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)
11 * linux/fs/minix/file.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * ext4 fs regular file handling primitives
17 * 64-bit file support on 64-bit platforms by Jakub Jelinek
18 * (jj@sunsite.ms.mff.cuni.cz)
21 #include <linux/time.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>
30 #include "ext4_jbd2.h"
35 static ssize_t
ext4_dax_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
37 struct inode
*inode
= file_inode(iocb
->ki_filp
);
40 inode_lock_shared(inode
);
42 * Recheck under inode lock - at this point we are sure it cannot
46 inode_unlock_shared(inode
);
47 /* Fallback to buffered IO in case we cannot support DAX */
48 return generic_file_read_iter(iocb
, to
);
50 ret
= dax_iomap_rw(iocb
, to
, &ext4_iomap_ops
);
51 inode_unlock_shared(inode
);
53 file_accessed(iocb
->ki_filp
);
58 static ssize_t
ext4_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
60 if (unlikely(ext4_forced_shutdown(EXT4_SB(file_inode(iocb
->ki_filp
)->i_sb
))))
63 if (!iov_iter_count(to
))
64 return 0; /* skip atime */
67 if (IS_DAX(file_inode(iocb
->ki_filp
)))
68 return ext4_dax_read_iter(iocb
, to
);
70 return generic_file_read_iter(iocb
, to
);
74 * Called when an inode is released. Note that this is different
75 * from ext4_file_open: open gets called at every open, but release
76 * gets called only when /all/ the files are closed.
78 static int ext4_release_file(struct inode
*inode
, struct file
*filp
)
80 if (ext4_test_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
)) {
81 ext4_alloc_da_blocks(inode
);
82 ext4_clear_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
84 /* if we are the last writer on the inode, drop the block reservation */
85 if ((filp
->f_mode
& FMODE_WRITE
) &&
86 (atomic_read(&inode
->i_writecount
) == 1) &&
87 !EXT4_I(inode
)->i_reserved_data_blocks
)
89 down_write(&EXT4_I(inode
)->i_data_sem
);
90 ext4_discard_preallocations(inode
);
91 up_write(&EXT4_I(inode
)->i_data_sem
);
93 if (is_dx(inode
) && filp
->private_data
)
94 ext4_htree_free_dir_info(filp
->private_data
);
99 static void ext4_unwritten_wait(struct inode
*inode
)
101 wait_queue_head_t
*wq
= ext4_ioend_wq(inode
);
103 wait_event(*wq
, (atomic_read(&EXT4_I(inode
)->i_unwritten
) == 0));
107 * This tests whether the IO in question is block-aligned or not.
108 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
109 * are converted to written only after the IO is complete. Until they are
110 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
111 * it needs to zero out portions of the start and/or end block. If 2 AIO
112 * threads are at work on the same unwritten block, they must be synchronized
113 * or one thread will zero the other's data, causing corruption.
116 ext4_unaligned_aio(struct inode
*inode
, struct iov_iter
*from
, loff_t pos
)
118 struct super_block
*sb
= inode
->i_sb
;
119 int blockmask
= sb
->s_blocksize
- 1;
121 if (pos
>= i_size_read(inode
))
124 if ((pos
| iov_iter_alignment(from
)) & blockmask
)
130 /* Is IO overwriting allocated and initialized blocks? */
131 static bool ext4_overwrite_io(struct inode
*inode
, loff_t pos
, loff_t len
)
133 struct ext4_map_blocks map
;
134 unsigned int blkbits
= inode
->i_blkbits
;
137 if (pos
+ len
> i_size_read(inode
))
140 map
.m_lblk
= pos
>> blkbits
;
141 map
.m_len
= EXT4_MAX_BLOCKS(len
, pos
, blkbits
);
144 err
= ext4_map_blocks(NULL
, inode
, &map
, 0);
146 * 'err==len' means that all of the blocks have been preallocated,
147 * regardless of whether they have been initialized or not. To exclude
148 * unwritten extents, we need to check m_flags.
150 return err
== blklen
&& (map
.m_flags
& EXT4_MAP_MAPPED
);
153 static ssize_t
ext4_write_checks(struct kiocb
*iocb
, struct iov_iter
*from
)
155 struct inode
*inode
= file_inode(iocb
->ki_filp
);
158 ret
= generic_write_checks(iocb
, from
);
162 * If we have encountered a bitmap-format file, the size limit
163 * is smaller than s_maxbytes, which is for extent-mapped files.
