Loading...
1/*
2 * linux/fs/ext4/file.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/file.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * ext4 fs regular file handling primitives
16 *
17 * 64-bit file support on 64-bit platforms by Jakub Jelinek
18 * (jj@sunsite.ms.mff.cuni.cz)
19 */
20
21#include <linux/time.h>
22#include <linux/fs.h>
23#include <linux/mount.h>
24#include <linux/path.h>
25#include <linux/dax.h>
26#include <linux/quotaops.h>
27#include <linux/pagevec.h>
28#include <linux/uio.h>
29#include "ext4.h"
30#include "ext4_jbd2.h"
31#include "xattr.h"
32#include "acl.h"
33
34/*
35 * Called when an inode is released. Note that this is different
36 * from ext4_file_open: open gets called at every open, but release
37 * gets called only when /all/ the files are closed.
38 */
39static int ext4_release_file(struct inode *inode, struct file *filp)
40{
41 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
42 ext4_alloc_da_blocks(inode);
43 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
44 }
45 /* if we are the last writer on the inode, drop the block reservation */
46 if ((filp->f_mode & FMODE_WRITE) &&
47 (atomic_read(&inode->i_writecount) == 1) &&
48 !EXT4_I(inode)->i_reserved_data_blocks)
49 {
50 down_write(&EXT4_I(inode)->i_data_sem);
51 ext4_discard_preallocations(inode);
52 up_write(&EXT4_I(inode)->i_data_sem);
53 }
54 if (is_dx(inode) && filp->private_data)
55 ext4_htree_free_dir_info(filp->private_data);
56
57 return 0;
58}
59
60static void ext4_unwritten_wait(struct inode *inode)
61{
62 wait_queue_head_t *wq = ext4_ioend_wq(inode);
63
64 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
65}
66
67/*
68 * This tests whether the IO in question is block-aligned or not.
69 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
70 * are converted to written only after the IO is complete. Until they are
71 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
72 * it needs to zero out portions of the start and/or end block. If 2 AIO
73 * threads are at work on the same unwritten block, they must be synchronized
74 * or one thread will zero the other's data, causing corruption.
75 */
76static int
77ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
78{
79 struct super_block *sb = inode->i_sb;
80 int blockmask = sb->s_blocksize - 1;
81
82 if (pos >= i_size_read(inode))
83 return 0;
84
85 if ((pos | iov_iter_alignment(from)) & blockmask)
86 return 1;
87
88 return 0;
89}
90
91static ssize_t
92ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
93{
94 struct file *file = iocb->ki_filp;
95 struct inode *inode = file_inode(iocb->ki_filp);
96 struct blk_plug plug;
97 int o_direct = iocb->ki_flags & IOCB_DIRECT;
98 int unaligned_aio = 0;
99 int overwrite = 0;
100 ssize_t ret;
101
102 inode_lock(inode);
103 ret = generic_write_checks(iocb, from);
104 if (ret <= 0)
105 goto out;
106
107 /*
108 * Unaligned direct AIO must be serialized among each other as zeroing
109 * of partial blocks of two competing unaligned AIOs can result in data
110 * corruption.
111 */
112 if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
113 !is_sync_kiocb(iocb) &&
114 ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
115 unaligned_aio = 1;
116 ext4_unwritten_wait(inode);
117 }
118
119 /*
120 * If we have encountered a bitmap-format file, the size limit
121 * is smaller than s_maxbytes, which is for extent-mapped files.
122 */
123 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
124 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
125
126 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes) {
127 ret = -EFBIG;
128 goto out;
129 }
130 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
131 }
132
133 iocb->private = &overwrite;
134 if (o_direct) {
135 size_t length = iov_iter_count(from);
136 loff_t pos = iocb->ki_pos;
137 blk_start_plug(&plug);
138
139 /* check whether we do a DIO overwrite or not */
140 if (ext4_should_dioread_nolock(inode) && !unaligned_aio &&
141 !file->f_mapping->nrpages && pos + length <= i_size_read(inode)) {
142 struct ext4_map_blocks map;
143 unsigned int blkbits = inode->i_blkbits;
144 int err, len;
145
146 map.m_lblk = pos >> blkbits;
147 map.m_len = (EXT4_BLOCK_ALIGN(pos + length, blkbits) >> blkbits)
148 - map.m_lblk;
149 len = map.m_len;
150
151 err = ext4_map_blocks(NULL, inode, &map, 0);
152 /*
153 * 'err==len' means that all of blocks has
154 * been preallocated no matter they are
155 * initialized or not. For excluding
156 * unwritten extents, we need to check
157 * m_flags. There are two conditions that
158 * indicate for initialized extents. 1) If we
159 * hit extent cache, EXT4_MAP_MAPPED flag is
160 * returned; 2) If we do a real lookup,
161 * non-flags are returned. So we should check
162 * these two conditions.
