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 kiocb * iocb,struct iov_iter * iter)39 static bool ext4_dio_supported(struct kiocb *iocb, struct iov_iter *iter)
40 {
41 struct inode *inode = file_inode(iocb->ki_filp);
42
43 if (!fscrypt_dio_supported(iocb, iter))
44 return false;
45 if (fsverity_active(inode))
46 return false;
47 if (ext4_should_journal_data(inode))
48 return false;
49 if (ext4_has_inline_data(inode))
50 return false;
51 return true;
52 }
53
ext4_dio_read_iter(struct kiocb * iocb,struct iov_iter * to)54 static ssize_t ext4_dio_read_iter(struct kiocb *iocb, struct iov_iter *to)
55 {
56 ssize_t ret;
57 struct inode *inode = file_inode(iocb->ki_filp);
58
59 if (iocb->ki_flags & IOCB_NOWAIT) {
60 if (!inode_trylock_shared(inode))
61 return -EAGAIN;
62 } else {
63 inode_lock_shared(inode);
64 }
65
66 if (!ext4_dio_supported(iocb, to)) {
67 inode_unlock_shared(inode);
68 /*
69 * Fallback to buffered I/O if the operation being performed on
70 * the inode is not supported by direct I/O. The IOCB_DIRECT
71 * flag needs to be cleared here in order to ensure that the
72 * direct I/O path within generic_file_read_iter() is not
73 * taken.
74 */
75 iocb->ki_flags &= ~IOCB_DIRECT;
76 return generic_file_read_iter(iocb, to);
77 }
78
79 ret = iomap_dio_rw(iocb, to, &ext4_iomap_ops, NULL,
80 is_sync_kiocb(iocb));
81 inode_unlock_shared(inode);
82
83 file_accessed(iocb->ki_filp);
84 return ret;
85 }
86
87 #ifdef CONFIG_FS_DAX
ext4_dax_read_iter(struct kiocb * iocb,struct iov_iter * to)88 static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
89 {
90 struct inode *inode = file_inode(iocb->ki_filp);
91 ssize_t ret;
92
93 if (iocb->ki_flags & IOCB_NOWAIT) {
94 if (!inode_trylock_shared(inode))
95 return -EAGAIN;
96 } else {
97 inode_lock_shared(inode);
98 }
99 /*
100 * Recheck under inode lock - at this point we are sure it cannot
101 * change anymore
102 */
103 if (!IS_DAX(inode)) {
104 inode_unlock_shared(inode);
105 /* Fallback to buffered IO in case we cannot support DAX */
106 return generic_file_read_iter(iocb, to);
107 }
108 ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
109 inode_unlock_shared(inode);
110
111 file_accessed(iocb->ki_filp);
112 return ret;
113 }
114 #endif
115
ext4_file_read_iter(struct kiocb * iocb,struct iov_iter * to)116 static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
117 {
118 struct inode *inode = file_inode(iocb->ki_filp);
119
120 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
121 return -EIO;
122
123 if (!iov_iter_count(to))
124 return 0; /* skip atime */
125
126 #ifdef CONFIG_FS_DAX
127 if (IS_DAX(inode))
128 return ext4_dax_read_iter(iocb, to);
129 #endif
130 if (iocb->ki_flags & IOCB_DIRECT)
131 return ext4_dio_read_iter(iocb, to);
132
133 return generic_file_read_iter(iocb, to);
134 }
135
136 /*
137 * Called when an inode is released. Note that this is different
138 * from ext4_file_open: open gets called at every open, but release
139 * gets called only when /all/ the files are closed.
