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
39/*
40 * Returns %true if the given DIO request should be attempted with DIO, or
41 * %false if it should fall back to buffered I/O.
42 *
43 * DIO isn't well specified; when it's unsupported (either due to the request
44 * being misaligned, or due to the file not supporting DIO at all), filesystems
45 * either fall back to buffered I/O or return EINVAL. For files that don't use
46 * any special features like encryption or verity, ext4 has traditionally
47 * returned EINVAL for misaligned DIO. iomap_dio_rw() uses this convention too.
48 * In this case, we should attempt the DIO, *not* fall back to buffered I/O.
49 *
50 * In contrast, in cases where DIO is unsupported due to ext4 features, ext4
51 * traditionally falls back to buffered I/O.
52 *
53 * This function implements the traditional ext4 behavior in all these cases.
54 */
55static bool ext4_should_use_dio(struct kiocb *iocb, struct iov_iter *iter)
56{
57 struct inode *inode = file_inode(iocb->ki_filp);
58 u32 dio_align = ext4_dio_alignment(inode);
59
60 if (dio_align == 0)
61 return false;
62
63 if (dio_align == 1)
64 return true;
65
66 return IS_ALIGNED(iocb->ki_pos | iov_iter_alignment(iter), dio_align);
67}
68
69static ssize_t ext4_dio_read_iter(struct kiocb *iocb, struct iov_iter *to)
70{
71 ssize_t ret;
72 struct inode *inode = file_inode(iocb->ki_filp);
73
74 if (iocb->ki_flags & IOCB_NOWAIT) {
75 if (!inode_trylock_shared(inode))
76 return -EAGAIN;
77 } else {
78 inode_lock_shared(inode);
79 }
80
81 if (!ext4_should_use_dio(iocb, to)) {
82 inode_unlock_shared(inode);
83 /*
84 * Fallback to buffered I/O if the operation being performed on
85 * the inode is not supported by direct I/O. The IOCB_DIRECT
86 * flag needs to be cleared here in order to ensure that the
87 * direct I/O path within generic_file_read_iter() is not
88 * taken.
89 */
90 iocb->ki_flags &= ~IOCB_DIRECT;
91 return generic_file_read_iter(iocb, to);
92 }
93
94 ret = iomap_dio_rw(iocb, to, &ext4_iomap_ops, NULL, 0, NULL, 0);
95 inode_unlock_shared(inode);
96
97 file_accessed(iocb->ki_filp);
98 return ret;
99}
100
101#ifdef CONFIG_FS_DAX
102static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
103{
104 struct inode *inode = file_inode(iocb->ki_filp);
105 ssize_t ret;
106
107 if (iocb->ki_flags & IOCB_NOWAIT) {
108 if (!inode_trylock_shared(inode))
109 return -EAGAIN;
110 } else {
111 inode_lock_shared(inode);
112 }
113 /*
114 * Recheck under inode lock - at this point we are sure it cannot
115 * change anymore
116 */
117 if (!IS_DAX(inode)) {
118 inode_unlock_shared(inode);
119 /* Fallback to buffered IO in case we cannot support DAX */
120 return generic_file_read_iter(iocb, to);
121 }
122 ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
123 inode_unlock_shared(inode);
124
125 file_accessed(iocb->ki_filp);
126 return ret;
127}
128#endif
129
130static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
131{
132 struct inode *inode = file_inode(iocb->ki_filp);
133
134 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
135 return -EIO;
136
137 if (!iov_iter_count(to))
138 return 0; /* skip atime */
139
140#ifdef CONFIG_FS_DAX
141 if (IS_DAX(inode))
142 return ext4_dax_read_iter(iocb, to);
143#endif
144 if (iocb->ki_flags & IOCB_DIRECT)
145 return ext4_dio_read_iter(iocb, to);
146
147 return generic_file_read_iter(iocb, to);
148}
149
150static ssize_t ext4_file_splice_read(struct file *in, loff_t *ppos,
151 struct pipe_inode_info *pipe,
152 size_t len, unsigned int flags)
153{
154 struct inode *inode = file_inode(in);
155
156 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
157 return -EIO;
158 return filemap_splice_read(in, ppos, pipe, len, flags);
159}
160
161/*
162 * Called when an inode is released. Note that this is different
163 * from ext4_file_open: open gets called at every open, but release
164 * gets called only when /all/ the files are closed.
