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 "ext4.h"
33#include "ext4_jbd2.h"
34#include "xattr.h"
35#include "acl.h"
36
37#ifdef CONFIG_FS_DAX
38static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
39{
40 struct inode *inode = file_inode(iocb->ki_filp);
41 ssize_t ret;
42
43 if (!inode_trylock_shared(inode)) {
44 if (iocb->ki_flags & IOCB_NOWAIT)
45 return -EAGAIN;
46 inode_lock_shared(inode);
47 }
48 /*
49 * Recheck under inode lock - at this point we are sure it cannot
50 * change anymore
51 */
52 if (!IS_DAX(inode)) {
53 inode_unlock_shared(inode);
54 /* Fallback to buffered IO in case we cannot support DAX */
55 return generic_file_read_iter(iocb, to);
56 }
57 ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
58 inode_unlock_shared(inode);
59
60 file_accessed(iocb->ki_filp);
61 return ret;
62}
63#endif
64
65static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
66{
67 if (unlikely(ext4_forced_shutdown(EXT4_SB(file_inode(iocb->ki_filp)->i_sb))))
68 return -EIO;
69
70 if (!iov_iter_count(to))
71 return 0; /* skip atime */
72
73#ifdef CONFIG_FS_DAX
74 if (IS_DAX(file_inode(iocb->ki_filp)))
75 return ext4_dax_read_iter(iocb, to);
76#endif
77 return generic_file_read_iter(iocb, to);
78}
79
80/*
81 * Called when an inode is released. Note that this is different
82 * from ext4_file_open: open gets called at every open, but release
83 * gets called only when /all/ the files are closed.
84 */
85static int ext4_release_file(struct inode *inode, struct file *filp)
86{
87 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
88 ext4_alloc_da_blocks(inode);
89 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
90 }
91 /* if we are the last writer on the inode, drop the block reservation */
92 if ((filp->f_mode & FMODE_WRITE) &&
93 (atomic_read(&inode->i_writecount) == 1) &&
94 !EXT4_I(inode)->i_reserved_data_blocks)
95 {
96 down_write(&EXT4_I(inode)->i_data_sem);
97 ext4_discard_preallocations(inode);
98 up_write(&EXT4_I(inode)->i_data_sem);
99 }
100 if (is_dx(inode) && filp->private_data)
101 ext4_htree_free_dir_info(filp->private_data);
102
103 return 0;
104}
105
106static void ext4_unwritten_wait(struct inode *inode)
107{
108 wait_queue_head_t *wq = ext4_ioend_wq(inode);
109
110 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
111}
112
113/*
114 * This tests whether the IO in question is block-aligned or not.
115 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
116 * are converted to written only after the IO is complete. Until they are
117 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
118 * it needs to zero out portions of the start and/or end block. If 2 AIO
119 * threads are at work on the same unwritten block, they must be synchronized
120 * or one thread will zero the other's data, causing corruption.
121 */
122static int
123ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
124{
125 struct super_block *sb = inode->i_sb;
126 int blockmask = sb->s_blocksize - 1;
127
128 if (pos >= ALIGN(i_size_read(inode), sb->s_blocksize))
129 return 0;
130
131 if ((pos | iov_iter_alignment(from)) & blockmask)
132 return 1;
133
134 return 0;
135}
136
137/* Is IO overwriting allocated and initialized blocks? */
138static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
139{
140 struct ext4_map_blocks map;
141 unsigned int blkbits = inode->i_blkbits;
142 int err, blklen;
143
144 if (pos + len > i_size_read(inode))
145 return false;
146
147 map.m_lblk = pos >> blkbits;
148 map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
149 blklen = map.m_len;
150
151 err = ext4_map_blocks(NULL, inode, &map, 0);
152 /*
153 * 'err==len' means that all of the blocks have been preallocated,
154 * regardless of whether they have been initialized or not. To exclude
155 * unwritten extents, we need to check m_flags.
