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/quotaops.h>
26 #include <linux/pagevec.h>
27 #include <linux/uio.h>
29 #include "ext4_jbd2.h"
34 * Called when an inode is released. Note that this is different
35 * from ext4_file_open: open gets called at every open, but release
36 * gets called only when /all/ the files are closed.
38 static int ext4_release_file(struct inode
*inode
, struct file
*filp
)
40 if (ext4_test_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
)) {
41 ext4_alloc_da_blocks(inode
);
42 ext4_clear_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
44 /* if we are the last writer on the inode, drop the block reservation */
45 if ((filp
->f_mode
& FMODE_WRITE
) &&
46 (atomic_read(&inode
->i_writecount
) == 1) &&
47 !EXT4_I(inode
)->i_reserved_data_blocks
)
49 down_write(&EXT4_I(inode
)->i_data_sem
);
50 ext4_discard_preallocations(inode
);
51 up_write(&EXT4_I(inode
)->i_data_sem
);
53 if (is_dx(inode
) && filp
->private_data
)
54 ext4_htree_free_dir_info(filp
->private_data
);
59 static void ext4_unwritten_wait(struct inode
*inode
)
61 wait_queue_head_t
*wq
= ext4_ioend_wq(inode
);
63 wait_event(*wq
, (atomic_read(&EXT4_I(inode
)->i_unwritten
) == 0));
67 * This tests whether the IO in question is block-aligned or not.
68 * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
69 * are converted to written only after the IO is complete. Until they are
70 * mapped, these blocks appear as holes, so dio_zero_block() will assume that
71 * it needs to zero out portions of the start and/or end block. If 2 AIO
72 * threads are at work on the same unwritten block, they must be synchronized
73 * or one thread will zero the other's data, causing corruption.
76 ext4_unaligned_aio(struct inode
*inode
, struct iov_iter
*from
, loff_t pos
)
78 struct super_block
*sb
= inode
->i_sb
;
79 int blockmask
= sb
->s_blocksize
- 1;
81 if (pos
>= i_size_read(inode
))
84 if ((pos
| iov_iter_alignment(from
)) & blockmask
)
91 ext4_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
93 struct file
*file
= iocb
->ki_filp
;
94 struct inode
*inode
= file_inode(iocb
->ki_filp
);
95 struct mutex
*aio_mutex
= NULL
;
97 int o_direct
= iocb
->ki_flags
& IOCB_DIRECT
;
102 * Unaligned direct AIO must be serialized; see comment above
103 * In the case of O_APPEND, assume that we must always serialize
106 ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) &&
107 !is_sync_kiocb(iocb
) &&
108 (iocb
->ki_flags
& IOCB_APPEND
||
109 ext4_unaligned_aio(inode
, from
, iocb
->ki_pos
))) {
110 aio_mutex
= ext4_aio_mutex(inode
);
111 mutex_lock(aio_mutex
);
112 ext4_unwritten_wait(inode
);
115 mutex_lock(&inode
->i_mutex
);
116 ret
= generic_write_checks(iocb
, from
);
121 * If we have encountered a bitmap-format file, the size limit
122 * is smaller than s_maxbytes, which is for extent-mapped files.
124 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
125 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
127 if (iocb
->ki_pos
>= sbi
->s_bitmap_maxbytes
) {
131 iov_iter_truncate(from
, sbi
->s_bitmap_maxbytes
- iocb
->ki_pos
);
134 iocb
->private = &overwrite
;
136 size_t length
= iov_iter_count(from
);
137 loff_t pos
= iocb
->ki_pos
;
138 blk_start_plug(&plug
);
140 /* check whether we do a DIO overwrite or not */
141 if (ext4_should_dioread_nolock(inode
) && !aio_mutex
&&
142 !file
->f_mapping
->nrpages
&& pos
+ length
<= i_size_read(inode
)) {
143 struct ext4_map_blocks map
;
144 unsigned int blkbits
= inode
->i_blkbits
;
147 map
.m_lblk
= pos
>> blkbits
;
148 map
.m_len
= (EXT4_BLOCK_ALIGN(pos
+ length
, blkbits
) >> blkbits
)
152 err
= ext4_map_blocks(NULL
, inode
, &map
, 0);
154 * 'err==len' means that all of blocks has
155 * been preallocated no matter they are
156 * initialized or not. For excluding
157 * unwritten extents, we need to check
158 * m_flags. There are two conditions that
159 * indicate for initialized extents. 1) If we
160 * hit extent cache, EXT4_MAP_MAPPED flag is
161 * returned; 2) If we do a real lookup,
162 * non-flags are returned. So we should check
163 * these two conditions.
