xref: /linux/fs/ext4/file.c (revision 5ea5880764cbb164afb17a62e76ca75dc371409d)
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/filelock.h>
29 #include <linux/quotaops.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  */
55 static 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 
69 static 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
102 static 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 
130 static 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 
150 static 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  */
166 static 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  */
195 static bool
196 ext4_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 
207 static bool
208 ext4_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? */
217 static 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 
243 static 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 
271 static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
272 {
273 	struct inode *inode = file_inode(iocb->ki_filp);
274 	loff_t old_size = i_size_read(inode);
275 	ssize_t ret, count;
276 
277 	count = ext4_generic_write_checks(iocb, from);
278 	if (count <= 0)
279 		return count;
280 
281 	ret = file_modified(iocb->ki_filp);
282 	if (ret)
283 		return ret;
284 
285 	/*
286 	 * If the position is beyond the EOF, it is necessary to zero out the
287 	 * partial block that beyond the existing EOF, as it may contains
288 	 * stale data written through mmap.
289 	 */
290 	if (iocb->ki_pos > old_size && !ext4_verity_in_progress(inode)) {
291 		if (iocb->ki_flags & IOCB_NOWAIT)
292 			return -EAGAIN;
293 
294 		ret = ext4_block_zero_eof(inode, old_size, iocb->ki_pos);
295 		if (ret)
296 			return ret;
297 	}
298 
299 	return count;
300 }
301 
302 static ssize_t ext4_buffered_write_iter(struct kiocb *iocb,
303 					struct iov_iter *from)
304 {
305 	ssize_t ret;
306 	struct inode *inode = file_inode(iocb->ki_filp);
307 
308 	if (iocb->ki_flags & IOCB_NOWAIT)
309 		return -EOPNOTSUPP;
310 
311 	inode_lock(inode);
312 	ret = ext4_write_checks(iocb, from);
313 	if (ret <= 0)
314 		goto out;
315 
316 	ret = generic_perform_write(iocb, from);
317 
318 out:
319 	inode_unlock(inode);
320 	if (unlikely(ret <= 0))
321 		return ret;
322 	return generic_write_sync(iocb, ret);
323 }
324 
325 static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
326 					   ssize_t written, ssize_t count)
327 {
328 	handle_t *handle;
329 
330 	lockdep_assert_held_write(&inode->i_rwsem);
331 	handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
332 	if (IS_ERR(handle))
333 		return PTR_ERR(handle);
334 
335 	if (ext4_update_inode_size(inode, offset + written)) {
336 		int ret = ext4_mark_inode_dirty(handle, inode);
337 		if (unlikely(ret)) {
338 			ext4_journal_stop(handle);
339 			return ret;
340 		}
341 	}
342 
343 	if ((written == count) && inode->i_nlink)
344 		ext4_orphan_del(handle, inode);
345 	ext4_journal_stop(handle);
346 
347 	return written;
348 }
349 
350 /*
351  * Clean up the inode after DIO or DAX extending write has completed and the
352  * inode size has been updated using ext4_handle_inode_extension().
353  */
354 static void ext4_inode_extension_cleanup(struct inode *inode, bool need_trunc)
355 {
356 	lockdep_assert_held_write(&inode->i_rwsem);
357 	if (need_trunc) {
358 		ext4_truncate_failed_write(inode);
359 		/*
360 		 * If the truncate operation failed early, then the inode may
361 		 * still be on the orphan list. In that case, we need to try
362 		 * remove the inode from the in-memory linked list.
363 		 */
364 		if (inode->i_nlink)
365 			ext4_orphan_del(NULL, inode);
366 		return;
367 	}
368 	/*
369 	 * If i_disksize got extended either due to writeback of delalloc
370 	 * blocks or extending truncate while the DIO was running we could fail
371 	 * to cleanup the orphan list in ext4_handle_inode_extension(). Do it
372 	 * now.