140 */
ext4_release_file(struct inode * inode,struct file * filp)141 static int ext4_release_file(struct inode *inode, struct file *filp)
142 {
143 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
144 ext4_alloc_da_blocks(inode);
145 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
146 }
147 /* if we are the last writer on the inode, drop the block reservation */
148 if ((filp->f_mode & FMODE_WRITE) &&
149 (atomic_read(&inode->i_writecount) == 1) &&
150 !EXT4_I(inode)->i_reserved_data_blocks) {
151 down_write(&EXT4_I(inode)->i_data_sem);
152 ext4_discard_preallocations(inode, 0);
153 up_write(&EXT4_I(inode)->i_data_sem);
154 }
155 if (is_dx(inode) && filp->private_data)
156 ext4_htree_free_dir_info(filp->private_data);
157
158 return 0;
159 }
160
161 /*
162 * This tests whether the IO in question is block-aligned or not.
163 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
164 * are converted to written only after the IO is complete. Until they are
165 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
166 * it needs to zero out portions of the start and/or end block. If 2 AIO
167 * threads are at work on the same unwritten block, they must be synchronized
168 * or one thread will zero the other's data, causing corruption.
169 */
170 static bool
ext4_unaligned_io(struct inode * inode,struct iov_iter * from,loff_t pos)171 ext4_unaligned_io(struct inode *inode, struct iov_iter *from, loff_t pos)
172 {
173 struct super_block *sb = inode->i_sb;
174 unsigned long blockmask = sb->s_blocksize - 1;
175
176 if ((pos | iov_iter_alignment(from)) & blockmask)
177 return true;
178
179 return false;
180 }
181
182 static bool
ext4_extending_io(struct inode * inode,loff_t offset,size_t len)183 ext4_extending_io(struct inode *inode, loff_t offset, size_t len)
184 {
185 if (offset + len > i_size_read(inode) ||
186 offset + len > EXT4_I(inode)->i_disksize)
187 return true;
188 return false;
189 }
190
191 /* Is IO overwriting allocated and initialized blocks? */
ext4_overwrite_io(struct inode * inode,loff_t pos,loff_t len)192 static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
193 {
194 struct ext4_map_blocks map;
195 unsigned int blkbits = inode->i_blkbits;
196 int err, blklen;
197
198 if (pos + len > i_size_read(inode))
199 return false;
200
201 map.m_lblk = pos >> blkbits;
202 map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
203 blklen = map.m_len;
204
205 err = ext4_map_blocks(NULL, inode, &map, 0);
206 /*
207 * 'err==len' means that all of the blocks have been preallocated,
208 * regardless of whether they have been initialized or not. To exclude
209 * unwritten extents, we need to check m_flags.
210 */
211 return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
212 }
213
ext4_generic_write_checks(struct kiocb * iocb,struct iov_iter * from)214 static ssize_t ext4_generic_write_checks(struct kiocb *iocb,
215 struct iov_iter *from)
216 {
217 struct inode *inode = file_inode(iocb->ki_filp);
218 ssize_t ret;
219
220 if (unlikely(IS_IMMUTABLE(inode)))
221 return -EPERM;
222
223 ret = generic_write_checks(iocb, from);
224 if (ret <= 0)
225 return ret;
226
227 /*
228 * If we have encountered a bitmap-format file, the size limit
229 * is smaller than s_maxbytes, which is for extent-mapped files.