165 if (err
== len
&& (map
.m_flags
& EXT4_MAP_MAPPED
))
170 ret
= __generic_file_write_iter(iocb
, from
);
171 mutex_unlock(&inode
->i_mutex
);
176 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
181 blk_finish_plug(&plug
);
184 mutex_unlock(aio_mutex
);
188 mutex_unlock(&inode
->i_mutex
);
190 mutex_unlock(aio_mutex
);
195 static int ext4_dax_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
197 return dax_fault(vma
, vmf
, ext4_get_block
);
198 /* Is this the right get_block? */
201 static int ext4_dax_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
203 return dax_mkwrite(vma
, vmf
, ext4_get_block
);
206 static const struct vm_operations_struct ext4_dax_vm_ops
= {
207 .fault
= ext4_dax_fault
,
208 .page_mkwrite
= ext4_dax_mkwrite
,
209 .pfn_mkwrite
= dax_pfn_mkwrite
,
212 #define ext4_dax_vm_ops ext4_file_vm_ops
215 static const struct vm_operations_struct ext4_file_vm_ops
= {
216 .fault
= filemap_fault
,
217 .map_pages
= filemap_map_pages
,
218 .page_mkwrite
= ext4_page_mkwrite
,
221 static int ext4_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
223 struct inode
*inode
= file
->f_mapping
->host
;
225 if (ext4_encrypted_inode(inode
)) {
226 int err
= ext4_generate_encryption_key(inode
);
231 if (IS_DAX(file_inode(file
))) {
232 vma
->vm_ops
= &ext4_dax_vm_ops
;
233 vma
->vm_flags
|= VM_MIXEDMAP
;
235 vma
->vm_ops
= &ext4_file_vm_ops
;
240 static int ext4_file_open(struct inode
* inode
, struct file
* filp
)
242 struct super_block
*sb
= inode
->i_sb
;
243 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
244 struct vfsmount
*mnt
= filp
->f_path
.mnt
;
249 if (unlikely(!(sbi
->s_mount_flags
& EXT4_MF_MNTDIR_SAMPLED
) &&
250 !(sb
->s_flags
& MS_RDONLY
))) {
251 sbi
->s_mount_flags
|= EXT4_MF_MNTDIR_SAMPLED
;
253 * Sample where the filesystem has been mounted and
254 * store it in the superblock for sysadmin convenience
255 * when trying to sort through large numbers of block
256 * devices or filesystem images.
258 memset(buf
, 0, sizeof(buf
));
260 path
.dentry
= mnt
->mnt_root
;
261 cp
= d_path(&path
, buf
, sizeof(buf
));
266 handle
= ext4_journal_start_sb(sb
, EXT4_HT_MISC
, 1);
268 return PTR_ERR(handle
);
269 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
270 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
272 ext4_journal_stop(handle
);
275 strlcpy(sbi
->s_es
->s_last_mounted
, cp
,
276 sizeof(sbi
->s_es
->s_last_mounted
));
277 ext4_handle_dirty_super(handle
, sb
);
278 ext4_journal_stop(handle
);
282 * Set up the jbd2_inode if we are opening the inode for
283 * writing and the journal is present
285 if (filp
->f_mode
& FMODE_WRITE
) {
286 ret
= ext4_inode_attach_jinode(inode
);
290 ret
= dquot_file_open(inode
, filp
);
291 if (!ret
&& ext4_encrypted_inode(inode
)) {
292 ret
= ext4_generate_encryption_key(inode
);
300 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
301 * file rather than ext4_ext_walk_space() because we can introduce
302 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
303 * function. When extent status tree has been fully implemented, it will
304 * track all extent status for a file and we can directly use it to
305 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
309 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
310 * lookup page cache to check whether or not there has some data between
311 * [startoff, endoff] because, if this range contains an unwritten extent,
312 * we determine this extent as a data or a hole according to whether the
313 * page cache has data or not.