373 	 */
374 	if (ext4_inode_orphan_tracked(inode) && inode->i_nlink) {
375 		handle_t *handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
376 
377 		if (IS_ERR(handle)) {
378 			/*
379 			 * The write has successfully completed. Not much to
380 			 * do with the error here so just cleanup the orphan
381 			 * list and hope for the best.
382 			 */
383 			ext4_orphan_del(NULL, inode);
384 			return;
385 		}
386 		ext4_orphan_del(handle, inode);
387 		ext4_journal_stop(handle);
388 	}
389 }
390 
391 static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
392 				 int error, unsigned int flags)
393 {
394 	loff_t pos = iocb->ki_pos;
395 	struct inode *inode = file_inode(iocb->ki_filp);
396 
397 
398 	if (!error && size && (flags & IOMAP_DIO_UNWRITTEN) &&
399 			(iocb->ki_flags & IOCB_ATOMIC))
400 		error = ext4_convert_unwritten_extents_atomic(NULL, inode, pos,
401 							      size);
402 	else if (!error && size && flags & IOMAP_DIO_UNWRITTEN)
403 		error = ext4_convert_unwritten_extents(NULL, inode, pos, size);
404 	if (error)
405 		return error;
406 	/*
407 	 * Note that EXT4_I(inode)->i_disksize can get extended up to
408 	 * inode->i_size while the I/O was running due to writeback of delalloc
409 	 * blocks. But the code in ext4_iomap_alloc() is careful to use
410 	 * zeroed/unwritten extents if this is possible; thus we won't leave
411 	 * uninitialized blocks in a file even if we didn't succeed in writing
412 	 * as much as we intended. Also we can race with truncate or write
413 	 * expanding the file so we have to be a bit careful here.
414 	 */
415 	if (pos + size <= READ_ONCE(EXT4_I(inode)->i_disksize) &&
416 	    pos + size <= i_size_read(inode))
417 		return 0;
418 	error = ext4_handle_inode_extension(inode, pos, size, size);
419 	return error < 0 ? error : 0;
420 }
421 
422 static const struct iomap_dio_ops ext4_dio_write_ops = {
423 	.end_io = ext4_dio_write_end_io,
424 };
425 
426 /*
427  * The intention here is to start with shared lock acquired then see if any
428  * condition requires an exclusive inode lock. If yes, then we restart the
429  * whole operation by releasing the shared lock and acquiring exclusive lock.
430  *
431  * - For unaligned_io we never take shared lock as it may cause data corruption
432  *   when two unaligned IO tries to modify the same block e.g. while zeroing.
433  *
434  * - For extending writes case we don't take the shared lock, since it requires
435  *   updating inode i_disksize and/or orphan handling with exclusive lock.
436  *
437  * - shared locking will only be true mostly with overwrites, including
438  *   initialized blocks and unwritten blocks.
439  *
440  * - Otherwise we will switch to exclusive i_rwsem lock.
441  */
442 static ssize_t ext4_dio_write_checks(struct kiocb *iocb, struct iov_iter *from,
443 				     bool *ilock_shared, bool *extend,
444 				     int *dio_flags)
445 {
446 	struct file *file = iocb->ki_filp;
447 	struct inode *inode = file_inode(file);
448 	loff_t offset;
449 	size_t count;
450 	ssize_t ret;
451 	bool overwrite, unaligned_io, unwritten;
452 
453 restart:
454 	ret = ext4_generic_write_checks(iocb, from);
455 	if (ret <= 0)
456 		goto out;
457 
458 	offset = iocb->ki_pos;
459 	count = ret;
460 
461 	unaligned_io = ext4_unaligned_io(inode, from, offset);
462 	*extend = ext4_extending_io(inode, offset, count);
463 	overwrite = ext4_overwrite_io(inode, offset, count, &unwritten);
464 
465 	/*
466 	 * Determine whether we need to upgrade to an exclusive lock. This is
467 	 * required to change security info in file_modified(), for extending
468 	 * I/O, any form of non-overwrite I/O, and unaligned I/O to unwritten
469 	 * extents (as partial block zeroing may be required).