165 */
166static int ext4_release_file(struct inode *inode, struct file *filp)
167{
168 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
169 ext4_alloc_da_blocks(inode);
170 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
171 }
172 /* if we are the last writer on the inode, drop the block reservation */
173 if ((filp->f_mode & FMODE_WRITE) &&
174 (atomic_read(&inode->i_writecount) == 1) &&
175 !EXT4_I(inode)->i_reserved_data_blocks) {
176 down_write(&EXT4_I(inode)->i_data_sem);
177 ext4_discard_preallocations(inode);
178 up_write(&EXT4_I(inode)->i_data_sem);
179 }
180 if (is_dx(inode) && filp->private_data)
181 ext4_htree_free_dir_info(filp->private_data);
182
183 return 0;
184}
185
186/*
187 * This tests whether the IO in question is block-aligned or not.
188 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
189 * are converted to written only after the IO is complete. Until they are
190 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
191 * it needs to zero out portions of the start and/or end block. If 2 AIO
192 * threads are at work on the same unwritten block, they must be synchronized
193 * or one thread will zero the other's data, causing corruption.
194 */
195static bool
196ext4_unaligned_io(struct inode *inode, struct iov_iter *from, loff_t pos)
197{
198 struct super_block *sb = inode->i_sb;
199 unsigned long blockmask = sb->s_blocksize - 1;
200
201 if ((pos | iov_iter_alignment(from)) & blockmask)
202 return true;
203
204 return false;
205}
206
207static bool
208ext4_extending_io(struct inode *inode, loff_t offset, size_t len)
209{
210 if (offset + len > i_size_read(inode) ||
211 offset + len > EXT4_I(inode)->i_disksize)
212 return true;
213 return false;
214}
215
216/* Is IO overwriting allocated or initialized blocks? */
217static bool ext4_overwrite_io(struct inode *inode,
218 loff_t pos, loff_t len, bool *unwritten)
219{
220 struct ext4_map_blocks map;
221 unsigned int blkbits = inode->i_blkbits;
222 int err, blklen;
223
224 if (pos + len > i_size_read(inode))
225 return false;
226
227 map.m_lblk = pos >> blkbits;
228 map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
229 blklen = map.m_len;
230
231 err = ext4_map_blocks(NULL, inode, &map, 0);
232 if (err != blklen)
233 return false;
234 /*
235 * 'err==len' means that all of the blocks have been preallocated,
236 * regardless of whether they have been initialized or not. We need to
237 * check m_flags to distinguish the unwritten extents.
238 */
239 *unwritten = !(map.m_flags & EXT4_MAP_MAPPED);
240 return true;
241}
242
243static ssize_t ext4_generic_write_checks(struct kiocb *iocb,
244 struct iov_iter *from)
245{
246 struct inode *inode = file_inode(iocb->ki_filp);
247 ssize_t ret;
248
249 if (unlikely(IS_IMMUTABLE(inode)))
250 return -EPERM;
251
252 ret = generic_write_checks(iocb, from);
253 if (ret <= 0)
254 return ret;
255
256 /*
257 * If we have encountered a bitmap-format file, the size limit
258 * is smaller than s_maxbytes, which is for extent-mapped files.