163 */
164 if (err == len && (map.m_flags & EXT4_MAP_MAPPED))
165 overwrite = 1;
166 }
167 }
168
169 ret = __generic_file_write_iter(iocb, from);
170 inode_unlock(inode);
171
172 if (ret > 0) {
173 ssize_t err;
174
175 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
176 if (err < 0)
177 ret = err;
178 }
179 if (o_direct)
180 blk_finish_plug(&plug);
181
182 return ret;
183
184out:
185 inode_unlock(inode);
186 return ret;
187}
188
189#ifdef CONFIG_FS_DAX
190static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
191{
192 int result;
193 handle_t *handle = NULL;
194 struct inode *inode = file_inode(vma->vm_file);
195 struct super_block *sb = inode->i_sb;
196 bool write = vmf->flags & FAULT_FLAG_WRITE;
197
198 if (write) {
199 sb_start_pagefault(sb);
200 file_update_time(vma->vm_file);
201 down_read(&EXT4_I(inode)->i_mmap_sem);
202 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
203 EXT4_DATA_TRANS_BLOCKS(sb));
204 } else
205 down_read(&EXT4_I(inode)->i_mmap_sem);
206
207 if (IS_ERR(handle))
208 result = VM_FAULT_SIGBUS;
209 else
210 result = __dax_fault(vma, vmf, ext4_dax_mmap_get_block, NULL);
211
212 if (write) {
213 if (!IS_ERR(handle))
214 ext4_journal_stop(handle);
215 up_read(&EXT4_I(inode)->i_mmap_sem);
216 sb_end_pagefault(sb);
217 } else
218 up_read(&EXT4_I(inode)->i_mmap_sem);
219
220 return result;
221}
222
223static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
224 pmd_t *pmd, unsigned int flags)
225{
226 int result;
227 handle_t *handle = NULL;
228 struct inode *inode = file_inode(vma->vm_file);
229 struct super_block *sb = inode->i_sb;
230 bool write = flags & FAULT_FLAG_WRITE;
231
232 if (write) {
233 sb_start_pagefault(sb);
234 file_update_time(vma->vm_file);
235 down_read(&EXT4_I(inode)->i_mmap_sem);
236 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
237 ext4_chunk_trans_blocks(inode,
238 PMD_SIZE / PAGE_SIZE));
239 } else
240 down_read(&EXT4_I(inode)->i_mmap_sem);
241
242 if (IS_ERR(handle))
243 result = VM_FAULT_SIGBUS;
244 else
245 result = __dax_pmd_fault(vma, addr, pmd, flags,
246 ext4_dax_mmap_get_block, NULL);
247
248 if (write) {
249 if (!IS_ERR(handle))
250 ext4_journal_stop(handle);
251 up_read(&EXT4_I(inode)->i_mmap_sem);
252 sb_end_pagefault(sb);
253 } else
254 up_read(&EXT4_I(inode)->i_mmap_sem);
255
256 return result;
257}
258
259/*
260 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
261 * handler we check for races agaist truncate. Note that since we cycle through
262 * i_mmap_sem, we are sure that also any hole punching that began before we
263 * were called is finished by now and so if it included part of the file we
264 * are working on, our pte will get unmapped and the check for pte_same() in
265 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
266 * desired.
