (*fn)(handle_t *handle, struct inode *inode,
1110	struct buffer_head *bh))
1111	{
1112	struct buffer_head *bh;
1113	unsigned block_start, block_end;
1114	unsigned blocksize = head->b_size;
1115	int err, ret = 0;
1116	struct buffer_head *next;
1117
1118	for (bh = head, block_start = 0;
1119	ret == 0 && (bh != head || !block_start);
1120	block_start = block_end, bh = next) {
1121	next = bh->b_this_page;
1122	block_end = block_start + blocksize;
1123	if (block_end <= from || block_start >= to) {
1124	if (partial && !buffer_uptodate(bh))
1125	*partial = 1;
1126	continue;
1127	}
1128	err = (*fn)(handle, inode, bh);
1129	if (!ret)
1130	ret = err;
1131	}
1132	return ret;
1133	}
1134
1135	/*
1136	* Helper for handling dirtying of journalled data. We also mark the folio as
1137	* dirty so that writeback code knows about this page (and inode) contains
1138	* dirty data. ext4_writepages() then commits appropriate transaction to
1139	* make data stable.
1140	*/
1141	static int ext4_dirty_journalled_data(handle_t *handle, struct buffer_head *bh)
1142	{
1143	struct folio *folio = bh->b_folio;
1144	struct inode *inode = folio->mapping->host;
1145
1146	/* only regular files have a_ops */
1147	if (S_ISREG(inode->i_mode))
1148	folio_mark_dirty(folio);
1149	return ext4_handle_dirty_metadata(handle, NULL, bh);
1150	}
1151
1152	int do_journal_get_write_access(handle_t *handle, struct inode *inode,
1153	struct buffer_head *bh)
1154	{
1155	if (!buffer_mapped(bh) || buffer_freed(bh))
1156	return 0;
1157	BUFFER_TRACE(bh, "get write access");
1158	return ext4_journal_get_write_access(handle, inode->i_sb, bh,
1159	EXT4_JTR_NONE);
1160	}
1161
1162	int ext4_block_write_begin(handle_t *handle, struct folio *folio,
1163	loff_t pos, unsigned len,
1164	get_block_t *get_block)
1165	{
1166	unsigned int from = offset_in_folio(folio, pos);
1167	unsigned to = from + len;
1168	struct inode *inode = folio->mapping->host;
1169	unsigned block_start, block_end;
1170	sector_t block;
1171	int err = 0;
1172	unsigned int blocksize = i_blocksize(node: inode);
1173	struct buffer_head *bh, *head, *wait[2];
1174	int nr_wait = 0;
1175	int i;
1176	bool should_journal_data = ext4_should_journal_data(inode);
1177
1178	BUG_ON(!folio_test_locked(folio));
1179	BUG_ON(to > folio_size(folio));
1180	BUG_ON(from > to);
1181	WARN_ON_ONCE(blocksize > folio_size(folio));
1182
1183	head = folio_buffers(folio);
1184	if (!head)
1185	head = create_empty_buffers(folio, blocksize, b_state: 0);
1186	block = EXT4_PG_TO_LBLK(inode, folio->index);
1187
1188	for (bh = head, block_start = 0; bh != head || !block_start;
1189	block++, block_start = block_end, bh = bh->b_this_page) {
1190	block_end = block_start + blocksize;
1191	if (block_end <= from || block_start >= to) {
1192	if (folio_test_uptodate(folio)) {
1193	set_buffer_uptodate(bh);
1194	}
1195	continue;
1196	}
1197	if (WARN_ON_ONCE(buffer_new(bh)))
1198	clear_buffer_new(bh);
1199	if (!buffer_mapped(bh)) {
1200	WARN_ON(bh->b_size != blocksize);
1201	err = ext4_journal_ensure_extent_credits(handle, inode);
1202	if (!err)
1203	err = get_block(inode, block, bh, 1);
1204	if (err)
1205	break;
1206	if (buffer_new(bh)) {
1207	/*
1208	* We may be zeroing partial buffers or all new
1209	* buffers in case of failure. Prepare JBD2 for
1210	* that.
1211	*/
1212	if (should_journal_data)
1213	do_journal_get_write_access(handle,
1214	inode, bh);
1215	if (folio_test_uptodate(folio)) {
1216	/*
1217	* Unlike __block_write_begin() we leave
1218	* dirtying of new uptodate buffers to
1219	* ->write_end() time or
1220	* folio_zero_new_buffers().
1221	*/
1222	set_buffer_uptodate(bh);
1223	continue;
1224	}
1225	if (block_end > to || block_start < from)
1226	folio_zero_segments(folio, start1: to,
1227	xend1: block_end,
1228	start2: block_start, xend2: from);
1229	continue;
1230	}
1231	}
1232	if (folio_test_uptodate(folio)) {
1233	set_buffer_uptodate(bh);
1234	continue;
1235	}
1236	if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1237	!buffer_unwritten(bh) &&
1238	(block_start < from || block_end > to)) {
1239	ext4_read_bh_lock(bh, op_flags: 0, wait: false);
1240	wait[nr_wait++] = bh;
1241	}
1242	}
1243	/*
1244	* If we issued read requests, let them complete.
1245	*/
1246	for (i = 0; i < nr_wait; i++) {
1247	wait_on_buffer(bh: wait[i]);
1248	if (!buffer_uptodate(bh: wait[i]))
1249	err = -EIO;
1250	}
1251	if (unlikely(err)) {
1252	if (should_journal_data)
1253	ext4_journalled_zero_new_buffers(handle, inode, folio,
1254	from, to);
1255	else
1256	folio_zero_new_buffers(folio, from, to);
1257	} else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1258	for (i = 0; i < nr_wait; i++) {
1259	int err2;
1260
1261	err2 = fscrypt_decrypt_pagecache_blocks(folio,
1262	len: blocksize, offs: bh_offset(bh: wait[i]));
1263	if (err2) {
1264	clear_buffer_uptodate(bh: wait[i]);
1265	err = err2;
1266	}
1267	}
1268	}
1269
1270	return err;
1271	}
1272
1273	/*
1274	* To preserve ordering, it is essential that the hole instantiation and
1275	* the data write be encapsulated in a single transaction. We cannot
1276	* close off a transaction and start a new one between the ext4_get_block()
1277	* and the ext4_write_end(). So doing the jbd2_journal_start at the start of
1278	* ext4_write_begin() is the right place.
1279	*/
1280	static int ext4_write_begin(const struct kiocb *iocb,
1281	struct address_space *mapping,
1282	loff_t pos, unsigned len,
1283	struct folio **foliop, void **fsdata)
1284	{
1285	struct inode *inode = mapping->host;
1286	int ret, needed_blocks;
1287	handle_t *handle;
1288	int retries = 0;
1289	struct folio *folio;
1290	pgoff_t index;
1291	unsigned from, to;
1292
1293	ret = ext4_emergency_state(sb: inode->i_sb);
1294	if (unlikely(ret))
1295	return ret;
1296
1297	trace_ext4_write_begin(inode, pos, len);
1298	/*
1299	* Reserve one block more for addition to orphan list in case
1300	* we allocate blocks but write fails for some reason
1301	*/
1302	needed_blocks = ext4_chunk_trans_extent(inode,
1303	nrblocks: ext4_journal_blocks_per_folio(inode)) + 1;
1304	index = pos >> PAGE_SHIFT;
1305
1306	if (ext4_test_inode_state(inode, bit: EXT4_STATE_MAY_INLINE_DATA)) {
1307	ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1308	foliop);
1309	if (ret < 0)
1310	return ret;
1311	if (ret == 1)
1312	return 0;
1313	}
1314
1315	/*
1316	* write_begin_get_folio() can take a long time if the
1317	* system is thrashing due to memory pressure, or if the folio
1318	* is being written back. So grab it first before we start
1319	* the transaction handle. This also allows us to allocate
1320	* the folio (if needed) without using GFP_NOFS.
1321	*/
1322	retry_grab:
1323	folio = write_begin_get_folio(iocb, mapping, index, len);
1324	if (IS_ERR(ptr: folio))
1325	return PTR_ERR(ptr: folio);
1326
1327	if (len > folio_next_pos(folio) - pos)
1328	len = folio_next_pos(folio) - pos;
1329
1330	from = offset_in_folio(folio, pos);
1331	to = from + len;
1332
1333	/*
1334	* The same as page allocation, we prealloc buffer heads before
1335	* starting the handle.
1336	*/
1337	if (!folio_buffers(folio))
1338	create_empty_buffers(folio, blocksize: inode->i_sb->s_blocksize, b_state: 0);
1339
1340	folio_unlock(folio);
1341
1342	retry_journal:
1343	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1344	if (IS_ERR(ptr: handle)) {
1345	folio_put(folio);
1346	return PTR_ERR(ptr: handle);
1347	}
1348
1349	folio_lock(folio);
1350	if (folio->mapping != mapping) {
1351	/* The folio got truncated from under us */
1352	folio_unlock(folio);
1353	folio_put(folio);
1354	ext4_journal_stop(handle);
1355	goto retry_grab;
1356	}
1357	/* In case writeback began while the folio was unlocked */
1358	folio_wait_stable(folio);
1359
1360	if (ext4_should_dioread_nolock(inode))
1361	ret = ext4_block_write_begin(handle, folio, pos, len,
1362	get_block: ext4_get_block_unwritten);
1363	else
1364	ret = ext4_block_write_begin(handle, folio, pos, len,
1365	get_block: ext4_get_block);
1366	if (!ret && ext4_should_journal_data(inode)) {
1367	ret = ext4_walk_page_buffers(handle, inode,
1368	folio_buffers(folio), from, to,
1369	NULL, fn: do_journal_get_write_access);
1370	}
1371
1372	if (ret) {
1373	bool extended = (pos + len > inode->i_size) &&
1374	!ext4_verity_in_progress(inode);
1375
1376	folio_unlock(folio);
1377	/*
1378	* ext4_block_write_begin may have instantiated a few blocks
1379	* outside i_size. Trim these off again. Don't need
1380	* i_size_read because we hold i_rwsem.
1381	*
1382	* Add inode to orphan list in case we crash before
1383	* truncate finishes
1384	*/
1385	if (extended && ext4_can_truncate(inode))
1386	ext4_orphan_add(handle, inode);
1387
1388	ext4_journal_stop(handle);
1389	if (extended) {
1390	ext4_truncate_failed_write(inode);
1391	/*
1392	* If truncate failed early the inode might
1393	* still be on the orphan list; we need to
1394	* make sure the inode is removed from the
1395	* orphan list in that case.
1396	*/
1397	if (inode->i_nlink)
1398	ext4_orphan_del(NULL, inode);
1399	}
1400
1401	if (ret == -EAGAIN ||
1402	(ret == -ENOSPC &&
1403	ext4_should_retry_alloc(sb: inode->i_sb, retries: &retries)))
1404	goto retry_journal;
1405	folio_put(folio);
1406	return ret;
1407	}
1408	*foliop = folio;
1409	return ret;
1410	}
1411
1412	/* For write_end() in data=journal mode */
1413	static int write_end_fn(handle_t *handle, struct inode *inode,
1414	struct buffer_head *bh)
1415	{
1416	int ret;
1417	if (!buffer_mapped(bh) || buffer_freed(bh))
1418	return 0;
1419	set_buffer_uptodate(bh);
1420	ret = ext4_dirty_journalled_data(handle, bh);
1421	clear_buffer_meta(bh);
1422	clear_buffer_prio(bh);
1423	clear_buffer_new(bh);
1424	return ret;
1425	}
1426
1427	/*
1428	* We need to pick up the new inode size which generic_commit_write gave us
1429	* `iocb` can be NULL - eg, when called from page_symlink().
1430	*
1431	* ext4 never places buffers on inode->i_mapping->i_private_list. metadata
1432	* buffers are managed internally.
1433	*/
1434	static int ext4_write_end(const struct kiocb *iocb,
1435	struct address_space *mapping,
1436	loff_t pos, unsigned len, unsigned copied,
1437	struct folio *folio, void *fsdata)
1438	{
1439	handle_t *handle = ext4_journal_current_handle();
1440	struct inode *inode = mapping->host;
1441	loff_t old_size = inode->i_size;
1442	int ret = 0, ret2;
1443	int i_size_changed = 0;
1444	bool verity = ext4_verity_in_progress(inode);
1445
1446	trace_ext4_write_end(inode, pos, len, copied);
1447
1448	if (ext4_has_inline_data(inode) &&
1449	ext4_test_inode_state(inode, bit: EXT4_STATE_MAY_INLINE_DATA))
1450	return ext4_write_inline_data_end(inode, pos, len, copied,
1451	folio);
1452
1453	copied = block_write_end(pos, len, copied, folio);
1454	/*
1455	* it's important to update i_size while still holding folio lock:
1456	* page writeout could otherwise come in and zero beyond i_size.
1457	*
1458	* If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1459	* blocks are being written past EOF, so skip the i_size update.
1460	*/
1461	if (!verity)
1462	i_size_changed = ext4_update_inode_size(inode, newsize: pos + copied);
1463	folio_unlock(folio);
1464	folio_put(folio);
1465
1466	if (old_size < pos && !verity) {
1467	pagecache_isize_extended(inode, from: old_size, to: pos);
1468	ext4_zero_partial_blocks(handle, inode, lstart: old_size, lend: pos - old_size);
1469	}
1470	/*
1471	* Don't mark the inode dirty under folio lock. First, it unnecessarily
1472	* makes the holding time of folio lock longer. Second, it forces lock
1473	* ordering of folio lock and transaction start for journaling
1474	* filesystems.
1475	*/
1476	if (i_size_changed)
1477	ret = ext4_mark_inode_dirty(handle, inode);
1478
1479	if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1480	/* if we have allocated more blocks and copied
1481	* less. We will have blocks allocated outside
1482	* inode->i_size. So truncate them
1483	*/
1484	ext4_orphan_add(handle, inode);
1485
1486	ret2 = ext4_journal_stop(handle);
1487	if (!ret)
1488	ret = ret2;
1489
1490	if (pos + len > inode->i_size && !verity) {
1491	ext4_truncate_failed_write(inode);
1492	/*
1493	* If truncate failed early the inode might still be
1494	* on the orphan list; we need to make sure the inode
1495	* is removed from the orphan list in that case.
1496	*/
1497	if (inode->i_nlink)
1498	ext4_orphan_del(NULL, inode);
1499	}
1500
1501	return ret ? ret : copied;
1502	}
1503
1504	/*
1505	* This is a private version of folio_zero_new_buffers() which doesn't
1506	* set the buffer to be dirty, since in data=journalled mode we need
1507	* to call ext4_dirty_journalled_data() instead.
1508	*/
1509	static void ext4_journalled_zero_new_buffers(handle_t *handle,
1510	struct inode *inode,
1511	struct folio *folio,
1512	unsigned from, unsigned to)
1513	{
1514	unsigned int block_start = 0, block_end;
1515	struct buffer_head *head, *bh;
1516
1517	bh = head = folio_buffers(folio);
1518	do {
1519	block_end = block_start + bh->b_size;
1520	if (buffer_new(bh)) {
1521	if (block_end > from && block_start < to) {
1522	if (!folio_test_uptodate(folio)) {
1523	unsigned start, size;
1524
1525	start = max(from, block_start);
1526	size = min(to, block_end) - start;
1527
1528	folio_zero_range(folio, start, length: size);
1529	}
1530	clear_buffer_new(bh);
1531	write_end_fn(handle, inode, bh);
1532	}
1533	}
1534	block_start = block_end;
1535	bh = bh->b_this_page;
1536	} while (bh != head);
1537	}
1538
1539	static int ext4_journalled_write_end(const struct kiocb *iocb,
1540	struct address_space *mapping,
1541	loff_t pos, unsigned len, unsigned copied,
1542	struct folio *folio, void *fsdata)
1543	{
1544	handle_t *handle = ext4_journal_current_handle();
1545	struct inode *inode = mapping->host;
1546	loff_t old_size = inode->i_size;
1547	int ret = 0, ret2;
1548	int partial = 0;
1549	unsigned from, to;
1550	int size_changed = 0;
1551	bool verity = ext4_verity_in_progress(inode);
1552
1553	trace_ext4_journalled_write_end(inode, pos, len, copied);
1554	from = pos & (PAGE_SIZE - 1);
1555	to = from + len;
1556
1557	BUG_ON(!ext4_handle_valid(handle));
1558
1559	if (ext4_has_inline_data(inode))
1560	return ext4_write_inline_data_end(inode, pos, len, copied,
1561	folio);
1562
1563	if (unlikely(copied < len) && !folio_test_uptodate(folio)) {
1564	copied = 0;
1565	ext4_journalled_zero_new_buffers(handle, inode, folio,
1566	from, to);
1567	} else {
1568	if (unlikely(copied < len))
1569	ext4_journalled_zero_new_buffers(handle, inode, folio,
1570	from: from + copied, to);
1571	ret = ext4_walk_page_buffers(handle, inode,
1572	folio_buffers(folio),
1573	from, to: from + copied, partial: &partial,
1574	fn: write_end_fn);
1575	if (!partial)
1576	folio_mark_uptodate(folio);
1577	}
1578	if (!verity)
1579	size_changed = ext4_update_inode_size(inode, newsize: pos + copied);
1580	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1581	folio_unlock(folio);
1582	folio_put(folio);
1583
1584	if (old_size < pos && !verity) {
1585	pagecache_isize_extended(inode, from: old_size, to: pos);
1586	ext4_zero_partial_blocks(handle, inode, lstart: old_size, lend: pos - old_size);
1587	}
1588
1589	if (size_changed) {
1590	ret2 = ext4_mark_inode_dirty(handle, inode);
1591	if (!ret)
1592	ret = ret2;
1593	}
1594
1595	if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1596	/* if we have allocated more blocks and copied
1597	* less. We will have blocks allocated outside
1598	* inode->i_size. So truncate them
1599	*/
1600	ext4_orphan_add(handle, inode);
1601
1602	ret2 = ext4_journal_stop(handle);
1603	if (!ret)
1604	ret = ret2;
1605	if (pos + len > inode->i_size && !verity) {
1606	ext4_truncate_failed_write(inode);
1607	/*
1608	* If truncate failed early the inode might still be
1609	* on the orphan list; we need to make sure the inode
1610	* is removed from the orphan list in that case.
1611	*/
1612	if (inode->i_nlink)
1613	ext4_orphan_del(NULL, inode);
1614	}
1615
1616	return ret ? ret : copied;
1617	}
1618
1619	/*
1620	* Reserve space for 'nr_resv' clusters
1621	*/
1622	static int ext4_da_reserve_space(struct inode *inode, int nr_resv)
1623	{
1624	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
1625	struct ext4_inode_info *ei = EXT4_I(inode);
1626	int ret;
1627
1628	/*
1629	* We will charge metadata quota at writeout time; this saves
1630	* us from metadata over-estimation, though we may go over by
1631	* a small amount in the end. Here we just reserve for data.
1632	*/
1633	ret = dquot_reserve_block(inode, EXT4_C2B(sbi, nr_resv));
1634	if (ret)
1635	return ret;
1636
1637	spin_lock(lock: &ei->i_block_reservation_lock);
1638	if (ext4_claim_free_clusters(sbi, nclusters: nr_resv, flags: 0)) {
1639	spin_unlock(lock: &ei->i_block_reservation_lock);
1640	dquot_release_reservation_block(inode, EXT4_C2B(sbi, nr_resv));
1641	return -ENOSPC;
1642	}
1643	ei->i_reserved_data_blocks += nr_resv;
1644	trace_ext4_da_reserve_space(inode, nr_resv);
1645	spin_unlock(lock: &ei->i_block_reservation_lock);
1646
1647	return 0; /* success */
1648	}
1649
1650	void ext4_da_release_space(struct inode *inode, int to_free)
1651	{
1652	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
1653	struct ext4_inode_info *ei = EXT4_I(inode);
1654
1655	if (!to_free)
1656	return; /* Nothing to release, exit */
1657
1658	spin_lock(lock: &EXT4_I(inode)->i_block_reservation_lock);
1659
1660	trace_ext4_da_release_space(inode, freed_blocks: to_free);
1661	if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1662	/*
1663	* if there aren't enough reserved blocks, then the
1664	* counter is messed up somewhere. Since this
1665	* function is called from invalidate page, it's
1666	* harmless to return without any action.
1667	*/
1668	ext4_warning(inode->i_sb, "ext4_da_release_space: "
1669	"ino %lu, to_free %d with only %d reserved "
1670	"data blocks", inode->i_ino, to_free,
1671	ei->i_reserved_data_blocks);
1672	WARN_ON(1);
1673	to_free = ei->i_reserved_data_blocks;
1674	}
1675	ei->i_reserved_data_blocks -= to_free;
1676
1677	/* update fs dirty data blocks counter */
1678	percpu_counter_sub(fbc: &sbi->s_dirtyclusters_counter, amount: to_free);
1679
1680	spin_unlock(lock: &EXT4_I(inode)->i_block_reservation_lock);
1681
1682	dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1683	}
1684
1685	/*
1686	* Delayed allocation stuff
1687	*/
1688
1689	struct mpage_da_data {
1690	/* These are input fields for ext4_do_writepages() */
1691	struct inode *inode;
1692	struct writeback_control *wbc;
1693	unsigned int can_map:1; /* Can writepages call map blocks? */
1694
1695	/* These are internal state of ext4_do_writepages() */
1696	loff_t start_pos; /* The start pos to write */
1697	loff_t next_pos; /* Current pos to examine */
1698	loff_t end_pos; /* Last pos to examine */
1699
1700	/*
1701	* Extent to map - this can be after start_pos because that can be
1702	* fully mapped. We somewhat abuse m_flags to store whether the extent
1703	* is delalloc or unwritten.
1704	*/
1705	struct ext4_map_blocks map;
1706	struct ext4_io_submit io_submit; /* IO submission data */
1707	unsigned int do_map:1;
1708	unsigned int scanned_until_end:1;
1709	unsigned int journalled_more_data:1;
1710	};
1711
1712	static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1713	bool invalidate)
1714	{
1715	unsigned nr, i;
1716	pgoff_t index, end;
1717	struct folio_batch fbatch;
1718	struct inode *inode = mpd->inode;
1719	struct address_space *mapping = inode->i_mapping;
1720
1721	/* This is necessary when next_pos == 0. */
1722	if (mpd->start_pos >= mpd->next_pos)
1723	return;
1724
1725	mpd->scanned_until_end = 0;
1726	if (invalidate) {
1727	ext4_lblk_t start, last;
1728	start = EXT4_B_TO_LBLK(inode, mpd->start_pos);
1729	last = mpd->next_pos >> inode->i_blkbits;
1730
1731	/*
1732	* avoid racing with extent status tree scans made by
1733	* ext4_insert_delayed_block()
1734	*/
1735	down_write(sem: &EXT4_I(inode)->i_data_sem);
1736	ext4_es_remove_extent(inode, lblk: start, len: last - start);
1737	up_write(sem: &EXT4_I(inode)->i_data_sem);
1738	}
1739
1740	folio_batch_init(fbatch: &fbatch);
1741	index = mpd->start_pos >> PAGE_SHIFT;
1742	end = mpd->next_pos >> PAGE_SHIFT;
1743	while (index < end) {
1744	nr = filemap_get_folios(mapping, start: &index, end: end - 1, fbatch: &fbatch);
1745	if (nr == 0)
1746	break;
1747	for (i = 0; i < nr; i++) {
1748	struct folio *folio = fbatch.folios[i];
1749
1750	if (folio_pos(folio) < mpd->start_pos)
1751	continue;
1752	if (folio_next_index(folio) > end)
1753	continue;
1754	BUG_ON(!folio_test_locked(folio));
1755	BUG_ON(folio_test_writeback(folio));
1756	if (invalidate) {
1757	if (folio_mapped(folio))
1758	folio_clear_dirty_for_io(folio);
1759	block_invalidate_folio(folio, offset: 0,
1760	length: folio_size(folio));
1761	folio_clear_uptodate(folio);
1762	}
1763	folio_unlock(folio);
1764	}
1765	folio_batch_release(fbatch: &fbatch);
1766	}
1767	}
1768
1769	static void ext4_print_free_blocks(struct inode *inode)
1770	{
1771	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
1772	struct super_block *sb = inode->i_sb;
1773	struct ext4_inode_info *ei = EXT4_I(inode);
1774
1775	ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1776	EXT4_C2B(EXT4_SB(inode->i_sb),
1777	ext4_count_free_clusters(sb)));
1778	ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1779	ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1780	(long long) EXT4_C2B(EXT4_SB(sb),
1781	percpu_counter_sum(&sbi->s_freeclusters_counter)));
1782	ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1783	(long long) EXT4_C2B(EXT4_SB(sb),
1784	percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1785	ext4_msg(sb, KERN_CRIT, "Block reservation details");
1786	ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1787	ei->i_reserved_data_blocks);
1788	return;
1789	}
1790
1791	/*
1792	* Check whether the cluster containing lblk has been allocated or has
1793	* delalloc reservation.
1794	*
1795	* Returns 0 if the cluster doesn't have either, 1 if it has delalloc
1796	* reservation, 2 if it's already been allocated, negative error code on
1797	* failure.
1798	*/
1799	static int ext4_clu_alloc_state(struct inode *inode, ext4_lblk_t lblk)
1800	{
1801	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
1802	int ret;
1803
1804	/* Has delalloc reservation? */
1805	if (ext4_es_scan_clu(inode, matching_fn: &ext4_es_is_delayed, lblk))
1806	return 1;
1807
1808	/* Already been allocated? */
1809	if (ext4_es_scan_clu(inode, matching_fn: &ext4_es_is_mapped, lblk))
1810	return 2;
1811	ret = ext4_clu_mapped(inode, EXT4_B2C(sbi, lblk));
1812	if (ret < 0)
1813	return ret;
1814	if (ret > 0)
1815	return 2;
1816
1817	return 0;
1818	}
1819
1820	/*
1821	* ext4_insert_delayed_blocks - adds a multiple delayed blocks to the extents
1822	* status tree, incrementing the reserved
1823	* cluster/block count or making pending
1824	* reservations where needed
1825	*
1826	* @inode - file containing the newly added block
1827	* @lblk - start logical block to be added
1828	* @len - length of blocks to be added
1829	*
1830	* Returns 0 on success, negative error code on failure.
