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(EXT4_SB(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
150/*
151 * Called when an inode is released. Note that this is different
152 * from ext4_file_open: open gets called at every open, but release
153 * gets called only when /all/ the files are closed.
154 */
155static int ext4_release_file(struct inode *inode, struct file *filp)
156{
157	if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
158		ext4_alloc_da_blocks(inode);
159		ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
160	}
161	/* if we are the last writer on the inode, drop the block reservation */
162	if ((filp->f_mode & FMODE_WRITE) &&
163			(atomic_read(&inode->i_writecount) == 1) &&
164			!EXT4_I(inode)->i_reserved_data_blocks) {
 
165		down_write(&EXT4_I(inode)->i_data_sem);
166		ext4_discard_preallocations(inode, 0);
167		up_write(&EXT4_I(inode)->i_data_sem);
168	}
169	if (is_dx(inode) && filp->private_data)
170		ext4_htree_free_dir_info(filp->private_data);
171
172	return 0;
173}
174
 
 
 
 
 
 
 
175/*
176 * This tests whether the IO in question is block-aligned or not.
177 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
178 * are converted to written only after the IO is complete.  Until they are
179 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
180 * it needs to zero out portions of the start and/or end block.  If 2 AIO
181 * threads are at work on the same unwritten block, they must be synchronized
182 * or one thread will zero the other's data, causing corruption.
183 */
184static bool
185ext4_unaligned_io(struct inode *inode, struct iov_iter *from, loff_t pos)
186{
187	struct super_block *sb = inode->i_sb;
188	unsigned long blockmask = sb->s_blocksize - 1;
189
190	if ((pos | iov_iter_alignment(from)) & blockmask)
191		return true;
192
193	return false;
194}
195
196static bool
197ext4_extending_io(struct inode *inode, loff_t offset, size_t len)
198{
199	if (offset + len > i_size_read(inode) ||
200	    offset + len > EXT4_I(inode)->i_disksize)
201		return true;
202	return false;
203}
204
205/* Is IO overwriting allocated and initialized blocks? */
206static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
207{
208	struct ext4_map_blocks map;
209	unsigned int blkbits = inode->i_blkbits;
210	int err, blklen;
211
212	if (pos + len > i_size_read(inode))
213		return false;
214
215	map.m_lblk = pos >> blkbits;
216	map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
217	blklen = map.m_len;
218
219	err = ext4_map_blocks(NULL, inode, &map, 0);
220	/*
221	 * 'err==len' means that all of the blocks have been preallocated,
222	 * regardless of whether they have been initialized or not. To exclude
223	 * unwritten extents, we need to check m_flags.
224	 */
225	return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
226}
227
228static ssize_t ext4_generic_write_checks(struct kiocb *iocb,
229					 struct iov_iter *from)
230{
231	struct inode *inode = file_inode(iocb->ki_filp);
232	ssize_t ret;
233
234	if (unlikely(IS_IMMUTABLE(inode)))
235		return -EPERM;
236
237	ret = generic_write_checks(iocb, from);
238	if (ret <= 0)
239		return ret;
240
 
 
 
241	/*
242	 * If we have encountered a bitmap-format file, the size limit
243	 * is smaller than s_maxbytes, which is for extent-mapped files.
244	 */
245	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
246		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
247
248		if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
249			return -EFBIG;
250		iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
251	}
252
253	return iov_iter_count(from);
254}
255
256static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
257{
258	ssize_t ret, count;
259
260	count = ext4_generic_write_checks(iocb, from);
261	if (count <= 0)
262		return count;
263
264	ret = file_modified(iocb->ki_filp);
265	if (ret)
266		return ret;
267	return count;
268}
269
270static ssize_t ext4_buffered_write_iter(struct kiocb *iocb,
271					struct iov_iter *from)
272{
273	ssize_t ret;
274	struct inode *inode = file_inode(iocb->ki_filp);
 
275
276	if (iocb->ki_flags & IOCB_NOWAIT)
277		return -EOPNOTSUPP;
278
279	inode_lock(inode);
 
280	ret = ext4_write_checks(iocb, from);
281	if (ret <= 0)
282		goto out;
 
 
 
 
 
 
283
284	current->backing_dev_info = inode_to_bdi(inode);
285	ret = generic_perform_write(iocb, from);
286	current->backing_dev_info = NULL;
287
288out:
289	inode_unlock(inode);
290	if (likely(ret > 0)) {
291		iocb->ki_pos += ret;
292		ret = generic_write_sync(iocb, ret);
293	}
294
295	return ret;
296}
 