267 */
268static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
269 struct vm_fault *vmf)
270{
271 struct inode *inode = file_inode(vma->vm_file);
272 struct super_block *sb = inode->i_sb;
273 loff_t size;
274 int ret;
275
276 sb_start_pagefault(sb);
277 file_update_time(vma->vm_file);
278 down_read(&EXT4_I(inode)->i_mmap_sem);
279 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
280 if (vmf->pgoff >= size)
281 ret = VM_FAULT_SIGBUS;
282 else
283 ret = dax_pfn_mkwrite(vma, vmf);
284 up_read(&EXT4_I(inode)->i_mmap_sem);
285 sb_end_pagefault(sb);
286
287 return ret;
288}
289
290static const struct vm_operations_struct ext4_dax_vm_ops = {
291 .fault = ext4_dax_fault,
292 .pmd_fault = ext4_dax_pmd_fault,
293 .page_mkwrite = ext4_dax_fault,
294 .pfn_mkwrite = ext4_dax_pfn_mkwrite,
295};
296#else
297#define ext4_dax_vm_ops ext4_file_vm_ops
298#endif
299
300static const struct vm_operations_struct ext4_file_vm_ops = {
301 .fault = ext4_filemap_fault,
302 .map_pages = filemap_map_pages,
303 .page_mkwrite = ext4_page_mkwrite,
304};
305
306static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
307{
308 struct inode *inode = file->f_mapping->host;
309
310 if (ext4_encrypted_inode(inode)) {
311 int err = ext4_get_encryption_info(inode);
312 if (err)
313 return 0;
314 if (ext4_encryption_info(inode) == NULL)
315 return -ENOKEY;
316 }
317 file_accessed(file);
318 if (IS_DAX(file_inode(file))) {
319 vma->vm_ops = &ext4_dax_vm_ops;
320 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
321 } else {
322 vma->vm_ops = &ext4_file_vm_ops;
323 }
324 return 0;
325}
326
327static int ext4_file_open(struct inode * inode, struct file * filp)
328{
329 struct super_block *sb = inode->i_sb;
330 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
331 struct vfsmount *mnt = filp->f_path.mnt;
332 struct dentry *dir;
333 struct path path;
334 char buf[64], *cp;
335 int ret;
336
337 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
338 !(sb->s_flags & MS_RDONLY))) {
339 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
340 /*
341 * Sample where the filesystem has been mounted and
342 * store it in the superblock for sysadmin convenience
343 * when trying to sort through large numbers of block
344 * devices or filesystem images.
345 */
346 memset(buf, 0, sizeof(buf));
347 path.mnt = mnt;
348 path.dentry = mnt->mnt_root;
349 cp = d_path(&path, buf, sizeof(buf));
350 if (!IS_ERR(cp)) {
351 handle_t *handle;
352 int err;
353
354 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
355 if (IS_ERR(handle))
356 return PTR_ERR(handle);
357 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
358 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
359 if (err) {
360 ext4_journal_stop(handle);
361 return err;
362 }
363 strlcpy(sbi->s_es->s_last_mounted, cp,
364 sizeof(sbi->s_es->s_last_mounted));
365 ext4_handle_dirty_super(handle, sb);
366 ext4_journal_stop(handle);
367 }
368 }
369 if (ext4_encrypted_inode(inode)) {
370 ret = ext4_get_encryption_info(inode);
371 if (ret)
372 return -EACCES;
373 if (ext4_encryption_info(inode) == NULL)
374 return -ENOKEY;
375 }
376
377 dir = dget_parent(file_dentry(filp));
378 if (ext4_encrypted_inode(d_inode(dir)) &&
379 !ext4_is_child_context_consistent_with_parent(d_inode(dir), inode)) {
380 ext4_warning(inode->i_sb,
381 "Inconsistent encryption contexts: %lu/%lu\n",
382 (unsigned long) d_inode(dir)->i_ino,
383 (unsigned long) inode->i_ino);
384 dput(dir);
385 return -EPERM;
386 }
387 dput(dir);
388 /*
389 * Set up the jbd2_inode if we are opening the inode for
390 * writing and the journal is present
391 */
392 if (filp->f_mode & FMODE_WRITE) {
393 ret = ext4_inode_attach_jinode(inode);
394 if (ret < 0)
395 return ret;
396 }
397 return dquot_file_open(inode, filp);
398}
399
400/*
401 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
402 * file rather than ext4_ext_walk_space() because we can introduce
403 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
404 * function. When extent status tree has been fully implemented, it will
405 * track all extent status for a file and we can directly use it to
406 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
407 */
408
409/*
410 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
411 * lookup page cache to check whether or not there has some data between
412 * [startoff, endoff] because, if this range contains an unwritten extent,
413 * we determine this extent as a data or a hole according to whether the
414 * page cache has data or not.
