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
ext4_dio_supported(struct inode * inode)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
ext4_dio_read_iter(struct kiocb * iocb,struct iov_iter * to)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, 0, 0);
78 inode_unlock_shared(inode);
79
80 file_accessed(iocb->ki_filp);
81 return ret;
82 }
83
84 #ifdef CONFIG_FS_DAX
ext4_dax_read_iter(struct kiocb * iocb,struct iov_iter * to)85 static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
86 {
87 struct inode *inode = file_inode(iocb->ki_filp);
88 ssize_t ret;
89
90 if (iocb->ki_flags & IOCB_NOWAIT) {
91 if (!inode_trylock_shared(inode))
92 return -EAGAIN;
93 } else {
94 inode_lock_shared(inode);
95 }
96 /*
97 * Recheck under inode lock - at this point we are sure it cannot
98 * change anymore
99 */
100 if (!IS_DAX(inode)) {
101 inode_unlock_shared(inode);
102 /* Fallback to buffered IO in case we cannot support DAX */
103 return generic_file_read_iter(iocb, to);
104 }
105 ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
106 inode_unlock_shared(inode);
107
108 file_accessed(iocb->ki_filp);
109 return ret;
110 }
111 #endif
112
ext4_file_read_iter(struct kiocb * iocb,struct iov_iter * to)113 static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
114 {
115 struct inode *inode = file_inode(iocb->ki_filp);
116
117 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
118 return -EIO;
119
120 if (!iov_iter_count(to))
121 return 0; /* skip atime */
122
123 #ifdef CONFIG_FS_DAX
124 if (IS_DAX(inode))
125 return ext4_dax_read_iter(iocb, to);
126 #endif
127 if (iocb->ki_flags & IOCB_DIRECT)
128 return ext4_dio_read_iter(iocb, to);
129
130 return generic_file_read_iter(iocb, to);
131 }
132
133 /*
134 * Called when an inode is released. Note that this is different
135 * from ext4_file_open: open gets called at every open, but release
136 * gets called only when /all/ the files are closed.
137 */
ext4_release_file(struct inode * inode,struct file * filp)138 static int ext4_release_file(struct inode *inode, struct file *filp)
139 {
140 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
141 ext4_alloc_da_blocks(inode);
142 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
143 }
144 /* if we are the last writer on the inode, drop the block reservation */
145 if ((filp->f_mode & FMODE_WRITE) &&
146 (atomic_read(&inode->i_writecount) == 1) &&
147 !EXT4_I(inode)->i_reserved_data_blocks) {
148 down_write(&EXT4_I(inode)->i_data_sem);
149 ext4_discard_preallocations(inode, 0);
150 up_write(&EXT4_I(inode)->i_data_sem);
151 }
152 if (is_dx(inode) && filp->private_data)
153 ext4_htree_free_dir_info(filp->private_data);
154
155 return 0;
156 }
157
158 /*
159 * This tests whether the IO in question is block-aligned or not.
160 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
161 * are converted to written only after the IO is complete. Until they are
162 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
163 * it needs to zero out portions of the start and/or end block. If 2 AIO
164 * threads are at work on the same unwritten block, they must be synchronized
165 * or one thread will zero the other's data, causing corruption.
166 */
167 static bool
ext4_unaligned_io(struct inode * inode,struct iov_iter * from,loff_t pos)168 ext4_unaligned_io(struct inode *inode, struct iov_iter *from, loff_t pos)
169 {
170 struct super_block *sb = inode->i_sb;
171 unsigned long blockmask = sb->s_blocksize - 1;
172
173 if ((pos | iov_iter_alignment(from)) & blockmask)
174 return true;
175
176 return false;
177 }
178
179 static bool
ext4_extending_io(struct inode * inode,loff_t offset,size_t len)180 ext4_extending_io(struct inode *inode, loff_t offset, size_t len)
181 {
182 if (offset + len > i_size_read(inode) ||
183 offset + len > EXT4_I(inode)->i_disksize)
184 return true;
185 return false;
186 }
187
188 /* Is IO overwriting allocated and initialized blocks? */
ext4_overwrite_io(struct inode * inode,loff_t pos,loff_t len)189 static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
190 {
191 struct ext4_map_blocks map;
192 unsigned int blkbits = inode->i_blkbits;
193 int err, blklen;
194
195 if (pos + len > i_size_read(inode))
196 return false;
197
198 map.m_lblk = pos >> blkbits;
199 map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
200 blklen = map.m_len;
201
202 err = ext4_map_blocks(NULL, inode, &map, 0);
203 /*
204 * 'err==len' means that all of the blocks have been preallocated,
205 * regardless of whether they have been initialized or not. To exclude
206 * unwritten extents, we need to check m_flags.