297
298static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
299					   ssize_t written, size_t count)
300{
301	handle_t *handle;
302	bool truncate = false;
303	u8 blkbits = inode->i_blkbits;
304	ext4_lblk_t written_blk, end_blk;
305	int ret;
306
307	/*
308	 * Note that EXT4_I(inode)->i_disksize can get extended up to
309	 * inode->i_size while the I/O was running due to writeback of delalloc
310	 * blocks. But, the code in ext4_iomap_alloc() is careful to use
311	 * zeroed/unwritten extents if this is possible; thus we won't leave
312	 * uninitialized blocks in a file even if we didn't succeed in writing
313	 * as much as we intended.
314	 */
315	WARN_ON_ONCE(i_size_read(inode) < EXT4_I(inode)->i_disksize);
316	if (offset + count <= EXT4_I(inode)->i_disksize) {
317		/*
318		 * We need to ensure that the inode is removed from the orphan
319		 * list if it has been added prematurely, due to writeback of
320		 * delalloc blocks.
321		 */
322		if (!list_empty(&EXT4_I(inode)->i_orphan) && inode->i_nlink) {
323			handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
324
325			if (IS_ERR(handle)) {
326				ext4_orphan_del(NULL, inode);
327				return PTR_ERR(handle);
328			}
329
330			ext4_orphan_del(handle, inode);
331			ext4_journal_stop(handle);
332		}
333
334		return written;
335	}
336
337	if (written < 0)
338		goto truncate;
339
340	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
341	if (IS_ERR(handle)) {
342		written = PTR_ERR(handle);
343		goto truncate;
344	}
345
346	if (ext4_update_inode_size(inode, offset + written)) {
347		ret = ext4_mark_inode_dirty(handle, inode);
348		if (unlikely(ret)) {
349			written = ret;
350			ext4_journal_stop(handle);
351			goto truncate;
352		}
353	}
354
355	/*
356	 * We may need to truncate allocated but not written blocks beyond EOF.
357	 */
358	written_blk = ALIGN(offset + written, 1 << blkbits);
359	end_blk = ALIGN(offset + count, 1 << blkbits);
360	if (written_blk < end_blk && ext4_can_truncate(inode))
361		truncate = true;
362
363	/*
364	 * Remove the inode from the orphan list if it has been extended and
365	 * everything went OK.
366	 */
367	if (!truncate && inode->i_nlink)
368		ext4_orphan_del(handle, inode);
369	ext4_journal_stop(handle);
370
371	if (truncate) {
372truncate:
373		ext4_truncate_failed_write(inode);
374		/*
375		 * If the truncate operation failed early, then the inode may
376		 * still be on the orphan list. In that case, we need to try
377		 * remove the inode from the in-memory linked list.
378		 */
379		if (inode->i_nlink)
380			ext4_orphan_del(NULL, inode);
381	}
382
383	return written;
384}
385
386static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
387				 int error, unsigned int flags)
388{
389	loff_t pos = iocb->ki_pos;
390	struct inode *inode = file_inode(iocb->ki_filp);
391
392	if (error)
393		return error;
394
395	if (size && flags & IOMAP_DIO_UNWRITTEN) {
396		error = ext4_convert_unwritten_extents(NULL, inode, pos, size);
397		if (error < 0)
398			return error;
399	}
400	/*
401	 * If we are extending the file, we have to update i_size here before
402	 * page cache gets invalidated in iomap_dio_rw(). Otherwise racing
403	 * buffered reads could zero out too much from page cache pages. Update
404	 * of on-disk size will happen later in ext4_dio_write_iter() where
405	 * we have enough information to also perform orphan list handling etc.
406	 * Note that we perform all extending writes synchronously under
407	 * i_rwsem held exclusively so i_size update is safe here in that case.
408	 * If the write was not extending, we cannot see pos > i_size here
409	 * because operations reducing i_size like truncate wait for all
410	 * outstanding DIO before updating i_size.
411	 */
412	pos += size;
413	if (pos > i_size_read(inode))
414		i_size_write(inode, pos);
415
416	return 0;
417}
418
419static const struct iomap_dio_ops ext4_dio_write_ops = {
420	.end_io = ext4_dio_write_end_io,
421};
422
423/*
424 * The intention here is to start with shared lock acquired then see if any
425 * condition requires an exclusive inode lock. If yes, then we restart the
426 * whole operation by releasing the shared lock and acquiring exclusive lock.
427 *
428 * - For unaligned_io we never take shared lock as it may cause data corruption
429 *   when two unaligned IO tries to modify the same block e.g. while zeroing.
430 *
431 * - For extending writes case we don't take the shared lock, since it requires
432 *   updating inode i_disksize and/or orphan handling with exclusive lock.
433 *
434 * - shared locking will only be true mostly with overwrites. Otherwise we will
435 *   switch to exclusive i_rwsem lock.
436 */
437static ssize_t ext4_dio_write_checks(struct kiocb *iocb, struct iov_iter *from,
438				     bool *ilock_shared, bool *extend)
439{
440	struct file *file = iocb->ki_filp;
441	struct inode *inode = file_inode(file);
442	loff_t offset;
443	size_t count;
444	ssize_t ret;
445
446restart:
447	ret = ext4_generic_write_checks(iocb, from);
448	if (ret <= 0)
449		goto out;
450
451	offset = iocb->ki_pos;
452	count = ret;
453	if (ext4_extending_io(inode, offset, count))
454		*extend = true;
455	/*
456	 * Determine whether the IO operation will overwrite allocated
457	 * and initialized blocks.
458	 * We need exclusive i_rwsem for changing security info
459	 * in file_modified().
460	 */
461	if (*ilock_shared && (!IS_NOSEC(inode) || *extend ||
462	     !ext4_overwrite_io(inode, offset, count))) {
463		if (iocb->ki_flags & IOCB_NOWAIT) {
464			ret = -EAGAIN;
465			goto out;
466		}
467		inode_unlock_shared(inode);
468		*ilock_shared = false;
469		inode_lock(inode);
470		goto restart;
471	}
472
473	ret = file_modified(file);
474	if (ret < 0)
475		goto out;
476
477	return count;
478out:
479	if (*ilock_shared)
480		inode_unlock_shared(inode);
481	else
482		inode_unlock(inode);
483	return ret;
484}
485
486static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from)
487{
488	ssize_t ret;
489	handle_t *handle;
490	struct inode *inode = file_inode(iocb->ki_filp);
491	loff_t offset = iocb->ki_pos;
492	size_t count = iov_iter_count(from);
493	const struct iomap_ops *iomap_ops = &ext4_iomap_ops;
494	bool extend = false, unaligned_io = false;
495	bool ilock_shared = true;
496
497	/*
498	 * We initially start with shared inode lock unless it is
499	 * unaligned IO which needs exclusive lock anyways.
 