415 */
416static int ext4_find_unwritten_pgoff(struct inode *inode,
417 int whence,
418 ext4_lblk_t end_blk,
419 loff_t *offset)
420{
421 struct pagevec pvec;
422 unsigned int blkbits;
423 pgoff_t index;
424 pgoff_t end;
425 loff_t endoff;
426 loff_t startoff;
427 loff_t lastoff;
428 int found = 0;
429
430 blkbits = inode->i_sb->s_blocksize_bits;
431 startoff = *offset;
432 lastoff = startoff;
433 endoff = (loff_t)end_blk << blkbits;
434
435 index = startoff >> PAGE_SHIFT;
436 end = endoff >> PAGE_SHIFT;
437
438 pagevec_init(&pvec, 0);
439 do {
440 int i, num;
441 unsigned long nr_pages;
442
443 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
444 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
445 (pgoff_t)num);
446 if (nr_pages == 0) {
447 if (whence == SEEK_DATA)
448 break;
449
450 BUG_ON(whence != SEEK_HOLE);
451 /*
452 * If this is the first time to go into the loop and
453 * offset is not beyond the end offset, it will be a
454 * hole at this offset
455 */
456 if (lastoff == startoff || lastoff < endoff)
457 found = 1;
458 break;
459 }
460
461 /*
462 * If this is the first time to go into the loop and
463 * offset is smaller than the first page offset, it will be a
464 * hole at this offset.
465 */
466 if (lastoff == startoff && whence == SEEK_HOLE &&
467 lastoff < page_offset(pvec.pages[0])) {
468 found = 1;
469 break;
470 }
471
472 for (i = 0; i < nr_pages; i++) {
473 struct page *page = pvec.pages[i];
474 struct buffer_head *bh, *head;
475
476 /*
477 * If the current offset is not beyond the end of given
478 * range, it will be a hole.
479 */
480 if (lastoff < endoff && whence == SEEK_HOLE &&
481 page->index > end) {
482 found = 1;
483 *offset = lastoff;
484 goto out;
485 }
486
487 lock_page(page);
488
489 if (unlikely(page->mapping != inode->i_mapping)) {
490 unlock_page(page);
491 continue;
492 }
493
494 if (!page_has_buffers(page)) {
495 unlock_page(page);
496 continue;
497 }
498
499 if (page_has_buffers(page)) {
500 lastoff = page_offset(page);
501 bh = head = page_buffers(page);
502 do {
503 if (buffer_uptodate(bh) ||
504 buffer_unwritten(bh)) {
505 if (whence == SEEK_DATA)
506 found = 1;
507 } else {
508 if (whence == SEEK_HOLE)
509 found = 1;
510 }
511 if (found) {
512 *offset = max_t(loff_t,
513 startoff, lastoff);
514 unlock_page(page);
515 goto out;
516 }
517 lastoff += bh->b_size;
518 bh = bh->b_this_page;
519 } while (bh != head);
520 }
521
522 lastoff = page_offset(page) + PAGE_SIZE;
523 unlock_page(page);
524 }
525
526 /*
527 * The no. of pages is less than our desired, that would be a
528 * hole in there.
529 */
530 if (nr_pages < num && whence == SEEK_HOLE) {
531 found = 1;
532 *offset = lastoff;
533 break;
534 }
535
536 index = pvec.pages[i - 1]->index + 1;
537 pagevec_release(&pvec);
538 } while (index <= end);
539
540out:
541 pagevec_release(&pvec);
542 return found;
543}
544
545/*
546 * ext4_seek_data() retrieves the offset for SEEK_DATA.
