xref: /linux/fs/ext4/file.c (revision fde668244d1d8d490b5b9daf53fe4f92a6751773)
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 static bool ext4_dio_supported(struct inode *inode)
40 {
41 	if (IS_ENABLED(CONFIG_FS_ENCRYPTION) && IS_ENCRYPTED(inode))
42 		return false;
43 	if (fsverity_active(inode))
44 		return false;
45 	if (ext4_should_journal_data(inode))
46 		return false;
47 	if (ext4_has_inline_data(inode))
48 		return false;
49 	return true;
50 }
51 
52 static ssize_t ext4_dio_read_iter(struct kiocb *iocb, struct iov_iter *to)
53 {
54 	ssize_t ret;
55 	struct inode *inode = file_inode(iocb->ki_filp);
56 
57 	if (iocb->ki_flags & IOCB_NOWAIT) {
58 		if (!inode_trylock_shared(inode))
59 			return -EAGAIN;
60 	} else {
61 		inode_lock_shared(inode);
62 	}
63 
64 	if (!ext4_dio_supported(inode)) {
65 		inode_unlock_shared(inode);
66 		/*
67 		 * Fallback to buffered I/O if the operation being performed on
68 		 * the inode is not supported by direct I/O. The IOCB_DIRECT
69 		 * flag needs to be cleared here in order to ensure that the
70 		 * direct I/O path within generic_file_read_iter() is not
71 		 * taken.
72 		 */
73 		iocb->ki_flags &= ~IOCB_DIRECT;
74 		return generic_file_read_iter(iocb, to);
75 	}
76 
77 	ret = iomap_dio_rw(iocb, to, &ext4_iomap_ops, NULL,
78 			   is_sync_kiocb(iocb));
79 	inode_unlock_shared(inode);
80 
81 	file_accessed(iocb->ki_filp);
82 	return ret;
83 }
84 
85 #ifdef CONFIG_FS_DAX
86 static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
87 {
88 	struct inode *inode = file_inode(iocb->ki_filp);
89 	ssize_t ret;
90 
91 	if (iocb->ki_flags & IOCB_NOWAIT) {
92 		if (!inode_trylock_shared(inode))
93 			return -EAGAIN;
94 	} else {
95 		inode_lock_shared(inode);
96 	}
97 	/*
98 	 * Recheck under inode lock - at this point we are sure it cannot
99 	 * change anymore
100 	 */
101 	if (!IS_DAX(inode)) {
102 		inode_unlock_shared(inode);
103 		/* Fallback to buffered IO in case we cannot support DAX */
104 		return generic_file_read_iter(iocb, to);
105 	}
106 	ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
107 	inode_unlock_shared(inode);
108 
109 	file_accessed(iocb->ki_filp);
110 	return ret;
111 }
112 #endif
113 
114 static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
115 {
116 	struct inode *inode = file_inode(iocb->ki_filp);
117 
118 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
119 		return -EIO;
120 
121 	if (!iov_iter_count(to))
122 		return 0; /* skip atime */
123 
124 #ifdef CONFIG_FS_DAX
125 	if (IS_DAX(inode))
126 		return ext4_dax_read_iter(iocb, to);
127 #endif
128 	if (iocb->ki_flags & IOCB_DIRECT)
129 		return ext4_dio_read_iter(iocb, to);
130 
131 	return generic_file_read_iter(iocb, to);
132 }
133 
134 /*
135  * Called when an inode is released. Note that this is different
136  * from ext4_file_open: open gets called at every open, but release
137  * gets called only when /all/ the files are closed.
138  */
139 static int ext4_release_file(struct inode *inode, struct file *filp)
140 {
141 	if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
142 		ext4_alloc_da_blocks(inode);
143 		ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
144 	}
145 	/* if we are the last writer on the inode, drop the block reservation */
146 	if ((filp->f_mode & FMODE_WRITE) &&
147 			(atomic_read(&inode->i_writecount) == 1) &&
148 			!EXT4_I(inode)->i_reserved_data_blocks) {
149 		down_write(&EXT4_I(inode)->i_data_sem);
150 		ext4_discard_preallocations(inode, 0);
151 		up_write(&EXT4_I(inode)->i_data_sem);
152 	}
153 	if (is_dx(inode) && filp->private_data)
154 		ext4_htree_free_dir_info(filp->private_data);
155 
156 	return 0;
157 }
158 
159 /*
160  * This tests whether the IO in question is block-aligned or not.
