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_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
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(&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(&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	if (!list_empty_careful(head: &inode->i_io_list))
206	inode_io_list_del(inode);
207
208	/*
209	* Protect us against freezing - iput() caller didn't have to have any
210	* protection against it. When we are in a running transaction though,
211	* we are already protected against freezing and we cannot grab further
212	* protection due to lock ordering constraints.
213	*/
214	if (!ext4_journal_current_handle()) {
215	sb_start_intwrite(sb: inode->i_sb);
216	freeze_protected = true;
217	}
218
219	if (!IS_NOQUOTA(inode))
220	extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
221
222	/*
223	* Block bitmap, group descriptor, and inode are accounted in both
224	* ext4_blocks_for_truncate() and extra_credits. So subtract 3.
225	*/
226	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
227	ext4_blocks_for_truncate(inode) + extra_credits - 3);
228	if (IS_ERR(ptr: handle)) {
229	ext4_std_error(inode->i_sb, PTR_ERR(handle));
230	/*
231	* If we're going to skip the normal cleanup, we still need to
232	* make sure that the in-core orphan linked list is properly
233	* cleaned up.
234	*/
235	ext4_orphan_del(NULL, inode);
236	if (freeze_protected)
237	sb_end_intwrite(sb: inode->i_sb);
238	goto no_delete;
239	}
240
241	if (IS_SYNC(inode))
242	ext4_handle_sync(handle);
243
244	/*
245	* Set inode->i_size to 0 before calling ext4_truncate(). We need
246	* special handling of symlinks here because i_size is used to
247	* determine whether ext4_inode_info->i_data contains symlink data or
248	* block mappings. Setting i_size to 0 will remove its fast symlink
249	* status. Erase i_data so that it becomes a valid empty block map.
250	*/
251	if (ext4_inode_is_fast_symlink(inode))
252	memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
253	inode->i_size = 0;
254	err = ext4_mark_inode_dirty(handle, inode);
255	if (err) {
256	ext4_warning(inode->i_sb,
257	"couldn't mark inode dirty (err %d)", err);
258	goto stop_handle;
259	}
260	if (inode->i_blocks) {
261	err = ext4_truncate(inode);
262	if (err) {
263	ext4_error_err(inode->i_sb, -err,
264	"couldn't truncate inode %lu (err %d)",
265	inode->i_ino, err);
266	goto stop_handle;
267	}
268	}
269
270	/* Remove xattr references. */
271	err = ext4_xattr_delete_inode(handle, inode, array: &ea_inode_array,
272	extra_credits);
273	if (err) {
274	ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
275	stop_handle:
276	ext4_journal_stop(handle);
277	ext4_orphan_del(NULL, inode);
278	if (freeze_protected)
279	sb_end_intwrite(sb: inode->i_sb);
280	ext4_xattr_inode_array_free(array: ea_inode_array);
281	goto no_delete;
282	}
283
284	/*
285	* Kill off the orphan record which ext4_truncate created.
286	* AKPM: I think this can be inside the above `if'.
287	* Note that ext4_orphan_del() has to be able to cope with the
288	* deletion of a non-existent orphan - this is because we don't
289	* know if ext4_truncate() actually created an orphan record.
290	* (Well, we could do this if we need to, but heck - it works)
291	*/
292	ext4_orphan_del(handle, inode);
293	EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
294
295	/*
296	* One subtle ordering requirement: if anything has gone wrong
297	* (transaction abort, IO errors, whatever), then we can still
298	* do these next steps (the fs will already have been marked as
299	* having errors), but we can't free the inode if the mark_dirty
300	* fails.
301	*/
302	if (ext4_mark_inode_dirty(handle, inode))
303	/* If that failed, just do the required in-core inode clear. */
304	ext4_clear_inode(inode);
305	else
306	ext4_free_inode(handle, inode);
307	ext4_journal_stop(handle);
308	if (freeze_protected)
309	sb_end_intwrite(sb: inode->i_sb);
310	ext4_xattr_inode_array_free(array: ea_inode_array);
311	return;
312	no_delete:
313	/*
314	* Check out some where else accidentally dirty the evicting inode,
315	* which may probably cause inode use-after-free issues later.
316	*/
317	WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list));
318
319	if (!list_empty(head: &EXT4_I(inode)->i_fc_list))
320	ext4_fc_mark_ineligible(sb: inode->i_sb, reason: EXT4_FC_REASON_NOMEM, NULL);
321	ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
322	}
323
324	#ifdef CONFIG_QUOTA
325	qsize_t *ext4_get_reserved_space(struct inode *inode)
326	{
327	return &EXT4_I(inode)->i_reserved_quota;
328	}
329	#endif
330
331	/*
332	* Called with i_data_sem down, which is important since we can call
333	* ext4_discard_preallocations() from here.
334	*/
335	void ext4_da_update_reserve_space(struct inode *inode,
336	int used, int quota_claim)
337	{
338	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
339	struct ext4_inode_info *ei = EXT4_I(inode);
340
341	spin_lock(lock: &ei->i_block_reservation_lock);
342	trace_ext4_da_update_reserve_space(inode, used_blocks: used, quota_claim);
343	if (unlikely(used > ei->i_reserved_data_blocks)) {
344	ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
345	"with only %d reserved data blocks",
346	__func__, inode->i_ino, used,
347	ei->i_reserved_data_blocks);
348	WARN_ON(1);
349	used = ei->i_reserved_data_blocks;
350	}
351
352	/* Update per-inode reservations */
353	ei->i_reserved_data_blocks -= used;
354	percpu_counter_sub(fbc: &sbi->s_dirtyclusters_counter, amount: used);
355
356	spin_unlock(lock: &ei->i_block_reservation_lock);
357
358	/* Update quota subsystem for data blocks */
359	if (quota_claim)
360	dquot_claim_block(inode, EXT4_C2B(sbi, used));
361	else {
362	/*
363	* We did fallocate with an offset that is already delayed
364	* allocated. So on delayed allocated writeback we should
365	* not re-claim the quota for fallocated blocks.
366	*/
367	dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
368	}
369
370	/*
371	* If we have done all the pending block allocations and if
372	* there aren't any writers on the inode, we can discard the
373	* inode's preallocations.
374	*/
375	if ((ei->i_reserved_data_blocks == 0) &&
376	!inode_is_open_for_write(inode))
377	ext4_discard_preallocations(inode);
378	}
379
380	static int __check_block_validity(struct inode *inode, const char *func,
381	unsigned int line,
382	struct ext4_map_blocks *map)
383	{
384	journal_t *journal = EXT4_SB(sb: inode->i_sb)->s_journal;
385
386	if (journal && inode == journal->j_inode)
387	return 0;
388
389	if (!ext4_inode_block_valid(inode, start_blk: map->m_pblk, count: map->m_len)) {
390	ext4_error_inode(inode, func, line, map->m_pblk,
391	"lblock %lu mapped to illegal pblock %llu "
392	"(length %d)", (unsigned long) map->m_lblk,
393	map->m_pblk, map->m_len);
394	return -EFSCORRUPTED;
395	}
396	return 0;
397	}
398
399	int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
400	ext4_lblk_t len)
401	{
402	int ret;
403
404	if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
405	return fscrypt_zeroout_range(inode, lblk, pblk, len);
406
407	ret = sb_issue_zeroout(sb: inode->i_sb, block: pblk, nr_blocks: len, GFP_NOFS);
408	if (ret > 0)
409	ret = 0;
410
411	return ret;
412	}
413
414	/*
415	* For generic regular files, when updating the extent tree, Ext4 should
416	* hold the i_rwsem and invalidate_lock exclusively. This ensures
417	* exclusion against concurrent page faults, as well as reads and writes.
418	*/
419	#ifdef CONFIG_EXT4_DEBUG
420	void ext4_check_map_extents_env(struct inode *inode)
421	{
422	if (EXT4_SB(sb: inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
423	return;
424
425	if (!S_ISREG(inode->i_mode) ||
426	IS_NOQUOTA(inode) || IS_VERITY(inode) ||
427	is_special_ino(sb: inode->i_sb, ino: inode->i_ino) ||
428	(inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW)) ||
429	ext4_test_inode_flag(inode, bit: EXT4_INODE_EA_INODE) ||
430	ext4_verity_in_progress(inode))
431	return;
432
433	WARN_ON_ONCE(!inode_is_locked(inode) &&
434	!rwsem_is_locked(&inode->i_mapping->invalidate_lock));
435	}
436	#else
437	void ext4_check_map_extents_env(struct inode *inode) {}
438	#endif
439
440	#define check_block_validity(inode, map) \
441	__check_block_validity((inode), __func__, __LINE__, (map))
442
443	#ifdef ES_AGGRESSIVE_TEST
444	static void ext4_map_blocks_es_recheck(handle_t *handle,
445	struct inode *inode,
446	struct ext4_map_blocks *es_map,
447	struct ext4_map_blocks *map,
448	int flags)
449	{
450	int retval;
451
452	map->m_flags = 0;
453	/*
454	* There is a race window that the result is not the same.
455	* e.g. xfstests #223 when dioread_nolock enables. The reason
456	* is that we lookup a block mapping in extent status tree with
457	* out taking i_data_sem. So at the time the unwritten extent
458	* could be converted.
459	*/
460	down_read(&EXT4_I(inode)->i_data_sem);
461	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
462	retval = ext4_ext_map_blocks(handle, inode, map, 0);
463	} else {
464	retval = ext4_ind_map_blocks(handle, inode, map, 0);
465	}
466	up_read((&EXT4_I(inode)->i_data_sem));
467
468	/*
469	* We don't check m_len because extent will be collpased in status
470	* tree. So the m_len might not equal.
471	*/
472	if (es_map->m_lblk != map->m_lblk ||
473	es_map->m_flags != map->m_flags ||
474	es_map->m_pblk != map->m_pblk) {
475	printk("ES cache assertion failed for inode: %lu "
476	"es_cached ex [%d/%d/%llu/%x] != "
477	"found ex [%d/%d/%llu/%x] retval %d flags %x\n",
478	inode->i_ino, es_map->m_lblk, es_map->m_len,
479	es_map->m_pblk, es_map->m_flags, map->m_lblk,
480	map->m_len, map->m_pblk, map->m_flags,
481	retval, flags);
482	}
483	}
484	#endif /* ES_AGGRESSIVE_TEST */
485
486	static int ext4_map_query_blocks_next_in_leaf(handle_t *handle,
487	struct inode *inode, struct ext4_map_blocks *map,
488	unsigned int orig_mlen)
489	{
490	struct ext4_map_blocks map2;
491	unsigned int status, status2;
492	int retval;
493
494	status = map->m_flags & EXT4_MAP_UNWRITTEN ?
495	EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
496
497	WARN_ON_ONCE(!(map->m_flags & EXT4_MAP_QUERY_LAST_IN_LEAF));
498	WARN_ON_ONCE(orig_mlen <= map->m_len);
499
500	/* Prepare map2 for lookup in next leaf block */
501	map2.m_lblk = map->m_lblk + map->m_len;
502	map2.m_len = orig_mlen - map->m_len;
503	map2.m_flags = 0;
504	retval = ext4_ext_map_blocks(handle, inode, map: &map2, flags: 0);
505
506	if (retval <= 0) {
507	ext4_es_insert_extent(inode, lblk: map->m_lblk, len: map->m_len,
508	pblk: map->m_pblk, status, delalloc_reserve_used: false);
509	return map->m_len;
510	}
511
512	if (unlikely(retval != map2.m_len)) {
513	ext4_warning(inode->i_sb,
514	"ES len assertion failed for inode "
515	"%lu: retval %d != map->m_len %d",
516	inode->i_ino, retval, map2.m_len);
517	WARN_ON(1);
518	}
519
520	status2 = map2.m_flags & EXT4_MAP_UNWRITTEN ?
521	EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
522
523	/*
524	* If map2 is contiguous with map, then let's insert it as a single
525	* extent in es cache and return the combined length of both the maps.
526	*/
527	if (map->m_pblk + map->m_len == map2.m_pblk &&
528	status == status2) {
529	ext4_es_insert_extent(inode, lblk: map->m_lblk,
530	len: map->m_len + map2.m_len, pblk: map->m_pblk,
531	status, delalloc_reserve_used: false);
532	map->m_len += map2.m_len;
533	} else {
534	ext4_es_insert_extent(inode, lblk: map->m_lblk, len: map->m_len,
535	pblk: map->m_pblk, status, delalloc_reserve_used: false);
536	}
537
538	return map->m_len;
539	}
540
541	static int ext4_map_query_blocks(handle_t *handle, struct inode *inode,
542	struct ext4_map_blocks *map, int flags)
543	{
544	unsigned int status;
545	int retval;
546	unsigned int orig_mlen = map->m_len;
547
548	flags &= EXT4_EX_QUERY_FILTER;
549	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
550	retval = ext4_ext_map_blocks(handle, inode, map, flags);
551	else
552	retval = ext4_ind_map_blocks(handle, inode, map, flags);
553
554	if (retval <= 0)
555	return retval;
556
557	if (unlikely(retval != map->m_len)) {
558	ext4_warning(inode->i_sb,
559	"ES len assertion failed for inode "
560	"%lu: retval %d != map->m_len %d",
561	inode->i_ino, retval, map->m_len);
562	WARN_ON(1);
563	}
564
565	/*
566	* No need to query next in leaf:
567	* - if returned extent is not last in leaf or
568	* - if the last in leaf is the full requested range
569	*/
570	if (!(map->m_flags & EXT4_MAP_QUERY_LAST_IN_LEAF) ||
571	map->m_len == orig_mlen) {
572	status = map->m_flags & EXT4_MAP_UNWRITTEN ?
573	EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
574	ext4_es_insert_extent(inode, lblk: map->m_lblk, len: map->m_len,
575	pblk: map->m_pblk, status, delalloc_reserve_used: false);
576	return retval;
577	}
578
579	return ext4_map_query_blocks_next_in_leaf(handle, inode, map,
580	orig_mlen);
581	}
582
583	static int ext4_map_create_blocks(handle_t *handle, struct inode *inode,
584	struct ext4_map_blocks *map, int flags)
585	{
586	struct extent_status es;
587	unsigned int status;
588	int err, retval = 0;
589
590	/*
591	* We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE
592	* indicates that the blocks and quotas has already been
593	* checked when the data was copied into the page cache.
594	*/
595	if (map->m_flags & EXT4_MAP_DELAYED)
596	flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
597
598	/*
599	* Here we clear m_flags because after allocating an new extent,
600	* it will be set again.
601	*/
602	map->m_flags &= ~EXT4_MAP_FLAGS;
603
604	/*
605	* We need to check for EXT4 here because migrate could have
606	* changed the inode type in between.
607	*/
608	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS)) {
609	retval = ext4_ext_map_blocks(handle, inode, map, flags);
610	} else {
611	retval = ext4_ind_map_blocks(handle, inode, map, flags);
612
613	/*
614	* We allocated new blocks which will result in i_data's
615	* format changing. Force the migrate to fail by clearing
616	* migrate flags.
617	*/
618	if (retval > 0 && map->m_flags & EXT4_MAP_NEW)
619	ext4_clear_inode_state(inode, bit: EXT4_STATE_EXT_MIGRATE);
620	}
621	if (retval <= 0)
622	return retval;
623
624	if (unlikely(retval != map->m_len)) {
625	ext4_warning(inode->i_sb,
626	"ES len assertion failed for inode %lu: "
627	"retval %d != map->m_len %d",
628	inode->i_ino, retval, map->m_len);
629	WARN_ON(1);
630	}
631
632	/*
633	* We have to zeroout blocks before inserting them into extent
634	* status tree. Otherwise someone could look them up there and
635	* use them before they are really zeroed. We also have to
636	* unmap metadata before zeroing as otherwise writeback can
637	* overwrite zeros with stale data from block device.
638	*/
639	if (flags & EXT4_GET_BLOCKS_ZERO &&
640	map->m_flags & EXT4_MAP_MAPPED && map->m_flags & EXT4_MAP_NEW) {
641	err = ext4_issue_zeroout(inode, lblk: map->m_lblk, pblk: map->m_pblk,
642	len: map->m_len);
643	if (err)
644	return err;
645	}
646
647	/*
648	* If the extent has been zeroed out, we don't need to update
649	* extent status tree.
650	*/
651	if (flags & EXT4_GET_BLOCKS_PRE_IO &&
652	ext4_es_lookup_extent(inode, lblk: map->m_lblk, NULL, es: &es)) {
653	if (ext4_es_is_written(es: &es))
654	return retval;
655	}
656
657	status = map->m_flags & EXT4_MAP_UNWRITTEN ?
658	EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
659	ext4_es_insert_extent(inode, lblk: map->m_lblk, len: map->m_len, pblk: map->m_pblk,
660	status, delalloc_reserve_used: flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE);
661
662	return retval;
663	}
664
665	/*
666	* The ext4_map_blocks() function tries to look up the requested blocks,
667	* and returns if the blocks are already mapped.
668	*
669	* Otherwise it takes the write lock of the i_data_sem and allocate blocks
670	* and store the allocated blocks in the result buffer head and mark it
671	* mapped.
672	*
673	* If file type is extents based, it will call ext4_ext_map_blocks(),
674	* Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
675	* based files
676	*
677	* On success, it returns the number of blocks being mapped or allocated.
678	* If flags doesn't contain EXT4_GET_BLOCKS_CREATE the blocks are
679	* pre-allocated and unwritten, the resulting @map is marked as unwritten.
680	* If the flags contain EXT4_GET_BLOCKS_CREATE, it will mark @map as mapped.
681	*
682	* It returns 0 if plain look up failed (blocks have not been allocated), in
683	* that case, @map is returned as unmapped but we still do fill map->m_len to
684	* indicate the length of a hole starting at map->m_lblk.
685	*
686	* It returns the error in case of allocation failure.
687	*/
688	int ext4_map_blocks(handle_t *handle, struct inode *inode,
689	struct ext4_map_blocks *map, int flags)
690	{
691	struct extent_status es;
692	int retval;
693	int ret = 0;
694	unsigned int orig_mlen = map->m_len;
695	#ifdef ES_AGGRESSIVE_TEST
696	struct ext4_map_blocks orig_map;
697
698	memcpy(&orig_map, map, sizeof(*map));
699	#endif
700
701	map->m_flags = 0;
702	ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
703	flags, map->m_len, (unsigned long) map->m_lblk);
704
705	/*
706	* ext4_map_blocks returns an int, and m_len is an unsigned int
707	*/
708	if (unlikely(map->m_len > INT_MAX))
709	map->m_len = INT_MAX;
710
711	/* We can handle the block number less than EXT_MAX_BLOCKS */
712	if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
713	return -EFSCORRUPTED;
714
715	/*
716	* Callers from the context of data submission are the only exceptions
717	* for regular files that do not hold the i_rwsem or invalidate_lock.
718	* However, caching unrelated ranges is not permitted.
719	*/
720	if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
721	WARN_ON_ONCE(!(flags & EXT4_EX_NOCACHE));
722	else
723	ext4_check_map_extents_env(inode);
724
725	/* Lookup extent status tree firstly */
726	if (!(EXT4_SB(sb: inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
727	ext4_es_lookup_extent(inode, lblk: map->m_lblk, NULL, es: &es)) {
728	if (ext4_es_is_written(es: &es) || ext4_es_is_unwritten(es: &es)) {
729	map->m_pblk = ext4_es_pblock(es: &es) +
730	map->m_lblk - es.es_lblk;
731	map->m_flags |= ext4_es_is_written(es: &es) ?
732	EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
733	retval = es.es_len - (map->m_lblk - es.es_lblk);
734	if (retval > map->m_len)
735	retval = map->m_len;
736	map->m_len = retval;
737	} else if (ext4_es_is_delayed(es: &es) || ext4_es_is_hole(es: &es)) {
738	map->m_pblk = 0;
739	map->m_flags |= ext4_es_is_delayed(es: &es) ?
740	EXT4_MAP_DELAYED : 0;
741	retval = es.es_len - (map->m_lblk - es.es_lblk);
742	if (retval > map->m_len)
743	retval = map->m_len;
744	map->m_len = retval;
745	retval = 0;
746	} else {
747	BUG();
748	}
749
750	if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
751	return retval;
752	#ifdef ES_AGGRESSIVE_TEST
753	ext4_map_blocks_es_recheck(handle, inode, map,
754	&orig_map, flags);
755	#endif
756	if (!(flags & EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF) ||
757	orig_mlen == map->m_len)
758	goto found;
759
760	if (flags & EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF)
761	map->m_len = orig_mlen;
762	}
763	/*
764	* In the query cache no-wait mode, nothing we can do more if we
765	* cannot find extent in the cache.
766	*/
767	if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
768	return 0;
769
770	/*
771	* Try to see if we can get the block without requesting a new
772	* file system block.
773	*/
774	down_read(sem: &EXT4_I(inode)->i_data_sem);
775	retval = ext4_map_query_blocks(handle, inode, map, flags);
776	up_read(sem: (&EXT4_I(inode)->i_data_sem));
777
778	found:
779	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
780	ret = check_block_validity(inode, map);
781	if (ret != 0)
782	return ret;
783	}
784
785	/* If it is only a block(s) look up */
786	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
787	return retval;
788
789	/*
790	* Returns if the blocks have already allocated
791	*
792	* Note that if blocks have been preallocated
793	* ext4_ext_map_blocks() returns with buffer head unmapped
794	*/
795	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
796	/*
797	* If we need to convert extent to unwritten
798	* we continue and do the actual work in
799	* ext4_ext_map_blocks()
800	*/
801	if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
802	return retval;
803
804
805	ext4_fc_track_inode(handle, inode);
806	/*
807	* New blocks allocate and/or writing to unwritten extent
808	* will possibly result in updating i_data, so we take
809	* the write lock of i_data_sem, and call get_block()
810	* with create == 1 flag.
811	*/
812	down_write(sem: &EXT4_I(inode)->i_data_sem);
813	retval = ext4_map_create_blocks(handle, inode, map, flags);
814	up_write(sem: (&EXT4_I(inode)->i_data_sem));
815	if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
816	ret = check_block_validity(inode, map);
817	if (ret != 0)
818	return ret;
819
820	/*
821	* Inodes with freshly allocated blocks where contents will be
822	* visible after transaction commit must be on transaction's
823	* ordered data list.
824	*/
825	if (map->m_flags & EXT4_MAP_NEW &&
826	!(map->m_flags & EXT4_MAP_UNWRITTEN) &&
827	!(flags & EXT4_GET_BLOCKS_ZERO) &&
828	!ext4_is_quota_file(inode) &&
829	ext4_should_order_data(inode)) {
830	loff_t start_byte =
831	(loff_t)map->m_lblk << inode->i_blkbits;
832	loff_t length = (loff_t)map->m_len << inode->i_blkbits;
833
834	if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
835	ret = ext4_jbd2_inode_add_wait(handle, inode,
836	start_byte, length);
837	else
838	ret = ext4_jbd2_inode_add_write(handle, inode,
839	start_byte, length);
840	if (ret)
841	return ret;
842	}
843	}
844	if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
845	map->m_flags & EXT4_MAP_MAPPED))
846	ext4_fc_track_range(handle, inode, start: map->m_lblk,
847	end: map->m_lblk + map->m_len - 1);
848	if (retval < 0)
849	ext_debug(inode, "failed with err %d\n", retval);
850	return retval;
851	}
852
853	/*
854	* Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
855	* we have to be careful as someone else may be manipulating b_state as well.
856	*/
857	static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
858	{
859	unsigned long old_state;
860	unsigned long new_state;
861
862	flags &= EXT4_MAP_FLAGS;
863
864	/* Dummy buffer_head? Set non-atomically. */
865	if (!bh->b_folio) {
866	bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
867	return;
868	}
869	/*
870	* Someone else may be modifying b_state. Be careful! This is ugly but
871	* once we get rid of using bh as a container for mapping information
872	* to pass to / from get_block functions, this can go away.
873	*/
874	old_state = READ_ONCE(bh->b_state);
875	do {
876	new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
877	} while (unlikely(!try_cmpxchg(&bh->b_state, &old_state, new_state)));
878	}
879
880	static int _ext4_get_block(struct inode *inode, sector_t iblock,
881	struct buffer_head *bh, int flags)
882	{
883	struct ext4_map_blocks map;
884	int ret = 0;
885
886	if (ext4_has_inline_data(inode))
887	return -ERANGE;
888
889	map.m_lblk = iblock;
890	map.m_len = bh->b_size >> inode->i_blkbits;
891
892	ret = ext4_map_blocks(handle: ext4_journal_current_handle(), inode, map: &map,
893	flags);
894	if (ret > 0) {
895	map_bh(bh, sb: inode->i_sb, block: map.m_pblk);
896	ext4_update_bh_state(bh, flags: map.m_flags);
897	bh->b_size = inode->i_sb->s_blocksize * map.m_len;
898	ret = 0;
899	} else if (ret == 0) {
900	/* hole case, need to fill in bh->b_size */
901	bh->b_size = inode->i_sb->s_blocksize * map.m_len;
902	}
903	return ret;
904	}
905
906	int ext4_get_block(struct inode *inode, sector_t iblock,
907	struct buffer_head *bh, int create)
908	{
909	return _ext4_get_block(inode, iblock, bh,
910	flags: create ? EXT4_GET_BLOCKS_CREATE : 0);
911	}
912
913	/*
914	* Get block function used when preparing for buffered write if we require
915	* creating an unwritten extent if blocks haven't been allocated. The extent
916	* will be converted to written after the IO is complete.
917	*/
918	int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
919	struct buffer_head *bh_result, int create)
920	{
921	int ret = 0;
922
923	ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
924	inode->i_ino, create);
925	ret = _ext4_get_block(inode, iblock, bh: bh_result,
926	EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
927
928	/*
929	* If the buffer is marked unwritten, mark it as new to make sure it is
930	* zeroed out correctly in case of partial writes. Otherwise, there is
931	* a chance of stale data getting exposed.
932	*/
933	if (ret == 0 && buffer_unwritten(bh: bh_result))
934	set_buffer_new(bh_result);
935
936	return ret;
937	}
938
939	/* Maximum number of blocks we map for direct IO at once. */
940	#define DIO_MAX_BLOCKS 4096
941
942	/*
943	* `handle' can be NULL if create is zero
944	*/
945	struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
946	ext4_lblk_t block, int map_flags)
947	{
948	struct ext4_map_blocks map;
949	struct buffer_head *bh;
950	int create = map_flags & EXT4_GET_BLOCKS_CREATE;
951	bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
952	int err;
953
954	ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
955	|| handle != NULL || create == 0);
956	ASSERT(create == 0 || !nowait);
957
958	map.m_lblk = block;
959	map.m_len = 1;
960	err = ext4_map_blocks(handle, inode, map: &map, flags: map_flags);
961
962	if (err == 0)
963	return create ? ERR_PTR(error: -ENOSPC) : NULL;
964	if (err < 0)
965	return ERR_PTR(error: err);
966
967	if (nowait)
968	return sb_find_get_block(sb: inode->i_sb, block: map.m_pblk);
969
970	/*
971	* Since bh could introduce extra ref count such as referred by
972	* journal_head etc. Try to avoid using __GFP_MOVABLE here
973	* as it may fail the migration when journal_head remains.
974	*/
975	bh = getblk_unmovable(bdev: inode->i_sb->s_bdev, block: map.m_pblk,
976	size: inode->i_sb->s_blocksize);
977
978	if (unlikely(!bh))
979	return ERR_PTR(error: -ENOMEM);
980	if (map.m_flags & EXT4_MAP_NEW) {
981	ASSERT(create != 0);
982	ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
983	|| (handle != NULL));
984
985	/*
986	* Now that we do not always journal data, we should
987	* keep in mind whether this should always journal the
988	* new buffer as metadata. For now, regular file
989	* writes use ext4_get_block instead, so it's not a
990	* problem.