315 static int ext4_find_unwritten_pgoff(struct inode
*inode
,
317 struct ext4_map_blocks
*map
,
321 unsigned int blkbits
;
329 blkbits
= inode
->i_sb
->s_blocksize_bits
;
332 endoff
= (loff_t
)(map
->m_lblk
+ map
->m_len
) << blkbits
;
334 index
= startoff
>> PAGE_CACHE_SHIFT
;
335 end
= endoff
>> PAGE_CACHE_SHIFT
;
337 pagevec_init(&pvec
, 0);
340 unsigned long nr_pages
;
342 num
= min_t(pgoff_t
, end
- index
, PAGEVEC_SIZE
);
343 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, index
,
346 if (whence
== SEEK_DATA
)
349 BUG_ON(whence
!= SEEK_HOLE
);
351 * If this is the first time to go into the loop and
352 * offset is not beyond the end offset, it will be a
353 * hole at this offset
355 if (lastoff
== startoff
|| lastoff
< endoff
)
361 * If this is the first time to go into the loop and
362 * offset is smaller than the first page offset, it will be a
363 * hole at this offset.
365 if (lastoff
== startoff
&& whence
== SEEK_HOLE
&&
366 lastoff
< page_offset(pvec
.pages
[0])) {
371 for (i
= 0; i
< nr_pages
; i
++) {
372 struct page
*page
= pvec
.pages
[i
];
373 struct buffer_head
*bh
, *head
;
376 * If the current offset is not beyond the end of given
377 * range, it will be a hole.
379 if (lastoff
< endoff
&& whence
== SEEK_HOLE
&&
388 if (unlikely(page
->mapping
!= inode
->i_mapping
)) {
393 if (!page_has_buffers(page
)) {
398 if (page_has_buffers(page
)) {
399 lastoff
= page_offset(page
);
400 bh
= head
= page_buffers(page
);
402 if (buffer_uptodate(bh
) ||
403 buffer_unwritten(bh
)) {
404 if (whence
== SEEK_DATA
)
407 if (whence
== SEEK_HOLE
)
411 *offset
= max_t(loff_t
,
416 lastoff
+= bh
->b_size
;
417 bh
= bh
->b_this_page
;
418 } while (bh
!= head
);
421 lastoff
= page_offset(page
) + PAGE_SIZE
;
426 * The no. of pages is less than our desired, that would be a
429 if (nr_pages
< num
&& whence
== SEEK_HOLE
) {
435 index
= pvec
.pages
[i
- 1]->index
+ 1;
436 pagevec_release(&pvec
);
437 } while (index
<= end
);
440 pagevec_release(&pvec
);
445 * ext4_seek_data() retrieves the offset for SEEK_DATA.
447 static loff_t
ext4_seek_data(struct file
*file
, loff_t offset
, loff_t maxsize
)
449 struct inode
*inode
= file
->f_mapping
->host
;
450 struct ext4_map_blocks map
;
451 struct extent_status es
;
452 ext4_lblk_t start
, last
, end
;
453 loff_t dataoff
, isize
;
457 mutex_lock(&inode
->i_mutex
);
459 isize
= i_size_read(inode
);
460 if (offset
>= isize
) {
461 mutex_unlock(&inode
->i_mutex
);
465 blkbits
= inode
->i_sb
->s_blocksize_bits
;
466 start
= offset
>> blkbits
;
468 end
= isize
>> blkbits
;
473 map
.m_len
= end
- last
+ 1;
474 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
475 if (ret
> 0 && !(map
.m_flags
& EXT4_MAP_UNWRITTEN
)) {
477 dataoff
= (loff_t
)last
<< blkbits
;
482 * If there is a delay extent at this offset,
483 * it will be as a data.
485 ext4_es_find_delayed_extent_range(inode
, last
, last
, &es
);
486 if (es
.es_len
!= 0 && in_range(last
, es
.es_lblk
, es
.es_len
)) {
488 dataoff
= (loff_t
)last
<< blkbits
;
493 * If there is a unwritten extent at this offset,
494 * it will be as a data or a hole according to page
495 * cache that has data or not.