165 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
166 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
168 if (iocb
->ki_pos
>= sbi
->s_bitmap_maxbytes
)
170 iov_iter_truncate(from
, sbi
->s_bitmap_maxbytes
- iocb
->ki_pos
);
172 return iov_iter_count(from
);
177 ext4_dax_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
179 struct inode
*inode
= file_inode(iocb
->ki_filp
);
183 ret
= ext4_write_checks(iocb
, from
);
186 ret
= file_remove_privs(iocb
->ki_filp
);
189 ret
= file_update_time(iocb
->ki_filp
);
193 ret
= dax_iomap_rw(iocb
, from
, &ext4_iomap_ops
);
197 ret
= generic_write_sync(iocb
, ret
);
203 ext4_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
205 struct inode
*inode
= file_inode(iocb
->ki_filp
);
206 int o_direct
= iocb
->ki_flags
& IOCB_DIRECT
;
207 int unaligned_aio
= 0;
211 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
216 return ext4_dax_write_iter(iocb
, from
);
220 ret
= ext4_write_checks(iocb
, from
);
225 * Unaligned direct AIO must be serialized among each other as zeroing
226 * of partial blocks of two competing unaligned AIOs can result in data
229 if (o_direct
&& ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) &&
230 !is_sync_kiocb(iocb
) &&
231 ext4_unaligned_aio(inode
, from
, iocb
->ki_pos
)) {
233 ext4_unwritten_wait(inode
);
236 iocb
->private = &overwrite
;
237 /* Check whether we do a DIO overwrite or not */
238 if (o_direct
&& ext4_should_dioread_nolock(inode
) && !unaligned_aio
&&
239 ext4_overwrite_io(inode
, iocb
->ki_pos
, iov_iter_count(from
)))
242 ret
= __generic_file_write_iter(iocb
, from
);
246 ret
= generic_write_sync(iocb
, ret
);
256 static int ext4_dax_huge_fault(struct vm_fault
*vmf
,
257 enum page_entry_size pe_size
)
260 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
261 struct super_block
*sb
= inode
->i_sb
;
262 bool write
= vmf
->flags
& FAULT_FLAG_WRITE
;
265 sb_start_pagefault(sb
);
266 file_update_time(vmf
->vma
->vm_file
);
268 down_read(&EXT4_I(inode
)->i_mmap_sem
);
269 result
= dax_iomap_fault(vmf
, pe_size
, &ext4_iomap_ops
);
270 up_read(&EXT4_I(inode
)->i_mmap_sem
);
272 sb_end_pagefault(sb
);
277 static int ext4_dax_fault(struct vm_fault
*vmf
)
279 return ext4_dax_huge_fault(vmf
, PE_SIZE_PTE
);
283 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_fault()
284 * handler we check for races agaist truncate. Note that since we cycle through
285 * i_mmap_sem, we are sure that also any hole punching that began before we
286 * were called is finished by now and so if it included part of the file we
287 * are working on, our pte will get unmapped and the check for pte_same() in
288 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
291 static int ext4_dax_pfn_mkwrite(struct vm_fault
*vmf
)
293 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
294 struct super_block
*sb
= inode
->i_sb
;
298 sb_start_pagefault(sb
);
299 file_update_time(vmf
->vma
->vm_file
);
300 down_read(&EXT4_I(inode
)->i_mmap_sem
);
301 size
= (i_size_read(inode
) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
302 if (vmf
->pgoff
>= size
)
303 ret
= VM_FAULT_SIGBUS
;
305 ret
= dax_pfn_mkwrite(vmf
);
306 up_read(&EXT4_I(inode
)->i_mmap_sem
);
307 sb_end_pagefault(sb
);
312 static const struct vm_operations_struct ext4_dax_vm_ops
= {
313 .fault
= ext4_dax_fault
,
314 .huge_fault
= ext4_dax_huge_fault
,
315 .page_mkwrite
= ext4_dax_fault
,
316 .pfn_mkwrite
= ext4_dax_pfn_mkwrite
,
319 #define ext4_dax_vm_ops ext4_file_vm_ops
322 static const struct vm_operations_struct ext4_file_vm_ops
= {
323 .fault
= ext4_filemap_fault
,
324 .map_pages
= filemap_map_pages
,
325 .page_mkwrite
= ext4_page_mkwrite
,
328 static int ext4_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
330 struct inode
*inode
= file
->f_mapping
->host
;
332 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
335 if (ext4_encrypted_inode(inode
)) {
336 int err
= fscrypt_get_encryption_info(inode
);
339 if (!fscrypt_has_encryption_key(inode
))
343 if (IS_DAX(file_inode(file
))) {
344 vma
->vm_ops
= &ext4_dax_vm_ops
;
345 vma
->vm_flags
|= VM_MIXEDMAP
| VM_HUGEPAGE
;
347 vma
->vm_ops
= &ext4_file_vm_ops
;
352 static int ext4_file_open(struct inode
* inode
, struct file
* filp
)
354 struct super_block
*sb
= inode
->i_sb
;
355 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
356 struct vfsmount
*mnt
= filp
->f_path
.mnt
;
362 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
365 if (unlikely(!(sbi
->s_mount_flags
& EXT4_MF_MNTDIR_SAMPLED
) &&
366 !(sb
->s_flags
& MS_RDONLY
))) {
367 sbi
->s_mount_flags
|= EXT4_MF_MNTDIR_SAMPLED
;
369 * Sample where the filesystem has been mounted and
370 * store it in the superblock for sysadmin convenience
371 * when trying to sort through large numbers of block
372 * devices or filesystem images.