991	*/
992	lock_buffer(bh);
993	BUFFER_TRACE(bh, "call get_create_access");
994	err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
995	EXT4_JTR_NONE);
996	if (unlikely(err)) {
997	unlock_buffer(bh);
998	goto errout;
999	}
1000	if (!buffer_uptodate(bh)) {
1001	memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1002	set_buffer_uptodate(bh);
1003	}
1004	unlock_buffer(bh);
1005	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1006	err = ext4_handle_dirty_metadata(handle, inode, bh);
1007	if (unlikely(err))
1008	goto errout;
1009	} else
1010	BUFFER_TRACE(bh, "not a new buffer");
1011	return bh;
1012	errout:
1013	brelse(bh);
1014	return ERR_PTR(error: err);
1015	}
1016
1017	struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1018	ext4_lblk_t block, int map_flags)
1019	{
1020	struct buffer_head *bh;
1021	int ret;
1022
1023	bh = ext4_getblk(handle, inode, block, map_flags);
1024	if (IS_ERR(ptr: bh))
1025	return bh;
1026	if (!bh || ext4_buffer_uptodate(bh))
1027	return bh;
1028
1029	ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, wait: true);
1030	if (ret) {
1031	put_bh(bh);
1032	return ERR_PTR(error: ret);
1033	}
1034	return bh;
1035	}
1036
1037	/* Read a contiguous batch of blocks. */
1038	int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1039	bool wait, struct buffer_head **bhs)
1040	{
1041	int i, err;
1042
1043	for (i = 0; i < bh_count; i++) {
1044	bhs[i] = ext4_getblk(NULL, inode, block: block + i, map_flags: 0 /* map_flags */);
1045	if (IS_ERR(ptr: bhs[i])) {
1046	err = PTR_ERR(ptr: bhs[i]);
1047	bh_count = i;
1048	goto out_brelse;
1049	}
1050	}
1051
1052	for (i = 0; i < bh_count; i++)
1053	/* Note that NULL bhs[i] is valid because of holes. */
1054	if (bhs[i] && !ext4_buffer_uptodate(bh: bhs[i]))
1055	ext4_read_bh_lock(bh: bhs[i], REQ_META | REQ_PRIO, wait: false);
1056
1057	if (!wait)
1058	return 0;
1059
1060	for (i = 0; i < bh_count; i++)
1061	if (bhs[i])
1062	wait_on_buffer(bh: bhs[i]);
1063
1064	for (i = 0; i < bh_count; i++) {
1065	if (bhs[i] && !buffer_uptodate(bh: bhs[i])) {
1066	err = -EIO;
1067	goto out_brelse;
1068	}
1069	}
1070	return 0;
1071
1072	out_brelse:
1073	for (i = 0; i < bh_count; i++) {
1074	brelse(bh: bhs[i]);
1075	bhs[i] = NULL;
1076	}
1077	return err;
1078	}
1079
1080	int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
1081	struct buffer_head *head,
1082	unsigned from,
1083	unsigned to,
1084	int *partial,
1085	int (*fn)(handle_t *handle, struct inode *inode,
1086	struct buffer_head *bh))
1087	{
1088	struct buffer_head *bh;
1089	unsigned block_start, block_end;
1090	unsigned blocksize = head->b_size;
1091	int err, ret = 0;
1092	struct buffer_head *next;
1093
1094	for (bh = head, block_start = 0;
1095	ret == 0 && (bh != head || !block_start);
1096	block_start = block_end, bh = next) {
1097	next = bh->b_this_page;
1098	block_end = block_start + blocksize;
1099	if (block_end <= from || block_start >= to) {
1100	if (partial && !buffer_uptodate(bh))
1101	*partial = 1;
1102	continue;
1103	}
1104	err = (*fn)(handle, inode, bh);
1105	if (!ret)
1106	ret = err;
1107	}
1108	return ret;
1109	}
1110
1111	/*
1112	* Helper for handling dirtying of journalled data. We also mark the folio as
1113	* dirty so that writeback code knows about this page (and inode) contains
1114	* dirty data. ext4_writepages() then commits appropriate transaction to
1115	* make data stable.
1116	*/
1117	static int ext4_dirty_journalled_data(handle_t *handle, struct buffer_head *bh)
1118	{
1119	struct folio *folio = bh->b_folio;
1120	struct inode *inode = folio->mapping->host;
1121
1122	/* only regular files have a_ops */
1123	if (S_ISREG(inode->i_mode))
1124	folio_mark_dirty(folio);
1125	return ext4_handle_dirty_metadata(handle, NULL, bh);
1126	}
1127
1128	int do_journal_get_write_access(handle_t *handle, struct inode *inode,
1129	struct buffer_head *bh)
1130	{
1131	if (!buffer_mapped(bh) || buffer_freed(bh))
1132	return 0;
1133	BUFFER_TRACE(bh, "get write access");
1134	return ext4_journal_get_write_access(handle, inode->i_sb, bh,
1135	EXT4_JTR_NONE);
1136	}
1137
1138	int ext4_block_write_begin(handle_t *handle, struct folio *folio,
1139	loff_t pos, unsigned len,
1140	get_block_t *get_block)
1141	{
1142	unsigned int from = offset_in_folio(folio, pos);
1143	unsigned to = from + len;
1144	struct inode *inode = folio->mapping->host;
1145	unsigned block_start, block_end;
1146	sector_t block;
1147	int err = 0;
1148	unsigned blocksize = inode->i_sb->s_blocksize;
1149	unsigned bbits;
1150	struct buffer_head *bh, *head, *wait[2];
1151	int nr_wait = 0;
1152	int i;
1153	bool should_journal_data = ext4_should_journal_data(inode);
1154
1155	BUG_ON(!folio_test_locked(folio));
1156	BUG_ON(to > folio_size(folio));
1157	BUG_ON(from > to);
1158
1159	head = folio_buffers(folio);
1160	if (!head)
1161	head = create_empty_buffers(folio, blocksize, b_state: 0);
1162	bbits = ilog2(blocksize);
1163	block = (sector_t)folio->index << (PAGE_SHIFT - bbits);
1164
1165	for (bh = head, block_start = 0; bh != head || !block_start;
1166	block++, block_start = block_end, bh = bh->b_this_page) {
1167	block_end = block_start + blocksize;
1168	if (block_end <= from || block_start >= to) {
1169	if (folio_test_uptodate(folio)) {
1170	set_buffer_uptodate(bh);
1171	}
1172	continue;
1173	}
1174	if (buffer_new(bh))
1175	clear_buffer_new(bh);
1176	if (!buffer_mapped(bh)) {
1177	WARN_ON(bh->b_size != blocksize);
1178	err = get_block(inode, block, bh, 1);
1179	if (err)
1180	break;
1181	if (buffer_new(bh)) {
1182	/*
1183	* We may be zeroing partial buffers or all new
1184	* buffers in case of failure. Prepare JBD2 for
1185	* that.
1186	*/
1187	if (should_journal_data)
1188	do_journal_get_write_access(handle,
1189	inode, bh);
1190	if (folio_test_uptodate(folio)) {
1191	/*
1192	* Unlike __block_write_begin() we leave
1193	* dirtying of new uptodate buffers to
1194	* ->write_end() time or
1195	* folio_zero_new_buffers().
1196	*/
1197	set_buffer_uptodate(bh);
1198	continue;
1199	}
1200	if (block_end > to || block_start < from)
1201	folio_zero_segments(folio, start1: to,
1202	xend1: block_end,
1203	start2: block_start, xend2: from);
1204	continue;
1205	}
1206	}
1207	if (folio_test_uptodate(folio)) {
1208	set_buffer_uptodate(bh);
1209	continue;
1210	}
1211	if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1212	!buffer_unwritten(bh) &&
1213	(block_start < from || block_end > to)) {
1214	ext4_read_bh_lock(bh, op_flags: 0, wait: false);
1215	wait[nr_wait++] = bh;
1216	}
1217	}
1218	/*
1219	* If we issued read requests, let them complete.
1220	*/
1221	for (i = 0; i < nr_wait; i++) {
1222	wait_on_buffer(bh: wait[i]);
1223	if (!buffer_uptodate(bh: wait[i]))
1224	err = -EIO;
1225	}
1226	if (unlikely(err)) {
1227	if (should_journal_data)
1228	ext4_journalled_zero_new_buffers(handle, inode, folio,
1229	from, to);
1230	else
1231	folio_zero_new_buffers(folio, from, to);
1232	} else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1233	for (i = 0; i < nr_wait; i++) {
1234	int err2;
1235
1236	err2 = fscrypt_decrypt_pagecache_blocks(folio,
1237	len: blocksize, offs: bh_offset(bh: wait[i]));
1238	if (err2) {
1239	clear_buffer_uptodate(bh: wait[i]);
1240	err = err2;
1241	}
1242	}
1243	}
1244
1245	return err;
1246	}
1247
1248	/*
1249	* To preserve ordering, it is essential that the hole instantiation and
1250	* the data write be encapsulated in a single transaction. We cannot
1251	* close off a transaction and start a new one between the ext4_get_block()
1252	* and the ext4_write_end(). So doing the jbd2_journal_start at the start of
1253	* ext4_write_begin() is the right place.
1254	*/
1255	static int ext4_write_begin(struct file *file, struct address_space *mapping,
1256	loff_t pos, unsigned len,
1257	struct folio **foliop, void **fsdata)
1258	{
1259	struct inode *inode = mapping->host;
1260	int ret, needed_blocks;
1261	handle_t *handle;
1262	int retries = 0;
1263	struct folio *folio;
1264	pgoff_t index;
1265	unsigned from, to;
1266	fgf_t fgp = FGP_WRITEBEGIN;
1267
1268	ret = ext4_emergency_state(sb: inode->i_sb);
1269	if (unlikely(ret))
1270	return ret;
1271
1272	trace_ext4_write_begin(inode, pos, len);
1273	/*
1274	* Reserve one block more for addition to orphan list in case
1275	* we allocate blocks but write fails for some reason
1276	*/
1277	needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1278	index = pos >> PAGE_SHIFT;
1279
1280	if (ext4_test_inode_state(inode, bit: EXT4_STATE_MAY_INLINE_DATA)) {
1281	ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1282	foliop);
1283	if (ret < 0)
1284	return ret;
1285	if (ret == 1)
1286	return 0;
1287	}
1288
1289	/*
1290	* __filemap_get_folio() can take a long time if the
1291	* system is thrashing due to memory pressure, or if the folio
1292	* is being written back. So grab it first before we start
1293	* the transaction handle. This also allows us to allocate
1294	* the folio (if needed) without using GFP_NOFS.
1295	*/
1296	retry_grab:
1297	fgp |= fgf_set_order(size: len);
1298	folio = __filemap_get_folio(mapping, index, fgp_flags: fgp,
1299	gfp: mapping_gfp_mask(mapping));
1300	if (IS_ERR(ptr: folio))
1301	return PTR_ERR(ptr: folio);
1302
1303	if (pos + len > folio_pos(folio) + folio_size(folio))
1304	len = folio_pos(folio) + folio_size(folio) - pos;
1305
1306	from = offset_in_folio(folio, pos);
1307	to = from + len;
1308
1309	/*
1310	* The same as page allocation, we prealloc buffer heads before
1311	* starting the handle.
1312	*/
1313	if (!folio_buffers(folio))
1314	create_empty_buffers(folio, blocksize: inode->i_sb->s_blocksize, b_state: 0);
1315
1316	folio_unlock(folio);
1317
1318	retry_journal:
1319	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1320	if (IS_ERR(ptr: handle)) {
1321	folio_put(folio);
1322	return PTR_ERR(ptr: handle);
1323	}
1324
1325	folio_lock(folio);
1326	if (folio->mapping != mapping) {
1327	/* The folio got truncated from under us */
1328	folio_unlock(folio);
1329	folio_put(folio);
1330	ext4_journal_stop(handle);
1331	goto retry_grab;
1332	}
1333	/* In case writeback began while the folio was unlocked */
1334	folio_wait_stable(folio);
1335
1336	if (ext4_should_dioread_nolock(inode))
1337	ret = ext4_block_write_begin(handle, folio, pos, len,
1338	get_block: ext4_get_block_unwritten);
1339	else
1340	ret = ext4_block_write_begin(handle, folio, pos, len,
1341	get_block: ext4_get_block);
1342	if (!ret && ext4_should_journal_data(inode)) {
1343	ret = ext4_walk_page_buffers(handle, inode,
1344	folio_buffers(folio), from, to,
1345	NULL, fn: do_journal_get_write_access);
1346	}
1347
1348	if (ret) {
1349	bool extended = (pos + len > inode->i_size) &&
1350	!ext4_verity_in_progress(inode);
1351
1352	folio_unlock(folio);
1353	/*
1354	* ext4_block_write_begin may have instantiated a few blocks
1355	* outside i_size. Trim these off again. Don't need
1356	* i_size_read because we hold i_rwsem.
1357	*
1358	* Add inode to orphan list in case we crash before
1359	* truncate finishes
1360	*/
1361	if (extended && ext4_can_truncate(inode))
1362	ext4_orphan_add(handle, inode);
1363
1364	ext4_journal_stop(handle);
1365	if (extended) {
1366	ext4_truncate_failed_write(inode);
1367	/*
1368	* If truncate failed early the inode might
1369	* still be on the orphan list; we need to
1370	* make sure the inode is removed from the
1371	* orphan list in that case.
1372	*/
1373	if (inode->i_nlink)
1374	ext4_orphan_del(NULL, inode);
1375	}
1376
1377	if (ret == -ENOSPC &&
1378	ext4_should_retry_alloc(sb: inode->i_sb, retries: &retries))
1379	goto retry_journal;
1380	folio_put(folio);
1381	return ret;
1382	}
1383	*foliop = folio;
1384	return ret;
1385	}
1386
1387	/* For write_end() in data=journal mode */
1388	static int write_end_fn(handle_t *handle, struct inode *inode,
1389	struct buffer_head *bh)
1390	{
1391	int ret;
1392	if (!buffer_mapped(bh) || buffer_freed(bh))
1393	return 0;
1394	set_buffer_uptodate(bh);
1395	ret = ext4_dirty_journalled_data(handle, bh);
1396	clear_buffer_meta(bh);
1397	clear_buffer_prio(bh);
1398	return ret;
1399	}
1400
1401	/*
1402	* We need to pick up the new inode size which generic_commit_write gave us
1403	* `file' can be NULL - eg, when called from page_symlink().
1404	*
1405	* ext4 never places buffers on inode->i_mapping->i_private_list. metadata
1406	* buffers are managed internally.
1407	*/
1408	static int ext4_write_end(struct file *file,
1409	struct address_space *mapping,
1410	loff_t pos, unsigned len, unsigned copied,
1411	struct folio *folio, void *fsdata)
1412	{
1413	handle_t *handle = ext4_journal_current_handle();
1414	struct inode *inode = mapping->host;
1415	loff_t old_size = inode->i_size;
1416	int ret = 0, ret2;
1417	int i_size_changed = 0;
1418	bool verity = ext4_verity_in_progress(inode);
1419
1420	trace_ext4_write_end(inode, pos, len, copied);
1421
1422	if (ext4_has_inline_data(inode) &&
1423	ext4_test_inode_state(inode, bit: EXT4_STATE_MAY_INLINE_DATA))
1424	return ext4_write_inline_data_end(inode, pos, len, copied,
1425	folio);
1426
1427	copied = block_write_end(file, mapping, pos, len, copied, folio, fsdata);
1428	/*
1429	* it's important to update i_size while still holding folio lock:
1430	* page writeout could otherwise come in and zero beyond i_size.
1431	*
1432	* If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1433	* blocks are being written past EOF, so skip the i_size update.
1434	*/
1435	if (!verity)
1436	i_size_changed = ext4_update_inode_size(inode, newsize: pos + copied);
1437	folio_unlock(folio);
1438	folio_put(folio);
1439
1440	if (old_size < pos && !verity) {
1441	pagecache_isize_extended(inode, from: old_size, to: pos);
1442	ext4_zero_partial_blocks(handle, inode, lstart: old_size, lend: pos - old_size);
1443	}
1444	/*
1445	* Don't mark the inode dirty under folio lock. First, it unnecessarily
1446	* makes the holding time of folio lock longer. Second, it forces lock
1447	* ordering of folio lock and transaction start for journaling
1448	* filesystems.
1449	*/
1450	if (i_size_changed)
1451	ret = ext4_mark_inode_dirty(handle, inode);
1452
1453	if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1454	/* if we have allocated more blocks and copied
1455	* less. We will have blocks allocated outside
1456	* inode->i_size. So truncate them
1457	*/
1458	ext4_orphan_add(handle, inode);
1459
1460	ret2 = ext4_journal_stop(handle);
1461	if (!ret)
1462	ret = ret2;
1463
1464	if (pos + len > inode->i_size && !verity) {
1465	ext4_truncate_failed_write(inode);
1466	/*
1467	* If truncate failed early the inode might still be
1468	* on the orphan list; we need to make sure the inode
1469	* is removed from the orphan list in that case.
1470	*/
1471	if (inode->i_nlink)
1472	ext4_orphan_del(NULL, inode);
1473	}
1474
1475	return ret ? ret : copied;
1476	}
1477
1478	/*
1479	* This is a private version of folio_zero_new_buffers() which doesn't
1480	* set the buffer to be dirty, since in data=journalled mode we need
1481	* to call ext4_dirty_journalled_data() instead.
1482	*/
1483	static void ext4_journalled_zero_new_buffers(handle_t *handle,
1484	struct inode *inode,
1485	struct folio *folio,
1486	unsigned from, unsigned to)
1487	{
1488	unsigned int block_start = 0, block_end;
1489	struct buffer_head *head, *bh;
1490
1491	bh = head = folio_buffers(folio);
1492	do {
1493	block_end = block_start + bh->b_size;
1494	if (buffer_new(bh)) {
1495	if (block_end > from && block_start < to) {
1496	if (!folio_test_uptodate(folio)) {
1497	unsigned start, size;
1498
1499	start = max(from, block_start);
1500	size = min(to, block_end) - start;
1501
1502	folio_zero_range(folio, start, length: size);
1503	}
1504	clear_buffer_new(bh);
1505	write_end_fn(handle, inode, bh);
1506	}
1507	}
1508	block_start = block_end;
1509	bh = bh->b_this_page;
1510	} while (bh != head);
1511	}
1512
1513	static int ext4_journalled_write_end(struct file *file,
1514	struct address_space *mapping,
1515	loff_t pos, unsigned len, unsigned copied,
1516	struct folio *folio, void *fsdata)
1517	{
1518	handle_t *handle = ext4_journal_current_handle();
1519	struct inode *inode = mapping->host;
1520	loff_t old_size = inode->i_size;
1521	int ret = 0, ret2;
1522	int partial = 0;
1523	unsigned from, to;
1524	int size_changed = 0;
1525	bool verity = ext4_verity_in_progress(inode);
1526
1527	trace_ext4_journalled_write_end(inode, pos, len, copied);
1528	from = pos & (PAGE_SIZE - 1);
1529	to = from + len;
1530
1531	BUG_ON(!ext4_handle_valid(handle));
1532
1533	if (ext4_has_inline_data(inode))
1534	return ext4_write_inline_data_end(inode, pos, len, copied,
1535	folio);
1536
1537	if (unlikely(copied < len) && !folio_test_uptodate(folio)) {
1538	copied = 0;
1539	ext4_journalled_zero_new_buffers(handle, inode, folio,
1540	from, to);
1541	} else {
1542	if (unlikely(copied < len))
1543	ext4_journalled_zero_new_buffers(handle, inode, folio,
1544	from: from + copied, to);
1545	ret = ext4_walk_page_buffers(handle, inode,
1546	folio_buffers(folio),
1547	from, to: from + copied, partial: &partial,
1548	fn: write_end_fn);
1549	if (!partial)
1550	folio_mark_uptodate(folio);
1551	}
1552	if (!verity)
1553	size_changed = ext4_update_inode_size(inode, newsize: pos + copied);
1554	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1555	folio_unlock(folio);
1556	folio_put(folio);
1557
1558	if (old_size < pos && !verity) {
1559	pagecache_isize_extended(inode, from: old_size, to: pos);
1560	ext4_zero_partial_blocks(handle, inode, lstart: old_size, lend: pos - old_size);
1561	}
1562
1563	if (size_changed) {
1564	ret2 = ext4_mark_inode_dirty(handle, inode);
1565	if (!ret)
1566	ret = ret2;
1567	}
1568
1569	if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1570	/* if we have allocated more blocks and copied
1571	* less. We will have blocks allocated outside
1572	* inode->i_size. So truncate them
1573	*/
1574	ext4_orphan_add(handle, inode);
1575
1576	ret2 = ext4_journal_stop(handle);
1577	if (!ret)
1578	ret = ret2;
1579	if (pos + len > inode->i_size && !verity) {
1580	ext4_truncate_failed_write(inode);
1581	/*
1582	* If truncate failed early the inode might still be
1583	* on the orphan list; we need to make sure the inode
1584	* is removed from the orphan list in that case.
1585	*/
1586	if (inode->i_nlink)
1587	ext4_orphan_del(NULL, inode);
1588	}
1589
1590	return ret ? ret : copied;
1591	}
1592
1593	/*
1594	* Reserve space for 'nr_resv' clusters
1595	*/
1596	static int ext4_da_reserve_space(struct inode *inode, int nr_resv)
1597	{
1598	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
1599	struct ext4_inode_info *ei = EXT4_I(inode);
1600	int ret;
1601
1602	/*
1603	* We will charge metadata quota at writeout time; this saves
1604	* us from metadata over-estimation, though we may go over by
1605	* a small amount in the end. Here we just reserve for data.
1606	*/
1607	ret = dquot_reserve_block(inode, EXT4_C2B(sbi, nr_resv));
1608	if (ret)
1609	return ret;
1610
1611	spin_lock(lock: &ei->i_block_reservation_lock);
1612	if (ext4_claim_free_clusters(sbi, nclusters: nr_resv, flags: 0)) {
1613	spin_unlock(lock: &ei->i_block_reservation_lock);
1614	dquot_release_reservation_block(inode, EXT4_C2B(sbi, nr_resv));
1615	return -ENOSPC;
1616	}
1617	ei->i_reserved_data_blocks += nr_resv;
1618	trace_ext4_da_reserve_space(inode, nr_resv);
1619	spin_unlock(lock: &ei->i_block_reservation_lock);
1620
1621	return 0; /* success */
1622	}
1623
1624	void ext4_da_release_space(struct inode *inode, int to_free)
1625	{
1626	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
1627	struct ext4_inode_info *ei = EXT4_I(inode);
1628
1629	if (!to_free)
1630	return; /* Nothing to release, exit */
1631
1632	spin_lock(lock: &EXT4_I(inode)->i_block_reservation_lock);
1633
1634	trace_ext4_da_release_space(inode, freed_blocks: to_free);
1635	if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1636	/*
1637	* if there aren't enough reserved blocks, then the
1638	* counter is messed up somewhere. Since this
1639	* function is called from invalidate page, it's
1640	* harmless to return without any action.
1641	*/
1642	ext4_warning(inode->i_sb, "ext4_da_release_space: "
1643	"ino %lu, to_free %d with only %d reserved "
1644	"data blocks", inode->i_ino, to_free,
1645	ei->i_reserved_data_blocks);
1646	WARN_ON(1);
1647	to_free = ei->i_reserved_data_blocks;
1648	}
1649	ei->i_reserved_data_blocks -= to_free;
1650
1651	/* update fs dirty data blocks counter */
1652	percpu_counter_sub(fbc: &sbi->s_dirtyclusters_counter, amount: to_free);
1653
1654	spin_unlock(lock: &EXT4_I(inode)->i_block_reservation_lock);
1655
1656	dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1657	}
1658
1659	/*
1660	* Delayed allocation stuff
1661	*/
1662
1663	struct mpage_da_data {
1664	/* These are input fields for ext4_do_writepages() */
1665	struct inode *inode;
1666	struct writeback_control *wbc;
1667	unsigned int can_map:1; /* Can writepages call map blocks? */
1668
1669	/* These are internal state of ext4_do_writepages() */
1670	pgoff_t first_page; /* The first page to write */
1671	pgoff_t next_page; /* Current page to examine */
1672	pgoff_t last_page; /* Last page to examine */
1673	/*
1674	* Extent to map - this can be after first_page because that can be
1675	* fully mapped. We somewhat abuse m_flags to store whether the extent
1676	* is delalloc or unwritten.
1677	*/
1678	struct ext4_map_blocks map;
1679	struct ext4_io_submit io_submit; /* IO submission data */
1680	unsigned int do_map:1;
1681	unsigned int scanned_until_end:1;
1682	unsigned int journalled_more_data:1;
1683	};
1684
1685	static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1686	bool invalidate)
1687	{
1688	unsigned nr, i;
1689	pgoff_t index, end;
1690	struct folio_batch fbatch;
1691	struct inode *inode = mpd->inode;
1692	struct address_space *mapping = inode->i_mapping;
1693
1694	/* This is necessary when next_page == 0. */
1695	if (mpd->first_page >= mpd->next_page)
1696	return;
1697
1698	mpd->scanned_until_end = 0;
1699	index = mpd->first_page;
1700	end = mpd->next_page - 1;
1701	if (invalidate) {
1702	ext4_lblk_t start, last;
1703	start = index << (PAGE_SHIFT - inode->i_blkbits);
1704	last = end << (PAGE_SHIFT - inode->i_blkbits);
1705
1706	/*
1707	* avoid racing with extent status tree scans made by
1708	* ext4_insert_delayed_block()
1709	*/
1710	down_write(sem: &EXT4_I(inode)->i_data_sem);
1711	ext4_es_remove_extent(inode, lblk: start, len: last - start + 1);
1712	up_write(sem: &EXT4_I(inode)->i_data_sem);
1713	}
1714
1715	folio_batch_init(fbatch: &fbatch);
1716	while (index <= end) {
1717	nr = filemap_get_folios(mapping, start: &index, end, fbatch: &fbatch);
1718	if (nr == 0)
1719	break;
1720	for (i = 0; i < nr; i++) {
1721	struct folio *folio = fbatch.folios[i];
1722
1723	if (folio->index < mpd->first_page)
1724	continue;
1725	if (folio_next_index(folio) - 1 > end)
1726	continue;
1727	BUG_ON(!folio_test_locked(folio));
1728	BUG_ON(folio_test_writeback(folio));
1729	if (invalidate) {
1730	if (folio_mapped(folio))
1731	folio_clear_dirty_for_io(folio);
1732	block_invalidate_folio(folio, offset: 0,
1733	length: folio_size(folio));
1734	folio_clear_uptodate(folio);
1735	}
1736	folio_unlock(folio);
1737	}
1738	folio_batch_release(fbatch: &fbatch);
1739	}
1740	}
1741
1742	static void ext4_print_free_blocks(struct inode *inode)
1743	{
1744	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
1745	struct super_block *sb = inode->i_sb;
1746	struct ext4_inode_info *ei = EXT4_I(inode);
1747
1748	ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1749	EXT4_C2B(EXT4_SB(inode->i_sb),
1750	ext4_count_free_clusters(sb)));
1751	ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1752	ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1753	(long long) EXT4_C2B(EXT4_SB(sb),
1754	percpu_counter_sum(&sbi->s_freeclusters_counter)));
1755	ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1756	(long long) EXT4_C2B(EXT4_SB(sb),
1757	percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1758	ext4_msg(sb, KERN_CRIT, "Block reservation details");
1759	ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1760	ei->i_reserved_data_blocks);
1761	return;
1762	}
1763
1764	/*
1765	* Check whether the cluster containing lblk has been allocated or has
1766	* delalloc reservation.
1767	*
1768	* Returns 0 if the cluster doesn't have either, 1 if it has delalloc
1769	* reservation, 2 if it's already been allocated, negative error code on
1770	* failure.
1771	*/
1772	static int ext4_clu_alloc_state(struct inode *inode, ext4_lblk_t lblk)
1773	{
1774	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
1775	int ret;
1776
1777	/* Has delalloc reservation? */
1778	if (ext4_es_scan_clu(inode, matching_fn: &ext4_es_is_delayed, lblk))
1779	return 1;
1780
1781	/* Already been allocated? */
1782	if (ext4_es_scan_clu(inode, matching_fn: &ext4_es_is_mapped, lblk))
1783	return 2;
1784	ret = ext4_clu_mapped(inode, EXT4_B2C(sbi, lblk));
1785	if (ret < 0)
1786	return ret;
1787	if (ret > 0)
1788	return 2;
1789
1790	return 0;
1791	}
1792
1793	/*
1794	* ext4_insert_delayed_blocks - adds a multiple delayed blocks to the extents
1795	* status tree, incrementing the reserved
1796	* cluster/block count or making pending
1797	* reservations where needed
1798	*
1799	* @inode - file containing the newly added block
1800	* @lblk - start logical block to be added
1801	* @len - length of blocks to be added
1802	*
1803	* Returns 0 on success, negative error code on failure.
1804	*/
1805	static int ext4_insert_delayed_blocks(struct inode *inode, ext4_lblk_t lblk,
1806	ext4_lblk_t len)
1807	{
1808	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
1809	int ret;
1810	bool lclu_allocated = false;
1811	bool end_allocated = false;
1812	ext4_lblk_t resv_clu;
1813	ext4_lblk_t end = lblk + len - 1;
1814
1815	/*
1816	* If the cluster containing lblk or end is shared with a delayed,
1817	* written, or unwritten extent in a bigalloc file system, it's
1818	* already been accounted for and does not need to be reserved.
1819	* A pending reservation must be made for the cluster if it's
1820	* shared with a written or unwritten extent and doesn't already
1821	* have one. Written and unwritten extents can be purged from the
1822	* extents status tree if the system is under memory pressure, so
1823	* it's necessary to examine the extent tree if a search of the
1824	* extents status tree doesn't get a match.
1825	*/
1826	if (sbi->s_cluster_ratio == 1) {
1827	ret = ext4_da_reserve_space(inode, nr_resv: len);
1828	if (ret != 0) /* ENOSPC */
1829	return ret;
1830	} else { /* bigalloc */
1831	resv_clu = EXT4_B2C(sbi, end) - EXT4_B2C(sbi, lblk) + 1;
1832
1833	ret = ext4_clu_alloc_state(inode, lblk);
1834	if (ret < 0)
1835	return ret;
1836	if (ret > 0) {
1837	resv_clu--;
1838	lclu_allocated = (ret == 2);
1839	}
1840
1841	if (EXT4_B2C(sbi, lblk) != EXT4_B2C(sbi, end)) {
1842	ret = ext4_clu_alloc_state(inode, lblk: end);
1843	if (ret < 0)
1844	return ret;
1845	if (ret > 0) {
1846	resv_clu--;
1847	end_allocated = (ret == 2);
1848	}
1849	}
1850
1851	if (resv_clu) {
1852	ret = ext4_da_reserve_space(inode, nr_resv: resv_clu);
1853	if (ret != 0) /* ENOSPC */
1854	return ret;
1855	}
1856	}
1857
1858	ext4_es_insert_delayed_extent(inode, lblk, len, lclu_allocated,
1859	end_allocated);
1860	return 0;
1861	}
1862
1863	/*
1864	* Looks up the requested blocks and sets the delalloc extent map.