161  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
162  * are converted to written only after the IO is complete.  Until they are
163  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
164  * it needs to zero out portions of the start and/or end block.  If 2 AIO
165  * threads are at work on the same unwritten block, they must be synchronized
166  * or one thread will zero the other's data, causing corruption.
167  */
168 static bool
169 ext4_unaligned_io(struct inode *inode, struct iov_iter *from, loff_t pos)
170 {
171 	struct super_block *sb = inode->i_sb;
172 	unsigned long blockmask = sb->s_blocksize - 1;
173 
174 	if ((pos | iov_iter_alignment(from)) & blockmask)
175 		return true;
176 
177 	return false;
178 }
179 
180 static bool
181 ext4_extending_io(struct inode *inode, loff_t offset, size_t len)
182 {
183 	if (offset + len > i_size_read(inode) ||
184 	    offset + len > EXT4_I(inode)->i_disksize)
185 		return true;
186 	return false;
187 }
188 
189 /* Is IO overwriting allocated and initialized blocks? */
190 static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
191 {
192 	struct ext4_map_blocks map;
193 	unsigned int blkbits = inode->i_blkbits;
194 	int err, blklen;
195 
196 	if (pos + len > i_size_read(inode))
197 		return false;
198 
199 	map.m_lblk = pos >> blkbits;
200 	map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
201 	blklen = map.m_len;
202 
203 	err = ext4_map_blocks(NULL, inode, &map, 0);
204 	/*
205 	 * 'err==len' means that all of the blocks have been preallocated,
206 	 * regardless of whether they have been initialized or not. To exclude
207 	 * unwritten extents, we need to check m_flags.
208 	 */
209 	return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
210 }
211 
212 static ssize_t ext4_generic_write_checks(struct kiocb *iocb,
213 					 struct iov_iter *from)
214 {
215 	struct inode *inode = file_inode(iocb->ki_filp);
216 	ssize_t ret;
217 
218 	if (unlikely(IS_IMMUTABLE(inode)))
219 		return -EPERM;
220 
221 	ret = generic_write_checks(iocb, from);
222 	if (ret <= 0)
223 		return ret;
224 
225 	/*
226 	 * If we have encountered a bitmap-format file, the size limit
227 	 * is smaller than s_maxbytes, which is for extent-mapped files.
228 	 */
229 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
230 		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
231 
232 		if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
233 			return -EFBIG;
234 		iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
235 	}
236 
237 	return iov_iter_count(from);
238 }
239 
240 static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
241 {
242 	ssize_t ret, count;
243 
244 	count = ext4_generic_write_checks(iocb, from);
245 	if (count <= 0)
246 		return count;
247 
248 	ret = file_modified(iocb->ki_filp);
249 	if (ret)
250 		return ret;
251 	return count;
252 }
253 
254 static ssize_t ext4_buffered_write_iter(struct kiocb *iocb,
255 					struct iov_iter *from)
256 {
257 	ssize_t ret;
258 	struct inode *inode = file_inode(iocb->ki_filp);
259 
260 	if (iocb->ki_flags & IOCB_NOWAIT)
261 		return -EOPNOTSUPP;
262 
263 	ext4_fc_start_update(inode);
264 	inode_lock(inode);
265 	ret = ext4_write_checks(iocb, from);
266 	if (ret <= 0)
267 		goto out;
268 
269 	current->backing_dev_info = inode_to_bdi(inode);
270 	ret = generic_perform_write(iocb->ki_filp, from, iocb->ki_pos);
271 	current->backing_dev_info = NULL;
272 
273 out:
274 	inode_unlock(inode);
275 	ext4_fc_stop_update(inode);
276 	if (likely(ret > 0)) {
277 		iocb->ki_pos += ret;
278 		ret = generic_write_sync(iocb, ret);
279 	}
280 
281 	return ret;
282 }
283 
284 static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
285 					   ssize_t written, size_t count)
286 {
287 	handle_t *handle;
288 	bool truncate = false;
289 	u8 blkbits = inode->i_blkbits;
290 	ext4_lblk_t written_blk, end_blk;
291 	int ret;
292 
293 	/*
294 	 * Note that EXT4_I(inode)->i_disksize can get extended up to
295 	 * inode->i_size while the I/O was running due to writeback of delalloc
296 	 * blocks. But, the code in ext4_iomap_alloc() is careful to use
297 	 * zeroed/unwritten extents if this is possible; thus we won't leave
298 	 * uninitialized blocks in a file even if we didn't succeed in writing
299 	 * as much as we intended.