500	 */
501	if (ext4_unaligned_io(inode, from, offset)) {
502		unaligned_io = true;
503		ilock_shared = false;
 
 
504	}
505	/*
506	 * Quick check here without any i_rwsem lock to see if it is extending
507	 * IO. A more reliable check is done in ext4_dio_write_checks() with
508	 * proper locking in place.
509	 */
510	if (offset + count > i_size_read(inode))
511		ilock_shared = false;
512
513	if (iocb->ki_flags & IOCB_NOWAIT) {
514		if (ilock_shared) {
515			if (!inode_trylock_shared(inode))
516				return -EAGAIN;
517		} else {
518			if (!inode_trylock(inode))
519				return -EAGAIN;
 
 
520		}
521	} else {
522		if (ilock_shared)
523			inode_lock_shared(inode);
524		else
525			inode_lock(inode);
526	}
527
528	/* Fallback to buffered I/O if the inode does not support direct I/O. */
529	if (!ext4_should_use_dio(iocb, from)) {
530		if (ilock_shared)
531			inode_unlock_shared(inode);
532		else
533			inode_unlock(inode);
534		return ext4_buffered_write_iter(iocb, from);
535	}
536
537	ret = ext4_dio_write_checks(iocb, from, &ilock_shared, &extend);
538	if (ret <= 0)
539		return ret;
540
541	/* if we're going to block and IOCB_NOWAIT is set, return -EAGAIN */
542	if ((iocb->ki_flags & IOCB_NOWAIT) && (unaligned_io || extend)) {
543		ret = -EAGAIN;
544		goto out;
545	}
546	/*
547	 * Make sure inline data cannot be created anymore since we are going
548	 * to allocate blocks for DIO. We know the inode does not have any
549	 * inline data now because ext4_dio_supported() checked for that.
550	 */
551	ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
552
553	offset = iocb->ki_pos;
554	count = ret;
555
 
