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 * 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.
39 static int ext4_release_file(struct inode
*inode
, struct file
*filp
)
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
);
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
)
50 down_write(&EXT4_I(inode
)->i_data_sem
);
51 ext4_discard_preallocations(inode
);
52 up_write(&EXT4_I(inode
)->i_data_sem
);
54 if (is_dx(inode
) && filp
->private_data
)
55 ext4_htree_free_dir_info(filp
->private_data
);
60 static void ext4_unwritten_wait(struct inode
*inode
)
62 wait_queue_head_t
*wq
= ext4_ioend_wq(inode
);
64 wait_event(*wq
, (atomic_read(&EXT4_I(inode
)->i_unwritten
) == 0));
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.
77 ext4_unaligned_aio(struct inode
*inode
, struct iov_iter
*from
, loff_t pos
)
79 struct super_block
*sb
= inode
->i_sb
;
80 int blockmask
= sb
->s_blocksize
- 1;
82 if (pos
>= i_size_read(inode
))
85 if ((pos
| iov_iter_alignment(from
)) & blockmask
)
92 ext4_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
94 struct file
*file
= iocb
->ki_filp
;
95 struct inode
*inode
= file_inode(iocb
->ki_filp
);
96 struct mutex
*aio_mutex
= NULL
;
98 int o_direct
= iocb
->ki_flags
& IOCB_DIRECT
;
103 * Unaligned direct AIO must be serialized; see comment above
104 * In the case of O_APPEND, assume that we must always serialize
107 ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) &&
108 !is_sync_kiocb(iocb
) &&
109 (iocb
->ki_flags
& IOCB_APPEND
||
110 ext4_unaligned_aio(inode
, from
, iocb
->ki_pos
))) {
111 aio_mutex
= ext4_aio_mutex(inode
);
112 mutex_lock(aio_mutex
);
113 ext4_unwritten_wait(inode
);
116 mutex_lock(&inode
->i_mutex
);
117 ret
= generic_write_checks(iocb
, from
);
122 * If we have encountered a bitmap-format file, the size limit
123 * is smaller than s_maxbytes, which is for extent-mapped files.
125 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
126 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
128 if (iocb
->ki_pos
>= sbi
->s_bitmap_maxbytes
) {
132 iov_iter_truncate(from
, sbi
->s_bitmap_maxbytes
- iocb
->ki_pos
);
135 iocb
->private = &overwrite
;
137 size_t length
= iov_iter_count(from
);
138 loff_t pos
= iocb
->ki_pos
;
139 blk_start_plug(&plug
);
141 /* check whether we do a DIO overwrite or not */
142 if (ext4_should_dioread_nolock(inode
) && !aio_mutex
&&
143 !file
->f_mapping
->nrpages
&& pos
+ length
<= i_size_read(inode
)) {
144 struct ext4_map_blocks map
;
145 unsigned int blkbits
= inode
->i_blkbits
;
148 map
.m_lblk
= pos
>> blkbits
;
149 map
.m_len
= (EXT4_BLOCK_ALIGN(pos
+ length
, blkbits
) >> blkbits
)
153 err
= ext4_map_blocks(NULL
, inode
, &map
, 0);
155 * 'err==len' means that all of blocks has
156 * been preallocated no matter they are
157 * initialized or not. For excluding
158 * unwritten extents, we need to check
159 * m_flags. There are two conditions that
160 * indicate for initialized extents. 1) If we
161 * hit extent cache, EXT4_MAP_MAPPED flag is
162 * returned; 2) If we do a real lookup,
163 * non-flags are returned. So we should check
164 * these two conditions.