300 	 */
301 	WARN_ON_ONCE(i_size_read(inode) < EXT4_I(inode)->i_disksize);
302 	if (offset + count <= EXT4_I(inode)->i_disksize) {
303 		/*
304 		 * We need to ensure that the inode is removed from the orphan
305 		 * list if it has been added prematurely, due to writeback of
306 		 * delalloc blocks.
307 		 */
308 		if (!list_empty(&EXT4_I(inode)->i_orphan) && inode->i_nlink) {
309 			handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
310 
311 			if (IS_ERR(handle)) {
312 				ext4_orphan_del(NULL, inode);
313 				return PTR_ERR(handle);
314 			}
315 
316 			ext4_orphan_del(handle, inode);
317 			ext4_journal_stop(handle);
318 		}
319 
320 		return written;
321 	}
322 
323 	if (written < 0)
324 		goto truncate;
325 
326 	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
327 	if (IS_ERR(handle)) {
328 		written = PTR_ERR(handle);
329 		goto truncate;
330 	}
331 
332 	if (ext4_update_inode_size(inode, offset + written)) {
333 		ret = ext4_mark_inode_dirty(handle, inode);
334 		if (unlikely(ret)) {
335 			written = ret;
336 			ext4_journal_stop(handle);
337 			goto truncate;
338 		}
339 	}
340 
341 	/*
342 	 * We may need to truncate allocated but not written blocks beyond EOF.
343 	 */
344 	written_blk = ALIGN(offset + written, 1 << blkbits);
345 	end_blk = ALIGN(offset + count, 1 << blkbits);
346 	if (written_blk < end_blk && ext4_can_truncate(inode))
347 		truncate = true;
348 
349 	/*
350 	 * Remove the inode from the orphan list if it has been extended and
351 	 * everything went OK.
352 	 */
353 	if (!truncate && inode->i_nlink)
354 		ext4_orphan_del(handle, inode);
355 	ext4_journal_stop(handle);
356 
357 	if (truncate) {
358 truncate:
359 		ext4_truncate_failed_write(inode);
360 		/*
361 		 * If the truncate operation failed early, then the inode may
362 		 * still be on the orphan list. In that case, we need to try
363 		 * remove the inode from the in-memory linked list.
364 		 */
365 		if (inode->i_nlink)
366 			ext4_orphan_del(NULL, inode);
367 	}
368 
369 	return written;
370 }
371 
372 static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
373 				 int error, unsigned int flags)
374 {
375 	loff_t offset = iocb->ki_pos;
376 	struct inode *inode = file_inode(iocb->ki_filp);
377 
378 	if (error)
379 		return error;
380 
381 	if (size && flags & IOMAP_DIO_UNWRITTEN)
382 		return ext4_convert_unwritten_extents(NULL, inode,
383 						      offset, size);
384 
385 	return 0;
386 }
387 
388 static const struct iomap_dio_ops ext4_dio_write_ops = {
389 	.end_io = ext4_dio_write_end_io,
390 };
391 
392 /*
393  * The intention here is to start with shared lock acquired then see if any
394  * condition requires an exclusive inode lock. If yes, then we restart the
395  * whole operation by releasing the shared lock and acquiring exclusive lock.
396  *
397  * - For unaligned_io we never take shared lock as it may cause data corruption
398  *   when two unaligned IO tries to modify the same block e.g. while zeroing.
399  *
400  * - For extending writes case we don't take the shared lock, since it requires
401  *   updating inode i_disksize and/or orphan handling with exclusive lock.
402  *
403  * - shared locking will only be true mostly with overwrites. Otherwise we will
404  *   switch to exclusive i_rwsem lock.
405  */
406 static ssize_t ext4_dio_write_checks(struct kiocb *iocb, struct iov_iter *from,
407 				     bool *ilock_shared, bool *extend)
408 {
409 	struct file *file = iocb->ki_filp;
410 	struct inode *inode = file_inode(file);
411 	loff_t offset;
412 	size_t count;
413 	ssize_t ret;
414 
415 restart:
416 	ret = ext4_generic_write_checks(iocb, from);
417 	if (ret <= 0)
418 		goto out;
419 
420 	offset = iocb->ki_pos;
421 	count = ret;
422 	if (ext4_extending_io(inode, offset, count))
423 		*extend = true;
424 	/*
425 	 * Determine whether the IO operation will overwrite allocated
426 	 * and initialized blocks.
