file.c

  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *  linux/fs/ext4/file.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/file.c
 13 *
 14 *  Copyright (C) 1991, 1992  Linus Torvalds
 15 *
 16 *  ext4 fs regular file handling primitives
 17 *
 18 *  64-bit file support on 64-bit platforms by Jakub Jelinek
 19 *	(jj@sunsite.ms.mff.cuni.cz)
 20 */
 21
 22#include <linux/time.h>
 23#include <linux/fs.h>
 24#include <linux/iomap.h>
 25#include <linux/mount.h>
 26#include <linux/path.h>
 27#include <linux/dax.h>
 28#include <linux/quotaops.h>
 29#include <linux/pagevec.h>
 30#include <linux/uio.h>
 31#include <linux/mman.h>
 32#include <linux/backing-dev.h>
 33#include "ext4.h"
 34#include "ext4_jbd2.h"
 35#include "xattr.h"
 36#include "acl.h"
 37#include "truncate.h"
 38
 39/*
 40 * Returns %true if the given DIO request should be attempted with DIO, or
 41 * %false if it should fall back to buffered I/O.
 42 *
 43 * DIO isn't well specified; when it's unsupported (either due to the request
 44 * being misaligned, or due to the file not supporting DIO at all), filesystems
 45 * either fall back to buffered I/O or return EINVAL.  For files that don't use
 46 * any special features like encryption or verity, ext4 has traditionally
 47 * returned EINVAL for misaligned DIO.  iomap_dio_rw() uses this convention too.
 48 * In this case, we should attempt the DIO, *not* fall back to buffered I/O.
 49 *
 50 * In contrast, in cases where DIO is unsupported due to ext4 features, ext4
 51 * traditionally falls back to buffered I/O.
 52 *
 53 * This function implements the traditional ext4 behavior in all these cases.
 54 */
 55static bool ext4_should_use_dio(struct kiocb *iocb, struct iov_iter *iter)
 56{
 57	struct inode *inode = file_inode(iocb->ki_filp);
 58	u32 dio_align = ext4_dio_alignment(inode);
 59
 60	if (dio_align == 0)
 61		return false;
 62
 63	if (dio_align == 1)
 64		return true;
 65
 66	return IS_ALIGNED(iocb->ki_pos | iov_iter_alignment(iter), dio_align);
 67}
 68
 69static ssize_t ext4_dio_read_iter(struct kiocb *iocb, struct iov_iter *to)
 70{
 71	ssize_t ret;
 72	struct inode *inode = file_inode(iocb->ki_filp);
 73
 74	if (iocb->ki_flags & IOCB_NOWAIT) {
 75		if (!inode_trylock_shared(inode))
 76			return -EAGAIN;
 77	} else {
 78		inode_lock_shared(inode);
 79	}
 80
 81	if (!ext4_should_use_dio(iocb, to)) {
 82		inode_unlock_shared(inode);
 83		/*
 84		 * Fallback to buffered I/O if the operation being performed on
 85		 * the inode is not supported by direct I/O. The IOCB_DIRECT
 86		 * flag needs to be cleared here in order to ensure that the
 87		 * direct I/O path within generic_file_read_iter() is not
 88		 * taken.