1865	* First try to look up for the extent entry that contains the requested
1866	* blocks in the extent status tree without i_data_sem, then try to look
1867	* up for the ondisk extent mapping with i_data_sem in read mode,
1868	* finally hold i_data_sem in write mode, looks up again and add a
1869	* delalloc extent entry if it still couldn't find any extent. Pass out
1870	* the mapped extent through @map and return 0 on success.
1871	*/
1872	static int ext4_da_map_blocks(struct inode *inode, struct ext4_map_blocks *map)
1873	{
1874	struct extent_status es;
1875	int retval;
1876	#ifdef ES_AGGRESSIVE_TEST
1877	struct ext4_map_blocks orig_map;
1878
1879	memcpy(&orig_map, map, sizeof(*map));
1880	#endif
1881
1882	map->m_flags = 0;
1883	ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1884	(unsigned long) map->m_lblk);
1885
1886	ext4_check_map_extents_env(inode);
1887
1888	/* Lookup extent status tree firstly */
1889	if (ext4_es_lookup_extent(inode, lblk: map->m_lblk, NULL, es: &es)) {
1890	map->m_len = min_t(unsigned int, map->m_len,
1891	es.es_len - (map->m_lblk - es.es_lblk));
1892
1893	if (ext4_es_is_hole(es: &es))
1894	goto add_delayed;
1895
1896	found:
1897	/*
1898	* Delayed extent could be allocated by fallocate.
1899	* So we need to check it.
1900	*/
1901	if (ext4_es_is_delayed(es: &es)) {
1902	map->m_flags |= EXT4_MAP_DELAYED;
1903	return 0;
1904	}
1905
1906	map->m_pblk = ext4_es_pblock(es: &es) + map->m_lblk - es.es_lblk;
1907	if (ext4_es_is_written(es: &es))
1908	map->m_flags |= EXT4_MAP_MAPPED;
1909	else if (ext4_es_is_unwritten(es: &es))
1910	map->m_flags |= EXT4_MAP_UNWRITTEN;
1911	else
1912	BUG();
1913
1914	#ifdef ES_AGGRESSIVE_TEST
1915	ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1916	#endif
1917	return 0;
1918	}
1919
1920	/*
1921	* Try to see if we can get the block without requesting a new
1922	* file system block.
1923	*/
1924	down_read(sem: &EXT4_I(inode)->i_data_sem);
1925	if (ext4_has_inline_data(inode))
1926	retval = 0;
1927	else
1928	retval = ext4_map_query_blocks(NULL, inode, map, flags: 0);
1929	up_read(sem: &EXT4_I(inode)->i_data_sem);
1930	if (retval)
1931	return retval < 0 ? retval : 0;
1932
1933	add_delayed:
1934	down_write(sem: &EXT4_I(inode)->i_data_sem);
1935	/*
1936	* Page fault path (ext4_page_mkwrite does not take i_rwsem)
1937	* and fallocate path (no folio lock) can race. Make sure we
1938	* lookup the extent status tree here again while i_data_sem
1939	* is held in write mode, before inserting a new da entry in
1940	* the extent status tree.
1941	*/
1942	if (ext4_es_lookup_extent(inode, lblk: map->m_lblk, NULL, es: &es)) {
1943	map->m_len = min_t(unsigned int, map->m_len,
1944	es.es_len - (map->m_lblk - es.es_lblk));
1945
1946	if (!ext4_es_is_hole(es: &es)) {
1947	up_write(sem: &EXT4_I(inode)->i_data_sem);
1948	goto found;
1949	}
1950	} else if (!ext4_has_inline_data(inode)) {
1951	retval = ext4_map_query_blocks(NULL, inode, map, flags: 0);
1952	if (retval) {
1953	up_write(sem: &EXT4_I(inode)->i_data_sem);
1954	return retval < 0 ? retval : 0;
1955	}
1956	}
1957
1958	map->m_flags |= EXT4_MAP_DELAYED;
1959	retval = ext4_insert_delayed_blocks(inode, lblk: map->m_lblk, len: map->m_len);
1960	up_write(sem: &EXT4_I(inode)->i_data_sem);
1961
1962	return retval;
1963	}
1964
1965	/*
1966	* This is a special get_block_t callback which is used by
1967	* ext4_da_write_begin(). It will either return mapped block or
1968	* reserve space for a single block.
1969	*
1970	* For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1971	* We also have b_blocknr = -1 and b_bdev initialized properly
1972	*
1973	* For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1974	* We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1975	* initialized properly.
1976	*/
1977	int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1978	struct buffer_head *bh, int create)
1979	{
1980	struct ext4_map_blocks map;
1981	sector_t invalid_block = ~((sector_t) 0xffff);
1982	int ret = 0;
1983
1984	BUG_ON(create == 0);
1985	BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1986
1987	if (invalid_block < ext4_blocks_count(es: EXT4_SB(sb: inode->i_sb)->s_es))
1988	invalid_block = ~0;
1989
1990	map.m_lblk = iblock;
1991	map.m_len = 1;
1992
1993	/*
1994	* first, we need to know whether the block is allocated already
1995	* preallocated blocks are unmapped but should treated
1996	* the same as allocated blocks.
1997	*/
1998	ret = ext4_da_map_blocks(inode, map: &map);
1999	if (ret < 0)
2000	return ret;
2001
2002	if (map.m_flags & EXT4_MAP_DELAYED) {
2003	map_bh(bh, sb: inode->i_sb, block: invalid_block);
2004	set_buffer_new(bh);
2005	set_buffer_delay(bh);
2006	return 0;
2007	}
2008
2009	map_bh(bh, sb: inode->i_sb, block: map.m_pblk);
2010	ext4_update_bh_state(bh, flags: map.m_flags);
2011
2012	if (buffer_unwritten(bh)) {
2013	/* A delayed write to unwritten bh should be marked
2014	* new and mapped. Mapped ensures that we don't do
2015	* get_block multiple times when we write to the same
2016	* offset and new ensures that we do proper zero out
2017	* for partial write.
2018	*/
2019	set_buffer_new(bh);
2020	set_buffer_mapped(bh);
2021	}
2022	return 0;
2023	}
2024
2025	static void mpage_folio_done(struct mpage_da_data *mpd, struct folio *folio)
2026	{
2027	mpd->first_page += folio_nr_pages(folio);
2028	folio_unlock(folio);
2029	}
2030
2031	static int mpage_submit_folio(struct mpage_da_data *mpd, struct folio *folio)
2032	{
2033	size_t len;
2034	loff_t size;
2035	int err;
2036
2037	BUG_ON(folio->index != mpd->first_page);
2038	folio_clear_dirty_for_io(folio);
2039	/*
2040	* We have to be very careful here! Nothing protects writeback path
2041	* against i_size changes and the page can be writeably mapped into
2042	* page tables. So an application can be growing i_size and writing
2043	* data through mmap while writeback runs. folio_clear_dirty_for_io()
2044	* write-protects our page in page tables and the page cannot get
2045	* written to again until we release folio lock. So only after
2046	* folio_clear_dirty_for_io() we are safe to sample i_size for
2047	* ext4_bio_write_folio() to zero-out tail of the written page. We rely
2048	* on the barrier provided by folio_test_clear_dirty() in
2049	* folio_clear_dirty_for_io() to make sure i_size is really sampled only
2050	* after page tables are updated.
2051	*/
2052	size = i_size_read(inode: mpd->inode);
2053	len = folio_size(folio);
2054	if (folio_pos(folio) + len > size &&
2055	!ext4_verity_in_progress(inode: mpd->inode))
2056	len = size & (len - 1);
2057	err = ext4_bio_write_folio(io: &mpd->io_submit, page: folio, len);
2058	if (!err)
2059	mpd->wbc->nr_to_write -= folio_nr_pages(folio);
2060
2061	return err;
2062	}
2063
2064	#define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2065
2066	/*
2067	* mballoc gives us at most this number of blocks...
2068	* XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2069	* The rest of mballoc seems to handle chunks up to full group size.
2070	*/
2071	#define MAX_WRITEPAGES_EXTENT_LEN 2048
2072
2073	/*
2074	* mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2075	*
2076	* @mpd - extent of blocks
2077	* @lblk - logical number of the block in the file
2078	* @bh - buffer head we want to add to the extent
2079	*
2080	* The function is used to collect contig. blocks in the same state. If the
2081	* buffer doesn't require mapping for writeback and we haven't started the
2082	* extent of buffers to map yet, the function returns 'true' immediately - the
2083	* caller can write the buffer right away. Otherwise the function returns true
2084	* if the block has been added to the extent, false if the block couldn't be
2085	* added.
2086	*/
2087	static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2088	struct buffer_head *bh)
2089	{
2090	struct ext4_map_blocks *map = &mpd->map;
2091
2092	/* Buffer that doesn't need mapping for writeback? */
2093	if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2094	(!buffer_delay(bh) && !buffer_unwritten(bh))) {
2095	/* So far no extent to map => we write the buffer right away */
2096	if (map->m_len == 0)
2097	return true;
2098	return false;
2099	}
2100
2101	/* First block in the extent? */
2102	if (map->m_len == 0) {
2103	/* We cannot map unless handle is started... */
2104	if (!mpd->do_map)
2105	return false;
2106	map->m_lblk = lblk;
2107	map->m_len = 1;
2108	map->m_flags = bh->b_state & BH_FLAGS;
2109	return true;
2110	}
2111
2112	/* Don't go larger than mballoc is willing to allocate */
2113	if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2114	return false;
2115
2116	/* Can we merge the block to our big extent? */
2117	if (lblk == map->m_lblk + map->m_len &&
2118	(bh->b_state & BH_FLAGS) == map->m_flags) {
2119	map->m_len++;
2120	return true;
2121	}
2122	return false;
2123	}
2124
2125	/*
2126	* mpage_process_page_bufs - submit page buffers for IO or add them to extent
2127	*
2128	* @mpd - extent of blocks for mapping
2129	* @head - the first buffer in the page
2130	* @bh - buffer we should start processing from
2131	* @lblk - logical number of the block in the file corresponding to @bh
2132	*
2133	* Walk through page buffers from @bh upto @head (exclusive) and either submit
2134	* the page for IO if all buffers in this page were mapped and there's no
2135	* accumulated extent of buffers to map or add buffers in the page to the
2136	* extent of buffers to map. The function returns 1 if the caller can continue
2137	* by processing the next page, 0 if it should stop adding buffers to the
2138	* extent to map because we cannot extend it anymore. It can also return value
2139	* < 0 in case of error during IO submission.
2140	*/
2141	static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2142	struct buffer_head *head,
2143	struct buffer_head *bh,
2144	ext4_lblk_t lblk)
2145	{
2146	struct inode *inode = mpd->inode;
2147	int err;
2148	ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(node: inode) - 1)
2149	>> inode->i_blkbits;
2150
2151	if (ext4_verity_in_progress(inode))
2152	blocks = EXT_MAX_BLOCKS;
2153
2154	do {
2155	BUG_ON(buffer_locked(bh));
2156
2157	if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2158	/* Found extent to map? */
2159	if (mpd->map.m_len)
2160	return 0;
2161	/* Buffer needs mapping and handle is not started? */
2162	if (!mpd->do_map)
2163	return 0;
2164	/* Everything mapped so far and we hit EOF */
2165	break;
2166	}
2167	} while (lblk++, (bh = bh->b_this_page) != head);
2168	/* So far everything mapped? Submit the page for IO. */
2169	if (mpd->map.m_len == 0) {
2170	err = mpage_submit_folio(mpd, folio: head->b_folio);
2171	if (err < 0)
2172	return err;
2173	mpage_folio_done(mpd, folio: head->b_folio);
2174	}
2175	if (lblk >= blocks) {
2176	mpd->scanned_until_end = 1;
2177	return 0;
2178	}
2179	return 1;
2180	}
2181
2182	/*
2183	* mpage_process_folio - update folio buffers corresponding to changed extent
2184	* and may submit fully mapped page for IO
2185	* @mpd: description of extent to map, on return next extent to map
2186	* @folio: Contains these buffers.
2187	* @m_lblk: logical block mapping.
2188	* @m_pblk: corresponding physical mapping.
2189	* @map_bh: determines on return whether this page requires any further
2190	* mapping or not.
2191	*
2192	* Scan given folio buffers corresponding to changed extent and update buffer
2193	* state according to new extent state.
2194	* We map delalloc buffers to their physical location, clear unwritten bits.
2195	* If the given folio is not fully mapped, we update @mpd to the next extent in
2196	* the given folio that needs mapping & return @map_bh as true.
2197	*/
2198	static int mpage_process_folio(struct mpage_da_data *mpd, struct folio *folio,
2199	ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2200	bool *map_bh)
2201	{
2202	struct buffer_head *head, *bh;
2203	ext4_io_end_t *io_end = mpd->io_submit.io_end;
2204	ext4_lblk_t lblk = *m_lblk;
2205	ext4_fsblk_t pblock = *m_pblk;
2206	int err = 0;
2207	int blkbits = mpd->inode->i_blkbits;
2208	ssize_t io_end_size = 0;
2209	struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2210
2211	bh = head = folio_buffers(folio);
2212	do {
2213	if (lblk < mpd->map.m_lblk)
2214	continue;
2215	if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2216	/*
2217	* Buffer after end of mapped extent.
2218	* Find next buffer in the folio to map.
2219	*/
2220	mpd->map.m_len = 0;
2221	mpd->map.m_flags = 0;
2222	io_end_vec->size += io_end_size;
2223
2224	err = mpage_process_page_bufs(mpd, head, bh, lblk);
2225	if (err > 0)
2226	err = 0;
2227	if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2228	io_end_vec = ext4_alloc_io_end_vec(io_end);
2229	if (IS_ERR(ptr: io_end_vec)) {
2230	err = PTR_ERR(ptr: io_end_vec);
2231	goto out;
2232	}
2233	io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2234	}
2235	*map_bh = true;
2236	goto out;
2237	}
2238	if (buffer_delay(bh)) {
2239	clear_buffer_delay(bh);
2240	bh->b_blocknr = pblock++;
2241	}
2242	clear_buffer_unwritten(bh);
2243	io_end_size += (1 << blkbits);
2244	} while (lblk++, (bh = bh->b_this_page) != head);
2245
2246	io_end_vec->size += io_end_size;
2247	*map_bh = false;
2248	out:
2249	*m_lblk = lblk;
2250	*m_pblk = pblock;
2251	return err;
2252	}
2253
2254	/*
2255	* mpage_map_buffers - update buffers corresponding to changed extent and
2256	* submit fully mapped pages for IO
2257	*
2258	* @mpd - description of extent to map, on return next extent to map
2259	*
2260	* Scan buffers corresponding to changed extent (we expect corresponding pages
2261	* to be already locked) and update buffer state according to new extent state.
2262	* We map delalloc buffers to their physical location, clear unwritten bits,
2263	* and mark buffers as uninit when we perform writes to unwritten extents
2264	* and do extent conversion after IO is finished. If the last page is not fully
2265	* mapped, we update @map to the next extent in the last page that needs
2266	* mapping. Otherwise we submit the page for IO.
2267	*/
2268	static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2269	{
2270	struct folio_batch fbatch;
2271	unsigned nr, i;
2272	struct inode *inode = mpd->inode;
2273	int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2274	pgoff_t start, end;
2275	ext4_lblk_t lblk;
2276	ext4_fsblk_t pblock;
2277	int err;
2278	bool map_bh = false;
2279
2280	start = mpd->map.m_lblk >> bpp_bits;
2281	end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2282	pblock = mpd->map.m_pblk;
2283
2284	folio_batch_init(fbatch: &fbatch);
2285	while (start <= end) {
2286	nr = filemap_get_folios(mapping: inode->i_mapping, start: &start, end, fbatch: &fbatch);
2287	if (nr == 0)
2288	break;
2289	for (i = 0; i < nr; i++) {
2290	struct folio *folio = fbatch.folios[i];
2291
2292	lblk = folio->index << bpp_bits;
2293	err = mpage_process_folio(mpd, folio, m_lblk: &lblk, m_pblk: &pblock,
2294	map_bh: &map_bh);
2295	/*
2296	* If map_bh is true, means page may require further bh
2297	* mapping, or maybe the page was submitted for IO.
2298	* So we return to call further extent mapping.
2299	*/
2300	if (err < 0 || map_bh)
2301	goto out;
2302	/* Page fully mapped - let IO run! */
2303	err = mpage_submit_folio(mpd, folio);
2304	if (err < 0)
2305	goto out;
2306	mpage_folio_done(mpd, folio);
2307	}
2308	folio_batch_release(fbatch: &fbatch);
2309	}
2310	/* Extent fully mapped and matches with page boundary. We are done. */
2311	mpd->map.m_len = 0;
2312	mpd->map.m_flags = 0;
2313	return 0;
2314	out:
2315	folio_batch_release(fbatch: &fbatch);
2316	return err;
2317	}
2318
2319	static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2320	{
2321	struct inode *inode = mpd->inode;
2322	struct ext4_map_blocks *map = &mpd->map;
2323	int get_blocks_flags;
2324	int err, dioread_nolock;
2325
2326	trace_ext4_da_write_pages_extent(inode, map);
2327	/*
2328	* Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2329	* to convert an unwritten extent to be initialized (in the case
2330	* where we have written into one or more preallocated blocks). It is
2331	* possible that we're going to need more metadata blocks than
2332	* previously reserved. However we must not fail because we're in
2333	* writeback and there is nothing we can do about it so it might result
2334	* in data loss. So use reserved blocks to allocate metadata if
2335	* possible. In addition, do not cache any unrelated extents, as it
2336	* only holds the folio lock but does not hold the i_rwsem or
2337	* invalidate_lock, which could corrupt the extent status tree.
2338	*/
2339	get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2340	EXT4_GET_BLOCKS_METADATA_NOFAIL |
2341	EXT4_GET_BLOCKS_IO_SUBMIT |
2342	EXT4_EX_NOCACHE;
2343
2344	dioread_nolock = ext4_should_dioread_nolock(inode);
2345	if (dioread_nolock)
2346	get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2347
2348	err = ext4_map_blocks(handle, inode, map, flags: get_blocks_flags);
2349	if (err < 0)
2350	return err;
2351	if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2352	if (!mpd->io_submit.io_end->handle &&
2353	ext4_handle_valid(handle)) {
2354	mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2355	handle->h_rsv_handle = NULL;
2356	}
2357	ext4_set_io_unwritten_flag(io_end: mpd->io_submit.io_end);
2358	}
2359
2360	BUG_ON(map->m_len == 0);
2361	return 0;
2362	}
2363
2364	/*
2365	* mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2366	* mpd->len and submit pages underlying it for IO
2367	*
2368	* @handle - handle for journal operations
2369	* @mpd - extent to map
2370	* @give_up_on_write - we set this to true iff there is a fatal error and there
2371	* is no hope of writing the data. The caller should discard
2372	* dirty pages to avoid infinite loops.
2373	*
2374	* The function maps extent starting at mpd->lblk of length mpd->len. If it is
2375	* delayed, blocks are allocated, if it is unwritten, we may need to convert
2376	* them to initialized or split the described range from larger unwritten
2377	* extent. Note that we need not map all the described range since allocation
2378	* can return less blocks or the range is covered by more unwritten extents. We
2379	* cannot map more because we are limited by reserved transaction credits. On
2380	* the other hand we always make sure that the last touched page is fully
2381	* mapped so that it can be written out (and thus forward progress is
2382	* guaranteed). After mapping we submit all mapped pages for IO.
2383	*/
2384	static int mpage_map_and_submit_extent(handle_t *handle,
2385	struct mpage_da_data *mpd,
2386	bool *give_up_on_write)
2387	{
2388	struct inode *inode = mpd->inode;
2389	struct ext4_map_blocks *map = &mpd->map;
2390	int err;
2391	loff_t disksize;
2392	int progress = 0;
2393	ext4_io_end_t *io_end = mpd->io_submit.io_end;
2394	struct ext4_io_end_vec *io_end_vec;
2395
2396	io_end_vec = ext4_alloc_io_end_vec(io_end);
2397	if (IS_ERR(ptr: io_end_vec))
2398	return PTR_ERR(ptr: io_end_vec);
2399	io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2400	do {
2401	err = mpage_map_one_extent(handle, mpd);
2402	if (err < 0) {
2403	struct super_block *sb = inode->i_sb;
2404
2405	if (ext4_emergency_state(sb))
2406	goto invalidate_dirty_pages;
2407	/*
2408	* Let the uper layers retry transient errors.
2409	* In the case of ENOSPC, if ext4_count_free_blocks()
2410	* is non-zero, a commit should free up blocks.
2411	*/
2412	if ((err == -ENOMEM) ||
2413	(err == -ENOSPC && ext4_count_free_clusters(sb))) {
2414	if (progress)
2415	goto update_disksize;
2416	return err;
2417	}
2418	ext4_msg(sb, KERN_CRIT,
2419	"Delayed block allocation failed for "
2420	"inode %lu at logical offset %llu with"
2421	" max blocks %u with error %d",
2422	inode->i_ino,
2423	(unsigned long long)map->m_lblk,
2424	(unsigned)map->m_len, -err);
2425	ext4_msg(sb, KERN_CRIT,
2426	"This should not happen!! Data will "
2427	"be lost\n");
2428	if (err == -ENOSPC)
2429	ext4_print_free_blocks(inode);
2430	invalidate_dirty_pages:
2431	*give_up_on_write = true;
2432	return err;
2433	}
2434	progress = 1;
2435	/*
2436	* Update buffer state, submit mapped pages, and get us new
2437	* extent to map
2438	*/
2439	err = mpage_map_and_submit_buffers(mpd);
2440	if (err < 0)
2441	goto update_disksize;
2442	} while (map->m_len);
2443
2444	update_disksize:
2445	/*
2446	* Update on-disk size after IO is submitted. Races with
2447	* truncate are avoided by checking i_size under i_data_sem.
2448	*/
2449	disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2450	if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2451	int err2;
2452	loff_t i_size;
2453
2454	down_write(sem: &EXT4_I(inode)->i_data_sem);
2455	i_size = i_size_read(inode);
2456	if (disksize > i_size)
2457	disksize = i_size;
2458	if (disksize > EXT4_I(inode)->i_disksize)
2459	EXT4_I(inode)->i_disksize = disksize;
2460	up_write(sem: &EXT4_I(inode)->i_data_sem);
2461	err2 = ext4_mark_inode_dirty(handle, inode);
2462	if (err2) {
2463	ext4_error_err(inode->i_sb, -err2,
2464	"Failed to mark inode %lu dirty",
2465	inode->i_ino);
2466	}
2467	if (!err)
2468	err = err2;
2469	}
2470	return err;
2471	}
2472
2473	/*
2474	* Calculate the total number of credits to reserve for one writepages
2475	* iteration. This is called from ext4_writepages(). We map an extent of
2476	* up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2477	* the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2478	* bpp - 1 blocks in bpp different extents.
2479	*/
2480	static int ext4_da_writepages_trans_blocks(struct inode *inode)
2481	{
2482	int bpp = ext4_journal_blocks_per_folio(inode);
2483
2484	return ext4_meta_trans_blocks(inode,
2485	MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, pextents: bpp);
2486	}
2487
2488	static int ext4_journal_folio_buffers(handle_t *handle, struct folio *folio,
2489	size_t len)
2490	{
2491	struct buffer_head *page_bufs = folio_buffers(folio);
2492	struct inode *inode = folio->mapping->host;
2493	int ret, err;
2494
2495	ret = ext4_walk_page_buffers(handle, inode, head: page_bufs, from: 0, to: len,
2496	NULL, fn: do_journal_get_write_access);
2497	err = ext4_walk_page_buffers(handle, inode, head: page_bufs, from: 0, to: len,
2498	NULL, fn: write_end_fn);
2499	if (ret == 0)
2500	ret = err;
2501	err = ext4_jbd2_inode_add_write(handle, inode, start_byte: folio_pos(folio), length: len);
2502	if (ret == 0)
2503	ret = err;
2504	EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2505
2506	return ret;
2507	}
2508
2509	static int mpage_journal_page_buffers(handle_t *handle,
2510	struct mpage_da_data *mpd,
2511	struct folio *folio)
2512	{
2513	struct inode *inode = mpd->inode;
2514	loff_t size = i_size_read(inode);
2515	size_t len = folio_size(folio);
2516
2517	folio_clear_checked(folio);
2518	mpd->wbc->nr_to_write -= folio_nr_pages(folio);
2519
2520	if (folio_pos(folio) + len > size &&
2521	!ext4_verity_in_progress(inode))
2522	len = size & (len - 1);
2523
2524	return ext4_journal_folio_buffers(handle, folio, len);
2525	}
2526
2527	/*
2528	* mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2529	* needing mapping, submit mapped pages
2530	*
2531	* @mpd - where to look for pages
2532	*
2533	* Walk dirty pages in the mapping. If they are fully mapped, submit them for
2534	* IO immediately. If we cannot map blocks, we submit just already mapped
2535	* buffers in the page for IO and keep page dirty. When we can map blocks and
2536	* we find a page which isn't mapped we start accumulating extent of buffers
2537	* underlying these pages that needs mapping (formed by either delayed or
2538	* unwritten buffers). We also lock the pages containing these buffers. The
2539	* extent found is returned in @mpd structure (starting at mpd->lblk with
2540	* length mpd->len blocks).
2541	*
2542	* Note that this function can attach bios to one io_end structure which are
2543	* neither logically nor physically contiguous. Although it may seem as an
2544	* unnecessary complication, it is actually inevitable in blocksize < pagesize
2545	* case as we need to track IO to all buffers underlying a page in one io_end.
2546	*/
2547	static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2548	{
2549	struct address_space *mapping = mpd->inode->i_mapping;
2550	struct folio_batch fbatch;
2551	unsigned int nr_folios;
2552	pgoff_t index = mpd->first_page;
2553	pgoff_t end = mpd->last_page;
2554	xa_mark_t tag;
2555	int i, err = 0;
2556	int blkbits = mpd->inode->i_blkbits;
2557	ext4_lblk_t lblk;
2558	struct buffer_head *head;
2559	handle_t *handle = NULL;
2560	int bpp = ext4_journal_blocks_per_folio(inode: mpd->inode);
2561
2562	if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2563	tag = PAGECACHE_TAG_TOWRITE;
2564	else
2565	tag = PAGECACHE_TAG_DIRTY;
2566
2567	mpd->map.m_len = 0;
2568	mpd->next_page = index;
2569	if (ext4_should_journal_data(inode: mpd->inode)) {
2570	handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE,
2571	bpp);
2572	if (IS_ERR(ptr: handle))
2573	return PTR_ERR(ptr: handle);
2574	}
2575	folio_batch_init(fbatch: &fbatch);
2576	while (index <= end) {
2577	nr_folios = filemap_get_folios_tag(mapping, start: &index, end,
2578	tag, fbatch: &fbatch);
2579	if (nr_folios == 0)
2580	break;
2581
2582	for (i = 0; i < nr_folios; i++) {
2583	struct folio *folio = fbatch.folios[i];
2584
2585	/*
2586	* Accumulated enough dirty pages? This doesn't apply
2587	* to WB_SYNC_ALL mode. For integrity sync we have to
2588	* keep going because someone may be concurrently
2589	* dirtying pages, and we might have synced a lot of
2590	* newly appeared dirty pages, but have not synced all
2591	* of the old dirty pages.
2592	*/
2593	if (mpd->wbc->sync_mode == WB_SYNC_NONE &&
2594	mpd->wbc->nr_to_write <=
2595	mpd->map.m_len >> (PAGE_SHIFT - blkbits))
2596	goto out;
2597
2598	/* If we can't merge this page, we are done. */
2599	if (mpd->map.m_len > 0 && mpd->next_page != folio->index)
2600	goto out;
2601
2602	if (handle) {
2603	err = ext4_journal_ensure_credits(handle, credits: bpp,
2604	revoke_creds: 0);
2605	if (err < 0)
2606	goto out;
2607	}
2608
2609	folio_lock(folio);
2610	/*
2611	* If the page is no longer dirty, or its mapping no
2612	* longer corresponds to inode we are writing (which
2613	* means it has been truncated or invalidated), or the
2614	* page is already under writeback and we are not doing
2615	* a data integrity writeback, skip the page
2616	*/
2617	if (!folio_test_dirty(folio) ||
2618	(folio_test_writeback(folio) &&
2619	(mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2620	unlikely(folio->mapping != mapping)) {
2621	folio_unlock(folio);
2622	continue;
2623	}
2624
2625	folio_wait_writeback(folio);
2626	BUG_ON(folio_test_writeback(folio));
2627
2628	/*
2629	* Should never happen but for buggy code in
2630	* other subsystems that call
2631	* set_page_dirty() without properly warning
2632	* the file system first. See [1] for more
2633	* information.
2634	*
2635	* [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2636	*/
2637	if (!folio_buffers(folio)) {
2638	ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", folio->index);
2639	folio_clear_dirty(folio);
2640	folio_unlock(folio);
2641	continue;
2642	}
2643
2644	if (mpd->map.m_len == 0)
2645	mpd->first_page = folio->index;
2646	mpd->next_page = folio_next_index(folio);
2647	/*
2648	* Writeout when we cannot modify metadata is simple.