156 */
157 return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
158}
159
160static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
161{
162 struct inode *inode = file_inode(iocb->ki_filp);
163 ssize_t ret;
164
165 ret = generic_write_checks(iocb, from);
166 if (ret <= 0)
167 return ret;
168
169 if (unlikely(IS_IMMUTABLE(inode)))
170 return -EPERM;
171
172 /*
173 * If we have encountered a bitmap-format file, the size limit
174 * is smaller than s_maxbytes, which is for extent-mapped files.
175 */
176 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
177 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
178
179 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
180 return -EFBIG;
181 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
182 }
183 return iov_iter_count(from);
184}
185
186#ifdef CONFIG_FS_DAX
187static ssize_t
188ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
189{
190 struct inode *inode = file_inode(iocb->ki_filp);
191 ssize_t ret;
192
193 if (!inode_trylock(inode)) {
194 if (iocb->ki_flags & IOCB_NOWAIT)
195 return -EAGAIN;
196 inode_lock(inode);
197 }
198 ret = ext4_write_checks(iocb, from);
199 if (ret <= 0)
200 goto out;
201 ret = file_remove_privs(iocb->ki_filp);
202 if (ret)
203 goto out;
204 ret = file_update_time(iocb->ki_filp);
205 if (ret)
206 goto out;
207
208 ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
209out:
210 inode_unlock(inode);
211 if (ret > 0)
212 ret = generic_write_sync(iocb, ret);
213 return ret;
214}
215#endif
216
217static ssize_t
218ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
219{
220 struct inode *inode = file_inode(iocb->ki_filp);
221 int o_direct = iocb->ki_flags & IOCB_DIRECT;
222 int unaligned_aio = 0;
223 int overwrite = 0;
224 ssize_t ret;
225
226 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
227 return -EIO;
228
229#ifdef CONFIG_FS_DAX
230 if (IS_DAX(inode))
231 return ext4_dax_write_iter(iocb, from);
232#endif
233
234 if (!inode_trylock(inode)) {
235 if (iocb->ki_flags & IOCB_NOWAIT)
236 return -EAGAIN;
237 inode_lock(inode);
238 }
239
240 ret = ext4_write_checks(iocb, from);
241 if (ret <= 0)
242 goto out;
243
244 /*
245 * Unaligned direct AIO must be serialized among each other as zeroing
246 * of partial blocks of two competing unaligned AIOs can result in data
247 * corruption.
248 */
249 if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
250 !is_sync_kiocb(iocb) &&
251 ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
252 unaligned_aio = 1;
253 ext4_unwritten_wait(inode);
254 }
255
256 iocb->private = &overwrite;
257 /* Check whether we do a DIO overwrite or not */
258 if (o_direct && !unaligned_aio) {
259 if (ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from))) {
260 if (ext4_should_dioread_nolock(inode))
261 overwrite = 1;
262 } else if (iocb->ki_flags & IOCB_NOWAIT) {
263 ret = -EAGAIN;
264 goto out;
265 }
266 }
267
268 ret = __generic_file_write_iter(iocb, from);
269 /*
270 * Unaligned direct AIO must be the only IO in flight. Otherwise
271 * overlapping aligned IO after unaligned might result in data
272 * corruption.
273 */
274 if (ret == -EIOCBQUEUED && unaligned_aio)
275 ext4_unwritten_wait(inode);
276 inode_unlock(inode);
277
278 if (ret > 0)
279 ret = generic_write_sync(iocb, ret);
280
281 return ret;
282
283out:
284 inode_unlock(inode);
285 return ret;
286}
287
288#ifdef CONFIG_FS_DAX
289static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf,
290 enum page_entry_size pe_size)
291{
292 int error = 0;
293 vm_fault_t result;
294 int retries = 0;
295 handle_t *handle = NULL;
296 struct inode *inode = file_inode(vmf->vma->vm_file);
297 struct super_block *sb = inode->i_sb;
298
299 /*
300 * We have to distinguish real writes from writes which will result in a
301 * COW page; COW writes should *not* poke the journal (the file will not
302 * be changed). Doing so would cause unintended failures when mounted
303 * read-only.