207 */
208 return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
209 }
210
ext4_generic_write_checks(struct kiocb * iocb,struct iov_iter * from)211 static ssize_t ext4_generic_write_checks(struct kiocb *iocb,
212 struct iov_iter *from)
213 {
214 struct inode *inode = file_inode(iocb->ki_filp);
215 ssize_t ret;
216
217 if (unlikely(IS_IMMUTABLE(inode)))
218 return -EPERM;
219
220 ret = generic_write_checks(iocb, from);
221 if (ret <= 0)
222 return ret;
223
224 /*
225 * If we have encountered a bitmap-format file, the size limit
226 * is smaller than s_maxbytes, which is for extent-mapped files.
227 */
228 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
229 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
230
231 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
232 return -EFBIG;
233 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
234 }
235
236 return iov_iter_count(from);
237 }
238
ext4_write_checks(struct kiocb * iocb,struct iov_iter * from)239 static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
240 {
241 ssize_t ret, count;
242
243 count = ext4_generic_write_checks(iocb, from);
244 if (count <= 0)
245 return count;
246
247 ret = file_modified(iocb->ki_filp);
248 if (ret)
249 return ret;
250 return count;
251 }
252
ext4_buffered_write_iter(struct kiocb * iocb,struct iov_iter * from)253 static ssize_t ext4_buffered_write_iter(struct kiocb *iocb,
254 struct iov_iter *from)
255 {
256 ssize_t ret;
257 struct inode *inode = file_inode(iocb->ki_filp);
258
259 if (iocb->ki_flags & IOCB_NOWAIT)
260 return -EOPNOTSUPP;
261
262 ext4_fc_start_update(inode);
263 inode_lock(inode);
264 ret = ext4_write_checks(iocb, from);
265 if (ret <= 0)
266 goto out;
267
268 current->backing_dev_info = inode_to_bdi(inode);
269 ret = generic_perform_write(iocb->ki_filp, from, iocb->ki_pos);
270 current->backing_dev_info = NULL;
271
272 out:
273 inode_unlock(inode);
274 ext4_fc_stop_update(inode);
275 if (likely(ret > 0)) {
276 iocb->ki_pos += ret;
277 ret = generic_write_sync(iocb, ret);
278 }
279
280 return ret;
281 }
282
ext4_handle_inode_extension(struct inode * inode,loff_t offset,ssize_t written,size_t count)283 static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
284 ssize_t written, size_t count)
285 {
286 handle_t *handle;
287 bool truncate = false;
288 u8 blkbits = inode->i_blkbits;
289 ext4_lblk_t written_blk, end_blk;
290 int ret;
291
292 /*
293 * Note that EXT4_I(inode)->i_disksize can get extended up to
294 * inode->i_size while the I/O was running due to writeback of delalloc
295 * blocks. But, the code in ext4_iomap_alloc() is careful to use
296 * zeroed/unwritten extents if this is possible; thus we won't leave
297 * uninitialized blocks in a file even if we didn't succeed in writing
298 * as much as we intended.
299 */
300 WARN_ON_ONCE(i_size_read(inode) < EXT4_I(inode)->i_disksize);
301 if (offset + count <= EXT4_I(inode)->i_disksize) {
302 /*
303 * We need to ensure that the inode is removed from the orphan
304 * list if it has been added prematurely, due to writeback of
305 * delalloc blocks.