470 	 *
471 	 * Note that unaligned writes are allowed under shared lock so long as
472 	 * they are pure overwrites. Otherwise, concurrent unaligned writes risk
473 	 * data corruption due to partial block zeroing in the dio layer, and so
474 	 * the I/O must occur exclusively.
475 	 */
476 	if (*ilock_shared &&
477 	    ((!IS_NOSEC(inode) || *extend || !overwrite ||
478 	     (unaligned_io && unwritten)))) {
479 		if (iocb->ki_flags & IOCB_NOWAIT) {
480 			ret = -EAGAIN;
481 			goto out;
482 		}
483 		inode_unlock_shared(inode);
484 		*ilock_shared = false;
485 		inode_lock(inode);
486 		goto restart;
487 	}
488 
489 	/*
490 	 * Now that locking is settled, determine dio flags and exclusivity
491 	 * requirements. We don't use DIO_OVERWRITE_ONLY because we enforce
492 	 * behavior already. The inode lock is already held exclusive if the
493 	 * write is non-overwrite or extending, so drain all outstanding dio and
494 	 * set the force wait dio flag.
495 	 */
496 	if (!*ilock_shared && (unaligned_io || *extend)) {
497 		if (iocb->ki_flags & IOCB_NOWAIT) {
498 			ret = -EAGAIN;
499 			goto out;
500 		}
501 		if (unaligned_io && (!overwrite || unwritten))
502 			inode_dio_wait(inode);
503 		*dio_flags = IOMAP_DIO_FORCE_WAIT;
504 	}
505 
506 	ret = file_modified(file);
507 	if (ret < 0)
508 		goto out;
509 
510 	return count;
511 out:
512 	if (*ilock_shared)
513 		inode_unlock_shared(inode);
514 	else
515 		inode_unlock(inode);
516 	return ret;
517 }
518 
519 static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from)
520 {
521 	ssize_t ret;
522 	handle_t *handle;
523 	struct inode *inode = file_inode(iocb->ki_filp);
524 	loff_t offset = iocb->ki_pos;
525 	size_t count = iov_iter_count(from);
526 	bool extend = false;
527 	bool ilock_shared = true;
528 	int dio_flags = 0;
529 
530 	/*
531 	 * Quick check here without any i_rwsem lock to see if it is extending
532 	 * IO. A more reliable check is done in ext4_dio_write_checks() with
533 	 * proper locking in place.
534 	 */
535 	if (offset + count > i_size_read(inode))
536 		ilock_shared = false;
537 
538 	if (iocb->ki_flags & IOCB_NOWAIT) {
539 		if (ilock_shared) {
540 			if (!inode_trylock_shared(inode))
541 				return -EAGAIN;
542 		} else {
543 			if (!inode_trylock(inode))
544 				return -EAGAIN;
545 		}
546 	} else {
547 		if (ilock_shared)
548 			inode_lock_shared(inode);
549 		else
550 			inode_lock(inode);
551 	}
552 
553 	/* Fallback to buffered I/O if the inode does not support direct I/O. */
554 	if (!ext4_should_use_dio(iocb, from)) {
555 		if (ilock_shared)
556 			inode_unlock_shared(inode);
557 		else
558 			inode_unlock(inode);
559 		return ext4_buffered_write_iter(iocb, from);
560 	}
561 
562 	/*
563 	 * Prevent inline data from being created since we are going to allocate
564 	 * blocks for DIO. We know the inode does not currently have inline data
565 	 * because ext4_should_use_dio() checked for it, but we have to clear
566 	 * the state flag before the write checks because a lock cycle could
567 	 * introduce races with other writers.