230 */
231 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
232 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
233
234 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
235 return -EFBIG;
236 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
237 }
238
239 return iov_iter_count(from);
240 }
241
ext4_write_checks(struct kiocb * iocb,struct iov_iter * from)242 static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
243 {
244 ssize_t ret, count;
245
246 count = ext4_generic_write_checks(iocb, from);
247 if (count <= 0)
248 return count;
249
250 ret = file_modified(iocb->ki_filp);
251 if (ret)
252 return ret;
253 return count;
254 }
255
ext4_buffered_write_iter(struct kiocb * iocb,struct iov_iter * from)256 static ssize_t ext4_buffered_write_iter(struct kiocb *iocb,
257 struct iov_iter *from)
258 {
259 ssize_t ret;
260 struct inode *inode = file_inode(iocb->ki_filp);
261
262 if (iocb->ki_flags & IOCB_NOWAIT)
263 return -EOPNOTSUPP;
264
265 ext4_fc_start_update(inode);
266 inode_lock(inode);
267 ret = ext4_write_checks(iocb, from);
268 if (ret <= 0)
269 goto out;
270
271 current->backing_dev_info = inode_to_bdi(inode);
272 ret = generic_perform_write(iocb->ki_filp, from, iocb->ki_pos);
273 current->backing_dev_info = NULL;
274
275 out:
276 inode_unlock(inode);
277 ext4_fc_stop_update(inode);
278 if (likely(ret > 0)) {
279 iocb->ki_pos += ret;
280 ret = generic_write_sync(iocb, ret);
281 }
282
283 return ret;
284 }
285
ext4_handle_inode_extension(struct inode * inode,loff_t offset,ssize_t written,size_t count)286 static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
287 ssize_t written, size_t count)
288 {
289 handle_t *handle;
290 bool truncate = false;
291 u8 blkbits = inode->i_blkbits;
292 ext4_lblk_t written_blk, end_blk;
293 int ret;
294
295 /*
296 * Note that EXT4_I(inode)->i_disksize can get extended up to
297 * inode->i_size while the I/O was running due to writeback of delalloc
298 * blocks. But, the code in ext4_iomap_alloc() is careful to use
299 * zeroed/unwritten extents if this is possible; thus we won't leave
300 * uninitialized blocks in a file even if we didn't succeed in writing
301 * as much as we intended.
302 */
303 WARN_ON_ONCE(i_size_read(inode) < EXT4_I(inode)->i_disksize);
304 if (offset + count <= EXT4_I(inode)->i_disksize) {
305 /*
306 * We need to ensure that the inode is removed from the orphan
307 * list if it has been added prematurely, due to writeback of
308 * delalloc blocks.
309 */
310 if (!list_empty(&EXT4_I(inode)->i_orphan) && inode->i_nlink) {
311 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
312
313 if (IS_ERR(handle)) {
314 ext4_orphan_del(NULL, inode);
315 return PTR_ERR(handle);
316 }
317
318 ext4_orphan_del(handle, inode);
319 ext4_journal_stop(handle);
320 }
321
322 return written;
323 }
324
325 if (written < 0)
326 goto truncate;
327
328 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
329 if (IS_ERR(handle)) {
330 written = PTR_ERR(handle);
331 goto truncate;
332 }
333
334 if (ext4_update_inode_size(inode, offset + written)) {
335 ret = ext4_mark_inode_dirty(handle, inode);
336 if (unlikely(ret)) {
337 written = ret;
338 ext4_journal_stop(handle);
339 goto truncate;
340 }
341 }
342
343 /*
344 * We may need to truncate allocated but not written blocks beyond EOF.
345 */
346 written_blk = ALIGN(offset + written, 1 << blkbits);
347 end_blk = ALIGN(offset + count, 1 << blkbits);
348 if (written_blk < end_blk && ext4_can_truncate(inode))
349 truncate = true;
350
351 /*
352 * Remove the inode from the orphan list if it has been extended and
353 * everything went OK.
354 */
355 if (!truncate && inode->i_nlink)
356 ext4_orphan_del(handle, inode);
357 ext4_journal_stop(handle);
358
359 if (truncate) {
360 truncate:
361 ext4_truncate_failed_write(inode);
362 /*
363 * If the truncate operation failed early, then the inode may
364 * still be on the orphan list. In that case, we need to try
365 * remove the inode from the in-memory linked list.
