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#ifdef CONFIG_FS_DAX
35static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
36{
37 struct inode *inode = file_inode(iocb->ki_filp);
38 ssize_t ret;
39
40 inode_lock_shared(inode);
41 /*
42 * Recheck under inode lock - at this point we are sure it cannot
43 * change anymore
44 */
45 if (!IS_DAX(inode)) {
46 inode_unlock_shared(inode);
47 /* Fallback to buffered IO in case we cannot support DAX */
48 return generic_file_read_iter(iocb, to);
49 }
50 ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
51 inode_unlock_shared(inode);
52
53 file_accessed(iocb->ki_filp);
54 return ret;
55}
56#endif
57
58static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
59{
60 if (!iov_iter_count(to))
61 return 0; /* skip atime */
62
63#ifdef CONFIG_FS_DAX
64 if (IS_DAX(file_inode(iocb->ki_filp)))
65 return ext4_dax_read_iter(iocb, to);
66#endif
67 return generic_file_read_iter(iocb, to);
68}
69
70/*
71 * Called when an inode is released. Note that this is different
72 * from ext4_file_open: open gets called at every open, but release
73 * gets called only when /all/ the files are closed.
74 */
75static int ext4_release_file(struct inode *inode, struct file *filp)
76{
77 if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
78 ext4_alloc_da_blocks(inode);
79 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
80 }
81 /* if we are the last writer on the inode, drop the block reservation */
82 if ((filp->f_mode & FMODE_WRITE) &&
83 (atomic_read(&inode->i_writecount) == 1) &&
84 !EXT4_I(inode)->i_reserved_data_blocks)
85 {
86 down_write(&EXT4_I(inode)->i_data_sem);
87 ext4_discard_preallocations(inode);
88 up_write(&EXT4_I(inode)->i_data_sem);
89 }
90 if (is_dx(inode) && filp->private_data)
91 ext4_htree_free_dir_info(filp->private_data);
92
93 return 0;
94}
95
96static void ext4_unwritten_wait(struct inode *inode)
97{
98 wait_queue_head_t *wq = ext4_ioend_wq(inode);
99
100 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
101}
102
103/*
104 * This tests whether the IO in question is block-aligned or not.
105 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
106 * are converted to written only after the IO is complete. Until they are
107 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
108 * it needs to zero out portions of the start and/or end block. If 2 AIO
109 * threads are at work on the same unwritten block, they must be synchronized
110 * or one thread will zero the other's data, causing corruption.
111 */
112static int
113ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
114{
115 struct super_block *sb = inode->i_sb;
116 int blockmask = sb->s_blocksize - 1;
117
118 if (pos >= i_size_read(inode))
119 return 0;
120
121 if ((pos | iov_iter_alignment(from)) & blockmask)
122 return 1;
123
124 return 0;
125}
126
127/* Is IO overwriting allocated and initialized blocks? */
128static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
129{
130 struct ext4_map_blocks map;
131 unsigned int blkbits = inode->i_blkbits;
132 int err, blklen;
133
134 if (pos + len > i_size_read(inode))
135 return false;
136
137 map.m_lblk = pos >> blkbits;
138 map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
139 blklen = map.m_len;
140
141 err = ext4_map_blocks(NULL, inode, &map, 0);
142 /*
143 * 'err==len' means that all of the blocks have been preallocated,
144 * regardless of whether they have been initialized or not. To exclude
145 * unwritten extents, we need to check m_flags.
146 */
147 return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
148}
149
150static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
151{
152 struct inode *inode = file_inode(iocb->ki_filp);
153 ssize_t ret;
154
155 ret = generic_write_checks(iocb, from);
156 if (ret <= 0)
157 return ret;
158 /*
159 * If we have encountered a bitmap-format file, the size limit
160 * is smaller than s_maxbytes, which is for extent-mapped files.