 89		 */
 90		iocb->ki_flags &= ~IOCB_DIRECT;
 91		return generic_file_read_iter(iocb, to);
 92	}
 93
 94	ret = iomap_dio_rw(iocb, to, &ext4_iomap_ops, NULL, 0, NULL, 0);
 95	inode_unlock_shared(inode);
 96
 97	file_accessed(iocb->ki_filp);
 98	return ret;
 99}
100
101#ifdef CONFIG_FS_DAX
102static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
103{
104	struct inode *inode = file_inode(iocb->ki_filp);
105	ssize_t ret;
106
107	if (iocb->ki_flags & IOCB_NOWAIT) {
108		if (!inode_trylock_shared(inode))
109			return -EAGAIN;
110	} else {
111		inode_lock_shared(inode);
112	}
113	/*
114	 * Recheck under inode lock - at this point we are sure it cannot
115	 * change anymore
116	 */
117	if (!IS_DAX(inode)) {
118		inode_unlock_shared(inode);
119		/* Fallback to buffered IO in case we cannot support DAX */
120		return generic_file_read_iter(iocb, to);
121	}
122	ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
123	inode_unlock_shared(inode);
124
125	file_accessed(iocb->ki_filp);
126	return ret;
127}
128#endif
129
130static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
131{
132	struct inode *inode = file_inode(iocb->ki_filp);
133
134	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
135		return -EIO;
136
137	if (!iov_iter_count(to))
138		return 0; /* skip atime */
139
140#ifdef CONFIG_FS_DAX
141	if (IS_DAX(inode))
142		return ext4_dax_read_iter(iocb, to);
143#endif
144	if (iocb->ki_flags & IOCB_DIRECT)
145		return ext4_dio_read_iter(iocb, to);
146
147	return generic_file_read_iter(iocb, to);
148}
149
150static ssize_t ext4_file_splice_read(struct file *in, loff_t *ppos,
151				     struct pipe_inode_info *pipe,
152				     size_t len, unsigned int flags)
153{
154	struct inode *inode = file_inode(in);
155
156	if (unlikely(ext4_forced_shutdown(inode->i_sb)))
157		return -EIO;
158	return filemap_splice_read(in, ppos, pipe, len, flags);
159}
160
161/*
162 * Called when an inode is released. Note that this is different
163 * from ext4_file_open: open gets called at every open, but release
164 * gets called only when /all/ the files are closed.
165 */
166static int ext4_release_file(struct inode *inode, struct file *filp)
167{
168	if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
169		ext4_alloc_da_blocks(inode);
170		ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
171	}
172	/* if we are the last writer on the inode, drop the block reservation */
173	if ((filp->f_mode & FMODE_WRITE) &&
174			(atomic_read(&inode->i_writecount) == 1) &&
175			!EXT4_I(inode)->i_reserved_data_blocks) {
176		down_write(&EXT4_I(inode)->i_data_sem);
177		ext4_discard_preallocations(inode);
178		up_write(&EXT4_I(inode)->i_data_sem);
179	}
180	if (is_dx(inode) && filp->private_data)
181		ext4_htree_free_dir_info(filp->private_data);
182
183	return 0;
184}
185
186/*
187 * This tests whether the IO in question is block-aligned or not.
188 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
189 * are converted to written only after the IO is complete.  Until they are
190 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
191 * it needs to zero out portions of the start and/or end block.  If 2 AIO
192 * threads are at work on the same unwritten block, they must be synchronized
193 * or one thread will zero the other's data, causing corruption.
194 */
195static bool
196ext4_unaligned_io(struct inode *inode, struct iov_iter *from, loff_t pos)
197{
198	struct super_block *sb = inode->i_sb;
199	unsigned long blockmask = sb->s_blocksize - 1;
200
201	if ((pos | iov_iter_alignment(from)) & blockmask)
202		return true;
203
204	return false;
205}
206
207static bool
208ext4_extending_io(struct inode *inode, loff_t offset, size_t len)
209{
210	if (offset + len > i_size_read(inode) ||
211	    offset + len > EXT4_I(inode)->i_disksize)
212		return true;
213	return false;
214}
215
216/* Is IO overwriting allocated or initialized blocks? */
217static bool ext4_overwrite_io(struct inode *inode,
218			      loff_t pos, loff_t len, bool *unwritten)
219{
220	struct ext4_map_blocks map;
221	unsigned int blkbits = inode->i_blkbits;
222	int err, blklen;
223
224	if (pos + len > i_size_read(inode))
225		return false;
226
227	map.m_lblk = pos >> blkbits;
228	map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
229	blklen = map.m_len;
230
231	err = ext4_map_blocks(NULL, inode, &map, 0);
232	if (err != blklen)
233		return false;
234	/*
235	 * 'err==len' means that all of the blocks have been preallocated,
236	 * regardless of whether they have been initialized or not. We need to
237	 * check m_flags to distinguish the unwritten extents.