306 */
307 if (!list_empty(&EXT4_I(inode)->i_orphan) && inode->i_nlink) {
308 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
309
310 if (IS_ERR(handle)) {
311 ext4_orphan_del(NULL, inode);
312 return PTR_ERR(handle);
313 }
314
315 ext4_orphan_del(handle, inode);
316 ext4_journal_stop(handle);
317 }
318
319 return written;
320 }
321
322 if (written < 0)
323 goto truncate;
324
325 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
326 if (IS_ERR(handle)) {
327 written = PTR_ERR(handle);
328 goto truncate;
329 }
330
331 if (ext4_update_inode_size(inode, offset + written)) {
332 ret = ext4_mark_inode_dirty(handle, inode);
333 if (unlikely(ret)) {
334 written = ret;
335 ext4_journal_stop(handle);
336 goto truncate;
337 }
338 }
339
340 /*
341 * We may need to truncate allocated but not written blocks beyond EOF.
342 */
343 written_blk = ALIGN(offset + written, 1 << blkbits);
344 end_blk = ALIGN(offset + count, 1 << blkbits);
345 if (written_blk < end_blk && ext4_can_truncate(inode))
346 truncate = true;
347
348 /*
349 * Remove the inode from the orphan list if it has been extended and
350 * everything went OK.
351 */
352 if (!truncate && inode->i_nlink)
353 ext4_orphan_del(handle, inode);
354 ext4_journal_stop(handle);
355
356 if (truncate) {
357 truncate:
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 }
367
368 return written;
369 }
370
ext4_dio_write_end_io(struct kiocb * iocb,ssize_t size,int error,unsigned int flags)371 static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
372 int error, unsigned int flags)
373 {
374 loff_t pos = iocb->ki_pos;
375 struct inode *inode = file_inode(iocb->ki_filp);
376
377 if (error)
378 return error;
379
380 if (size && flags & IOMAP_DIO_UNWRITTEN) {
381 error = ext4_convert_unwritten_extents(NULL, inode, pos, size);
382 if (error < 0)
383 return error;
384 }
385 /*
386 * If we are extending the file, we have to update i_size here before
387 * page cache gets invalidated in iomap_dio_rw(). Otherwise racing
388 * buffered reads could zero out too much from page cache pages. Update
389 * of on-disk size will happen later in ext4_dio_write_iter() where
390 * we have enough information to also perform orphan list handling etc.
391 * Note that we perform all extending writes synchronously under
392 * i_rwsem held exclusively so i_size update is safe here in that case.
393 * If the write was not extending, we cannot see pos > i_size here
394 * because operations reducing i_size like truncate wait for all
395 * outstanding DIO before updating i_size.
396 */
397 pos += size;
398 if (pos > i_size_read(inode))
399 i_size_write(inode, pos);
400
401 return 0;
402 }
403
404 static const struct iomap_dio_ops ext4_dio_write_ops = {
405 .end_io = ext4_dio_write_end_io,
406 };
407
408 /*
409 * The intention here is to start with shared lock acquired then see if any
410 * condition requires an exclusive inode lock. If yes, then we restart the
411 * whole operation by releasing the shared lock and acquiring exclusive lock.
412 *
413 * - For unaligned_io we never take shared lock as it may cause data corruption
414 * when two unaligned IO tries to modify the same block e.g. while zeroing.
415 *
416 * - For extending writes case we don't take the shared lock, since it requires
417 * updating inode i_disksize and/or orphan handling with exclusive lock.
418 *
419 * - shared locking will only be true mostly with overwrites. Otherwise we will
420 * switch to exclusive i_rwsem lock.
421 */
ext4_dio_write_checks(struct kiocb * iocb,struct iov_iter * from,bool * ilock_shared,bool * extend)422 static ssize_t ext4_dio_write_checks(struct kiocb *iocb, struct iov_iter *from,
423 bool *ilock_shared, bool *extend)
424 {
425 struct file *file = iocb->ki_filp;
426 struct inode *inode = file_inode(file);
427 loff_t offset;
428 size_t count;
429 ssize_t ret;
430
431 restart:
432 ret = ext4_generic_write_checks(iocb, from);
433 if (ret <= 0)
434 goto out;
435
436 offset = iocb->ki_pos;
437 count = ret;
438 if (ext4_extending_io(inode, offset, count))
439 *extend = true;
440 /*
441 * Determine whether the IO operation will overwrite allocated
442 * and initialized blocks.