374 memset(buf
, 0, sizeof(buf
));
376 path
.dentry
= mnt
->mnt_root
;
377 cp
= d_path(&path
, buf
, sizeof(buf
));
382 handle
= ext4_journal_start_sb(sb
, EXT4_HT_MISC
, 1);
384 return PTR_ERR(handle
);
385 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
386 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
388 ext4_journal_stop(handle
);
391 strlcpy(sbi
->s_es
->s_last_mounted
, cp
,
392 sizeof(sbi
->s_es
->s_last_mounted
));
393 ext4_handle_dirty_super(handle
, sb
);
394 ext4_journal_stop(handle
);
397 if (ext4_encrypted_inode(inode
)) {
398 ret
= fscrypt_get_encryption_info(inode
);
401 if (!fscrypt_has_encryption_key(inode
))
405 dir
= dget_parent(file_dentry(filp
));
406 if (ext4_encrypted_inode(d_inode(dir
)) &&
407 !fscrypt_has_permitted_context(d_inode(dir
), inode
)) {
408 ext4_warning(inode
->i_sb
,
409 "Inconsistent encryption contexts: %lu/%lu",
410 (unsigned long) d_inode(dir
)->i_ino
,
411 (unsigned long) inode
->i_ino
);
417 * Set up the jbd2_inode if we are opening the inode for
418 * writing and the journal is present
420 if (filp
->f_mode
& FMODE_WRITE
) {
421 ret
= ext4_inode_attach_jinode(inode
);
425 return dquot_file_open(inode
, filp
);
429 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
430 * file rather than ext4_ext_walk_space() because we can introduce
431 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
432 * function. When extent status tree has been fully implemented, it will
433 * track all extent status for a file and we can directly use it to
434 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
438 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
439 * lookup page cache to check whether or not there has some data between
440 * [startoff, endoff] because, if this range contains an unwritten extent,
441 * we determine this extent as a data or a hole according to whether the
442 * page cache has data or not.
444 static int ext4_find_unwritten_pgoff(struct inode
*inode
,
450 unsigned int blkbits
;
458 blkbits
= inode
->i_sb
->s_blocksize_bits
;
461 endoff
= (loff_t
)end_blk
<< blkbits
;
463 index
= startoff
>> PAGE_SHIFT
;
464 end
= endoff
>> PAGE_SHIFT
;
466 pagevec_init(&pvec
, 0);
469 unsigned long nr_pages
;
471 num
= min_t(pgoff_t
, end
- index
, PAGEVEC_SIZE
);
472 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, index
,
475 if (whence
== SEEK_DATA
)
478 BUG_ON(whence
!= SEEK_HOLE
);
480 * If this is the first time to go into the loop and
481 * offset is not beyond the end offset, it will be a
482 * hole at this offset
484 if (lastoff
== startoff
|| lastoff
< endoff
)
490 * If this is the first time to go into the loop and
491 * offset is smaller than the first page offset, it will be a
492 * hole at this offset.
494 if (lastoff
== startoff
&& whence
== SEEK_HOLE
&&
495 lastoff
< page_offset(pvec
.pages
[0])) {
500 for (i
= 0; i
< nr_pages
; i
++) {
501 struct page
*page
= pvec
.pages
[i
];
502 struct buffer_head
*bh
, *head
;
505 * If the current offset is not beyond the end of given
506 * range, it will be a hole.
508 if (lastoff
< endoff
&& whence
== SEEK_HOLE
&&
517 if (unlikely(page
->mapping
!= inode
->i_mapping
)) {
522 if (!page_has_buffers(page
)) {
527 if (page_has_buffers(page
)) {
528 lastoff
= page_offset(page
);
529 bh
= head
= page_buffers(page
);
531 if (buffer_uptodate(bh
) ||
532 buffer_unwritten(bh
)) {
533 if (whence
== SEEK_DATA
)
536 if (whence
== SEEK_HOLE
)
540 *offset
= max_t(loff_t
,
545 lastoff
+= bh
->b_size
;
546 bh
= bh
->b_this_page
;
547 } while (bh
!= head
);
550 lastoff
= page_offset(page
) + PAGE_SIZE
;
555 * The no. of pages is less than our desired, that would be a
558 if (nr_pages
< num
&& whence
== SEEK_HOLE
) {
564 index
= pvec
.pages
[i
- 1]->index
+ 1;
565 pagevec_release(&pvec
);
566 } while (index
<= end
);
569 pagevec_release(&pvec
);
574 * ext4_seek_data() retrieves the offset for SEEK_DATA.