2649	* Just submit the page. For data=journal mode we
2650	* first handle writeout of the page for checkpoint and
2651	* only after that handle delayed page dirtying. This
2652	* makes sure current data is checkpointed to the final
2653	* location before possibly journalling it again which
2654	* is desirable when the page is frequently dirtied
2655	* through a pin.
2656	*/
2657	if (!mpd->can_map) {
2658	err = mpage_submit_folio(mpd, folio);
2659	if (err < 0)
2660	goto out;
2661	/* Pending dirtying of journalled data? */
2662	if (folio_test_checked(folio)) {
2663	err = mpage_journal_page_buffers(handle,
2664	mpd, folio);
2665	if (err < 0)
2666	goto out;
2667	mpd->journalled_more_data = 1;
2668	}
2669	mpage_folio_done(mpd, folio);
2670	} else {
2671	/* Add all dirty buffers to mpd */
2672	lblk = ((ext4_lblk_t)folio->index) <<
2673	(PAGE_SHIFT - blkbits);
2674	head = folio_buffers(folio);
2675	err = mpage_process_page_bufs(mpd, head, bh: head,
2676	lblk);
2677	if (err <= 0)
2678	goto out;
2679	err = 0;
2680	}
2681	}
2682	folio_batch_release(fbatch: &fbatch);
2683	cond_resched();
2684	}
2685	mpd->scanned_until_end = 1;
2686	if (handle)
2687	ext4_journal_stop(handle);
2688	return 0;
2689	out:
2690	folio_batch_release(fbatch: &fbatch);
2691	if (handle)
2692	ext4_journal_stop(handle);
2693	return err;
2694	}
2695
2696	static int ext4_do_writepages(struct mpage_da_data *mpd)
2697	{
2698	struct writeback_control *wbc = mpd->wbc;
2699	pgoff_t writeback_index = 0;
2700	long nr_to_write = wbc->nr_to_write;
2701	int range_whole = 0;
2702	int cycled = 1;
2703	handle_t *handle = NULL;
2704	struct inode *inode = mpd->inode;
2705	struct address_space *mapping = inode->i_mapping;
2706	int needed_blocks, rsv_blocks = 0, ret = 0;
2707	struct ext4_sb_info *sbi = EXT4_SB(sb: mapping->host->i_sb);
2708	struct blk_plug plug;
2709	bool give_up_on_write = false;
2710
2711	trace_ext4_writepages(inode, wbc);
2712
2713	/*
2714	* No pages to write? This is mainly a kludge to avoid starting
2715	* a transaction for special inodes like journal inode on last iput()
2716	* because that could violate lock ordering on umount
2717	*/
2718	if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2719	goto out_writepages;
2720
2721	/*
2722	* If the filesystem has aborted, it is read-only, so return
2723	* right away instead of dumping stack traces later on that
2724	* will obscure the real source of the problem. We test
2725	* fs shutdown state instead of sb->s_flag's SB_RDONLY because
2726	* the latter could be true if the filesystem is mounted
2727	* read-only, and in that case, ext4_writepages should
2728	* *never* be called, so if that ever happens, we would want
2729	* the stack trace.
2730	*/
2731	ret = ext4_emergency_state(sb: mapping->host->i_sb);
2732	if (unlikely(ret))
2733	goto out_writepages;
2734
2735	/*
2736	* If we have inline data and arrive here, it means that
2737	* we will soon create the block for the 1st page, so
2738	* we'd better clear the inline data here.
2739	*/
2740	if (ext4_has_inline_data(inode)) {
2741	/* Just inode will be modified... */
2742	handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2743	if (IS_ERR(ptr: handle)) {
2744	ret = PTR_ERR(ptr: handle);
2745	goto out_writepages;
2746	}
2747	BUG_ON(ext4_test_inode_state(inode,
2748	EXT4_STATE_MAY_INLINE_DATA));
2749	ext4_destroy_inline_data(handle, inode);
2750	ext4_journal_stop(handle);
2751	}
2752
2753	/*
2754	* data=journal mode does not do delalloc so we just need to writeout /
2755	* journal already mapped buffers. On the other hand we need to commit
2756	* transaction to make data stable. We expect all the data to be
2757	* already in the journal (the only exception are DMA pinned pages
2758	* dirtied behind our back) so we commit transaction here and run the
2759	* writeback loop to checkpoint them. The checkpointing is not actually
2760	* necessary to make data persistent *but* quite a few places (extent
2761	* shifting operations, fsverity, ...) depend on being able to drop
2762	* pagecache pages after calling filemap_write_and_wait() and for that
2763	* checkpointing needs to happen.
2764	*/
2765	if (ext4_should_journal_data(inode)) {
2766	mpd->can_map = 0;
2767	if (wbc->sync_mode == WB_SYNC_ALL)
2768	ext4_fc_commit(journal: sbi->s_journal,
2769	EXT4_I(inode)->i_datasync_tid);
2770	}
2771	mpd->journalled_more_data = 0;
2772
2773	if (ext4_should_dioread_nolock(inode)) {
2774	/*
2775	* We may need to convert up to one extent per block in
2776	* the page and we may dirty the inode.
2777	*/
2778	rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2779	PAGE_SIZE >> inode->i_blkbits);
2780	}
2781
2782	if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2783	range_whole = 1;
2784
2785	if (wbc->range_cyclic) {
2786	writeback_index = mapping->writeback_index;
2787	if (writeback_index)
2788	cycled = 0;
2789	mpd->first_page = writeback_index;
2790	mpd->last_page = -1;
2791	} else {
2792	mpd->first_page = wbc->range_start >> PAGE_SHIFT;
2793	mpd->last_page = wbc->range_end >> PAGE_SHIFT;
2794	}
2795
2796	ext4_io_submit_init(io: &mpd->io_submit, wbc);
2797	retry:
2798	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2799	tag_pages_for_writeback(mapping, start: mpd->first_page,
2800	end: mpd->last_page);
2801	blk_start_plug(&plug);
2802
2803	/*
2804	* First writeback pages that don't need mapping - we can avoid
2805	* starting a transaction unnecessarily and also avoid being blocked
2806	* in the block layer on device congestion while having transaction
2807	* started.
2808	*/
2809	mpd->do_map = 0;
2810	mpd->scanned_until_end = 0;
2811	mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2812	if (!mpd->io_submit.io_end) {
2813	ret = -ENOMEM;
2814	goto unplug;
2815	}
2816	ret = mpage_prepare_extent_to_map(mpd);
2817	/* Unlock pages we didn't use */
2818	mpage_release_unused_pages(mpd, invalidate: false);
2819	/* Submit prepared bio */
2820	ext4_io_submit(io: &mpd->io_submit);
2821	ext4_put_io_end_defer(io_end: mpd->io_submit.io_end);
2822	mpd->io_submit.io_end = NULL;
2823	if (ret < 0)
2824	goto unplug;
2825
2826	while (!mpd->scanned_until_end && wbc->nr_to_write > 0) {
2827	/* For each extent of pages we use new io_end */
2828	mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2829	if (!mpd->io_submit.io_end) {
2830	ret = -ENOMEM;
2831	break;
2832	}
2833
2834	WARN_ON_ONCE(!mpd->can_map);
2835	/*
2836	* We have two constraints: We find one extent to map and we
2837	* must always write out whole page (makes a difference when
2838	* blocksize < pagesize) so that we don't block on IO when we
2839	* try to write out the rest of the page. Journalled mode is
2840	* not supported by delalloc.
2841	*/
2842	BUG_ON(ext4_should_journal_data(inode));
2843	needed_blocks = ext4_da_writepages_trans_blocks(inode);
2844
2845	/* start a new transaction */
2846	handle = ext4_journal_start_with_reserve(inode,
2847	EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2848	if (IS_ERR(ptr: handle)) {
2849	ret = PTR_ERR(ptr: handle);
2850	ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2851	"%ld pages, ino %lu; err %d", __func__,
2852	wbc->nr_to_write, inode->i_ino, ret);
2853	/* Release allocated io_end */
2854	ext4_put_io_end(io_end: mpd->io_submit.io_end);
2855	mpd->io_submit.io_end = NULL;
2856	break;
2857	}
2858	mpd->do_map = 1;
2859
2860	trace_ext4_da_write_pages(inode, first_page: mpd->first_page, wbc);
2861	ret = mpage_prepare_extent_to_map(mpd);
2862	if (!ret && mpd->map.m_len)
2863	ret = mpage_map_and_submit_extent(handle, mpd,
2864	give_up_on_write: &give_up_on_write);
2865	/*
2866	* Caution: If the handle is synchronous,
2867	* ext4_journal_stop() can wait for transaction commit
2868	* to finish which may depend on writeback of pages to
2869	* complete or on page lock to be released. In that
2870	* case, we have to wait until after we have
2871	* submitted all the IO, released page locks we hold,
2872	* and dropped io_end reference (for extent conversion
2873	* to be able to complete) before stopping the handle.
2874	*/
2875	if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2876	ext4_journal_stop(handle);
2877	handle = NULL;
2878	mpd->do_map = 0;
2879	}
2880	/* Unlock pages we didn't use */
2881	mpage_release_unused_pages(mpd, invalidate: give_up_on_write);
2882	/* Submit prepared bio */
2883	ext4_io_submit(io: &mpd->io_submit);
2884
2885	/*
2886	* Drop our io_end reference we got from init. We have
2887	* to be careful and use deferred io_end finishing if
2888	* we are still holding the transaction as we can
2889	* release the last reference to io_end which may end
2890	* up doing unwritten extent conversion.
2891	*/
2892	if (handle) {
2893	ext4_put_io_end_defer(io_end: mpd->io_submit.io_end);
2894	ext4_journal_stop(handle);
2895	} else
2896	ext4_put_io_end(io_end: mpd->io_submit.io_end);
2897	mpd->io_submit.io_end = NULL;
2898
2899	if (ret == -ENOSPC && sbi->s_journal) {
2900	/*
2901	* Commit the transaction which would
2902	* free blocks released in the transaction
2903	* and try again
2904	*/
2905	jbd2_journal_force_commit_nested(sbi->s_journal);
2906	ret = 0;
2907	continue;
2908	}
2909	/* Fatal error - ENOMEM, EIO... */
2910	if (ret)
2911	break;
2912	}
2913	unplug:
2914	blk_finish_plug(&plug);
2915	if (!ret && !cycled && wbc->nr_to_write > 0) {
2916	cycled = 1;
2917	mpd->last_page = writeback_index - 1;
2918	mpd->first_page = 0;
2919	goto retry;
2920	}
2921
2922	/* Update index */
2923	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2924	/*
2925	* Set the writeback_index so that range_cyclic
2926	* mode will write it back later
2927	*/
2928	mapping->writeback_index = mpd->first_page;
2929
2930	out_writepages:
2931	trace_ext4_writepages_result(inode, wbc, ret,
2932	pages_written: nr_to_write - wbc->nr_to_write);
2933	return ret;
2934	}
2935
2936	static int ext4_writepages(struct address_space *mapping,
2937	struct writeback_control *wbc)
2938	{
2939	struct super_block *sb = mapping->host->i_sb;
2940	struct mpage_da_data mpd = {
2941	.inode = mapping->host,
2942	.wbc = wbc,
2943	.can_map = 1,
2944	};
2945	int ret;
2946	int alloc_ctx;
2947
2948	ret = ext4_emergency_state(sb);
2949	if (unlikely(ret))
2950	return ret;
2951
2952	alloc_ctx = ext4_writepages_down_read(sb);
2953	ret = ext4_do_writepages(mpd: &mpd);
2954	/*
2955	* For data=journal writeback we could have come across pages marked
2956	* for delayed dirtying (PageChecked) which were just added to the
2957	* running transaction. Try once more to get them to stable storage.
2958	*/
2959	if (!ret && mpd.journalled_more_data)
2960	ret = ext4_do_writepages(mpd: &mpd);
2961	ext4_writepages_up_read(sb, ctx: alloc_ctx);
2962
2963	return ret;
2964	}
2965
2966	int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode)
2967	{
2968	struct writeback_control wbc = {
2969	.sync_mode = WB_SYNC_ALL,
2970	.nr_to_write = LONG_MAX,
2971	.range_start = jinode->i_dirty_start,
2972	.range_end = jinode->i_dirty_end,
2973	};
2974	struct mpage_da_data mpd = {
2975	.inode = jinode->i_vfs_inode,
2976	.wbc = &wbc,
2977	.can_map = 0,
2978	};
2979	return ext4_do_writepages(mpd: &mpd);
2980	}
2981
2982	static int ext4_dax_writepages(struct address_space *mapping,
2983	struct writeback_control *wbc)
2984	{
2985	int ret;
2986	long nr_to_write = wbc->nr_to_write;
2987	struct inode *inode = mapping->host;
2988	int alloc_ctx;
2989
2990	ret = ext4_emergency_state(sb: inode->i_sb);
2991	if (unlikely(ret))
2992	return ret;
2993
2994	alloc_ctx = ext4_writepages_down_read(sb: inode->i_sb);
2995	trace_ext4_writepages(inode, wbc);
2996
2997	ret = dax_writeback_mapping_range(mapping,
2998	dax_dev: EXT4_SB(sb: inode->i_sb)->s_daxdev, wbc);
2999	trace_ext4_writepages_result(inode, wbc, ret,
3000	pages_written: nr_to_write - wbc->nr_to_write);
3001	ext4_writepages_up_read(sb: inode->i_sb, ctx: alloc_ctx);
3002	return ret;
3003	}
3004
3005	static int ext4_nonda_switch(struct super_block *sb)
3006	{
3007	s64 free_clusters, dirty_clusters;
3008	struct ext4_sb_info *sbi = EXT4_SB(sb);
3009
3010	/*
3011	* switch to non delalloc mode if we are running low
3012	* on free block. The free block accounting via percpu
3013	* counters can get slightly wrong with percpu_counter_batch getting
3014	* accumulated on each CPU without updating global counters
3015	* Delalloc need an accurate free block accounting. So switch
3016	* to non delalloc when we are near to error range.
3017	*/
3018	free_clusters =
3019	percpu_counter_read_positive(fbc: &sbi->s_freeclusters_counter);
3020	dirty_clusters =
3021	percpu_counter_read_positive(fbc: &sbi->s_dirtyclusters_counter);
3022	/*
3023	* Start pushing delalloc when 1/2 of free blocks are dirty.
3024	*/
3025	if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
3026	try_to_writeback_inodes_sb(sb, reason: WB_REASON_FS_FREE_SPACE);
3027
3028	if (2 * free_clusters < 3 * dirty_clusters ||
3029	free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
3030	/*
3031	* free block count is less than 150% of dirty blocks
3032	* or free blocks is less than watermark
3033	*/
3034	return 1;
3035	}
3036	return 0;
3037	}
3038
3039	static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3040	loff_t pos, unsigned len,
3041	struct folio **foliop, void **fsdata)
3042	{
3043	int ret, retries = 0;
3044	struct folio *folio;
3045	pgoff_t index;
3046	struct inode *inode = mapping->host;
3047	fgf_t fgp = FGP_WRITEBEGIN;
3048
3049	ret = ext4_emergency_state(sb: inode->i_sb);
3050	if (unlikely(ret))
3051	return ret;
3052
3053	index = pos >> PAGE_SHIFT;
3054
3055	if (ext4_nonda_switch(sb: inode->i_sb) || ext4_verity_in_progress(inode)) {
3056	*fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3057	return ext4_write_begin(file, mapping, pos,
3058	len, foliop, fsdata);
3059	}
3060	*fsdata = (void *)0;
3061	trace_ext4_da_write_begin(inode, pos, len);
3062
3063	if (ext4_test_inode_state(inode, bit: EXT4_STATE_MAY_INLINE_DATA)) {
3064	ret = ext4_generic_write_inline_data(mapping, inode, pos, len,
3065	foliop, fsdata, da: true);
3066	if (ret < 0)
3067	return ret;
3068	if (ret == 1)
3069	return 0;
3070	}
3071
3072	retry:
3073	fgp |= fgf_set_order(size: len);
3074	folio = __filemap_get_folio(mapping, index, fgp_flags: fgp,
3075	gfp: mapping_gfp_mask(mapping));
3076	if (IS_ERR(ptr: folio))
3077	return PTR_ERR(ptr: folio);
3078
3079	if (pos + len > folio_pos(folio) + folio_size(folio))
3080	len = folio_pos(folio) + folio_size(folio) - pos;
3081
3082	ret = ext4_block_write_begin(NULL, folio, pos, len,
3083	get_block: ext4_da_get_block_prep);
3084	if (ret < 0) {
3085	folio_unlock(folio);
3086	folio_put(folio);
3087	/*
3088	* block_write_begin may have instantiated a few blocks
3089	* outside i_size. Trim these off again. Don't need
3090	* i_size_read because we hold inode lock.
3091	*/
3092	if (pos + len > inode->i_size)
3093	ext4_truncate_failed_write(inode);
3094
3095	if (ret == -ENOSPC &&
3096	ext4_should_retry_alloc(sb: inode->i_sb, retries: &retries))
3097	goto retry;
3098	return ret;
3099	}
3100
3101	*foliop = folio;
3102	return ret;
3103	}
3104
3105	/*
3106	* Check if we should update i_disksize
3107	* when write to the end of file but not require block allocation
3108	*/
3109	static int ext4_da_should_update_i_disksize(struct folio *folio,
3110	unsigned long offset)
3111	{
3112	struct buffer_head *bh;
3113	struct inode *inode = folio->mapping->host;
3114	unsigned int idx;
3115	int i;
3116
3117	bh = folio_buffers(folio);
3118	idx = offset >> inode->i_blkbits;
3119
3120	for (i = 0; i < idx; i++)
3121	bh = bh->b_this_page;
3122
3123	if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3124	return 0;
3125	return 1;
3126	}
3127
3128	static int ext4_da_do_write_end(struct address_space *mapping,
3129	loff_t pos, unsigned len, unsigned copied,
3130	struct folio *folio)
3131	{
3132	struct inode *inode = mapping->host;
3133	loff_t old_size = inode->i_size;
3134	bool disksize_changed = false;
3135	loff_t new_i_size, zero_len = 0;
3136	handle_t *handle;
3137
3138	if (unlikely(!folio_buffers(folio))) {
3139	folio_unlock(folio);
3140	folio_put(folio);
3141	return -EIO;
3142	}
3143	/*
3144	* block_write_end() will mark the inode as dirty with I_DIRTY_PAGES
3145	* flag, which all that's needed to trigger page writeback.
3146	*/
3147	copied = block_write_end(NULL, mapping, pos, len, copied,
3148	folio, NULL);
3149	new_i_size = pos + copied;
3150
3151	/*
3152	* It's important to update i_size while still holding folio lock,
3153	* because folio writeout could otherwise come in and zero beyond
3154	* i_size.
3155	*
3156	* Since we are holding inode lock, we are sure i_disksize <=
3157	* i_size. We also know that if i_disksize < i_size, there are
3158	* delalloc writes pending in the range up to i_size. If the end of
3159	* the current write is <= i_size, there's no need to touch
3160	* i_disksize since writeback will push i_disksize up to i_size
3161	* eventually. If the end of the current write is > i_size and
3162	* inside an allocated block which ext4_da_should_update_i_disksize()
3163	* checked, we need to update i_disksize here as certain
3164	* ext4_writepages() paths not allocating blocks and update i_disksize.
3165	*/
3166	if (new_i_size > inode->i_size) {
3167	unsigned long end;
3168
3169	i_size_write(inode, i_size: new_i_size);
3170	end = offset_in_folio(folio, new_i_size - 1);
3171	if (copied && ext4_da_should_update_i_disksize(folio, offset: end)) {
3172	ext4_update_i_disksize(inode, newsize: new_i_size);
3173	disksize_changed = true;
3174	}
3175	}
3176
3177	folio_unlock(folio);
3178	folio_put(folio);
3179
3180	if (pos > old_size) {
3181	pagecache_isize_extended(inode, from: old_size, to: pos);
3182	zero_len = pos - old_size;
3183	}
3184
3185	if (!disksize_changed && !zero_len)
3186	return copied;
3187
3188	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3189	if (IS_ERR(ptr: handle))
3190	return PTR_ERR(ptr: handle);
3191	if (zero_len)
3192	ext4_zero_partial_blocks(handle, inode, lstart: old_size, lend: zero_len);
3193	ext4_mark_inode_dirty(handle, inode);
3194	ext4_journal_stop(handle);
3195
3196	return copied;
3197	}
3198
3199	static int ext4_da_write_end(struct file *file,
3200	struct address_space *mapping,
3201	loff_t pos, unsigned len, unsigned copied,
3202	struct folio *folio, void *fsdata)
3203	{
3204	struct inode *inode = mapping->host;
3205	int write_mode = (int)(unsigned long)fsdata;
3206
3207	if (write_mode == FALL_BACK_TO_NONDELALLOC)
3208	return ext4_write_end(file, mapping, pos,
3209	len, copied, folio, fsdata);
3210
3211	trace_ext4_da_write_end(inode, pos, len, copied);
3212
3213	if (write_mode != CONVERT_INLINE_DATA &&
3214	ext4_test_inode_state(inode, bit: EXT4_STATE_MAY_INLINE_DATA) &&
3215	ext4_has_inline_data(inode))
3216	return ext4_write_inline_data_end(inode, pos, len, copied,
3217	folio);
3218
3219	if (unlikely(copied < len) && !folio_test_uptodate(folio))
3220	copied = 0;
3221
3222	return ext4_da_do_write_end(mapping, pos, len, copied, folio);
3223	}
3224
3225	/*
3226	* Force all delayed allocation blocks to be allocated for a given inode.
3227	*/
3228	int ext4_alloc_da_blocks(struct inode *inode)
3229	{
3230	trace_ext4_alloc_da_blocks(inode);
3231
3232	if (!EXT4_I(inode)->i_reserved_data_blocks)
3233	return 0;
3234
3235	/*
3236	* We do something simple for now. The filemap_flush() will
3237	* also start triggering a write of the data blocks, which is
3238	* not strictly speaking necessary (and for users of
3239	* laptop_mode, not even desirable). However, to do otherwise
3240	* would require replicating code paths in:
3241	*
3242	* ext4_writepages() ->
3243	* write_cache_pages() ---> (via passed in callback function)
3244	* __mpage_da_writepage() -->
3245	* mpage_add_bh_to_extent()
3246	* mpage_da_map_blocks()
3247	*
3248	* The problem is that write_cache_pages(), located in
3249	* mm/page-writeback.c, marks pages clean in preparation for
3250	* doing I/O, which is not desirable if we're not planning on
3251	* doing I/O at all.
3252	*
3253	* We could call write_cache_pages(), and then redirty all of
3254	* the pages by calling redirty_page_for_writepage() but that
3255	* would be ugly in the extreme. So instead we would need to
3256	* replicate parts of the code in the above functions,
3257	* simplifying them because we wouldn't actually intend to
3258	* write out the pages, but rather only collect contiguous
3259	* logical block extents, call the multi-block allocator, and
3260	* then update the buffer heads with the block allocations.
3261	*
3262	* For now, though, we'll cheat by calling filemap_flush(),
3263	* which will map the blocks, and start the I/O, but not
3264	* actually wait for the I/O to complete.
3265	*/
3266	return filemap_flush(inode->i_mapping);
3267	}
3268
3269	/*
3270	* bmap() is special. It gets used by applications such as lilo and by
3271	* the swapper to find the on-disk block of a specific piece of data.
3272	*
3273	* Naturally, this is dangerous if the block concerned is still in the
3274	* journal. If somebody makes a swapfile on an ext4 data-journaling
3275	* filesystem and enables swap, then they may get a nasty shock when the
3276	* data getting swapped to that swapfile suddenly gets overwritten by
3277	* the original zero's written out previously to the journal and
3278	* awaiting writeback in the kernel's buffer cache.
3279	*
3280	* So, if we see any bmap calls here on a modified, data-journaled file,
3281	* take extra steps to flush any blocks which might be in the cache.
3282	*/
3283	static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3284	{
3285	struct inode *inode = mapping->host;
3286	sector_t ret = 0;
3287
3288	inode_lock_shared(inode);
3289	/*
3290	* We can get here for an inline file via the FIBMAP ioctl
3291	*/
3292	if (ext4_has_inline_data(inode))
3293	goto out;
3294
3295	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3296	(test_opt(inode->i_sb, DELALLOC) ||
3297	ext4_should_journal_data(inode))) {
3298	/*
3299	* With delalloc or journalled data we want to sync the file so
3300	* that we can make sure we allocate blocks for file and data
3301	* is in place for the user to see it
3302	*/
3303	filemap_write_and_wait(mapping);
3304	}
3305
3306	ret = iomap_bmap(mapping, bno: block, ops: &ext4_iomap_ops);
3307
3308	out:
3309	inode_unlock_shared(inode);
3310	return ret;
3311	}
3312
3313	static int ext4_read_folio(struct file *file, struct folio *folio)
3314	{
3315	int ret = -EAGAIN;
3316	struct inode *inode = folio->mapping->host;
3317
3318	trace_ext4_read_folio(inode, folio);
3319
3320	if (ext4_has_inline_data(inode))
3321	ret = ext4_readpage_inline(inode, folio);
3322
3323	if (ret == -EAGAIN)
3324	return ext4_mpage_readpages(inode, NULL, folio);
3325
3326	return ret;
3327	}
3328
3329	static void ext4_readahead(struct readahead_control *rac)
3330	{
3331	struct inode *inode = rac->mapping->host;
3332
3333	/* If the file has inline data, no need to do readahead. */
3334	if (ext4_has_inline_data(inode))
3335	return;
3336
3337	ext4_mpage_readpages(inode, rac, NULL);
3338	}
3339
3340	static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3341	size_t length)
3342	{
3343	trace_ext4_invalidate_folio(folio, offset, length);
3344
3345	/* No journalling happens on data buffers when this function is used */
3346	WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3347
3348	block_invalidate_folio(folio, offset, length);
3349	}
3350
3351	static int __ext4_journalled_invalidate_folio(struct folio *folio,
3352	size_t offset, size_t length)
3353	{
3354	journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3355
3356	trace_ext4_journalled_invalidate_folio(folio, offset, length);
3357
3358	/*
3359	* If it's a full truncate we just forget about the pending dirtying
3360	*/
3361	if (offset == 0 && length == folio_size(folio))
3362	folio_clear_checked(folio);
3363
3364	return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3365	}
3366
3367	/* Wrapper for aops... */
3368	static void ext4_journalled_invalidate_folio(struct folio *folio,
3369	size_t offset,
3370	size_t length)
3371	{
3372	WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3373	}
3374
3375	static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3376	{
3377	struct inode *inode = folio->mapping->host;
3378	journal_t *journal = EXT4_JOURNAL(inode);
3379
3380	trace_ext4_release_folio(inode, folio);
3381
3382	/* Page has dirty journalled data -> cannot release */
3383	if (folio_test_checked(folio))
3384	return false;
3385	if (journal)
3386	return jbd2_journal_try_to_free_buffers(journal, folio);
3387	else
3388	return try_to_free_buffers(folio);
3389	}
3390
3391	static bool ext4_inode_datasync_dirty(struct inode *inode)
3392	{
3393	journal_t *journal = EXT4_SB(sb: inode->i_sb)->s_journal;
3394
3395	if (journal) {
3396	if (jbd2_transaction_committed(journal,
3397	EXT4_I(inode)->i_datasync_tid))
3398	return false;
3399	if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3400	return !list_empty(head: &EXT4_I(inode)->i_fc_list);
3401	return true;
3402	}
3403
3404	/* Any metadata buffers to write? */
3405	if (!list_empty(head: &inode->i_mapping->i_private_list))
3406	return true;
3407	return inode->i_state & I_DIRTY_DATASYNC;
3408	}
3409
3410	static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3411	struct ext4_map_blocks *map, loff_t offset,
3412	loff_t length, unsigned int flags)
3413	{
3414	u8 blkbits = inode->i_blkbits;
3415
3416	/*
3417	* Writes that span EOF might trigger an I/O size update on completion,
3418	* so consider them to be dirty for the purpose of O_DSYNC, even if
3419	* there is no other metadata changes being made or are pending.
3420	*/
3421	iomap->flags = 0;
3422	if (ext4_inode_datasync_dirty(inode) ||
3423	offset + length > i_size_read(inode))
3424	iomap->flags |= IOMAP_F_DIRTY;
3425
3426	if (map->m_flags & EXT4_MAP_NEW)
3427	iomap->flags |= IOMAP_F_NEW;
3428
3429	/* HW-offload atomics are always used */
3430	if (flags & IOMAP_ATOMIC)
3431	iomap->flags |= IOMAP_F_ATOMIC_BIO;
3432
3433	if (flags & IOMAP_DAX)
3434	iomap->dax_dev = EXT4_SB(sb: inode->i_sb)->s_daxdev;
3435	else
3436	iomap->bdev = inode->i_sb->s_bdev;
3437	iomap->offset = (u64) map->m_lblk << blkbits;
3438	iomap->length = (u64) map->m_len << blkbits;
3439
3440	if ((map->m_flags & EXT4_MAP_MAPPED) &&
3441	!ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
3442	iomap->flags |= IOMAP_F_MERGED;
3443
3444	/*
3445	* Flags passed to ext4_map_blocks() for direct I/O writes can result
3446	* in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3447	* set. In order for any allocated unwritten extents to be converted
3448	* into written extents correctly within the ->end_io() handler, we
3449	* need to ensure that the iomap->type is set appropriately. Hence, the
3450	* reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3451	* been set first.