443 * We need exclusive i_rwsem for changing security info
444 * in file_modified().
445 */
446 if (*ilock_shared && (!IS_NOSEC(inode) || *extend ||
447 !ext4_overwrite_io(inode, offset, count))) {
448 if (iocb->ki_flags & IOCB_NOWAIT) {
449 ret = -EAGAIN;
450 goto out;
451 }
452 inode_unlock_shared(inode);
453 *ilock_shared = false;
454 inode_lock(inode);
455 goto restart;
456 }
457
458 ret = file_modified(file);
459 if (ret < 0)
460 goto out;
461
462 return count;
463 out:
464 if (*ilock_shared)
465 inode_unlock_shared(inode);
466 else
467 inode_unlock(inode);
468 return ret;
469 }
470
ext4_dio_write_iter(struct kiocb * iocb,struct iov_iter * from)471 static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from)
472 {
473 ssize_t ret;
474 handle_t *handle;
475 struct inode *inode = file_inode(iocb->ki_filp);
476 loff_t offset = iocb->ki_pos;
477 size_t count = iov_iter_count(from);
478 const struct iomap_ops *iomap_ops = &ext4_iomap_ops;
479 bool extend = false, unaligned_io = false;
480 bool ilock_shared = true;
481
482 /*
483 * We initially start with shared inode lock unless it is
484 * unaligned IO which needs exclusive lock anyways.
485 */
486 if (ext4_unaligned_io(inode, from, offset)) {
487 unaligned_io = true;
488 ilock_shared = false;
489 }
490 /*
491 * Quick check here without any i_rwsem lock to see if it is extending
492 * IO. A more reliable check is done in ext4_dio_write_checks() with
493 * proper locking in place.
494 */
495 if (offset + count > i_size_read(inode))
496 ilock_shared = false;
497
498 if (iocb->ki_flags & IOCB_NOWAIT) {
499 if (ilock_shared) {
500 if (!inode_trylock_shared(inode))
501 return -EAGAIN;
502 } else {
503 if (!inode_trylock(inode))
504 return -EAGAIN;
505 }
506 } else {
507 if (ilock_shared)
508 inode_lock_shared(inode);
509 else
510 inode_lock(inode);
511 }
512
513 /* Fallback to buffered I/O if the inode does not support direct I/O. */
514 if (!ext4_dio_supported(inode)) {
515 if (ilock_shared)
516 inode_unlock_shared(inode);
517 else
518 inode_unlock(inode);
519 return ext4_buffered_write_iter(iocb, from);
520 }
521
522 ret = ext4_dio_write_checks(iocb, from, &ilock_shared, &extend);
523 if (ret <= 0)
524 return ret;
525
526 /* if we're going to block and IOCB_NOWAIT is set, return -EAGAIN */
527 if ((iocb->ki_flags & IOCB_NOWAIT) && (unaligned_io || extend)) {
528 ret = -EAGAIN;
529 goto out;
530 }
531
532 offset = iocb->ki_pos;
533 count = ret;
534
535 /*
536 * Unaligned direct IO must be serialized among each other as zeroing
537 * of partial blocks of two competing unaligned IOs can result in data
538 * corruption.
539 *
540 * So we make sure we don't allow any unaligned IO in flight.
541 * For IOs where we need not wait (like unaligned non-AIO DIO),
542 * below inode_dio_wait() may anyway become a no-op, since we start
543 * with exclusive lock.