547 */
548static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
549{
550 struct inode *inode = file->f_mapping->host;
551 struct extent_status es;
552 ext4_lblk_t start, last, end;
553 loff_t dataoff, isize;
554 int blkbits;
555 int ret;
556
557 inode_lock(inode);
558
559 isize = i_size_read(inode);
560 if (offset >= isize) {
561 inode_unlock(inode);
562 return -ENXIO;
563 }
564
565 blkbits = inode->i_sb->s_blocksize_bits;
566 start = offset >> blkbits;
567 last = start;
568 end = isize >> blkbits;
569 dataoff = offset;
570
571 do {
572 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
573 if (ret <= 0) {
574 /* No extent found -> no data */
575 if (ret == 0)
576 ret = -ENXIO;
577 inode_unlock(inode);
578 return ret;
579 }
580
581 last = es.es_lblk;
582 if (last != start)
583 dataoff = (loff_t)last << blkbits;
584 if (!ext4_es_is_unwritten(&es))
585 break;
586
587 /*
588 * If there is a unwritten extent at this offset,
589 * it will be as a data or a hole according to page
590 * cache that has data or not.
591 */
592 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
593 es.es_lblk + es.es_len, &dataoff))
594 break;
595 last += es.es_len;
596 dataoff = (loff_t)last << blkbits;
597 cond_resched();
598 } while (last <= end);
599
600 inode_unlock(inode);
601
602 if (dataoff > isize)
603 return -ENXIO;
604
605 return vfs_setpos(file, dataoff, maxsize);
606}
607
608/*
609 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
610 */
611static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
612{
613 struct inode *inode = file->f_mapping->host;
614 struct extent_status es;
615 ext4_lblk_t start, last, end;
616 loff_t holeoff, isize;
617 int blkbits;
618 int ret;
619
620 inode_lock(inode);
621
622 isize = i_size_read(inode);
623 if (offset >= isize) {
624 inode_unlock(inode);
625 return -ENXIO;
626 }
627
628 blkbits = inode->i_sb->s_blocksize_bits;
629 start = offset >> blkbits;
630 last = start;
631 end = isize >> blkbits;
632 holeoff = offset;
633
634 do {
635 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
636 if (ret < 0) {
637 inode_unlock(inode);
638 return ret;
639 }
640 /* Found a hole? */
641 if (ret == 0 || es.es_lblk > last) {
642 if (last != start)
643 holeoff = (loff_t)last << blkbits;
644 break;
645 }
646 /*
647 * If there is a unwritten extent at this offset,
648 * it will be as a data or a hole according to page
649 * cache that has data or not.
650 */
651 if (ext4_es_is_unwritten(&es) &&
652 ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
653 last + es.es_len, &holeoff))
654 break;
655
656 last += es.es_len;
657 holeoff = (loff_t)last << blkbits;
658 cond_resched();
659 } while (last <= end);
660
661 inode_unlock(inode);
662
663 if (holeoff > isize)
664 holeoff = isize;
665
666 return vfs_setpos(file, holeoff, maxsize);
667}
668
669/*
670 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
671 * by calling generic_file_llseek_size() with the appropriate maxbytes
672 * value for each.
673 */
674loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
675{
676 struct inode *inode = file->f_mapping->host;
677 loff_t maxbytes;
678
679 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
680 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
681 else
682 maxbytes = inode->i_sb->s_maxbytes;
683
684 switch (whence) {
685 case SEEK_SET:
686 case SEEK_CUR:
687 case SEEK_END:
688 return generic_file_llseek_size(file, offset, whence,
689 maxbytes, i_size_read(inode));
690 case SEEK_DATA:
691 return ext4_seek_data(file, offset, maxbytes);
692 case SEEK_HOLE:
693 return ext4_seek_hole(file, offset, maxbytes);
694 }
695
696 return -EINVAL;
697}
698
699const struct file_operations ext4_file_operations = {
700 .llseek = ext4_llseek,
701 .read_iter = generic_file_read_iter,
702 .write_iter = ext4_file_write_iter,
703 .unlocked_ioctl = ext4_ioctl,
704#ifdef CONFIG_COMPAT
705 .compat_ioctl = ext4_compat_ioctl,
706#endif
707 .mmap = ext4_file_mmap,
708 .open = ext4_file_open,
709 .release = ext4_release_file,
710 .fsync = ext4_sync_file,
711 .splice_read = generic_file_splice_read,
712 .splice_write = iter_file_splice_write,
713 .fallocate = ext4_fallocate,
714};
715
716const struct inode_operations ext4_file_inode_operations = {
717 .setattr = ext4_setattr,
718 .getattr = ext4_getattr,
719 .setxattr = generic_setxattr,
720 .getxattr = generic_getxattr,
721 .listxattr = ext4_listxattr,
722 .removexattr = generic_removexattr,
723 .get_acl = ext4_get_acl,
724 .set_acl = ext4_set_acl,
725 .fiemap = ext4_fiemap,
726};
727
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
39static 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