259 */
260 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
261 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
262
263 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
264 return -EFBIG;
265 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
266 }
267
268 return iov_iter_count(from);
269}
270
271static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
272{
273 ssize_t ret, count;
274
275 count = ext4_generic_write_checks(iocb, from);
276 if (count <= 0)
277 return count;
278
279 ret = file_modified(iocb->ki_filp);
280 if (ret)
281 return ret;
282 return count;
283}
284
285static ssize_t ext4_buffered_write_iter(struct kiocb *iocb,
286 struct iov_iter *from)
287{
288 ssize_t ret;
289 struct inode *inode = file_inode(iocb->ki_filp);
290
291 if (iocb->ki_flags & IOCB_NOWAIT)
292 return -EOPNOTSUPP;
293
294 inode_lock(inode);
295 ret = ext4_write_checks(iocb, from);
296 if (ret <= 0)
297 goto out;
298
299 ret = generic_perform_write(iocb, from);
300
301out:
302 inode_unlock(inode);
303 if (unlikely(ret <= 0))
304 return ret;
305 return generic_write_sync(iocb, ret);
306}
307
308static ssize_t ext4_handle_inode_extension(struct inode *inode, loff_t offset,
309 ssize_t written, ssize_t count)
310{
311 handle_t *handle;
312
313 lockdep_assert_held_write(&inode->i_rwsem);
314 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
315 if (IS_ERR(handle))
316 return PTR_ERR(handle);
317
318 if (ext4_update_inode_size(inode, offset + written)) {
319 int ret = ext4_mark_inode_dirty(handle, inode);
320 if (unlikely(ret)) {
321 ext4_journal_stop(handle);
322 return ret;
323 }
324 }
325
326 if ((written == count) && inode->i_nlink)
327 ext4_orphan_del(handle, inode);
328 ext4_journal_stop(handle);
329
330 return written;
331}
332
333/*
334 * Clean up the inode after DIO or DAX extending write has completed and the
335 * inode size has been updated using ext4_handle_inode_extension().
336 */
337static void ext4_inode_extension_cleanup(struct inode *inode, bool need_trunc)
338{
339 lockdep_assert_held_write(&inode->i_rwsem);
340 if (need_trunc) {
341 ext4_truncate_failed_write(inode);
342 /*
343 * If the truncate operation failed early, then the inode may
344 * still be on the orphan list. In that case, we need to try
345 * remove the inode from the in-memory linked list.
346 */
347 if (inode->i_nlink)
348 ext4_orphan_del(NULL, inode);
349 return;
350 }
351 /*
352 * If i_disksize got extended either due to writeback of delalloc
353 * blocks or extending truncate while the DIO was running we could fail
354 * to cleanup the orphan list in ext4_handle_inode_extension(). Do it
355 * now.
356 */
357 if (!list_empty(&EXT4_I(inode)->i_orphan) && inode->i_nlink) {
358 handle_t *handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
359
360 if (IS_ERR(handle)) {
361 /*
362 * The write has successfully completed. Not much to
363 * do with the error here so just cleanup the orphan
364 * list and hope for the best.
365 */
366 ext4_orphan_del(NULL, inode);
367 return;
368 }
369 ext4_orphan_del(handle, inode);
370 ext4_journal_stop(handle);
371 }
372}
373
374static 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 && size && flags & IOMAP_DIO_UNWRITTEN)
381 error = ext4_convert_unwritten_extents(NULL, inode, pos, size);
382 if (error)
383 return error;
384 /*
385 * Note that EXT4_I(inode)->i_disksize can get extended up to
386 * inode->i_size while the I/O was running due to writeback of delalloc
387 * blocks. But the code in ext4_iomap_alloc() is careful to use
388 * zeroed/unwritten extents if this is possible; thus we won't leave
389 * uninitialized blocks in a file even if we didn't succeed in writing
390 * as much as we intended. Also we can race with truncate or write
391 * expanding the file so we have to be a bit careful here.