3452	*/
3453	if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3454	iomap->type = IOMAP_UNWRITTEN;
3455	iomap->addr = (u64) map->m_pblk << blkbits;
3456	if (flags & IOMAP_DAX)
3457	iomap->addr += EXT4_SB(sb: inode->i_sb)->s_dax_part_off;
3458	} else if (map->m_flags & EXT4_MAP_MAPPED) {
3459	iomap->type = IOMAP_MAPPED;
3460	iomap->addr = (u64) map->m_pblk << blkbits;
3461	if (flags & IOMAP_DAX)
3462	iomap->addr += EXT4_SB(sb: inode->i_sb)->s_dax_part_off;
3463	} else if (map->m_flags & EXT4_MAP_DELAYED) {
3464	iomap->type = IOMAP_DELALLOC;
3465	iomap->addr = IOMAP_NULL_ADDR;
3466	} else {
3467	iomap->type = IOMAP_HOLE;
3468	iomap->addr = IOMAP_NULL_ADDR;
3469	}
3470	}
3471
3472	static int ext4_map_blocks_atomic_write_slow(handle_t *handle,
3473	struct inode *inode, struct ext4_map_blocks *map)
3474	{
3475	ext4_lblk_t m_lblk = map->m_lblk;
3476	unsigned int m_len = map->m_len;
3477	unsigned int mapped_len = 0, m_flags = 0;
3478	ext4_fsblk_t next_pblk;
3479	bool check_next_pblk = false;
3480	int ret = 0;
3481
3482	WARN_ON_ONCE(!ext4_has_feature_bigalloc(inode->i_sb));
3483
3484	/*
3485	* This is a slow path in case of mixed mapping. We use
3486	* EXT4_GET_BLOCKS_CREATE_ZERO flag here to make sure we get a single
3487	* contiguous mapped mapping. This will ensure any unwritten or hole
3488	* regions within the requested range is zeroed out and we return
3489	* a single contiguous mapped extent.
3490	*/
3491	m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3492
3493	do {
3494	ret = ext4_map_blocks(handle, inode, map, flags: m_flags);
3495	if (ret < 0 && ret != -ENOSPC)
3496	goto out_err;
3497	/*
3498	* This should never happen, but let's return an error code to
3499	* avoid an infinite loop in here.
3500	*/
3501	if (ret == 0) {
3502	ret = -EFSCORRUPTED;
3503	ext4_warning_inode(inode,
3504	"ext4_map_blocks() couldn't allocate blocks m_flags: 0x%x, ret:%d",
3505	m_flags, ret);
3506	goto out_err;
3507	}
3508	/*
3509	* With bigalloc we should never get ENOSPC nor discontiguous
3510	* physical extents.
3511	*/
3512	if ((check_next_pblk && next_pblk != map->m_pblk) ||
3513	ret == -ENOSPC) {
3514	ext4_warning_inode(inode,
3515	"Non-contiguous allocation detected: expected %llu, got %llu, "
3516	"or ext4_map_blocks() returned out of space ret: %d",
3517	next_pblk, map->m_pblk, ret);
3518	ret = -EFSCORRUPTED;
3519	goto out_err;
3520	}
3521	next_pblk = map->m_pblk + map->m_len;
3522	check_next_pblk = true;
3523
3524	mapped_len += map->m_len;
3525	map->m_lblk += map->m_len;
3526	map->m_len = m_len - mapped_len;
3527	} while (mapped_len < m_len);
3528
3529	/*
3530	* We might have done some work in above loop, so we need to query the
3531	* start of the physical extent, based on the origin m_lblk and m_len.
3532	* Let's also ensure we were able to allocate the required range for
3533	* mixed mapping case.
3534	*/
3535	map->m_lblk = m_lblk;
3536	map->m_len = m_len;
3537	map->m_flags = 0;
3538
3539	ret = ext4_map_blocks(handle, inode, map,
3540	EXT4_GET_BLOCKS_QUERY_LAST_IN_LEAF);
3541	if (ret != m_len) {
3542	ext4_warning_inode(inode,
3543	"allocation failed for atomic write request m_lblk:%u, m_len:%u, ret:%d\n",
3544	m_lblk, m_len, ret);
3545	ret = -EINVAL;
3546	}
3547	return ret;
3548
3549	out_err:
3550	/* reset map before returning an error */
3551	map->m_lblk = m_lblk;
3552	map->m_len = m_len;
3553	map->m_flags = 0;
3554	return ret;
3555	}
3556
3557	/*
3558	* ext4_map_blocks_atomic: Helper routine to ensure the entire requested
3559	* range in @map [lblk, lblk + len) is one single contiguous extent with no
3560	* mixed mappings.
3561	*
3562	* We first use m_flags passed to us by our caller (ext4_iomap_alloc()).
3563	* We only call EXT4_GET_BLOCKS_ZERO in the slow path, when the underlying
3564	* physical extent for the requested range does not have a single contiguous
3565	* mapping type i.e. (Hole, Mapped, or Unwritten) throughout.
3566	* In that case we will loop over the requested range to allocate and zero out
3567	* the unwritten / holes in between, to get a single mapped extent from
3568	* [m_lblk, m_lblk + m_len). Note that this is only possible because we know
3569	* this can be called only with bigalloc enabled filesystem where the underlying
3570	* cluster is already allocated. This avoids allocating discontiguous extents
3571	* in the slow path due to multiple calls to ext4_map_blocks().
3572	* The slow path is mostly non-performance critical path, so it should be ok to
3573	* loop using ext4_map_blocks() with appropriate flags to allocate & zero the
3574	* underlying short holes/unwritten extents within the requested range.
3575	*/
3576	static int ext4_map_blocks_atomic_write(handle_t *handle, struct inode *inode,
3577	struct ext4_map_blocks *map, int m_flags,
3578	bool *force_commit)
3579	{
3580	ext4_lblk_t m_lblk = map->m_lblk;
3581	unsigned int m_len = map->m_len;
3582	int ret = 0;
3583
3584	WARN_ON_ONCE(m_len > 1 && !ext4_has_feature_bigalloc(inode->i_sb));
3585
3586	ret = ext4_map_blocks(handle, inode, map, flags: m_flags);
3587	if (ret < 0 || ret == m_len)
3588	goto out;
3589	/*
3590	* This is a mixed mapping case where we were not able to allocate
3591	* a single contiguous extent. In that case let's reset requested
3592	* mapping and call the slow path.
3593	*/
3594	map->m_lblk = m_lblk;
3595	map->m_len = m_len;
3596	map->m_flags = 0;
3597
3598	/*
3599	* slow path means we have mixed mapping, that means we will need
3600	* to force txn commit.
3601	*/
3602	*force_commit = true;
3603	return ext4_map_blocks_atomic_write_slow(handle, inode, map);
3604	out:
3605	return ret;
3606	}
3607
3608	static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3609	unsigned int flags)
3610	{
3611	handle_t *handle;
3612	u8 blkbits = inode->i_blkbits;
3613	int ret, dio_credits, m_flags = 0, retries = 0;
3614	bool force_commit = false;
3615
3616	/*
3617	* Trim the mapping request to the maximum value that we can map at
3618	* once for direct I/O.
3619	*/
3620	if (map->m_len > DIO_MAX_BLOCKS)
3621	map->m_len = DIO_MAX_BLOCKS;
3622
3623	/*
3624	* journal credits estimation for atomic writes. We call
3625	* ext4_map_blocks(), to find if there could be a mixed mapping. If yes,
3626	* then let's assume the no. of pextents required can be m_len i.e.
3627	* every alternate block can be unwritten and hole.
3628	*/
3629	if (flags & IOMAP_ATOMIC) {
3630	unsigned int orig_mlen = map->m_len;
3631
3632	ret = ext4_map_blocks(NULL, inode, map, flags: 0);
3633	if (ret < 0)
3634	return ret;
3635	if (map->m_len < orig_mlen) {
3636	map->m_len = orig_mlen;
3637	dio_credits = ext4_meta_trans_blocks(inode, lblocks: orig_mlen,
3638	pextents: map->m_len);
3639	} else {
3640	dio_credits = ext4_chunk_trans_blocks(inode,
3641	nrblocks: map->m_len);
3642	}
3643	} else {
3644	dio_credits = ext4_chunk_trans_blocks(inode, nrblocks: map->m_len);
3645	}
3646
3647	retry:
3648	/*
3649	* Either we allocate blocks and then don't get an unwritten extent, so
3650	* in that case we have reserved enough credits. Or, the blocks are
3651	* already allocated and unwritten. In that case, the extent conversion
3652	* fits into the credits as well.
3653	*/
3654	handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3655	if (IS_ERR(ptr: handle))
3656	return PTR_ERR(ptr: handle);
3657
3658	/*
3659	* DAX and direct I/O are the only two operations that are currently
3660	* supported with IOMAP_WRITE.
3661	*/
3662	WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3663	if (flags & IOMAP_DAX)
3664	m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3665	/*
3666	* We use i_size instead of i_disksize here because delalloc writeback
3667	* can complete at any point during the I/O and subsequently push the
3668	* i_disksize out to i_size. This could be beyond where direct I/O is
3669	* happening and thus expose allocated blocks to direct I/O reads.
3670	*/
3671	else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3672	m_flags = EXT4_GET_BLOCKS_CREATE;
3673	else if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
3674	m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3675
3676	if (flags & IOMAP_ATOMIC)
3677	ret = ext4_map_blocks_atomic_write(handle, inode, map, m_flags,
3678	force_commit: &force_commit);
3679	else
3680	ret = ext4_map_blocks(handle, inode, map, flags: m_flags);
3681
3682	/*
3683	* We cannot fill holes in indirect tree based inodes as that could
3684	* expose stale data in the case of a crash. Use the magic error code
3685	* to fallback to buffered I/O.
3686	*/
3687	if (!m_flags && !ret)
3688	ret = -ENOTBLK;
3689
3690	ext4_journal_stop(handle);
3691	if (ret == -ENOSPC && ext4_should_retry_alloc(sb: inode->i_sb, retries: &retries))
3692	goto retry;
3693
3694	/*
3695	* Force commit the current transaction if the allocation spans a mixed
3696	* mapping range. This ensures any pending metadata updates (like
3697	* unwritten to written extents conversion) in this range are in
3698	* consistent state with the file data blocks, before performing the
3699	* actual write I/O. If the commit fails, the whole I/O must be aborted
3700	* to prevent any possible torn writes.
3701	*/
3702	if (ret > 0 && force_commit) {
3703	int ret2;
3704
3705	ret2 = ext4_force_commit(sb: inode->i_sb);
3706	if (ret2)
3707	return ret2;
3708	}
3709
3710	return ret;
3711	}
3712
3713
3714	static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3715	unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3716	{
3717	int ret;
3718	struct ext4_map_blocks map;
3719	u8 blkbits = inode->i_blkbits;
3720	unsigned int orig_mlen;
3721
3722	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3723	return -EINVAL;
3724
3725	if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3726	return -ERANGE;
3727
3728	/*
3729	* Calculate the first and last logical blocks respectively.
3730	*/
3731	map.m_lblk = offset >> blkbits;
3732	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3733	EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3734	orig_mlen = map.m_len;
3735
3736	if (flags & IOMAP_WRITE) {
3737	/*
3738	* We check here if the blocks are already allocated, then we
3739	* don't need to start a journal txn and we can directly return
3740	* the mapping information. This could boost performance
3741	* especially in multi-threaded overwrite requests.
3742	*/
3743	if (offset + length <= i_size_read(inode)) {
3744	ret = ext4_map_blocks(NULL, inode, map: &map, flags: 0);
3745	/*
3746	* For atomic writes the entire requested length should
3747	* be mapped.
3748	*/
3749	if (map.m_flags & EXT4_MAP_MAPPED) {
3750	if ((!(flags & IOMAP_ATOMIC) && ret > 0) ||
3751	(flags & IOMAP_ATOMIC && ret >= orig_mlen))
3752	goto out;
3753	}
3754	map.m_len = orig_mlen;
3755	}
3756	ret = ext4_iomap_alloc(inode, map: &map, flags);
3757	} else {
3758	/*
3759	* This can be called for overwrites path from
3760	* ext4_iomap_overwrite_begin().
3761	*/
3762	ret = ext4_map_blocks(NULL, inode, map: &map, flags: 0);
3763	}
3764
3765	if (ret < 0)
3766	return ret;
3767	out:
3768	/*
3769	* When inline encryption is enabled, sometimes I/O to an encrypted file
3770	* has to be broken up to guarantee DUN contiguity. Handle this by
3771	* limiting the length of the mapping returned.
3772	*/
3773	map.m_len = fscrypt_limit_io_blocks(inode, lblk: map.m_lblk, nr_blocks: map.m_len);
3774
3775	/*
3776	* Before returning to iomap, let's ensure the allocated mapping
3777	* covers the entire requested length for atomic writes.
3778	*/
3779	if (flags & IOMAP_ATOMIC) {
3780	if (map.m_len < (length >> blkbits)) {
3781	WARN_ON_ONCE(1);
3782	return -EINVAL;
3783	}
3784	}
3785	ext4_set_iomap(inode, iomap, map: &map, offset, length, flags);
3786
3787	return 0;
3788	}
3789
3790	static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3791	loff_t length, unsigned flags, struct iomap *iomap,
3792	struct iomap *srcmap)
3793	{
3794	int ret;
3795
3796	/*
3797	* Even for writes we don't need to allocate blocks, so just pretend
3798	* we are reading to save overhead of starting a transaction.
3799	*/
3800	flags &= ~IOMAP_WRITE;
3801	ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3802	WARN_ON_ONCE(!ret && iomap->type != IOMAP_MAPPED);
3803	return ret;
3804	}
3805
3806	static inline bool ext4_want_directio_fallback(unsigned flags, ssize_t written)
3807	{
3808	/* must be a directio to fall back to buffered */
3809	if ((flags & (IOMAP_WRITE | IOMAP_DIRECT)) !=
3810	(IOMAP_WRITE | IOMAP_DIRECT))
3811	return false;
3812
3813	/* atomic writes are all-or-nothing */
3814	if (flags & IOMAP_ATOMIC)
3815	return false;
3816
3817	/* can only try again if we wrote nothing */
3818	return written == 0;
3819	}
3820
3821	static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3822	ssize_t written, unsigned flags, struct iomap *iomap)
3823	{
3824	/*
3825	* Check to see whether an error occurred while writing out the data to
3826	* the allocated blocks. If so, return the magic error code for
3827	* non-atomic write so that we fallback to buffered I/O and attempt to
3828	* complete the remainder of the I/O.
3829	* For non-atomic writes, any blocks that may have been
3830	* allocated in preparation for the direct I/O will be reused during
3831	* buffered I/O. For atomic write, we never fallback to buffered-io.
3832	*/
3833	if (ext4_want_directio_fallback(flags, written))
3834	return -ENOTBLK;
3835
3836	return 0;
3837	}
3838
3839	const struct iomap_ops ext4_iomap_ops = {
3840	.iomap_begin = ext4_iomap_begin,
3841	.iomap_end = ext4_iomap_end,
3842	};
3843
3844	const struct iomap_ops ext4_iomap_overwrite_ops = {
3845	.iomap_begin = ext4_iomap_overwrite_begin,
3846	.iomap_end = ext4_iomap_end,
3847	};
3848
3849	static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3850	loff_t length, unsigned int flags,
3851	struct iomap *iomap, struct iomap *srcmap)
3852	{
3853	int ret;
3854	struct ext4_map_blocks map;
3855	u8 blkbits = inode->i_blkbits;
3856
3857	if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3858	return -EINVAL;
3859
3860	if (ext4_has_inline_data(inode)) {
3861	ret = ext4_inline_data_iomap(inode, iomap);
3862	if (ret != -EAGAIN) {
3863	if (ret == 0 && offset >= iomap->length)
3864	ret = -ENOENT;
3865	return ret;
3866	}
3867	}
3868
3869	/*
3870	* Calculate the first and last logical block respectively.
3871	*/
3872	map.m_lblk = offset >> blkbits;
3873	map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3874	EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3875
3876	/*
3877	* Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3878	* So handle it here itself instead of querying ext4_map_blocks().
3879	* Since ext4_map_blocks() will warn about it and will return
3880	* -EIO error.
3881	*/
3882	if (!(ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))) {
3883	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
3884
3885	if (offset >= sbi->s_bitmap_maxbytes) {
3886	map.m_flags = 0;
3887	goto set_iomap;
3888	}
3889	}
3890
3891	ret = ext4_map_blocks(NULL, inode, map: &map, flags: 0);
3892	if (ret < 0)
3893	return ret;
3894	set_iomap:
3895	ext4_set_iomap(inode, iomap, map: &map, offset, length, flags);
3896
3897	return 0;
3898	}
3899
3900	const struct iomap_ops ext4_iomap_report_ops = {
3901	.iomap_begin = ext4_iomap_begin_report,
3902	};
3903
3904	/*
3905	* For data=journal mode, folio should be marked dirty only when it was
3906	* writeably mapped. When that happens, it was already attached to the
3907	* transaction and marked as jbddirty (we take care of this in
3908	* ext4_page_mkwrite()). On transaction commit, we writeprotect page mappings
3909	* so we should have nothing to do here, except for the case when someone
3910	* had the page pinned and dirtied the page through this pin (e.g. by doing
3911	* direct IO to it). In that case we'd need to attach buffers here to the
3912	* transaction but we cannot due to lock ordering. We cannot just dirty the
3913	* folio and leave attached buffers clean, because the buffers' dirty state is
3914	* "definitive". We cannot just set the buffers dirty or jbddirty because all
3915	* the journalling code will explode. So what we do is to mark the folio
3916	* "pending dirty" and next time ext4_writepages() is called, attach buffers
3917	* to the transaction appropriately.
3918	*/
3919	static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3920	struct folio *folio)
3921	{
3922	WARN_ON_ONCE(!folio_buffers(folio));
3923	if (folio_maybe_dma_pinned(folio))
3924	folio_set_checked(folio);
3925	return filemap_dirty_folio(mapping, folio);
3926	}
3927
3928	static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3929	{
3930	WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3931	WARN_ON_ONCE(!folio_buffers(folio));
3932	return block_dirty_folio(mapping, folio);
3933	}
3934
3935	static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3936	struct file *file, sector_t *span)
3937	{
3938	return iomap_swapfile_activate(sis, swap_file: file, pagespan: span,
3939	ops: &ext4_iomap_report_ops);
3940	}
3941
3942	static const struct address_space_operations ext4_aops = {
3943	.read_folio = ext4_read_folio,
3944	.readahead = ext4_readahead,
3945	.writepages = ext4_writepages,
3946	.write_begin = ext4_write_begin,
3947	.write_end = ext4_write_end,
3948	.dirty_folio = ext4_dirty_folio,
3949	.bmap = ext4_bmap,
3950	.invalidate_folio = ext4_invalidate_folio,
3951	.release_folio = ext4_release_folio,
3952	.migrate_folio = buffer_migrate_folio,
3953	.is_partially_uptodate = block_is_partially_uptodate,
3954	.error_remove_folio = generic_error_remove_folio,
3955	.swap_activate = ext4_iomap_swap_activate,
3956	};
3957
3958	static const struct address_space_operations ext4_journalled_aops = {
3959	.read_folio = ext4_read_folio,
3960	.readahead = ext4_readahead,
3961	.writepages = ext4_writepages,
3962	.write_begin = ext4_write_begin,
3963	.write_end = ext4_journalled_write_end,
3964	.dirty_folio = ext4_journalled_dirty_folio,
3965	.bmap = ext4_bmap,
3966	.invalidate_folio = ext4_journalled_invalidate_folio,
3967	.release_folio = ext4_release_folio,
3968	.migrate_folio = buffer_migrate_folio_norefs,
3969	.is_partially_uptodate = block_is_partially_uptodate,
3970	.error_remove_folio = generic_error_remove_folio,
3971	.swap_activate = ext4_iomap_swap_activate,
3972	};
3973
3974	static const struct address_space_operations ext4_da_aops = {
3975	.read_folio = ext4_read_folio,
3976	.readahead = ext4_readahead,
3977	.writepages = ext4_writepages,
3978	.write_begin = ext4_da_write_begin,
3979	.write_end = ext4_da_write_end,
3980	.dirty_folio = ext4_dirty_folio,
3981	.bmap = ext4_bmap,
3982	.invalidate_folio = ext4_invalidate_folio,
3983	.release_folio = ext4_release_folio,
3984	.migrate_folio = buffer_migrate_folio,
3985	.is_partially_uptodate = block_is_partially_uptodate,
3986	.error_remove_folio = generic_error_remove_folio,
3987	.swap_activate = ext4_iomap_swap_activate,
3988	};
3989
3990	static const struct address_space_operations ext4_dax_aops = {
3991	.writepages = ext4_dax_writepages,
3992	.dirty_folio = noop_dirty_folio,
3993	.bmap = ext4_bmap,
3994	.swap_activate = ext4_iomap_swap_activate,
3995	};
3996
3997	void ext4_set_aops(struct inode *inode)
3998	{
3999	switch (ext4_inode_journal_mode(inode)) {
4000	case EXT4_INODE_ORDERED_DATA_MODE:
4001	case EXT4_INODE_WRITEBACK_DATA_MODE:
4002	break;
4003	case EXT4_INODE_JOURNAL_DATA_MODE:
4004	inode->i_mapping->a_ops = &ext4_journalled_aops;
4005	return;
4006	default:
4007	BUG();
4008	}
4009	if (IS_DAX(inode))
4010	inode->i_mapping->a_ops = &ext4_dax_aops;
4011	else if (test_opt(inode->i_sb, DELALLOC))
4012	inode->i_mapping->a_ops = &ext4_da_aops;
4013	else
4014	inode->i_mapping->a_ops = &ext4_aops;
4015	}
4016
4017	/*
4018	* Here we can't skip an unwritten buffer even though it usually reads zero
4019	* because it might have data in pagecache (eg, if called from ext4_zero_range,
4020	* ext4_punch_hole, etc) which needs to be properly zeroed out. Otherwise a
4021	* racing writeback can come later and flush the stale pagecache to disk.
4022	*/
4023	static int __ext4_block_zero_page_range(handle_t *handle,
4024	struct address_space *mapping, loff_t from, loff_t length)
4025	{
4026	unsigned int offset, blocksize, pos;
4027	ext4_lblk_t iblock;
4028	struct inode *inode = mapping->host;
4029	struct buffer_head *bh;
4030	struct folio *folio;
4031	int err = 0;
4032
4033	folio = __filemap_get_folio(mapping, index: from >> PAGE_SHIFT,
4034	FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
4035	gfp: mapping_gfp_constraint(mapping, gfp_mask: ~__GFP_FS));
4036	if (IS_ERR(ptr: folio))
4037	return PTR_ERR(ptr: folio);
4038
4039	blocksize = inode->i_sb->s_blocksize;
4040
4041	iblock = folio->index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
4042
4043	bh = folio_buffers(folio);
4044	if (!bh)
4045	bh = create_empty_buffers(folio, blocksize, b_state: 0);
4046
4047	/* Find the buffer that contains "offset" */
4048	offset = offset_in_folio(folio, from);
4049	pos = blocksize;
4050	while (offset >= pos) {
4051	bh = bh->b_this_page;
4052	iblock++;
4053	pos += blocksize;
4054	}
4055	if (buffer_freed(bh)) {
4056	BUFFER_TRACE(bh, "freed: skip");
4057	goto unlock;
4058	}
4059	if (!buffer_mapped(bh)) {
4060	BUFFER_TRACE(bh, "unmapped");
4061	ext4_get_block(inode, iblock, bh, create: 0);
4062	/* unmapped? It's a hole - nothing to do */
4063	if (!buffer_mapped(bh)) {
4064	BUFFER_TRACE(bh, "still unmapped");
4065	goto unlock;
4066	}
4067	}
4068
4069	/* Ok, it's mapped. Make sure it's up-to-date */
4070	if (folio_test_uptodate(folio))
4071	set_buffer_uptodate(bh);
4072
4073	if (!buffer_uptodate(bh)) {
4074	err = ext4_read_bh_lock(bh, op_flags: 0, wait: true);
4075	if (err)
4076	goto unlock;
4077	if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
4078	/* We expect the key to be set. */
4079	BUG_ON(!fscrypt_has_encryption_key(inode));
4080	err = fscrypt_decrypt_pagecache_blocks(folio,
4081	len: blocksize,
4082	offs: bh_offset(bh));
4083	if (err) {
4084	clear_buffer_uptodate(bh);
4085	goto unlock;
4086	}
4087	}
4088	}
4089	if (ext4_should_journal_data(inode)) {
4090	BUFFER_TRACE(bh, "get write access");
4091	err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
4092	EXT4_JTR_NONE);
4093	if (err)
4094	goto unlock;
4095	}
4096	folio_zero_range(folio, start: offset, length);
4097	BUFFER_TRACE(bh, "zeroed end of block");
4098
4099	if (ext4_should_journal_data(inode)) {
4100	err = ext4_dirty_journalled_data(handle, bh);
4101	} else {
4102	err = 0;
4103	mark_buffer_dirty(bh);
4104	if (ext4_should_order_data(inode))
4105	err = ext4_jbd2_inode_add_write(handle, inode, start_byte: from,
4106	length);
4107	}
4108
4109	unlock:
4110	folio_unlock(folio);
4111	folio_put(folio);
4112	return err;
4113	}
4114
4115	/*
4116	* ext4_block_zero_page_range() zeros out a mapping of length 'length'
4117	* starting from file offset 'from'. The range to be zero'd must
4118	* be contained with in one block. If the specified range exceeds
4119	* the end of the block it will be shortened to end of the block
4120	* that corresponds to 'from'
4121	*/
4122	static int ext4_block_zero_page_range(handle_t *handle,
4123	struct address_space *mapping, loff_t from, loff_t length)
4124	{
4125	struct inode *inode = mapping->host;
4126	unsigned offset = from & (PAGE_SIZE-1);
4127	unsigned blocksize = inode->i_sb->s_blocksize;
4128	unsigned max = blocksize - (offset & (blocksize - 1));
4129
4130	/*
4131	* correct length if it does not fall between
4132	* 'from' and the end of the block
4133	*/
4134	if (length > max || length < 0)
4135	length = max;
4136
4137	if (IS_DAX(inode)) {
4138	return dax_zero_range(inode, pos: from, len: length, NULL,
4139	ops: &ext4_iomap_ops);
4140	}
4141	return __ext4_block_zero_page_range(handle, mapping, from, length);
4142	}
4143
4144	/*
4145	* ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4146	* up to the end of the block which corresponds to `from'.
4147	* This required during truncate. We need to physically zero the tail end
4148	* of that block so it doesn't yield old data if the file is later grown.
4149	*/
4150	static int ext4_block_truncate_page(handle_t *handle,
4151	struct address_space *mapping, loff_t from)
4152	{
4153	unsigned offset = from & (PAGE_SIZE-1);
4154	unsigned length;
4155	unsigned blocksize;
4156	struct inode *inode = mapping->host;
4157
4158	/* If we are processing an encrypted inode during orphan list handling */
4159	if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
4160	return 0;
4161
4162	blocksize = inode->i_sb->s_blocksize;
4163	length = blocksize - (offset & (blocksize - 1));
4164
4165	return ext4_block_zero_page_range(handle, mapping, from, length);
4166	}
4167
4168	int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4169	loff_t lstart, loff_t length)
4170	{
4171	struct super_block *sb = inode->i_sb;
4172	struct address_space *mapping = inode->i_mapping;
4173	unsigned partial_start, partial_end;
4174	ext4_fsblk_t start, end;
4175	loff_t byte_end = (lstart + length - 1);
4176	int err = 0;
4177
4178	partial_start = lstart & (sb->s_blocksize - 1);
4179	partial_end = byte_end & (sb->s_blocksize - 1);
4180
4181	start = lstart >> sb->s_blocksize_bits;
4182	end = byte_end >> sb->s_blocksize_bits;
4183
4184	/* Handle partial zero within the single block */
4185	if (start == end &&
4186	(partial_start || (partial_end != sb->s_blocksize - 1))) {
4187	err = ext4_block_zero_page_range(handle, mapping,
4188	from: lstart, length);
4189	return err;
4190	}
4191	/* Handle partial zero out on the start of the range */
4192	if (partial_start) {
4193	err = ext4_block_zero_page_range(handle, mapping,
4194	from: lstart, length: sb->s_blocksize);
4195	if (err)
4196	return err;
4197	}
4198	/* Handle partial zero out on the end of the range */
4199	if (partial_end != sb->s_blocksize - 1)
4200	err = ext4_block_zero_page_range(handle, mapping,
4201	from: byte_end - partial_end,
4202	length: partial_end + 1);
4203	return err;
4204	}
4205
4206	int ext4_can_truncate(struct inode *inode)
4207	{
4208	if (S_ISREG(inode->i_mode))
4209	return 1;
4210	if (S_ISDIR(inode->i_mode))
4211	return 1;
4212	if (S_ISLNK(inode->i_mode))
4213	return !ext4_inode_is_fast_symlink(inode);
4214	return 0;
4215	}
4216
4217	/*
4218	* We have to make sure i_disksize gets properly updated before we truncate
4219	* page cache due to hole punching or zero range. Otherwise i_disksize update
4220	* can get lost as it may have been postponed to submission of writeback but
4221	* that will never happen after we truncate page cache.