427 	 * We need exclusive i_rwsem for changing security info
428 	 * in file_modified().
429 	 */
430 	if (*ilock_shared && (!IS_NOSEC(inode) || *extend ||
431 	     !ext4_overwrite_io(inode, offset, count))) {
432 		if (iocb->ki_flags & IOCB_NOWAIT) {
433 			ret = -EAGAIN;
434 			goto out;
435 		}
436 		inode_unlock_shared(inode);
437 		*ilock_shared = false;
438 		inode_lock(inode);
439 		goto restart;
440 	}
441 
442 	ret = file_modified(file);
443 	if (ret < 0)
444 		goto out;
445 
446 	return count;
447 out:
448 	if (*ilock_shared)
449 		inode_unlock_shared(inode);
450 	else
451 		inode_unlock(inode);
452 	return ret;
453 }
454 
455 static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from)
456 {
457 	ssize_t ret;
458 	handle_t *handle;
459 	struct inode *inode = file_inode(iocb->ki_filp);
460 	loff_t offset = iocb->ki_pos;
461 	size_t count = iov_iter_count(from);
462 	const struct iomap_ops *iomap_ops = &ext4_iomap_ops;
463 	bool extend = false, unaligned_io = false;
464 	bool ilock_shared = true;
465 
466 	/*
467 	 * We initially start with shared inode lock unless it is
468 	 * unaligned IO which needs exclusive lock anyways.
469 	 */
470 	if (ext4_unaligned_io(inode, from, offset)) {
471 		unaligned_io = true;
472 		ilock_shared = false;
473 	}
474 	/*
475 	 * Quick check here without any i_rwsem lock to see if it is extending
476 	 * IO. A more reliable check is done in ext4_dio_write_checks() with
477 	 * proper locking in place.
478 	 */
479 	if (offset + count > i_size_read(inode))
480 		ilock_shared = false;
481 
482 	if (iocb->ki_flags & IOCB_NOWAIT) {
483 		if (ilock_shared) {
484 			if (!inode_trylock_shared(inode))
485 				return -EAGAIN;
486 		} else {
487 			if (!inode_trylock(inode))
488 				return -EAGAIN;
489 		}
490 	} else {
491 		if (ilock_shared)
492 			inode_lock_shared(inode);
493 		else
494 			inode_lock(inode);
495 	}
496 
497 	/* Fallback to buffered I/O if the inode does not support direct I/O. */
498 	if (!ext4_dio_supported(inode)) {
499 		if (ilock_shared)
500 			inode_unlock_shared(inode);
501 		else
502 			inode_unlock(inode);
503 		return ext4_buffered_write_iter(iocb, from);
504 	}
505 
506 	ret = ext4_dio_write_checks(iocb, from, &ilock_shared, &extend);
507 	if (ret <= 0)
508 		return ret;
509 
510 	/* if we're going to block and IOCB_NOWAIT is set, return -EAGAIN */
511 	if ((iocb->ki_flags & IOCB_NOWAIT) && (unaligned_io || extend)) {
512 		ret = -EAGAIN;
513 		goto out;
514 	}
515 
516 	offset = iocb->ki_pos;
517 	count = ret;
518 
519 	/*
520 	 * Unaligned direct IO must be serialized among each other as zeroing
521 	 * of partial blocks of two competing unaligned IOs can result in data
522 	 * corruption.
523 	 *
524 	 * So we make sure we don't allow any unaligned IO in flight.
525 	 * For IOs where we need not wait (like unaligned non-AIO DIO),
526 	 * below inode_dio_wait() may anyway become a no-op, since we start
527 	 * with exclusive lock.