1831	*/
1832	static int ext4_insert_delayed_blocks(struct inode *inode, ext4_lblk_t lblk,
1833	ext4_lblk_t len)
1834	{
1835	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
1836	int ret;
1837	bool lclu_allocated = false;
1838	bool end_allocated = false;
1839	ext4_lblk_t resv_clu;
1840	ext4_lblk_t end = lblk + len - 1;
1841
1842	/*
1843	* If the cluster containing lblk or end is shared with a delayed,
1844	* written, or unwritten extent in a bigalloc file system, it's
1845	* already been accounted for and does not need to be reserved.
1846	* A pending reservation must be made for the cluster if it's
1847	* shared with a written or unwritten extent and doesn't already
1848	* have one. Written and unwritten extents can be purged from the
1849	* extents status tree if the system is under memory pressure, so
1850	* it's necessary to examine the extent tree if a search of the
1851	* extents status tree doesn't get a match.
1852	*/
1853	if (sbi->s_cluster_ratio == 1) {
1854	ret = ext4_da_reserve_space(inode, nr_resv: len);
1855	if (ret != 0) /* ENOSPC */
1856	return ret;
1857	} else { /* bigalloc */
1858	resv_clu = EXT4_B2C(sbi, end) - EXT4_B2C(sbi, lblk) + 1;
1859
1860	ret = ext4_clu_alloc_state(inode, lblk);
1861	if (ret < 0)
1862	return ret;
1863	if (ret > 0) {
1864	resv_clu--;
1865	lclu_allocated = (ret == 2);
1866	}
1867
1868	if (EXT4_B2C(sbi, lblk) != EXT4_B2C(sbi, end)) {
1869	ret = ext4_clu_alloc_state(inode, lblk: end);
1870	if (ret < 0)
1871	return ret;
1872	if (ret > 0) {
1873	resv_clu--;
1874	end_allocated = (ret == 2);
1875	}
1876	}
1877
1878	if (resv_clu) {
1879	ret = ext4_da_reserve_space(inode, nr_resv: resv_clu);
1880	if (ret != 0) /* ENOSPC */
1881	return ret;
1882	}
1883	}
1884
1885	ext4_es_insert_delayed_extent(inode, lblk, len, lclu_allocated,
1886	end_allocated);
1887	return 0;
1888	}
1889
1890	/*
1891	* Looks up the requested blocks and sets the delalloc extent map.
1892	* First try to look up for the extent entry that contains the requested
1893	* blocks in the extent status tree without i_data_sem, then try to look
1894	* up for the ondisk extent mapping with i_data_sem in read mode,
1895	* finally hold i_data_sem in write mode, looks up again and add a
1896	* delalloc extent entry if it still couldn't find any extent. Pass out
1897	* the mapped extent through @map and return 0 on success.
1898	*/
1899	static int ext4_da_map_blocks(struct inode *inode, struct ext4_map_blocks *map)
1900	{
1901	struct extent_status es;
1902	int retval;
1903	#ifdef ES_AGGRESSIVE_TEST
1904	struct ext4_map_blocks orig_map;
1905
1906	memcpy(&orig_map, map, sizeof(*map));
1907	#endif
1908
1909	map->m_flags = 0;
1910	ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1911	(unsigned long) map->m_lblk);
1912
1913	ext4_check_map_extents_env(inode);
1914
1915	/* Lookup extent status tree firstly */
1916	if (ext4_es_lookup_extent(inode, lblk: map->m_lblk, NULL, es: &es, NULL)) {
1917	map->m_len = min_t(unsigned int, map->m_len,
1918	es.es_len - (map->m_lblk - es.es_lblk));
1919
1920	if (ext4_es_is_hole(es: &es))
1921	goto add_delayed;
1922
1923	found:
1924	/*
1925	* Delayed extent could be allocated by fallocate.
1926	* So we need to check it.
1927	*/
1928	if (ext4_es_is_delayed(es: &es)) {
1929	map->m_flags |= EXT4_MAP_DELAYED;
1930	return 0;
1931	}
1932
1933	map->m_pblk = ext4_es_pblock(es: &es) + map->m_lblk - es.es_lblk;
1934	if (ext4_es_is_written(es: &es))
1935	map->m_flags |= EXT4_MAP_MAPPED;
1936	else if (ext4_es_is_unwritten(es: &es))
1937	map->m_flags |= EXT4_MAP_UNWRITTEN;
1938	else
1939	BUG();
1940
1941	#ifdef ES_AGGRESSIVE_TEST
1942	ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1943	#endif
1944	return 0;
1945	}
1946
1947	/*
1948	* Try to see if we can get the block without requesting a new
1949	* file system block.
1950	*/
1951	down_read(sem: &EXT4_I(inode)->i_data_sem);
1952	if (ext4_has_inline_data(inode))
1953	retval = 0;
1954	else
1955	retval = ext4_map_query_blocks(NULL, inode, map, flags: 0);
1956	up_read(sem: &EXT4_I(inode)->i_data_sem);
1957	if (retval)
1958	return retval < 0 ? retval : 0;
1959
1960	add_delayed:
1961	down_write(sem: &EXT4_I(inode)->i_data_sem);
1962	/*
1963	* Page fault path (ext4_page_mkwrite does not take i_rwsem)
1964	* and fallocate path (no folio lock) can race. Make sure we
1965	* lookup the extent status tree here again while i_data_sem
1966	* is held in write mode, before inserting a new da entry in
1967	* the extent status tree.
1968	*/
1969	if (ext4_es_lookup_extent(inode, lblk: map->m_lblk, NULL, es: &es, NULL)) {
1970	map->m_len = min_t(unsigned int, map->m_len,
1971	es.es_len - (map->m_lblk - es.es_lblk));
1972
1973	if (!ext4_es_is_hole(es: &es)) {
1974	up_write(sem: &EXT4_I(inode)->i_data_sem);
1975	goto found;
1976	}
1977	} else if (!ext4_has_inline_data(inode)) {
1978	retval = ext4_map_query_blocks(NULL, inode, map, flags: 0);
1979	if (retval) {
1980	up_write(sem: &EXT4_I(inode)->i_data_sem);
1981	return retval < 0 ? retval : 0;
1982	}
1983	}
1984
1985	map->m_flags |= EXT4_MAP_DELAYED;
1986	retval = ext4_insert_delayed_blocks(inode, lblk: map->m_lblk, len: map->m_len);
1987	if (!retval)
1988	map->m_seq = READ_ONCE(EXT4_I(inode)->i_es_seq);
1989	up_write(sem: &EXT4_I(inode)->i_data_sem);
1990
1991	return retval;
1992	}
1993
1994	/*
1995	* This is a special get_block_t callback which is used by
1996	* ext4_da_write_begin(). It will either return mapped block or
1997	* reserve space for a single block.
1998	*
1999	* For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2000	* We also have b_blocknr = -1 and b_bdev initialized properly
2001	*
2002	* For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2003	* We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2004	* initialized properly.
2005	*/
2006	int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2007	struct buffer_head *bh, int create)
2008	{
2009	struct ext4_map_blocks map;
2010	sector_t invalid_block = ~((sector_t) 0xffff);
2011	int ret = 0;
2012
2013	BUG_ON(create == 0);
2014	BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2015
2016	if (invalid_block < ext4_blocks_count(es: EXT4_SB(sb: inode->i_sb)->s_es))
2017	invalid_block = ~0;
2018
2019	map.m_lblk = iblock;
2020	map.m_len = 1;
2021
2022	/*
2023	* first, we need to know whether the block is allocated already
2024	* preallocated blocks are unmapped but should treated
2025	* the same as allocated blocks.
2026	*/
2027	ret = ext4_da_map_blocks(inode, map: &map);
2028	if (ret < 0)
2029	return ret;
2030
2031	if (map.m_flags & EXT4_MAP_DELAYED) {
2032	map_bh(bh, sb: inode->i_sb, block: invalid_block);
2033	set_buffer_new(bh);
2034	set_buffer_delay(bh);
2035	return 0;
2036	}
2037
2038	map_bh(bh, sb: inode->i_sb, block: map.m_pblk);
2039	ext4_update_bh_state(bh, flags: map.m_flags);
2040
2041	if (buffer_unwritten(bh)) {
2042	/* A delayed write to unwritten bh should be marked
2043	* new and mapped. Mapped ensures that we don't do
2044	* get_block multiple times when we write to the same
2045	* offset and new ensures that we do proper zero out
2046	* for partial write.
2047	*/
2048	set_buffer_new(bh);
2049	set_buffer_mapped(bh);
2050	}
2051	return 0;
2052	}
2053
2054	static void mpage_folio_done(struct mpage_da_data *mpd, struct folio *folio)
2055	{
2056	mpd->start_pos += folio_size(folio);
2057	mpd->wbc->nr_to_write -= folio_nr_pages(folio);
2058	folio_unlock(folio);
2059	}
2060
2061	static int mpage_submit_folio(struct mpage_da_data *mpd, struct folio *folio)
2062	{
2063	size_t len;
2064	loff_t size;
2065	int err;
2066
2067	WARN_ON_ONCE(folio_pos(folio) != mpd->start_pos);
2068	folio_clear_dirty_for_io(folio);
2069	/*
2070	* We have to be very careful here! Nothing protects writeback path
2071	* against i_size changes and the page can be writeably mapped into
2072	* page tables. So an application can be growing i_size and writing
2073	* data through mmap while writeback runs. folio_clear_dirty_for_io()
2074	* write-protects our page in page tables and the page cannot get
2075	* written to again until we release folio lock. So only after
2076	* folio_clear_dirty_for_io() we are safe to sample i_size for
2077	* ext4_bio_write_folio() to zero-out tail of the written page. We rely
2078	* on the barrier provided by folio_test_clear_dirty() in
2079	* folio_clear_dirty_for_io() to make sure i_size is really sampled only
2080	* after page tables are updated.
2081	*/
2082	size = i_size_read(inode: mpd->inode);
2083	len = folio_size(folio);
2084	if (folio_pos(folio) + len > size &&
2085	!ext4_verity_in_progress(inode: mpd->inode))
2086	len = size & (len - 1);
2087	err = ext4_bio_write_folio(io: &mpd->io_submit, page: folio, len);
2088
2089	return err;
2090	}
2091
2092	#define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2093
2094	/*
2095	* mballoc gives us at most this number of blocks...
2096	* XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2097	* The rest of mballoc seems to handle chunks up to full group size.
2098	*/
2099	#define MAX_WRITEPAGES_EXTENT_LEN 2048
2100
2101	/*
2102	* mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2103	*
2104	* @mpd - extent of blocks
2105	* @lblk - logical number of the block in the file
2106	* @bh - buffer head we want to add to the extent
2107	*
2108	* The function is used to collect contig. blocks in the same state. If the
2109	* buffer doesn't require mapping for writeback and we haven't started the
2110	* extent of buffers to map yet, the function returns 'true' immediately - the
2111	* caller can write the buffer right away. Otherwise the function returns true
2112	* if the block has been added to the extent, false if the block couldn't be
2113	* added.
2114	*/
2115	static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2116	struct buffer_head *bh)
2117	{
2118	struct ext4_map_blocks *map = &mpd->map;
2119
2120	/* Buffer that doesn't need mapping for writeback? */
2121	if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2122	(!buffer_delay(bh) && !buffer_unwritten(bh))) {
2123	/* So far no extent to map => we write the buffer right away */
2124	if (map->m_len == 0)
2125	return true;
2126	return false;
2127	}
2128
2129	/* First block in the extent? */
2130	if (map->m_len == 0) {
2131	/* We cannot map unless handle is started... */
2132	if (!mpd->do_map)
2133	return false;
2134	map->m_lblk = lblk;
2135	map->m_len = 1;
2136	map->m_flags = bh->b_state & BH_FLAGS;
2137	return true;
2138	}
2139
2140	/* Don't go larger than mballoc is willing to allocate */
2141	if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2142	return false;
2143
2144	/* Can we merge the block to our big extent? */
2145	if (lblk == map->m_lblk + map->m_len &&
2146	(bh->b_state & BH_FLAGS) == map->m_flags) {
2147	map->m_len++;
2148	return true;
2149	}
2150	return false;
2151	}
2152
2153	/*
2154	* mpage_process_page_bufs - submit page buffers for IO or add them to extent
2155	*
2156	* @mpd - extent of blocks for mapping
2157	* @head - the first buffer in the page
2158	* @bh - buffer we should start processing from
2159	* @lblk - logical number of the block in the file corresponding to @bh
2160	*
2161	* Walk through page buffers from @bh upto @head (exclusive) and either submit
2162	* the page for IO if all buffers in this page were mapped and there's no
2163	* accumulated extent of buffers to map or add buffers in the page to the
2164	* extent of buffers to map. The function returns 1 if the caller can continue
2165	* by processing the next page, 0 if it should stop adding buffers to the
2166	* extent to map because we cannot extend it anymore. It can also return value
2167	* < 0 in case of error during IO submission.
2168	*/
2169	static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2170	struct buffer_head *head,
2171	struct buffer_head *bh,
2172	ext4_lblk_t lblk)
2173	{
2174	struct inode *inode = mpd->inode;
2175	int err;
2176	ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(node: inode) - 1)
2177	>> inode->i_blkbits;
2178
2179	if (ext4_verity_in_progress(inode))
2180	blocks = EXT_MAX_BLOCKS;
2181
2182	do {
2183	BUG_ON(buffer_locked(bh));
2184
2185	if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2186	/* Found extent to map? */
2187	if (mpd->map.m_len)
2188	return 0;
2189	/* Buffer needs mapping and handle is not started? */
2190	if (!mpd->do_map)
2191	return 0;
2192	/* Everything mapped so far and we hit EOF */
2193	break;
2194	}
2195	} while (lblk++, (bh = bh->b_this_page) != head);
2196	/* So far everything mapped? Submit the page for IO. */
2197	if (mpd->map.m_len == 0) {
2198	err = mpage_submit_folio(mpd, folio: head->b_folio);
2199	if (err < 0)
2200	return err;
2201	mpage_folio_done(mpd, folio: head->b_folio);
2202	}
2203	if (lblk >= blocks) {
2204	mpd->scanned_until_end = 1;
2205	return 0;
2206	}
2207	return 1;
2208	}
2209
2210	/*
2211	* mpage_process_folio - update folio buffers corresponding to changed extent
2212	* and may submit fully mapped page for IO
2213	* @mpd: description of extent to map, on return next extent to map
2214	* @folio: Contains these buffers.
2215	* @m_lblk: logical block mapping.
2216	* @m_pblk: corresponding physical mapping.
2217	* @map_bh: determines on return whether this page requires any further
2218	* mapping or not.
2219	*
2220	* Scan given folio buffers corresponding to changed extent and update buffer
2221	* state according to new extent state.
2222	* We map delalloc buffers to their physical location, clear unwritten bits.
2223	* If the given folio is not fully mapped, we update @mpd to the next extent in
2224	* the given folio that needs mapping & return @map_bh as true.
2225	*/
2226	static int mpage_process_folio(struct mpage_da_data *mpd, struct folio *folio,
2227	ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2228	bool *map_bh)
2229	{
2230	struct buffer_head *head, *bh;
2231	ext4_io_end_t *io_end = mpd->io_submit.io_end;
2232	ext4_lblk_t lblk = *m_lblk;
2233	ext4_fsblk_t pblock = *m_pblk;
2234	int err = 0;
2235	ssize_t io_end_size = 0;
2236	struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2237
2238	bh = head = folio_buffers(folio);
2239	do {
2240	if (lblk < mpd->map.m_lblk)
2241	continue;
2242	if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2243	/*
2244	* Buffer after end of mapped extent.
2245	* Find next buffer in the folio to map.
2246	*/
2247	mpd->map.m_len = 0;
2248	mpd->map.m_flags = 0;
2249	io_end_vec->size += io_end_size;
2250
2251	err = mpage_process_page_bufs(mpd, head, bh, lblk);
2252	if (err > 0)
2253	err = 0;
2254	if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2255	io_end_vec = ext4_alloc_io_end_vec(io_end);
2256	if (IS_ERR(ptr: io_end_vec)) {
2257	err = PTR_ERR(ptr: io_end_vec);
2258	goto out;
2259	}
2260	io_end_vec->offset = EXT4_LBLK_TO_B(mpd->inode,
2261	mpd->map.m_lblk);
2262	}
2263	*map_bh = true;
2264	goto out;
2265	}
2266	if (buffer_delay(bh)) {
2267	clear_buffer_delay(bh);
2268	bh->b_blocknr = pblock++;
2269	}
2270	clear_buffer_unwritten(bh);
2271	io_end_size += i_blocksize(node: mpd->inode);
2272	} while (lblk++, (bh = bh->b_this_page) != head);
2273
2274	io_end_vec->size += io_end_size;
2275	*map_bh = false;
2276	out:
2277	*m_lblk = lblk;
2278	*m_pblk = pblock;
2279	return err;
2280	}
2281
2282	/*
2283	* mpage_map_buffers - update buffers corresponding to changed extent and
2284	* submit fully mapped pages for IO
2285	*
2286	* @mpd - description of extent to map, on return next extent to map
2287	*
2288	* Scan buffers corresponding to changed extent (we expect corresponding pages
2289	* to be already locked) and update buffer state according to new extent state.
2290	* We map delalloc buffers to their physical location, clear unwritten bits,
2291	* and mark buffers as uninit when we perform writes to unwritten extents
2292	* and do extent conversion after IO is finished. If the last page is not fully
2293	* mapped, we update @map to the next extent in the last page that needs
2294	* mapping. Otherwise we submit the page for IO.
2295	*/
2296	static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2297	{
2298	struct folio_batch fbatch;
2299	unsigned nr, i;
2300	struct inode *inode = mpd->inode;
2301	pgoff_t start, end;
2302	ext4_lblk_t lblk;
2303	ext4_fsblk_t pblock;
2304	int err;
2305	bool map_bh = false;
2306
2307	start = EXT4_LBLK_TO_PG(inode, mpd->map.m_lblk);
2308	end = EXT4_LBLK_TO_PG(inode, mpd->map.m_lblk + mpd->map.m_len - 1);
2309	pblock = mpd->map.m_pblk;
2310
2311	folio_batch_init(fbatch: &fbatch);
2312	while (start <= end) {
2313	nr = filemap_get_folios(mapping: inode->i_mapping, start: &start, end, fbatch: &fbatch);
2314	if (nr == 0)
2315	break;
2316	for (i = 0; i < nr; i++) {
2317	struct folio *folio = fbatch.folios[i];
2318
2319	lblk = EXT4_PG_TO_LBLK(inode, folio->index);
2320	err = mpage_process_folio(mpd, folio, m_lblk: &lblk, m_pblk: &pblock,
2321	map_bh: &map_bh);
2322	/*
2323	* If map_bh is true, means page may require further bh
2324	* mapping, or maybe the page was submitted for IO.
2325	* So we return to call further extent mapping.
2326	*/
2327	if (err < 0 || map_bh)
2328	goto out;
2329	/* Page fully mapped - let IO run! */
2330	err = mpage_submit_folio(mpd, folio);
2331	if (err < 0)
2332	goto out;
2333	mpage_folio_done(mpd, folio);
2334	}
2335	folio_batch_release(fbatch: &fbatch);
2336	}
2337	/* Extent fully mapped and matches with page boundary. We are done. */
2338	mpd->map.m_len = 0;
2339	mpd->map.m_flags = 0;
2340	return 0;
2341	out:
2342	folio_batch_release(fbatch: &fbatch);
2343	return err;
2344	}
2345
2346	static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2347	{
2348	struct inode *inode = mpd->inode;
2349	struct ext4_map_blocks *map = &mpd->map;
2350	int get_blocks_flags;
2351	int err, dioread_nolock;
2352
2353	/* Make sure transaction has enough credits for this extent */
2354	err = ext4_journal_ensure_extent_credits(handle, inode);
2355	if (err < 0)
2356	return err;
2357
2358	trace_ext4_da_write_pages_extent(inode, map);
2359	/*
2360	* Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2361	* to convert an unwritten extent to be initialized (in the case
2362	* where we have written into one or more preallocated blocks). It is
2363	* possible that we're going to need more metadata blocks than
2364	* previously reserved. However we must not fail because we're in
2365	* writeback and there is nothing we can do about it so it might result
2366	* in data loss. So use reserved blocks to allocate metadata if
2367	* possible. In addition, do not cache any unrelated extents, as it
2368	* only holds the folio lock but does not hold the i_rwsem or
2369	* invalidate_lock, which could corrupt the extent status tree.
2370	*/
2371	get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2372	EXT4_GET_BLOCKS_METADATA_NOFAIL |
2373	EXT4_GET_BLOCKS_IO_SUBMIT |
2374	EXT4_EX_NOCACHE;
2375
2376	dioread_nolock = ext4_should_dioread_nolock(inode);
2377	if (dioread_nolock)
2378	get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2379
2380	err = ext4_map_blocks(handle, inode, map, flags: get_blocks_flags);
2381	if (err < 0)
2382	return err;
2383	if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2384	if (!mpd->io_submit.io_end->handle &&
2385	ext4_handle_valid(handle)) {
2386	mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2387	handle->h_rsv_handle = NULL;
2388	}
2389	ext4_set_io_unwritten_flag(io_end: mpd->io_submit.io_end);
2390	}
2391
2392	BUG_ON(map->m_len == 0);
2393	return 0;
2394	}
2395
2396	/*
2397	* This is used to submit mapped buffers in a single folio that is not fully
2398	* mapped for various reasons, such as insufficient space or journal credits.
2399	*/
2400	static int mpage_submit_partial_folio(struct mpage_da_data *mpd)
2401	{
2402	struct inode *inode = mpd->inode;
2403	struct folio *folio;
2404	loff_t pos;
2405	int ret;
2406
2407	folio = filemap_get_folio(mapping: inode->i_mapping,
2408	index: mpd->start_pos >> PAGE_SHIFT);
2409	if (IS_ERR(ptr: folio))
2410	return PTR_ERR(ptr: folio);
2411	/*
2412	* The mapped position should be within the current processing folio
2413	* but must not be the folio start position.
2414	*/
2415	pos = ((loff_t)mpd->map.m_lblk) << inode->i_blkbits;
2416	if (WARN_ON_ONCE((folio_pos(folio) == pos) ||
2417	!folio_contains(folio, pos >> PAGE_SHIFT)))
2418	return -EINVAL;
2419
2420	ret = mpage_submit_folio(mpd, folio);
2421	if (ret)
2422	goto out;
2423	/*
2424	* Update start_pos to prevent this folio from being released in
2425	* mpage_release_unused_pages(), it will be reset to the aligned folio
2426	* pos when this folio is written again in the next round. Additionally,
2427	* do not update wbc->nr_to_write here, as it will be updated once the
2428	* entire folio has finished processing.
2429	*/
2430	mpd->start_pos = pos;
2431	out:
2432	folio_unlock(folio);
2433	folio_put(folio);
2434	return ret;
2435	}
2436
2437	/*
2438	* mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2439	* mpd->len and submit pages underlying it for IO
2440	*
2441	* @handle - handle for journal operations
2442	* @mpd - extent to map
2443	* @give_up_on_write - we set this to true iff there is a fatal error and there
2444	* is no hope of writing the data. The caller should discard
2445	* dirty pages to avoid infinite loops.
2446	*
2447	* The function maps extent starting at mpd->lblk of length mpd->len. If it is
2448	* delayed, blocks are allocated, if it is unwritten, we may need to convert
2449	* them to initialized or split the described range from larger unwritten
2450	* extent. Note that we need not map all the described range since allocation
2451	* can return less blocks or the range is covered by more unwritten extents. We
2452	* cannot map more because we are limited by reserved transaction credits. On
2453	* the other hand we always make sure that the last touched page is fully
2454	* mapped so that it can be written out (and thus forward progress is
2455	* guaranteed). After mapping we submit all mapped pages for IO.
2456	*/
2457	static int mpage_map_and_submit_extent(handle_t *handle,
2458	struct mpage_da_data *mpd,
2459	bool *give_up_on_write)
2460	{
2461	struct inode *inode = mpd->inode;
2462	struct ext4_map_blocks *map = &mpd->map;
2463	int err;
2464	loff_t disksize;
2465	int progress = 0;
2466	ext4_io_end_t *io_end = mpd->io_submit.io_end;
2467	struct ext4_io_end_vec *io_end_vec;
2468
2469	io_end_vec = ext4_alloc_io_end_vec(io_end);
2470	if (IS_ERR(ptr: io_end_vec))
2471	return PTR_ERR(ptr: io_end_vec);
2472	io_end_vec->offset = EXT4_LBLK_TO_B(inode, map->m_lblk);
2473	do {
2474	err = mpage_map_one_extent(handle, mpd);
2475	if (err < 0) {
2476	struct super_block *sb = inode->i_sb;
2477
2478	if (ext4_emergency_state(sb))
2479	goto invalidate_dirty_pages;
2480	/*
2481	* Let the uper layers retry transient errors.
2482	* In the case of ENOSPC, if ext4_count_free_blocks()
2483	* is non-zero, a commit should free up blocks.
2484	*/
2485	if ((err == -ENOMEM) || (err == -EAGAIN) ||
2486	(err == -ENOSPC && ext4_count_free_clusters(sb))) {
2487	/*
2488	* We may have already allocated extents for
2489	* some bhs inside the folio, issue the
2490	* corresponding data to prevent stale data.