392 */
393 if (pos + size <= READ_ONCE(EXT4_I(inode)->i_disksize) &&
394 pos + size <= i_size_read(inode))
395 return 0;
396 error = ext4_handle_inode_extension(inode, pos, size, size);
397 return error < 0 ? error : 0;
398}
399
400static const struct iomap_dio_ops ext4_dio_write_ops = {
401 .end_io = ext4_dio_write_end_io,
402};
403
404/*
405 * The intention here is to start with shared lock acquired then see if any
406 * condition requires an exclusive inode lock. If yes, then we restart the
407 * whole operation by releasing the shared lock and acquiring exclusive lock.
408 *
409 * - For unaligned_io we never take shared lock as it may cause data corruption
410 * when two unaligned IO tries to modify the same block e.g. while zeroing.
411 *
412 * - For extending writes case we don't take the shared lock, since it requires
413 * updating inode i_disksize and/or orphan handling with exclusive lock.
414 *
415 * - shared locking will only be true mostly with overwrites, including
416 * initialized blocks and unwritten blocks. For overwrite unwritten blocks
417 * we protect splitting extents by i_data_sem in ext4_inode_info, so we can
418 * also release exclusive i_rwsem lock.
419 *
420 * - Otherwise we will 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 bool *unwritten, int *dio_flags)
425{
426 struct file *file = iocb->ki_filp;
427 struct inode *inode = file_inode(file);
428 loff_t offset;
429 size_t count;
430 ssize_t ret;
431 bool overwrite, unaligned_io;
432
433restart:
434 ret = ext4_generic_write_checks(iocb, from);
435 if (ret <= 0)
436 goto out;
437
438 offset = iocb->ki_pos;
439 count = ret;
440
441 unaligned_io = ext4_unaligned_io(inode, from, offset);
442 *extend = ext4_extending_io(inode, offset, count);
443 overwrite = ext4_overwrite_io(inode, offset, count, unwritten);
444
445 /*
446 * Determine whether we need to upgrade to an exclusive lock. This is
447 * required to change security info in file_modified(), for extending
448 * I/O, any form of non-overwrite I/O, and unaligned I/O to unwritten
449 * extents (as partial block zeroing may be required).
450 *
451 * Note that unaligned writes are allowed under shared lock so long as
452 * they are pure overwrites. Otherwise, concurrent unaligned writes risk
453 * data corruption due to partial block zeroing in the dio layer, and so
454 * the I/O must occur exclusively.
455 */
456 if (*ilock_shared &&
457 ((!IS_NOSEC(inode) || *extend || !overwrite ||
458 (unaligned_io && *unwritten)))) {
459 if (iocb->ki_flags & IOCB_NOWAIT) {
460 ret = -EAGAIN;
461 goto out;
462 }
463 inode_unlock_shared(inode);
464 *ilock_shared = false;
465 inode_lock(inode);
466 goto restart;
467 }
468
469 /*
470 * Now that locking is settled, determine dio flags and exclusivity
471 * requirements. We don't use DIO_OVERWRITE_ONLY because we enforce
472 * behavior already. The inode lock is already held exclusive if the
473 * write is non-overwrite or extending, so drain all outstanding dio and
474 * set the force wait dio flag.
475 */
476 if (!*ilock_shared && (unaligned_io || *extend)) {
477 if (iocb->ki_flags & IOCB_NOWAIT) {
478 ret = -EAGAIN;
479 goto out;
480 }
481 if (unaligned_io && (!overwrite || *unwritten))
482 inode_dio_wait(inode);
483 *dio_flags = IOMAP_DIO_FORCE_WAIT;
484 }
485
486 ret = file_modified(file);
487 if (ret < 0)
488 goto out;
489
490 return count;
491out:
492 if (*ilock_shared)
493 inode_unlock_shared(inode);
494 else
495 inode_unlock(inode);
496 return ret;
497}
498
499static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from)
500{
501 ssize_t ret;
502 handle_t *handle;
503 struct inode *inode = file_inode(iocb->ki_filp);
504 loff_t offset = iocb->ki_pos;
505 size_t count = iov_iter_count(from);
506 const struct iomap_ops *iomap_ops = &ext4_iomap_ops;
507 bool extend = false, unwritten = false;
508 bool ilock_shared = true;
509 int dio_flags = 0;
510
511 /*
512 * Quick check here without any i_rwsem lock to see if it is extending
513 * IO. A more reliable check is done in ext4_dio_write_checks() with
514 * proper locking in place.