528 	 */
529 	if (unaligned_io)
530 		inode_dio_wait(inode);
531 
532 	if (extend) {
533 		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
534 		if (IS_ERR(handle)) {
535 			ret = PTR_ERR(handle);
536 			goto out;
537 		}
538 
539 		ext4_fc_start_update(inode);
540 		ret = ext4_orphan_add(handle, inode);
541 		ext4_fc_stop_update(inode);
542 		if (ret) {
543 			ext4_journal_stop(handle);
544 			goto out;
545 		}
546 
547 		ext4_journal_stop(handle);
548 	}
549 
550 	if (ilock_shared)
551 		iomap_ops = &ext4_iomap_overwrite_ops;
552 	ret = iomap_dio_rw(iocb, from, iomap_ops, &ext4_dio_write_ops,
553 			   is_sync_kiocb(iocb) || unaligned_io || extend);
554 	if (ret == -ENOTBLK)
555 		ret = 0;
556 
557 	if (extend)
558 		ret = ext4_handle_inode_extension(inode, offset, ret, count);
559 
560 out:
561 	if (ilock_shared)
562 		inode_unlock_shared(inode);
563 	else
564 		inode_unlock(inode);
565 
566 	if (ret >= 0 && iov_iter_count(from)) {
567 		ssize_t err;
568 		loff_t endbyte;
569 
570 		offset = iocb->ki_pos;
571 		err = ext4_buffered_write_iter(iocb, from);
572 		if (err < 0)
573 			return err;
574 
575 		/*
576 		 * We need to ensure that the pages within the page cache for
577 		 * the range covered by this I/O are written to disk and
578 		 * invalidated. This is in attempt to preserve the expected
579 		 * direct I/O semantics in the case we fallback to buffered I/O
580 		 * to complete off the I/O request.
581 		 */
582 		ret += err;
583 		endbyte = offset + err - 1;
584 		err = filemap_write_and_wait_range(iocb->ki_filp->f_mapping,
585 						   offset, endbyte);
586 		if (!err)
587 			invalidate_mapping_pages(iocb->ki_filp->f_mapping,
588 						 offset >> PAGE_SHIFT,
589 						 endbyte >> PAGE_SHIFT);
590 	}
591 
592 	return ret;
593 }
594 
595 #ifdef CONFIG_FS_DAX
596 static ssize_t
597 ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
598 {
599 	ssize_t ret;
600 	size_t count;
601 	loff_t offset;
602 	handle_t *handle;
603 	bool extend = false;
604 	struct inode *inode = file_inode(iocb->ki_filp);
605 
606 	if (iocb->ki_flags & IOCB_NOWAIT) {
607 		if (!inode_trylock(inode))
608 			return -EAGAIN;
609 	} else {
610 		inode_lock(inode);
611 	}
612 
613 	ret = ext4_write_checks(iocb, from);
614 	if (ret <= 0)
615 		goto out;
616 
617 	offset = iocb->ki_pos;
618 	count = iov_iter_count(from);
619 
620 	if (offset + count > EXT4_I(inode)->i_disksize) {
621 		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
622 		if (IS_ERR(handle)) {
623 			ret = PTR_ERR(handle);
624 			goto out;
625 		}
626 
627 		ret = ext4_orphan_add(handle, inode);
628 		if (ret) {
629 			ext4_journal_stop(handle);
630 			goto out;
631 		}
632 
633 		extend = true;
634 		ext4_journal_stop(handle);
635 	}
636 
637 	ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
638 
639 	if (extend)
640 		ret = ext4_handle_inode_extension(inode, offset, ret, count);
641 out:
642 	inode_unlock(inode);
643 	if (ret > 0)
644 		ret = generic_write_sync(iocb, ret);
645 	return ret;
646 }
647 #endif
648 
649 static ssize_t
650 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
651 {
652 	struct inode *inode = file_inode(iocb->ki_filp);
653 
654 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
655 		return -EIO;
656 
657 #ifdef CONFIG_FS_DAX
658 	if (IS_DAX(inode))
659 		return ext4_dax_write_iter(iocb, from);
660 #endif
661 	if (iocb->ki_flags & IOCB_DIRECT)
662 		return ext4_dio_write_iter(iocb, from);
663 	else
664 		return ext4_buffered_write_iter(iocb, from);
665 }
666 
667 #ifdef CONFIG_FS_DAX
668 static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf,
669 		enum page_entry_size pe_size)
670 {
671 	int error = 0;
672 	vm_fault_t result;
673 	int retries = 0;
674 	handle_t *handle = NULL;
675 	struct inode *inode = file_inode(vmf->vma->vm_file);
676 	struct super_block *sb = inode->i_sb;
677 
678 	/*
679 	 * We have to distinguish real writes from writes which will result in a
680 	 * COW page; COW writes should *not* poke the journal (the file will not
681 	 * be changed). Doing so would cause unintended failures when mounted
682 	 * read-only.