568 	 */
569 	ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
570 
571 	ret = ext4_dio_write_checks(iocb, from, &ilock_shared, &extend,
572 				    &dio_flags);
573 	if (ret <= 0)
574 		return ret;
575 
576 	offset = iocb->ki_pos;
577 	count = ret;
578 
579 	if (extend) {
580 		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
581 		if (IS_ERR(handle)) {
582 			ret = PTR_ERR(handle);
583 			goto out;
584 		}
585 
586 		ret = ext4_orphan_add(handle, inode);
587 		ext4_journal_stop(handle);
588 		if (ret)
589 			goto out;
590 	}
591 
592 	ret = iomap_dio_rw(iocb, from, &ext4_iomap_ops, &ext4_dio_write_ops,
593 			   dio_flags, NULL, 0);
594 	if (ret == -ENOTBLK)
595 		ret = 0;
596 	if (extend) {
597 		/*
598 		 * We always perform extending DIO write synchronously so by
599 		 * now the IO is completed and ext4_handle_inode_extension()
600 		 * was called. Cleanup the inode in case of error or race with
601 		 * writeback of delalloc blocks.
602 		 */
603 		WARN_ON_ONCE(ret == -EIOCBQUEUED);
604 		ext4_inode_extension_cleanup(inode, ret < 0);
605 	}
606 
607 out:
608 	if (ilock_shared)
609 		inode_unlock_shared(inode);
610 	else
611 		inode_unlock(inode);
612 
613 	if (ret >= 0 && iov_iter_count(from)) {
614 		ssize_t err;
615 		loff_t endbyte;
616 
617 		/*
618 		 * There is no support for atomic writes on buffered-io yet,
619 		 * we should never fallback to buffered-io for DIO atomic
620 		 * writes.
621 		 */
622 		WARN_ON_ONCE(iocb->ki_flags & IOCB_ATOMIC);
623 
624 		offset = iocb->ki_pos;
625 		err = ext4_buffered_write_iter(iocb, from);
626 		if (err < 0)
627 			return err;
628 
629 		/*
630 		 * We need to ensure that the pages within the page cache for
631 		 * the range covered by this I/O are written to disk and
632 		 * invalidated. This is in attempt to preserve the expected
633 		 * direct I/O semantics in the case we fallback to buffered I/O
634 		 * to complete off the I/O request.
635 		 */
636 		ret += err;
637 		endbyte = offset + err - 1;
638 		err = filemap_write_and_wait_range(iocb->ki_filp->f_mapping,
639 						   offset, endbyte);
640 		if (!err)
641 			invalidate_mapping_pages(iocb->ki_filp->f_mapping,
642 						 offset >> PAGE_SHIFT,
643 						 endbyte >> PAGE_SHIFT);
644 	}
645 
646 	return ret;
647 }
648 
649 #ifdef CONFIG_FS_DAX
650 static ssize_t
651 ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
652 {
653 	ssize_t ret;
654 	size_t count;
655 	loff_t offset;
656 	handle_t *handle;
657 	bool extend = false;
658 	struct inode *inode = file_inode(iocb->ki_filp);
659 
660 	if (iocb->ki_flags & IOCB_NOWAIT) {
661 		if (!inode_trylock(inode))
662 			return -EAGAIN;
663 	} else {
664 		inode_lock(inode);
665 	}
666 
667 	ret = ext4_write_checks(iocb, from);
668 	if (ret <= 0)
669 		goto out;
670 
671 	offset = iocb->ki_pos;
672 	count = iov_iter_count(from);
673 
674 	if (offset + count > EXT4_I(inode)->i_disksize) {
675 		handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