2491	*/
2492	if (progress) {
2493	if (mpage_submit_partial_folio(mpd))
2494	goto invalidate_dirty_pages;
2495	goto update_disksize;
2496	}
2497	return err;
2498	}
2499	ext4_msg(sb, KERN_CRIT,
2500	"Delayed block allocation failed for "
2501	"inode %lu at logical offset %llu with"
2502	" max blocks %u with error %d",
2503	inode->i_ino,
2504	(unsigned long long)map->m_lblk,
2505	(unsigned)map->m_len, -err);
2506	ext4_msg(sb, KERN_CRIT,
2507	"This should not happen!! Data will "
2508	"be lost\n");
2509	if (err == -ENOSPC)
2510	ext4_print_free_blocks(inode);
2511	invalidate_dirty_pages:
2512	*give_up_on_write = true;
2513	return err;
2514	}
2515	progress = 1;
2516	/*
2517	* Update buffer state, submit mapped pages, and get us new
2518	* extent to map
2519	*/
2520	err = mpage_map_and_submit_buffers(mpd);
2521	if (err < 0)
2522	goto update_disksize;
2523	} while (map->m_len);
2524
2525	update_disksize:
2526	/*
2527	* Update on-disk size after IO is submitted. Races with
2528	* truncate are avoided by checking i_size under i_data_sem.
2529	*/
2530	disksize = mpd->start_pos;
2531	if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2532	int err2;
2533	loff_t i_size;
2534
2535	down_write(sem: &EXT4_I(inode)->i_data_sem);
2536	i_size = i_size_read(inode);
2537	if (disksize > i_size)
2538	disksize = i_size;
2539	if (disksize > EXT4_I(inode)->i_disksize)
2540	EXT4_I(inode)->i_disksize = disksize;
2541	up_write(sem: &EXT4_I(inode)->i_data_sem);
2542	err2 = ext4_mark_inode_dirty(handle, inode);
2543	if (err2) {
2544	ext4_error_err(inode->i_sb, -err2,
2545	"Failed to mark inode %lu dirty",
2546	inode->i_ino);
2547	}
2548	if (!err)
2549	err = err2;
2550	}
2551	return err;
2552	}
2553
2554	static int ext4_journal_folio_buffers(handle_t *handle, struct folio *folio,
2555	size_t len)
2556	{
2557	struct buffer_head *page_bufs = folio_buffers(folio);
2558	struct inode *inode = folio->mapping->host;
2559	int ret, err;
2560
2561	ret = ext4_walk_page_buffers(handle, inode, head: page_bufs, from: 0, to: len,
2562	NULL, fn: do_journal_get_write_access);
2563	err = ext4_walk_page_buffers(handle, inode, head: page_bufs, from: 0, to: len,
2564	NULL, fn: write_end_fn);
2565	if (ret == 0)
2566	ret = err;
2567	err = ext4_jbd2_inode_add_write(handle, inode, start_byte: folio_pos(folio), length: len);
2568	if (ret == 0)
2569	ret = err;
2570	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2571
2572	return ret;
2573	}
2574
2575	static int mpage_journal_page_buffers(handle_t *handle,
2576	struct mpage_da_data *mpd,
2577	struct folio *folio)
2578	{
2579	struct inode *inode = mpd->inode;
2580	loff_t size = i_size_read(inode);
2581	size_t len = folio_size(folio);
2582
2583	folio_clear_checked(folio);
2584	mpd->wbc->nr_to_write -= folio_nr_pages(folio);
2585
2586	if (folio_pos(folio) + len > size &&
2587	!ext4_verity_in_progress(inode))
2588	len = size & (len - 1);
2589
2590	return ext4_journal_folio_buffers(handle, folio, len);
2591	}
2592
2593	/*
2594	* mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2595	* needing mapping, submit mapped pages
2596	*
2597	* @mpd - where to look for pages
2598	*
2599	* Walk dirty pages in the mapping. If they are fully mapped, submit them for
2600	* IO immediately. If we cannot map blocks, we submit just already mapped
2601	* buffers in the page for IO and keep page dirty. When we can map blocks and
2602	* we find a page which isn't mapped we start accumulating extent of buffers
2603	* underlying these pages that needs mapping (formed by either delayed or
2604	* unwritten buffers). We also lock the pages containing these buffers. The
2605	* extent found is returned in @mpd structure (starting at mpd->lblk with
2606	* length mpd->len blocks).
2607	*
2608	* Note that this function can attach bios to one io_end structure which are
2609	* neither logically nor physically contiguous. Although it may seem as an
2610	* unnecessary complication, it is actually inevitable in blocksize < pagesize
2611	* case as we need to track IO to all buffers underlying a page in one io_end.
2612	*/
2613	static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2614	{
2615	struct address_space *mapping = mpd->inode->i_mapping;
2616	struct folio_batch fbatch;
2617	unsigned int nr_folios;
2618	pgoff_t index = mpd->start_pos >> PAGE_SHIFT;
2619	pgoff_t end = mpd->end_pos >> PAGE_SHIFT;
2620	xa_mark_t tag;
2621	int i, err = 0;
2622	ext4_lblk_t lblk;
2623	struct buffer_head *head;
2624	handle_t *handle = NULL;
2625	int bpp = ext4_journal_blocks_per_folio(inode: mpd->inode);
2626
2627	tag = wbc_to_tag(wbc: mpd->wbc);
2628
2629	mpd->map.m_len = 0;
2630	mpd->next_pos = mpd->start_pos;
2631	if (ext4_should_journal_data(inode: mpd->inode)) {
2632	handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE,
2633	bpp);
2634	if (IS_ERR(ptr: handle))
2635	return PTR_ERR(ptr: handle);
2636	}
2637	folio_batch_init(fbatch: &fbatch);
2638	while (index <= end) {
2639	nr_folios = filemap_get_folios_tag(mapping, start: &index, end,
2640	tag, fbatch: &fbatch);
2641	if (nr_folios == 0)
2642	break;
2643
2644	for (i = 0; i < nr_folios; i++) {
2645	struct folio *folio = fbatch.folios[i];
2646
2647	/*
2648	* Accumulated enough dirty pages? This doesn't apply
2649	* to WB_SYNC_ALL mode. For integrity sync we have to
2650	* keep going because someone may be concurrently
2651	* dirtying pages, and we might have synced a lot of
2652	* newly appeared dirty pages, but have not synced all
2653	* of the old dirty pages.
2654	*/
2655	if (mpd->wbc->sync_mode == WB_SYNC_NONE &&
2656	mpd->wbc->nr_to_write <=
2657	EXT4_LBLK_TO_PG(mpd->inode, mpd->map.m_len))
2658	goto out;
2659
2660	/* If we can't merge this page, we are done. */
2661	if (mpd->map.m_len > 0 &&
2662	mpd->next_pos != folio_pos(folio))
2663	goto out;
2664
2665	if (handle) {
2666	err = ext4_journal_ensure_credits(handle, credits: bpp,
2667	revoke_creds: 0);
2668	if (err < 0)
2669	goto out;
2670	}
2671
2672	folio_lock(folio);
2673	/*
2674	* If the page is no longer dirty, or its mapping no
2675	* longer corresponds to inode we are writing (which
2676	* means it has been truncated or invalidated), or the
2677	* page is already under writeback and we are not doing
2678	* a data integrity writeback, skip the page
2679	*/
2680	if (!folio_test_dirty(folio) ||
2681	(folio_test_writeback(folio) &&
2682	(mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2683	unlikely(folio->mapping != mapping)) {
2684	folio_unlock(folio);
2685	continue;
2686	}
2687
2688	folio_wait_writeback(folio);
2689	BUG_ON(folio_test_writeback(folio));
2690
2691	/*
2692	* Should never happen but for buggy code in
2693	* other subsystems that call
2694	* set_page_dirty() without properly warning
2695	* the file system first. See [1] for more
2696	* information.
2697	*
2698	* [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2699	*/
2700	if (!folio_buffers(folio)) {
2701	ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", folio->index);
2702	folio_clear_dirty(folio);
2703	folio_unlock(folio);
2704	continue;
2705	}
2706
2707	if (mpd->map.m_len == 0)
2708	mpd->start_pos = folio_pos(folio);
2709	mpd->next_pos = folio_next_pos(folio);
2710	/*
2711	* Writeout when we cannot modify metadata is simple.
2712	* Just submit the page. For data=journal mode we
2713	* first handle writeout of the page for checkpoint and
2714	* only after that handle delayed page dirtying. This
2715	* makes sure current data is checkpointed to the final
2716	* location before possibly journalling it again which
2717	* is desirable when the page is frequently dirtied
2718	* through a pin.
2719	*/
2720	if (!mpd->can_map) {
2721	err = mpage_submit_folio(mpd, folio);
2722	if (err < 0)
2723	goto out;
2724	/* Pending dirtying of journalled data? */
2725	if (folio_test_checked(folio)) {
2726	err = mpage_journal_page_buffers(handle,
2727	mpd, folio);
2728	if (err < 0)
2729	goto out;
2730	mpd->journalled_more_data = 1;
2731	}
2732	mpage_folio_done(mpd, folio);
2733	} else {
2734	/* Add all dirty buffers to mpd */
2735	lblk = EXT4_PG_TO_LBLK(mpd->inode, folio->index);
2736	head = folio_buffers(folio);
2737	err = mpage_process_page_bufs(mpd, head, bh: head,
2738	lblk);
2739	if (err <= 0)
2740	goto out;
2741	err = 0;
2742	}
2743	}
2744	folio_batch_release(fbatch: &fbatch);
2745	cond_resched();
2746	}
2747	mpd->scanned_until_end = 1;
2748	if (handle)
2749	ext4_journal_stop(handle);
2750	return 0;
2751	out:
2752	folio_batch_release(fbatch: &fbatch);
2753	if (handle)
2754	ext4_journal_stop(handle);
2755	return err;
2756	}
2757
2758	static int ext4_do_writepages(struct mpage_da_data *mpd)
2759	{
2760	struct writeback_control *wbc = mpd->wbc;
2761	pgoff_t writeback_index = 0;
2762	long nr_to_write = wbc->nr_to_write;
2763	int range_whole = 0;
2764	int cycled = 1;
2765	handle_t *handle = NULL;
2766	struct inode *inode = mpd->inode;
2767	struct address_space *mapping = inode->i_mapping;
2768	int needed_blocks, rsv_blocks = 0, ret = 0;
2769	struct ext4_sb_info *sbi = EXT4_SB(sb: mapping->host->i_sb);
2770	struct blk_plug plug;
2771	bool give_up_on_write = false;
2772
2773	trace_ext4_writepages(inode, wbc);
2774
2775	/*
2776	* No pages to write? This is mainly a kludge to avoid starting
2777	* a transaction for special inodes like journal inode on last iput()
2778	* because that could violate lock ordering on umount
2779	*/
2780	if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2781	goto out_writepages;
2782
2783	/*
2784	* If the filesystem has aborted, it is read-only, so return
2785	* right away instead of dumping stack traces later on that
2786	* will obscure the real source of the problem. We test
2787	* fs shutdown state instead of sb->s_flag's SB_RDONLY because
2788	* the latter could be true if the filesystem is mounted
2789	* read-only, and in that case, ext4_writepages should
2790	* *never* be called, so if that ever happens, we would want
2791	* the stack trace.
2792	*/
2793	ret = ext4_emergency_state(sb: mapping->host->i_sb);
2794	if (unlikely(ret))
2795	goto out_writepages;
2796
2797	/*
2798	* If we have inline data and arrive here, it means that
2799	* we will soon create the block for the 1st page, so
2800	* we'd better clear the inline data here.
2801	*/
2802	if (ext4_has_inline_data(inode)) {
2803	/* Just inode will be modified... */
2804	handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2805	if (IS_ERR(ptr: handle)) {
2806	ret = PTR_ERR(ptr: handle);
2807	goto out_writepages;
2808	}
2809	BUG_ON(ext4_test_inode_state(inode,
2810	EXT4_STATE_MAY_INLINE_DATA));
2811	ext4_destroy_inline_data(handle, inode);
2812	ext4_journal_stop(handle);
2813	}
2814
2815	/*
2816	* data=journal mode does not do delalloc so we just need to writeout /
2817	* journal already mapped buffers. On the other hand we need to commit
2818	* transaction to make data stable. We expect all the data to be
2819	* already in the journal (the only exception are DMA pinned pages
2820	* dirtied behind our back) so we commit transaction here and run the
2821	* writeback loop to checkpoint them. The checkpointing is not actually
2822	* necessary to make data persistent *but* quite a few places (extent
2823	* shifting operations, fsverity, ...) depend on being able to drop
2824	* pagecache pages after calling filemap_write_and_wait() and for that
2825	* checkpointing needs to happen.
2826	*/
2827	if (ext4_should_journal_data(inode)) {
2828	mpd->can_map = 0;
2829	if (wbc->sync_mode == WB_SYNC_ALL)
2830	ext4_fc_commit(journal: sbi->s_journal,
2831	EXT4_I(inode)->i_datasync_tid);
2832	}
2833	mpd->journalled_more_data = 0;
2834
2835	if (ext4_should_dioread_nolock(inode)) {
2836	int bpf = ext4_journal_blocks_per_folio(inode);
2837	/*
2838	* We may need to convert up to one extent per block in
2839	* the folio and we may dirty the inode.
2840	*/
2841	rsv_blocks = 1 + ext4_ext_index_trans_blocks(inode, extents: bpf);
2842	}
2843
2844	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2845	range_whole = 1;
2846
2847	if (wbc->range_cyclic) {
2848	writeback_index = mapping->writeback_index;
2849	if (writeback_index)
2850	cycled = 0;
2851	mpd->start_pos = writeback_index << PAGE_SHIFT;
2852	mpd->end_pos = LLONG_MAX;
2853	} else {
2854	mpd->start_pos = wbc->range_start;
2855	mpd->end_pos = wbc->range_end;
2856	}
2857
2858	ext4_io_submit_init(io: &mpd->io_submit, wbc);
2859	retry:
2860	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2861	tag_pages_for_writeback(mapping, start: mpd->start_pos >> PAGE_SHIFT,
2862	end: mpd->end_pos >> PAGE_SHIFT);
2863	blk_start_plug(&plug);
2864
2865	/*
2866	* First writeback pages that don't need mapping - we can avoid
2867	* starting a transaction unnecessarily and also avoid being blocked
2868	* in the block layer on device congestion while having transaction
2869	* started.
2870	*/
2871	mpd->do_map = 0;
2872	mpd->scanned_until_end = 0;
2873	mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2874	if (!mpd->io_submit.io_end) {
2875	ret = -ENOMEM;
2876	goto unplug;
2877	}
2878	ret = mpage_prepare_extent_to_map(mpd);
2879	/* Unlock pages we didn't use */
2880	mpage_release_unused_pages(mpd, invalidate: false);
2881	/* Submit prepared bio */
2882	ext4_io_submit(io: &mpd->io_submit);
2883	ext4_put_io_end_defer(io_end: mpd->io_submit.io_end);
2884	mpd->io_submit.io_end = NULL;
2885	if (ret < 0)
2886	goto unplug;
2887
2888	while (!mpd->scanned_until_end && wbc->nr_to_write > 0) {
2889	/* For each extent of pages we use new io_end */
2890	mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2891	if (!mpd->io_submit.io_end) {
2892	ret = -ENOMEM;
2893	break;
2894	}
2895
2896	WARN_ON_ONCE(!mpd->can_map);
2897	/*
2898	* We have two constraints: We find one extent to map and we
2899	* must always write out whole page (makes a difference when
2900	* blocksize < pagesize) so that we don't block on IO when we
2901	* try to write out the rest of the page. Journalled mode is
2902	* not supported by delalloc.
2903	*/
2904	BUG_ON(ext4_should_journal_data(inode));
2905	/*
2906	* Calculate the number of credits needed to reserve for one
2907	* extent of up to MAX_WRITEPAGES_EXTENT_LEN blocks. It will
2908	* attempt to extend the transaction or start a new iteration
2909	* if the reserved credits are insufficient.
2910	*/
2911	needed_blocks = ext4_chunk_trans_blocks(inode,
2912	MAX_WRITEPAGES_EXTENT_LEN);
2913	/* start a new transaction */
2914	handle = ext4_journal_start_with_reserve(inode,
2915	EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2916	if (IS_ERR(ptr: handle)) {
2917	ret = PTR_ERR(ptr: handle);
2918	ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2919	"%ld pages, ino %lu; err %d", __func__,
2920	wbc->nr_to_write, inode->i_ino, ret);
2921	/* Release allocated io_end */
2922	ext4_put_io_end(io_end: mpd->io_submit.io_end);
2923	mpd->io_submit.io_end = NULL;
2924	break;
2925	}
2926	mpd->do_map = 1;
2927
2928	trace_ext4_da_write_folios_start(inode, start_pos: mpd->start_pos,
2929	next_pos: mpd->next_pos, wbc);
2930	ret = mpage_prepare_extent_to_map(mpd);
2931	if (!ret && mpd->map.m_len)
2932	ret = mpage_map_and_submit_extent(handle, mpd,
2933	give_up_on_write: &give_up_on_write);
2934	/*
2935	* Caution: If the handle is synchronous,
2936	* ext4_journal_stop() can wait for transaction commit
2937	* to finish which may depend on writeback of pages to
2938	* complete or on page lock to be released. In that
2939	* case, we have to wait until after we have
2940	* submitted all the IO, released page locks we hold,
2941	* and dropped io_end reference (for extent conversion
2942	* to be able to complete) before stopping the handle.
2943	*/
2944	if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2945	ext4_journal_stop(handle);
2946	handle = NULL;
2947	mpd->do_map = 0;
2948	}
2949	/* Unlock pages we didn't use */
2950	mpage_release_unused_pages(mpd, invalidate: give_up_on_write);
2951	/* Submit prepared bio */
2952	ext4_io_submit(io: &mpd->io_submit);
2953
2954	/*
2955	* Drop our io_end reference we got from init. We have
2956	* to be careful and use deferred io_end finishing if
2957	* we are still holding the transaction as we can
2958	* release the last reference to io_end which may end
2959	* up doing unwritten extent conversion.
2960	*/
2961	if (handle) {
2962	ext4_put_io_end_defer(io_end: mpd->io_submit.io_end);
2963	ext4_journal_stop(handle);
2964	} else
2965	ext4_put_io_end(io_end: mpd->io_submit.io_end);
2966	mpd->io_submit.io_end = NULL;
2967	trace_ext4_da_write_folios_end(inode, start_pos: mpd->start_pos,
2968	next_pos: mpd->next_pos, wbc, ret);
2969
2970	if (ret == -ENOSPC && sbi->s_journal) {
2971	/*
2972	* Commit the transaction which would
2973	* free blocks released in the transaction
2974	* and try again
2975	*/
2976	jbd2_journal_force_commit_nested(sbi->s_journal);
2977	ret = 0;
2978	continue;
2979	}
2980	if (ret == -EAGAIN)
2981	ret = 0;
2982	/* Fatal error - ENOMEM, EIO... */
2983	if (ret)
2984	break;
2985	}
2986	unplug:
2987	blk_finish_plug(&plug);
2988	if (!ret && !cycled && wbc->nr_to_write > 0) {
2989	cycled = 1;
2990	mpd->end_pos = (writeback_index << PAGE_SHIFT) - 1;
2991	mpd->start_pos = 0;
2992	goto retry;
2993	}
2994
2995	/* Update index */
2996	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2997	/*
2998	* Set the writeback_index so that range_cyclic
2999	* mode will write it back later
3000	*/
3001	mapping->writeback_index = mpd->start_pos >> PAGE_SHIFT;
3002
3003	out_writepages:
3004	trace_ext4_writepages_result(inode, wbc, ret,
3005	pages_written: nr_to_write - wbc->nr_to_write);
3006	return ret;
3007	}
3008
3009	static int ext4_writepages(struct address_space *mapping,
3010	struct writeback_control *wbc)
3011	{
3012	struct super_block *sb = mapping->host->i_sb;
3013	struct mpage_da_data mpd = {
3014	.inode = mapping->host,
3015	.wbc = wbc,
3016	.can_map = 1,
3017	};
3018	int ret;
3019	int alloc_ctx;
3020
3021	ret = ext4_emergency_state(sb);
3022	if (unlikely(ret))
3023	return ret;
3024
3025	alloc_ctx = ext4_writepages_down_read(sb);
3026	ret = ext4_do_writepages(mpd: &mpd);
3027	/*
3028	* For data=journal writeback we could have come across pages marked
3029	* for delayed dirtying (PageChecked) which were just added to the
3030	* running transaction. Try once more to get them to stable storage.
3031	*/
3032	if (!ret && mpd.journalled_more_data)
3033	ret = ext4_do_writepages(mpd: &mpd);
3034	ext4_writepages_up_read(sb, ctx: alloc_ctx);
3035
3036	return ret;
3037	}
3038
3039	int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode)
3040	{
3041	struct writeback_control wbc = {
3042	.sync_mode = WB_SYNC_ALL,
3043	.nr_to_write = LONG_MAX,
3044	.range_start = jinode->i_dirty_start,
3045	.range_end = jinode->i_dirty_end,
3046	};
3047	struct mpage_da_data mpd = {
3048	.inode = jinode->i_vfs_inode,
3049	.wbc = &wbc,
3050	.can_map = 0,
3051	};
3052	return ext4_do_writepages(mpd: &mpd);
3053	}
3054
3055	static int ext4_dax_writepages(struct address_space *mapping,
3056	struct writeback_control *wbc)
3057	{
3058	int ret;
3059	long nr_to_write = wbc->nr_to_write;
3060	struct inode *inode = mapping->host;
3061	int alloc_ctx;
3062
3063	ret = ext4_emergency_state(sb: inode->i_sb);
3064	if (unlikely(ret))
3065	return ret;
3066
3067	alloc_ctx = ext4_writepages_down_read(sb: inode->i_sb);
3068	trace_ext4_writepages(inode, wbc);
3069
3070	ret = dax_writeback_mapping_range(mapping,
3071	dax_dev: EXT4_SB(sb: inode->i_sb)->s_daxdev, wbc);
3072	trace_ext4_writepages_result(inode, wbc, ret,
3073	pages_written: nr_to_write - wbc->nr_to_write);
3074	ext4_writepages_up_read(sb: inode->i_sb, ctx: alloc_ctx);
3075	return ret;
3076	}
3077
3078	static int ext4_nonda_switch(struct super_block *sb)
3079	{
3080	s64 free_clusters, dirty_clusters;
3081	struct ext4_sb_info *sbi = EXT4_SB(sb);
3082
3083	/*
3084	* switch to non delalloc mode if we are running low
3085	* on free block. The free block accounting via percpu
3086	* counters can get slightly wrong with percpu_counter_batch getting
3087	* accumulated on each CPU without updating global counters
3088	* Delalloc need an accurate free block accounting. So switch
3089	* to non delalloc when we are near to error range.
3090	*/
3091	free_clusters =
3092	percpu_counter_read_positive(fbc: &sbi->s_freeclusters_counter);
3093	dirty_clusters =
3094	percpu_counter_read_positive(fbc: &sbi->s_dirtyclusters_counter);
3095	/*
3096	* Start pushing delalloc when 1/2 of free blocks are dirty.
3097	*/
3098	if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3099	try_to_writeback_inodes_sb(sb, reason: WB_REASON_FS_FREE_SPACE);
3100
3101	if (2 * free_clusters < 3 * dirty_clusters ||
3102	free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3103	/*
3104	* free block count is less than 150% of dirty blocks
3105	* or free blocks is less than watermark
3106	*/
3107	return 1;
3108	}
3109	return 0;
3110	}
3111
3112	static int ext4_da_write_begin(const struct kiocb *iocb,
3113	struct address_space *mapping,
3114	loff_t pos, unsigned len,
3115	struct folio **foliop, void **fsdata)
3116	{
3117	int ret, retries = 0;
3118	struct folio *folio;
3119	pgoff_t index;
3120	struct inode *inode = mapping->host;
3121
3122	ret = ext4_emergency_state(sb: inode->i_sb);
3123	if (unlikely(ret))
3124	return ret;
3125
3126	index = pos >> PAGE_SHIFT;
3127
3128	if (ext4_nonda_switch(sb: inode->i_sb) || ext4_verity_in_progress(inode)) {
3129	*fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3130	return ext4_write_begin(iocb, mapping, pos,
3131	len, foliop, fsdata);
3132	}
3133	*fsdata = (void *)0;
3134	trace_ext4_da_write_begin(inode, pos, len);
3135
3136	if (ext4_test_inode_state(inode, bit: EXT4_STATE_MAY_INLINE_DATA)) {
3137	ret = ext4_generic_write_inline_data(mapping, inode, pos, len,
3138	foliop, fsdata, da: true);
3139	if (ret < 0)
3140	return ret;
3141	if (ret == 1)
3142	return 0;
3143	}
3144
3145	retry:
3146	folio = write_begin_get_folio(iocb, mapping, index, len);
3147	if (IS_ERR(ptr: folio))
3148	return PTR_ERR(ptr: folio);
3149
3150	if (len > folio_next_pos(folio) - pos)
3151	len = folio_next_pos(folio) - pos;
3152
3153	ret = ext4_block_write_begin(NULL, folio, pos, len,
3154	get_block: ext4_da_get_block_prep);
3155	if (ret < 0) {
3156	folio_unlock(folio);
3157	folio_put(folio);
3158	/*
3159	* ext4_block_write_begin may have instantiated a few blocks
3160	* outside i_size. Trim these off again. Don't need
3161	* i_size_read because we hold inode lock.