4222	*/
4223	int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4224	loff_t len)
4225	{
4226	handle_t *handle;
4227	int ret;
4228
4229	loff_t size = i_size_read(inode);
4230
4231	WARN_ON(!inode_is_locked(inode));
4232	if (offset > size || offset + len < size)
4233	return 0;
4234
4235	if (EXT4_I(inode)->i_disksize >= size)
4236	return 0;
4237
4238	handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4239	if (IS_ERR(ptr: handle))
4240	return PTR_ERR(ptr: handle);
4241	ext4_update_i_disksize(inode, newsize: size);
4242	ret = ext4_mark_inode_dirty(handle, inode);
4243	ext4_journal_stop(handle);
4244
4245	return ret;
4246	}
4247
4248	static inline void ext4_truncate_folio(struct inode *inode,
4249	loff_t start, loff_t end)
4250	{
4251	unsigned long blocksize = i_blocksize(node: inode);
4252	struct folio *folio;
4253
4254	/* Nothing to be done if no complete block needs to be truncated. */
4255	if (round_up(start, blocksize) >= round_down(end, blocksize))
4256	return;
4257
4258	folio = filemap_lock_folio(mapping: inode->i_mapping, index: start >> PAGE_SHIFT);
4259	if (IS_ERR(ptr: folio))
4260	return;
4261
4262	if (folio_mkclean(folio))
4263	folio_mark_dirty(folio);
4264	folio_unlock(folio);
4265	folio_put(folio);
4266	}
4267
4268	int ext4_truncate_page_cache_block_range(struct inode *inode,
4269	loff_t start, loff_t end)
4270	{
4271	unsigned long blocksize = i_blocksize(node: inode);
4272	int ret;
4273
4274	/*
4275	* For journalled data we need to write (and checkpoint) pages
4276	* before discarding page cache to avoid inconsitent data on disk
4277	* in case of crash before freeing or unwritten converting trans
4278	* is committed.
4279	*/
4280	if (ext4_should_journal_data(inode)) {
4281	ret = filemap_write_and_wait_range(mapping: inode->i_mapping, lstart: start,
4282	lend: end - 1);
4283	if (ret)
4284	return ret;
4285	goto truncate_pagecache;
4286	}
4287
4288	/*
4289	* If the block size is less than the page size, the file's mapped
4290	* blocks within one page could be freed or converted to unwritten.
4291	* So it's necessary to remove writable userspace mappings, and then
4292	* ext4_page_mkwrite() can be called during subsequent write access
4293	* to these partial folios.
4294	*/
4295	if (!IS_ALIGNED(start | end, PAGE_SIZE) &&
4296	blocksize < PAGE_SIZE && start < inode->i_size) {
4297	loff_t page_boundary = round_up(start, PAGE_SIZE);
4298
4299	ext4_truncate_folio(inode, start, min(page_boundary, end));
4300	if (end > page_boundary)
4301	ext4_truncate_folio(inode,
4302	round_down(end, PAGE_SIZE), end);
4303	}
4304
4305	truncate_pagecache:
4306	truncate_pagecache_range(inode, offset: start, end: end - 1);
4307	return 0;
4308	}
4309
4310	static void ext4_wait_dax_page(struct inode *inode)
4311	{
4312	filemap_invalidate_unlock(mapping: inode->i_mapping);
4313	schedule();
4314	filemap_invalidate_lock(mapping: inode->i_mapping);
4315	}
4316
4317	int ext4_break_layouts(struct inode *inode)
4318	{
4319	if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
4320	return -EINVAL;
4321
4322	return dax_break_layout_inode(inode, cb: ext4_wait_dax_page);
4323	}
4324
4325	/*
4326	* ext4_punch_hole: punches a hole in a file by releasing the blocks
4327	* associated with the given offset and length
4328	*
4329	* @inode: File inode
4330	* @offset: The offset where the hole will begin
4331	* @len: The length of the hole
4332	*
4333	* Returns: 0 on success or negative on failure
4334	*/
4335
4336	int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
4337	{
4338	struct inode *inode = file_inode(f: file);
4339	struct super_block *sb = inode->i_sb;
4340	ext4_lblk_t start_lblk, end_lblk;
4341	loff_t max_end = sb->s_maxbytes;
4342	loff_t end = offset + length;
4343	handle_t *handle;
4344	unsigned int credits;
4345	int ret;
4346
4347	trace_ext4_punch_hole(inode, offset, len: length, mode: 0);
4348	WARN_ON_ONCE(!inode_is_locked(inode));
4349
4350	/*
4351	* For indirect-block based inodes, make sure that the hole within
4352	* one block before last range.
4353	*/
4354	if (!ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
4355	max_end = EXT4_SB(sb)->s_bitmap_maxbytes - sb->s_blocksize;
4356
4357	/* No need to punch hole beyond i_size */
4358	if (offset >= inode->i_size || offset >= max_end)
4359	return 0;
4360
4361	/*
4362	* If the hole extends beyond i_size, set the hole to end after
4363	* the page that contains i_size.
4364	*/
4365	if (end > inode->i_size)
4366	end = round_up(inode->i_size, PAGE_SIZE);
4367	if (end > max_end)
4368	end = max_end;
4369	length = end - offset;
4370
4371	/*
4372	* Attach jinode to inode for jbd2 if we do any zeroing of partial
4373	* block.
4374	*/
4375	if (!IS_ALIGNED(offset | end, sb->s_blocksize)) {
4376	ret = ext4_inode_attach_jinode(inode);
4377	if (ret < 0)
4378	return ret;
4379	}
4380
4381
4382	ret = ext4_update_disksize_before_punch(inode, offset, len: length);
4383	if (ret)
4384	return ret;
4385
4386	/* Now release the pages and zero block aligned part of pages*/
4387	ret = ext4_truncate_page_cache_block_range(inode, start: offset, end);
4388	if (ret)
4389	return ret;
4390
4391	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
4392	credits = ext4_writepage_trans_blocks(inode);
4393	else
4394	credits = ext4_blocks_for_truncate(inode);
4395	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4396	if (IS_ERR(ptr: handle)) {
4397	ret = PTR_ERR(ptr: handle);
4398	ext4_std_error(sb, ret);
4399	return ret;
4400	}
4401
4402	ret = ext4_zero_partial_blocks(handle, inode, lstart: offset, length);
4403	if (ret)
4404	goto out_handle;
4405
4406	/* If there are blocks to remove, do it */
4407	start_lblk = EXT4_B_TO_LBLK(inode, offset);
4408	end_lblk = end >> inode->i_blkbits;
4409
4410	if (end_lblk > start_lblk) {
4411	ext4_lblk_t hole_len = end_lblk - start_lblk;
4412
4413	ext4_fc_track_inode(handle, inode);
4414	ext4_check_map_extents_env(inode);
4415	down_write(sem: &EXT4_I(inode)->i_data_sem);
4416	ext4_discard_preallocations(inode);
4417
4418	ext4_es_remove_extent(inode, lblk: start_lblk, len: hole_len);
4419
4420	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
4421	ret = ext4_ext_remove_space(inode, start: start_lblk,
4422	end: end_lblk - 1);
4423	else
4424	ret = ext4_ind_remove_space(handle, inode, start: start_lblk,
4425	end: end_lblk);
4426	if (ret) {
4427	up_write(sem: &EXT4_I(inode)->i_data_sem);
4428	goto out_handle;
4429	}
4430
4431	ext4_es_insert_extent(inode, lblk: start_lblk, len: hole_len, pblk: ~0,
4432	EXTENT_STATUS_HOLE, delalloc_reserve_used: 0);
4433	up_write(sem: &EXT4_I(inode)->i_data_sem);
4434	}
4435	ext4_fc_track_range(handle, inode, start: start_lblk, end: end_lblk);
4436
4437	ret = ext4_mark_inode_dirty(handle, inode);
4438	if (unlikely(ret))
4439	goto out_handle;
4440
4441	ext4_update_inode_fsync_trans(handle, inode, datasync: 1);
4442	if (IS_SYNC(inode))
4443	ext4_handle_sync(handle);
4444	out_handle:
4445	ext4_journal_stop(handle);
4446	return ret;
4447	}
4448
4449	int ext4_inode_attach_jinode(struct inode *inode)
4450	{
4451	struct ext4_inode_info *ei = EXT4_I(inode);
4452	struct jbd2_inode *jinode;
4453
4454	if (ei->jinode || !EXT4_SB(sb: inode->i_sb)->s_journal)
4455	return 0;
4456
4457	jinode = jbd2_alloc_inode(GFP_KERNEL);
4458	spin_lock(lock: &inode->i_lock);
4459	if (!ei->jinode) {
4460	if (!jinode) {
4461	spin_unlock(lock: &inode->i_lock);
4462	return -ENOMEM;
4463	}
4464	ei->jinode = jinode;
4465	jbd2_journal_init_jbd_inode(jinode: ei->jinode, inode);
4466	jinode = NULL;
4467	}
4468	spin_unlock(lock: &inode->i_lock);
4469	if (unlikely(jinode != NULL))
4470	jbd2_free_inode(jinode);
4471	return 0;
4472	}
4473
4474	/*
4475	* ext4_truncate()
4476	*
4477	* We block out ext4_get_block() block instantiations across the entire
4478	* transaction, and VFS/VM ensures that ext4_truncate() cannot run
4479	* simultaneously on behalf of the same inode.
4480	*
4481	* As we work through the truncate and commit bits of it to the journal there
4482	* is one core, guiding principle: the file's tree must always be consistent on
4483	* disk. We must be able to restart the truncate after a crash.
4484	*
4485	* The file's tree may be transiently inconsistent in memory (although it
4486	* probably isn't), but whenever we close off and commit a journal transaction,
4487	* the contents of (the filesystem + the journal) must be consistent and
4488	* restartable. It's pretty simple, really: bottom up, right to left (although
4489	* left-to-right works OK too).
4490	*
4491	* Note that at recovery time, journal replay occurs *before* the restart of
4492	* truncate against the orphan inode list.
4493	*
4494	* The committed inode has the new, desired i_size (which is the same as
4495	* i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4496	* that this inode's truncate did not complete and it will again call
4497	* ext4_truncate() to have another go. So there will be instantiated blocks
4498	* to the right of the truncation point in a crashed ext4 filesystem. But
4499	* that's fine - as long as they are linked from the inode, the post-crash
4500	* ext4_truncate() run will find them and release them.
4501	*/
4502	int ext4_truncate(struct inode *inode)
4503	{
4504	struct ext4_inode_info *ei = EXT4_I(inode);
4505	unsigned int credits;
4506	int err = 0, err2;
4507	handle_t *handle;
4508	struct address_space *mapping = inode->i_mapping;
4509
4510	/*
4511	* There is a possibility that we're either freeing the inode
4512	* or it's a completely new inode. In those cases we might not
4513	* have i_rwsem locked because it's not necessary.
4514	*/
4515	if (!(inode->i_state & (I_NEW|I_FREEING)))
4516	WARN_ON(!inode_is_locked(inode));
4517	trace_ext4_truncate_enter(inode);
4518
4519	if (!ext4_can_truncate(inode))
4520	goto out_trace;
4521
4522	if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4523	ext4_set_inode_state(inode, bit: EXT4_STATE_DA_ALLOC_CLOSE);
4524
4525	if (ext4_has_inline_data(inode)) {
4526	int has_inline = 1;
4527
4528	err = ext4_inline_data_truncate(inode, has_inline: &has_inline);
4529	if (err || has_inline)
4530	goto out_trace;
4531	}
4532
4533	/* If we zero-out tail of the page, we have to create jinode for jbd2 */
4534	if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4535	err = ext4_inode_attach_jinode(inode);
4536	if (err)
4537	goto out_trace;
4538	}
4539
4540	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
4541	credits = ext4_writepage_trans_blocks(inode);
4542	else
4543	credits = ext4_blocks_for_truncate(inode);
4544
4545	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4546	if (IS_ERR(ptr: handle)) {
4547	err = PTR_ERR(ptr: handle);
4548	goto out_trace;
4549	}
4550
4551	if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4552	ext4_block_truncate_page(handle, mapping, from: inode->i_size);
4553
4554	/*
4555	* We add the inode to the orphan list, so that if this
4556	* truncate spans multiple transactions, and we crash, we will
4557	* resume the truncate when the filesystem recovers. It also
4558	* marks the inode dirty, to catch the new size.
4559	*
4560	* Implication: the file must always be in a sane, consistent
4561	* truncatable state while each transaction commits.
4562	*/
4563	err = ext4_orphan_add(handle, inode);
4564	if (err)
4565	goto out_stop;
4566
4567	ext4_fc_track_inode(handle, inode);
4568	ext4_check_map_extents_env(inode);
4569
4570	down_write(sem: &EXT4_I(inode)->i_data_sem);
4571	ext4_discard_preallocations(inode);
4572
4573	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
4574	err = ext4_ext_truncate(handle, inode);
4575	else
4576	ext4_ind_truncate(handle, inode);
4577
4578	up_write(sem: &ei->i_data_sem);
4579	if (err)
4580	goto out_stop;
4581
4582	if (IS_SYNC(inode))
4583	ext4_handle_sync(handle);
4584
4585	out_stop:
4586	/*
4587	* If this was a simple ftruncate() and the file will remain alive,
4588	* then we need to clear up the orphan record which we created above.
4589	* However, if this was a real unlink then we were called by
4590	* ext4_evict_inode(), and we allow that function to clean up the
4591	* orphan info for us.
4592	*/
4593	if (inode->i_nlink)
4594	ext4_orphan_del(handle, inode);
4595
4596	inode_set_mtime_to_ts(inode, ts: inode_set_ctime_current(inode));
4597	err2 = ext4_mark_inode_dirty(handle, inode);
4598	if (unlikely(err2 && !err))
4599	err = err2;
4600	ext4_journal_stop(handle);
4601
4602	out_trace:
4603	trace_ext4_truncate_exit(inode);
4604	return err;
4605	}
4606
4607	static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4608	{
4609	if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4610	return inode_peek_iversion_raw(inode);
4611	else
4612	return inode_peek_iversion(inode);
4613	}
4614
4615	static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4616	struct ext4_inode_info *ei)
4617	{
4618	struct inode *inode = &(ei->vfs_inode);
4619	u64 i_blocks = READ_ONCE(inode->i_blocks);
4620	struct super_block *sb = inode->i_sb;
4621
4622	if (i_blocks <= ~0U) {
4623	/*
4624	* i_blocks can be represented in a 32 bit variable
4625	* as multiple of 512 bytes
4626	*/
4627	raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4628	raw_inode->i_blocks_high = 0;
4629	ext4_clear_inode_flag(inode, bit: EXT4_INODE_HUGE_FILE);
4630	return 0;
4631	}
4632
4633	/*
4634	* This should never happen since sb->s_maxbytes should not have
4635	* allowed this, sb->s_maxbytes was set according to the huge_file
4636	* feature in ext4_fill_super().
4637	*/
4638	if (!ext4_has_feature_huge_file(sb))
4639	return -EFSCORRUPTED;
4640
4641	if (i_blocks <= 0xffffffffffffULL) {
4642	/*
4643	* i_blocks can be represented in a 48 bit variable
4644	* as multiple of 512 bytes
4645	*/
4646	raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4647	raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4648	ext4_clear_inode_flag(inode, bit: EXT4_INODE_HUGE_FILE);
4649	} else {
4650	ext4_set_inode_flag(inode, bit: EXT4_INODE_HUGE_FILE);
4651	/* i_block is stored in file system block size */
4652	i_blocks = i_blocks >> (inode->i_blkbits - 9);
4653	raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4654	raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4655	}
4656	return 0;
4657	}
4658
4659	static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4660	{
4661	struct ext4_inode_info *ei = EXT4_I(inode);
4662	uid_t i_uid;
4663	gid_t i_gid;
4664	projid_t i_projid;
4665	int block;
4666	int err;
4667
4668	err = ext4_inode_blocks_set(raw_inode, ei);
4669
4670	raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4671	i_uid = i_uid_read(inode);
4672	i_gid = i_gid_read(inode);
4673	i_projid = from_kprojid(to: &init_user_ns, projid: ei->i_projid);
4674	if (!(test_opt(inode->i_sb, NO_UID32))) {
4675	raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4676	raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4677	/*
4678	* Fix up interoperability with old kernels. Otherwise,
4679	* old inodes get re-used with the upper 16 bits of the
4680	* uid/gid intact.
4681	*/
4682	if (ei->i_dtime && list_empty(head: &ei->i_orphan)) {
4683	raw_inode->i_uid_high = 0;
4684	raw_inode->i_gid_high = 0;
4685	} else {
4686	raw_inode->i_uid_high =
4687	cpu_to_le16(high_16_bits(i_uid));
4688	raw_inode->i_gid_high =
4689	cpu_to_le16(high_16_bits(i_gid));
4690	}
4691	} else {
4692	raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4693	raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4694	raw_inode->i_uid_high = 0;
4695	raw_inode->i_gid_high = 0;
4696	}
4697	raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4698
4699	EXT4_INODE_SET_CTIME(inode, raw_inode);
4700	EXT4_INODE_SET_MTIME(inode, raw_inode);
4701	EXT4_INODE_SET_ATIME(inode, raw_inode);
4702	EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4703
4704	raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4705	raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4706	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4707	raw_inode->i_file_acl_high =
4708	cpu_to_le16(ei->i_file_acl >> 32);
4709	raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4710	ext4_isize_set(raw_inode, i_size: ei->i_disksize);
4711
4712	raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4713	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4714	if (old_valid_dev(dev: inode->i_rdev)) {
4715	raw_inode->i_block[0] =
4716	cpu_to_le32(old_encode_dev(inode->i_rdev));
4717	raw_inode->i_block[1] = 0;
4718	} else {
4719	raw_inode->i_block[0] = 0;
4720	raw_inode->i_block[1] =
4721	cpu_to_le32(new_encode_dev(inode->i_rdev));
4722	raw_inode->i_block[2] = 0;
4723	}
4724	} else if (!ext4_has_inline_data(inode)) {
4725	for (block = 0; block < EXT4_N_BLOCKS; block++)
4726	raw_inode->i_block[block] = ei->i_data[block];
4727	}
4728
4729	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4730	u64 ivers = ext4_inode_peek_iversion(inode);
4731
4732	raw_inode->i_disk_version = cpu_to_le32(ivers);
4733	if (ei->i_extra_isize) {
4734	if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4735	raw_inode->i_version_hi =
4736	cpu_to_le32(ivers >> 32);
4737	raw_inode->i_extra_isize =
4738	cpu_to_le16(ei->i_extra_isize);
4739	}
4740	}
4741
4742	if (i_projid != EXT4_DEF_PROJID &&
4743	!ext4_has_feature_project(sb: inode->i_sb))
4744	err = err ?: -EFSCORRUPTED;
4745
4746	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4747	EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4748	raw_inode->i_projid = cpu_to_le32(i_projid);
4749
4750	ext4_inode_csum_set(inode, raw: raw_inode, ei);
4751	return err;
4752	}
4753
4754	/*
4755	* ext4_get_inode_loc returns with an extra refcount against the inode's
4756	* underlying buffer_head on success. If we pass 'inode' and it does not
4757	* have in-inode xattr, we have all inode data in memory that is needed
4758	* to recreate the on-disk version of this inode.
4759	*/
4760	static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4761	struct inode *inode, struct ext4_iloc *iloc,
4762	ext4_fsblk_t *ret_block)
4763	{
4764	struct ext4_group_desc *gdp;
4765	struct buffer_head *bh;
4766	ext4_fsblk_t block;
4767	struct blk_plug plug;
4768	int inodes_per_block, inode_offset;
4769
4770	iloc->bh = NULL;
4771	if (ino < EXT4_ROOT_INO ||
4772	ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4773	return -EFSCORRUPTED;
4774
4775	iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4776	gdp = ext4_get_group_desc(sb, block_group: iloc->block_group, NULL);
4777	if (!gdp)
4778	return -EIO;
4779
4780	/*
4781	* Figure out the offset within the block group inode table
4782	*/
4783	inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4784	inode_offset = ((ino - 1) %
4785	EXT4_INODES_PER_GROUP(sb));
4786	iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4787
4788	block = ext4_inode_table(sb, bg: gdp);
4789	if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4790	(block >= ext4_blocks_count(es: EXT4_SB(sb)->s_es))) {
4791	ext4_error(sb, "Invalid inode table block %llu in "
4792	"block_group %u", block, iloc->block_group);
4793	return -EFSCORRUPTED;
4794	}
4795	block += (inode_offset / inodes_per_block);
4796
4797	bh = sb_getblk(sb, block);
4798	if (unlikely(!bh))
4799	return -ENOMEM;
4800	if (ext4_buffer_uptodate(bh))
4801	goto has_buffer;
4802
4803	lock_buffer(bh);
4804	if (ext4_buffer_uptodate(bh)) {
4805	/* Someone brought it uptodate while we waited */
4806	unlock_buffer(bh);
4807	goto has_buffer;
4808	}
4809
4810	/*
4811	* If we have all information of the inode in memory and this
4812	* is the only valid inode in the block, we need not read the
4813	* block.
4814	*/
4815	if (inode && !ext4_test_inode_state(inode, bit: EXT4_STATE_XATTR)) {
4816	struct buffer_head *bitmap_bh;
4817	int i, start;
4818
4819	start = inode_offset & ~(inodes_per_block - 1);
4820
4821	/* Is the inode bitmap in cache? */
4822	bitmap_bh = sb_getblk(sb, block: ext4_inode_bitmap(sb, bg: gdp));
4823	if (unlikely(!bitmap_bh))
4824	goto make_io;
4825
4826	/*
4827	* If the inode bitmap isn't in cache then the
4828	* optimisation may end up performing two reads instead
4829	* of one, so skip it.
4830	*/
4831	if (!buffer_uptodate(bh: bitmap_bh)) {
4832	brelse(bh: bitmap_bh);
4833	goto make_io;
4834	}
4835	for (i = start; i < start + inodes_per_block; i++) {
4836	if (i == inode_offset)
4837	continue;
4838	if (ext4_test_bit(nr: i, addr: bitmap_bh->b_data))
4839	break;
4840	}
4841	brelse(bh: bitmap_bh);
4842	if (i == start + inodes_per_block) {
4843	struct ext4_inode *raw_inode =
4844	(struct ext4_inode *) (bh->b_data + iloc->offset);
4845
4846	/* all other inodes are free, so skip I/O */
4847	memset(bh->b_data, 0, bh->b_size);
4848	if (!ext4_test_inode_state(inode, bit: EXT4_STATE_NEW))
4849	ext4_fill_raw_inode(inode, raw_inode);
4850	set_buffer_uptodate(bh);
4851	unlock_buffer(bh);
4852	goto has_buffer;
4853	}
4854	}
4855
4856	make_io:
4857	/*
4858	* If we need to do any I/O, try to pre-readahead extra
4859	* blocks from the inode table.
4860	*/
4861	blk_start_plug(&plug);
4862	if (EXT4_SB(sb)->s_inode_readahead_blks) {
4863	ext4_fsblk_t b, end, table;
4864	unsigned num;
4865	__u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4866
4867	table = ext4_inode_table(sb, bg: gdp);
4868	/* s_inode_readahead_blks is always a power of 2 */
4869	b = block & ~((ext4_fsblk_t) ra_blks - 1);
4870	if (table > b)
4871	b = table;
4872	end = b + ra_blks;
4873	num = EXT4_INODES_PER_GROUP(sb);
4874	if (ext4_has_group_desc_csum(sb))
4875	num -= ext4_itable_unused_count(sb, bg: gdp);
4876	table += num / inodes_per_block;
4877	if (end > table)
4878	end = table;
4879	while (b <= end)
4880	ext4_sb_breadahead_unmovable(sb, block: b++);
4881	}
4882
4883	/*
4884	* There are other valid inodes in the buffer, this inode
4885	* has in-inode xattrs, or we don't have this inode in memory.
4886	* Read the block from disk.
4887	*/
4888	trace_ext4_load_inode(sb, ino);
4889	ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL,
4890	simu_fail: ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO));
4891	blk_finish_plug(&plug);
4892	wait_on_buffer(bh);
4893	if (!buffer_uptodate(bh)) {
4894	if (ret_block)
4895	*ret_block = block;
4896	brelse(bh);
4897	return -EIO;
4898	}
4899	has_buffer:
4900	iloc->bh = bh;
4901	return 0;
4902	}
4903
4904	static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4905	struct ext4_iloc *iloc)
4906	{
4907	ext4_fsblk_t err_blk = 0;
4908	int ret;
4909
4910	ret = __ext4_get_inode_loc(sb: inode->i_sb, ino: inode->i_ino, NULL, iloc,
4911	ret_block: &err_blk);
4912
4913	if (ret == -EIO)
4914	ext4_error_inode_block(inode, err_blk, EIO,
4915	"unable to read itable block");
4916
4917	return ret;
4918	}
4919
4920	int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4921	{
4922	ext4_fsblk_t err_blk = 0;
4923	int ret;
4924
4925	ret = __ext4_get_inode_loc(sb: inode->i_sb, ino: inode->i_ino, inode, iloc,
4926	ret_block: &err_blk);
4927
4928	if (ret == -EIO)
4929	ext4_error_inode_block(inode, err_blk, EIO,
4930	"unable to read itable block");
4931
4932	return ret;
4933	}
4934
4935
4936	int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4937	struct ext4_iloc *iloc)
4938	{
4939	return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4940	}
4941
4942	static bool ext4_should_enable_dax(struct inode *inode)
4943	{
4944	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
4945
4946	if (test_opt2(inode->i_sb, DAX_NEVER))
4947	return false;
4948	if (!S_ISREG(inode->i_mode))
4949	return false;
4950	if (ext4_should_journal_data(inode))
4951	return false;
4952	if (ext4_has_inline_data(inode))
4953	return false;
4954	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_ENCRYPT))
4955	return false;
4956	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_VERITY))
4957	return false;
4958	if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4959	return false;
4960	if (test_opt(inode->i_sb, DAX_ALWAYS))
4961	return true;
4962
4963	return ext4_test_inode_flag(inode, bit: EXT4_INODE_DAX);
4964	}
4965
4966	void ext4_set_inode_flags(struct inode *inode, bool init)
4967	{
4968	unsigned int flags = EXT4_I(inode)->i_flags;
4969	unsigned int new_fl = 0;
4970
4971	WARN_ON_ONCE(IS_DAX(inode) && init);
4972
4973	if (flags & EXT4_SYNC_FL)
4974	new_fl |= S_SYNC;
4975	if (flags & EXT4_APPEND_FL)
4976	new_fl |= S_APPEND;
4977	if (flags & EXT4_IMMUTABLE_FL)
4978	new_fl |= S_IMMUTABLE;
4979	if (flags & EXT4_NOATIME_FL)
4980	new_fl |= S_NOATIME;
4981	if (flags & EXT4_DIRSYNC_FL)
4982	new_fl |= S_DIRSYNC;
4983
4984	/* Because of the way inode_set_flags() works we must preserve S_DAX
4985	* here if already set. */
4986	new_fl |= (inode->i_flags & S_DAX);
4987	if (init && ext4_should_enable_dax(inode))
4988	new_fl |= S_DAX;
4989
4990	if (flags & EXT4_ENCRYPT_FL)
4991	new_fl |= S_ENCRYPTED;
4992	if (flags & EXT4_CASEFOLD_FL)
4993	new_fl |= S_CASEFOLD;
4994	if (flags & EXT4_VERITY_FL)
4995	new_fl |= S_VERITY;
4996	inode_set_flags(inode, flags: new_fl,
4997	S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4998	S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4999	}
5000
5001	static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
5002	struct ext4_inode_info *ei)
5003	{
5004	blkcnt_t i_blocks ;
5005	struct inode *inode = &(ei->vfs_inode);
5006	struct super_block *sb = inode->i_sb;
5007
5008	if (ext4_has_feature_huge_file(sb)) {
5009	/* we are using combined 48 bit field */
5010	i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
5011	le32_to_cpu(raw_inode->i_blocks_lo);
5012	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_HUGE_FILE)) {
5013	/* i_blocks represent file system block size */
5014	return i_blocks << (inode->i_blkbits - 9);
5015	} else {
5016	return i_blocks;
5017	}
5018	} else {
5019	return le32_to_cpu(raw_inode->i_blocks_lo);
5020	}
5021	}
5022
5023	static inline int ext4_iget_extra_inode(struct inode *inode,
5024	struct ext4_inode *raw_inode,
5025	struct ext4_inode_info *ei)
5026	{
5027	__le32 *magic = (void *)raw_inode +
5028	EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
5029
5030	if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
5031	*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
5032	int err;
5033
5034	err = xattr_check_inode(inode, IHDR(inode, raw_inode),
5035	ITAIL(inode, raw_inode));
5036	if (err)
5037	return err;
5038
5039	ext4_set_inode_state(inode, bit: EXT4_STATE_XATTR);
5040	err = ext4_find_inline_data_nolock(inode);
5041	if (!err && ext4_has_inline_data(inode))
5042	ext4_set_inode_state(inode, bit: EXT4_STATE_MAY_INLINE_DATA);
5043	return err;
5044	} else
5045	EXT4_I(inode)->i_inline_off = 0;
5046	return 0;
5047	}
5048
5049	int ext4_get_projid(struct inode *inode, kprojid_t *projid)
5050	{
5051	if (!ext4_has_feature_project(sb: inode->i_sb))
5052	return -EOPNOTSUPP;
5053	*projid = EXT4_I(inode)->i_projid;
5054	return 0;
5055	}
5056
5057	/*
5058	* ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
5059	* refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
5060	* set.