366 */
367 if (inode->i_nlink)
368 ext4_orphan_del(NULL, inode);
369 }
370
371 return written;
372 }
373
ext4_dio_write_end_io(struct kiocb * iocb,ssize_t size,int error,unsigned int flags)374 static int ext4_dio_write_end_io(struct kiocb *iocb, ssize_t size,
375 int error, unsigned int flags)
376 {
377 loff_t pos = iocb->ki_pos;
378 struct inode *inode = file_inode(iocb->ki_filp);
379
380 if (error)
381 return error;
382
383 if (size && flags & IOMAP_DIO_UNWRITTEN) {
384 error = ext4_convert_unwritten_extents(NULL, inode, pos, size);
385 if (error < 0)
386 return error;
387 }
388 /*
389 * If we are extending the file, we have to update i_size here before
390 * page cache gets invalidated in iomap_dio_rw(). Otherwise racing
391 * buffered reads could zero out too much from page cache pages. Update
392 * of on-disk size will happen later in ext4_dio_write_iter() where
393 * we have enough information to also perform orphan list handling etc.
394 * Note that we perform all extending writes synchronously under
395 * i_rwsem held exclusively so i_size update is safe here in that case.
396 * If the write was not extending, we cannot see pos > i_size here
397 * because operations reducing i_size like truncate wait for all
398 * outstanding DIO before updating i_size.
399 */
400 pos += size;
401 if (pos > i_size_read(inode))
402 i_size_write(inode, pos);
403
404 return 0;
405 }
406
407 static const struct iomap_dio_ops ext4_dio_write_ops = {
408 .end_io = ext4_dio_write_end_io,
409 };
410
411 /*
412 * The intention here is to start with shared lock acquired then see if any
413 * condition requires an exclusive inode lock. If yes, then we restart the
414 * whole operation by releasing the shared lock and acquiring exclusive lock.
415 *
416 * - For unaligned_io we never take shared lock as it may cause data corruption
417 * when two unaligned IO tries to modify the same block e.g. while zeroing.
418 *
419 * - For extending writes case we don't take the shared lock, since it requires
420 * updating inode i_disksize and/or orphan handling with exclusive lock.
421 *
422 * - shared locking will only be true mostly with overwrites. Otherwise we will
423 * switch to exclusive i_rwsem lock.
424 */
ext4_dio_write_checks(struct kiocb * iocb,struct iov_iter * from,bool * ilock_shared,bool * extend)425 static ssize_t ext4_dio_write_checks(struct kiocb *iocb, struct iov_iter *from,
426 bool *ilock_shared, bool *extend)
427 {
428 struct file *file = iocb->ki_filp;
429 struct inode *inode = file_inode(file);
430 loff_t offset;
431 size_t count;
432 ssize_t ret;
433
434 restart:
435 ret = ext4_generic_write_checks(iocb, from);
436 if (ret <= 0)
437 goto out;
438
439 offset = iocb->ki_pos;
440 count = ret;
441 if (ext4_extending_io(inode, offset, count))
442 *extend = true;
443 /*
444 * Determine whether the IO operation will overwrite allocated
445 * and initialized blocks.
446 * We need exclusive i_rwsem for changing security info
447 * in file_modified().
448 */
449 if (*ilock_shared && (!IS_NOSEC(inode) || *extend ||
450 !ext4_overwrite_io(inode, offset, count))) {
451 if (iocb->ki_flags & IOCB_NOWAIT) {
452 ret = -EAGAIN;
453 goto out;
454 }
455 inode_unlock_shared(inode);
456 *ilock_shared = false;
457 inode_lock(inode);
458 goto restart;
459 }
460
461 ret = file_modified(file);
462 if (ret < 0)
463 goto out;
464
465 return count;
466 out:
467 if (*ilock_shared)
468 inode_unlock_shared(inode);
469 else
470 inode_unlock(inode);
471 return ret;
472 }
473
ext4_dio_write_iter(struct kiocb * iocb,struct iov_iter * from)474 static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from)
475 {
476 ssize_t ret;
477 handle_t *handle;
478 struct inode *inode = file_inode(iocb->ki_filp);
479 loff_t offset = iocb->ki_pos;
480 size_t count = iov_iter_count(from);
481 const struct iomap_ops *iomap_ops = &ext4_iomap_ops;
482 bool extend = false, unaligned_io = false;
483 bool ilock_shared = true;
484
485 /*
486 * We initially start with shared inode lock unless it is
487 * unaligned IO which needs exclusive lock anyways.