515 */
516 if (offset + count > i_size_read(inode))
517 ilock_shared = false;
518
519 if (iocb->ki_flags & IOCB_NOWAIT) {
520 if (ilock_shared) {
521 if (!inode_trylock_shared(inode))
522 return -EAGAIN;
523 } else {
524 if (!inode_trylock(inode))
525 return -EAGAIN;
526 }
527 } else {
528 if (ilock_shared)
529 inode_lock_shared(inode);
530 else
531 inode_lock(inode);
532 }
533
534 /* Fallback to buffered I/O if the inode does not support direct I/O. */
535 if (!ext4_should_use_dio(iocb, from)) {
536 if (ilock_shared)
537 inode_unlock_shared(inode);
538 else
539 inode_unlock(inode);
540 return ext4_buffered_write_iter(iocb, from);
541 }
542
543 /*
544 * Prevent inline data from being created since we are going to allocate
545 * blocks for DIO. We know the inode does not currently have inline data
546 * because ext4_should_use_dio() checked for it, but we have to clear
547 * the state flag before the write checks because a lock cycle could
548 * introduce races with other writers.
549 */
550 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
551
552 ret = ext4_dio_write_checks(iocb, from, &ilock_shared, &extend,
553 &unwritten, &dio_flags);
554 if (ret <= 0)
555 return ret;
556
557 offset = iocb->ki_pos;
558 count = ret;
559
560 if (extend) {
561 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
562 if (IS_ERR(handle)) {
563 ret = PTR_ERR(handle);
564 goto out;
565 }
566
567 ret = ext4_orphan_add(handle, inode);
568 ext4_journal_stop(handle);
569 if (ret)
570 goto out;
571 }
572
573 if (ilock_shared && !unwritten)
574 iomap_ops = &ext4_iomap_overwrite_ops;
575 ret = iomap_dio_rw(iocb, from, iomap_ops, &ext4_dio_write_ops,
576 dio_flags, NULL, 0);
577 if (ret == -ENOTBLK)
578 ret = 0;
579 if (extend) {
580 /*
581 * We always perform extending DIO write synchronously so by
582 * now the IO is completed and ext4_handle_inode_extension()
583 * was called. Cleanup the inode in case of error or race with
584 * writeback of delalloc blocks.
585 */
586 WARN_ON_ONCE(ret == -EIOCBQUEUED);
587 ext4_inode_extension_cleanup(inode, ret < 0);
588 }
589
590out:
591 if (ilock_shared)
592 inode_unlock_shared(inode);
593 else
594 inode_unlock(inode);
595
596 if (ret >= 0 && iov_iter_count(from)) {
597 ssize_t err;
598 loff_t endbyte;
599
600 /*
601 * There is no support for atomic writes on buffered-io yet,
602 * we should never fallback to buffered-io for DIO atomic
603 * writes.
604 */
605 WARN_ON_ONCE(iocb->ki_flags & IOCB_ATOMIC);
606
607 offset = iocb->ki_pos;
608 err = ext4_buffered_write_iter(iocb, from);
609 if (err < 0)
610 return err;
611
612 /*
613 * We need to ensure that the pages within the page cache for
614 * the range covered by this I/O are written to disk and
615 * invalidated. This is in attempt to preserve the expected
616 * direct I/O semantics in the case we fallback to buffered I/O
617 * to complete off the I/O request.