304 *
305 * We check for VM_SHARED rather than vmf->cow_page since the latter is
306 * unset for pe_size != PE_SIZE_PTE (i.e. only in do_cow_fault); for
307 * other sizes, dax_iomap_fault will handle splitting / fallback so that
308 * we eventually come back with a COW page.
309 */
310 bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
311 (vmf->vma->vm_flags & VM_SHARED);
312 pfn_t pfn;
313
314 if (write) {
315 sb_start_pagefault(sb);
316 file_update_time(vmf->vma->vm_file);
317 down_read(&EXT4_I(inode)->i_mmap_sem);
318retry:
319 handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
320 EXT4_DATA_TRANS_BLOCKS(sb));
321 if (IS_ERR(handle)) {
322 up_read(&EXT4_I(inode)->i_mmap_sem);
323 sb_end_pagefault(sb);
324 return VM_FAULT_SIGBUS;
325 }
326 } else {
327 down_read(&EXT4_I(inode)->i_mmap_sem);
328 }
329 result = dax_iomap_fault(vmf, pe_size, &pfn, &error, &ext4_iomap_ops);
330 if (write) {
331 ext4_journal_stop(handle);
332
333 if ((result & VM_FAULT_ERROR) && error == -ENOSPC &&
334 ext4_should_retry_alloc(sb, &retries))
335 goto retry;
336 /* Handling synchronous page fault? */
337 if (result & VM_FAULT_NEEDDSYNC)
338 result = dax_finish_sync_fault(vmf, pe_size, pfn);
339 up_read(&EXT4_I(inode)->i_mmap_sem);
340 sb_end_pagefault(sb);
341 } else {
342 up_read(&EXT4_I(inode)->i_mmap_sem);
343 }
344
345 return result;
346}
347
348static vm_fault_t ext4_dax_fault(struct vm_fault *vmf)
349{
350 return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
351}
352
353static const struct vm_operations_struct ext4_dax_vm_ops = {
354 .fault = ext4_dax_fault,
355 .huge_fault = ext4_dax_huge_fault,
356 .page_mkwrite = ext4_dax_fault,
357 .pfn_mkwrite = ext4_dax_fault,
358};
359#else
360#define ext4_dax_vm_ops ext4_file_vm_ops
361#endif
362
363static const struct vm_operations_struct ext4_file_vm_ops = {
364 .fault = ext4_filemap_fault,
365 .map_pages = filemap_map_pages,
366 .page_mkwrite = ext4_page_mkwrite,
367};
368
369static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
370{
371 struct inode *inode = file->f_mapping->host;
372 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
373 struct dax_device *dax_dev = sbi->s_daxdev;
374
375 if (unlikely(ext4_forced_shutdown(sbi)))
376 return -EIO;
377
378 /*
379 * We don't support synchronous mappings for non-DAX files and
380 * for DAX files if underneath dax_device is not synchronous.
381 */
382 if (!daxdev_mapping_supported(vma, dax_dev))
383 return -EOPNOTSUPP;
384
385 file_accessed(file);
386 if (IS_DAX(file_inode(file))) {
387 vma->vm_ops = &ext4_dax_vm_ops;
388 vma->vm_flags |= VM_HUGEPAGE;
389 } else {
390 vma->vm_ops = &ext4_file_vm_ops;
391 }
392 return 0;
393}
394
395static int ext4_sample_last_mounted(struct super_block *sb,
396 struct vfsmount *mnt)
397{
398 struct ext4_sb_info *sbi = EXT4_SB(sb);
399 struct path path;
400 char buf[64], *cp;
401 handle_t *handle;
402 int err;
403
404 if (likely(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED))
405 return 0;
406
407 if (sb_rdonly(sb) || !sb_start_intwrite_trylock(sb))
408 return 0;
409
410 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
411 /*
412 * Sample where the filesystem has been mounted and
413 * store it in the superblock for sysadmin convenience
414 * when trying to sort through large numbers of block
415 * devices or filesystem images.