488 */
489 if (ext4_unaligned_io(inode, from, offset)) {
490 unaligned_io = true;
491 ilock_shared = false;
492 }
493 /*
494 * Quick check here without any i_rwsem lock to see if it is extending
495 * IO. A more reliable check is done in ext4_dio_write_checks() with
496 * proper locking in place.
497 */
498 if (offset + count > i_size_read(inode))
499 ilock_shared = false;
500
501 if (iocb->ki_flags & IOCB_NOWAIT) {
502 if (ilock_shared) {
503 if (!inode_trylock_shared(inode))
504 return -EAGAIN;
505 } else {
506 if (!inode_trylock(inode))
507 return -EAGAIN;
508 }
509 } else {
510 if (ilock_shared)
511 inode_lock_shared(inode);
512 else
513 inode_lock(inode);
514 }
515
516 /* Fallback to buffered I/O if the inode does not support direct I/O. */
517 if (!ext4_dio_supported(iocb, from)) {
518 if (ilock_shared)
519 inode_unlock_shared(inode);
520 else
521 inode_unlock(inode);
522 return ext4_buffered_write_iter(iocb, from);
523 }
524
525 ret = ext4_dio_write_checks(iocb, from, &ilock_shared, &extend);
526 if (ret <= 0)
527 return ret;
528
529 /* if we're going to block and IOCB_NOWAIT is set, return -EAGAIN */
530 if ((iocb->ki_flags & IOCB_NOWAIT) && (unaligned_io || extend)) {
531 ret = -EAGAIN;
532 goto out;
533 }
534 /*
535 * Make sure inline data cannot be created anymore since we are going
536 * to allocate blocks for DIO. We know the inode does not have any
537 * inline data now because ext4_dio_supported() checked for that.
538 */
539 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
540
541 offset = iocb->ki_pos;
542 count = ret;
543
544 /*
545 * Unaligned direct IO must be serialized among each other as zeroing
546 * of partial blocks of two competing unaligned IOs can result in data
547 * corruption.
548 *
549 * So we make sure we don't allow any unaligned IO in flight.
550 * For IOs where we need not wait (like unaligned non-AIO DIO),
551 * below inode_dio_wait() may anyway become a no-op, since we start
552 * with exclusive lock.
553 */
554 if (unaligned_io)
555 inode_dio_wait(inode);
556
557 if (extend) {
558 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
559 if (IS_ERR(handle)) {
560 ret = PTR_ERR(handle);
561 goto out;
562 }
563
564 ext4_fc_start_update(inode);
565 ret = ext4_orphan_add(handle, inode);
566 ext4_fc_stop_update(inode);
567 if (ret) {
568 ext4_journal_stop(handle);
569 goto out;
570 }
571
572 ext4_journal_stop(handle);
573 }
574
575 if (ilock_shared)
576 iomap_ops = &ext4_iomap_overwrite_ops;
577 ret = iomap_dio_rw(iocb, from, iomap_ops, &ext4_dio_write_ops,
578 is_sync_kiocb(iocb) || unaligned_io || extend);
579 if (ret == -ENOTBLK)
580 ret = 0;
581
582 if (extend)
583 ret = ext4_handle_inode_extension(inode, offset, ret, count);
584
585 out:
586 if (ilock_shared)
587 inode_unlock_shared(inode);
588 else
589 inode_unlock(inode);
590
591 if (ret >= 0 && iov_iter_count(from)) {
592 ssize_t err;
593 loff_t endbyte;
594
595 offset = iocb->ki_pos;
596 err = ext4_buffered_write_iter(iocb, from);
597 if (err < 0)
598 return err;
599
600 /*
601 * We need to ensure that the pages within the page cache for
602 * the range covered by this I/O are written to disk and
603 * invalidated. This is in attempt to preserve the expected
604 * direct I/O semantics in the case we fallback to buffered I/O
605 * to complete off the I/O request.