3162	*/
3163	if (pos + len > inode->i_size)
3164	ext4_truncate_failed_write(inode);
3165
3166	if (ret == -ENOSPC &&
3167	ext4_should_retry_alloc(sb: inode->i_sb, retries: &retries))
3168	goto retry;
3169	return ret;
3170	}
3171
3172	*foliop = folio;
3173	return ret;
3174	}
3175
3176	/*
3177	* Check if we should update i_disksize
3178	* when write to the end of file but not require block allocation
3179	*/
3180	static int ext4_da_should_update_i_disksize(struct folio *folio,
3181	unsigned long offset)
3182	{
3183	struct buffer_head *bh;
3184	struct inode *inode = folio->mapping->host;
3185	unsigned int idx;
3186	int i;
3187
3188	bh = folio_buffers(folio);
3189	idx = offset >> inode->i_blkbits;
3190
3191	for (i = 0; i < idx; i++)
3192	bh = bh->b_this_page;
3193
3194	if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3195	return 0;
3196	return 1;
3197	}
3198
3199	static int ext4_da_do_write_end(struct address_space *mapping,
3200	loff_t pos, unsigned len, unsigned copied,
3201	struct folio *folio)
3202	{
3203	struct inode *inode = mapping->host;
3204	loff_t old_size = inode->i_size;
3205	bool disksize_changed = false;
3206	loff_t new_i_size, zero_len = 0;
3207	handle_t *handle;
3208
3209	if (unlikely(!folio_buffers(folio))) {
3210	folio_unlock(folio);
3211	folio_put(folio);
3212	return -EIO;
3213	}
3214	/*
3215	* block_write_end() will mark the inode as dirty with I_DIRTY_PAGES
3216	* flag, which all that's needed to trigger page writeback.
3217	*/
3218	copied = block_write_end(pos, len, copied, folio);
3219	new_i_size = pos + copied;
3220
3221	/*
3222	* It's important to update i_size while still holding folio lock,
3223	* because folio writeout could otherwise come in and zero beyond
3224	* i_size.
3225	*
3226	* Since we are holding inode lock, we are sure i_disksize <=
3227	* i_size. We also know that if i_disksize < i_size, there are
3228	* delalloc writes pending in the range up to i_size. If the end of
3229	* the current write is <= i_size, there's no need to touch
3230	* i_disksize since writeback will push i_disksize up to i_size
3231	* eventually. If the end of the current write is > i_size and
3232	* inside an allocated block which ext4_da_should_update_i_disksize()
3233	* checked, we need to update i_disksize here as certain
3234	* ext4_writepages() paths not allocating blocks and update i_disksize.
3235	*/
3236	if (new_i_size > inode->i_size) {
3237	unsigned long end;
3238
3239	i_size_write(inode, i_size: new_i_size);
3240	end = offset_in_folio(folio, new_i_size - 1);
3241	if (copied && ext4_da_should_update_i_disksize(folio, offset: end)) {
3242	ext4_update_i_disksize(inode, newsize: new_i_size);
3243	disksize_changed = true;
3244	}
3245	}
3246
3247	folio_unlock(folio);
3248	folio_put(folio);
3249
3250	if (pos > old_size) {
3251	pagecache_isize_extended(inode, from: old_size, to: pos);
3252	zero_len = pos - old_size;
3253	}
3254
3255	if (!disksize_changed && !zero_len)
3256	return copied;
3257
3258	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3259	if (IS_ERR(ptr: handle))
3260	return PTR_ERR(ptr: handle);
3261	if (zero_len)
3262	ext4_zero_partial_blocks(handle, inode, lstart: old_size, lend: zero_len);
3263	ext4_mark_inode_dirty(handle, inode);
3264	ext4_journal_stop(handle);
3265
3266	return copied;
3267	}
3268
3269	static int ext4_da_write_end(const struct kiocb *iocb,
3270	struct address_space *mapping,
3271	loff_t pos, unsigned len, unsigned copied,
3272	struct folio *folio, void *fsdata)
3273	{
3274	struct inode *inode = mapping->host;
3275	int write_mode = (int)(unsigned long)fsdata;
3276
3277	if (write_mode == FALL_BACK_TO_NONDELALLOC)
3278	return ext4_write_end(iocb, mapping, pos,
3279	len, copied, folio, fsdata);
3280
3281	trace_ext4_da_write_end(inode, pos, len, copied);
3282
3283	if (write_mode != CONVERT_INLINE_DATA &&
3284	ext4_test_inode_state(inode, bit: EXT4_STATE_MAY_INLINE_DATA) &&
3285	ext4_has_inline_data(inode))
3286	return ext4_write_inline_data_end(inode, pos, len, copied,
3287	folio);
3288
3289	if (unlikely(copied < len) && !folio_test_uptodate(folio))
3290	copied = 0;
3291
3292	return ext4_da_do_write_end(mapping, pos, len, copied, folio);
3293	}
3294
3295	/*
3296	* Force all delayed allocation blocks to be allocated for a given inode.
3297	*/
3298	int ext4_alloc_da_blocks(struct inode *inode)
3299	{
3300	trace_ext4_alloc_da_blocks(inode);
3301
3302	if (!EXT4_I(inode)->i_reserved_data_blocks)
3303	return 0;
3304
3305	/*
3306	* We do something simple for now. The filemap_flush() will
3307	* also start triggering a write of the data blocks, which is
3308	* not strictly speaking necessary (and for users of
3309	* laptop_mode, not even desirable). However, to do otherwise
3310	* would require replicating code paths in:
3311	*
3312	* ext4_writepages() ->
3313	* write_cache_pages() ---> (via passed in callback function)
3314	* __mpage_da_writepage() -->
3315	* mpage_add_bh_to_extent()
3316	* mpage_da_map_blocks()
3317	*
3318	* The problem is that write_cache_pages(), located in
3319	* mm/page-writeback.c, marks pages clean in preparation for
3320	* doing I/O, which is not desirable if we're not planning on
3321	* doing I/O at all.
3322	*
3323	* We could call write_cache_pages(), and then redirty all of
3324	* the pages by calling redirty_page_for_writepage() but that
3325	* would be ugly in the extreme. So instead we would need to
3326	* replicate parts of the code in the above functions,
3327	* simplifying them because we wouldn't actually intend to
3328	* write out the pages, but rather only collect contiguous
3329	* logical block extents, call the multi-block allocator, and
3330	* then update the buffer heads with the block allocations.
3331	*
3332	* For now, though, we'll cheat by calling filemap_flush(),
3333	* which will map the blocks, and start the I/O, but not
3334	* actually wait for the I/O to complete.
3335	*/
3336	return filemap_flush(inode->i_mapping);
3337	}
3338
3339	/*
3340	* bmap() is special. It gets used by applications such as lilo and by
3341	* the swapper to find the on-disk block of a specific piece of data.
3342	*
3343	* Naturally, this is dangerous if the block concerned is still in the
3344	* journal. If somebody makes a swapfile on an ext4 data-journaling
3345	* filesystem and enables swap, then they may get a nasty shock when the
3346	* data getting swapped to that swapfile suddenly gets overwritten by
3347	* the original zero's written out previously to the journal and
3348	* awaiting writeback in the kernel's buffer cache.
3349	*
3350	* So, if we see any bmap calls here on a modified, data-journaled file,
3351	* take extra steps to flush any blocks which might be in the cache.
3352	*/
3353	static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3354	{
3355	struct inode *inode = mapping->host;
3356	sector_t ret = 0;
3357
3358	inode_lock_shared(inode);
3359	/*
3360	* We can get here for an inline file via the FIBMAP ioctl
3361	*/
3362	if (ext4_has_inline_data(inode))
3363	goto out;
3364
3365	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3366	(test_opt(inode->i_sb, DELALLOC) ||
3367	ext4_should_journal_data(inode))) {
3368	/*
3369	* With delalloc or journalled data we want to sync the file so
3370	* that we can make sure we allocate blocks for file and data
3371	* is in place for the user to see it
3372	*/
3373	filemap_write_and_wait(mapping);
3374	}
3375
3376	ret = iomap_bmap(mapping, bno: block, ops: &ext4_iomap_ops);
3377
3378	out:
3379	inode_unlock_shared(inode);
3380	return ret;
3381	}
3382
3383	static int ext4_read_folio(struct file *file, struct folio *folio)
3384	{
3385	int ret = -EAGAIN;
3386	struct inode *inode = folio->mapping->host;
3387
3388	trace_ext4_read_folio(inode, folio);
3389
3390	if (ext4_has_inline_data(inode))
3391	ret = ext4_readpage_inline(inode, folio);
3392
3393	if (ret == -EAGAIN)
3394	return ext4_mpage_readpages(inode, NULL, folio);
3395
3396	return ret;
3397	}
3398
3399	static void ext4_readahead(struct readahead_control *rac)
3400	{
3401	struct inode *inode = rac->mapping->host;
3402
3403	/* If the file has inline data, no need to do readahead. */
3404	if (ext4_has_inline_data(inode))
3405	return;
3406
3407	ext4_mpage_readpages(inode, rac, NULL);
3408	}
3409
3410	static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3411	size_t length)
3412	{
3413	trace_ext4_invalidate_folio(folio, offset, length);
3414
3415	/* No journalling happens on data buffers when this function is used */
3416	WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3417
3418	block_invalidate_folio(folio, offset, length);
3419	}
3420
3421	static int __ext4_journalled_invalidate_folio(struct folio *folio,
3422	size_t offset, size_t length)
3423	{
3424	journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3425
3426	trace_ext4_journalled_invalidate_folio(folio, offset, length);
3427
3428	/*
3429	* If it's a full truncate we just forget about the pending dirtying
3430	*/
3431	if (offset == 0 && length == folio_size(folio))
3432	folio_clear_checked(folio);
3433
3434	return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3435	}
3436
3437	/* Wrapper for aops... */
3438	static void ext4_journalled_invalidate_folio(struct folio *folio,
3439	size_t offset,
3440	size_t length)
3441	{
3442	WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3443	}
3444
3445	static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3446	{
3447	struct inode *inode = folio->mapping->host;
3448	journal_t *journal = EXT4_JOURNAL(inode);
3449
3450	trace_ext4_release_folio(inode, folio);
3451
3452	/* Page has dirty journalled data -> cannot release */
3453	if (folio_test_checked(folio))
3454	return false;
3455	if (journal)
3456	return jbd2_journal_try_to_free_buffers(journal, folio);
3457	else
3458	return try_to_free_buffers(folio);
3459	}
3460
3461	static bool ext4_inode_datasync_dirty(struct inode *inode)
3462	{
3463	journal_t *journal = EXT4_SB(sb: inode->i_sb)->s_journal;
3464
3465	if (journal) {
3466	if (jbd2_transaction_committed(journal,
3467	EXT4_I(inode)->i_datasync_tid))
3468	return false;
3469	if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3470	return !list_empty(head: &EXT4_I(inode)->i_fc_list);
3471	return true;
3472	}
3473
3474	/* Any metadata buffers to write? */
3475	if (!list_empty(head: &inode->i_mapping->i_private_list))
3476	return true;
3477	return inode_state_read_once(inode) & I_DIRTY_DATASYNC;
3478	}
3479
3480	static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3481	struct ext4_map_blocks *map, loff_t offset,
3482	loff_t length, unsigned int flags)
3483	{
3484	u8 blkbits = inode->i_blkbits;
3485
3486	/*
3487	* Writes that span EOF might trigger an I/O size update on completion,
3488	* so consider them to be dirty for the purpose of O_DSYNC, even if
3489	* there is no other metadata changes being made or are pending.
3490	*/
3491	iomap->flags = 0;
3492	if (ext4_inode_datasync_dirty(inode) ||
3493	offset + length > i_size_read(inode))
3494	iomap->flags |= IOMAP_F_DIRTY;
3495
3496	if (map->m_flags & EXT4_MAP_NEW)
3497	iomap->flags |= IOMAP_F_NEW;
3498
3499	/* HW-offload atomics are always used */
3500	if (flags & IOMAP_ATOMIC)
3501	iomap->flags |= IOMAP_F_ATOMIC_BIO;
3502
3503	if (flags & IOMAP_DAX)
3504	iomap->dax_dev = EXT4_SB(sb: inode->i_sb)->s_daxdev;
3505	else
3506	iomap->bdev = inode->i_sb->s_bdev;
3507	iomap->offset = EXT4_LBLK_TO_B(inode, map->m_lblk);
3508	iomap->length = EXT4_LBLK_TO_B(inode, map->m_len);
3509
3510	if ((map->m_flags & EXT4_MAP_MAPPED) &&
3511	!ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
3512	iomap->flags |= IOMAP_F_MERGED;
3513
3514	/*
3515	* Flags passed to ext4_map_blocks() for direct I/O writes can result
3516	* in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3517	* set. In order for any allocated unwritten extents to be converted
3518	* into written extents correctly within the ->end_io() handler, we
3519	* need to ensure that the iomap->type is set appropriately. Hence, the
3520	* reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3521	* been set first.
3522	*/
3523	if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3524	iomap->type = IOMAP_UNWRITTEN;
3525	iomap->addr = (u64) map->m_pblk << blkbits;
3526	if (flags & IOMAP_DAX)
3527	iomap->addr += EXT4_SB(sb: inode->i_sb)->s_dax_part_off;
3528	} else if (map->m_flags & EXT4_MAP_MAPPED) {
3529	iomap->type = IOMAP_MAPPED;
3530	iomap->addr = (u64) map->m_pblk << blkbits;
3531	if (flags & IOMAP_DAX)
3532	iomap->addr += EXT4_SB(sb: inode->i_sb)->s_dax_part_off;
3533	} else if (map->m_flags & EXT4_MAP_DELAYED) {
3534	iomap->type = IOMAP_DELALLOC;
3535	iomap->addr = IOMAP_NULL_ADDR;
3536	} else {
3537	iomap->type = IOMAP_HOLE;
3538	iomap->addr = IOMAP_NULL_ADDR;
3539	}
3540	}
3541
3542	static int ext4_map_blocks_atomic_write_slow(handle_t *handle,
3543	struct inode *inode, struct ext4_map_blocks *map)
3544	{
3545	ext4_lblk_t m_lblk = map->m_lblk;
3546	unsigned int m_len = map->m_len;
3547	unsigned int mapped_len = 0, m_flags = 0;
3548	ext4_fsblk_t next_pblk = 0;
3549	bool check_next_pblk = false;
3550	int ret = 0;
3551
3552	WARN_ON_ONCE(!ext4_has_feature_bigalloc(inode->i_sb));
3553
3554	/*
3555	* This is a slow path in case of mixed mapping. We use
3556	* EXT4_GET_BLOCKS_CREATE_ZERO flag here to make sure we get a single
3557	* contiguous mapped mapping. This will ensure any unwritten or hole
3558	* regions within the requested range is zeroed out and we return
3559	* a single contiguous mapped extent.
3560	*/
3561	m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3562
3563	do {
3564	ret = ext4_map_blocks(handle, inode, map, flags: m_flags);
3565	if (ret < 0 && ret != -ENOSPC)
3566	goto out_err;
3567	/*
3568	* This should never happen, but let's return an error code to
3569	* avoid an infinite loop in here.
3570	*/
3571	if (ret == 0) {
3572	ret = -EFSCORRUPTED;
3573	ext4_warning_inode(inode,
3574	"ext4_map_blocks() couldn't allocate blocks m_flags: 0x%x, ret:%d",
3575	m_flags, ret);
3576	goto out_err;
3577	}
3578	/*
3579	* With bigalloc we should never get ENOSPC nor discontiguous
3580	* physical extents.
3581	*/
3582	if ((check_next_pblk && next_pblk != map->m_pblk) ||
3583	ret == -ENOSPC) {
3584	ext4_warning_inode(inode,
3585	"Non-contiguous allocation detected: expected %llu, got %llu, "
3586	"or ext4_map_blocks() returned out of space ret: %d",
3587	next_pblk, map->m_pblk, ret);
3588	ret = -EFSCORRUPTED;
3589	goto out_err;
3590	}
3591	next_pblk = map->m_pblk + map->m_len;
3592	check_next_pblk = true;
3593
3594	mapped_len += map->m_len;
3595	map->m_lblk += map->m_len;
3596	map->m_len = m_len - mapped_len;
3597	} while (mapped_len < m_len);
3598
3599	/*
3600	* We might have done some work in above loop, so we need to query the
3601	* start of the physical extent, based on the origin m_lblk and m_len.
3602	* Let's also ensure we were able to allocate the required range for
3603	* mixed mapping case.
3604	*/
3605	map->m_lblk = m_lblk;
3606	map->m_len = m_len;
3607	map->m_flags = 0;
3608
3609	ret = ext4_map_blocks(handle, inode, map,
3610	EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF);
3611	if (ret != m_len) {
3612	ext4_warning_inode(inode,
3613	"allocation failed for atomic write request m_lblk:%u, m_len:%u, ret:%d\n",
3614	m_lblk, m_len, ret);
3615	ret = -EINVAL;
3616	}
3617	return ret;
3618
3619	out_err:
3620	/* reset map before returning an error */
3621	map->m_lblk = m_lblk;
3622	map->m_len = m_len;
3623	map->m_flags = 0;
3624	return ret;
3625	}
3626
3627	/*
3628	* ext4_map_blocks_atomic: Helper routine to ensure the entire requested
3629	* range in @map [lblk, lblk + len) is one single contiguous extent with no
3630	* mixed mappings.
3631	*
3632	* We first use m_flags passed to us by our caller (ext4_iomap_alloc()).
3633	* We only call EXT4_GET_BLOCKS_ZERO in the slow path, when the underlying
3634	* physical extent for the requested range does not have a single contiguous
3635	* mapping type i.e. (Hole, Mapped, or Unwritten) throughout.
3636	* In that case we will loop over the requested range to allocate and zero out
3637	* the unwritten / holes in between, to get a single mapped extent from
3638	* [m_lblk, m_lblk + m_len). Note that this is only possible because we know
3639	* this can be called only with bigalloc enabled filesystem where the underlying
3640	* cluster is already allocated. This avoids allocating discontiguous extents
3641	* in the slow path due to multiple calls to ext4_map_blocks().
3642	* The slow path is mostly non-performance critical path, so it should be ok to
3643	* loop using ext4_map_blocks() with appropriate flags to allocate & zero the
3644	* underlying short holes/unwritten extents within the requested range.
3645	*/
3646	static int ext4_map_blocks_atomic_write(handle_t *handle, struct inode *inode,
3647	struct ext4_map_blocks *map, int m_flags,
3648	bool *force_commit)
3649	{
3650	ext4_lblk_t m_lblk = map->m_lblk;
3651	unsigned int m_len = map->m_len;
3652	int ret = 0;
3653
3654	WARN_ON_ONCE(m_len > 1 && !ext4_has_feature_bigalloc(inode->i_sb));
3655
3656	ret = ext4_map_blocks(handle, inode, map, flags: m_flags);
3657	if (ret < 0 || ret == m_len)
3658	goto out;
3659	/*
3660	* This is a mixed mapping case where we were not able to allocate
3661	* a single contiguous extent. In that case let's reset requested
3662	* mapping and call the slow path.
3663	*/
3664	map->m_lblk = m_lblk;
3665	map->m_len = m_len;
3666	map->m_flags = 0;
3667
3668	/*
3669	* slow path means we have mixed mapping, that means we will need
3670	* to force txn commit.
3671	*/
3672	*force_commit = true;
3673	return ext4_map_blocks_atomic_write_slow(handle, inode, map);
3674	out:
3675	return ret;
3676	}
3677
3678	static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3679	unsigned int flags)
3680	{
3681	handle_t *handle;
3682	int ret, dio_credits, m_flags = 0, retries = 0;
3683	bool force_commit = false;
3684
3685	/*
3686	* Trim the mapping request to the maximum value that we can map at
3687	* once for direct I/O.
3688	*/
3689	if (map->m_len > DIO_MAX_BLOCKS)
3690	map->m_len = DIO_MAX_BLOCKS;
3691
3692	/*
3693	* journal credits estimation for atomic writes. We call
3694	* ext4_map_blocks(), to find if there could be a mixed mapping. If yes,
3695	* then let's assume the no. of pextents required can be m_len i.e.
3696	* every alternate block can be unwritten and hole.
3697	*/
3698	if (flags & IOMAP_ATOMIC) {
3699	unsigned int orig_mlen = map->m_len;
3700
3701	ret = ext4_map_blocks(NULL, inode, map, flags: 0);
3702	if (ret < 0)
3703	return ret;
3704	if (map->m_len < orig_mlen) {
3705	map->m_len = orig_mlen;
3706	dio_credits = ext4_meta_trans_blocks(inode, lblocks: orig_mlen,
3707	pextents: map->m_len);
3708	} else {
3709	dio_credits = ext4_chunk_trans_blocks(inode,
3710	nrblocks: map->m_len);
3711	}
3712	} else {
3713	dio_credits = ext4_chunk_trans_blocks(inode, nrblocks: map->m_len);
3714	}
3715
3716	retry:
3717	/*
3718	* Either we allocate blocks and then don't get an unwritten extent, so
3719	* in that case we have reserved enough credits. Or, the blocks are
3720	* already allocated and unwritten. In that case, the extent conversion
3721	* fits into the credits as well.
3722	*/
3723	handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3724	if (IS_ERR(ptr: handle))
3725	return PTR_ERR(ptr: handle);
3726
3727	/*
3728	* DAX and direct I/O are the only two operations that are currently
3729	* supported with IOMAP_WRITE.
3730	*/
3731	WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3732	if (flags & IOMAP_DAX)
3733	m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3734	/*
3735	* We use i_size instead of i_disksize here because delalloc writeback
3736	* can complete at any point during the I/O and subsequently push the
3737	* i_disksize out to i_size. This could be beyond where direct I/O is
3738	* happening and thus expose allocated blocks to direct I/O reads.
3739	*/
3740	else if (EXT4_LBLK_TO_B(inode, map->m_lblk) >= i_size_read(inode))
3741	m_flags = EXT4_GET_BLOCKS_CREATE;
3742	else if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
3743	m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3744
3745	if (flags & IOMAP_ATOMIC)
3746	ret = ext4_map_blocks_atomic_write(handle, inode, map, m_flags,
3747	force_commit: &force_commit);
3748	else
3749	ret = ext4_map_blocks(handle, inode, map, flags: m_flags);
3750
3751	/*
3752	* We cannot fill holes in indirect tree based inodes as that could
3753	* expose stale data in the case of a crash. Use the magic error code
3754	* to fallback to buffered I/O.
3755	*/
3756	if (!m_flags && !ret)
3757	ret = -ENOTBLK;
3758
3759	ext4_journal_stop(handle);
3760	if (ret == -ENOSPC && ext4_should_retry_alloc(sb: inode->i_sb, retries: &retries))
3761	goto retry;
3762
3763	/*
3764	* Force commit the current transaction if the allocation spans a mixed
3765	* mapping range. This ensures any pending metadata updates (like
3766	* unwritten to written extents conversion) in this range are in
3767	* consistent state with the file data blocks, before performing the
3768	* actual write I/O. If the commit fails, the whole I/O must be aborted
3769	* to prevent any possible torn writes.
3770	*/
3771	if (ret > 0 && force_commit) {
3772	int ret2;
3773
3774	ret2 = ext4_force_commit(sb: inode->i_sb);
3775	if (ret2)
3776	return ret2;
3777	}
3778
3779	return ret;
3780	}
3781
3782
3783	static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3784	unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3785	{
3786	int ret;
3787	struct ext4_map_blocks map;
3788	u8 blkbits = inode->i_blkbits;
3789	unsigned int orig_mlen;
3790
3791	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3792	return -EINVAL;
3793
3794	if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3795	return -ERANGE;
3796
3797	/*
3798	* Calculate the first and last logical blocks respectively.
3799	*/
3800	map.m_lblk = offset >> blkbits;
3801	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3802	EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3803	orig_mlen = map.m_len;
3804
3805	if (flags & IOMAP_WRITE) {
3806	/*
3807	* We check here if the blocks are already allocated, then we
3808	* don't need to start a journal txn and we can directly return
3809	* the mapping information. This could boost performance
3810	* especially in multi-threaded overwrite requests.
3811	*/
3812	if (offset + length <= i_size_read(inode)) {
3813	ret = ext4_map_blocks(NULL, inode, map: &map, flags: 0);
3814	/*
3815	* For atomic writes the entire requested length should
3816	* be mapped.
3817	*/
3818	if (map.m_flags & EXT4_MAP_MAPPED) {
3819	if ((!(flags & IOMAP_ATOMIC) && ret > 0) ||
3820	(flags & IOMAP_ATOMIC && ret >= orig_mlen))
3821	goto out;
3822	}
3823	map.m_len = orig_mlen;
3824	}
3825	ret = ext4_iomap_alloc(inode, map: &map, flags);
3826	} else {
3827	/*
3828	* This can be called for overwrites path from
3829	* ext4_iomap_overwrite_begin().
3830	*/
3831	ret = ext4_map_blocks(NULL, inode, map: &map, flags: 0);
3832	}
3833
3834	if (ret < 0)
3835	return ret;
3836	out:
3837	/*
3838	* When inline encryption is enabled, sometimes I/O to an encrypted file
3839	* has to be broken up to guarantee DUN contiguity. Handle this by
3840	* limiting the length of the mapping returned.
3841	*/
3842	map.m_len = fscrypt_limit_io_blocks(inode, lblk: map.m_lblk, nr_blocks: map.m_len);
3843
3844	/*
3845	* Before returning to iomap, let's ensure the allocated mapping
3846	* covers the entire requested length for atomic writes.
3847	*/
3848	if (flags & IOMAP_ATOMIC) {
3849	if (map.m_len < (length >> blkbits)) {
3850	WARN_ON_ONCE(1);
3851	return -EINVAL;
3852	}
3853	}
3854	ext4_set_iomap(inode, iomap, map: &map, offset, length, flags);
3855
3856	return 0;
3857	}
3858
3859	static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3860	loff_t length, unsigned flags, struct iomap *iomap,
3861	struct iomap *srcmap)
3862	{
3863	int ret;
3864
3865	/*
3866	* Even for writes we don't need to allocate blocks, so just pretend
3867	* we are reading to save overhead of starting a transaction.