238	 */
239	*unwritten = !(map.m_flags & EXT4_MAP_MAPPED);
240	return true;
241}
242
243static ssize_t ext4_generic_write_checks(struct kiocb *iocb,
244					 struct iov_iter *from)
245{
246	struct inode *inode = file_inode(iocb->ki_filp);
247	ssize_t ret;
248
249	if (unlikely(IS_IMMUTABLE(inode)))
250		return -EPERM;
251
252	ret = generic_write_checks(iocb, from);
253	if (ret <= 0)
254		return ret;
255
256	/*
257	 * If we have encountered a bitmap-format file, the size limit
258	 * is smaller than s_maxbytes, which is for extent-mapped files.
259	 */
260	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
261		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
262
263		if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
264			return -EFBIG;
265		iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
266	}
267
268	return iov_iter_count(from);
269}
270
271static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
272{
273	ssize_t ret, count;
274
275	count = ext4_generic_write_checks(iocb, from);
276	if (count <= 0)
277		return count;
278
279	ret = file_modified(iocb->ki_filp);
280	if (ret)
281		return ret;
282	return count;
283}
284
285static ssize_t ext4_buffered_write_iter(struct kiocb *iocb,
286					struct iov_iter *from)
287{
288	ssize_t ret;
289	struct inode *inode = file_inode(iocb->ki_filp);
290
291	if (iocb->ki_flags & IOCB_NOWAIT)
292		return -EOPNOTSUPP;
293
294	inode_lock(inode);
295	ret = ext4_write_checks(iocb, from);
296	if (ret <= 0)
297		goto out;
298
299	ret = generic_perform_write(iocb, from);
300
301out:
302	inode_unlock(inode);
303	if (unlikely(ret <= 0))
304		return ret;
305	return generic_write_sync(iocb, ret);
306}
307
308static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
309					   ssize_t written, ssize_t count)
310{
311	handle_t *handle;
312
313	lockdep_assert_held_write(&inode->i_rwsem);
314	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
315	if (IS_ERR(handle))
316		return PTR_ERR(handle);
317
318	if (ext4_update_inode_size(inode, offset + written)) {
319		int ret = ext4_mark_inode_dirty(handle, inode);
320		if (unlikely(ret)) {
321			ext4_journal_stop(handle);
322			return ret;
323		}
324	}
325
326	if ((written == count) && inode->i_nlink)
327		ext4_orphan_del(handle, inode);
328	ext4_journal_stop(handle);
329
330	return written;
331}
332
333/*
334 * Clean up the inode after DIO or DAX extending write has completed and the
335 * inode size has been updated using ext4_handle_inode_extension().
336 */
337static void ext4_inode_extension_cleanup(struct inode *inode, bool need_trunc)
338{
339	lockdep_assert_held_write(&inode->i_rwsem);
340	if (need_trunc) {
341		ext4_truncate_failed_write(inode);
342		/*
343		 * If the truncate operation failed early, then the inode may
344		 * still be on the orphan list. In that case, we need to try
345		 * remove the inode from the in-memory linked list.
346		 */
347		if (inode->i_nlink)
348			ext4_orphan_del(NULL, inode);
349		return;
350	}
351	/*
352	 * If i_disksize got extended either due to writeback of delalloc
353	 * blocks or extending truncate while the DIO was running we could fail
354	 * to cleanup the orphan list in ext4_handle_inode_extension(). Do it
355	 * now.
356	 */
357	if (ext4_inode_orphan_tracked(inode) && inode->i_nlink) {
358		handle_t *handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
359
360		if (IS_ERR(handle)) {
361			/*
362			 * The write has successfully completed. Not much to
363			 * do with the error here so just cleanup the orphan
364			 * list and hope for the best.
365			 */
366			ext4_orphan_del(NULL, inode);
367			return;
368		}
369		ext4_orphan_del(handle, inode);
370		ext4_journal_stop(handle);
371	}
372}
373
374static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
375				 int error, unsigned int flags)
376{
377	loff_t pos = iocb->ki_pos;
378	struct inode *inode = file_inode(iocb->ki_filp);
379
380
381	if (!error && size && (flags & IOMAP_DIO_UNWRITTEN) &&
382			(iocb->ki_flags & IOCB_ATOMIC))
383		error = ext4_convert_unwritten_extents_atomic(NULL, inode, pos,
384							      size);
385	else if (!error && size && flags & IOMAP_DIO_UNWRITTEN)
386		error = ext4_convert_unwritten_extents(NULL, inode, pos, size);
387	if (error)
388		return error;
389	/*
390	 * Note that EXT4_I(inode)->i_disksize can get extended up to
391	 * inode->i_size while the I/O was running due to writeback of delalloc
392	 * blocks. But the code in ext4_iomap_alloc() is careful to use
393	 * zeroed/unwritten extents if this is possible; thus we won't leave
394	 * uninitialized blocks in a file even if we didn't succeed in writing
395	 * as much as we intended. Also we can race with truncate or write
396	 * expanding the file so we have to be a bit careful here.