618 */
619 ret += err;
620 endbyte = offset + err - 1;
621 err = filemap_write_and_wait_range(iocb->ki_filp->f_mapping,
622 offset, endbyte);
623 if (!err)
624 invalidate_mapping_pages(iocb->ki_filp->f_mapping,
625 offset >> PAGE_SHIFT,
626 endbyte >> PAGE_SHIFT);
627 }
628
629 return ret;
630}
631
632#ifdef CONFIG_FS_DAX
633static ssize_t
634ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
635{
636 ssize_t ret;
637 size_t count;
638 loff_t offset;
639 handle_t *handle;
640 bool extend = false;
641 struct inode *inode = file_inode(iocb->ki_filp);
642
643 if (iocb->ki_flags & IOCB_NOWAIT) {
644 if (!inode_trylock(inode))
645 return -EAGAIN;
646 } else {
647 inode_lock(inode);
648 }
649
650 ret = ext4_write_checks(iocb, from);
651 if (ret <= 0)
652 goto out;
653
654 offset = iocb->ki_pos;
655 count = iov_iter_count(from);
656
657 if (offset + count > EXT4_I(inode)->i_disksize) {
658 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
659 if (IS_ERR(handle)) {
660 ret = PTR_ERR(handle);
661 goto out;
662 }
663
664 ret = ext4_orphan_add(handle, inode);
665 if (ret) {
666 ext4_journal_stop(handle);
667 goto out;
668 }
669
670 extend = true;
671 ext4_journal_stop(handle);
672 }
673
674 ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
675
676 if (extend) {
677 ret = ext4_handle_inode_extension(inode, offset, ret, count);
678 ext4_inode_extension_cleanup(inode, ret < (ssize_t)count);
679 }
680out:
681 inode_unlock(inode);
682 if (ret > 0)
683 ret = generic_write_sync(iocb, ret);
684 return ret;
685}
686#endif
687
688static ssize_t
689ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
690{
691 struct inode *inode = file_inode(iocb->ki_filp);
692
693 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
694 return -EIO;
695
696#ifdef CONFIG_FS_DAX
697 if (IS_DAX(inode))
698 return ext4_dax_write_iter(iocb, from);
699#endif
700
701 if (iocb->ki_flags & IOCB_ATOMIC) {
702 size_t len = iov_iter_count(from);
703 int ret;
704
705 if (len < EXT4_SB(inode->i_sb)->s_awu_min ||
706 len > EXT4_SB(inode->i_sb)->s_awu_max)
707 return -EINVAL;
708
709 ret = generic_atomic_write_valid(iocb, from);
710 if (ret)
711 return ret;
712 }
713
714 if (iocb->ki_flags & IOCB_DIRECT)
715 return ext4_dio_write_iter(iocb, from);
716 else
717 return ext4_buffered_write_iter(iocb, from);
718}
719
720#ifdef CONFIG_FS_DAX
721static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf, unsigned int order)
722{
723 int error = 0;
724 vm_fault_t result;
725 int retries = 0;
726 handle_t *handle = NULL;
727 struct inode *inode = file_inode(vmf->vma->vm_file);
728 struct super_block *sb = inode->i_sb;
729
730 /*
731 * We have to distinguish real writes from writes which will result in a
732 * COW page; COW writes should *not* poke the journal (the file will not
733 * be changed). Doing so would cause unintended failures when mounted
734 * read-only.
735 *
736 * We check for VM_SHARED rather than vmf->cow_page since the latter is
737 * unset for order != 0 (i.e. only in do_cow_fault); for
738 * other sizes, dax_iomap_fault will handle splitting / fallback so that
739 * we eventually come back with a COW page.