5061	*/
5062	static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
5063	{
5064	if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
5065	inode_set_iversion_raw(inode, val);
5066	else
5067	inode_set_iversion_queried(inode, val);
5068	}
5069
5070	static int check_igot_inode(struct inode *inode, ext4_iget_flags flags,
5071	const char *function, unsigned int line)
5072	{
5073	const char *err_str;
5074
5075	if (flags & EXT4_IGET_EA_INODE) {
5076	if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
5077	err_str = "missing EA_INODE flag";
5078	goto error;
5079	}
5080	if (ext4_test_inode_state(inode, bit: EXT4_STATE_XATTR) ||
5081	EXT4_I(inode)->i_file_acl) {
5082	err_str = "ea_inode with extended attributes";
5083	goto error;
5084	}
5085	} else {
5086	if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
5087	/*
5088	* open_by_handle_at() could provide an old inode number
5089	* that has since been reused for an ea_inode; this does
5090	* not indicate filesystem corruption
5091	*/
5092	if (flags & EXT4_IGET_HANDLE)
5093	return -ESTALE;
5094	err_str = "unexpected EA_INODE flag";
5095	goto error;
5096	}
5097	}
5098	if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD)) {
5099	err_str = "unexpected bad inode w/o EXT4_IGET_BAD";
5100	goto error;
5101	}
5102	return 0;
5103
5104	error:
5105	ext4_error_inode(inode, function, line, 0, "%s", err_str);
5106	return -EFSCORRUPTED;
5107	}
5108
5109	bool ext4_should_enable_large_folio(struct inode *inode)
5110	{
5111	struct super_block *sb = inode->i_sb;
5112
5113	if (!S_ISREG(inode->i_mode))
5114	return false;
5115	if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
5116	ext4_test_inode_flag(inode, bit: EXT4_INODE_JOURNAL_DATA))
5117	return false;
5118	if (ext4_has_feature_verity(sb))
5119	return false;
5120	if (ext4_has_feature_encrypt(sb))
5121	return false;
5122
5123	return true;
5124	}
5125
5126	struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
5127	ext4_iget_flags flags, const char *function,
5128	unsigned int line)
5129	{
5130	struct ext4_iloc iloc;
5131	struct ext4_inode *raw_inode;
5132	struct ext4_inode_info *ei;
5133	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
5134	struct inode *inode;
5135	journal_t *journal = EXT4_SB(sb)->s_journal;
5136	long ret;
5137	loff_t size;
5138	int block;
5139	uid_t i_uid;
5140	gid_t i_gid;
5141	projid_t i_projid;
5142
5143	if ((!(flags & EXT4_IGET_SPECIAL) && is_special_ino(sb, ino)) ||
5144	(ino < EXT4_ROOT_INO) ||
5145	(ino > le32_to_cpu(es->s_inodes_count))) {
5146	if (flags & EXT4_IGET_HANDLE)
5147	return ERR_PTR(error: -ESTALE);
5148	__ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
5149	"inode #%lu: comm %s: iget: illegal inode #",
5150	ino, current->comm);
5151	return ERR_PTR(error: -EFSCORRUPTED);
5152	}
5153
5154	inode = iget_locked(sb, ino);
5155	if (!inode)
5156	return ERR_PTR(error: -ENOMEM);
5157	if (!(inode->i_state & I_NEW)) {
5158	ret = check_igot_inode(inode, flags, function, line);
5159	if (ret) {
5160	iput(inode);
5161	return ERR_PTR(error: ret);
5162	}
5163	return inode;
5164	}
5165
5166	ei = EXT4_I(inode);
5167	iloc.bh = NULL;
5168
5169	ret = __ext4_get_inode_loc_noinmem(inode, iloc: &iloc);
5170	if (ret < 0)
5171	goto bad_inode;
5172	raw_inode = ext4_raw_inode(iloc: &iloc);
5173
5174	if ((flags & EXT4_IGET_HANDLE) &&
5175	(raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
5176	ret = -ESTALE;
5177	goto bad_inode;
5178	}
5179
5180	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5181	ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
5182	if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
5183	EXT4_INODE_SIZE(inode->i_sb) ||
5184	(ei->i_extra_isize & 3)) {
5185	ext4_error_inode(inode, function, line, 0,
5186	"iget: bad extra_isize %u "
5187	"(inode size %u)",
5188	ei->i_extra_isize,
5189	EXT4_INODE_SIZE(inode->i_sb));
5190	ret = -EFSCORRUPTED;
5191	goto bad_inode;
5192	}
5193	} else
5194	ei->i_extra_isize = 0;
5195
5196	/* Precompute checksum seed for inode metadata */
5197	if (ext4_has_feature_metadata_csum(sb)) {
5198	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
5199	__u32 csum;
5200	__le32 inum = cpu_to_le32(inode->i_ino);
5201	__le32 gen = raw_inode->i_generation;
5202	csum = ext4_chksum(crc: sbi->s_csum_seed, address: (__u8 *)&inum,
5203	length: sizeof(inum));
5204	ei->i_csum_seed = ext4_chksum(crc: csum, address: (__u8 *)&gen, length: sizeof(gen));
5205	}
5206
5207	if ((!ext4_inode_csum_verify(inode, raw: raw_inode, ei) ||
5208	ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
5209	(!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
5210	ext4_error_inode_err(inode, function, line, 0,
5211	EFSBADCRC, "iget: checksum invalid");
5212	ret = -EFSBADCRC;
5213	goto bad_inode;
5214	}
5215
5216	inode->i_mode = le16_to_cpu(raw_inode->i_mode);
5217	i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
5218	i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
5219	if (ext4_has_feature_project(sb) &&
5220	EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5221	EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5222	i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
5223	else
5224	i_projid = EXT4_DEF_PROJID;
5225
5226	if (!(test_opt(inode->i_sb, NO_UID32))) {
5227	i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
5228	i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
5229	}
5230	i_uid_write(inode, uid: i_uid);
5231	i_gid_write(inode, gid: i_gid);
5232	ei->i_projid = make_kprojid(from: &init_user_ns, projid: i_projid);
5233	set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
5234
5235	ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
5236	ei->i_inline_off = 0;
5237	ei->i_dir_start_lookup = 0;
5238	ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
5239	/* We now have enough fields to check if the inode was active or not.
5240	* This is needed because nfsd might try to access dead inodes
5241	* the test is that same one that e2fsck uses
5242	* NeilBrown 1999oct15
5243	*/
5244	if (inode->i_nlink == 0) {
5245	if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
5246	!(EXT4_SB(sb: inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
5247	ino != EXT4_BOOT_LOADER_INO) {
5248	/* this inode is deleted or unallocated */
5249	if (flags & EXT4_IGET_SPECIAL) {
5250	ext4_error_inode(inode, function, line, 0,
5251	"iget: special inode unallocated");
5252	ret = -EFSCORRUPTED;
5253	} else
5254	ret = -ESTALE;
5255	goto bad_inode;
5256	}
5257	/* The only unlinked inodes we let through here have
5258	* valid i_mode and are being read by the orphan
5259	* recovery code: that's fine, we're about to complete
5260	* the process of deleting those.
5261	* OR it is the EXT4_BOOT_LOADER_INO which is
5262	* not initialized on a new filesystem. */
5263	}
5264	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
5265	ext4_set_inode_flags(inode, init: true);
5266	inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
5267	ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
5268	if (ext4_has_feature_64bit(sb))
5269	ei->i_file_acl |=
5270	((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
5271	inode->i_size = ext4_isize(sb, raw_inode);
5272	size = i_size_read(inode);
5273	if (size < 0 || size > ext4_get_maxbytes(inode)) {
5274	ext4_error_inode(inode, function, line, 0,
5275	"iget: bad i_size value: %lld", size);
5276	ret = -EFSCORRUPTED;
5277	goto bad_inode;
5278	}
5279	/*
5280	* If dir_index is not enabled but there's dir with INDEX flag set,
5281	* we'd normally treat htree data as empty space. But with metadata
5282	* checksumming that corrupts checksums so forbid that.
5283	*/
5284	if (!ext4_has_feature_dir_index(sb) &&
5285	ext4_has_feature_metadata_csum(sb) &&
5286	ext4_test_inode_flag(inode, bit: EXT4_INODE_INDEX)) {
5287	ext4_error_inode(inode, function, line, 0,
5288	"iget: Dir with htree data on filesystem without dir_index feature.");
5289	ret = -EFSCORRUPTED;
5290	goto bad_inode;
5291	}
5292	ei->i_disksize = inode->i_size;
5293	#ifdef CONFIG_QUOTA
5294	ei->i_reserved_quota = 0;
5295	#endif
5296	inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5297	ei->i_block_group = iloc.block_group;
5298	ei->i_last_alloc_group = ~0;
5299	/*
5300	* NOTE! The in-memory inode i_data array is in little-endian order
5301	* even on big-endian machines: we do NOT byteswap the block numbers!
5302	*/
5303	for (block = 0; block < EXT4_N_BLOCKS; block++)
5304	ei->i_data[block] = raw_inode->i_block[block];
5305	INIT_LIST_HEAD(list: &ei->i_orphan);
5306	ext4_fc_init_inode(inode: &ei->vfs_inode);
5307
5308	/*
5309	* Set transaction id's of transactions that have to be committed
5310	* to finish f[data]sync. We set them to currently running transaction
5311	* as we cannot be sure that the inode or some of its metadata isn't
5312	* part of the transaction - the inode could have been reclaimed and
5313	* now it is reread from disk.
5314	*/
5315	if (journal) {
5316	transaction_t *transaction;
5317	tid_t tid;
5318
5319	read_lock(&journal->j_state_lock);
5320	if (journal->j_running_transaction)
5321	transaction = journal->j_running_transaction;
5322	else
5323	transaction = journal->j_committing_transaction;
5324	if (transaction)
5325	tid = transaction->t_tid;
5326	else
5327	tid = journal->j_commit_sequence;
5328	read_unlock(&journal->j_state_lock);
5329	ei->i_sync_tid = tid;
5330	ei->i_datasync_tid = tid;
5331	}
5332
5333	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5334	if (ei->i_extra_isize == 0) {
5335	/* The extra space is currently unused. Use it. */
5336	BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
5337	ei->i_extra_isize = sizeof(struct ext4_inode) -
5338	EXT4_GOOD_OLD_INODE_SIZE;
5339	} else {
5340	ret = ext4_iget_extra_inode(inode, raw_inode, ei);
5341	if (ret)
5342	goto bad_inode;
5343	}
5344	}
5345
5346	EXT4_INODE_GET_CTIME(inode, raw_inode);
5347	EXT4_INODE_GET_ATIME(inode, raw_inode);
5348	EXT4_INODE_GET_MTIME(inode, raw_inode);
5349	EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5350
5351	if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5352	u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5353
5354	if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5355	if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5356	ivers |=
5357	(__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5358	}
5359	ext4_inode_set_iversion_queried(inode, val: ivers);
5360	}
5361
5362	ret = 0;
5363	if (ei->i_file_acl &&
5364	!ext4_inode_block_valid(inode, start_blk: ei->i_file_acl, count: 1)) {
5365	ext4_error_inode(inode, function, line, 0,
5366	"iget: bad extended attribute block %llu",
5367	ei->i_file_acl);
5368	ret = -EFSCORRUPTED;
5369	goto bad_inode;
5370	} else if (!ext4_has_inline_data(inode)) {
5371	/* validate the block references in the inode */
5372	if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
5373	(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5374	(S_ISLNK(inode->i_mode) &&
5375	!ext4_inode_is_fast_symlink(inode)))) {
5376	if (ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))
5377	ret = ext4_ext_check_inode(inode);
5378	else
5379	ret = ext4_ind_check_inode(inode);
5380	}
5381	}
5382	if (ret)
5383	goto bad_inode;
5384
5385	if (S_ISREG(inode->i_mode)) {
5386	inode->i_op = &ext4_file_inode_operations;
5387	inode->i_fop = &ext4_file_operations;
5388	ext4_set_aops(inode);
5389	} else if (S_ISDIR(inode->i_mode)) {
5390	inode->i_op = &ext4_dir_inode_operations;
5391	inode->i_fop = &ext4_dir_operations;
5392	} else if (S_ISLNK(inode->i_mode)) {
5393	/* VFS does not allow setting these so must be corruption */
5394	if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5395	ext4_error_inode(inode, function, line, 0,
5396	"iget: immutable or append flags "
5397	"not allowed on symlinks");
5398	ret = -EFSCORRUPTED;
5399	goto bad_inode;
5400	}
5401	if (IS_ENCRYPTED(inode)) {
5402	inode->i_op = &ext4_encrypted_symlink_inode_operations;
5403	} else if (ext4_inode_is_fast_symlink(inode)) {
5404	inode->i_op = &ext4_fast_symlink_inode_operations;
5405	if (inode->i_size == 0 ||
5406	inode->i_size >= sizeof(ei->i_data) ||
5407	strnlen(p: (char *)ei->i_data, maxlen: inode->i_size + 1) !=
5408	inode->i_size) {
5409	ext4_error_inode(inode, function, line, 0,
5410	"invalid fast symlink length %llu",
5411	(unsigned long long)inode->i_size);
5412	ret = -EFSCORRUPTED;
5413	goto bad_inode;
5414	}
5415	inode_set_cached_link(inode, link: (char *)ei->i_data,
5416	linklen: inode->i_size);
5417	} else {
5418	inode->i_op = &ext4_symlink_inode_operations;
5419	}
5420	} else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5421	S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5422	inode->i_op = &ext4_special_inode_operations;
5423	if (raw_inode->i_block[0])
5424	init_special_inode(inode, inode->i_mode,
5425	old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5426	else
5427	init_special_inode(inode, inode->i_mode,
5428	new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5429	} else if (ino == EXT4_BOOT_LOADER_INO) {
5430	make_bad_inode(inode);
5431	} else {
5432	ret = -EFSCORRUPTED;
5433	ext4_error_inode(inode, function, line, 0,
5434	"iget: bogus i_mode (%o)", inode->i_mode);
5435	goto bad_inode;
5436	}
5437	if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(sb: inode->i_sb)) {
5438	ext4_error_inode(inode, function, line, 0,
5439	"casefold flag without casefold feature");
5440	ret = -EFSCORRUPTED;
5441	goto bad_inode;
5442	}
5443	if (ext4_should_enable_large_folio(inode))
5444	mapping_set_large_folios(mapping: inode->i_mapping);
5445
5446	ret = check_igot_inode(inode, flags, function, line);
5447	/*
5448	* -ESTALE here means there is nothing inherently wrong with the inode,
5449	* it's just not an inode we can return for an fhandle lookup.
5450	*/
5451	if (ret == -ESTALE) {
5452	brelse(bh: iloc.bh);
5453	unlock_new_inode(inode);
5454	iput(inode);
5455	return ERR_PTR(error: -ESTALE);
5456	}
5457	if (ret)
5458	goto bad_inode;
5459	brelse(bh: iloc.bh);
5460
5461	unlock_new_inode(inode);
5462	return inode;
5463
5464	bad_inode:
5465	brelse(bh: iloc.bh);
5466	iget_failed(inode);
5467	return ERR_PTR(error: ret);
5468	}
5469
5470	static void __ext4_update_other_inode_time(struct super_block *sb,
5471	unsigned long orig_ino,
5472	unsigned long ino,
5473	struct ext4_inode *raw_inode)
5474	{
5475	struct inode *inode;
5476
5477	inode = find_inode_by_ino_rcu(sb, ino);
5478	if (!inode)
5479	return;
5480
5481	if (!inode_is_dirtytime_only(inode))
5482	return;
5483
5484	spin_lock(lock: &inode->i_lock);
5485	if (inode_is_dirtytime_only(inode)) {
5486	struct ext4_inode_info *ei = EXT4_I(inode);
5487
5488	inode->i_state &= ~I_DIRTY_TIME;
5489	spin_unlock(lock: &inode->i_lock);
5490
5491	spin_lock(lock: &ei->i_raw_lock);
5492	EXT4_INODE_SET_CTIME(inode, raw_inode);
5493	EXT4_INODE_SET_MTIME(inode, raw_inode);
5494	EXT4_INODE_SET_ATIME(inode, raw_inode);
5495	ext4_inode_csum_set(inode, raw: raw_inode, ei);
5496	spin_unlock(lock: &ei->i_raw_lock);
5497	trace_ext4_other_inode_update_time(inode, orig_ino);
5498	return;
5499	}
5500	spin_unlock(lock: &inode->i_lock);
5501	}
5502
5503	/*
5504	* Opportunistically update the other time fields for other inodes in
5505	* the same inode table block.
5506	*/
5507	static void ext4_update_other_inodes_time(struct super_block *sb,
5508	unsigned long orig_ino, char *buf)
5509	{
5510	unsigned long ino;
5511	int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5512	int inode_size = EXT4_INODE_SIZE(sb);
5513
5514	/*
5515	* Calculate the first inode in the inode table block. Inode
5516	* numbers are one-based. That is, the first inode in a block
5517	* (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5518	*/
5519	ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5520	rcu_read_lock();
5521	for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5522	if (ino == orig_ino)
5523	continue;
5524	__ext4_update_other_inode_time(sb, orig_ino, ino,
5525	raw_inode: (struct ext4_inode *)buf);
5526	}
5527	rcu_read_unlock();
5528	}
5529
5530	/*
5531	* Post the struct inode info into an on-disk inode location in the
5532	* buffer-cache. This gobbles the caller's reference to the
5533	* buffer_head in the inode location struct.
5534	*
5535	* The caller must have write access to iloc->bh.
5536	*/
5537	static int ext4_do_update_inode(handle_t *handle,
5538	struct inode *inode,
5539	struct ext4_iloc *iloc)
5540	{
5541	struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5542	struct ext4_inode_info *ei = EXT4_I(inode);
5543	struct buffer_head *bh = iloc->bh;
5544	struct super_block *sb = inode->i_sb;
5545	int err;
5546	int need_datasync = 0, set_large_file = 0;
5547
5548	spin_lock(lock: &ei->i_raw_lock);
5549
5550	/*
5551	* For fields not tracked in the in-memory inode, initialise them
5552	* to zero for new inodes.
5553	*/
5554	if (ext4_test_inode_state(inode, bit: EXT4_STATE_NEW))
5555	memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5556
5557	if (READ_ONCE(ei->i_disksize) != ext4_isize(sb: inode->i_sb, raw_inode))
5558	need_datasync = 1;
5559	if (ei->i_disksize > 0x7fffffffULL) {
5560	if (!ext4_has_feature_large_file(sb) ||
5561	EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5562	set_large_file = 1;
5563	}
5564
5565	err = ext4_fill_raw_inode(inode, raw_inode);
5566	spin_unlock(lock: &ei->i_raw_lock);
5567	if (err) {
5568	EXT4_ERROR_INODE(inode, "corrupted inode contents");
5569	goto out_brelse;
5570	}
5571
5572	if (inode->i_sb->s_flags & SB_LAZYTIME)
5573	ext4_update_other_inodes_time(sb: inode->i_sb, orig_ino: inode->i_ino,
5574	buf: bh->b_data);
5575
5576	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5577	err = ext4_handle_dirty_metadata(handle, NULL, bh);
5578	if (err)
5579	goto out_error;
5580	ext4_clear_inode_state(inode, bit: EXT4_STATE_NEW);
5581	if (set_large_file) {
5582	BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5583	err = ext4_journal_get_write_access(handle, sb,
5584	EXT4_SB(sb)->s_sbh,
5585	EXT4_JTR_NONE);
5586	if (err)
5587	goto out_error;
5588	lock_buffer(bh: EXT4_SB(sb)->s_sbh);
5589	ext4_set_feature_large_file(sb);
5590	ext4_superblock_csum_set(sb);
5591	unlock_buffer(bh: EXT4_SB(sb)->s_sbh);
5592	ext4_handle_sync(handle);
5593	err = ext4_handle_dirty_metadata(handle, NULL,
5594	EXT4_SB(sb)->s_sbh);
5595	}
5596	ext4_update_inode_fsync_trans(handle, inode, datasync: need_datasync);
5597	out_error:
5598	ext4_std_error(inode->i_sb, err);
5599	out_brelse:
5600	brelse(bh);
5601	return err;
5602	}
5603
5604	/*
5605	* ext4_write_inode()
5606	*
5607	* We are called from a few places:
5608	*
5609	* - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5610	* Here, there will be no transaction running. We wait for any running
5611	* transaction to commit.
5612	*
5613	* - Within flush work (sys_sync(), kupdate and such).
5614	* We wait on commit, if told to.
5615	*
5616	* - Within iput_final() -> write_inode_now()
5617	* We wait on commit, if told to.
5618	*
5619	* In all cases it is actually safe for us to return without doing anything,
5620	* because the inode has been copied into a raw inode buffer in
5621	* ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5622	* writeback.
5623	*
5624	* Note that we are absolutely dependent upon all inode dirtiers doing the
5625	* right thing: they *must* call mark_inode_dirty() after dirtying info in
5626	* which we are interested.
5627	*
5628	* It would be a bug for them to not do this. The code:
5629	*
5630	* mark_inode_dirty(inode)
5631	* stuff();
5632	* inode->i_size = expr;
5633	*
5634	* is in error because write_inode() could occur while `stuff()' is running,
5635	* and the new i_size will be lost. Plus the inode will no longer be on the
5636	* superblock's dirty inode list.
5637	*/
5638	int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5639	{
5640	int err;
5641
5642	if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5643	return 0;
5644
5645	err = ext4_emergency_state(sb: inode->i_sb);
5646	if (unlikely(err))
5647	return err;
5648
5649	if (EXT4_SB(sb: inode->i_sb)->s_journal) {
5650	if (ext4_journal_current_handle()) {
5651	ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5652	dump_stack();
5653	return -EIO;
5654	}
5655
5656	/*
5657	* No need to force transaction in WB_SYNC_NONE mode. Also
5658	* ext4_sync_fs() will force the commit after everything is
5659	* written.
5660	*/
5661	if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5662	return 0;
5663
5664	err = ext4_fc_commit(journal: EXT4_SB(sb: inode->i_sb)->s_journal,
5665	EXT4_I(inode)->i_sync_tid);
5666	} else {
5667	struct ext4_iloc iloc;
5668
5669	err = __ext4_get_inode_loc_noinmem(inode, iloc: &iloc);
5670	if (err)
5671	return err;
5672	/*
5673	* sync(2) will flush the whole buffer cache. No need to do
5674	* it here separately for each inode.
5675	*/
5676	if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5677	sync_dirty_buffer(bh: iloc.bh);
5678	if (buffer_req(bh: iloc.bh) && !buffer_uptodate(bh: iloc.bh)) {
5679	ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5680	"IO error syncing inode");
5681	err = -EIO;
5682	}
5683	brelse(bh: iloc.bh);
5684	}
5685	return err;
5686	}
5687
5688	/*
5689	* In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5690	* buffers that are attached to a folio straddling i_size and are undergoing
5691	* commit. In that case we have to wait for commit to finish and try again.
5692	*/
5693	static void ext4_wait_for_tail_page_commit(struct inode *inode)
5694	{
5695	unsigned offset;
5696	journal_t *journal = EXT4_SB(sb: inode->i_sb)->s_journal;
5697	tid_t commit_tid;
5698	int ret;
5699	bool has_transaction;
5700
5701	offset = inode->i_size & (PAGE_SIZE - 1);
5702	/*
5703	* If the folio is fully truncated, we don't need to wait for any commit
5704	* (and we even should not as __ext4_journalled_invalidate_folio() may
5705	* strip all buffers from the folio but keep the folio dirty which can then
5706	* confuse e.g. concurrent ext4_writepages() seeing dirty folio without
5707	* buffers). Also we don't need to wait for any commit if all buffers in
5708	* the folio remain valid. This is most beneficial for the common case of
5709	* blocksize == PAGESIZE.
5710	*/
5711	if (!offset || offset > (PAGE_SIZE - i_blocksize(node: inode)))
5712	return;
5713	while (1) {
5714	struct folio *folio = filemap_lock_folio(mapping: inode->i_mapping,
5715	index: inode->i_size >> PAGE_SHIFT);
5716	if (IS_ERR(ptr: folio))
5717	return;
5718	ret = __ext4_journalled_invalidate_folio(folio, offset,
5719	length: folio_size(folio) - offset);
5720	folio_unlock(folio);
5721	folio_put(folio);
5722	if (ret != -EBUSY)
5723	return;
5724	has_transaction = false;
5725	read_lock(&journal->j_state_lock);
5726	if (journal->j_committing_transaction) {
5727	commit_tid = journal->j_committing_transaction->t_tid;
5728	has_transaction = true;
5729	}
5730	read_unlock(&journal->j_state_lock);
5731	if (has_transaction)
5732	jbd2_log_wait_commit(journal, tid: commit_tid);
5733	}
5734	}
5735
5736	/*
5737	* ext4_setattr()
5738	*
5739	* Called from notify_change.
5740	*
5741	* We want to trap VFS attempts to truncate the file as soon as
5742	* possible. In particular, we want to make sure that when the VFS
5743	* shrinks i_size, we put the inode on the orphan list and modify
5744	* i_disksize immediately, so that during the subsequent flushing of
5745	* dirty pages and freeing of disk blocks, we can guarantee that any
5746	* commit will leave the blocks being flushed in an unused state on
5747	* disk. (On recovery, the inode will get truncated and the blocks will
5748	* be freed, so we have a strong guarantee that no future commit will
5749	* leave these blocks visible to the user.)
5750	*
5751	* Another thing we have to assure is that if we are in ordered mode
5752	* and inode is still attached to the committing transaction, we must
5753	* we start writeout of all the dirty pages which are being truncated.
5754	* This way we are sure that all the data written in the previous
5755	* transaction are already on disk (truncate waits for pages under
5756	* writeback).
5757	*
5758	* Called with inode->i_rwsem down.
5759	*/
5760	int ext4_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5761	struct iattr *attr)
5762	{
5763	struct inode *inode = d_inode(dentry);
5764	int error, rc = 0;
5765	int orphan = 0;
5766	const unsigned int ia_valid = attr->ia_valid;
5767	bool inc_ivers = true;
5768
5769	error = ext4_emergency_state(sb: inode->i_sb);
5770	if (unlikely(error))
5771	return error;
5772
5773	if (unlikely(IS_IMMUTABLE(inode)))
5774	return -EPERM;
5775
5776	if (unlikely(IS_APPEND(inode) &&
5777	(ia_valid & (ATTR_MODE | ATTR_UID |
5778	ATTR_GID | ATTR_TIMES_SET))))
5779	return -EPERM;
5780
5781	error = setattr_prepare(idmap, dentry, attr);
5782	if (error)
5783	return error;
5784
5785	error = fscrypt_prepare_setattr(dentry, attr);
5786	if (error)
5787	return error;
5788
5789	error = fsverity_prepare_setattr(dentry, attr);
5790	if (error)
5791	return error;
5792
5793	if (is_quota_modification(idmap, inode, ia: attr)) {
5794	error = dquot_initialize(inode);
5795	if (error)
5796	return error;
5797	}
5798
5799	if (i_uid_needs_update(idmap, attr, inode) ||
5800	i_gid_needs_update(idmap, attr, inode)) {
5801	handle_t *handle;
5802
5803	/* (user+group)*(old+new) structure, inode write (sb,
5804	* inode block, ? - but truncate inode update has it) */
5805	handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5806	(EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5807	EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5808	if (IS_ERR(ptr: handle)) {
5809	error = PTR_ERR(ptr: handle);
5810	goto err_out;
5811	}
5812
5813	/* dquot_transfer() calls back ext4_get_inode_usage() which
5814	* counts xattr inode references.
5815	*/
5816	down_read(sem: &EXT4_I(inode)->xattr_sem);
5817	error = dquot_transfer(idmap, inode, iattr: attr);
5818	up_read(sem: &EXT4_I(inode)->xattr_sem);
5819
5820	if (error) {
5821	ext4_journal_stop(handle);
5822	return error;
5823	}
5824	/* Update corresponding info in inode so that everything is in
5825	* one transaction */
5826	i_uid_update(idmap, attr, inode);
5827	i_gid_update(idmap, attr, inode);
5828	error = ext4_mark_inode_dirty(handle, inode);
5829	ext4_journal_stop(handle);
5830	if (unlikely(error)) {
5831	return error;
5832	}
5833	}
5834
5835	if (attr->ia_valid & ATTR_SIZE) {
5836	handle_t *handle;
5837	loff_t oldsize = inode->i_size;
5838	loff_t old_disksize;
5839	int shrink = (attr->ia_size < inode->i_size);
5840
5841	if (!(ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS))) {
5842	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
5843
5844	if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5845	return -EFBIG;
5846	}
5847	}
5848	if (!S_ISREG(inode->i_mode)) {
5849	return -EINVAL;
5850	}
5851
5852	if (attr->ia_size == inode->i_size)
5853	inc_ivers = false;
5854
5855	if (shrink) {
5856	if (ext4_should_order_data(inode)) {
5857	error = ext4_begin_ordered_truncate(inode,
5858	new_size: attr->ia_size);
5859	if (error)
5860	goto err_out;
5861	}
5862	/*
5863	* Blocks are going to be removed from the inode. Wait
5864	* for dio in flight.