397	 */
398	if (pos + size <= READ_ONCE(EXT4_I(inode)->i_disksize) &&
399	    pos + size <= i_size_read(inode))
400		return 0;
401	error = ext4_handle_inode_extension(inode, pos, size, size);
402	return error < 0 ? error : 0;
403}
404
405static const struct iomap_dio_ops ext4_dio_write_ops = {
406	.end_io = ext4_dio_write_end_io,
407};
408
409/*
410 * The intention here is to start with shared lock acquired then see if any
411 * condition requires an exclusive inode lock. If yes, then we restart the
412 * whole operation by releasing the shared lock and acquiring exclusive lock.
413 *
414 * - For unaligned_io we never take shared lock as it may cause data corruption
415 *   when two unaligned IO tries to modify the same block e.g. while zeroing.
416 *
417 * - For extending writes case we don't take the shared lock, since it requires
418 *   updating inode i_disksize and/or orphan handling with exclusive lock.
419 *
420 * - shared locking will only be true mostly with overwrites, including
421 *   initialized blocks and unwritten blocks. For overwrite unwritten blocks
422 *   we protect splitting extents by i_data_sem in ext4_inode_info, so we can
423 *   also release exclusive i_rwsem lock.
424 *
425 * - Otherwise we will switch to exclusive i_rwsem lock.
426 */
427static ssize_t ext4_dio_write_checks(struct kiocb *iocb, struct iov_iter *from,
428				     bool *ilock_shared, bool *extend,
429				     bool *unwritten, int *dio_flags)
430{
431	struct file *file = iocb->ki_filp;
432	struct inode *inode = file_inode(file);
433	loff_t offset;
434	size_t count;
435	ssize_t ret;
436	bool overwrite, unaligned_io;
437
438restart:
439	ret = ext4_generic_write_checks(iocb, from);
440	if (ret <= 0)
441		goto out;
442
443	offset = iocb->ki_pos;
444	count = ret;
445
446	unaligned_io = ext4_unaligned_io(inode, from, offset);
447	*extend = ext4_extending_io(inode, offset, count);
448	overwrite = ext4_overwrite_io(inode, offset, count, unwritten);
449
450	/*
451	 * Determine whether we need to upgrade to an exclusive lock. This is
452	 * required to change security info in file_modified(), for extending
453	 * I/O, any form of non-overwrite I/O, and unaligned I/O to unwritten
454	 * extents (as partial block zeroing may be required).
455	 *
456	 * Note that unaligned writes are allowed under shared lock so long as
457	 * they are pure overwrites. Otherwise, concurrent unaligned writes risk
458	 * data corruption due to partial block zeroing in the dio layer, and so
459	 * the I/O must occur exclusively.
460	 */
461	if (*ilock_shared &&
462	    ((!IS_NOSEC(inode) || *extend || !overwrite ||
463	     (unaligned_io && *unwritten)))) {
464		if (iocb->ki_flags & IOCB_NOWAIT) {
465			ret = -EAGAIN;
466			goto out;
467		}
468		inode_unlock_shared(inode);
469		*ilock_shared = false;
470		inode_lock(inode);
471		goto restart;
472	}
473
474	/*
475	 * Now that locking is settled, determine dio flags and exclusivity
476	 * requirements. We don't use DIO_OVERWRITE_ONLY because we enforce
477	 * behavior already. The inode lock is already held exclusive if the
478	 * write is non-overwrite or extending, so drain all outstanding dio and
479	 * set the force wait dio flag.