161 */
162 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
163 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
164
165 if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
166 return -EFBIG;
167 iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
168 }
169 return iov_iter_count(from);
170}
171
172#ifdef CONFIG_FS_DAX
173static ssize_t
174ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
175{
176 struct inode *inode = file_inode(iocb->ki_filp);
177 ssize_t ret;
178 bool overwrite = false;
179
180 inode_lock(inode);
181 ret = ext4_write_checks(iocb, from);
182 if (ret <= 0)
183 goto out;
184 ret = file_remove_privs(iocb->ki_filp);
185 if (ret)
186 goto out;
187 ret = file_update_time(iocb->ki_filp);
188 if (ret)
189 goto out;
190
191 if (ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from))) {
192 overwrite = true;
193 downgrade_write(&inode->i_rwsem);
194 }
195 ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
196out:
197 if (!overwrite)
198 inode_unlock(inode);
199 else
200 inode_unlock_shared(inode);
201 if (ret > 0)
202 ret = generic_write_sync(iocb, ret);
203 return ret;
204}
205#endif
206
207static ssize_t
208ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
209{
210 struct inode *inode = file_inode(iocb->ki_filp);
211 int o_direct = iocb->ki_flags & IOCB_DIRECT;
212 int unaligned_aio = 0;
213 int overwrite = 0;
214 ssize_t ret;
215
216#ifdef CONFIG_FS_DAX
217 if (IS_DAX(inode))
218 return ext4_dax_write_iter(iocb, from);
219#endif
220
221 inode_lock(inode);
222 ret = ext4_write_checks(iocb, from);
223 if (ret <= 0)
224 goto out;
225
226 /*
227 * Unaligned direct AIO must be serialized among each other as zeroing
228 * of partial blocks of two competing unaligned AIOs can result in data
229 * corruption.
230 */
231 if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
232 !is_sync_kiocb(iocb) &&
233 ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
234 unaligned_aio = 1;
235 ext4_unwritten_wait(inode);
236 }
237
238 iocb->private = &overwrite;
239 /* Check whether we do a DIO overwrite or not */
240 if (o_direct && ext4_should_dioread_nolock(inode) && !unaligned_aio &&
241 ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from)))
242 overwrite = 1;
243
244 ret = __generic_file_write_iter(iocb, from);
245 inode_unlock(inode);
246
247 if (ret > 0)
248 ret = generic_write_sync(iocb, ret);
249
250 return ret;
251
252out:
253 inode_unlock(inode);
254 return ret;
255}
256
257#ifdef CONFIG_FS_DAX
258static int ext4_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
259{
260 int result;
261 struct inode *inode = file_inode(vma->vm_file);
262 struct super_block *sb = inode->i_sb;
263 bool write = vmf->flags & FAULT_FLAG_WRITE;
264
265 if (write) {
266 sb_start_pagefault(sb);
267 file_update_time(vma->vm_file);
268 }
269 down_read(&EXT4_I(inode)->i_mmap_sem);
270 result = dax_iomap_fault(vma, vmf, &ext4_iomap_ops);
271 up_read(&EXT4_I(inode)->i_mmap_sem);
272 if (write)
273 sb_end_pagefault(sb);
274
275 return result;
276}
277
278static int ext4_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
279 pmd_t *pmd, unsigned int flags)
280{
281 int result;
282 struct inode *inode = file_inode(vma->vm_file);
283 struct super_block *sb = inode->i_sb;
284 bool write = flags & FAULT_FLAG_WRITE;
285
286 if (write) {
287 sb_start_pagefault(sb);
288 file_update_time(vma->vm_file);
289 }
290 down_read(&EXT4_I(inode)->i_mmap_sem);
291 result = dax_iomap_pmd_fault(vma, addr, pmd, flags,
292 &ext4_iomap_ops);
293 up_read(&EXT4_I(inode)->i_mmap_sem);
294 if (write)
295 sb_end_pagefault(sb);
296
297 return result;
298}
299
300/*
301 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
302 * handler we check for races agaist truncate. Note that since we cycle through
303 * i_mmap_sem, we are sure that also any hole punching that began before we
304 * were called is finished by now and so if it included part of the file we
305 * are working on, our pte will get unmapped and the check for pte_same() in
306 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
307 * desired.