497 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
499 unwritten
= ext4_find_unwritten_pgoff(inode
, SEEK_DATA
,
506 dataoff
= (loff_t
)last
<< blkbits
;
507 } while (last
<= end
);
509 mutex_unlock(&inode
->i_mutex
);
514 return vfs_setpos(file
, dataoff
, maxsize
);
518 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
520 static loff_t
ext4_seek_hole(struct file
*file
, loff_t offset
, loff_t maxsize
)
522 struct inode
*inode
= file
->f_mapping
->host
;
523 struct ext4_map_blocks map
;
524 struct extent_status es
;
525 ext4_lblk_t start
, last
, end
;
526 loff_t holeoff
, isize
;
530 mutex_lock(&inode
->i_mutex
);
532 isize
= i_size_read(inode
);
533 if (offset
>= isize
) {
534 mutex_unlock(&inode
->i_mutex
);
538 blkbits
= inode
->i_sb
->s_blocksize_bits
;
539 start
= offset
>> blkbits
;
541 end
= isize
>> blkbits
;
546 map
.m_len
= end
- last
+ 1;
547 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
548 if (ret
> 0 && !(map
.m_flags
& EXT4_MAP_UNWRITTEN
)) {
550 holeoff
= (loff_t
)last
<< blkbits
;
555 * If there is a delay extent at this offset,
556 * we will skip this extent.
558 ext4_es_find_delayed_extent_range(inode
, last
, last
, &es
);
559 if (es
.es_len
!= 0 && in_range(last
, es
.es_lblk
, es
.es_len
)) {
560 last
= es
.es_lblk
+ es
.es_len
;
561 holeoff
= (loff_t
)last
<< blkbits
;
566 * If there is a unwritten extent at this offset,
567 * it will be as a data or a hole according to page
568 * cache that has data or not.
570 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
572 unwritten
= ext4_find_unwritten_pgoff(inode
, SEEK_HOLE
,
576 holeoff
= (loff_t
)last
<< blkbits
;
583 } while (last
<= end
);
585 mutex_unlock(&inode
->i_mutex
);
590 return vfs_setpos(file
, holeoff
, maxsize
);
594 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
595 * by calling generic_file_llseek_size() with the appropriate maxbytes
598 loff_t
ext4_llseek(struct file
*file
, loff_t offset
, int whence
)
600 struct inode
*inode
= file
->f_mapping
->host
;
603 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
604 maxbytes
= EXT4_SB(inode
->i_sb
)->s_bitmap_maxbytes
;
606 maxbytes
= inode
->i_sb
->s_maxbytes
;
612 return generic_file_llseek_size(file
, offset
, whence
,
613 maxbytes
, i_size_read(inode
));
615 return ext4_seek_data(file
, offset
, maxbytes
);
617 return ext4_seek_hole(file
, offset
, maxbytes
);
623 const struct file_operations ext4_file_operations
= {
624 .llseek
= ext4_llseek
,
625 .read_iter
= generic_file_read_iter
,
626 .write_iter
= ext4_file_write_iter
,
627 .unlocked_ioctl
= ext4_ioctl
,
629 .compat_ioctl
= ext4_compat_ioctl
,
631 .mmap
= ext4_file_mmap
,
632 .open
= ext4_file_open
,
633 .release
= ext4_release_file
,
634 .fsync
= ext4_sync_file
,
635 .splice_read
= generic_file_splice_read
,
636 .splice_write
= iter_file_splice_write
,
637 .fallocate
= ext4_fallocate
,
640 const struct inode_operations ext4_file_inode_operations
= {
641 .setattr
= ext4_setattr
,
642 .getattr
= ext4_getattr
,
643 .setxattr
= generic_setxattr
,
644 .getxattr
= generic_getxattr
,
645 .listxattr
= ext4_listxattr
,
646 .removexattr
= generic_removexattr
,
647 .get_acl
= ext4_get_acl
,
648 .set_acl
= ext4_set_acl
,
649 .fiemap
= ext4_fiemap
,