480	 */
481	if (!*ilock_shared && (unaligned_io || *extend)) {
482		if (iocb->ki_flags & IOCB_NOWAIT) {
483			ret = -EAGAIN;
484			goto out;
485		}
486		if (unaligned_io && (!overwrite || *unwritten))
487			inode_dio_wait(inode);
488		*dio_flags = IOMAP_DIO_FORCE_WAIT;
489	}
490
491	ret = file_modified(file);
492	if (ret < 0)
493		goto out;
494
495	return count;
496out:
497	if (*ilock_shared)
498		inode_unlock_shared(inode);
499	else
500		inode_unlock(inode);
501	return ret;
502}
503
504static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from)
505{
506	ssize_t ret;
507	handle_t *handle;
508	struct inode *inode = file_inode(iocb->ki_filp);
509	loff_t offset = iocb->ki_pos;
510	size_t count = iov_iter_count(from);
511	const struct iomap_ops *iomap_ops = &ext4_iomap_ops;
512	bool extend = false, unwritten = false;
513	bool ilock_shared = true;
514	int dio_flags = 0;
515
516	/*
517	 * Quick check here without any i_rwsem lock to see if it is extending
518	 * IO. A more reliable check is done in ext4_dio_write_checks() with
519	 * proper locking in place.
520	 */
521	if (offset + count > i_size_read(inode))
522		ilock_shared = false;
523
524	if (iocb->ki_flags & IOCB_NOWAIT) {
525		if (ilock_shared) {
526			if (!inode_trylock_shared(inode))
527				return -EAGAIN;
528		} else {
529			if (!inode_trylock(inode))
530				return -EAGAIN;
531		}
532	} else {
533		if (ilock_shared)
534			inode_lock_shared(inode);
535		else
536			inode_lock(inode);
537	}
538
539	/* Fallback to buffered I/O if the inode does not support direct I/O. */
540	if (!ext4_should_use_dio(iocb, from)) {
541		if (ilock_shared)
542			inode_unlock_shared(inode);
543		else
544			inode_unlock(inode);
545		return ext4_buffered_write_iter(iocb, from);
546	}
547
548	/*
549	 * Prevent inline data from being created since we are going to allocate
550	 * blocks for DIO. We know the inode does not currently have inline data
551	 * because ext4_should_use_dio() checked for it, but we have to clear
552	 * the state flag before the write checks because a lock cycle could
553	 * introduce races with other writers.
554	 */
555	ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
556
557	ret = ext4_dio_write_checks(iocb, from, &ilock_shared, &extend,
558				    &unwritten, &dio_flags);
559	if (ret <= 0)
560		return ret;
561
562	offset = iocb->ki_pos;
563	count = ret;
564
565	if (extend) {
566		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
567		if (IS_ERR(handle)) {
568			ret = PTR_ERR(handle);
569			goto out;
570		}
571
572		ret = ext4_orphan_add(handle, inode);
573		ext4_journal_stop(handle);
574		if (ret)
575			goto out;
576	}
577
578	if (ilock_shared && !unwritten)
579		iomap_ops = &ext4_iomap_overwrite_ops;
580	ret = iomap_dio_rw(iocb, from, iomap_ops, &ext4_dio_write_ops,
581			   dio_flags, NULL, 0);
582	if (ret == -ENOTBLK)
583		ret = 0;
584	if (extend) {
585		/*
586		 * We always perform extending DIO write synchronously so by
587		 * now the IO is completed and ext4_handle_inode_extension()
588		 * was called. Cleanup the inode in case of error or race with
589		 * writeback of delalloc blocks.
590		 */
591		WARN_ON_ONCE(ret == -EIOCBQUEUED);
592		ext4_inode_extension_cleanup(inode, ret < 0);
593	}
594
595out:
596	if (ilock_shared)
597		inode_unlock_shared(inode);
598	else
599		inode_unlock(inode);
600
601	if (ret >= 0 && iov_iter_count(from)) {
602		ssize_t err;
603		loff_t endbyte;
604
605		/*
606		 * There is no support for atomic writes on buffered-io yet,
607		 * we should never fallback to buffered-io for DIO atomic
608		 * writes.
609		 */
610		WARN_ON_ONCE(iocb->ki_flags & IOCB_ATOMIC);
611
612		offset = iocb->ki_pos;
613		err = ext4_buffered_write_iter(iocb, from);
614		if (err < 0)
615			return err;
616
617		/*
618		 * We need to ensure that the pages within the page cache for
619		 * the range covered by this I/O are written to disk and
620		 * invalidated. This is in attempt to preserve the expected
621		 * direct I/O semantics in the case we fallback to buffered I/O
622		 * to complete off the I/O request.