308 */
309static int ext4_dax_pfn_mkwrite(struct vm_area_struct *vma,
310 struct vm_fault *vmf)
311{
312 struct inode *inode = file_inode(vma->vm_file);
313 struct super_block *sb = inode->i_sb;
314 loff_t size;
315 int ret;
316
317 sb_start_pagefault(sb);
318 file_update_time(vma->vm_file);
319 down_read(&EXT4_I(inode)->i_mmap_sem);
320 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
321 if (vmf->pgoff >= size)
322 ret = VM_FAULT_SIGBUS;
323 else
324 ret = dax_pfn_mkwrite(vma, vmf);
325 up_read(&EXT4_I(inode)->i_mmap_sem);
326 sb_end_pagefault(sb);
327
328 return ret;
329}
330
331static const struct vm_operations_struct ext4_dax_vm_ops = {
332 .fault = ext4_dax_fault,
333 .pmd_fault = ext4_dax_pmd_fault,
334 .page_mkwrite = ext4_dax_fault,
335 .pfn_mkwrite = ext4_dax_pfn_mkwrite,
336};
337#else
338#define ext4_dax_vm_ops ext4_file_vm_ops
339#endif
340
341static const struct vm_operations_struct ext4_file_vm_ops = {
342 .fault = ext4_filemap_fault,
343 .map_pages = filemap_map_pages,
344 .page_mkwrite = ext4_page_mkwrite,
345};
346
347static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
348{
349 struct inode *inode = file->f_mapping->host;
350
351 if (ext4_encrypted_inode(inode)) {
352 int err = fscrypt_get_encryption_info(inode);
353 if (err)
354 return 0;
355 if (!fscrypt_has_encryption_key(inode))
356 return -ENOKEY;
357 }
358 file_accessed(file);
359 if (IS_DAX(file_inode(file))) {
360 vma->vm_ops = &ext4_dax_vm_ops;
361 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
362 } else {
363 vma->vm_ops = &ext4_file_vm_ops;
364 }
365 return 0;
366}
367
368static int ext4_file_open(struct inode * inode, struct file * filp)
369{
370 struct super_block *sb = inode->i_sb;
371 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
372 struct vfsmount *mnt = filp->f_path.mnt;
373 struct dentry *dir;
374 struct path path;
375 char buf[64], *cp;
376 int ret;
377
378 if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
379 !(sb->s_flags & MS_RDONLY))) {
380 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
381 /*
382 * Sample where the filesystem has been mounted and
383 * store it in the superblock for sysadmin convenience
384 * when trying to sort through large numbers of block
385 * devices or filesystem images.
386 */
387 memset(buf, 0, sizeof(buf));
388 path.mnt = mnt;
389 path.dentry = mnt->mnt_root;
390 cp = d_path(&path, buf, sizeof(buf));
391 if (!IS_ERR(cp)) {
392 handle_t *handle;
393 int err;
394
395 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
396 if (IS_ERR(handle))
397 return PTR_ERR(handle);
398 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
399 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
400 if (err) {
401 ext4_journal_stop(handle);
402 return err;
403 }
404 strlcpy(sbi->s_es->s_last_mounted, cp,
405 sizeof(sbi->s_es->s_last_mounted));
406 ext4_handle_dirty_super(handle, sb);
407 ext4_journal_stop(handle);
408 }
409 }
410 if (ext4_encrypted_inode(inode)) {
411 ret = fscrypt_get_encryption_info(inode);
412 if (ret)
413 return -EACCES;
414 if (!fscrypt_has_encryption_key(inode))
415 return -ENOKEY;
416 }
417
418 dir = dget_parent(file_dentry(filp));
419 if (ext4_encrypted_inode(d_inode(dir)) &&
420 !fscrypt_has_permitted_context(d_inode(dir), inode)) {
421 ext4_warning(inode->i_sb,
422 "Inconsistent encryption contexts: %lu/%lu",
423 (unsigned long) d_inode(dir)->i_ino,
424 (unsigned long) inode->i_ino);
425 dput(dir);
426 return -EPERM;
427 }
428 dput(dir);
429 /*
430 * Set up the jbd2_inode if we are opening the inode for
431 * writing and the journal is present
432 */
433 if (filp->f_mode & FMODE_WRITE) {
434 ret = ext4_inode_attach_jinode(inode);
435 if (ret < 0)
436 return ret;
437 }
438 return dquot_file_open(inode, filp);
439}
440
441/*
442 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
443 * file rather than ext4_ext_walk_space() because we can introduce
444 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
445 * function. When extent status tree has been fully implemented, it will
446 * track all extent status for a file and we can directly use it to
447 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
448 */
449
450/*
451 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
452 * lookup page cache to check whether or not there has some data between
453 * [startoff, endoff] because, if this range contains an unwritten extent,
454 * we determine this extent as a data or a hole according to whether the
455 * page cache has data or not.