740 */
741 bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
742 (vmf->vma->vm_flags & VM_SHARED);
743 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
744 pfn_t pfn;
745
746 if (write) {
747 sb_start_pagefault(sb);
748 file_update_time(vmf->vma->vm_file);
749 filemap_invalidate_lock_shared(mapping);
750retry:
751 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
752 EXT4_DATA_TRANS_BLOCKS(sb));
753 if (IS_ERR(handle)) {
754 filemap_invalidate_unlock_shared(mapping);
755 sb_end_pagefault(sb);
756 return VM_FAULT_SIGBUS;
757 }
758 } else {
759 filemap_invalidate_lock_shared(mapping);
760 }
761 result = dax_iomap_fault(vmf, order, &pfn, &error, &ext4_iomap_ops);
762 if (write) {
763 ext4_journal_stop(handle);
764
765 if ((result & VM_FAULT_ERROR) && error == -ENOSPC &&
766 ext4_should_retry_alloc(sb, &retries))
767 goto retry;
768 /* Handling synchronous page fault? */
769 if (result & VM_FAULT_NEEDDSYNC)
770 result = dax_finish_sync_fault(vmf, order, pfn);
771 filemap_invalidate_unlock_shared(mapping);
772 sb_end_pagefault(sb);
773 } else {
774 filemap_invalidate_unlock_shared(mapping);
775 }
776
777 return result;
778}
779
780static vm_fault_t ext4_dax_fault(struct vm_fault *vmf)
781{
782 return ext4_dax_huge_fault(vmf, 0);
783}
784
785static const struct vm_operations_struct ext4_dax_vm_ops = {
786 .fault = ext4_dax_fault,
787 .huge_fault = ext4_dax_huge_fault,
788 .page_mkwrite = ext4_dax_fault,
789 .pfn_mkwrite = ext4_dax_fault,
790};
791#else
792#define ext4_dax_vm_ops ext4_file_vm_ops
793#endif
794
795static const struct vm_operations_struct ext4_file_vm_ops = {
796 .fault = filemap_fault,
797 .map_pages = filemap_map_pages,
798 .page_mkwrite = ext4_page_mkwrite,
799};
800
801static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
802{
803 struct inode *inode = file->f_mapping->host;
804 struct dax_device *dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
805
806 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
807 return -EIO;
808
809 /*
810 * We don't support synchronous mappings for non-DAX files and
811 * for DAX files if underneath dax_device is not synchronous.
812 */
813 if (!daxdev_mapping_supported(vma, dax_dev))
814 return -EOPNOTSUPP;
815
816 file_accessed(file);
817 if (IS_DAX(file_inode(file))) {
818 vma->vm_ops = &ext4_dax_vm_ops;
819 vm_flags_set(vma, VM_HUGEPAGE);
820 } else {
821 vma->vm_ops = &ext4_file_vm_ops;
822 }
823 return 0;
824}
825
826static int ext4_sample_last_mounted(struct super_block *sb,
827 struct vfsmount *mnt)
828{
829 struct ext4_sb_info *sbi = EXT4_SB(sb);
830 struct path path;
831 char buf[64], *cp;
832 handle_t *handle;
833 int err;
834
835 if (likely(ext4_test_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED)))
836 return 0;
837
838 if (sb_rdonly(sb) || !sb_start_intwrite_trylock(sb))
839 return 0;
840
841 ext4_set_mount_flag(sb, EXT4_MF_MNTDIR_SAMPLED);
842 /*
843 * Sample where the filesystem has been mounted and
844 * store it in the superblock for sysadmin convenience
845 * when trying to sort through large numbers of block
846 * devices or filesystem images.