676 		if (IS_ERR(handle)) {
677 			ret = PTR_ERR(handle);
678 			goto out;
679 		}
680 
681 		ret = ext4_orphan_add(handle, inode);
682 		if (ret) {
683 			ext4_journal_stop(handle);
684 			goto out;
685 		}
686 
687 		extend = true;
688 		ext4_journal_stop(handle);
689 	}
690 
691 	ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
692 
693 	if (extend) {
694 		ret = ext4_handle_inode_extension(inode, offset, ret, count);
695 		ext4_inode_extension_cleanup(inode, ret < (ssize_t)count);
696 	}
697 out:
698 	inode_unlock(inode);
699 	if (ret > 0)
700 		ret = generic_write_sync(iocb, ret);
701 	return ret;
702 }
703 #endif
704 
705 static ssize_t
706 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
707 {
708 	int ret;
709 	struct inode *inode = file_inode(iocb->ki_filp);
710 
711 	ret = ext4_emergency_state(inode->i_sb);
712 	if (unlikely(ret))
713 		return ret;
714 
715 #ifdef CONFIG_FS_DAX
716 	if (IS_DAX(inode))
717 		return ext4_dax_write_iter(iocb, from);
718 #endif
719 
720 	if (iocb->ki_flags & IOCB_ATOMIC) {
721 		size_t len = iov_iter_count(from);
722 
723 		if (len < EXT4_SB(inode->i_sb)->s_awu_min ||
724 		    len > EXT4_SB(inode->i_sb)->s_awu_max)
725 			return -EINVAL;
726 
727 		ret = generic_atomic_write_valid(iocb, from);
728 		if (ret)
729 			return ret;
730 	}
731 
732 	if (iocb->ki_flags & IOCB_DIRECT)
733 		return ext4_dio_write_iter(iocb, from);
734 	else
735 		return ext4_buffered_write_iter(iocb, from);
736 }
737 
738 #ifdef CONFIG_FS_DAX
739 static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf, unsigned int order)
740 {
741 	int error = 0;
742 	vm_fault_t result;
743 	int retries = 0;
744 	handle_t *handle = NULL;
745 	struct inode *inode = file_inode(vmf->vma->vm_file);
746 	struct super_block *sb = inode->i_sb;
747 
748 	/*
749 	 * We have to distinguish real writes from writes which will result in a
750 	 * COW page; COW writes should *not* poke the journal (the file will not
751 	 * be changed). Doing so would cause unintended failures when mounted
752 	 * read-only.
753 	 *
754 	 * We check for VM_SHARED rather than vmf->cow_page since the latter is
755 	 * unset for order != 0 (i.e. only in do_cow_fault); for
756 	 * other sizes, dax_iomap_fault will handle splitting / fallback so that
757 	 * we eventually come back with a COW page.
758 	 */
759 	bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
760 		(vmf->vma->vm_flags & VM_SHARED);
761 	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
762 	unsigned long pfn;
763 
764 	if (write) {
765 		sb_start_pagefault(sb);
766 		file_update_time(vmf->vma->vm_file);
767 		filemap_invalidate_lock_shared(mapping);
768 retry:
769 		handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
770 					       EXT4_DATA_TRANS_BLOCKS(sb));
771 		if (IS_ERR(handle)) {
772 			filemap_invalidate_unlock_shared(mapping);
773 			sb_end_pagefault(sb);
774 			return VM_FAULT_SIGBUS;