456 */
457static int ext4_find_unwritten_pgoff(struct inode *inode,
458 int whence,
459 ext4_lblk_t end_blk,
460 loff_t *offset)
461{
462 struct pagevec pvec;
463 unsigned int blkbits;
464 pgoff_t index;
465 pgoff_t end;
466 loff_t endoff;
467 loff_t startoff;
468 loff_t lastoff;
469 int found = 0;
470
471 blkbits = inode->i_sb->s_blocksize_bits;
472 startoff = *offset;
473 lastoff = startoff;
474 endoff = (loff_t)end_blk << blkbits;
475
476 index = startoff >> PAGE_SHIFT;
477 end = endoff >> PAGE_SHIFT;
478
479 pagevec_init(&pvec, 0);
480 do {
481 int i, num;
482 unsigned long nr_pages;
483
484 num = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
485 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
486 (pgoff_t)num);
487 if (nr_pages == 0) {
488 if (whence == SEEK_DATA)
489 break;
490
491 BUG_ON(whence != SEEK_HOLE);
492 /*
493 * If this is the first time to go into the loop and
494 * offset is not beyond the end offset, it will be a
495 * hole at this offset
496 */
497 if (lastoff == startoff || lastoff < endoff)
498 found = 1;
499 break;
500 }
501
502 /*
503 * If this is the first time to go into the loop and
504 * offset is smaller than the first page offset, it will be a
505 * hole at this offset.
506 */
507 if (lastoff == startoff && whence == SEEK_HOLE &&
508 lastoff < page_offset(pvec.pages[0])) {
509 found = 1;
510 break;
511 }
512
513 for (i = 0; i < nr_pages; i++) {
514 struct page *page = pvec.pages[i];
515 struct buffer_head *bh, *head;
516
517 /*
518 * If the current offset is not beyond the end of given
519 * range, it will be a hole.
520 */
521 if (lastoff < endoff && whence == SEEK_HOLE &&
522 page->index > end) {
523 found = 1;
524 *offset = lastoff;
525 goto out;
526 }
527
528 lock_page(page);
529
530 if (unlikely(page->mapping != inode->i_mapping)) {
531 unlock_page(page);
532 continue;
533 }
534
535 if (!page_has_buffers(page)) {
536 unlock_page(page);
537 continue;
538 }
539
540 if (page_has_buffers(page)) {
541 lastoff = page_offset(page);
542 bh = head = page_buffers(page);
543 do {
544 if (buffer_uptodate(bh) ||
545 buffer_unwritten(bh)) {
546 if (whence == SEEK_DATA)
547 found = 1;
548 } else {
549 if (whence == SEEK_HOLE)
550 found = 1;
551 }
552 if (found) {
553 *offset = max_t(loff_t,
554 startoff, lastoff);
555 unlock_page(page);
556 goto out;
557 }
558 lastoff += bh->b_size;
559 bh = bh->b_this_page;
560 } while (bh != head);
561 }
562
563 lastoff = page_offset(page) + PAGE_SIZE;
564 unlock_page(page);
565 }
566
567 /*
568 * The no. of pages is less than our desired, that would be a
569 * hole in there.
570 */
571 if (nr_pages < num && whence == SEEK_HOLE) {
572 found = 1;
573 *offset = lastoff;
574 break;
575 }
576
577 index = pvec.pages[i - 1]->index + 1;
578 pagevec_release(&pvec);
579 } while (index <= end);
580
581out:
582 pagevec_release(&pvec);
583 return found;
584}
585
586/*
587 * ext4_seek_data() retrieves the offset for SEEK_DATA.
588 */
589static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
590{
591 struct inode *inode = file->f_mapping->host;
592 struct extent_status es;
593 ext4_lblk_t start, last, end;
594 loff_t dataoff, isize;
595 int blkbits;
596 int ret;
597
598 inode_lock(inode);
599
600 isize = i_size_read(inode);
601 if (offset >= isize) {
602 inode_unlock(inode);
603 return -ENXIO;
604 }
605
606 blkbits = inode->i_sb->s_blocksize_bits;
607 start = offset >> blkbits;
608 last = start;
609 end = isize >> blkbits;
610 dataoff = offset;
611
612 do {
613 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
614 if (ret <= 0) {
615 /* No extent found -> no data */
616 if (ret == 0)
617 ret = -ENXIO;
618 inode_unlock(inode);
619 return ret;
620 }
621
622 last = es.es_lblk;
623 if (last != start)
624 dataoff = (loff_t)last << blkbits;
625 if (!ext4_es_is_unwritten(&es))
626 break;
627
628 /*
629 * If there is a unwritten extent at this offset,
630 * it will be as a data or a hole according to page
631 * cache that has data or not.