544 */
545 if (unaligned_io)
546 inode_dio_wait(inode);
547
548 if (extend) {
549 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
550 if (IS_ERR(handle)) {
551 ret = PTR_ERR(handle);
552 goto out;
553 }
554
555 ext4_fc_start_update(inode);
556 ret = ext4_orphan_add(handle, inode);
557 ext4_fc_stop_update(inode);
558 if (ret) {
559 ext4_journal_stop(handle);
560 goto out;
561 }
562
563 ext4_journal_stop(handle);
564 }
565
566 if (ilock_shared)
567 iomap_ops = &ext4_iomap_overwrite_ops;
568 ret = iomap_dio_rw(iocb, from, iomap_ops, &ext4_dio_write_ops,
569 (unaligned_io || extend) ? IOMAP_DIO_FORCE_WAIT : 0,
570 0);
571 if (ret == -ENOTBLK)
572 ret = 0;
573
574 if (extend)
575 ret = ext4_handle_inode_extension(inode, offset, ret, count);
576
577 out:
578 if (ilock_shared)
579 inode_unlock_shared(inode);
580 else
581 inode_unlock(inode);
582
583 if (ret >= 0 && iov_iter_count(from)) {
584 ssize_t err;
585 loff_t endbyte;
586
587 offset = iocb->ki_pos;
588 err = ext4_buffered_write_iter(iocb, from);
589 if (err < 0)
590 return err;
591
592 /*
593 * We need to ensure that the pages within the page cache for
594 * the range covered by this I/O are written to disk and
595 * invalidated. This is in attempt to preserve the expected
596 * direct I/O semantics in the case we fallback to buffered I/O
597 * to complete off the I/O request.
598 */
599 ret += err;
600 endbyte = offset + err - 1;
601 err = filemap_write_and_wait_range(iocb->ki_filp->f_mapping,
602 offset, endbyte);
603 if (!err)
604 invalidate_mapping_pages(iocb->ki_filp->f_mapping,
605 offset >> PAGE_SHIFT,
606 endbyte >> PAGE_SHIFT);
607 }
608
609 return ret;
610 }
611
612 #ifdef CONFIG_FS_DAX
613 static ssize_t
ext4_dax_write_iter(struct kiocb * iocb,struct iov_iter * from)614 ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
615 {
616 ssize_t ret;
617 size_t count;
618 loff_t offset;
619 handle_t *handle;
620 bool extend = false;
621 struct inode *inode = file_inode(iocb->ki_filp);
622
623 if (iocb->ki_flags & IOCB_NOWAIT) {
624 if (!inode_trylock(inode))
625 return -EAGAIN;
626 } else {
627 inode_lock(inode);
628 }
629
630 ret = ext4_write_checks(iocb, from);
631 if (ret <= 0)
632 goto out;
633
634 offset = iocb->ki_pos;
635 count = iov_iter_count(from);
636
637 if (offset + count > EXT4_I(inode)->i_disksize) {
638 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
639 if (IS_ERR(handle)) {
640 ret = PTR_ERR(handle);
641 goto out;
642 }
643
644 ret = ext4_orphan_add(handle, inode);
645 if (ret) {
646 ext4_journal_stop(handle);
647 goto out;
648 }
649
650 extend = true;
651 ext4_journal_stop(handle);
652 }
653
654 ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
655
656 if (extend)
657 ret = ext4_handle_inode_extension(inode, offset, ret, count);
658 out:
659 inode_unlock(inode);
660 if (ret > 0)
661 ret = generic_write_sync(iocb, ret);
662 return ret;
663 }
664 #endif
665
666 static ssize_t
ext4_file_write_iter(struct kiocb * iocb,struct iov_iter * from)667 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
668 {
669 struct inode *inode = file_inode(iocb->ki_filp);
670
671 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
672 return -EIO;
673
674 #ifdef CONFIG_FS_DAX
675 if (IS_DAX(inode))
676 return ext4_dax_write_iter(iocb, from);
677 #endif
678 if (iocb->ki_flags & IOCB_DIRECT)
679 return ext4_dio_write_iter(iocb, from);
680 else
681 return ext4_buffered_write_iter(iocb, from);
682 }
683
684 #ifdef CONFIG_FS_DAX
ext4_dax_huge_fault(struct vm_fault * vmf,enum page_entry_size pe_size)685 static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf,
686 enum page_entry_size pe_size)
687 {
688 int error = 0;
689 vm_fault_t result;
690 int retries = 0;
691 handle_t *handle = NULL;
692 struct inode *inode = file_inode(vmf->vma->vm_file);
693 struct super_block *sb = inode->i_sb;
694
695 /*
696 * We have to distinguish real writes from writes which will result in a
697 * COW page; COW writes should *not* poke the journal (the file will not
698 * be changed). Doing so would cause unintended failures when mounted
699 * read-only.