556	/*
557	 * Unaligned direct IO must be serialized among each other as zeroing
558	 * of partial blocks of two competing unaligned IOs can result in data
559	 * corruption.
560	 *
561	 * So we make sure we don't allow any unaligned IO in flight.
562	 * For IOs where we need not wait (like unaligned non-AIO DIO),
563	 * below inode_dio_wait() may anyway become a no-op, since we start
564	 * with exclusive lock.
565	 */
566	if (unaligned_io)
567		inode_dio_wait(inode);
568
569	if (extend) {
570		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
571		if (IS_ERR(handle)) {
572			ret = PTR_ERR(handle);
573			goto out;
574		}
575
576		ret = ext4_orphan_add(handle, inode);
577		if (ret) {
578			ext4_journal_stop(handle);
579			goto out;
580		}
581
582		ext4_journal_stop(handle);
583	}
584
585	if (ilock_shared)
586		iomap_ops = &ext4_iomap_overwrite_ops;
587	ret = iomap_dio_rw(iocb, from, iomap_ops, &ext4_dio_write_ops,
588			   (unaligned_io || extend) ? IOMAP_DIO_FORCE_WAIT : 0,
589			   NULL, 0);
590	if (ret == -ENOTBLK)
591		ret = 0;
592
593	if (extend)
594		ret = ext4_handle_inode_extension(inode, offset, ret, count);
595
596out:
597	if (ilock_shared)
598		inode_unlock_shared(inode);
599	else
600		inode_unlock(inode);
601
602	if (ret >= 0 && iov_iter_count(from)) {
603		ssize_t err;
604		loff_t endbyte;
605
606		offset = iocb->ki_pos;
607		err = ext4_buffered_write_iter(iocb, from);
608		if (err < 0)
609			return err;
610
611		/*
612		 * We need to ensure that the pages within the page cache for
613		 * the range covered by this I/O are written to disk and
614		 * invalidated. This is in attempt to preserve the expected
615		 * direct I/O semantics in the case we fallback to buffered I/O
616		 * to complete off the I/O request.
617		 */
618		ret += err;
619		endbyte = offset + err - 1;
620		err = filemap_write_and_wait_range(iocb->ki_filp->f_mapping,
621						   offset, endbyte);
622		if (!err)
623			invalidate_mapping_pages(iocb->ki_filp->f_mapping,
624						 offset >> PAGE_SHIFT,
625						 endbyte >> PAGE_SHIFT);
626	}
627
628	return ret;
629}
630
631#ifdef CONFIG_FS_DAX
632static ssize_t
633ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
634{
635	ssize_t ret;
636	size_t count;
637	loff_t offset;
638	handle_t *handle;
639	bool extend = false;
640	struct inode *inode = file_inode(iocb->ki_filp);
641
642	if (iocb->ki_flags & IOCB_NOWAIT) {
643		if (!inode_trylock(inode))
644			return -EAGAIN;
645	} else {
646		inode_lock(inode);
647	}
648
649	ret = ext4_write_checks(iocb, from);
650	if (ret <= 0)
651		goto out;
652
653	offset = iocb->ki_pos;
654	count = iov_iter_count(from);
655
656	if (offset + count > EXT4_I(inode)->i_disksize) {
657		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
658		if (IS_ERR(handle)) {
659			ret = PTR_ERR(handle);
660			goto out;
661		}
662
663		ret = ext4_orphan_add(handle, inode);
664		if (ret) {
665			ext4_journal_stop(handle);
666			goto out;
667		}
668
669		extend = true;
670		ext4_journal_stop(handle);
671	}
672
673	ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
674
675	if (extend)
676		ret = ext4_handle_inode_extension(inode, offset, ret, count);
677out:
678	inode_unlock(inode);
679	if (ret > 0)
680		ret = generic_write_sync(iocb, ret);
681	return ret;
682}
683#endif
684
685static ssize_t
686ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
687{
688	struct inode *inode = file_inode(iocb->ki_filp);
689
690	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
691		return -EIO;
692
693#ifdef CONFIG_FS_DAX
694	if (IS_DAX(inode))
695		return ext4_dax_write_iter(iocb, from);
696#endif
697	if (iocb->ki_flags & IOCB_DIRECT)
698		return ext4_dio_write_iter(iocb, from);
699	else
700		return ext4_buffered_write_iter(iocb, from);
701}
702
703#ifdef CONFIG_FS_DAX
704static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf,