632 */
633 if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
634 es.es_lblk + es.es_len, &dataoff))
635 break;
636 last += es.es_len;
637 dataoff = (loff_t)last << blkbits;
638 cond_resched();
639 } while (last <= end);
640
641 inode_unlock(inode);
642
643 if (dataoff > isize)
644 return -ENXIO;
645
646 return vfs_setpos(file, dataoff, maxsize);
647}
648
649/*
650 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
651 */
652static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
653{
654 struct inode *inode = file->f_mapping->host;
655 struct extent_status es;
656 ext4_lblk_t start, last, end;
657 loff_t holeoff, isize;
658 int blkbits;
659 int ret;
660
661 inode_lock(inode);
662
663 isize = i_size_read(inode);
664 if (offset >= isize) {
665 inode_unlock(inode);
666 return -ENXIO;
667 }
668
669 blkbits = inode->i_sb->s_blocksize_bits;
670 start = offset >> blkbits;
671 last = start;
672 end = isize >> blkbits;
673 holeoff = offset;
674
675 do {
676 ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
677 if (ret < 0) {
678 inode_unlock(inode);
679 return ret;
680 }
681 /* Found a hole? */
682 if (ret == 0 || es.es_lblk > last) {
683 if (last != start)
684 holeoff = (loff_t)last << blkbits;
685 break;
686 }
687 /*
688 * If there is a unwritten extent at this offset,
689 * it will be as a data or a hole according to page
690 * cache that has data or not.
691 */
692 if (ext4_es_is_unwritten(&es) &&
693 ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
694 last + es.es_len, &holeoff))
695 break;
696
697 last += es.es_len;
698 holeoff = (loff_t)last << blkbits;
699 cond_resched();
700 } while (last <= end);
701
702 inode_unlock(inode);
703
704 if (holeoff > isize)
705 holeoff = isize;
706
707 return vfs_setpos(file, holeoff, maxsize);
708}
709
710/*
711 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
712 * by calling generic_file_llseek_size() with the appropriate maxbytes
713 * value for each.
714 */
715loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
716{
717 struct inode *inode = file->f_mapping->host;
718 loff_t maxbytes;
719
720 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
721 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
722 else
723 maxbytes = inode->i_sb->s_maxbytes;
724
725 switch (whence) {
726 case SEEK_SET:
727 case SEEK_CUR:
728 case SEEK_END:
729 return generic_file_llseek_size(file, offset, whence,
730 maxbytes, i_size_read(inode));
731 case SEEK_DATA:
732 return ext4_seek_data(file, offset, maxbytes);
733 case SEEK_HOLE:
734 return ext4_seek_hole(file, offset, maxbytes);
735 }
736
737 return -EINVAL;
738}
739
740const struct file_operations ext4_file_operations = {
741 .llseek = ext4_llseek,
742 .read_iter = ext4_file_read_iter,
743 .write_iter = ext4_file_write_iter,
744 .unlocked_ioctl = ext4_ioctl,
745#ifdef CONFIG_COMPAT
746 .compat_ioctl = ext4_compat_ioctl,
747#endif
748 .mmap = ext4_file_mmap,
749 .open = ext4_file_open,
750 .release = ext4_release_file,
751 .fsync = ext4_sync_file,
752 .get_unmapped_area = thp_get_unmapped_area,
753 .splice_read = generic_file_splice_read,
754 .splice_write = iter_file_splice_write,
755 .fallocate = ext4_fallocate,
756};
757
758const struct inode_operations ext4_file_inode_operations = {
759 .setattr = ext4_setattr,
760 .getattr = ext4_getattr,
761 .listxattr = ext4_listxattr,
762 .get_acl = ext4_get_acl,
763 .set_acl = ext4_set_acl,
764 .fiemap = ext4_fiemap,
765};
766
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