52static 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);
78 inode_unlock_shared(inode);
79
80 file_accessed(iocb->ki_filp);
81 return ret;
82}
83
84#ifdef CONFIG_FS_DAX
85static 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
113static 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 */
138static 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 */
167static bool
168ext4_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
179static bool
180ext4_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? */
189static 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
211static 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
239static 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
253static 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
272out:
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
283static 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) {
357truncate:
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
371static 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
404static 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 */
422static 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
431restart:
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;
463out:
464 if (*ilock_shared)
465 inode_unlock_shared(inode);
466 else
467 inode_unlock(inode);
468 return ret;
469}
470
471static 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 if (ret == -ENOTBLK)
571 ret = 0;
572
573 if (extend)
574 ret = ext4_handle_inode_extension(inode, offset, ret, count);
575
576out:
577 if (ilock_shared)
578 inode_unlock_shared(inode);
579 else
580 inode_unlock(inode);
581
582 if (ret >= 0 && iov_iter_count(from)) {
583 ssize_t err;
584 loff_t endbyte;
585
586 offset = iocb->ki_pos;
587 err = ext4_buffered_write_iter(iocb, from);
588 if (err < 0)
589 return err;
590
591 /*
592 * We need to ensure that the pages within the page cache for
593 * the range covered by this I/O are written to disk and
594 * invalidated. This is in attempt to preserve the expected
595 * direct I/O semantics in the case we fallback to buffered I/O
596 * to complete off the I/O request.
597 */
598 ret += err;
599 endbyte = offset + err - 1;
600 err = filemap_write_and_wait_range(iocb->ki_filp->f_mapping,
601 offset, endbyte);
602 if (!err)
603 invalidate_mapping_pages(iocb->ki_filp->f_mapping,
604 offset >> PAGE_SHIFT,
605 endbyte >> PAGE_SHIFT);
606 }
607
608 return ret;
609}
610
611#ifdef CONFIG_FS_DAX
612static ssize_t
613ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
614{
615 ssize_t ret;
616 size_t count;
617 loff_t offset;
618 handle_t *handle;
619 bool extend = false;
620 struct inode *inode = file_inode(iocb->ki_filp);
621
622 if (iocb->ki_flags & IOCB_NOWAIT) {
623 if (!inode_trylock(inode))
624 return -EAGAIN;
625 } else {
626 inode_lock(inode);
627 }
628
629 ret = ext4_write_checks(iocb, from);
630 if (ret <= 0)
631 goto out;
632
633 offset = iocb->ki_pos;
634 count = iov_iter_count(from);
635
636 if (offset + count > EXT4_I(inode)->i_disksize) {
637 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
638 if (IS_ERR(handle)) {
639 ret = PTR_ERR(handle);
640 goto out;
641 }
642
643 ret = ext4_orphan_add(handle, inode);
644 if (ret) {
645 ext4_journal_stop(handle);
646 goto out;
647 }
648
649 extend = true;
650 ext4_journal_stop(handle);
651 }
652
653 ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
654
655 if (extend)
656 ret = ext4_handle_inode_extension(inode, offset, ret, count);
657out:
658 inode_unlock(inode);
659 if (ret > 0)
660 ret = generic_write_sync(iocb, ret);
661 return ret;
662}
663#endif
664
665static ssize_t
666ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
667{
668 struct inode *inode = file_inode(iocb->ki_filp);
669
670 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
671 return -EIO;
672
673#ifdef CONFIG_FS_DAX
674 if (IS_DAX(inode))
675 return ext4_dax_write_iter(iocb, from);
676#endif
677 if (iocb->ki_flags & IOCB_DIRECT)
678 return ext4_dio_write_iter(iocb, from);
679 else
680 return ext4_buffered_write_iter(iocb, from);
681}
682
683#ifdef CONFIG_FS_DAX
684static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf,
685 enum page_entry_size pe_size)
686{
687 int error = 0;
688 vm_fault_t result;
689 int retries = 0;
690 handle_t *handle = NULL;
691 struct inode *inode = file_inode(vmf->vma->vm_file);
692 struct super_block *sb = inode->i_sb;
693
694 /*
695 * We have to distinguish real writes from writes which will result in a
696 * COW page; COW writes should *not* poke the journal (the file will not
697 * be changed). Doing so would cause unintended failures when mounted
698 * read-only.