3868	*/
3869	flags &= ~IOMAP_WRITE;
3870	ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3871	WARN_ON_ONCE(!ret && iomap->type != IOMAP_MAPPED);
3872	return ret;
3873	}
3874
3875	const struct iomap_ops ext4_iomap_ops = {
3876	.iomap_begin = ext4_iomap_begin,
3877	};
3878
3879	const struct iomap_ops ext4_iomap_overwrite_ops = {
3880	.iomap_begin = ext4_iomap_overwrite_begin,
3881	};
3882
3883	static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3884	loff_t length, unsigned int flags,
3885	struct iomap *iomap, struct iomap *srcmap)
3886	{
3887	int ret;
3888	struct ext4_map_blocks map;
3889	u8 blkbits = inode->i_blkbits;
3890
3891	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3892	return -EINVAL;
3893
3894	if (ext4_has_inline_data(inode)) {
3895	ret = ext4_inline_data_iomap(inode, iomap);
3896	if (ret != -EAGAIN) {
3897	if (ret == 0 && offset >= iomap->length)
3898	ret = -ENOENT;
3899	return ret;
3900	}
3901	}
3902
3903	/*
3904	* Calculate the first and last logical block respectively.
3905	*/
3906	map.m_lblk = offset >> blkbits;
3907	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3908	EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3909
3910	/*
3911	* Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3912	* So handle it here itself instead of querying ext4_map_blocks().
3913	* Since ext4_map_blocks() will warn about it and will return
3914	* -EIO error.
3915	*/
3916	if (!(ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))) {
3917	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
3918
3919	if (offset >= sbi->s_bitmap_maxbytes) {
3920	map.m_flags = 0;
3921	goto set_iomap;
3922	}
3923	}
3924
3925	ret = ext4_map_blocks(NULL, inode, map: &map, flags: 0);
3926	if (ret < 0)
3927	return ret;
3928	set_iomap:
3929	ext4_set_iomap(inode, iomap, map: &map, offset, length, flags);
3930
3931	return 0;
3932	}
3933
3934	const struct iomap_ops ext4_iomap_report_ops = {
3935	.iomap_begin = ext4_iomap_begin_report,
3936	};
3937
3938	/*
3939	* For data=journal mode, folio should be marked dirty only when it was
3940	* writeably mapped. When that happens, it was already attached to the
3941	* transaction and marked as jbddirty (we take care of this in
3942	* ext4_page_mkwrite()). On transaction commit, we writeprotect page mappings
3943	* so we should have nothing to do here, except for the case when someone
3944	* had the page pinned and dirtied the page through this pin (e.g. by doing
3945	* direct IO to it). In that case we'd need to attach buffers here to the
3946	* transaction but we cannot due to lock ordering. We cannot just dirty the
3947	* folio and leave attached buffers clean, because the buffers' dirty state is
3948	* "definitive". We cannot just set the buffers dirty or jbddirty because all
3949	* the journalling code will explode. So what we do is to mark the folio
3950	* "pending dirty" and next time ext4_writepages() is called, attach buffers
3951	* to the transaction appropriately.
3952	*/
3953	static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3954	struct folio *folio)
3955	{
3956	WARN_ON_ONCE(!folio_buffers(folio));
3957	if (folio_maybe_dma_pinned(folio))
3958	folio_set_checked(folio);
3959	return filemap_dirty_folio(mapping, folio);
3960	}
3961
3962	static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3963	{
3964	WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3965	WARN_ON_ONCE(!folio_buffers(folio));
3966	return block_dirty_folio(mapping, folio);
3967	}
3968
3969	static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3970	struct file *file, sector_t *span)
3971	{
3972	return iomap_swapfile_activate(sis, swap_file: file, pagespan: span,
3973	ops: &ext4_iomap_report_ops);
3974	}
3975
3976	static const struct address_space_operations ext4_aops = {
3977	.read_folio = ext4_read_folio,
3978	.readahead = ext4_readahead,
3979	.writepages = ext4_writepages,
3980	.write_begin = ext4_write_begin,
3981	.write_end = ext4_write_end,
3982	.dirty_folio = ext4_dirty_folio,
3983	.bmap = ext4_bmap,
3984	.invalidate_folio = ext4_invalidate_folio,
3985	.release_folio = ext4_release_folio,
3986	.migrate_folio = buffer_migrate_folio,
3987	.is_partially_uptodate = block_is_partially_uptodate,
3988	.error_remove_folio = generic_error_remove_folio,
3989	.swap_activate = ext4_iomap_swap_activate,
3990	};
3991
3992	static const struct address_space_operations ext4_journalled_aops = {
3993	.read_folio = ext4_read_folio,
3994	.readahead = ext4_readahead,
3995	.writepages = ext4_writepages,
3996	.write_begin = ext4_write_begin,
3997	.write_end = ext4_journalled_write_end,
3998	.dirty_folio = ext4_journalled_dirty_folio,
3999	.bmap = ext4_bmap,
4000	.invalidate_folio = ext4_journalled_invalidate_folio,
4001	.release_folio = ext4_release_folio,
4002	.migrate_folio = buffer_migrate_folio_norefs,
4003	.is_partially_uptodate = block_is_partially_uptodate,
4004	.error_remove_folio = generic_error_remove_folio,
4005	.swap_activate = ext4_iomap_swap_activate,
4006	};
4007
4008	static const struct address_space_operations ext4_da_aops = {
4009	.read_folio = ext4_read_folio,
4010	.readahead = ext4_readahead,
4011	.writepages = ext4_writepages,
4012	.write_begin = ext4_da_write_begin,
4013	.write_end = ext4_da_write_end,
4014	.dirty_folio = ext4_dirty_folio,
4015	.bmap = ext4_bmap,
4016	.invalidate_folio = ext4_invalidate_folio,
4017	.release_folio = ext4_release_folio,
4018	.migrate_folio = buffer_migrate_folio,
4019	.is_partially_uptodate = block_is_partially_uptodate,
4020	.error_remove_folio = generic_error_remove_folio,
4021	.swap_activate = ext4_iomap_swap_activate,
4022	};
4023
4024	static const struct address_space_operations ext4_dax_aops = {
4025	.writepages = ext4_dax_writepages,
4026	.dirty_folio = noop_dirty_folio,
4027	.bmap = ext4_bmap,
4028	.swap_activate = ext4_iomap_swap_activate,
4029	};
4030
4031	void ext4_set_aops(struct inode *inode)
4032	{
4033	switch (ext4_inode_journal_mode(inode)) {
4034	case EXT4_INODE_ORDERED_DATA_MODE:
4035	case EXT4_INODE_WRITEBACK_DATA_MODE:
4036	break;
4037	case EXT4_INODE_JOURNAL_DATA_MODE:
4038	inode->i_mapping->a_ops = &ext4_journalled_aops;
4039	return;
4040	default:
4041	BUG();
4042	}
4043	if (IS_DAX(inode))
4044	inode->i_mapping->a_ops = &ext4_dax_aops;
4045	else if (test_opt(inode->i_sb, DELALLOC))
4046	inode->i_mapping->a_ops = &ext4_da_aops;
4047	else
4048	inode->i_mapping->a_ops = &ext4_aops;
4049	}
4050
4051	/*
4052	* Here we can't skip an unwritten buffer even though it usually reads zero
4053	* because it might have data in pagecache (eg, if called from ext4_zero_range,
4054	* ext4_punch_hole, etc) which needs to be properly zeroed out. Otherwise a
4055	* racing writeback can come later and flush the stale pagecache to disk.
4056	*/
4057	static int __ext4_block_zero_page_range(handle_t *handle,
4058	struct address_space *mapping, loff_t from, loff_t length)
4059	{
4060	unsigned int offset, blocksize, pos;
4061	ext4_lblk_t iblock;
4062	struct inode *inode = mapping->host;
4063	struct buffer_head *bh;
4064	struct folio *folio;
4065	int err = 0;
4066
4067	folio = __filemap_get_folio(mapping, index: from >> PAGE_SHIFT,
4068	FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
4069	gfp: mapping_gfp_constraint(mapping, gfp_mask: ~__GFP_FS));
4070	if (IS_ERR(ptr: folio))
4071	return PTR_ERR(ptr: folio);
4072
4073	blocksize = inode->i_sb->s_blocksize;
4074
4075	iblock = EXT4_PG_TO_LBLK(inode, folio->index);
4076
4077	bh = folio_buffers(folio);
4078	if (!bh)
4079	bh = create_empty_buffers(folio, blocksize, b_state: 0);
4080
4081	/* Find the buffer that contains "offset" */
4082	offset = offset_in_folio(folio, from);
4083	pos = blocksize;
4084	while (offset >= pos) {
4085	bh = bh->b_this_page;
4086	iblock++;
4087	pos += blocksize;
4088	}
4089	if (buffer_freed(bh)) {
4090	BUFFER_TRACE(bh, "freed: skip");
4091	goto unlock;
4092	}
4093	if (!buffer_mapped(bh)) {
4094	BUFFER_TRACE(bh, "unmapped");
4095	ext4_get_block(inode, iblock, bh, create: 0);
4096	/* unmapped? It's a hole - nothing to do */
4097	if (!buffer_mapped(bh)) {
4098	BUFFER_TRACE(bh, "still unmapped");
4099	goto unlock;
4100	}
4101	}
4102
4103	/* Ok, it's mapped. Make sure it's up-to-date */
4104	if (folio_test_uptodate(folio))
4105	set_buffer_uptodate(bh);
4106
4107	if (!buffer_uptodate(bh)) {
4108	err = ext4_read_bh_lock(bh, op_flags: 0, wait: true);
4109	if (err)
4110	goto unlock;
4111	if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
4112	/* We expect the key to be set. */
4113	BUG_ON(!fscrypt_has_encryption_key(inode));
4114	err = fscrypt_decrypt_pagecache_blocks(folio,
4115	len: blocksize,
4116	offs: bh_offset(bh));
4117	if (err) {
4118	clear_buffer_uptodate(bh);
4119	goto unlock;
4120	}
4121	}
4122	}
4123	if (ext4_should_journal_data(inode)) {
4124	BUFFER_TRACE(bh, "get write access");
4125	err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
4126	EXT4_JTR_NONE);
4127	if (err)
4128	goto unlock;
4129	}
4130	folio_zero_range(folio, start: offset, length);
4131	BUFFER_TRACE(bh, "zeroed end of block");
4132
4133	if (ext4_should_journal_data(inode)) {
4134	err = ext4_dirty_journalled_data(handle, bh);
4135	} else {
4136	err = 0;
4137	mark_buffer_dirty(bh);
4138	if (ext4_should_order_data(inode))
4139	err = ext4_jbd2_inode_add_write(handle, inode, start_byte: from,
4140	length);
4141	}
4142
4143	unlock:
4144	folio_unlock(folio);
4145	folio_put(folio);
4146	return err;
4147	}
4148
4149	/*
4150	* ext4_block_zero_page_range() zeros out a mapping of length 'length'
4151	* starting from file offset 'from'. The range to be zero'd must
4152	* be contained with in one block. If the specified range exceeds
4153	* the end of the block it will be shortened to end of the block
4154	* that corresponds to 'from'
4155	*/
4156	static int ext4_block_zero_page_range(handle_t *handle,
4157	struct address_space *mapping, loff_t from, loff_t length)
4158	{
4159	struct inode *inode = mapping->host;
4160	unsigned blocksize = inode->i_sb->s_blocksize;
4161	unsigned int max = blocksize - (from & (blocksize - 1));
4162
4163	/*
4164	* correct length if it does not fall between
4165	* 'from' and the end of the block
4166	*/
4167	if (length > max || length < 0)
4168	length = max;
4169
4170	if (IS_DAX(inode)) {
4171	return dax_zero_range(inode, pos: from, len: length, NULL,
4172	ops: &ext4_iomap_ops);
4173	}
4174	return __ext4_block_zero_page_range(handle, mapping, from, length);
4175	}
4176
4177	/*
4178	* ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4179	* up to the end of the block which corresponds to `from'.
4180	* This required during truncate. We need to physically zero the tail end
4181	* of that block so it doesn't yield old data if the file is later grown.
4182	*/
4183	static int ext4_block_truncate_page(handle_t *handle,
4184	struct address_space *mapping, loff_t from)
4185	{
4186	unsigned length;
4187	unsigned blocksize;
4188	struct inode *inode = mapping->host;
4189
4190	/* If we are processing an encrypted inode during orphan list handling */
4191	if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
4192	return 0;
4193
4194	blocksize = i_blocksize(node: inode);
4195	length = blocksize - (from & (blocksize - 1));
4196
4197	return ext4_block_zero_page_range(handle, mapping, from, length);
4198	}
4199
4200	int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4201	loff_t lstart, loff_t length)
4202	{
4203	struct super_block *sb = inode->i_sb;
4204	struct address_space *mapping = inode->i_mapping;
4205	unsigned partial_start, partial_end;
4206	ext4_fsblk_t start, end;
4207	loff_t byte_end = (lstart + length - 1);
4208	int err = 0;
4209
4210	partial_start = lstart & (sb->s_blocksize - 1);
4211	partial_end = byte_end & (sb->s_blocksize - 1);
4212
4213	start = lstart >> sb->s_blocksize_bits;
4214	end = byte_end >> sb->s_blocksize_bits;
4215
4216	/* Handle partial zero within the single block */
4217	if (start == end &&
4218	(partial_start || (partial_end != sb->s_blocksize - 1))) {
4219	err = ext4_block_zero_page_range(handle, mapping,
4220	from: lstart, length);
4221	return err;
4222	}
4223	/* Handle partial zero out on the start of the range */
4224	if (partial_start) {
4225	err = ext4_block_zero_page_range(handle, mapping,
4226	from: lstart, length: sb->s_blocksize);
4227	if (err)
4228	return err;
4229	}
4230	/* Handle partial zero out on the end of the range */
4231	if (partial_end != sb->s_blocksize - 1)
4232	err = ext4_block_zero_page_range(handle, mapping,
4233	from: byte_end - partial_end,
4234	length: partial_end + 1);
4235	return err;
4236	}
4237
4238	int ext4_can_truncate(struct inode *inode)
4239	{
4240	if (S_ISREG(inode->i_mode))
4241	return 1;
4242	if (S_ISDIR(inode->i_mode))
4243	return 1;
4244	if (S_ISLNK(inode->i_mode))
4245	return !ext4_inode_is_fast_symlink(inode);
4246	return 0;
4247	}
4248
4249	/*
4250	* We have to make sure i_disksize gets properly updated before we truncate
4251	* page cache due to hole punching or zero range. Otherwise i_disksize update
4252	* can get lost as it may have been postponed to submission of writeback but
4253	* that will never happen if we remove the folio containing i_size from the
4254	* page cache. Also if we punch hole within i_size but above i_disksize,
4255	* following ext4_page_mkwrite() may mistakenly allocate written blocks over
4256	* the hole and thus introduce allocated blocks beyond i_disksize which is
4257	* not allowed (e2fsck would complain in case of crash).
4258	*/
4259	int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4260	loff_t len)
4261	{
4262	handle_t *handle;
4263	int ret;
4264
4265	loff_t size = i_size_read(inode);
4266
4267	WARN_ON(!inode_is_locked(inode));
4268	if (offset > size)
4269	return 0;
4270
4271	if (offset + len < size)
4272	size = offset + len;
4273	if (EXT4_I(inode)->i_disksize >= size)
4274	return 0;
4275
4276	handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4277	if (IS_ERR(ptr: handle))
4278	return PTR_ERR(ptr: handle);
4279	ext4_update_i_disksize(inode, newsize: size);
4280	ret = ext4_mark_inode_dirty(handle, inode);
4281	ext4_journal_stop(handle);
4282
4283	return ret;
4284	}
4285
4286	static inline void ext4_truncate_folio(struct inode *inode,
4287	loff_t start, loff_t end)
4288	{
4289	unsigned long blocksize = i_blocksize(node: inode);
4290	struct folio *folio;
4291
4292	/* Nothing to be done if no complete block needs to be truncated. */
4293	if (round_up(start, blocksize) >= round_down(end, blocksize))
4294	return;
4295
4296	folio = filemap_lock_folio(mapping: inode->i_mapping, index: start >> PAGE_SHIFT);
4297	if (IS_ERR(ptr: folio))
4298	return;
4299
4300	if (folio_mkclean(folio))
4301	folio_mark_dirty(folio);
4302	folio_unlock(folio);
4303	folio_put(folio);
4304	}
4305
4306	int ext4_truncate_page_cache_block_range(struct inode *inode,
4307	loff_t start, loff_t end)
4308	{
4309	unsigned long blocksize = i_blocksize(node: inode);
4310	int ret;
4311
4312	/*
4313	* For journalled data we need to write (and checkpoint) pages
4314	* before discarding page cache to avoid inconsitent data on disk
4315	* in case of crash before freeing or unwritten converting trans
4316	* is committed.
4317	*/
4318	if (ext4_should_journal_data(inode)) {
4319	ret = filemap_write_and_wait_range(mapping: inode->i_mapping, lstart: start,
4320	lend: end - 1);
4321	if (ret)
4322	return ret;
4323	goto truncate_pagecache;
4324	}
4325
4326	/*
4327	* If the block size is less than the page size, the file's mapped
4328	* blocks within one page could be freed or converted to unwritten.
4329	* So it's necessary to remove writable userspace mappings, and then
4330	* ext4_page_mkwrite() can be called during subsequent write access
4331	* to these partial folios.
4332	*/
4333	if (!IS_ALIGNED(start | end, PAGE_SIZE) &&
4334	blocksize < PAGE_SIZE && start < inode->i_size) {
4335	loff_t page_boundary = round_up(start, PAGE_SIZE);
4336
4337	ext4_truncate_folio(inode, start, min(page_boundary, end));
4338	if (end > page_boundary)
4339	ext4_truncate_folio(inode,
4340	round_down(end, PAGE_SIZE), end);
4341	}
4342
4343	truncate_pagecache:
4344	truncate_pagecache_range(inode, offset: start, end: end - 1);
4345	return 0;
4346	}
4347
4348	static void ext4_wait_dax_page(struct inode *inode)
4349	{
4350	filemap_invalidate_unlock(mapping: inode->i_mapping);
4351	schedule();
4352	filemap_invalidate_lock(mapping: inode->i_mapping);
4353	}
4354
4355	int ext4_break_layouts(struct inode *inode)
4356	{
4357	if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
4358	return -EINVAL;
4359
4360	return dax_break_layout_inode(inode, cb: ext4_wait_dax_page);
4361	}
4362
4363	/*
4364	* ext4_punch_hole: punches a hole in a file by releasing the blocks
4365	* associated with the given offset and length
4366	*
4367	* @inode: File inode
4368	* @offset: The offset where the hole will begin
4369	* @len: The length of the hole
4370	*
4371	* Returns: 0 on success or negative on failure
4372	*/
4373
4374	int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
4375	{
4376	struct inode *inode = file_inode(f: file);
4377	struct super_block *sb = inode->i_sb;
4378	ext4_lblk_t start_lblk, end_lblk;
4379	loff_t max_end = sb->s_maxbytes;
4380	loff_t end = offset + length;
4381	handle_t *handle;
4382	unsigned int credits;
4383	int ret;
4384
4385	trace_ext4_punch_hole(inode, offset, len: length, mode: 0);
4386	WARN_ON_ONCE(!inode_is_locked(inode));
4387
4388	/*
4389	* For indirect-block based inodes, make sure that the hole within
4390	* one block before last range.
4391	*/
4392	if (!ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
4393	max_end = EXT4_SB(sb)->s_bitmap_maxbytes - sb->s_blocksize;
4394
4395	/* No need to punch hole beyond i_size */
4396	if (offset >= inode->i_size || offset >= max_end)
4397	return 0;
4398
4399	/*
4400	* If the hole extends beyond i_size, set the hole to end after
4401	* the block that contains i_size to save pointless tail block zeroing.
4402	*/
4403	if (end >= inode->i_size)
4404	end = round_up(inode->i_size, sb->s_blocksize);
4405	if (end > max_end)
4406	end = max_end;
4407	length = end - offset;
4408
4409	/*
4410	* Attach jinode to inode for jbd2 if we do any zeroing of partial
4411	* block.
4412	*/
4413	if (!IS_ALIGNED(offset | end, sb->s_blocksize)) {
4414	ret = ext4_inode_attach_jinode(inode);
4415	if (ret < 0)
4416	return ret;
4417	}
4418
4419
4420	ret = ext4_update_disksize_before_punch(inode, offset, len: length);
4421	if (ret)
4422	return ret;
4423
4424	/* Now release the pages and zero block aligned part of pages*/
4425	ret = ext4_truncate_page_cache_block_range(inode, start: offset, end);
4426	if (ret)
4427	return ret;
4428
4429	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
4430	credits = ext4_chunk_trans_extent(inode, nrblocks: 2);
4431	else
4432	credits = ext4_blocks_for_truncate(inode);
4433	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4434	if (IS_ERR(ptr: handle)) {
4435	ret = PTR_ERR(ptr: handle);
4436	ext4_std_error(sb, ret);
4437	return ret;
4438	}
4439
4440	ret = ext4_zero_partial_blocks(handle, inode, lstart: offset, length);
4441	if (ret)
4442	goto out_handle;
4443
4444	/* If there are blocks to remove, do it */
4445	start_lblk = EXT4_B_TO_LBLK(inode, offset);
4446	end_lblk = end >> inode->i_blkbits;
4447
4448	if (end_lblk > start_lblk) {
4449	ext4_lblk_t hole_len = end_lblk - start_lblk;
4450
4451	ext4_fc_track_inode(handle, inode);
4452	ext4_check_map_extents_env(inode);
4453	down_write(sem: &EXT4_I(inode)->i_data_sem);
4454	ext4_discard_preallocations(inode);
4455
4456	ext4_es_remove_extent(inode, lblk: start_lblk, len: hole_len);
4457
4458	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
4459	ret = ext4_ext_remove_space(inode, start: start_lblk,
4460	end: end_lblk - 1);
4461	else
4462	ret = ext4_ind_remove_space(handle, inode, start: start_lblk,
4463	end: end_lblk);
4464	if (ret) {
4465	up_write(sem: &EXT4_I(inode)->i_data_sem);
4466	goto out_handle;
4467	}
4468
4469	ext4_es_insert_extent(inode, lblk: start_lblk, len: hole_len, pblk: ~0,
4470	EXTENT_STATUS_HOLE, delalloc_reserve_used: 0);
4471	up_write(sem: &EXT4_I(inode)->i_data_sem);
4472	}
4473	ext4_fc_track_range(handle, inode, start: start_lblk, end: end_lblk);
4474
4475	ret = ext4_mark_inode_dirty(handle, inode);
4476	if (unlikely(ret))
4477	goto out_handle;
4478
4479	ext4_update_inode_fsync_trans(handle, inode, datasync: 1);
4480	if (IS_SYNC(inode))
4481	ext4_handle_sync(handle);
4482	out_handle:
4483	ext4_journal_stop(handle);
4484	return ret;
4485	}
4486
4487	int ext4_inode_attach_jinode(struct inode *inode)
4488	{
4489	struct ext4_inode_info *ei = EXT4_I(inode);
4490	struct jbd2_inode *jinode;
4491
4492	if (ei->jinode || !EXT4_SB(sb: inode->i_sb)->s_journal)
4493	return 0;
4494
4495	jinode = jbd2_alloc_inode(GFP_KERNEL);
4496	spin_lock(lock: &inode->i_lock);
4497	if (!ei->jinode) {
4498	if (!jinode) {
4499	spin_unlock(lock: &inode->i_lock);
4500	return -ENOMEM;
4501	}
4502	ei->jinode = jinode;
4503	jbd2_journal_init_jbd_inode(jinode: ei->jinode, inode);
4504	jinode = NULL;
4505	}
4506	spin_unlock(lock: &inode->i_lock);
4507	if (unlikely(jinode != NULL))
4508	jbd2_free_inode(jinode);
4509	return 0;
4510	}
4511
4512	/*
4513	* ext4_truncate()
4514	*
4515	* We block out ext4_get_block() block instantiations across the entire
4516	* transaction, and VFS/VM ensures that ext4_truncate() cannot run
4517	* simultaneously on behalf of the same inode.
4518	*
4519	* As we work through the truncate and commit bits of it to the journal there
4520	* is one core, guiding principle: the file's tree must always be consistent on
4521	* disk. We must be able to restart the truncate after a crash.
4522	*
4523	* The file's tree may be transiently inconsistent in memory (although it
4524	* probably isn't), but whenever we close off and commit a journal transaction,
4525	* the contents of (the filesystem + the journal) must be consistent and
4526	* restartable. It's pretty simple, really: bottom up, right to left (although
4527	* left-to-right works OK too).
4528	*
4529	* Note that at recovery time, journal replay occurs *before* the restart of
4530	* truncate against the orphan inode list.
4531	*
4532	* The committed inode has the new, desired i_size (which is the same as
4533	* i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4534	* that this inode's truncate did not complete and it will again call
4535	* ext4_truncate() to have another go. So there will be instantiated blocks
4536	* to the right of the truncation point in a crashed ext4 filesystem. But
4537	* that's fine - as long as they are linked from the inode, the post-crash
4538	* ext4_truncate() run will find them and release them.
4539	*/
4540	int ext4_truncate(struct inode *inode)
4541	{
4542	struct ext4_inode_info *ei = EXT4_I(inode);
4543	unsigned int credits;
4544	int err = 0, err2;
4545	handle_t *handle;
4546	struct address_space *mapping = inode->i_mapping;
4547
4548	/*
4549	* There is a possibility that we're either freeing the inode
4550	* or it's a completely new inode. In those cases we might not
4551	* have i_rwsem locked because it's not necessary.
4552	*/
4553	if (!(inode_state_read_once(inode) & (I_NEW | I_FREEING)))
4554	WARN_ON(!inode_is_locked(inode));
4555	trace_ext4_truncate_enter(inode);
4556
4557	if (!ext4_can_truncate(inode))
4558	goto out_trace;
4559
4560	if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4561	ext4_set_inode_state(inode, bit: EXT4_STATE_DA_ALLOC_CLOSE);
4562
4563	if (ext4_has_inline_data(inode)) {
4564	int has_inline = 1;
4565
4566	err = ext4_inline_data_truncate(inode, has_inline: &has_inline);
4567	if (err || has_inline)
4568	goto out_trace;
4569	}
4570
4571	/* If we zero-out tail of the page, we have to create jinode for jbd2 */
4572	if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4573	err = ext4_inode_attach_jinode(inode);
4574	if (err)
4575	goto out_trace;
4576	}
4577
4578	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
4579	credits = ext4_chunk_trans_extent(inode, nrblocks: 1);
4580	else
4581	credits = ext4_blocks_for_truncate(inode);
4582
4583	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4584	if (IS_ERR(ptr: handle)) {
4585	err = PTR_ERR(ptr: handle);
4586	goto out_trace;
4587	}
4588
4589	if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4590	ext4_block_truncate_page(handle, mapping, from: inode->i_size);
4591
4592	/*
4593	* We add the inode to the orphan list, so that if this
4594	* truncate spans multiple transactions, and we crash, we will
4595	* resume the truncate when the filesystem recovers. It also
4596	* marks the inode dirty, to catch the new size.