606 */
607 ret += err;
608 endbyte = offset + err - 1;
609 err = filemap_write_and_wait_range(iocb->ki_filp->f_mapping,
610 offset, endbyte);
611 if (!err)
612 invalidate_mapping_pages(iocb->ki_filp->f_mapping,
613 offset >> PAGE_SHIFT,
614 endbyte >> PAGE_SHIFT);
615 }
616
617 return ret;
618 }
619
620 #ifdef CONFIG_FS_DAX
621 static ssize_t
ext4_dax_write_iter(struct kiocb * iocb,struct iov_iter * from)622 ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
623 {
624 ssize_t ret;
625 size_t count;
626 loff_t offset;
627 handle_t *handle;
628 bool extend = false;
629 struct inode *inode = file_inode(iocb->ki_filp);
630
631 if (iocb->ki_flags & IOCB_NOWAIT) {
632 if (!inode_trylock(inode))
633 return -EAGAIN;
634 } else {
635 inode_lock(inode);
636 }
637
638 ret = ext4_write_checks(iocb, from);
639 if (ret <= 0)
640 goto out;
641
642 offset = iocb->ki_pos;
643 count = iov_iter_count(from);
644
645 if (offset + count > EXT4_I(inode)->i_disksize) {
646 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
647 if (IS_ERR(handle)) {
648 ret = PTR_ERR(handle);
649 goto out;
650 }
651
652 ret = ext4_orphan_add(handle, inode);
653 if (ret) {
654 ext4_journal_stop(handle);
655 goto out;
656 }
657
658 extend = true;
659 ext4_journal_stop(handle);
660 }
661
662 ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
663
664 if (extend)
665 ret = ext4_handle_inode_extension(inode, offset, ret, count);
666 out:
667 inode_unlock(inode);
668 if (ret > 0)
669 ret = generic_write_sync(iocb, ret);
670 return ret;
671 }
672 #endif
673
674 static ssize_t
ext4_file_write_iter(struct kiocb * iocb,struct iov_iter * from)675 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
676 {
677 struct inode *inode = file_inode(iocb->ki_filp);
678
679 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
680 return -EIO;
681
682 #ifdef CONFIG_FS_DAX
683 if (IS_DAX(inode))
684 return ext4_dax_write_iter(iocb, from);
685 #endif
686 if (iocb->ki_flags & IOCB_DIRECT)
687 return ext4_dio_write_iter(iocb, from);
688 else
689 return ext4_buffered_write_iter(iocb, from);
690 }
691
692 #ifdef CONFIG_FS_DAX
ext4_dax_huge_fault(struct vm_fault * vmf,enum page_entry_size pe_size)693 static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf,
694 enum page_entry_size pe_size)
695 {
696 int error = 0;
697 vm_fault_t result;
698 int retries = 0;
699 handle_t *handle = NULL;
700 struct inode *inode = file_inode(vmf->vma->vm_file);
701 struct super_block *sb = inode->i_sb;
702
703 /*
704 * We have to distinguish real writes from writes which will result in a
705 * COW page; COW writes should *not* poke the journal (the file will not
706 * be changed). Doing so would cause unintended failures when mounted
707 * read-only.
708 *
709 * We check for VM_SHARED rather than vmf->cow_page since the latter is
710 * unset for pe_size != PE_SIZE_PTE (i.e. only in do_cow_fault); for
711 * other sizes, dax_iomap_fault will handle splitting / fallback so that
712 * we eventually come back with a COW page.