699 *
700 * We check for VM_SHARED rather than vmf->cow_page since the latter is
701 * unset for pe_size != PE_SIZE_PTE (i.e. only in do_cow_fault); for
702 * other sizes, dax_iomap_fault will handle splitting / fallback so that
703 * we eventually come back with a COW page.
704 */
705 bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
706 (vmf->vma->vm_flags & VM_SHARED);
707 pfn_t pfn;
708
709 if (write) {
710 sb_start_pagefault(sb);
711 file_update_time(vmf->vma->vm_file);
712 down_read(&EXT4_I(inode)->i_mmap_sem);
713retry:
714 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
715 EXT4_DATA_TRANS_BLOCKS(sb));
716 if (IS_ERR(handle)) {
717 up_read(&EXT4_I(inode)->i_mmap_sem);
718 sb_end_pagefault(sb);
719 return VM_FAULT_SIGBUS;
720 }
721 } else {
722 down_read(&EXT4_I(inode)->i_mmap_sem);
723 }
724 result = dax_iomap_fault(vmf, pe_size, &pfn, &error, &ext4_iomap_ops);
725 if (write) {
726 ext4_journal_stop(handle);
727
728 if ((result & VM_FAULT_ERROR) && error == -ENOSPC &&
729 ext4_should_retry_alloc(sb, &retries))
730 goto retry;
731 /* Handling synchronous page fault? */
732 if (result & VM_FAULT_NEEDDSYNC)
733 result = dax_finish_sync_fault(vmf, pe_size, pfn);
734 up_read(&EXT4_I(inode)->i_mmap_sem);
735 sb_end_pagefault(sb);
736 } else {
737 up_read(&EXT4_I(inode)->i_mmap_sem);
738 }
739
740 return result;
741}
742
743static vm_fault_t ext4_dax_fault(struct vm_fault *vmf)
744{
745 return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
746}
747
748static const struct vm_operations_struct ext4_dax_vm_ops = {
749 .fault = ext4_dax_fault,
750 .huge_fault = ext4_dax_huge_fault,
751 .page_mkwrite = ext4_dax_fault,
752 .pfn_mkwrite = ext4_dax_fault,
753};
754#else
755#define ext4_dax_vm_ops ext4_file_vm_ops
756#endif
757
758static const struct vm_operations_struct ext4_file_vm_ops = {
759 .fault = ext4_filemap_fault,
760 .map_pages = filemap_map_pages,
761 .page_mkwrite = ext4_page_mkwrite,
762};
763
764static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
765{
766 struct inode *inode = file->f_mapping->host;
767 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
768 struct dax_device *dax_dev = sbi->s_daxdev;
769
770 if (unlikely(ext4_forced_shutdown(sbi)))
771 return -EIO;
772
773 /*
774 * We don't support synchronous mappings for non-DAX files and
775 * for DAX files if underneath dax_device is not synchronous.
776 */
777 if (!daxdev_mapping_supported(vma, dax_dev))
778 return -EOPNOTSUPP;
779
780 file_accessed(file);
781 if (IS_DAX(file_inode(file))) {
782 vma->vm_ops = &ext4_dax_vm_ops;
783 vma->vm_flags |= VM_HUGEPAGE;
784 } else {
785 vma->vm_ops = &ext4_file_vm_ops;
786 }
787 return 0;
788}
789
790static int ext4_sample_last_mounted(struct super_block *sb,
791 struct vfsmount *mnt)
792{
793 struct ext4_sb_info *sbi = EXT4_SB(sb);
794 struct path path;
795 char buf[64], *cp;
796 handle_t *handle;
797 int err;
798
799 if (likely(ext4_test_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED)))
800 return 0;
801
802 if (sb_rdonly(sb) || !sb_start_intwrite_trylock(sb))
803 return 0;
804
805 ext4_set_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED);
806 /*
807 * Sample where the filesystem has been mounted and
808 * store it in the superblock for sysadmin convenience
809 * when trying to sort through large numbers of block
810 * devices or filesystem images.