4597	*
4598	* Implication: the file must always be in a sane, consistent
4599	* truncatable state while each transaction commits.
4600	*/
4601	err = ext4_orphan_add(handle, inode);
4602	if (err)
4603	goto out_stop;
4604
4605	ext4_fc_track_inode(handle, inode);
4606	ext4_check_map_extents_env(inode);
4607
4608	down_write(sem: &EXT4_I(inode)->i_data_sem);
4609	ext4_discard_preallocations(inode);
4610
4611	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
4612	err = ext4_ext_truncate(handle, inode);
4613	else
4614	ext4_ind_truncate(handle, inode);
4615
4616	up_write(sem: &ei->i_data_sem);
4617	if (err)
4618	goto out_stop;
4619
4620	if (IS_SYNC(inode))
4621	ext4_handle_sync(handle);
4622
4623	out_stop:
4624	/*
4625	* If this was a simple ftruncate() and the file will remain alive,
4626	* then we need to clear up the orphan record which we created above.
4627	* However, if this was a real unlink then we were called by
4628	* ext4_evict_inode(), and we allow that function to clean up the
4629	* orphan info for us.
4630	*/
4631	if (inode->i_nlink)
4632	ext4_orphan_del(handle, inode);
4633
4634	inode_set_mtime_to_ts(inode, ts: inode_set_ctime_current(inode));
4635	err2 = ext4_mark_inode_dirty(handle, inode);
4636	if (unlikely(err2 && !err))
4637	err = err2;
4638	ext4_journal_stop(handle);
4639
4640	out_trace:
4641	trace_ext4_truncate_exit(inode);
4642	return err;
4643	}
4644
4645	static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4646	{
4647	if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4648	return inode_peek_iversion_raw(inode);
4649	else
4650	return inode_peek_iversion(inode);
4651	}
4652
4653	static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4654	struct ext4_inode_info *ei)
4655	{
4656	struct inode *inode = &(ei->vfs_inode);
4657	u64 i_blocks = READ_ONCE(inode->i_blocks);
4658	struct super_block *sb = inode->i_sb;
4659
4660	if (i_blocks <= ~0U) {
4661	/*
4662	* i_blocks can be represented in a 32 bit variable
4663	* as multiple of 512 bytes
4664	*/
4665	raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4666	raw_inode->i_blocks_high = 0;
4667	ext4_clear_inode_flag(inode, bit: EXT4_INODE_HUGE_FILE);
4668	return 0;
4669	}
4670
4671	/*
4672	* This should never happen since sb->s_maxbytes should not have
4673	* allowed this, sb->s_maxbytes was set according to the huge_file
4674	* feature in ext4_fill_super().
4675	*/
4676	if (!ext4_has_feature_huge_file(sb))
4677	return -EFSCORRUPTED;
4678
4679	if (i_blocks <= 0xffffffffffffULL) {
4680	/*
4681	* i_blocks can be represented in a 48 bit variable
4682	* as multiple of 512 bytes
4683	*/
4684	raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4685	raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4686	ext4_clear_inode_flag(inode, bit: EXT4_INODE_HUGE_FILE);
4687	} else {
4688	ext4_set_inode_flag(inode, bit: EXT4_INODE_HUGE_FILE);
4689	/* i_block is stored in file system block size */
4690	i_blocks = i_blocks >> (inode->i_blkbits - 9);
4691	raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4692	raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4693	}
4694	return 0;
4695	}
4696
4697	static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4698	{
4699	struct ext4_inode_info *ei = EXT4_I(inode);
4700	uid_t i_uid;
4701	gid_t i_gid;
4702	projid_t i_projid;
4703	int block;
4704	int err;
4705
4706	err = ext4_inode_blocks_set(raw_inode, ei);
4707
4708	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4709	i_uid = i_uid_read(inode);
4710	i_gid = i_gid_read(inode);
4711	i_projid = from_kprojid(to: &init_user_ns, projid: ei->i_projid);
4712	if (!(test_opt(inode->i_sb, NO_UID32))) {
4713	raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4714	raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4715	/*
4716	* Fix up interoperability with old kernels. Otherwise,
4717	* old inodes get re-used with the upper 16 bits of the
4718	* uid/gid intact.
4719	*/
4720	if (ei->i_dtime && !ext4_inode_orphan_tracked(inode)) {
4721	raw_inode->i_uid_high = 0;
4722	raw_inode->i_gid_high = 0;
4723	} else {
4724	raw_inode->i_uid_high =
4725	cpu_to_le16(high_16_bits(i_uid));
4726	raw_inode->i_gid_high =
4727	cpu_to_le16(high_16_bits(i_gid));
4728	}
4729	} else {
4730	raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4731	raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4732	raw_inode->i_uid_high = 0;
4733	raw_inode->i_gid_high = 0;
4734	}
4735	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4736
4737	EXT4_INODE_SET_CTIME(inode, raw_inode);
4738	EXT4_INODE_SET_MTIME(inode, raw_inode);
4739	EXT4_INODE_SET_ATIME(inode, raw_inode);
4740	EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4741
4742	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4743	raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4744	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4745	raw_inode->i_file_acl_high =
4746	cpu_to_le16(ei->i_file_acl >> 32);
4747	raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4748	ext4_isize_set(raw_inode, i_size: ei->i_disksize);
4749
4750	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4751	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4752	if (old_valid_dev(dev: inode->i_rdev)) {
4753	raw_inode->i_block[0] =
4754	cpu_to_le32(old_encode_dev(inode->i_rdev));
4755	raw_inode->i_block[1] = 0;
4756	} else {
4757	raw_inode->i_block[0] = 0;
4758	raw_inode->i_block[1] =
4759	cpu_to_le32(new_encode_dev(inode->i_rdev));
4760	raw_inode->i_block[2] = 0;
4761	}
4762	} else if (!ext4_has_inline_data(inode)) {
4763	for (block = 0; block < EXT4_N_BLOCKS; block++)
4764	raw_inode->i_block[block] = ei->i_data[block];
4765	}
4766
4767	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4768	u64 ivers = ext4_inode_peek_iversion(inode);
4769
4770	raw_inode->i_disk_version = cpu_to_le32(ivers);
4771	if (ei->i_extra_isize) {
4772	if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4773	raw_inode->i_version_hi =
4774	cpu_to_le32(ivers >> 32);
4775	raw_inode->i_extra_isize =
4776	cpu_to_le16(ei->i_extra_isize);
4777	}
4778	}
4779
4780	if (i_projid != EXT4_DEF_PROJID &&
4781	!ext4_has_feature_project(sb: inode->i_sb))
4782	err = err ?: -EFSCORRUPTED;
4783
4784	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4785	EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4786	raw_inode->i_projid = cpu_to_le32(i_projid);
4787
4788	ext4_inode_csum_set(inode, raw: raw_inode, ei);
4789	return err;
4790	}
4791
4792	/*
4793	* ext4_get_inode_loc returns with an extra refcount against the inode's
4794	* underlying buffer_head on success. If we pass 'inode' and it does not
4795	* have in-inode xattr, we have all inode data in memory that is needed
4796	* to recreate the on-disk version of this inode.
4797	*/
4798	static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4799	struct inode *inode, struct ext4_iloc *iloc,
4800	ext4_fsblk_t *ret_block)
4801	{
4802	struct ext4_group_desc *gdp;
4803	struct buffer_head *bh;
4804	ext4_fsblk_t block;
4805	struct blk_plug plug;
4806	int inodes_per_block, inode_offset;
4807
4808	iloc->bh = NULL;
4809	if (ino < EXT4_ROOT_INO ||
4810	ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4811	return -EFSCORRUPTED;
4812
4813	iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4814	gdp = ext4_get_group_desc(sb, block_group: iloc->block_group, NULL);
4815	if (!gdp)
4816	return -EIO;
4817
4818	/*
4819	* Figure out the offset within the block group inode table
4820	*/
4821	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4822	inode_offset = ((ino - 1) %
4823	EXT4_INODES_PER_GROUP(sb));
4824	iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4825
4826	block = ext4_inode_table(sb, bg: gdp);
4827	if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4828	(block >= ext4_blocks_count(es: EXT4_SB(sb)->s_es))) {
4829	ext4_error(sb, "Invalid inode table block %llu in "
4830	"block_group %u", block, iloc->block_group);
4831	return -EFSCORRUPTED;
4832	}
4833	block += (inode_offset / inodes_per_block);
4834
4835	bh = sb_getblk(sb, block);
4836	if (unlikely(!bh))
4837	return -ENOMEM;
4838	if (ext4_buffer_uptodate(bh))
4839	goto has_buffer;
4840
4841	lock_buffer(bh);
4842	if (ext4_buffer_uptodate(bh)) {
4843	/* Someone brought it uptodate while we waited */
4844	unlock_buffer(bh);
4845	goto has_buffer;
4846	}
4847
4848	/*
4849	* If we have all information of the inode in memory and this
4850	* is the only valid inode in the block, we need not read the
4851	* block.
4852	*/
4853	if (inode && !ext4_test_inode_state(inode, bit: EXT4_STATE_XATTR)) {
4854	struct buffer_head *bitmap_bh;
4855	int i, start;
4856
4857	start = inode_offset & ~(inodes_per_block - 1);
4858
4859	/* Is the inode bitmap in cache? */
4860	bitmap_bh = sb_getblk(sb, block: ext4_inode_bitmap(sb, bg: gdp));
4861	if (unlikely(!bitmap_bh))
4862	goto make_io;
4863
4864	/*
4865	* If the inode bitmap isn't in cache then the
4866	* optimisation may end up performing two reads instead
4867	* of one, so skip it.
4868	*/
4869	if (!buffer_uptodate(bh: bitmap_bh)) {
4870	brelse(bh: bitmap_bh);
4871	goto make_io;
4872	}
4873	for (i = start; i < start + inodes_per_block; i++) {
4874	if (i == inode_offset)
4875	continue;
4876	if (ext4_test_bit(nr: i, addr: bitmap_bh->b_data))
4877	break;
4878	}
4879	brelse(bh: bitmap_bh);
4880	if (i == start + inodes_per_block) {
4881	struct ext4_inode *raw_inode =
4882	(struct ext4_inode *) (bh->b_data + iloc->offset);
4883
4884	/* all other inodes are free, so skip I/O */
4885	memset(bh->b_data, 0, bh->b_size);
4886	if (!ext4_test_inode_state(inode, bit: EXT4_STATE_NEW))
4887	ext4_fill_raw_inode(inode, raw_inode);
4888	set_buffer_uptodate(bh);
4889	unlock_buffer(bh);
4890	goto has_buffer;
4891	}
4892	}
4893
4894	make_io:
4895	/*
4896	* If we need to do any I/O, try to pre-readahead extra
4897	* blocks from the inode table.
4898	*/
4899	blk_start_plug(&plug);
4900	if (EXT4_SB(sb)->s_inode_readahead_blks) {
4901	ext4_fsblk_t b, end, table;
4902	unsigned num;
4903	__u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4904
4905	table = ext4_inode_table(sb, bg: gdp);
4906	/* s_inode_readahead_blks is always a power of 2 */
4907	b = block & ~((ext4_fsblk_t) ra_blks - 1);
4908	if (table > b)
4909	b = table;
4910	end = b + ra_blks;
4911	num = EXT4_INODES_PER_GROUP(sb);
4912	if (ext4_has_group_desc_csum(sb))
4913	num -= ext4_itable_unused_count(sb, bg: gdp);
4914	table += num / inodes_per_block;
4915	if (end > table)
4916	end = table;
4917	while (b <= end)
4918	ext4_sb_breadahead_unmovable(sb, block: b++);
4919	}
4920
4921	/*
4922	* There are other valid inodes in the buffer, this inode
4923	* has in-inode xattrs, or we don't have this inode in memory.
4924	* Read the block from disk.
4925	*/
4926	trace_ext4_load_inode(sb, ino);
4927	ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL,
4928	simu_fail: ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO));
4929	blk_finish_plug(&plug);
4930	wait_on_buffer(bh);
4931	if (!buffer_uptodate(bh)) {
4932	if (ret_block)
4933	*ret_block = block;
4934	brelse(bh);
4935	return -EIO;
4936	}
4937	has_buffer:
4938	iloc->bh = bh;
4939	return 0;
4940	}
4941
4942	static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4943	struct ext4_iloc *iloc)
4944	{
4945	ext4_fsblk_t err_blk = 0;
4946	int ret;
4947
4948	ret = __ext4_get_inode_loc(sb: inode->i_sb, ino: inode->i_ino, NULL, iloc,
4949	ret_block: &err_blk);
4950
4951	if (ret == -EIO)
4952	ext4_error_inode_block(inode, err_blk, EIO,
4953	"unable to read itable block");
4954
4955	return ret;
4956	}
4957
4958	int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4959	{
4960	ext4_fsblk_t err_blk = 0;
4961	int ret;
4962
4963	ret = __ext4_get_inode_loc(sb: inode->i_sb, ino: inode->i_ino, inode, iloc,
4964	ret_block: &err_blk);
4965
4966	if (ret == -EIO)
4967	ext4_error_inode_block(inode, err_blk, EIO,
4968	"unable to read itable block");
4969
4970	return ret;
4971	}
4972
4973
4974	int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4975	struct ext4_iloc *iloc)
4976	{
4977	return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4978	}
4979
4980	static bool ext4_should_enable_dax(struct inode *inode)
4981	{
4982	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
4983
4984	if (test_opt2(inode->i_sb, DAX_NEVER))
4985	return false;
4986	if (!S_ISREG(inode->i_mode))
4987	return false;
4988	if (ext4_should_journal_data(inode))
4989	return false;
4990	if (ext4_has_inline_data(inode))
4991	return false;
4992	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_ENCRYPT))
4993	return false;
4994	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_VERITY))
4995	return false;
4996	if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4997	return false;
4998	if (test_opt(inode->i_sb, DAX_ALWAYS))
4999	return true;
5000
5001	return ext4_test_inode_flag(inode, bit: EXT4_INODE_DAX);
5002	}
5003
5004	void ext4_set_inode_flags(struct inode *inode, bool init)
5005	{
5006	unsigned int flags = EXT4_I(inode)->i_flags;
5007	unsigned int new_fl = 0;
5008
5009	WARN_ON_ONCE(IS_DAX(inode) && init);
5010
5011	if (flags & EXT4_SYNC_FL)
5012	new_fl |= S_SYNC;
5013	if (flags & EXT4_APPEND_FL)
5014	new_fl |= S_APPEND;
5015	if (flags & EXT4_IMMUTABLE_FL)
5016	new_fl |= S_IMMUTABLE;
5017	if (flags & EXT4_NOATIME_FL)
5018	new_fl |= S_NOATIME;
5019	if (flags & EXT4_DIRSYNC_FL)
5020	new_fl |= S_DIRSYNC;
5021
5022	/* Because of the way inode_set_flags() works we must preserve S_DAX
5023	* here if already set. */
5024	new_fl |= (inode->i_flags & S_DAX);
5025	if (init && ext4_should_enable_dax(inode))
5026	new_fl |= S_DAX;
5027
5028	if (flags & EXT4_ENCRYPT_FL)
5029	new_fl |= S_ENCRYPTED;
5030	if (flags & EXT4_CASEFOLD_FL)
5031	new_fl |= S_CASEFOLD;
5032	if (flags & EXT4_VERITY_FL)
5033	new_fl |= S_VERITY;
5034	inode_set_flags(inode, flags: new_fl,
5035	S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
5036	S_ENCRYPTED|S_CASEFOLD|S_VERITY);
5037	}
5038
5039	static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
5040	struct ext4_inode_info *ei)
5041	{
5042	blkcnt_t i_blocks ;
5043	struct inode *inode = &(ei->vfs_inode);
5044	struct super_block *sb = inode->i_sb;
5045
5046	if (ext4_has_feature_huge_file(sb)) {
5047	/* we are using combined 48 bit field */
5048	i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
5049	le32_to_cpu(raw_inode->i_blocks_lo);
5050	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_HUGE_FILE)) {
5051	/* i_blocks represent file system block size */
5052	return i_blocks << (inode->i_blkbits - 9);
5053	} else {
5054	return i_blocks;
5055	}
5056	} else {
5057	return le32_to_cpu(raw_inode->i_blocks_lo);
5058	}
5059	}
5060
5061	static inline int ext4_iget_extra_inode(struct inode *inode,
5062	struct ext4_inode *raw_inode,
5063	struct ext4_inode_info *ei)
5064	{
5065	__le32 *magic = (void *)raw_inode +
5066	EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
5067
5068	if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
5069	*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
5070	int err;
5071
5072	err = xattr_check_inode(inode, IHDR(inode, raw_inode),
5073	ITAIL(inode, raw_inode));
5074	if (err)
5075	return err;
5076
5077	ext4_set_inode_state(inode, bit: EXT4_STATE_XATTR);
5078	err = ext4_find_inline_data_nolock(inode);
5079	if (!err && ext4_has_inline_data(inode))
5080	ext4_set_inode_state(inode, bit: EXT4_STATE_MAY_INLINE_DATA);
5081	return err;
5082	} else
5083	EXT4_I(inode)->i_inline_off = 0;
5084	return 0;
5085	}
5086
5087	int ext4_get_projid(struct inode *inode, kprojid_t *projid)
5088	{
5089	if (!ext4_has_feature_project(sb: inode->i_sb))
5090	return -EOPNOTSUPP;
5091	*projid = EXT4_I(inode)->i_projid;
5092	return 0;
5093	}
5094
5095	/*
5096	* ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
5097	* refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
5098	* set.
5099	*/
5100	static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
5101	{
5102	if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
5103	inode_set_iversion_raw(inode, val);
5104	else
5105	inode_set_iversion_queried(inode, val);
5106	}
5107
5108	static int check_igot_inode(struct inode *inode, ext4_iget_flags flags,
5109	const char *function, unsigned int line)
5110	{
5111	const char *err_str;
5112
5113	if (flags & EXT4_IGET_EA_INODE) {
5114	if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
5115	err_str = "missing EA_INODE flag";
5116	goto error;
5117	}
5118	if (ext4_test_inode_state(inode, bit: EXT4_STATE_XATTR) ||
5119	EXT4_I(inode)->i_file_acl) {
5120	err_str = "ea_inode with extended attributes";
5121	goto error;
5122	}
5123	} else {
5124	if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
5125	/*
5126	* open_by_handle_at() could provide an old inode number
5127	* that has since been reused for an ea_inode; this does
5128	* not indicate filesystem corruption
5129	*/
5130	if (flags & EXT4_IGET_HANDLE)
5131	return -ESTALE;
5132	err_str = "unexpected EA_INODE flag";
5133	goto error;
5134	}
5135	}
5136	if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD)) {
5137	err_str = "unexpected bad inode w/o EXT4_IGET_BAD";
5138	goto error;
5139	}
5140	return 0;
5141
5142	error:
5143	ext4_error_inode(inode, function, line, 0, "%s", err_str);
5144	return -EFSCORRUPTED;
5145	}
5146
5147	void ext4_set_inode_mapping_order(struct inode *inode)
5148	{
5149	struct super_block *sb = inode->i_sb;
5150	u16 min_order, max_order;
5151
5152	max_order = EXT4_SB(sb)->s_max_folio_order;
5153	if (!max_order)
5154	return;
5155
5156	min_order = EXT4_SB(sb)->s_min_folio_order;
5157	if (!min_order && !S_ISREG(inode->i_mode))
5158	return;
5159
5160	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_JOURNAL_DATA))
5161	max_order = min_order;
5162
5163	mapping_set_folio_order_range(mapping: inode->i_mapping, min: min_order, max: max_order);
5164	}
5165
5166	struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
5167	ext4_iget_flags flags, const char *function,
5168	unsigned int line)
5169	{
5170	struct ext4_iloc iloc;
5171	struct ext4_inode *raw_inode;
5172	struct ext4_inode_info *ei;
5173	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
5174	struct inode *inode;
5175	journal_t *journal = EXT4_SB(sb)->s_journal;
5176	long ret;
5177	loff_t size;
5178	int block;
5179	uid_t i_uid;
5180	gid_t i_gid;
5181	projid_t i_projid;
5182
5183	if ((!(flags & EXT4_IGET_SPECIAL) && is_special_ino(sb, ino)) ||
5184	(ino < EXT4_ROOT_INO) ||
5185	(ino > le32_to_cpu(es->s_inodes_count))) {
5186	if (flags & EXT4_IGET_HANDLE)
5187	return ERR_PTR(error: -ESTALE);
5188	__ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
5189	"inode #%lu: comm %s: iget: illegal inode #",
5190	ino, current->comm);
5191	return ERR_PTR(error: -EFSCORRUPTED);
5192	}
5193
5194	inode = iget_locked(sb, ino);
5195	if (!inode)
5196	return ERR_PTR(error: -ENOMEM);
5197	if (!(inode_state_read_once(inode) & I_NEW)) {
5198	ret = check_igot_inode(inode, flags, function, line);
5199	if (ret) {
5200	iput(inode);
5201	return ERR_PTR(error: ret);
5202	}
5203	return inode;
5204	}
5205
5206	ei = EXT4_I(inode);
5207	iloc.bh = NULL;
5208
5209	ret = __ext4_get_inode_loc_noinmem(inode, iloc: &iloc);
5210	if (ret < 0)
5211	goto bad_inode;
5212	raw_inode = ext4_raw_inode(iloc: &iloc);
5213
5214	if ((flags & EXT4_IGET_HANDLE) &&
5215	(raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
5216	ret = -ESTALE;
5217	goto bad_inode;
5218	}
5219
5220	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5221	ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
5222	if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
5223	EXT4_INODE_SIZE(inode->i_sb) ||
5224	(ei->i_extra_isize & 3)) {
5225	ext4_error_inode(inode, function, line, 0,
5226	"iget: bad extra_isize %u "
5227	"(inode size %u)",
5228	ei->i_extra_isize,
5229	EXT4_INODE_SIZE(inode->i_sb));
5230	ret = -EFSCORRUPTED;
5231	goto bad_inode;
5232	}
5233	} else
5234	ei->i_extra_isize = 0;
5235
5236	/* Precompute checksum seed for inode metadata */
5237	if (ext4_has_feature_metadata_csum(sb)) {
5238	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
5239	__u32 csum;
5240	__le32 inum = cpu_to_le32(inode->i_ino);
5241	__le32 gen = raw_inode->i_generation;
5242	csum = ext4_chksum(crc: sbi->s_csum_seed, address: (__u8 *)&inum,
5243	length: sizeof(inum));
5244	ei->i_csum_seed = ext4_chksum(crc: csum, address: (__u8 *)&gen, length: sizeof(gen));
5245	}
5246
5247	if ((!ext4_inode_csum_verify(inode, raw: raw_inode, ei) ||
5248	ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
5249	(!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
5250	ext4_error_inode_err(inode, function, line, 0,
5251	EFSBADCRC, "iget: checksum invalid");
5252	ret = -EFSBADCRC;
5253	goto bad_inode;
5254	}
5255
5256	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
5257	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
5258	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
5259	if (ext4_has_feature_project(sb) &&
5260	EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5261	EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5262	i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
5263	else
5264	i_projid = EXT4_DEF_PROJID;
5265
5266	if (!(test_opt(inode->i_sb, NO_UID32))) {
5267	i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
5268	i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
5269	}
5270	i_uid_write(inode, uid: i_uid);
5271	i_gid_write(inode, gid: i_gid);
5272	ei->i_projid = make_kprojid(from: &init_user_ns, projid: i_projid);
5273	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
5274
5275	ei->i_inline_off = 0;
5276	ei->i_dir_start_lookup = 0;
5277	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
5278	/* We now have enough fields to check if the inode was active or not.
5279	* This is needed because nfsd might try to access dead inodes
5280	* the test is that same one that e2fsck uses
5281	* NeilBrown 1999oct15
5282	*/
5283	if (inode->i_nlink == 0) {
5284	if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
5285	!(EXT4_SB(sb: inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
5286	ino != EXT4_BOOT_LOADER_INO) {
5287	/* this inode is deleted or unallocated */
5288	if (flags & EXT4_IGET_SPECIAL) {
5289	ext4_error_inode(inode, function, line, 0,
5290	"iget: special inode unallocated");
5291	ret = -EFSCORRUPTED;
5292	} else
5293	ret = -ESTALE;
5294	goto bad_inode;
5295	}
5296	/* The only unlinked inodes we let through here have
5297	* valid i_mode and are being read by the orphan
5298	* recovery code: that's fine, we're about to complete
5299	* the process of deleting those.
5300	* OR it is the EXT4_BOOT_LOADER_INO which is
5301	* not initialized on a new filesystem. */
5302	}
5303	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
5304	ext4_set_inode_flags(inode, init: true);
5305	/* Detect invalid flag combination - can't have both inline data and extents */
5306	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_INLINE_DATA) &&
5307	ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS)) {
5308	ext4_error_inode(inode, function, line, 0,
5309	"inode has both inline data and extents flags");
5310	ret = -EFSCORRUPTED;
5311	goto bad_inode;
5312	}
5313	inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
5314	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
5315	if (ext4_has_feature_64bit(sb))
5316	ei->i_file_acl |=
5317	((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
5318	inode->i_size = ext4_isize(sb, raw_inode);
5319	size = i_size_read(inode);
5320	if (size < 0 || size > ext4_get_maxbytes(inode)) {
5321	ext4_error_inode(inode, function, line, 0,
5322	"iget: bad i_size value: %lld", size);
5323	ret = -EFSCORRUPTED;
5324	goto bad_inode;
5325	}
5326	/*
5327	* If dir_index is not enabled but there's dir with INDEX flag set,
5328	* we'd normally treat htree data as empty space. But with metadata
5329	* checksumming that corrupts checksums so forbid that.