623		 */
624		ret += err;
625		endbyte = offset + err - 1;
626		err = filemap_write_and_wait_range(iocb->ki_filp->f_mapping,
627						   offset, endbyte);
628		if (!err)
629			invalidate_mapping_pages(iocb->ki_filp->f_mapping,
630						 offset >> PAGE_SHIFT,
631						 endbyte >> PAGE_SHIFT);
632	}
633
634	return ret;
635}
636
637#ifdef CONFIG_FS_DAX
638static ssize_t
639ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
640{
641	ssize_t ret;
642	size_t count;
643	loff_t offset;
644	handle_t *handle;
645	bool extend = false;
646	struct inode *inode = file_inode(iocb->ki_filp);
647
648	if (iocb->ki_flags & IOCB_NOWAIT) {
649		if (!inode_trylock(inode))
650			return -EAGAIN;
651	} else {
652		inode_lock(inode);
653	}
654
655	ret = ext4_write_checks(iocb, from);
656	if (ret <= 0)
657		goto out;
658
659	offset = iocb->ki_pos;
660	count = iov_iter_count(from);
661
662	if (offset + count > EXT4_I(inode)->i_disksize) {
663		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
664		if (IS_ERR(handle)) {
665			ret = PTR_ERR(handle);
666			goto out;
667		}
668
669		ret = ext4_orphan_add(handle, inode);
670		if (ret) {
671			ext4_journal_stop(handle);
672			goto out;
673		}
674
675		extend = true;
676		ext4_journal_stop(handle);
677	}
678
679	ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
680
681	if (extend) {
682		ret = ext4_handle_inode_extension(inode, offset, ret, count);
683		ext4_inode_extension_cleanup(inode, ret < (ssize_t)count);
684	}
685out:
686	inode_unlock(inode);
687	if (ret > 0)
688		ret = generic_write_sync(iocb, ret);
689	return ret;
690}
691#endif
692
693static ssize_t
694ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
695{
696	int ret;
697	struct inode *inode = file_inode(iocb->ki_filp);
698
699	ret = ext4_emergency_state(inode->i_sb);
700	if (unlikely(ret))
701		return ret;
702
703#ifdef CONFIG_FS_DAX
704	if (IS_DAX(inode))
705		return ext4_dax_write_iter(iocb, from);
706#endif
707
708	if (iocb->ki_flags & IOCB_ATOMIC) {
709		size_t len = iov_iter_count(from);
710
711		if (len < EXT4_SB(inode->i_sb)->s_awu_min ||
712		    len > EXT4_SB(inode->i_sb)->s_awu_max)
713			return -EINVAL;
714
715		ret = generic_atomic_write_valid(iocb, from);
716		if (ret)
717			return ret;
718	}
719
720	if (iocb->ki_flags & IOCB_DIRECT)
721		return ext4_dio_write_iter(iocb, from);
722	else
723		return ext4_buffered_write_iter(iocb, from);
724}
725
726#ifdef CONFIG_FS_DAX
727static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf, unsigned int order)
728{
729	int error = 0;
730	vm_fault_t result;
731	int retries = 0;
732	handle_t *handle = NULL;
733	struct inode *inode = file_inode(vmf->vma->vm_file);
734	struct super_block *sb = inode->i_sb;
735
736	/*
737	 * We have to distinguish real writes from writes which will result in a
738	 * COW page; COW writes should *not* poke the journal (the file will not
739	 * be changed). Doing so would cause unintended failures when mounted
740	 * read-only.
741	 *
742	 * We check for VM_SHARED rather than vmf->cow_page since the latter is
743	 * unset for order != 0 (i.e. only in do_cow_fault); for
744	 * other sizes, dax_iomap_fault will handle splitting / fallback so that
745	 * we eventually come back with a COW page.