700 *
701 * We check for VM_SHARED rather than vmf->cow_page since the latter is
702 * unset for pe_size != PE_SIZE_PTE (i.e. only in do_cow_fault); for
703 * other sizes, dax_iomap_fault will handle splitting / fallback so that
704 * we eventually come back with a COW page.
705 */
706 bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
707 (vmf->vma->vm_flags & VM_SHARED);
708 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
709 pfn_t pfn;
710
711 if (write) {
712 sb_start_pagefault(sb);
713 file_update_time(vmf->vma->vm_file);
714 filemap_invalidate_lock_shared(mapping);
715 retry:
716 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
717 EXT4_DATA_TRANS_BLOCKS(sb));
718 if (IS_ERR(handle)) {
719 filemap_invalidate_unlock_shared(mapping);
720 sb_end_pagefault(sb);
721 return VM_FAULT_SIGBUS;
722 }
723 } else {
724 filemap_invalidate_lock_shared(mapping);
725 }
726 result = dax_iomap_fault(vmf, pe_size, &pfn, &error, &ext4_iomap_ops);
727 if (write) {
728 ext4_journal_stop(handle);
729
730 if ((result & VM_FAULT_ERROR) && error == -ENOSPC &&
731 ext4_should_retry_alloc(sb, &retries))
732 goto retry;
733 /* Handling synchronous page fault? */
734 if (result & VM_FAULT_NEEDDSYNC)
735 result = dax_finish_sync_fault(vmf, pe_size, pfn);
736 filemap_invalidate_unlock_shared(mapping);
737 sb_end_pagefault(sb);
738 } else {
739 filemap_invalidate_unlock_shared(mapping);
740 }
741
742 return result;
743 }
744
ext4_dax_fault(struct vm_fault * vmf)745 static vm_fault_t ext4_dax_fault(struct vm_fault *vmf)
746 {
747 return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
748 }
749
750 static const struct vm_operations_struct ext4_dax_vm_ops = {
751 .fault = ext4_dax_fault,
752 .huge_fault = ext4_dax_huge_fault,
753 .page_mkwrite = ext4_dax_fault,
754 .pfn_mkwrite = ext4_dax_fault,
755 };
756 #else
757 #define ext4_dax_vm_ops ext4_file_vm_ops
758 #endif
759
760 static const struct vm_operations_struct ext4_file_vm_ops = {
761 .fault = filemap_fault,
762 .map_pages = filemap_map_pages,
763 .page_mkwrite = ext4_page_mkwrite,
764 };
765
ext4_file_mmap(struct file * file,struct vm_area_struct * vma)766 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
767 {
768 struct inode *inode = file->f_mapping->host;
769 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
770 struct dax_device *dax_dev = sbi->s_daxdev;
771
772 if (unlikely(ext4_forced_shutdown(sbi)))
773 return -EIO;
774
775 /*
776 * We don't support synchronous mappings for non-DAX files and
777 * for DAX files if underneath dax_device is not synchronous.
778 */
779 if (!daxdev_mapping_supported(vma, dax_dev))
780 return -EOPNOTSUPP;
781
782 file_accessed(file);
783 if (IS_DAX(file_inode(file))) {
784 vma->vm_ops = &ext4_dax_vm_ops;
785 vma->vm_flags |= VM_HUGEPAGE;
786 } else {
787 vma->vm_ops = &ext4_file_vm_ops;
788 }
789 return 0;
790 }
791
ext4_sample_last_mounted(struct super_block * sb,struct vfsmount * mnt)792 static int ext4_sample_last_mounted(struct super_block *sb,
793 struct vfsmount *mnt)
794 {
795 struct ext4_sb_info *sbi = EXT4_SB(sb);
796 struct path path;
797 char buf[64], *cp;
798 handle_t *handle;
799 int err;
800
801 if (likely(ext4_test_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED)))
802 return 0;
803
804 if (sb_rdonly(sb) || !sb_start_intwrite_trylock(sb))
805 return 0;
806
807 ext4_set_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED);
808 /*
809 * Sample where the filesystem has been mounted and
810 * store it in the superblock for sysadmin convenience
811 * when trying to sort through large numbers of block
812 * devices or filesystem images.