847 */
848 memset(buf, 0, sizeof(buf));
849 path.mnt = mnt;
850 path.dentry = mnt->mnt_root;
851 cp = d_path(&path, buf, sizeof(buf));
852 err = 0;
853 if (IS_ERR(cp))
854 goto out;
855
856 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
857 err = PTR_ERR(handle);
858 if (IS_ERR(handle))
859 goto out;
860 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
861 err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
862 EXT4_JTR_NONE);
863 if (err)
864 goto out_journal;
865 lock_buffer(sbi->s_sbh);
866 strtomem_pad(sbi->s_es->s_last_mounted, cp, 0);
867 ext4_superblock_csum_set(sb);
868 unlock_buffer(sbi->s_sbh);
869 ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
870out_journal:
871 ext4_journal_stop(handle);
872out:
873 sb_end_intwrite(sb);
874 return err;
875}
876
877static int ext4_file_open(struct inode *inode, struct file *filp)
878{
879 int ret;
880
881 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
882 return -EIO;
883
884 ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt);
885 if (ret)
886 return ret;
887
888 ret = fscrypt_file_open(inode, filp);
889 if (ret)
890 return ret;
891
892 ret = fsverity_file_open(inode, filp);
893 if (ret)
894 return ret;
895
896 /*
897 * Set up the jbd2_inode if we are opening the inode for
898 * writing and the journal is present
899 */
900 if (filp->f_mode & FMODE_WRITE) {
901 ret = ext4_inode_attach_jinode(inode);
902 if (ret < 0)
903 return ret;
904 }
905
906 if (ext4_inode_can_atomic_write(inode))
907 filp->f_mode |= FMODE_CAN_ATOMIC_WRITE;
908
909 filp->f_mode |= FMODE_NOWAIT | FMODE_CAN_ODIRECT;
910 return dquot_file_open(inode, filp);
911}
912
913/*
914 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
915 * by calling generic_file_llseek_size() with the appropriate maxbytes
916 * value for each.
917 */
918loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
919{
920 struct inode *inode = file->f_mapping->host;
921 loff_t maxbytes;
922
923 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
924 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
925 else
926 maxbytes = inode->i_sb->s_maxbytes;
927
928 switch (whence) {
929 default:
930 return generic_file_llseek_size(file, offset, whence,
931 maxbytes, i_size_read(inode));
932 case SEEK_HOLE:
933 inode_lock_shared(inode);
934 offset = iomap_seek_hole(inode, offset,
935 &ext4_iomap_report_ops);
936 inode_unlock_shared(inode);
937 break;
938 case SEEK_DATA:
939 inode_lock_shared(inode);
940 offset = iomap_seek_data(inode, offset,
941 &ext4_iomap_report_ops);
942 inode_unlock_shared(inode);
943 break;
944 }
945
946 if (offset < 0)
947 return offset;
948 return vfs_setpos(file, offset, maxbytes);
949}
950
951const struct file_operations ext4_file_operations = {
952 .llseek = ext4_llseek,
953 .read_iter = ext4_file_read_iter,
954 .write_iter = ext4_file_write_iter,
955 .iopoll = iocb_bio_iopoll,
956 .unlocked_ioctl = ext4_ioctl,
957#ifdef CONFIG_COMPAT
958 .compat_ioctl = ext4_compat_ioctl,
959#endif
960 .mmap = ext4_file_mmap,
961 .open = ext4_file_open,
962 .release = ext4_release_file,
963 .fsync = ext4_sync_file,
964 .get_unmapped_area = thp_get_unmapped_area,
965 .splice_read = ext4_file_splice_read,
966 .splice_write = iter_file_splice_write,
967 .fallocate = ext4_fallocate,
968 .fop_flags = FOP_MMAP_SYNC | FOP_BUFFER_RASYNC |
969 FOP_DIO_PARALLEL_WRITE,
970};
971
972const struct inode_operations ext4_file_inode_operations = {
973 .setattr = ext4_setattr,
974 .getattr = ext4_file_getattr,
975 .listxattr = ext4_listxattr,
976 .get_inode_acl = ext4_get_acl,
977 .set_acl = ext4_set_acl,
978 .fiemap = ext4_fiemap,
979 .fileattr_get = ext4_fileattr_get,
980 .fileattr_set = ext4_fileattr_set,
981};
982