5865	*/
5866	inode_dio_wait(inode);
5867	}
5868
5869	filemap_invalidate_lock(mapping: inode->i_mapping);
5870
5871	rc = ext4_break_layouts(inode);
5872	if (rc) {
5873	filemap_invalidate_unlock(mapping: inode->i_mapping);
5874	goto err_out;
5875	}
5876
5877	if (attr->ia_size != inode->i_size) {
5878	/* attach jbd2 jinode for EOF folio tail zeroing */
5879	if (attr->ia_size & (inode->i_sb->s_blocksize - 1) ||
5880	oldsize & (inode->i_sb->s_blocksize - 1)) {
5881	error = ext4_inode_attach_jinode(inode);
5882	if (error)
5883	goto out_mmap_sem;
5884	}
5885
5886	handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5887	if (IS_ERR(ptr: handle)) {
5888	error = PTR_ERR(ptr: handle);
5889	goto out_mmap_sem;
5890	}
5891	if (ext4_handle_valid(handle) && shrink) {
5892	error = ext4_orphan_add(handle, inode);
5893	orphan = 1;
5894	}
5895	/*
5896	* Update c/mtime and tail zero the EOF folio on
5897	* truncate up. ext4_truncate() handles the shrink case
5898	* below.
5899	*/
5900	if (!shrink) {
5901	inode_set_mtime_to_ts(inode,
5902	ts: inode_set_ctime_current(inode));
5903	if (oldsize & (inode->i_sb->s_blocksize - 1))
5904	ext4_block_truncate_page(handle,
5905	mapping: inode->i_mapping, from: oldsize);
5906	}
5907
5908	if (shrink)
5909	ext4_fc_track_range(handle, inode,
5910	start: (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5911	inode->i_sb->s_blocksize_bits,
5912	EXT_MAX_BLOCKS - 1);
5913	else
5914	ext4_fc_track_range(
5915	handle, inode,
5916	start: (oldsize > 0 ? oldsize - 1 : oldsize) >>
5917	inode->i_sb->s_blocksize_bits,
5918	end: (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5919	inode->i_sb->s_blocksize_bits);
5920
5921	down_write(sem: &EXT4_I(inode)->i_data_sem);
5922	old_disksize = EXT4_I(inode)->i_disksize;
5923	EXT4_I(inode)->i_disksize = attr->ia_size;
5924
5925	/*
5926	* We have to update i_size under i_data_sem together
5927	* with i_disksize to avoid races with writeback code
5928	* running ext4_wb_update_i_disksize().
5929	*/
5930	if (!error)
5931	i_size_write(inode, i_size: attr->ia_size);
5932	else
5933	EXT4_I(inode)->i_disksize = old_disksize;
5934	up_write(sem: &EXT4_I(inode)->i_data_sem);
5935	rc = ext4_mark_inode_dirty(handle, inode);
5936	if (!error)
5937	error = rc;
5938	ext4_journal_stop(handle);
5939	if (error)
5940	goto out_mmap_sem;
5941	if (!shrink) {
5942	pagecache_isize_extended(inode, from: oldsize,
5943	to: inode->i_size);
5944	} else if (ext4_should_journal_data(inode)) {
5945	ext4_wait_for_tail_page_commit(inode);
5946	}
5947	}
5948
5949	/*
5950	* Truncate pagecache after we've waited for commit
5951	* in data=journal mode to make pages freeable.
5952	*/
5953	truncate_pagecache(inode, new: inode->i_size);
5954	/*
5955	* Call ext4_truncate() even if i_size didn't change to
5956	* truncate possible preallocated blocks.
5957	*/
5958	if (attr->ia_size <= oldsize) {
5959	rc = ext4_truncate(inode);
5960	if (rc)
5961	error = rc;
5962	}
5963	out_mmap_sem:
5964	filemap_invalidate_unlock(mapping: inode->i_mapping);
5965	}
5966
5967	if (!error) {
5968	if (inc_ivers)
5969	inode_inc_iversion(inode);
5970	setattr_copy(idmap, inode, attr);
5971	mark_inode_dirty(inode);
5972	}
5973
5974	/*
5975	* If the call to ext4_truncate failed to get a transaction handle at
5976	* all, we need to clean up the in-core orphan list manually.
5977	*/
5978	if (orphan && inode->i_nlink)
5979	ext4_orphan_del(NULL, inode);
5980
5981	if (!error && (ia_valid & ATTR_MODE))
5982	rc = posix_acl_chmod(idmap, dentry, inode->i_mode);
5983
5984	err_out:
5985	if (error)
5986	ext4_std_error(inode->i_sb, error);
5987	if (!error)
5988	error = rc;
5989	return error;
5990	}
5991
5992	u32 ext4_dio_alignment(struct inode *inode)
5993	{
5994	if (fsverity_active(inode))
5995	return 0;
5996	if (ext4_should_journal_data(inode))
5997	return 0;
5998	if (ext4_has_inline_data(inode))
5999	return 0;
6000	if (IS_ENCRYPTED(inode)) {
6001	if (!fscrypt_dio_supported(inode))
6002	return 0;
6003	return i_blocksize(node: inode);
6004	}
6005	return 1; /* use the iomap defaults */
6006	}
6007
6008	int ext4_getattr(struct mnt_idmap *idmap, const struct path *path,
6009	struct kstat *stat, u32 request_mask, unsigned int query_flags)
6010	{
6011	struct inode *inode = d_inode(dentry: path->dentry);
6012	struct ext4_inode *raw_inode;
6013	struct ext4_inode_info *ei = EXT4_I(inode);
6014	unsigned int flags;
6015
6016	if ((request_mask & STATX_BTIME) &&
6017	EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
6018	stat->result_mask |= STATX_BTIME;
6019	stat->btime.tv_sec = ei->i_crtime.tv_sec;
6020	stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
6021	}
6022
6023	/*
6024	* Return the DIO alignment restrictions if requested. We only return
6025	* this information when requested, since on encrypted files it might
6026	* take a fair bit of work to get if the file wasn't opened recently.
6027	*/
6028	if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
6029	u32 dio_align = ext4_dio_alignment(inode);
6030
6031	stat->result_mask |= STATX_DIOALIGN;
6032	if (dio_align == 1) {
6033	struct block_device *bdev = inode->i_sb->s_bdev;
6034
6035	/* iomap defaults */
6036	stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
6037	stat->dio_offset_align = bdev_logical_block_size(bdev);
6038	} else {
6039	stat->dio_mem_align = dio_align;
6040	stat->dio_offset_align = dio_align;
6041	}
6042	}
6043
6044	if ((request_mask & STATX_WRITE_ATOMIC) && S_ISREG(inode->i_mode)) {
6045	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
6046	unsigned int awu_min = 0, awu_max = 0;
6047
6048	if (ext4_inode_can_atomic_write(inode)) {
6049	awu_min = sbi->s_awu_min;
6050	awu_max = sbi->s_awu_max;
6051	}
6052
6053	generic_fill_statx_atomic_writes(stat, unit_min: awu_min, unit_max: awu_max, unit_max_opt: 0);
6054	}
6055
6056	flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
6057	if (flags & EXT4_APPEND_FL)
6058	stat->attributes |= STATX_ATTR_APPEND;
6059	if (flags & EXT4_COMPR_FL)
6060	stat->attributes |= STATX_ATTR_COMPRESSED;
6061	if (flags & EXT4_ENCRYPT_FL)
6062	stat->attributes |= STATX_ATTR_ENCRYPTED;
6063	if (flags & EXT4_IMMUTABLE_FL)
6064	stat->attributes |= STATX_ATTR_IMMUTABLE;
6065	if (flags & EXT4_NODUMP_FL)
6066	stat->attributes |= STATX_ATTR_NODUMP;
6067	if (flags & EXT4_VERITY_FL)
6068	stat->attributes |= STATX_ATTR_VERITY;
6069
6070	stat->attributes_mask |= (STATX_ATTR_APPEND |
6071	STATX_ATTR_COMPRESSED |
6072	STATX_ATTR_ENCRYPTED |
6073	STATX_ATTR_IMMUTABLE |
6074	STATX_ATTR_NODUMP |
6075	STATX_ATTR_VERITY);
6076
6077	generic_fillattr(idmap, request_mask, inode, stat);
6078	return 0;
6079	}
6080
6081	int ext4_file_getattr(struct mnt_idmap *idmap,
6082	const struct path *path, struct kstat *stat,
6083	u32 request_mask, unsigned int query_flags)
6084	{
6085	struct inode *inode = d_inode(dentry: path->dentry);
6086	u64 delalloc_blocks;
6087
6088	ext4_getattr(idmap, path, stat, request_mask, query_flags);
6089
6090	/*
6091	* If there is inline data in the inode, the inode will normally not
6092	* have data blocks allocated (it may have an external xattr block).
6093	* Report at least one sector for such files, so tools like tar, rsync,
6094	* others don't incorrectly think the file is completely sparse.
6095	*/
6096	if (unlikely(ext4_has_inline_data(inode)))
6097	stat->blocks += (stat->size + 511) >> 9;
6098
6099	/*
6100	* We can't update i_blocks if the block allocation is delayed
6101	* otherwise in the case of system crash before the real block
6102	* allocation is done, we will have i_blocks inconsistent with
6103	* on-disk file blocks.
6104	* We always keep i_blocks updated together with real
6105	* allocation. But to not confuse with user, stat
6106	* will return the blocks that include the delayed allocation
6107	* blocks for this file.
6108	*/
6109	delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
6110	EXT4_I(inode)->i_reserved_data_blocks);
6111	stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
6112	return 0;
6113	}
6114
6115	static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
6116	int pextents)
6117	{
6118	if (!(ext4_test_inode_flag(inode, bit: EXT4_INODE_EXTENTS)))
6119	return ext4_ind_trans_blocks(inode, nrblocks: lblocks);
6120	return ext4_ext_index_trans_blocks(inode, extents: pextents);
6121	}
6122
6123	/*
6124	* Account for index blocks, block groups bitmaps and block group
6125	* descriptor blocks if modify datablocks and index blocks
6126	* worse case, the indexs blocks spread over different block groups
6127	*
6128	* If datablocks are discontiguous, they are possible to spread over
6129	* different block groups too. If they are contiguous, with flexbg,
6130	* they could still across block group boundary.
6131	*
6132	* Also account for superblock, inode, quota and xattr blocks
6133	*/
6134	int ext4_meta_trans_blocks(struct inode *inode, int lblocks, int pextents)
6135	{
6136	ext4_group_t groups, ngroups = ext4_get_groups_count(sb: inode->i_sb);
6137	int gdpblocks;
6138	int idxblocks;
6139	int ret;
6140
6141	/*
6142	* How many index and lead blocks need to touch to map @lblocks
6143	* logical blocks to @pextents physical extents?
6144	*/
6145	idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
6146
6147	/*
6148	* Now let's see how many group bitmaps and group descriptors need
6149	* to account
6150	*/
6151	groups = idxblocks;
6152	gdpblocks = groups;
6153	if (groups > ngroups)
6154	groups = ngroups;
6155	if (groups > EXT4_SB(sb: inode->i_sb)->s_gdb_count)
6156	gdpblocks = EXT4_SB(sb: inode->i_sb)->s_gdb_count;
6157
6158	/* bitmaps and block group descriptor blocks */
6159	ret = idxblocks + groups + gdpblocks;
6160
6161	/* Blocks for super block, inode, quota and xattr blocks */
6162	ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
6163
6164	return ret;
6165	}
6166
6167	/*
6168	* Calculate the total number of credits to reserve to fit
6169	* the modification of a single pages into a single transaction,
6170	* which may include multiple chunks of block allocations.
6171	*
6172	* This could be called via ext4_write_begin()
6173	*
6174	* We need to consider the worse case, when
6175	* one new block per extent.
6176	*/
6177	int ext4_writepage_trans_blocks(struct inode *inode)
6178	{
6179	int bpp = ext4_journal_blocks_per_folio(inode);
6180	int ret;
6181
6182	ret = ext4_meta_trans_blocks(inode, lblocks: bpp, pextents: bpp);
6183
6184	/* Account for data blocks for journalled mode */
6185	if (ext4_should_journal_data(inode))
6186	ret += bpp;
6187	return ret;
6188	}
6189
6190	/*
6191	* Calculate the journal credits for a chunk of data modification.
6192	*
6193	* This is called from DIO, fallocate or whoever calling
6194	* ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
6195	*
6196	* journal buffers for data blocks are not included here, as DIO
6197	* and fallocate do no need to journal data buffers.
6198	*/
6199	int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
6200	{
6201	return ext4_meta_trans_blocks(inode, lblocks: nrblocks, pextents: 1);
6202	}
6203
6204	/*
6205	* The caller must have previously called ext4_reserve_inode_write().
6206	* Give this, we know that the caller already has write access to iloc->bh.
6207	*/
6208	int ext4_mark_iloc_dirty(handle_t *handle,
6209	struct inode *inode, struct ext4_iloc *iloc)
6210	{
6211	int err = 0;
6212
6213	err = ext4_emergency_state(sb: inode->i_sb);
6214	if (unlikely(err)) {
6215	put_bh(bh: iloc->bh);
6216	return err;
6217	}
6218	ext4_fc_track_inode(handle, inode);
6219
6220	/* the do_update_inode consumes one bh->b_count */
6221	get_bh(bh: iloc->bh);
6222
6223	/* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
6224	err = ext4_do_update_inode(handle, inode, iloc);
6225	put_bh(bh: iloc->bh);
6226	return err;
6227	}
6228
6229	/*
6230	* On success, We end up with an outstanding reference count against
6231	* iloc->bh. This _must_ be cleaned up later.
6232	*/
6233
6234	int
6235	ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
6236	struct ext4_iloc *iloc)
6237	{
6238	int err;
6239
6240	err = ext4_emergency_state(sb: inode->i_sb);
6241	if (unlikely(err))
6242	return err;
6243
6244	err = ext4_get_inode_loc(inode, iloc);
6245	if (!err) {
6246	BUFFER_TRACE(iloc->bh, "get_write_access");
6247	err = ext4_journal_get_write_access(handle, inode->i_sb,
6248	iloc->bh, EXT4_JTR_NONE);
6249	if (err) {
6250	brelse(bh: iloc->bh);
6251	iloc->bh = NULL;
6252	}
6253	ext4_fc_track_inode(handle, inode);
6254	}
6255	ext4_std_error(inode->i_sb, err);
6256	return err;
6257	}
6258
6259	static int __ext4_expand_extra_isize(struct inode *inode,
6260	unsigned int new_extra_isize,
6261	struct ext4_iloc *iloc,
6262	handle_t *handle, int *no_expand)
6263	{
6264	struct ext4_inode *raw_inode;
6265	struct ext4_xattr_ibody_header *header;
6266	unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
6267	struct ext4_inode_info *ei = EXT4_I(inode);
6268	int error;
6269
6270	/* this was checked at iget time, but double check for good measure */
6271	if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
6272	(ei->i_extra_isize & 3)) {
6273	EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
6274	ei->i_extra_isize,
6275	EXT4_INODE_SIZE(inode->i_sb));
6276	return -EFSCORRUPTED;
6277	}
6278	if ((new_extra_isize < ei->i_extra_isize) ||
6279	(new_extra_isize < 4) ||
6280	(new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
6281	return -EINVAL; /* Should never happen */
6282
6283	raw_inode = ext4_raw_inode(iloc);
6284
6285	header = IHDR(inode, raw_inode);
6286
6287	/* No extended attributes present */
6288	if (!ext4_test_inode_state(inode, bit: EXT4_STATE_XATTR) ||
6289	header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
6290	memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
6291	EXT4_I(inode)->i_extra_isize, 0,
6292	new_extra_isize - EXT4_I(inode)->i_extra_isize);
6293	EXT4_I(inode)->i_extra_isize = new_extra_isize;
6294	return 0;
6295	}
6296
6297	/*
6298	* We may need to allocate external xattr block so we need quotas
6299	* initialized. Here we can be called with various locks held so we
6300	* cannot affort to initialize quotas ourselves. So just bail.
6301	*/
6302	if (dquot_initialize_needed(inode))
6303	return -EAGAIN;
6304
6305	/* try to expand with EAs present */
6306	error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
6307	raw_inode, handle);
6308	if (error) {
6309	/*
6310	* Inode size expansion failed; don't try again
6311	*/
6312	*no_expand = 1;
6313	}
6314
6315	return error;
6316	}
6317
6318	/*
6319	* Expand an inode by new_extra_isize bytes.
6320	* Returns 0 on success or negative error number on failure.
6321	*/
6322	static int ext4_try_to_expand_extra_isize(struct inode *inode,
6323	unsigned int new_extra_isize,
6324	struct ext4_iloc iloc,
6325	handle_t *handle)
6326	{
6327	int no_expand;
6328	int error;
6329
6330	if (ext4_test_inode_state(inode, bit: EXT4_STATE_NO_EXPAND))
6331	return -EOVERFLOW;
6332
6333	/*
6334	* In nojournal mode, we can immediately attempt to expand
6335	* the inode. When journaled, we first need to obtain extra
6336	* buffer credits since we may write into the EA block
6337	* with this same handle. If journal_extend fails, then it will
6338	* only result in a minor loss of functionality for that inode.
6339	* If this is felt to be critical, then e2fsck should be run to
6340	* force a large enough s_min_extra_isize.
6341	*/
6342	if (ext4_journal_extend(handle,
6343	EXT4_DATA_TRANS_BLOCKS(inode->i_sb), revoke: 0) != 0)
6344	return -ENOSPC;
6345
6346	if (ext4_write_trylock_xattr(inode, save: &no_expand) == 0)
6347	return -EBUSY;
6348
6349	error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc: &iloc,
6350	handle, no_expand: &no_expand);
6351	ext4_write_unlock_xattr(inode, save: &no_expand);
6352
6353	return error;
6354	}
6355
6356	int ext4_expand_extra_isize(struct inode *inode,
6357	unsigned int new_extra_isize,
6358	struct ext4_iloc *iloc)
6359	{
6360	handle_t *handle;
6361	int no_expand;
6362	int error, rc;
6363
6364	if (ext4_test_inode_state(inode, bit: EXT4_STATE_NO_EXPAND)) {
6365	brelse(bh: iloc->bh);
6366	return -EOVERFLOW;
6367	}
6368
6369	handle = ext4_journal_start(inode, EXT4_HT_INODE,
6370	EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
6371	if (IS_ERR(ptr: handle)) {
6372	error = PTR_ERR(ptr: handle);
6373	brelse(bh: iloc->bh);
6374	return error;
6375	}
6376
6377	ext4_write_lock_xattr(inode, save: &no_expand);
6378
6379	BUFFER_TRACE(iloc->bh, "get_write_access");
6380	error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
6381	EXT4_JTR_NONE);
6382	if (error) {
6383	brelse(bh: iloc->bh);
6384	goto out_unlock;
6385	}
6386
6387	error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
6388	handle, no_expand: &no_expand);
6389
6390	rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6391	if (!error)
6392	error = rc;
6393
6394	out_unlock:
6395	ext4_write_unlock_xattr(inode, save: &no_expand);
6396	ext4_journal_stop(handle);
6397	return error;
6398	}
6399
6400	/*
6401	* What we do here is to mark the in-core inode as clean with respect to inode
6402	* dirtiness (it may still be data-dirty).
6403	* This means that the in-core inode may be reaped by prune_icache
6404	* without having to perform any I/O. This is a very good thing,
6405	* because *any* task may call prune_icache - even ones which
6406	* have a transaction open against a different journal.
6407	*
6408	* Is this cheating? Not really. Sure, we haven't written the
6409	* inode out, but prune_icache isn't a user-visible syncing function.
6410	* Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6411	* we start and wait on commits.
6412	*/
6413	int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
6414	const char *func, unsigned int line)
6415	{
6416	struct ext4_iloc iloc;
6417	struct ext4_sb_info *sbi = EXT4_SB(sb: inode->i_sb);
6418	int err;
6419
6420	might_sleep();
6421	trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6422	err = ext4_reserve_inode_write(handle, inode, iloc: &iloc);
6423	if (err)
6424	goto out;
6425
6426	if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6427	ext4_try_to_expand_extra_isize(inode, new_extra_isize: sbi->s_want_extra_isize,
6428	iloc, handle);
6429
6430	err = ext4_mark_iloc_dirty(handle, inode, iloc: &iloc);
6431	out:
6432	if (unlikely(err))
6433	ext4_error_inode_err(inode, func, line, 0, err,
6434	"mark_inode_dirty error");
6435	return err;
6436	}
6437
6438	/*
6439	* ext4_dirty_inode() is called from __mark_inode_dirty()
6440	*
6441	* We're really interested in the case where a file is being extended.
6442	* i_size has been changed by generic_commit_write() and we thus need
6443	* to include the updated inode in the current transaction.
6444	*
6445	* Also, dquot_alloc_block() will always dirty the inode when blocks
6446	* are allocated to the file.
6447	*
6448	* If the inode is marked synchronous, we don't honour that here - doing
6449	* so would cause a commit on atime updates, which we don't bother doing.
6450	* We handle synchronous inodes at the highest possible level.
6451	*/
6452	void ext4_dirty_inode(struct inode *inode, int flags)
6453	{
6454	handle_t *handle;
6455
6456	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6457	if (IS_ERR(ptr: handle))
6458	return;
6459	ext4_mark_inode_dirty(handle, inode);
6460	ext4_journal_stop(handle);
6461	}
6462
6463	int ext4_change_inode_journal_flag(struct inode *inode, int val)
6464	{
6465	journal_t *journal;
6466	handle_t *handle;
6467	int err;
6468	int alloc_ctx;
6469
6470	/*
6471	* We have to be very careful here: changing a data block's
6472	* journaling status dynamically is dangerous. If we write a
6473	* data block to the journal, change the status and then delete
6474	* that block, we risk forgetting to revoke the old log record
6475	* from the journal and so a subsequent replay can corrupt data.
6476	* So, first we make sure that the journal is empty and that
6477	* nobody is changing anything.
6478	*/
6479
6480	journal = EXT4_JOURNAL(inode);
6481	if (!journal)
6482	return 0;
6483	if (is_journal_aborted(journal))
6484	return -EROFS;
6485
6486	/* Wait for all existing dio workers */
6487	inode_dio_wait(inode);
6488
6489	/*
6490	* Before flushing the journal and switching inode's aops, we have
6491	* to flush all dirty data the inode has. There can be outstanding
6492	* delayed allocations, there can be unwritten extents created by
6493	* fallocate or buffered writes in dioread_nolock mode covered by
6494	* dirty data which can be converted only after flushing the dirty
6495	* data (and journalled aops don't know how to handle these cases).
6496	*/
6497	if (val) {
6498	filemap_invalidate_lock(mapping: inode->i_mapping);
6499	err = filemap_write_and_wait(mapping: inode->i_mapping);
6500	if (err < 0) {
6501	filemap_invalidate_unlock(mapping: inode->i_mapping);
6502	return err;
6503	}
6504	}
6505
6506	alloc_ctx = ext4_writepages_down_write(sb: inode->i_sb);
6507	jbd2_journal_lock_updates(journal);
6508
6509	/*
6510	* OK, there are no updates running now, and all cached data is
6511	* synced to disk. We are now in a completely consistent state
6512	* which doesn't have anything in the journal, and we know that
6513	* no filesystem updates are running, so it is safe to modify
6514	* the inode's in-core data-journaling state flag now.
6515	*/
6516
6517	if (val)
6518	ext4_set_inode_flag(inode, bit: EXT4_INODE_JOURNAL_DATA);
6519	else {
6520	err = jbd2_journal_flush(journal, flags: 0);
6521	if (err < 0) {
6522	jbd2_journal_unlock_updates(journal);
6523	ext4_writepages_up_write(sb: inode->i_sb, ctx: alloc_ctx);
6524	return err;
6525	}
6526	ext4_clear_inode_flag(inode, bit: EXT4_INODE_JOURNAL_DATA);
6527	}
6528	ext4_set_aops(inode);
6529
6530	jbd2_journal_unlock_updates(journal);
6531	ext4_writepages_up_write(sb: inode->i_sb, ctx: alloc_ctx);
6532
6533	if (val)
6534	filemap_invalidate_unlock(mapping: inode->i_mapping);
6535
6536	/* Finally we can mark the inode as dirty. */
6537
6538	handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6539	if (IS_ERR(ptr: handle))
6540	return PTR_ERR(ptr: handle);
6541
6542	ext4_fc_mark_ineligible(sb: inode->i_sb,
6543	reason: EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6544	err = ext4_mark_inode_dirty(handle, inode);
6545	ext4_handle_sync(handle);
6546	ext4_journal_stop(handle);
6547	ext4_std_error(inode->i_sb, err);
6548
6549	return err;
6550	}
6551
6552	static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6553	struct buffer_head *bh)
6554	{
6555	return !buffer_mapped(bh);
6556	}
6557
6558	vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6559	{
6560	struct vm_area_struct *vma = vmf->vma;
6561	struct folio *folio = page_folio(vmf->page);
6562	loff_t size;
6563	unsigned long len;
6564	int err;
6565	vm_fault_t ret;
6566	struct file *file = vma->vm_file;
6567	struct inode *inode = file_inode(f: file);
6568	struct address_space *mapping = inode->i_mapping;
6569	handle_t *handle;
6570	get_block_t *get_block;
6571	int retries = 0;
6572
6573	if (unlikely(IS_IMMUTABLE(inode)))
6574	return VM_FAULT_SIGBUS;
6575
6576	sb_start_pagefault(sb: inode->i_sb);
6577	file_update_time(file: vma->vm_file);
6578
6579	filemap_invalidate_lock_shared(mapping);
6580
6581	err = ext4_convert_inline_data(inode);
6582	if (err)
6583	goto out_ret;
6584
6585	/*
6586	* On data journalling we skip straight to the transaction handle:
6587	* there's no delalloc; page truncated will be checked later; the
6588	* early return w/ all buffers mapped (calculates size/len) can't
6589	* be used; and there's no dioread_nolock, so only ext4_get_block.
6590	*/
6591	if (ext4_should_journal_data(inode))
6592	goto retry_alloc;
6593
6594	/* Delalloc case is easy... */
6595	if (test_opt(inode->i_sb, DELALLOC) &&
6596	!ext4_nonda_switch(sb: inode->i_sb)) {
6597	do {
6598	err = block_page_mkwrite(vma, vmf,
6599	get_block: ext4_da_get_block_prep);
6600	} while (err == -ENOSPC &&
6601	ext4_should_retry_alloc(sb: inode->i_sb, retries: &retries));
6602	goto out_ret;
6603	}
6604
6605	folio_lock(folio);
6606	size = i_size_read(inode);
6607	/* Page got truncated from under us? */
6608	if (folio->mapping != mapping || folio_pos(folio) > size) {
6609	folio_unlock(folio);
6610	ret = VM_FAULT_NOPAGE;
6611	goto out;
6612	}
6613
6614	len = folio_size(folio);
6615	if (folio_pos(folio) + len > size)
6616	len = size - folio_pos(folio);
6617	/*
6618	* Return if we have all the buffers mapped. This avoids the need to do
6619	* journal_start/journal_stop which can block and take a long time
6620	*
6621	* This cannot be done for data journalling, as we have to add the
6622	* inode to the transaction's list to writeprotect pages on commit.
6623	*/
6624	if (folio_buffers(folio)) {
6625	if (!ext4_walk_page_buffers(NULL, inode, folio_buffers(folio),
6626	from: 0, to: len, NULL,
6627	fn: ext4_bh_unmapped)) {
6628	/* Wait so that we don't change page under IO */
6629	folio_wait_stable(folio);
6630	ret = VM_FAULT_LOCKED;
6631	goto out;
6632	}
6633	}
6634	folio_unlock(folio);
6635	/* OK, we need to fill the hole... */
6636	if (ext4_should_dioread_nolock(inode))
6637	get_block = ext4_get_block_unwritten;
6638	else
6639	get_block = ext4_get_block;
6640	retry_alloc:
6641	handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6642	ext4_writepage_trans_blocks(inode));
6643	if (IS_ERR(ptr: handle)) {
6644	ret = VM_FAULT_SIGBUS;
6645	goto out;
6646	}
6647	/*
6648	* Data journalling can't use block_page_mkwrite() because it
6649	* will set_buffer_dirty() before do_journal_get_write_access()
6650	* thus might hit warning messages for dirty metadata buffers.
6651	*/
6652	if (!ext4_should_journal_data(inode)) {
6653	err = block_page_mkwrite(vma, vmf, get_block);
6654	} else {
6655	folio_lock(folio);
6656	size = i_size_read(inode);
6657	/* Page got truncated from under us? */
6658	if (folio->mapping != mapping || folio_pos(folio) > size) {
6659	ret = VM_FAULT_NOPAGE;
6660	goto out_error;
6661	}
6662
6663	len = folio_size(folio);
6664	if (folio_pos(folio) + len > size)
6665	len = size - folio_pos(folio);
6666
6667	err = ext4_block_write_begin(handle, folio, pos: 0, len,
6668	get_block: ext4_get_block);
6669	if (!err) {
6670	ret = VM_FAULT_SIGBUS;
6671	if (ext4_journal_folio_buffers(handle, folio, len))
6672	goto out_error;
6673	} else {
6674	folio_unlock(folio);
6675	}
6676	}
6677	ext4_journal_stop(handle);
6678	if (err == -ENOSPC && ext4_should_retry_alloc(sb: inode->i_sb, retries: &retries))
6679	goto retry_alloc;
6680	out_ret:
6681	ret = vmf_fs_error(err);
6682	out:
6683	filemap_invalidate_unlock_shared(mapping);
6684	sb_end_pagefault(sb: inode->i_sb);
6685	return ret;
6686	out_error:
6687	folio_unlock(folio);
6688	ext4_journal_stop(handle);
6689	goto out;
6690	}
6691