576 static loff_t
ext4_seek_data(struct file
*file
, loff_t offset
, loff_t maxsize
)
578 struct inode
*inode
= file
->f_mapping
->host
;
579 struct extent_status es
;
580 ext4_lblk_t start
, last
, end
;
581 loff_t dataoff
, isize
;
587 isize
= i_size_read(inode
);
588 if (offset
>= isize
) {
593 blkbits
= inode
->i_sb
->s_blocksize_bits
;
594 start
= offset
>> blkbits
;
596 end
= isize
>> blkbits
;
600 ret
= ext4_get_next_extent(inode
, last
, end
- last
+ 1, &es
);
602 /* No extent found -> no data */
611 dataoff
= (loff_t
)last
<< blkbits
;
612 if (!ext4_es_is_unwritten(&es
))
616 * If there is a unwritten extent at this offset,
617 * it will be as a data or a hole according to page
618 * cache that has data or not.
620 if (ext4_find_unwritten_pgoff(inode
, SEEK_DATA
,
621 es
.es_lblk
+ es
.es_len
, &dataoff
))
624 dataoff
= (loff_t
)last
<< blkbits
;
626 } while (last
<= end
);
633 return vfs_setpos(file
, dataoff
, maxsize
);
637 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
639 static loff_t
ext4_seek_hole(struct file
*file
, loff_t offset
, loff_t maxsize
)
641 struct inode
*inode
= file
->f_mapping
->host
;
642 struct extent_status es
;
643 ext4_lblk_t start
, last
, end
;
644 loff_t holeoff
, isize
;
650 isize
= i_size_read(inode
);
651 if (offset
>= isize
) {
656 blkbits
= inode
->i_sb
->s_blocksize_bits
;
657 start
= offset
>> blkbits
;
659 end
= isize
>> blkbits
;
663 ret
= ext4_get_next_extent(inode
, last
, end
- last
+ 1, &es
);
669 if (ret
== 0 || es
.es_lblk
> last
) {
671 holeoff
= (loff_t
)last
<< blkbits
;
675 * If there is a unwritten extent at this offset,
676 * it will be as a data or a hole according to page
677 * cache that has data or not.
679 if (ext4_es_is_unwritten(&es
) &&
680 ext4_find_unwritten_pgoff(inode
, SEEK_HOLE
,
681 last
+ es
.es_len
, &holeoff
))
685 holeoff
= (loff_t
)last
<< blkbits
;
687 } while (last
<= end
);
694 return vfs_setpos(file
, holeoff
, maxsize
);
698 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
699 * by calling generic_file_llseek_size() with the appropriate maxbytes
702 loff_t
ext4_llseek(struct file
*file
, loff_t offset
, int whence
)
704 struct inode
*inode
= file
->f_mapping
->host
;
707 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
708 maxbytes
= EXT4_SB(inode
->i_sb
)->s_bitmap_maxbytes
;
710 maxbytes
= inode
->i_sb
->s_maxbytes
;
716 return generic_file_llseek_size(file
, offset
, whence
,
717 maxbytes
, i_size_read(inode
));
719 return ext4_seek_data(file
, offset
, maxbytes
);
721 return ext4_seek_hole(file
, offset
, maxbytes
);
727 const struct file_operations ext4_file_operations
= {
728 .llseek
= ext4_llseek
,
729 .read_iter
= ext4_file_read_iter
,
730 .write_iter
= ext4_file_write_iter
,
731 .unlocked_ioctl
= ext4_ioctl
,
733 .compat_ioctl
= ext4_compat_ioctl
,
735 .mmap
= ext4_file_mmap
,
736 .open
= ext4_file_open
,
737 .release
= ext4_release_file
,
738 .fsync
= ext4_sync_file
,
739 .get_unmapped_area
= thp_get_unmapped_area
,
740 .splice_read
= generic_file_splice_read
,
741 .splice_write
= iter_file_splice_write
,
742 .fallocate
= ext4_fallocate
,
745 const struct inode_operations ext4_file_inode_operations
= {
746 .setattr
= ext4_setattr
,
747 .getattr
= ext4_getattr
,
748 .listxattr
= ext4_listxattr
,
749 .get_acl
= ext4_get_acl
,
750 .set_acl
= ext4_set_acl
,
751 .fiemap
= ext4_fiemap
,