705		enum page_entry_size pe_size)
706{
707	int error = 0;
708	vm_fault_t result;
709	int retries = 0;
710	handle_t *handle = NULL;
711	struct inode *inode = file_inode(vmf->vma->vm_file);
712	struct super_block *sb = inode->i_sb;
713
714	/*
715	 * We have to distinguish real writes from writes which will result in a
716	 * COW page; COW writes should *not* poke the journal (the file will not
717	 * be changed). Doing so would cause unintended failures when mounted
718	 * read-only.
719	 *
720	 * We check for VM_SHARED rather than vmf->cow_page since the latter is
721	 * unset for pe_size != PE_SIZE_PTE (i.e. only in do_cow_fault); for
722	 * other sizes, dax_iomap_fault will handle splitting / fallback so that
723	 * we eventually come back with a COW page.
724	 */
725	bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
726		(vmf->vma->vm_flags & VM_SHARED);
727	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
728	pfn_t pfn;
729
730	if (write) {
731		sb_start_pagefault(sb);
732		file_update_time(vmf->vma->vm_file);
733		filemap_invalidate_lock_shared(mapping);
734retry:
735		handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
736					       EXT4_DATA_TRANS_BLOCKS(sb));
737		if (IS_ERR(handle)) {
738			filemap_invalidate_unlock_shared(mapping);
739			sb_end_pagefault(sb);
740			return VM_FAULT_SIGBUS;
741		}
742	} else {
743		filemap_invalidate_lock_shared(mapping);
744	}
745	result = dax_iomap_fault(vmf, pe_size, &pfn, &error, &ext4_iomap_ops);
746	if (write) {
747		ext4_journal_stop(handle);
748
749		if ((result & VM_FAULT_ERROR) && error == -ENOSPC &&
750		    ext4_should_retry_alloc(sb, &retries))
751			goto retry;
752		/* Handling synchronous page fault? */
753		if (result & VM_FAULT_NEEDDSYNC)
754			result = dax_finish_sync_fault(vmf, pe_size, pfn);
755		filemap_invalidate_unlock_shared(mapping);
756		sb_end_pagefault(sb);
757	} else {
758		filemap_invalidate_unlock_shared(mapping);
759	}
760
761	return result;
762}
763
764static vm_fault_t ext4_dax_fault(struct vm_fault *vmf)
765{
766	return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
767}
768
769static const struct vm_operations_struct ext4_dax_vm_ops = {
770	.fault		= ext4_dax_fault,
771	.huge_fault	= ext4_dax_huge_fault,
772	.page_mkwrite	= ext4_dax_fault,
773	.pfn_mkwrite	= ext4_dax_fault,
774};
775#else
776#define ext4_dax_vm_ops	ext4_file_vm_ops
777#endif
778
779static const struct vm_operations_struct ext4_file_vm_ops = {
780	.fault		= filemap_fault,
781	.map_pages	= filemap_map_pages,
782	.page_mkwrite   = ext4_page_mkwrite,
783};
784
785static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
786{
787	struct inode *inode = file->f_mapping->host;
788	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
789	struct dax_device *dax_dev = sbi->s_daxdev;
790
791	if (unlikely(ext4_forced_shutdown(sbi)))
792		return -EIO;
793
794	/*
795	 * We don't support synchronous mappings for non-DAX files and
796	 * for DAX files if underneath dax_device is not synchronous.
797	 */
798	if (!daxdev_mapping_supported(vma, dax_dev))
799		return -EOPNOTSUPP;
800
801	file_accessed(file);
802	if (IS_DAX(file_inode(file))) {
803		vma->vm_ops = &ext4_dax_vm_ops;
804		vma->vm_flags |= VM_HUGEPAGE;
805	} else {
806		vma->vm_ops = &ext4_file_vm_ops;
807	}
808	return 0;
809}
810
811static int ext4_sample_last_mounted(struct super_block *sb,
812				    struct vfsmount *mnt)
813{
814	struct ext4_sb_info *sbi = EXT4_SB(sb);
815	struct path path;
816	char buf[64], *cp;
817	handle_t *handle;
818	int err;
819
820	if (likely(ext4_test_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED)))
821		return 0;
822
823	if (sb_rdonly(sb) || !sb_start_intwrite_trylock(sb))
824		return 0;
825
826	ext4_set_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED);
827	/*
828	 * Sample where the filesystem has been mounted and