713 */
714 bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
715 (vmf->vma->vm_flags & VM_SHARED);
716 pfn_t pfn;
717
718 if (write) {
719 sb_start_pagefault(sb);
720 file_update_time(vmf->vma->vm_file);
721 down_read(&EXT4_I(inode)->i_mmap_sem);
722 retry:
723 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
724 EXT4_DATA_TRANS_BLOCKS(sb));
725 if (IS_ERR(handle)) {
726 up_read(&EXT4_I(inode)->i_mmap_sem);
727 sb_end_pagefault(sb);
728 return VM_FAULT_SIGBUS;
729 }
730 } else {
731 down_read(&EXT4_I(inode)->i_mmap_sem);
732 }
733 result = dax_iomap_fault(vmf, pe_size, &pfn, &error, &ext4_iomap_ops);
734 if (write) {
735 ext4_journal_stop(handle);
736
737 if ((result & VM_FAULT_ERROR) && error == -ENOSPC &&
738 ext4_should_retry_alloc(sb, &retries))
739 goto retry;
740 /* Handling synchronous page fault? */
741 if (result & VM_FAULT_NEEDDSYNC)
742 result = dax_finish_sync_fault(vmf, pe_size, pfn);
743 up_read(&EXT4_I(inode)->i_mmap_sem);
744 sb_end_pagefault(sb);
745 } else {
746 up_read(&EXT4_I(inode)->i_mmap_sem);
747 }
748
749 return result;
750 }
751
ext4_dax_fault(struct vm_fault * vmf)752 static vm_fault_t ext4_dax_fault(struct vm_fault *vmf)
753 {
754 return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
755 }
756
757 static const struct vm_operations_struct ext4_dax_vm_ops = {
758 .fault = ext4_dax_fault,
759 .huge_fault = ext4_dax_huge_fault,
760 .page_mkwrite = ext4_dax_fault,
761 .pfn_mkwrite = ext4_dax_fault,
762 };
763 #else
764 #define ext4_dax_vm_ops ext4_file_vm_ops
765 #endif
766
767 static const struct vm_operations_struct ext4_file_vm_ops = {
768 .fault = ext4_filemap_fault,
769 .map_pages = filemap_map_pages,
770 .page_mkwrite = ext4_page_mkwrite,
771 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
772 .allow_speculation = filemap_allow_speculation,
773 #endif
774 };
775
ext4_file_mmap(struct file * file,struct vm_area_struct * vma)776 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
777 {
778 struct inode *inode = file->f_mapping->host;
779 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
780 struct dax_device *dax_dev = sbi->s_daxdev;
781
782 if (unlikely(ext4_forced_shutdown(sbi)))
783 return -EIO;
784
785 /*
786 * We don't support synchronous mappings for non-DAX files and
787 * for DAX files if underneath dax_device is not synchronous.
788 */
789 if (!daxdev_mapping_supported(vma, dax_dev))
790 return -EOPNOTSUPP;
791
792 file_accessed(file);
793 if (IS_DAX(file_inode(file))) {
794 vma->vm_ops = &ext4_dax_vm_ops;
795 vma->vm_flags |= VM_HUGEPAGE;
796 } else {
797 vma->vm_ops = &ext4_file_vm_ops;
798 }
799 return 0;
800 }
801
ext4_sample_last_mounted(struct super_block * sb,struct vfsmount * mnt)802 static int ext4_sample_last_mounted(struct super_block *sb,
803 struct vfsmount *mnt)
804 {
805 struct ext4_sb_info *sbi = EXT4_SB(sb);
806 struct path path;
807 char buf[64], *cp;
808 handle_t *handle;
809 int err;
810
811 if (likely(ext4_test_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED)))
812 return 0;
813
814 if (sb_rdonly(sb) || !sb_start_intwrite_trylock(sb))
815 return 0;
816
817 ext4_set_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED);
818 /*
819 * Sample where the filesystem has been mounted and
820 * store it in the superblock for sysadmin convenience
821 * when trying to sort through large numbers of block
822 * devices or filesystem images.