775 		}
776 	} else {
777 		filemap_invalidate_lock_shared(mapping);
778 	}
779 	result = dax_iomap_fault(vmf, order, &pfn, &error, &ext4_iomap_ops);
780 	if (write) {
781 		ext4_journal_stop(handle);
782 
783 		if ((result & VM_FAULT_ERROR) && error == -ENOSPC &&
784 		    ext4_should_retry_alloc(sb, &retries))
785 			goto retry;
786 		/* Handling synchronous page fault? */
787 		if (result & VM_FAULT_NEEDDSYNC)
788 			result = dax_finish_sync_fault(vmf, order, pfn);
789 		filemap_invalidate_unlock_shared(mapping);
790 		sb_end_pagefault(sb);
791 	} else {
792 		filemap_invalidate_unlock_shared(mapping);
793 	}
794 
795 	return result;
796 }
797 
798 static vm_fault_t ext4_dax_fault(struct vm_fault *vmf)
799 {
800 	return ext4_dax_huge_fault(vmf, 0);
801 }
802 
803 static const struct vm_operations_struct ext4_dax_vm_ops = {
804 	.fault		= ext4_dax_fault,
805 	.huge_fault	= ext4_dax_huge_fault,
806 	.page_mkwrite	= ext4_dax_fault,
807 	.pfn_mkwrite	= ext4_dax_fault,
808 };
809 #else
810 #define ext4_dax_vm_ops	ext4_file_vm_ops
811 #endif
812 
813 static const struct vm_operations_struct ext4_file_vm_ops = {
814 	.fault		= filemap_fault,
815 	.map_pages	= filemap_map_pages,
816 	.page_mkwrite   = ext4_page_mkwrite,
817 };
818 
819 static int ext4_file_mmap_prepare(struct vm_area_desc *desc)
820 {
821 	int ret;
822 	struct file *file = desc->file;
823 	struct inode *inode = file->f_mapping->host;
824 	struct dax_device *dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
825 
826 	if (file->f_mode & FMODE_WRITE)
827 		ret = ext4_emergency_state(inode->i_sb);
828 	else
829 		ret = ext4_forced_shutdown(inode->i_sb) ? -EIO : 0;
830 	if (unlikely(ret))
831 		return ret;
832 
833 	/*
834 	 * We don't support synchronous mappings for non-DAX files and
835 	 * for DAX files if underneath dax_device is not synchronous.
836 	 */
837 	if (!daxdev_mapping_supported(desc, file_inode(file), dax_dev))
838 		return -EOPNOTSUPP;
839 
840 	file_accessed(file);
841 	if (IS_DAX(file_inode(file))) {
842 		desc->vm_ops = &ext4_dax_vm_ops;
843 		vma_desc_set_flags(desc, VMA_HUGEPAGE_BIT);
844 	} else {
845 		desc->vm_ops = &ext4_file_vm_ops;
846 	}
847 	return 0;
848 }
849 
850 static int ext4_sample_last_mounted(struct super_block *sb,
851 				    struct vfsmount *mnt)
852 {
853 	struct ext4_sb_info *sbi = EXT4_SB(sb);
854 	struct path path;
855 	char buf[64], *cp;
856 	handle_t *handle;
857 	int err;
858 
859 	if (likely(ext4_test_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED)))
860 		return 0;
861 
862 	if (ext4_emergency_state(sb) || sb_rdonly(sb) ||
863 	    !sb_start_intwrite_trylock(sb))
864 		return 0;
865 
866 	ext4_set_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED);
867 	/*
868 	 * Sample where the filesystem has been mounted and
869 	 * store it in the superblock for sysadmin convenience
870 	 * when trying to sort through large numbers of block
871 	 * devices or filesystem images.