5330	*/
5331	if (!ext4_has_feature_dir_index(sb) &&
5332	ext4_has_feature_metadata_csum(sb) &&
5333	ext4_test_inode_flag(inode, bit: EXT4_INODE_INDEX)) {
5334	ext4_error_inode(inode, function, line, 0,
5335	"iget: Dir with htree data on filesystem without dir_index feature.");
5336	ret = -EFSCORRUPTED;
5337	goto bad_inode;
5338	}
5339	ei->i_disksize = inode->i_size;
5340	#ifdef CONFIG_QUOTA
5341	ei->i_reserved_quota = 0;
5342	#endif
5343	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5344	ei->i_block_group = iloc.block_group;
5345	ei->i_last_alloc_group = ~0;
5346	/*
5347	* NOTE! The in-memory inode i_data array is in little-endian order
5348	* even on big-endian machines: we do NOT byteswap the block numbers!
5349	*/
5350	for (block = 0; block < EXT4_N_BLOCKS; block++)
5351	ei->i_data[block] = raw_inode->i_block[block];
5352	INIT_LIST_HEAD(list: &ei->i_orphan);
5353	ext4_fc_init_inode(inode: &ei->vfs_inode);
5354
5355	/*
5356	* Set transaction id's of transactions that have to be committed
5357	* to finish f[data]sync. We set them to currently running transaction
5358	* as we cannot be sure that the inode or some of its metadata isn't
5359	* part of the transaction - the inode could have been reclaimed and
5360	* now it is reread from disk.
5361	*/
5362	if (journal) {
5363	transaction_t *transaction;
5364	tid_t tid;
5365
5366	read_lock(&journal->j_state_lock);
5367	if (journal->j_running_transaction)
5368	transaction = journal->j_running_transaction;
5369	else
5370	transaction = journal->j_committing_transaction;
5371	if (transaction)
5372	tid = transaction->t_tid;
5373	else
5374	tid = journal->j_commit_sequence;
5375	read_unlock(&journal->j_state_lock);
5376	ei->i_sync_tid = tid;
5377	ei->i_datasync_tid = tid;
5378	}
5379
5380	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5381	if (ei->i_extra_isize == 0) {
5382	/* The extra space is currently unused. Use it. */
5383	BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
5384	ei->i_extra_isize = sizeof(struct ext4_inode) -
5385	EXT4_GOOD_OLD_INODE_SIZE;
5386	} else {
5387	ret = ext4_iget_extra_inode(inode, raw_inode, ei);
5388	if (ret)
5389	goto bad_inode;
5390	}
5391	}
5392
5393	EXT4_INODE_GET_CTIME(inode, raw_inode);
5394	EXT4_INODE_GET_ATIME(inode, raw_inode);
5395	EXT4_INODE_GET_MTIME(inode, raw_inode);
5396	EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5397
5398	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5399	u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5400
5401	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5402	if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5403	ivers |=
5404	(__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5405	}
5406	ext4_inode_set_iversion_queried(inode, val: ivers);
5407	}
5408
5409	ret = 0;
5410	if (ei->i_file_acl &&
5411	!ext4_inode_block_valid(inode, start_blk: ei->i_file_acl, count: 1)) {
5412	ext4_error_inode(inode, function, line, 0,
5413	"iget: bad extended attribute block %llu",
5414	ei->i_file_acl);
5415	ret = -EFSCORRUPTED;
5416	goto bad_inode;
5417	} else if (!ext4_has_inline_data(inode)) {
5418	/* validate the block references in the inode */
5419	if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
5420	(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5421	(S_ISLNK(inode->i_mode) &&
5422	!ext4_inode_is_fast_symlink(inode)))) {
5423	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
5424	ret = ext4_ext_check_inode(inode);
5425	else
5426	ret = ext4_ind_check_inode(inode);
5427	}
5428	}
5429	if (ret)
5430	goto bad_inode;
5431
5432	if (S_ISREG(inode->i_mode)) {
5433	inode->i_op = &ext4_file_inode_operations;
5434	inode->i_fop = &ext4_file_operations;
5435	ext4_set_aops(inode);
5436	} else if (S_ISDIR(inode->i_mode)) {
5437	inode->i_op = &ext4_dir_inode_operations;
5438	inode->i_fop = &ext4_dir_operations;
5439	} else if (S_ISLNK(inode->i_mode)) {
5440	/* VFS does not allow setting these so must be corruption */
5441	if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5442	ext4_error_inode(inode, function, line, 0,
5443	"iget: immutable or append flags "
5444	"not allowed on symlinks");
5445	ret = -EFSCORRUPTED;
5446	goto bad_inode;
5447	}
5448	if (IS_ENCRYPTED(inode)) {
5449	inode->i_op = &ext4_encrypted_symlink_inode_operations;
5450	} else if (ext4_inode_is_fast_symlink(inode)) {
5451	inode->i_op = &ext4_fast_symlink_inode_operations;
5452	if (inode->i_size == 0 ||
5453	inode->i_size >= sizeof(ei->i_data) ||
5454	strnlen(p: (char *)ei->i_data, maxlen: inode->i_size + 1) !=
5455	inode->i_size) {
5456	ext4_error_inode(inode, function, line, 0,
5457	"invalid fast symlink length %llu",
5458	(unsigned long long)inode->i_size);
5459	ret = -EFSCORRUPTED;
5460	goto bad_inode;
5461	}
5462	inode_set_cached_link(inode, link: (char *)ei->i_data,
5463	linklen: inode->i_size);
5464	} else {
5465	inode->i_op = &ext4_symlink_inode_operations;
5466	}
5467	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5468	S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5469	inode->i_op = &ext4_special_inode_operations;
5470	if (raw_inode->i_block[0])
5471	init_special_inode(inode, inode->i_mode,
5472	old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5473	else
5474	init_special_inode(inode, inode->i_mode,
5475	new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5476	} else if (ino == EXT4_BOOT_LOADER_INO) {
5477	make_bad_inode(inode);
5478	} else {
5479	ret = -EFSCORRUPTED;
5480	ext4_error_inode(inode, function, line, 0,
5481	"iget: bogus i_mode (%o)", inode->i_mode);
5482	goto bad_inode;
5483	}
5484	if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(sb: inode->i_sb)) {
5485	ext4_error_inode(inode, function, line, 0,
5486	"casefold flag without casefold feature");
5487	ret = -EFSCORRUPTED;
5488	goto bad_inode;
5489	}
5490
5491	ext4_set_inode_mapping_order(inode);
5492
5493	ret = check_igot_inode(inode, flags, function, line);
5494	/*
5495	* -ESTALE here means there is nothing inherently wrong with the inode,
5496	* it's just not an inode we can return for an fhandle lookup.
5497	*/
5498	if (ret == -ESTALE) {
5499	brelse(bh: iloc.bh);
5500	unlock_new_inode(inode);
5501	iput(inode);
5502	return ERR_PTR(error: -ESTALE);
5503	}
5504	if (ret)
5505	goto bad_inode;
5506	brelse(bh: iloc.bh);
5507	/* Initialize the "no ACL's" state for the simple cases */
5508	if (!ext4_test_inode_state(inode, bit: EXT4_STATE_XATTR) && !ei->i_file_acl)
5509	cache_no_acl(inode);
5510	unlock_new_inode(inode);
5511	return inode;
5512
5513	bad_inode:
5514	brelse(bh: iloc.bh);
5515	iget_failed(inode);
5516	return ERR_PTR(error: ret);
5517	}
5518
5519	static void __ext4_update_other_inode_time(struct super_block *sb,
5520	unsigned long orig_ino,
5521	unsigned long ino,
5522	struct ext4_inode *raw_inode)
5523	{
5524	struct inode *inode;
5525
5526	inode = find_inode_by_ino_rcu(sb, ino);
5527	if (!inode)
5528	return;
5529
5530	if (!inode_is_dirtytime_only(inode))
5531	return;
5532
5533	spin_lock(lock: &inode->i_lock);
5534	if (inode_is_dirtytime_only(inode)) {
5535	struct ext4_inode_info *ei = EXT4_I(inode);
5536
5537	inode_state_clear(inode, flags: I_DIRTY_TIME);
5538	spin_unlock(lock: &inode->i_lock);
5539
5540	spin_lock(lock: &ei->i_raw_lock);
5541	EXT4_INODE_SET_CTIME(inode, raw_inode);
5542	EXT4_INODE_SET_MTIME(inode, raw_inode);
5543	EXT4_INODE_SET_ATIME(inode, raw_inode);
5544	ext4_inode_csum_set(inode, raw: raw_inode, ei);
5545	spin_unlock(lock: &ei->i_raw_lock);
5546	trace_ext4_other_inode_update_time(inode, orig_ino);
5547	return;
5548	}
5549	spin_unlock(lock: &inode->i_lock);
5550	}
5551
5552	/*
5553	* Opportunistically update the other time fields for other inodes in
5554	* the same inode table block.
5555	*/
5556	static void ext4_update_other_inodes_time(struct super_block *sb,
5557	unsigned long orig_ino, char *buf)
5558	{
5559	unsigned long ino;
5560	int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5561	int inode_size = EXT4_INODE_SIZE(sb);
5562
5563	/*
5564	* Calculate the first inode in the inode table block. Inode
5565	* numbers are one-based. That is, the first inode in a block
5566	* (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5567	*/
5568	ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5569	rcu_read_lock();
5570	for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5571	if (ino == orig_ino)
5572	continue;
5573	__ext4_update_other_inode_time(sb, orig_ino, ino,
5574	raw_inode: (struct ext4_inode *)buf);
5575	}
5576	rcu_read_unlock();
5577	}
5578
5579	/*
5580	* Post the struct inode info into an on-disk inode location in the
5581	* buffer-cache. This gobbles the caller's reference to the
5582	* buffer_head in the inode location struct.
5583	*
5584	* The caller must have write access to iloc->bh.
5585	*/
5586	static int ext4_do_update_inode(handle_t *handle,
5587	struct inode *inode,
5588	struct ext4_iloc *iloc)
5589	{
5590	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5591	struct ext4_inode_info *ei = EXT4_I(inode);
5592	struct buffer_head *bh = iloc->bh;
5593	struct super_block *sb = inode->i_sb;
5594	int err;
5595	int need_datasync = 0, set_large_file = 0;
5596
5597	spin_lock(lock: &ei->i_raw_lock);
5598
5599	/*
5600	* For fields not tracked in the in-memory inode, initialise them
5601	* to zero for new inodes.
5602	*/
5603	if (ext4_test_inode_state(inode, bit: EXT4_STATE_NEW))
5604	memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5605
5606	if (READ_ONCE(ei->i_disksize) != ext4_isize(sb: inode->i_sb, raw_inode))
5607	need_datasync = 1;
5608	if (ei->i_disksize > 0x7fffffffULL) {
5609	if (!ext4_has_feature_large_file(sb) ||
5610	EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5611	set_large_file = 1;
5612	}
5613
5614	err = ext4_fill_raw_inode(inode, raw_inode);
5615	spin_unlock(lock: &ei->i_raw_lock);
5616	if (err) {
5617	EXT4_ERROR_INODE(inode, "corrupted inode contents");
5618	goto out_brelse;
5619	}
5620
5621	if (inode->i_sb->s_flags & SB_LAZYTIME)
5622	ext4_update_other_inodes_time(sb: inode->i_sb, orig_ino: inode->i_ino,
5623	buf: bh->b_data);
5624
5625	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5626	err = ext4_handle_dirty_metadata(handle, NULL, bh);
5627	if (err)
5628	goto out_error;
5629	ext4_clear_inode_state(inode, bit: EXT4_STATE_NEW);
5630	if (set_large_file) {
5631	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5632	err = ext4_journal_get_write_access(handle, sb,
5633	EXT4_SB(sb)->s_sbh,
5634	EXT4_JTR_NONE);
5635	if (err)
5636	goto out_error;
5637	lock_buffer(bh: EXT4_SB(sb)->s_sbh);
5638	ext4_set_feature_large_file(sb);
5639	ext4_superblock_csum_set(sb);
5640	unlock_buffer(bh: EXT4_SB(sb)->s_sbh);
5641	ext4_handle_sync(handle);
5642	err = ext4_handle_dirty_metadata(handle, NULL,
5643	EXT4_SB(sb)->s_sbh);
5644	}
5645	ext4_update_inode_fsync_trans(handle, inode, datasync: need_datasync);
5646	out_error:
5647	ext4_std_error(inode->i_sb, err);
5648	out_brelse:
5649	brelse(bh);
5650	return err;
5651	}
5652
5653	/*
5654	* ext4_write_inode()
5655	*
5656	* We are called from a few places:
5657	*
5658	* - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5659	* Here, there will be no transaction running. We wait for any running
5660	* transaction to commit.
5661	*
5662	* - Within flush work (sys_sync(), kupdate and such).
5663	* We wait on commit, if told to.
5664	*
5665	* - Within iput_final() -> write_inode_now()
5666	* We wait on commit, if told to.
5667	*
5668	* In all cases it is actually safe for us to return without doing anything,
5669	* because the inode has been copied into a raw inode buffer in
5670	* ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5671	* writeback.
5672	*
5673	* Note that we are absolutely dependent upon all inode dirtiers doing the
5674	* right thing: they *must* call mark_inode_dirty() after dirtying info in
5675	* which we are interested.
5676	*
5677	* It would be a bug for them to not do this. The code:
5678	*
5679	* mark_inode_dirty(inode)
5680	* stuff();
5681	* inode->i_size = expr;
5682	*
5683	* is in error because write_inode() could occur while `stuff()' is running,
5684	* and the new i_size will be lost. Plus the inode will no longer be on the
5685	* superblock's dirty inode list.
5686	*/
5687	int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5688	{
5689	int err;
5690
5691	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5692	return 0;
5693
5694	err = ext4_emergency_state(sb: inode->i_sb);
5695	if (unlikely(err))
5696	return err;
5697
5698	if (EXT4_SB(sb: inode->i_sb)->s_journal) {
5699	if (ext4_journal_current_handle()) {
5700	ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5701	dump_stack();
5702	return -EIO;
5703	}
5704
5705	/*
5706	* No need to force transaction in WB_SYNC_NONE mode. Also
5707	* ext4_sync_fs() will force the commit after everything is
5708	* written.
5709	*/
5710	if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5711	return 0;
5712
5713	err = ext4_fc_commit(journal: EXT4_SB(sb: inode->i_sb)->s_journal,
5714	EXT4_I(inode)->i_sync_tid);
5715	} else {
5716	struct ext4_iloc iloc;
5717
5718	err = __ext4_get_inode_loc_noinmem(inode, iloc: &iloc);
5719	if (err)
5720	return err;
5721	/*
5722	* sync(2) will flush the whole buffer cache. No need to do
5723	* it here separately for each inode.
5724	*/
5725	if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5726	sync_dirty_buffer(bh: iloc.bh);
5727	if (buffer_req(bh: iloc.bh) && !buffer_uptodate(bh: iloc.bh)) {
5728	ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5729	"IO error syncing inode");
5730	err = -EIO;
5731	}
5732	brelse(bh: iloc.bh);
5733	}
5734	return err;
5735	}
5736
5737	/*
5738	* In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5739	* buffers that are attached to a folio straddling i_size and are undergoing
5740	* commit. In that case we have to wait for commit to finish and try again.
5741	*/
5742	static void ext4_wait_for_tail_page_commit(struct inode *inode)
5743	{
5744	unsigned offset;
5745	journal_t *journal = EXT4_SB(sb: inode->i_sb)->s_journal;
5746	tid_t commit_tid;
5747	int ret;
5748	bool has_transaction;
5749
5750	offset = inode->i_size & (PAGE_SIZE - 1);
5751	/*
5752	* If the folio is fully truncated, we don't need to wait for any commit
5753	* (and we even should not as __ext4_journalled_invalidate_folio() may
5754	* strip all buffers from the folio but keep the folio dirty which can then
5755	* confuse e.g. concurrent ext4_writepages() seeing dirty folio without
5756	* buffers). Also we don't need to wait for any commit if all buffers in
5757	* the folio remain valid. This is most beneficial for the common case of
5758	* blocksize == PAGESIZE.
5759	*/
5760	if (!offset || offset > (PAGE_SIZE - i_blocksize(node: inode)))
5761	return;
5762	while (1) {
5763	struct folio *folio = filemap_lock_folio(mapping: inode->i_mapping,
5764	index: inode->i_size >> PAGE_SHIFT);
5765	if (IS_ERR(ptr: folio))
5766	return;
5767	ret = __ext4_journalled_invalidate_folio(folio, offset,
5768	length: folio_size(folio) - offset);
5769	folio_unlock(folio);
5770	folio_put(folio);
5771	if (ret != -EBUSY)
5772	return;
5773	has_transaction = false;
5774	read_lock(&journal->j_state_lock);
5775	if (journal->j_committing_transaction) {
5776	commit_tid = journal->j_committing_transaction->t_tid;
5777	has_transaction = true;
5778	}
5779	read_unlock(&journal->j_state_lock);
5780	if (has_transaction)
5781	jbd2_log_wait_commit(journal, tid: commit_tid);
5782	}
5783	}
5784
5785	/*
5786	* ext4_setattr()
5787	*
5788	* Called from notify_change.
5789	*
5790	* We want to trap VFS attempts to truncate the file as soon as
5791	* possible. In particular, we want to make sure that when the VFS
5792	* shrinks i_size, we put the inode on the orphan list and modify
5793	* i_disksize immediately, so that during the subsequent flushing of
5794	* dirty pages and freeing of disk blocks, we can guarantee that any
5795	* commit will leave the blocks being flushed in an unused state on
5796	* disk. (On recovery, the inode will get truncated and the blocks will
5797	* be freed, so we have a strong guarantee that no future commit will
5798	* leave these blocks visible to the user.)
5799	*
5800	* Another thing we have to assure is that if we are in ordered mode
5801	* and inode is still attached to the committing transaction, we must
5802	* we start writeout of all the dirty pages which are being truncated.
5803	* This way we are sure that all the data written in the previous
5804	* transaction are already on disk (truncate waits for pages under
5805	* writeback).
5806	*
5807	* Called with inode->i_rwsem down.
5808	*/
5809	int ext4_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5810	struct iattr *attr)
5811	{
5812	struct inode *inode = d_inode(dentry);
5813	int error, rc = 0;
5814	int orphan = 0;
5815	const unsigned int ia_valid = attr->ia_valid;
5816	bool inc_ivers = true;
5817
5818	error = ext4_emergency_state(sb: inode->i_sb);
5819	if (unlikely(error))
5820	return error;
5821
5822	if (unlikely(IS_IMMUTABLE(inode)))
5823	return -EPERM;
5824
5825	if (unlikely(IS_APPEND(inode) &&
5826	(ia_valid & (ATTR_MODE | ATTR_UID |
5827	ATTR_GID | ATTR_TIMES_SET))))
5828	return -EPERM;
5829
5830	error = setattr_prepare(idmap, dentry, attr);
5831	if (error)
5832	return error;
5833
5834	error = fscrypt_prepare_setattr(dentry, attr);
5835	if (error)
5836	return error;
5837
5838	error = fsverity_prepare_setattr(dentry, attr);
5839	if (error)
5840	return error;
5841
5842	if (is_quota_modification(idmap, inode, ia: attr)) {
5843	error = dquot_initialize(inode);
5844	if (error)
5845	return error;
5846	}
5847
5848	if (i_uid_needs_update(idmap, attr, inode) ||
5849	i_gid_needs_update(idmap, attr, inode)) {
5850	handle_t *handle;
5851
5852	/* (user+group)*(old+new) structure, inode write (sb,
5853	* inode block, ? - but truncate inode update has it) */
5854	handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5855	(EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5856	EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5857	if (IS_ERR(ptr: handle)) {
5858	error = PTR_ERR(ptr: handle);
5859	goto err_out;
5860	}
5861
5862	/* dquot_transfer() calls back ext4_get_inode_usage() which
5863	* counts xattr inode references.
5864	*/
5865	down_read(sem: &EXT4_I(inode)->xattr_sem);
5866	error = dquot_transfer(idmap, inode, iattr: attr);
5867	up_read(sem: &EXT4_I(inode)->xattr_sem);
5868
5869	if (error) {
5870	ext4_journal_stop(handle);
5871	return error;
5872	}
5873	/* Update corresponding info in inode so that everything is in
5874	* one transaction */
5875	i_uid_update(idmap, attr, inode);
5876	i_gid_update(idmap, attr, inode);
5877	error = ext4_mark_inode_dirty(handle, inode);
5878	ext4_journal_stop(handle);
5879	if (unlikely(error)) {
5880	return error;
5881	}
5882	}
5883
5884	if (attr->ia_valid & ATTR_SIZE) {
5885	handle_t *handle;
5886	loff_t oldsize = inode->i_size;
5887	loff_t old_disksize;
5888	int shrink = (attr->ia_size < inode->i_size);
5889
5890	if (!(ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))) {
5891	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
5892
5893	if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5894	return -EFBIG;
5895	}
5896	}
5897	if (!S_ISREG(inode->i_mode)) {
5898	return -EINVAL;
5899	}
5900
5901	if (attr->ia_size == inode->i_size)
5902	inc_ivers = false;
5903
5904	if (shrink) {
5905	if (ext4_should_order_data(inode)) {
5906	error = ext4_begin_ordered_truncate(inode,
5907	new_size: attr->ia_size);
5908	if (error)
5909	goto err_out;
5910	}
5911	/*
5912	* Blocks are going to be removed from the inode. Wait
5913	* for dio in flight.
5914	*/
5915	inode_dio_wait(inode);
5916	}
5917
5918	filemap_invalidate_lock(mapping: inode->i_mapping);
5919
5920	rc = ext4_break_layouts(inode);
5921	if (rc) {
5922	filemap_invalidate_unlock(mapping: inode->i_mapping);
5923	goto err_out;
5924	}
5925
5926	if (attr->ia_size != inode->i_size) {
5927	/* attach jbd2 jinode for EOF folio tail zeroing */
5928	if (attr->ia_size & (inode->i_sb->s_blocksize - 1) ||
5929	oldsize & (inode->i_sb->s_blocksize - 1)) {
5930	error = ext4_inode_attach_jinode(inode);
5931	if (error)
5932	goto out_mmap_sem;
5933	}
5934
5935	handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5936	if (IS_ERR(ptr: handle)) {
5937	error = PTR_ERR(ptr: handle);
5938	goto out_mmap_sem;
5939	}
5940	if (ext4_handle_valid(handle) && shrink) {
5941	error = ext4_orphan_add(handle, inode);
5942	orphan = 1;
5943	}
5944	/*
5945	* Update c/mtime and tail zero the EOF folio on
5946	* truncate up. ext4_truncate() handles the shrink case
5947	* below.
5948	*/
5949	if (!shrink) {
5950	inode_set_mtime_to_ts(inode,
5951	ts: inode_set_ctime_current(inode));
5952	if (oldsize & (inode->i_sb->s_blocksize - 1))
5953	ext4_block_truncate_page(handle,
5954	mapping: inode->i_mapping, from: oldsize);
5955	}
5956
5957	if (shrink)
5958	ext4_fc_track_range(handle, inode,
5959	start: (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5960	inode->i_sb->s_blocksize_bits,
5961	EXT_MAX_BLOCKS - 1);
5962	else
5963	ext4_fc_track_range(
5964	handle, inode,
5965	start: (oldsize > 0 ? oldsize - 1 : oldsize) >>
5966	inode->i_sb->s_blocksize_bits,
5967	end: (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5968	inode->i_sb->s_blocksize_bits);
5969
5970	down_write(sem: &EXT4_I(inode)->i_data_sem);
5971	old_disksize = EXT4_I(inode)->i_disksize;
5972	EXT4_I(inode)->i_disksize = attr->ia_size;
5973
5974	/*
5975	* We have to update i_size under i_data_sem together
5976	* with i_disksize to avoid races with writeback code
5977	* running ext4_wb_update_i_disksize().
5978	*/
5979	if (!error)
5980	i_size_write(inode, i_size: attr->ia_size);
5981	else
5982	EXT4_I(inode)->i_disksize = old_disksize;
5983	up_write(sem: &EXT4_I(inode)->i_data_sem);
5984	rc = ext4_mark_inode_dirty(handle, inode);
5985	if (!error)
5986	error = rc;
5987	ext4_journal_stop(handle);
5988	if (error)
5989	goto out_mmap_sem;
5990	if (!shrink) {
5991	pagecache_isize_extended(inode, from: oldsize,
5992	to: inode->i_size);
5993	} else if (ext4_should_journal_data(inode)) {
5994	ext4_wait_for_tail_page_commit(inode);
5995	}
5996	}
5997
5998	/*
5999	* Truncate pagecache after we've waited for commit
6000	* in data=journal mode to make pages freeable.
6001	*/
6002	truncate_pagecache(inode, new: inode->i_size);
6003	/*
6004	* Call ext4_truncate() even if i_size didn't change to
6005	* truncate possible preallocated blocks.
6006	*/
6007	if (attr->ia_size <= oldsize) {
6008	rc = ext4_truncate(inode);
6009	if (rc)
6010	error = rc;
6011	}
6012	out_mmap_sem:
6013	filemap_invalidate_unlock(mapping: inode->i_mapping);
6014	}
6015
6016	if (!error) {
6017	if (inc_ivers)
6018	inode_inc_iversion(inode);
6019	setattr_copy(idmap, inode, attr);
6020	mark_inode_dirty(inode);
6021	}
6022
6023	/*
6024	* If the call to ext4_truncate failed to get a transaction handle at
6025	* all, we need to clean up the in-core orphan list manually.