823 */
824 memset(buf, 0, sizeof(buf));
825 path.mnt = mnt;
826 path.dentry = mnt->mnt_root;
827 cp = d_path(&path, buf, sizeof(buf));
828 err = 0;
829 if (IS_ERR(cp))
830 goto out;
831
832 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
833 err = PTR_ERR(handle);
834 if (IS_ERR(handle))
835 goto out;
836 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
837 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
838 if (err)
839 goto out_journal;
840 strncpy(sbi->s_es->s_last_mounted, cp,
841 sizeof(sbi->s_es->s_last_mounted));
842 ext4_handle_dirty_super(handle, sb);
843 out_journal:
844 ext4_journal_stop(handle);
845 out:
846 sb_end_intwrite(sb);
847 return err;
848 }
849
ext4_file_open(struct inode * inode,struct file * filp)850 static int ext4_file_open(struct inode *inode, struct file *filp)
851 {
852 int ret;
853
854 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
855 return -EIO;
856
857 ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt);
858 if (ret)
859 return ret;
860
861 ret = fscrypt_file_open(inode, filp);
862 if (ret)
863 return ret;
864
865 ret = fsverity_file_open(inode, filp);
866 if (ret)
867 return ret;
868
869 /*
870 * Set up the jbd2_inode if we are opening the inode for
871 * writing and the journal is present
872 */
873 if (filp->f_mode & FMODE_WRITE) {
874 ret = ext4_inode_attach_jinode(inode);
875 if (ret < 0)
876 return ret;
877 }
878
879 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
880 return dquot_file_open(inode, filp);
881 }
882
883 /*
884 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
885 * by calling generic_file_llseek_size() with the appropriate maxbytes
886 * value for each.
887 */
ext4_llseek(struct file * file,loff_t offset,int whence)888 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
889 {
890 struct inode *inode = file->f_mapping->host;
891 loff_t maxbytes;
892
893 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
894 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
895 else
896 maxbytes = inode->i_sb->s_maxbytes;
897
898 switch (whence) {
899 default:
900 return generic_file_llseek_size(file, offset, whence,
901 maxbytes, i_size_read(inode));
902 case SEEK_HOLE:
903 inode_lock_shared(inode);
904 offset = iomap_seek_hole(inode, offset,
905 &ext4_iomap_report_ops);
906 inode_unlock_shared(inode);
907 break;
908 case SEEK_DATA:
909 inode_lock_shared(inode);
910 offset = iomap_seek_data(inode, offset,
911 &ext4_iomap_report_ops);
912 inode_unlock_shared(inode);
913 break;
914 }
915
916 if (offset < 0)
917 return offset;
918 return vfs_setpos(file, offset, maxbytes);
919 }
920
921 const struct file_operations ext4_file_operations = {
922 .llseek = ext4_llseek,
923 .read_iter = ext4_file_read_iter,
924 .write_iter = ext4_file_write_iter,
925 .iopoll = iomap_dio_iopoll,
926 .unlocked_ioctl = ext4_ioctl,
927 #ifdef CONFIG_COMPAT
928 .compat_ioctl = ext4_compat_ioctl,
929 #endif
930 .mmap = ext4_file_mmap,
931 .mmap_supported_flags = MAP_SYNC,
932 .open = ext4_file_open,
933 .release = ext4_release_file,
934 .fsync = ext4_sync_file,
935 .get_unmapped_area = thp_get_unmapped_area,
936 .splice_read = generic_file_splice_read,
937 .splice_write = iter_file_splice_write,
938 .fallocate = ext4_fallocate,
939 };
940
941 const struct inode_operations ext4_file_inode_operations = {
942 .setattr = ext4_setattr,
943 .getattr = ext4_file_getattr,
944 .listxattr = ext4_listxattr,
945 .get_acl = ext4_get_acl,
946 .set_acl = ext4_set_acl,
947 .fiemap = ext4_fiemap,
948 };
949
950