683 	 *
684 	 * We check for VM_SHARED rather than vmf->cow_page since the latter is
685 	 * unset for pe_size != PE_SIZE_PTE (i.e. only in do_cow_fault); for
686 	 * other sizes, dax_iomap_fault will handle splitting / fallback so that
687 	 * we eventually come back with a COW page.
688 	 */
689 	bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
690 		(vmf->vma->vm_flags & VM_SHARED);
691 	pfn_t pfn;
692 
693 	if (write) {
694 		sb_start_pagefault(sb);
695 		file_update_time(vmf->vma->vm_file);
696 		down_read(&EXT4_I(inode)->i_mmap_sem);
697 retry:
698 		handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
699 					       EXT4_DATA_TRANS_BLOCKS(sb));
700 		if (IS_ERR(handle)) {
701 			up_read(&EXT4_I(inode)->i_mmap_sem);
702 			sb_end_pagefault(sb);
703 			return VM_FAULT_SIGBUS;
704 		}
705 	} else {
706 		down_read(&EXT4_I(inode)->i_mmap_sem);
707 	}
708 	result = dax_iomap_fault(vmf, pe_size, &pfn, &error, &ext4_iomap_ops);
709 	if (write) {
710 		ext4_journal_stop(handle);
711 
712 		if ((result & VM_FAULT_ERROR) && error == -ENOSPC &&
713 		    ext4_should_retry_alloc(sb, &retries))
714 			goto retry;
715 		/* Handling synchronous page fault? */
716 		if (result & VM_FAULT_NEEDDSYNC)
717 			result = dax_finish_sync_fault(vmf, pe_size, pfn);
718 		up_read(&EXT4_I(inode)->i_mmap_sem);
719 		sb_end_pagefault(sb);
720 	} else {
721 		up_read(&EXT4_I(inode)->i_mmap_sem);
722 	}
723 
724 	return result;
725 }
726 
727 static vm_fault_t ext4_dax_fault(struct vm_fault *vmf)
728 {
729 	return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
730 }
731 
732 static const struct vm_operations_struct ext4_dax_vm_ops = {
733 	.fault		= ext4_dax_fault,
734 	.huge_fault	= ext4_dax_huge_fault,
735 	.page_mkwrite	= ext4_dax_fault,
736 	.pfn_mkwrite	= ext4_dax_fault,
737 };
738 #else
739 #define ext4_dax_vm_ops	ext4_file_vm_ops
740 #endif
741 
742 static const struct vm_operations_struct ext4_file_vm_ops = {
743 	.fault		= ext4_filemap_fault,
744 	.map_pages	= filemap_map_pages,
745 	.page_mkwrite   = ext4_page_mkwrite,
746 };
747 
748 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
749 {
750 	struct inode *inode = file->f_mapping->host;
751 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
752 	struct dax_device *dax_dev = sbi->s_daxdev;
753 
754 	if (unlikely(ext4_forced_shutdown(sbi)))
755 		return -EIO;
756 
757 	/*
758 	 * We don't support synchronous mappings for non-DAX files and
759 	 * for DAX files if underneath dax_device is not synchronous.
760 	 */
761 	if (!daxdev_mapping_supported(vma, dax_dev))
762 		return -EOPNOTSUPP;
763 
764 	file_accessed(file);
765 	if (IS_DAX(file_inode(file))) {
766 		vma->vm_ops = &ext4_dax_vm_ops;
767 		vma->vm_flags |= VM_HUGEPAGE;
768 	} else {
769 		vma->vm_ops = &ext4_file_vm_ops;
770 	}
771 	return 0;
772 }
773 
774 static int ext4_sample_last_mounted(struct super_block *sb,
775 				    struct vfsmount *mnt)
776 {
777 	struct ext4_sb_info *sbi = EXT4_SB(sb);
778 	struct path path;
779 	char buf[64], *cp;
780 	handle_t *handle;
781 	int err;
782 
783 	if (likely(ext4_test_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED)))
784 		return 0;
785 
786 	if (sb_rdonly(sb) || !sb_start_intwrite_trylock(sb))
787 		return 0;
788 
789 	ext4_set_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED);
790 	/*
791 	 * Sample where the filesystem has been mounted and
792 	 * store it in the superblock for sysadmin convenience
793 	 * when trying to sort through large numbers of block
794 	 * devices or filesystem images.