872 	 */
873 	path.mnt = mnt;
874 	path.dentry = mnt->mnt_root;
875 	cp = d_path(&path, buf, sizeof(buf));
876 	err = 0;
877 	if (IS_ERR(cp))
878 		goto out;
879 
880 	handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
881 	err = PTR_ERR(handle);
882 	if (IS_ERR(handle))
883 		goto out;
884 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
885 	err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
886 					    EXT4_JTR_NONE);
887 	if (err)
888 		goto out_journal;
889 	lock_buffer(sbi->s_sbh);
890 	strtomem_pad(sbi->s_es->s_last_mounted, cp, 0);
891 	ext4_superblock_csum_set(sb);
892 	unlock_buffer(sbi->s_sbh);
893 	ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
894 out_journal:
895 	ext4_journal_stop(handle);
896 out:
897 	sb_end_intwrite(sb);
898 	return err;
899 }
900 
901 static int ext4_file_open(struct inode *inode, struct file *filp)
902 {
903 	int ret;
904 
905 	if (filp->f_mode & FMODE_WRITE)
906 		ret = ext4_emergency_state(inode->i_sb);
907 	else
908 		ret = ext4_forced_shutdown(inode->i_sb) ? -EIO : 0;
909 	if (unlikely(ret))
910 		return ret;
911 
912 	ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt);
913 	if (ret)
914 		return ret;
915 
916 	ret = fscrypt_file_open(inode, filp);
917 	if (ret)
918 		return ret;
919 
920 	ret = fsverity_file_open(inode, filp);
921 	if (ret)
922 		return ret;
923 
924 	/*
925 	 * Set up the jbd2_inode if we are opening the inode for
926 	 * writing and the journal is present
927 	 */
928 	if (filp->f_mode & FMODE_WRITE) {
929 		ret = ext4_inode_attach_jinode(inode);
930 		if (ret < 0)
931 			return ret;
932 	}
933 
934 	if (ext4_inode_can_atomic_write(inode))
935 		filp->f_mode |= FMODE_CAN_ATOMIC_WRITE;
936 
937 	filp->f_mode |= FMODE_NOWAIT | FMODE_CAN_ODIRECT;
938 	return dquot_file_open(inode, filp);
939 }
940 
941 /*
942  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
943  * by calling generic_file_llseek_size() with the appropriate maxbytes
944  * value for each.
945  */
946 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
947 {
948 	struct inode *inode = file->f_mapping->host;
949 	loff_t maxbytes = ext4_get_maxbytes(inode);
950 
951 	switch (whence) {
952 	default:
953 		return generic_file_llseek_size(file, offset, whence,
954 						maxbytes, i_size_read(inode));
955 	case SEEK_HOLE:
956 		inode_lock_shared(inode);
957 		offset = iomap_seek_hole(inode, offset,
958 					 &ext4_iomap_report_ops);
959 		inode_unlock_shared(inode);
960 		break;
961 	case SEEK_DATA:
962 		inode_lock_shared(inode);
963 		offset = iomap_seek_data(inode, offset,
964 					 &ext4_iomap_report_ops);
965 		inode_unlock_shared(inode);
966 		break;
967 	}
968 
969 	if (offset < 0)
970 		return offset;
971 	return vfs_setpos(file, offset, maxbytes);
972 }
973 
974 const struct file_operations ext4_file_operations = {
975 	.llseek		= ext4_llseek,
976 	.read_iter	= ext4_file_read_iter,
977 	.write_iter	= ext4_file_write_iter,
978 	.iopoll		= iocb_bio_iopoll,
979 	.unlocked_ioctl = ext4_ioctl,
980 #ifdef CONFIG_COMPAT
981 	.compat_ioctl	= ext4_compat_ioctl,
982 #endif
983 	.mmap_prepare	= ext4_file_mmap_prepare,
984 	.open		= ext4_file_open,
985 	.release	= ext4_release_file,
986 	.fsync		= ext4_sync_file,
987 	.get_unmapped_area = thp_get_unmapped_area,
988 	.splice_read	= ext4_file_splice_read,
989 	.splice_write	= iter_file_splice_write,
990 	.fallocate	= ext4_fallocate,
991 	.fop_flags	= FOP_MMAP_SYNC | FOP_BUFFER_RASYNC |
992 			  FOP_DIO_PARALLEL_WRITE |
993 			  FOP_DONTCACHE,
994 	.setlease	= generic_setlease,
995 };
996 
997 const struct inode_operations ext4_file_inode_operations = {
998 	.setattr	= ext4_setattr,
999 	.getattr	= ext4_file_getattr,
1000 	.listxattr	= ext4_listxattr,
1001 	.get_inode_acl	= ext4_get_acl,
1002 	.set_acl	= ext4_set_acl,
1003 	.fiemap		= ext4_fiemap,
1004 	.fileattr_get	= ext4_fileattr_get,
1005 	.fileattr_set	= ext4_fileattr_set,
1006 };
1007 
1008