746	 */
747	bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
748		(vmf->vma->vm_flags & VM_SHARED);
749	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
750	unsigned long pfn;
751
752	if (write) {
753		sb_start_pagefault(sb);
754		file_update_time(vmf->vma->vm_file);
755		filemap_invalidate_lock_shared(mapping);
756retry:
757		handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
758					       EXT4_DATA_TRANS_BLOCKS(sb));
759		if (IS_ERR(handle)) {
760			filemap_invalidate_unlock_shared(mapping);
761			sb_end_pagefault(sb);
762			return VM_FAULT_SIGBUS;
763		}
764	} else {
765		filemap_invalidate_lock_shared(mapping);
766	}
767	result = dax_iomap_fault(vmf, order, &pfn, &error, &ext4_iomap_ops);
768	if (write) {
769		ext4_journal_stop(handle);
770
771		if ((result & VM_FAULT_ERROR) && error == -ENOSPC &&
772		    ext4_should_retry_alloc(sb, &retries))
773			goto retry;
774		/* Handling synchronous page fault? */
775		if (result & VM_FAULT_NEEDDSYNC)
776			result = dax_finish_sync_fault(vmf, order, pfn);
777		filemap_invalidate_unlock_shared(mapping);
778		sb_end_pagefault(sb);
779	} else {
780		filemap_invalidate_unlock_shared(mapping);
781	}
782
783	return result;
784}
785
786static vm_fault_t ext4_dax_fault(struct vm_fault *vmf)
787{
788	return ext4_dax_huge_fault(vmf, 0);
789}
790
791static const struct vm_operations_struct ext4_dax_vm_ops = {
792	.fault		= ext4_dax_fault,
793	.huge_fault	= ext4_dax_huge_fault,
794	.page_mkwrite	= ext4_dax_fault,
795	.pfn_mkwrite	= ext4_dax_fault,
796};
797#else
798#define ext4_dax_vm_ops	ext4_file_vm_ops
799#endif
800
801static const struct vm_operations_struct ext4_file_vm_ops = {
802	.fault		= filemap_fault,
803	.map_pages	= filemap_map_pages,
804	.page_mkwrite   = ext4_page_mkwrite,
805};
806
807static int ext4_file_mmap_prepare(struct vm_area_desc *desc)
808{
809	int ret;
810	struct file *file = desc->file;
811	struct inode *inode = file->f_mapping->host;
812	struct dax_device *dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
813
814	if (file->f_mode & FMODE_WRITE)
815		ret = ext4_emergency_state(inode->i_sb);
816	else
817		ret = ext4_forced_shutdown(inode->i_sb) ? -EIO : 0;
818	if (unlikely(ret))
819		return ret;
820
821	/*
822	 * We don't support synchronous mappings for non-DAX files and
823	 * for DAX files if underneath dax_device is not synchronous.
824	 */
825	if (!daxdev_mapping_supported(desc->vm_flags, file_inode(file), dax_dev))
826		return -EOPNOTSUPP;
827
828	file_accessed(file);
829	if (IS_DAX(file_inode(file))) {
830		desc->vm_ops = &ext4_dax_vm_ops;
831		desc->vm_flags |= VM_HUGEPAGE;
832	} else {
833		desc->vm_ops = &ext4_file_vm_ops;
834	}
835	return 0;
836}
837
838static int ext4_sample_last_mounted(struct super_block *sb,
839				    struct vfsmount *mnt)
840{
841	struct ext4_sb_info *sbi = EXT4_SB(sb);
842	struct path path;
843	char buf[64], *cp;
844	handle_t *handle;
845	int err;
846
847	if (likely(ext4_test_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED)))
848		return 0;
849
850	if (ext4_emergency_state(sb) || sb_rdonly(sb) ||
851	    !sb_start_intwrite_trylock(sb))
852		return 0;
853
854	ext4_set_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED);
855	/*
856	 * Sample where the filesystem has been mounted and
857	 * store it in the superblock for sysadmin convenience
858	 * when trying to sort through large numbers of block
859	 * devices or filesystem images.