795 	 */
796 	memset(buf, 0, sizeof(buf));
797 	path.mnt = mnt;
798 	path.dentry = mnt->mnt_root;
799 	cp = d_path(&path, buf, sizeof(buf));
800 	err = 0;
801 	if (IS_ERR(cp))
802 		goto out;
803 
804 	handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
805 	err = PTR_ERR(handle);
806 	if (IS_ERR(handle))
807 		goto out;
808 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
809 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
810 	if (err)
811 		goto out_journal;
812 	strlcpy(sbi->s_es->s_last_mounted, cp,
813 		sizeof(sbi->s_es->s_last_mounted));
814 	ext4_handle_dirty_super(handle, sb);
815 out_journal:
816 	ext4_journal_stop(handle);
817 out:
818 	sb_end_intwrite(sb);
819 	return err;
820 }
821 
822 static int ext4_file_open(struct inode *inode, struct file *filp)
823 {
824 	int ret;
825 
826 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
827 		return -EIO;
828 
829 	ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt);
830 	if (ret)
831 		return ret;
832 
833 	ret = fscrypt_file_open(inode, filp);
834 	if (ret)
835 		return ret;
836 
837 	ret = fsverity_file_open(inode, filp);
838 	if (ret)
839 		return ret;
840 
841 	/*
842 	 * Set up the jbd2_inode if we are opening the inode for
843 	 * writing and the journal is present
844 	 */
845 	if (filp->f_mode & FMODE_WRITE) {
846 		ret = ext4_inode_attach_jinode(inode);
847 		if (ret < 0)
848 			return ret;
849 	}
850 
851 	filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
852 	return dquot_file_open(inode, filp);
853 }
854 
855 /*
856  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
857  * by calling generic_file_llseek_size() with the appropriate maxbytes
858  * value for each.
859  */
860 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
861 {
862 	struct inode *inode = file->f_mapping->host;
863 	loff_t maxbytes;
864 
865 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
866 		maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
867 	else
868 		maxbytes = inode->i_sb->s_maxbytes;
869 
870 	switch (whence) {
871 	default:
872 		return generic_file_llseek_size(file, offset, whence,
873 						maxbytes, i_size_read(inode));
874 	case SEEK_HOLE:
875 		inode_lock_shared(inode);
876 		offset = iomap_seek_hole(inode, offset,
877 					 &ext4_iomap_report_ops);
878 		inode_unlock_shared(inode);
879 		break;
880 	case SEEK_DATA:
881 		inode_lock_shared(inode);
882 		offset = iomap_seek_data(inode, offset,
883 					 &ext4_iomap_report_ops);
884 		inode_unlock_shared(inode);
885 		break;
886 	}
887 
888 	if (offset < 0)
889 		return offset;
890 	return vfs_setpos(file, offset, maxbytes);
891 }
892 
893 const struct file_operations ext4_file_operations = {
894 	.llseek		= ext4_llseek,
895 	.read_iter	= ext4_file_read_iter,
896 	.write_iter	= ext4_file_write_iter,
897 	.iopoll		= iomap_dio_iopoll,
898 	.unlocked_ioctl = ext4_ioctl,
899 #ifdef CONFIG_COMPAT
900 	.compat_ioctl	= ext4_compat_ioctl,
901 #endif
902 	.mmap		= ext4_file_mmap,
903 	.mmap_supported_flags = MAP_SYNC,
904 	.open		= ext4_file_open,
905 	.release	= ext4_release_file,
906 	.fsync		= ext4_sync_file,
907 	.get_unmapped_area = thp_get_unmapped_area,
908 	.splice_read	= generic_file_splice_read,
909 	.splice_write	= iter_file_splice_write,
910 	.fallocate	= ext4_fallocate,
911 };
912 
913 const struct inode_operations ext4_file_inode_operations = {
914 	.setattr	= ext4_setattr,
915 	.getattr	= ext4_file_getattr,
916 	.listxattr	= ext4_listxattr,
917 	.get_acl	= ext4_get_acl,
918 	.set_acl	= ext4_set_acl,
919 	.fiemap		= ext4_fiemap,
920 };
921 
922