416 */
417 memset(buf, 0, sizeof(buf));
418 path.mnt = mnt;
419 path.dentry = mnt->mnt_root;
420 cp = d_path(&path, buf, sizeof(buf));
421 err = 0;
422 if (IS_ERR(cp))
423 goto out;
424
425 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
426 err = PTR_ERR(handle);
427 if (IS_ERR(handle))
428 goto out;
429 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
430 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
431 if (err)
432 goto out_journal;
433 strlcpy(sbi->s_es->s_last_mounted, cp,
434 sizeof(sbi->s_es->s_last_mounted));
435 ext4_handle_dirty_super(handle, sb);
436out_journal:
437 ext4_journal_stop(handle);
438out:
439 sb_end_intwrite(sb);
440 return err;
441}
442
443static int ext4_file_open(struct inode * inode, struct file * filp)
444{
445 int ret;
446
447 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
448 return -EIO;
449
450 ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt);
451 if (ret)
452 return ret;
453
454 ret = fscrypt_file_open(inode, filp);
455 if (ret)
456 return ret;
457
458 ret = fsverity_file_open(inode, filp);
459 if (ret)
460 return ret;
461
462 /*
463 * Set up the jbd2_inode if we are opening the inode for
464 * writing and the journal is present
465 */
466 if (filp->f_mode & FMODE_WRITE) {
467 ret = ext4_inode_attach_jinode(inode);
468 if (ret < 0)
469 return ret;
470 }
471
472 filp->f_mode |= FMODE_NOWAIT;
473 return dquot_file_open(inode, filp);
474}
475
476/*
477 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
478 * by calling generic_file_llseek_size() with the appropriate maxbytes
479 * value for each.
480 */
481loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
482{
483 struct inode *inode = file->f_mapping->host;
484 loff_t maxbytes;
485
486 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
487 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
488 else
489 maxbytes = inode->i_sb->s_maxbytes;
490
491 switch (whence) {
492 default:
493 return generic_file_llseek_size(file, offset, whence,
494 maxbytes, i_size_read(inode));
495 case SEEK_HOLE:
496 inode_lock_shared(inode);
497 offset = iomap_seek_hole(inode, offset, &ext4_iomap_ops);
498 inode_unlock_shared(inode);
499 break;
500 case SEEK_DATA:
501 inode_lock_shared(inode);
502 offset = iomap_seek_data(inode, offset, &ext4_iomap_ops);
503 inode_unlock_shared(inode);
504 break;
505 }
506
507 if (offset < 0)
508 return offset;
509 return vfs_setpos(file, offset, maxbytes);
510}
511
512const struct file_operations ext4_file_operations = {
513 .llseek = ext4_llseek,
514 .read_iter = ext4_file_read_iter,
515 .write_iter = ext4_file_write_iter,
516 .unlocked_ioctl = ext4_ioctl,
517#ifdef CONFIG_COMPAT
518 .compat_ioctl = ext4_compat_ioctl,
519#endif
520 .mmap = ext4_file_mmap,
521 .mmap_supported_flags = MAP_SYNC,
522 .open = ext4_file_open,
523 .release = ext4_release_file,
524 .fsync = ext4_sync_file,
525 .get_unmapped_area = thp_get_unmapped_area,
526 .splice_read = generic_file_splice_read,
527 .splice_write = iter_file_splice_write,
528 .fallocate = ext4_fallocate,
529};
530
531const struct inode_operations ext4_file_inode_operations = {
532 .setattr = ext4_setattr,
533 .getattr = ext4_file_getattr,
534 .listxattr = ext4_listxattr,
535 .get_acl = ext4_get_acl,
536 .set_acl = ext4_set_acl,
537 .fiemap = ext4_fiemap,
538};
539