811 */
812 memset(buf, 0, sizeof(buf));
813 path.mnt = mnt;
814 path.dentry = mnt->mnt_root;
815 cp = d_path(&path, buf, sizeof(buf));
816 err = 0;
817 if (IS_ERR(cp))
818 goto out;
819
820 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
821 err = PTR_ERR(handle);
822 if (IS_ERR(handle))
823 goto out;
824 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
825 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
826 if (err)
827 goto out_journal;
828 lock_buffer(sbi->s_sbh);
829 strncpy(sbi->s_es->s_last_mounted, cp,
830 sizeof(sbi->s_es->s_last_mounted));
831 ext4_superblock_csum_set(sb);
832 unlock_buffer(sbi->s_sbh);
833 ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
834out_journal:
835 ext4_journal_stop(handle);
836out:
837 sb_end_intwrite(sb);
838 return err;
839}
840
841static int ext4_file_open(struct inode *inode, struct file *filp)
842{
843 int ret;
844
845 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
846 return -EIO;
847
848 ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt);
849 if (ret)
850 return ret;
851
852 ret = fscrypt_file_open(inode, filp);
853 if (ret)
854 return ret;
855
856 ret = fsverity_file_open(inode, filp);
857 if (ret)
858 return ret;
859
860 /*
861 * Set up the jbd2_inode if we are opening the inode for
862 * writing and the journal is present
863 */
864 if (filp->f_mode & FMODE_WRITE) {
865 ret = ext4_inode_attach_jinode(inode);
866 if (ret < 0)
867 return ret;
868 }
869
870 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
871 return dquot_file_open(inode, filp);
872}
873
874/*
875 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
876 * by calling generic_file_llseek_size() with the appropriate maxbytes
877 * value for each.
878 */
879loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
880{
881 struct inode *inode = file->f_mapping->host;
882 loff_t maxbytes;
883
884 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
885 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
886 else
887 maxbytes = inode->i_sb->s_maxbytes;
888
889 switch (whence) {
890 default:
891 return generic_file_llseek_size(file, offset, whence,
892 maxbytes, i_size_read(inode));
893 case SEEK_HOLE:
894 inode_lock_shared(inode);
895 offset = iomap_seek_hole(inode, offset,
896 &ext4_iomap_report_ops);
897 inode_unlock_shared(inode);
898 break;
899 case SEEK_DATA:
900 inode_lock_shared(inode);
901 offset = iomap_seek_data(inode, offset,
902 &ext4_iomap_report_ops);
903 inode_unlock_shared(inode);
904 break;
905 }
906
907 if (offset < 0)
908 return offset;
909 return vfs_setpos(file, offset, maxbytes);
910}
911
912const struct file_operations ext4_file_operations = {
913 .llseek = ext4_llseek,
914 .read_iter = ext4_file_read_iter,
915 .write_iter = ext4_file_write_iter,
916 .iopoll = iomap_dio_iopoll,
917 .unlocked_ioctl = ext4_ioctl,
918#ifdef CONFIG_COMPAT
919 .compat_ioctl = ext4_compat_ioctl,
920#endif
921 .mmap = ext4_file_mmap,
922 .mmap_supported_flags = MAP_SYNC,
923 .open = ext4_file_open,
924 .release = ext4_release_file,
925 .fsync = ext4_sync_file,
926 .get_unmapped_area = thp_get_unmapped_area,
927 .splice_read = generic_file_splice_read,
928 .splice_write = iter_file_splice_write,
929 .fallocate = ext4_fallocate,
930};
931
932const struct inode_operations ext4_file_inode_operations = {
933 .setattr = ext4_setattr,
934 .getattr = ext4_file_getattr,
935 .listxattr = ext4_listxattr,
936 .get_acl = ext4_get_acl,
937 .set_acl = ext4_set_acl,
938 .fiemap = ext4_fiemap,
939 .fileattr_get = ext4_fileattr_get,
940 .fileattr_set = ext4_fileattr_set,
941};
942