860	 */
861	memset(buf, 0, sizeof(buf));
862	path.mnt = mnt;
863	path.dentry = mnt->mnt_root;
864	cp = d_path(&path, buf, sizeof(buf));
865	err = 0;
866	if (IS_ERR(cp))
867		goto out;
868
869	handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
870	err = PTR_ERR(handle);
871	if (IS_ERR(handle))
872		goto out;
873	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
874	err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
875					    EXT4_JTR_NONE);
876	if (err)
877		goto out_journal;
878	lock_buffer(sbi->s_sbh);
879	strtomem_pad(sbi->s_es->s_last_mounted, cp, 0);
880	ext4_superblock_csum_set(sb);
881	unlock_buffer(sbi->s_sbh);
882	ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
883out_journal:
884	ext4_journal_stop(handle);
885out:
886	sb_end_intwrite(sb);
887	return err;
888}
889
890static int ext4_file_open(struct inode *inode, struct file *filp)
891{
892	int ret;
893
894	if (filp->f_mode & FMODE_WRITE)
895		ret = ext4_emergency_state(inode->i_sb);
896	else
897		ret = ext4_forced_shutdown(inode->i_sb) ? -EIO : 0;
898	if (unlikely(ret))
899		return ret;
900
901	ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt);
902	if (ret)
903		return ret;
904
905	ret = fscrypt_file_open(inode, filp);
906	if (ret)
907		return ret;
908
909	ret = fsverity_file_open(inode, filp);
910	if (ret)
911		return ret;
912
913	/*
914	 * Set up the jbd2_inode if we are opening the inode for
915	 * writing and the journal is present
916	 */
917	if (filp->f_mode & FMODE_WRITE) {
918		ret = ext4_inode_attach_jinode(inode);
919		if (ret < 0)
920			return ret;
921	}
922
923	if (ext4_inode_can_atomic_write(inode))
924		filp->f_mode |= FMODE_CAN_ATOMIC_WRITE;
925
926	filp->f_mode |= FMODE_NOWAIT | FMODE_CAN_ODIRECT;
927	return dquot_file_open(inode, filp);
928}
929
930/*
931 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
932 * by calling generic_file_llseek_size() with the appropriate maxbytes
933 * value for each.
934 */
935loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
936{
937	struct inode *inode = file->f_mapping->host;
938	loff_t maxbytes = ext4_get_maxbytes(inode);
939
940	switch (whence) {
941	default:
942		return generic_file_llseek_size(file, offset, whence,
943						maxbytes, i_size_read(inode));
944	case SEEK_HOLE:
945		inode_lock_shared(inode);
946		offset = iomap_seek_hole(inode, offset,
947					 &ext4_iomap_report_ops);
948		inode_unlock_shared(inode);
949		break;
950	case SEEK_DATA:
951		inode_lock_shared(inode);
952		offset = iomap_seek_data(inode, offset,
953					 &ext4_iomap_report_ops);
954		inode_unlock_shared(inode);
955		break;
956	}
957
958	if (offset < 0)
959		return offset;
960	return vfs_setpos(file, offset, maxbytes);
961}
962
963const struct file_operations ext4_file_operations = {
964	.llseek		= ext4_llseek,
965	.read_iter	= ext4_file_read_iter,
966	.write_iter	= ext4_file_write_iter,
967	.iopoll		= iocb_bio_iopoll,
968	.unlocked_ioctl = ext4_ioctl,
969#ifdef CONFIG_COMPAT
970	.compat_ioctl	= ext4_compat_ioctl,
971#endif
972	.mmap_prepare	= ext4_file_mmap_prepare,
973	.open		= ext4_file_open,
974	.release	= ext4_release_file,
975	.fsync		= ext4_sync_file,
976	.get_unmapped_area = thp_get_unmapped_area,
977	.splice_read	= ext4_file_splice_read,
978	.splice_write	= iter_file_splice_write,
979	.fallocate	= ext4_fallocate,
980	.fop_flags	= FOP_MMAP_SYNC | FOP_BUFFER_RASYNC |
981			  FOP_DIO_PARALLEL_WRITE |
982			  FOP_DONTCACHE,
983};
984
985const struct inode_operations ext4_file_inode_operations = {
986	.setattr	= ext4_setattr,
987	.getattr	= ext4_file_getattr,
988	.listxattr	= ext4_listxattr,
989	.get_inode_acl	= ext4_get_acl,
990	.set_acl	= ext4_set_acl,
991	.fiemap		= ext4_fiemap,
992	.fileattr_get	= ext4_fileattr_get,
993	.fileattr_set	= ext4_fileattr_set,
994};
995