813 */
814 memset(buf, 0, sizeof(buf));
815 path.mnt = mnt;
816 path.dentry = mnt->mnt_root;
817 cp = d_path(&path, buf, sizeof(buf));
818 err = 0;
819 if (IS_ERR(cp))
820 goto out;
821
822 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
823 err = PTR_ERR(handle);
824 if (IS_ERR(handle))
825 goto out;
826 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
827 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
828 EXT4_JTR_NONE);
829 if (err)
830 goto out_journal;
831 lock_buffer(sbi->s_sbh);
832 strncpy(sbi->s_es->s_last_mounted, cp,
833 sizeof(sbi->s_es->s_last_mounted));
834 ext4_superblock_csum_set(sb);
835 unlock_buffer(sbi->s_sbh);
836 ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
837 out_journal:
838 ext4_journal_stop(handle);
839 out:
840 sb_end_intwrite(sb);
841 return err;
842 }
843
ext4_file_open(struct inode * inode,struct file * filp)844 static int ext4_file_open(struct inode *inode, struct file *filp)
845 {
846 int ret;
847
848 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
849 return -EIO;
850
851 ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt);
852 if (ret)
853 return ret;
854
855 ret = fscrypt_file_open(inode, filp);
856 if (ret)
857 return ret;
858
859 ret = fsverity_file_open(inode, filp);
860 if (ret)
861 return ret;
862
863 /*
864 * Set up the jbd2_inode if we are opening the inode for
865 * writing and the journal is present
866 */
867 if (filp->f_mode & FMODE_WRITE) {
868 ret = ext4_inode_attach_jinode(inode);
869 if (ret < 0)
870 return ret;
871 }
872
873 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
874 return dquot_file_open(inode, filp);
875 }
876
877 /*
878 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
879 * by calling generic_file_llseek_size() with the appropriate maxbytes
880 * value for each.
881 */
ext4_llseek(struct file * file,loff_t offset,int whence)882 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
883 {
884 struct inode *inode = file->f_mapping->host;
885 loff_t maxbytes;
886
887 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
888 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
889 else
890 maxbytes = inode->i_sb->s_maxbytes;
891
892 switch (whence) {
893 default:
894 return generic_file_llseek_size(file, offset, whence,
895 maxbytes, i_size_read(inode));
896 case SEEK_HOLE:
897 inode_lock_shared(inode);
898 offset = iomap_seek_hole(inode, offset,
899 &ext4_iomap_report_ops);
900 inode_unlock_shared(inode);
901 break;
902 case SEEK_DATA:
903 inode_lock_shared(inode);
904 offset = iomap_seek_data(inode, offset,
905 &ext4_iomap_report_ops);
906 inode_unlock_shared(inode);
907 break;
908 }
909
910 if (offset < 0)
911 return offset;
912 return vfs_setpos(file, offset, maxbytes);
913 }
914
915 const struct file_operations ext4_file_operations = {
916 .llseek = ext4_llseek,
917 .read_iter = ext4_file_read_iter,
918 .write_iter = ext4_file_write_iter,
919 .iopoll = iocb_bio_iopoll,
920 .unlocked_ioctl = ext4_ioctl,
921 #ifdef CONFIG_COMPAT
922 .compat_ioctl = ext4_compat_ioctl,
923 #endif
924 .mmap = ext4_file_mmap,
925 .mmap_supported_flags = MAP_SYNC,
926 .open = ext4_file_open,
927 .release = ext4_release_file,
928 .fsync = ext4_sync_file,
929 .get_unmapped_area = thp_get_unmapped_area,
930 .splice_read = generic_file_splice_read,
931 .splice_write = iter_file_splice_write,
932 .fallocate = ext4_fallocate,
933 };
934
935 const struct inode_operations ext4_file_inode_operations = {
936 .setattr = ext4_setattr,
937 .getattr = ext4_file_getattr,
938 .listxattr = ext4_listxattr,
939 .get_acl = ext4_get_acl,
940 .set_acl = ext4_set_acl,
941 .fiemap = ext4_fiemap,
942 .fileattr_get = ext4_fileattr_get,
943 .fileattr_set = ext4_fileattr_set,
944 };
945
946