829	 * store it in the superblock for sysadmin convenience
830	 * when trying to sort through large numbers of block
831	 * devices or filesystem images.
832	 */
833	memset(buf, 0, sizeof(buf));
834	path.mnt = mnt;
835	path.dentry = mnt->mnt_root;
836	cp = d_path(&path, buf, sizeof(buf));
837	err = 0;
838	if (IS_ERR(cp))
839		goto out;
840
841	handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
842	err = PTR_ERR(handle);
843	if (IS_ERR(handle))
844		goto out;
845	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
846	err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
847					    EXT4_JTR_NONE);
848	if (err)
849		goto out_journal;
850	lock_buffer(sbi->s_sbh);
851	strncpy(sbi->s_es->s_last_mounted, cp,
852		sizeof(sbi->s_es->s_last_mounted));
853	ext4_superblock_csum_set(sb);
854	unlock_buffer(sbi->s_sbh);
855	ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
856out_journal:
857	ext4_journal_stop(handle);
858out:
859	sb_end_intwrite(sb);
860	return err;
861}
862
863static int ext4_file_open(struct inode *inode, struct file *filp)
864{
865	int ret;
866
867	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
868		return -EIO;
869
870	ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt);
871	if (ret)
872		return ret;
873
874	ret = fscrypt_file_open(inode, filp);
875	if (ret)
876		return ret;
877
878	ret = fsverity_file_open(inode, filp);
879	if (ret)
880		return ret;
881
882	/*
883	 * Set up the jbd2_inode if we are opening the inode for
884	 * writing and the journal is present
885	 */
886	if (filp->f_mode & FMODE_WRITE) {
887		ret = ext4_inode_attach_jinode(inode);
888		if (ret < 0)
889			return ret;
890	}
891
892	filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
893	return dquot_file_open(inode, filp);
894}
895
896/*
897 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
898 * by calling generic_file_llseek_size() with the appropriate maxbytes
899 * value for each.
900 */
901loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
902{
903	struct inode *inode = file->f_mapping->host;
904	loff_t maxbytes;
905
906	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
907		maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
908	else
909		maxbytes = inode->i_sb->s_maxbytes;
910
911	switch (whence) {
912	default:
913		return generic_file_llseek_size(file, offset, whence,
914						maxbytes, i_size_read(inode));
915	case SEEK_HOLE:
916		inode_lock_shared(inode);
917		offset = iomap_seek_hole(inode, offset,
918					 &ext4_iomap_report_ops);
919		inode_unlock_shared(inode);
920		break;
921	case SEEK_DATA:
922		inode_lock_shared(inode);
923		offset = iomap_seek_data(inode, offset,
924					 &ext4_iomap_report_ops);
925		inode_unlock_shared(inode);
926		break;
927	}
928
929	if (offset < 0)
930		return offset;
931	return vfs_setpos(file, offset, maxbytes);
932}
933
934const struct file_operations ext4_file_operations = {
935	.llseek		= ext4_llseek,
936	.read_iter	= ext4_file_read_iter,
937	.write_iter	= ext4_file_write_iter,
938	.iopoll		= iocb_bio_iopoll,
939	.unlocked_ioctl = ext4_ioctl,
940#ifdef CONFIG_COMPAT
941	.compat_ioctl	= ext4_compat_ioctl,
942#endif
943	.mmap		= ext4_file_mmap,
944	.mmap_supported_flags = MAP_SYNC,
945	.open		= ext4_file_open,
946	.release	= ext4_release_file,
947	.fsync		= ext4_sync_file,
948	.get_unmapped_area = thp_get_unmapped_area,
949	.splice_read	= generic_file_splice_read,
950	.splice_write	= iter_file_splice_write,
951	.fallocate	= ext4_fallocate,
952};
953
954const struct inode_operations ext4_file_inode_operations = {
955	.setattr	= ext4_setattr,
956	.getattr	= ext4_file_getattr,
957	.listxattr	= ext4_listxattr,
958	.get_inode_acl	= ext4_get_acl,
959	.set_acl	= ext4_set_acl,
960	.fiemap		= ext4_fiemap,
961	.fileattr_get	= ext4_fileattr_get,
962	.fileattr_set	= ext4_fileattr_set,
963};
964