6026	*/
6027	if (orphan && inode->i_nlink)
6028	ext4_orphan_del(NULL, inode);
6029
6030	if (!error && (ia_valid & ATTR_MODE))
6031	rc = posix_acl_chmod(idmap, dentry, inode->i_mode);
6032
6033	err_out:
6034	if (error)
6035	ext4_std_error(inode->i_sb, error);
6036	if (!error)
6037	error = rc;
6038	return error;
6039	}
6040
6041	u32 ext4_dio_alignment(struct inode *inode)
6042	{
6043	if (fsverity_active(inode))
6044	return 0;
6045	if (ext4_should_journal_data(inode))
6046	return 0;
6047	if (ext4_has_inline_data(inode))
6048	return 0;
6049	if (IS_ENCRYPTED(inode)) {
6050	if (!fscrypt_dio_supported(inode))
6051	return 0;
6052	return i_blocksize(node: inode);
6053	}
6054	return 1; /* use the iomap defaults */
6055	}
6056
6057	int ext4_getattr(struct mnt_idmap *idmap, const struct path *path,
6058	struct kstat *stat, u32 request_mask, unsigned int query_flags)
6059	{
6060	struct inode *inode = d_inode(dentry: path->dentry);
6061	struct ext4_inode *raw_inode;
6062	struct ext4_inode_info *ei = EXT4_I(inode);
6063	unsigned int flags;
6064
6065	if ((request_mask & STATX_BTIME) &&
6066	EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
6067	stat->result_mask |= STATX_BTIME;
6068	stat->btime.tv_sec = ei->i_crtime.tv_sec;
6069	stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
6070	}
6071
6072	/*
6073	* Return the DIO alignment restrictions if requested. We only return
6074	* this information when requested, since on encrypted files it might
6075	* take a fair bit of work to get if the file wasn't opened recently.
6076	*/
6077	if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
6078	u32 dio_align = ext4_dio_alignment(inode);
6079
6080	stat->result_mask |= STATX_DIOALIGN;
6081	if (dio_align == 1) {
6082	struct block_device *bdev = inode->i_sb->s_bdev;
6083
6084	/* iomap defaults */
6085	stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
6086	stat->dio_offset_align = bdev_logical_block_size(bdev);
6087	} else {
6088	stat->dio_mem_align = dio_align;
6089	stat->dio_offset_align = dio_align;
6090	}
6091	}
6092
6093	if ((request_mask & STATX_WRITE_ATOMIC) && S_ISREG(inode->i_mode)) {
6094	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
6095	unsigned int awu_min = 0, awu_max = 0;
6096
6097	if (ext4_inode_can_atomic_write(inode)) {
6098	awu_min = sbi->s_awu_min;
6099	awu_max = sbi->s_awu_max;
6100	}
6101
6102	generic_fill_statx_atomic_writes(stat, unit_min: awu_min, unit_max: awu_max, unit_max_opt: 0);
6103	}
6104
6105	flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
6106	if (flags & EXT4_APPEND_FL)
6107	stat->attributes |= STATX_ATTR_APPEND;
6108	if (flags & EXT4_COMPR_FL)
6109	stat->attributes |= STATX_ATTR_COMPRESSED;
6110	if (flags & EXT4_ENCRYPT_FL)
6111	stat->attributes |= STATX_ATTR_ENCRYPTED;
6112	if (flags & EXT4_IMMUTABLE_FL)
6113	stat->attributes |= STATX_ATTR_IMMUTABLE;
6114	if (flags & EXT4_NODUMP_FL)
6115	stat->attributes |= STATX_ATTR_NODUMP;
6116	if (flags & EXT4_VERITY_FL)
6117	stat->attributes |= STATX_ATTR_VERITY;
6118
6119	stat->attributes_mask |= (STATX_ATTR_APPEND |
6120	STATX_ATTR_COMPRESSED |
6121	STATX_ATTR_ENCRYPTED |
6122	STATX_ATTR_IMMUTABLE |
6123	STATX_ATTR_NODUMP |
6124	STATX_ATTR_VERITY);
6125
6126	generic_fillattr(idmap, request_mask, inode, stat);
6127	return 0;
6128	}
6129
6130	int ext4_file_getattr(struct mnt_idmap *idmap,
6131	const struct path *path, struct kstat *stat,
6132	u32 request_mask, unsigned int query_flags)
6133	{
6134	struct inode *inode = d_inode(dentry: path->dentry);
6135	u64 delalloc_blocks;
6136
6137	ext4_getattr(idmap, path, stat, request_mask, query_flags);
6138
6139	/*
6140	* If there is inline data in the inode, the inode will normally not
6141	* have data blocks allocated (it may have an external xattr block).
6142	* Report at least one sector for such files, so tools like tar, rsync,
6143	* others don't incorrectly think the file is completely sparse.
6144	*/
6145	if (unlikely(ext4_has_inline_data(inode)))
6146	stat->blocks += (stat->size + 511) >> 9;
6147
6148	/*
6149	* We can't update i_blocks if the block allocation is delayed
6150	* otherwise in the case of system crash before the real block
6151	* allocation is done, we will have i_blocks inconsistent with
6152	* on-disk file blocks.
6153	* We always keep i_blocks updated together with real
6154	* allocation. But to not confuse with user, stat
6155	* will return the blocks that include the delayed allocation
6156	* blocks for this file.
6157	*/
6158	delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
6159	EXT4_I(inode)->i_reserved_data_blocks);
6160	stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
6161	return 0;
6162	}
6163
6164	static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
6165	int pextents)
6166	{
6167	if (!(ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS)))
6168	return ext4_ind_trans_blocks(inode, nrblocks: lblocks);
6169	return ext4_ext_index_trans_blocks(inode, extents: pextents);
6170	}
6171
6172	/*
6173	* Account for index blocks, block groups bitmaps and block group
6174	* descriptor blocks if modify datablocks and index blocks
6175	* worse case, the indexs blocks spread over different block groups
6176	*
6177	* If datablocks are discontiguous, they are possible to spread over
6178	* different block groups too. If they are contiguous, with flexbg,
6179	* they could still across block group boundary.
6180	*
6181	* Also account for superblock, inode, quota and xattr blocks
6182	*/
6183	int ext4_meta_trans_blocks(struct inode *inode, int lblocks, int pextents)
6184	{
6185	ext4_group_t groups, ngroups = ext4_get_groups_count(sb: inode->i_sb);
6186	int gdpblocks;
6187	int idxblocks;
6188	int ret;
6189
6190	/*
6191	* How many index and leaf blocks need to touch to map @lblocks
6192	* logical blocks to @pextents physical extents?
6193	*/
6194	idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
6195
6196	/*
6197	* Now let's see how many group bitmaps and group descriptors need
6198	* to account
6199	*/
6200	groups = idxblocks + pextents;
6201	gdpblocks = groups;
6202	if (groups > ngroups)
6203	groups = ngroups;
6204	if (groups > EXT4_SB(sb: inode->i_sb)->s_gdb_count)
6205	gdpblocks = EXT4_SB(sb: inode->i_sb)->s_gdb_count;
6206
6207	/* bitmaps and block group descriptor blocks */
6208	ret = idxblocks + groups + gdpblocks;
6209
6210	/* Blocks for super block, inode, quota and xattr blocks */
6211	ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
6212
6213	return ret;
6214	}
6215
6216	/*
6217	* Calculate the journal credits for modifying the number of blocks
6218	* in a single extent within one transaction. 'nrblocks' is used only
6219	* for non-extent inodes. For extent type inodes, 'nrblocks' can be
6220	* zero if the exact number of blocks is unknown.
6221	*/
6222	int ext4_chunk_trans_extent(struct inode *inode, int nrblocks)
6223	{
6224	int ret;
6225
6226	ret = ext4_meta_trans_blocks(inode, lblocks: nrblocks, pextents: 1);
6227	/* Account for data blocks for journalled mode */
6228	if (ext4_should_journal_data(inode))
6229	ret += nrblocks;
6230	return ret;
6231	}
6232
6233	/*
6234	* Calculate the journal credits for a chunk of data modification.
6235	*
6236	* This is called from DIO, fallocate or whoever calling
6237	* ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
6238	*
6239	* journal buffers for data blocks are not included here, as DIO
6240	* and fallocate do no need to journal data buffers.
6241	*/
6242	int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
6243	{
6244	return ext4_meta_trans_blocks(inode, lblocks: nrblocks, pextents: 1);
6245	}
6246
6247	/*
6248	* The caller must have previously called ext4_reserve_inode_write().
6249	* Give this, we know that the caller already has write access to iloc->bh.
6250	*/
6251	int ext4_mark_iloc_dirty(handle_t *handle,
6252	struct inode *inode, struct ext4_iloc *iloc)
6253	{
6254	int err = 0;
6255
6256	err = ext4_emergency_state(sb: inode->i_sb);
6257	if (unlikely(err)) {
6258	put_bh(bh: iloc->bh);
6259	return err;
6260	}
6261	ext4_fc_track_inode(handle, inode);
6262
6263	/* the do_update_inode consumes one bh->b_count */
6264	get_bh(bh: iloc->bh);
6265
6266	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
6267	err = ext4_do_update_inode(handle, inode, iloc);
6268	put_bh(bh: iloc->bh);
6269	return err;
6270	}
6271
6272	/*
6273	* On success, We end up with an outstanding reference count against
6274	* iloc->bh. This _must_ be cleaned up later.
6275	*/
6276
6277	int
6278	ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
6279	struct ext4_iloc *iloc)
6280	{
6281	int err;
6282
6283	err = ext4_emergency_state(sb: inode->i_sb);
6284	if (unlikely(err))
6285	return err;
6286
6287	err = ext4_get_inode_loc(inode, iloc);
6288	if (!err) {
6289	BUFFER_TRACE(iloc->bh, "get_write_access");
6290	err = ext4_journal_get_write_access(handle, inode->i_sb,
6291	iloc->bh, EXT4_JTR_NONE);
6292	if (err) {
6293	brelse(bh: iloc->bh);
6294	iloc->bh = NULL;
6295	}
6296	ext4_fc_track_inode(handle, inode);
6297	}
6298	ext4_std_error(inode->i_sb, err);
6299	return err;
6300	}
6301
6302	static int __ext4_expand_extra_isize(struct inode *inode,
6303	unsigned int new_extra_isize,
6304	struct ext4_iloc *iloc,
6305	handle_t *handle, int *no_expand)
6306	{
6307	struct ext4_inode *raw_inode;
6308	struct ext4_xattr_ibody_header *header;
6309	unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
6310	struct ext4_inode_info *ei = EXT4_I(inode);
6311	int error;
6312
6313	/* this was checked at iget time, but double check for good measure */
6314	if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
6315	(ei->i_extra_isize & 3)) {
6316	EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
6317	ei->i_extra_isize,
6318	EXT4_INODE_SIZE(inode->i_sb));
6319	return -EFSCORRUPTED;
6320	}
6321	if ((new_extra_isize < ei->i_extra_isize) ||
6322	(new_extra_isize < 4) ||
6323	(new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
6324	return -EINVAL; /* Should never happen */
6325
6326	raw_inode = ext4_raw_inode(iloc);
6327
6328	header = IHDR(inode, raw_inode);
6329
6330	/* No extended attributes present */
6331	if (!ext4_test_inode_state(inode, bit: EXT4_STATE_XATTR) ||
6332	header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
6333	memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
6334	EXT4_I(inode)->i_extra_isize, 0,
6335	new_extra_isize - EXT4_I(inode)->i_extra_isize);
6336	EXT4_I(inode)->i_extra_isize = new_extra_isize;
6337	return 0;
6338	}
6339
6340	/*
6341	* We may need to allocate external xattr block so we need quotas
6342	* initialized. Here we can be called with various locks held so we
6343	* cannot affort to initialize quotas ourselves. So just bail.
6344	*/
6345	if (dquot_initialize_needed(inode))
6346	return -EAGAIN;
6347
6348	/* try to expand with EAs present */
6349	error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
6350	raw_inode, handle);
6351	if (error) {
6352	/*
6353	* Inode size expansion failed; don't try again
6354	*/
6355	*no_expand = 1;
6356	}
6357
6358	return error;
6359	}
6360
6361	/*
6362	* Expand an inode by new_extra_isize bytes.
6363	* Returns 0 on success or negative error number on failure.
6364	*/
6365	static int ext4_try_to_expand_extra_isize(struct inode *inode,
6366	unsigned int new_extra_isize,
6367	struct ext4_iloc iloc,
6368	handle_t *handle)
6369	{
6370	int no_expand;
6371	int error;
6372
6373	if (ext4_test_inode_state(inode, bit: EXT4_STATE_NO_EXPAND))
6374	return -EOVERFLOW;
6375
6376	/*
6377	* In nojournal mode, we can immediately attempt to expand
6378	* the inode. When journaled, we first need to obtain extra
6379	* buffer credits since we may write into the EA block
6380	* with this same handle. If journal_extend fails, then it will
6381	* only result in a minor loss of functionality for that inode.
6382	* If this is felt to be critical, then e2fsck should be run to
6383	* force a large enough s_min_extra_isize.
6384	*/
6385	if (ext4_journal_extend(handle,
6386	EXT4_DATA_TRANS_BLOCKS(inode->i_sb), revoke: 0) != 0)
6387	return -ENOSPC;
6388
6389	if (ext4_write_trylock_xattr(inode, save: &no_expand) == 0)
6390	return -EBUSY;
6391
6392	error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc: &iloc,
6393	handle, no_expand: &no_expand);
6394	ext4_write_unlock_xattr(inode, save: &no_expand);
6395
6396	return error;
6397	}
6398
6399	int ext4_expand_extra_isize(struct inode *inode,
6400	unsigned int new_extra_isize,
6401	struct ext4_iloc *iloc)
6402	{
6403	handle_t *handle;
6404	int no_expand;
6405	int error, rc;
6406
6407	if (ext4_test_inode_state(inode, bit: EXT4_STATE_NO_EXPAND)) {
6408	brelse(bh: iloc->bh);
6409	return -EOVERFLOW;
6410	}
6411
6412	handle = ext4_journal_start(inode, EXT4_HT_INODE,
6413	EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
6414	if (IS_ERR(ptr: handle)) {
6415	error = PTR_ERR(ptr: handle);
6416	brelse(bh: iloc->bh);
6417	return error;
6418	}
6419
6420	ext4_write_lock_xattr(inode, save: &no_expand);
6421
6422	BUFFER_TRACE(iloc->bh, "get_write_access");
6423	error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
6424	EXT4_JTR_NONE);
6425	if (error) {
6426	brelse(bh: iloc->bh);
6427	goto out_unlock;
6428	}
6429
6430	error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
6431	handle, no_expand: &no_expand);
6432
6433	rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6434	if (!error)
6435	error = rc;
6436
6437	out_unlock:
6438	ext4_write_unlock_xattr(inode, save: &no_expand);
6439	ext4_journal_stop(handle);
6440	return error;
6441	}
6442
6443	/*
6444	* What we do here is to mark the in-core inode as clean with respect to inode
6445	* dirtiness (it may still be data-dirty).
6446	* This means that the in-core inode may be reaped by prune_icache
6447	* without having to perform any I/O. This is a very good thing,
6448	* because *any* task may call prune_icache - even ones which
6449	* have a transaction open against a different journal.
6450	*
6451	* Is this cheating? Not really. Sure, we haven't written the
6452	* inode out, but prune_icache isn't a user-visible syncing function.
6453	* Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6454	* we start and wait on commits.
6455	*/
6456	int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
6457	const char *func, unsigned int line)
6458	{
6459	struct ext4_iloc iloc;
6460	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
6461	int err;
6462
6463	might_sleep();
6464	trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6465	err = ext4_reserve_inode_write(handle, inode, iloc: &iloc);
6466	if (err)
6467	goto out;
6468
6469	if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6470	ext4_try_to_expand_extra_isize(inode, new_extra_isize: sbi->s_want_extra_isize,
6471	iloc, handle);
6472
6473	err = ext4_mark_iloc_dirty(handle, inode, iloc: &iloc);
6474	out:
6475	if (unlikely(err))
6476	ext4_error_inode_err(inode, func, line, 0, err,
6477	"mark_inode_dirty error");
6478	return err;
6479	}
6480
6481	/*
6482	* ext4_dirty_inode() is called from __mark_inode_dirty()
6483	*
6484	* We're really interested in the case where a file is being extended.
6485	* i_size has been changed by generic_commit_write() and we thus need
6486	* to include the updated inode in the current transaction.
6487	*
6488	* Also, dquot_alloc_block() will always dirty the inode when blocks
6489	* are allocated to the file.
6490	*
6491	* If the inode is marked synchronous, we don't honour that here - doing
6492	* so would cause a commit on atime updates, which we don't bother doing.
6493	* We handle synchronous inodes at the highest possible level.
6494	*/
6495	void ext4_dirty_inode(struct inode *inode, int flags)
6496	{
6497	handle_t *handle;
6498
6499	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6500	if (IS_ERR(ptr: handle))
6501	return;
6502	ext4_mark_inode_dirty(handle, inode);
6503	ext4_journal_stop(handle);
6504	}
6505
6506	int ext4_change_inode_journal_flag(struct inode *inode, int val)
6507	{
6508	journal_t *journal;
6509	handle_t *handle;
6510	int err;
6511	int alloc_ctx;
6512
6513	/*
6514	* We have to be very careful here: changing a data block's
6515	* journaling status dynamically is dangerous. If we write a
6516	* data block to the journal, change the status and then delete
6517	* that block, we risk forgetting to revoke the old log record
6518	* from the journal and so a subsequent replay can corrupt data.
6519	* So, first we make sure that the journal is empty and that
6520	* nobody is changing anything.
6521	*/
6522
6523	journal = EXT4_JOURNAL(inode);
6524	if (!journal)
6525	return 0;
6526	if (is_journal_aborted(journal))
6527	return -EROFS;
6528
6529	/* Wait for all existing dio workers */
6530	inode_dio_wait(inode);
6531
6532	/*
6533	* Before flushing the journal and switching inode's aops, we have
6534	* to flush all dirty data the inode has. There can be outstanding
6535	* delayed allocations, there can be unwritten extents created by
6536	* fallocate or buffered writes in dioread_nolock mode covered by
6537	* dirty data which can be converted only after flushing the dirty
6538	* data (and journalled aops don't know how to handle these cases).
6539	*/
6540	filemap_invalidate_lock(mapping: inode->i_mapping);
6541	err = filemap_write_and_wait(mapping: inode->i_mapping);
6542	if (err < 0) {
6543	filemap_invalidate_unlock(mapping: inode->i_mapping);
6544	return err;
6545	}
6546	/* Before switch the inode journalling mode evict all the page cache. */
6547	truncate_pagecache(inode, new: 0);
6548
6549	alloc_ctx = ext4_writepages_down_write(sb: inode->i_sb);
6550	jbd2_journal_lock_updates(journal);
6551
6552	/*
6553	* OK, there are no updates running now, and all cached data is
6554	* synced to disk. We are now in a completely consistent state
6555	* which doesn't have anything in the journal, and we know that
6556	* no filesystem updates are running, so it is safe to modify
6557	* the inode's in-core data-journaling state flag now.
6558	*/
6559
6560	if (val)
6561	ext4_set_inode_flag(inode, bit: EXT4_INODE_JOURNAL_DATA);
6562	else {
6563	err = jbd2_journal_flush(journal, flags: 0);
6564	if (err < 0) {
6565	jbd2_journal_unlock_updates(journal);
6566	ext4_writepages_up_write(sb: inode->i_sb, ctx: alloc_ctx);
6567	filemap_invalidate_unlock(mapping: inode->i_mapping);
6568	return err;
6569	}
6570	ext4_clear_inode_flag(inode, bit: EXT4_INODE_JOURNAL_DATA);
6571	}
6572	ext4_set_aops(inode);
6573	ext4_set_inode_mapping_order(inode);
6574
6575	jbd2_journal_unlock_updates(journal);
6576	ext4_writepages_up_write(sb: inode->i_sb, ctx: alloc_ctx);
6577	filemap_invalidate_unlock(mapping: inode->i_mapping);
6578
6579	/* Finally we can mark the inode as dirty. */
6580
6581	handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6582	if (IS_ERR(ptr: handle))
6583	return PTR_ERR(ptr: handle);
6584
6585	ext4_fc_mark_ineligible(sb: inode->i_sb,
6586	reason: EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6587	err = ext4_mark_inode_dirty(handle, inode);
6588	ext4_handle_sync(handle);
6589	ext4_journal_stop(handle);
6590	ext4_std_error(inode->i_sb, err);
6591
6592	return err;
6593	}
6594
6595	static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6596	struct buffer_head *bh)
6597	{
6598	return !buffer_mapped(bh);
6599	}
6600
6601	static int ext4_block_page_mkwrite(struct inode *inode, struct folio *folio,
6602	get_block_t get_block)
6603	{
6604	handle_t *handle;
6605	loff_t size;
6606	unsigned long len;
6607	int credits;
6608	int ret;
6609
6610	credits = ext4_chunk_trans_extent(inode,
6611	nrblocks: ext4_journal_blocks_per_folio(inode));
6612	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, credits);
6613	if (IS_ERR(ptr: handle))
6614	return PTR_ERR(ptr: handle);
6615
6616	folio_lock(folio);
6617	size = i_size_read(inode);
6618	/* Page got truncated from under us? */
6619	if (folio->mapping != inode->i_mapping || folio_pos(folio) > size) {
6620	ret = -EFAULT;
6621	goto out_error;
6622	}
6623
6624	len = folio_size(folio);
6625	if (folio_pos(folio) + len > size)
6626	len = size - folio_pos(folio);
6627
6628	ret = ext4_block_write_begin(handle, folio, pos: 0, len, get_block);
6629	if (ret)
6630	goto out_error;
6631
6632	if (!ext4_should_journal_data(inode)) {
6633	block_commit_write(folio, from: 0, to: len);
6634	folio_mark_dirty(folio);
6635	} else {
6636	ret = ext4_journal_folio_buffers(handle, folio, len);
6637	if (ret)
6638	goto out_error;
6639	}
6640	ext4_journal_stop(handle);
6641	folio_wait_stable(folio);
6642	return ret;
6643
6644	out_error:
6645	folio_unlock(folio);
6646	ext4_journal_stop(handle);
6647	return ret;
6648	}
6649
6650	vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6651	{
6652	struct vm_area_struct *vma = vmf->vma;
6653	struct folio *folio = page_folio(vmf->page);
6654	loff_t size;
6655	unsigned long len;
6656	int err;
6657	vm_fault_t ret;
6658	struct file *file = vma->vm_file;
6659	struct inode *inode = file_inode(f: file);
6660	struct address_space *mapping = inode->i_mapping;
6661	get_block_t *get_block = ext4_get_block;
6662	int retries = 0;
6663
6664	if (unlikely(IS_IMMUTABLE(inode)))
6665	return VM_FAULT_SIGBUS;
6666
6667	sb_start_pagefault(sb: inode->i_sb);
6668	file_update_time(file: vma->vm_file);
6669
6670	filemap_invalidate_lock_shared(mapping);
6671
6672	err = ext4_convert_inline_data(inode);
6673	if (err)
6674	goto out_ret;
6675
6676	/*
6677	* On data journalling we skip straight to the transaction handle:
6678	* there's no delalloc; page truncated will be checked later; the
6679	* early return w/ all buffers mapped (calculates size/len) can't
6680	* be used; and there's no dioread_nolock, so only ext4_get_block.
6681	*/
6682	if (ext4_should_journal_data(inode))
6683	goto retry_alloc;
6684
6685	/* Delalloc case is easy... */
6686	if (test_opt(inode->i_sb, DELALLOC) &&
6687	!ext4_nonda_switch(sb: inode->i_sb)) {
6688	do {
6689	err = block_page_mkwrite(vma, vmf,
6690	get_block: ext4_da_get_block_prep);
6691	} while (err == -ENOSPC &&
6692	ext4_should_retry_alloc(sb: inode->i_sb, retries: &retries));
6693	goto out_ret;
6694	}
6695
6696	folio_lock(folio);
6697	size = i_size_read(inode);
6698	/* Page got truncated from under us? */
6699	if (folio->mapping != mapping || folio_pos(folio) > size) {
6700	folio_unlock(folio);
6701	ret = VM_FAULT_NOPAGE;
6702	goto out;
6703	}
6704
6705	len = folio_size(folio);
6706	if (folio_pos(folio) + len > size)
6707	len = size - folio_pos(folio);
6708	/*
6709	* Return if we have all the buffers mapped. This avoids the need to do
6710	* journal_start/journal_stop which can block and take a long time
6711	*
6712	* This cannot be done for data journalling, as we have to add the
6713	* inode to the transaction's list to writeprotect pages on commit.
6714	*/
6715	if (folio_buffers(folio)) {
6716	if (!ext4_walk_page_buffers(NULL, inode, folio_buffers(folio),
6717	from: 0, to: len, NULL,
6718	fn: ext4_bh_unmapped)) {
6719	/* Wait so that we don't change page under IO */
6720	folio_wait_stable(folio);
6721	ret = VM_FAULT_LOCKED;
6722	goto out;
6723	}
6724	}
6725	folio_unlock(folio);
6726	/* OK, we need to fill the hole... */
6727	if (ext4_should_dioread_nolock(inode))
6728	get_block = ext4_get_block_unwritten;
6729	retry_alloc:
6730	/* Start journal and allocate blocks */
6731	err = ext4_block_page_mkwrite(inode, folio, get_block);
6732	if (err == -EAGAIN ||
6733	(err == -ENOSPC && ext4_should_retry_alloc(sb: inode->i_sb, retries: &retries)))
6734	goto retry_alloc;
6735	out_ret:
6736	ret = vmf_fs_error(err);
6737	out:
6738	filemap_invalidate_unlock_shared(mapping);
6739	sb_end_pagefault(sb: inode->i_sb);
6740	return ret;
6741	}
6742