166 if (err
== len
&& (map
.m_flags
& EXT4_MAP_MAPPED
))
171 ret
= __generic_file_write_iter(iocb
, from
);
172 mutex_unlock(&inode
->i_mutex
);
177 err
= generic_write_sync(file
, iocb
->ki_pos
- ret
, ret
);
182 blk_finish_plug(&plug
);
185 mutex_unlock(aio_mutex
);
189 mutex_unlock(&inode
->i_mutex
);
191 mutex_unlock(aio_mutex
);
196 static int ext4_dax_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
199 handle_t
*handle
= NULL
;
200 struct inode
*inode
= file_inode(vma
->vm_file
);
201 struct super_block
*sb
= inode
->i_sb
;
202 bool write
= vmf
->flags
& FAULT_FLAG_WRITE
;
205 sb_start_pagefault(sb
);
206 file_update_time(vma
->vm_file
);
207 down_read(&EXT4_I(inode
)->i_mmap_sem
);
208 handle
= ext4_journal_start_sb(sb
, EXT4_HT_WRITE_PAGE
,
209 EXT4_DATA_TRANS_BLOCKS(sb
));
211 down_read(&EXT4_I(inode
)->i_mmap_sem
);
214 result
= VM_FAULT_SIGBUS
;
216 result
= __dax_fault(vma
, vmf
, ext4_dax_mmap_get_block
, NULL
);
220 ext4_journal_stop(handle
);
221 up_read(&EXT4_I(inode
)->i_mmap_sem
);
222 sb_end_pagefault(sb
);
224 up_read(&EXT4_I(inode
)->i_mmap_sem
);
229 static int ext4_dax_pmd_fault(struct vm_area_struct
*vma
, unsigned long addr
,
230 pmd_t
*pmd
, unsigned int flags
)
233 handle_t
*handle
= NULL
;
234 struct inode
*inode
= file_inode(vma
->vm_file
);
235 struct super_block
*sb
= inode
->i_sb
;
236 bool write
= flags
& FAULT_FLAG_WRITE
;
239 sb_start_pagefault(sb
);
240 file_update_time(vma
->vm_file
);
241 down_read(&EXT4_I(inode
)->i_mmap_sem
);
242 handle
= ext4_journal_start_sb(sb
, EXT4_HT_WRITE_PAGE
,
243 ext4_chunk_trans_blocks(inode
,
244 PMD_SIZE
/ PAGE_SIZE
));
246 down_read(&EXT4_I(inode
)->i_mmap_sem
);
249 result
= VM_FAULT_SIGBUS
;
251 result
= __dax_pmd_fault(vma
, addr
, pmd
, flags
,
252 ext4_dax_mmap_get_block
, NULL
);
256 ext4_journal_stop(handle
);
257 up_read(&EXT4_I(inode
)->i_mmap_sem
);
258 sb_end_pagefault(sb
);
260 up_read(&EXT4_I(inode
)->i_mmap_sem
);
265 static int ext4_dax_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
268 struct inode
*inode
= file_inode(vma
->vm_file
);
270 sb_start_pagefault(inode
->i_sb
);
271 file_update_time(vma
->vm_file
);
272 down_read(&EXT4_I(inode
)->i_mmap_sem
);
273 err
= __dax_mkwrite(vma
, vmf
, ext4_dax_mmap_get_block
, NULL
);
274 up_read(&EXT4_I(inode
)->i_mmap_sem
);
275 sb_end_pagefault(inode
->i_sb
);
281 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_mkwrite()
282 * handler we check for races agaist truncate. Note that since we cycle through
283 * i_mmap_sem, we are sure that also any hole punching that began before we
284 * were called is finished by now and so if it included part of the file we
285 * are working on, our pte will get unmapped and the check for pte_same() in
286 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
289 static int ext4_dax_pfn_mkwrite(struct vm_area_struct
*vma
,
290 struct vm_fault
*vmf
)
292 struct inode
*inode
= file_inode(vma
->vm_file
);
293 struct super_block
*sb
= inode
->i_sb
;
294 int ret
= VM_FAULT_NOPAGE
;
297 sb_start_pagefault(sb
);
298 file_update_time(vma
->vm_file
);
299 down_read(&EXT4_I(inode
)->i_mmap_sem
);
300 size
= (i_size_read(inode
) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
301 if (vmf
->pgoff
>= size
)
302 ret
= VM_FAULT_SIGBUS
;
303 up_read(&EXT4_I(inode
)->i_mmap_sem
);
304 sb_end_pagefault(sb
);
309 static const struct vm_operations_struct ext4_dax_vm_ops
= {
310 .fault
= ext4_dax_fault
,
311 .pmd_fault
= ext4_dax_pmd_fault
,
312 .page_mkwrite
= ext4_dax_mkwrite
,
313 .pfn_mkwrite
= ext4_dax_pfn_mkwrite
,
316 #define ext4_dax_vm_ops ext4_file_vm_ops
319 static const struct vm_operations_struct ext4_file_vm_ops
= {
320 .fault
= ext4_filemap_fault
,
321 .map_pages
= filemap_map_pages
,
322 .page_mkwrite
= ext4_page_mkwrite
,
325 static int ext4_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
327 struct inode
*inode
= file
->f_mapping
->host
;
329 if (ext4_encrypted_inode(inode
)) {
330 int err
= ext4_get_encryption_info(inode
);
333 if (ext4_encryption_info(inode
) == NULL
)
337 if (IS_DAX(file_inode(file
))) {
338 vma
->vm_ops
= &ext4_dax_vm_ops
;
339 vma
->vm_flags
|= VM_MIXEDMAP
| VM_HUGEPAGE
;
341 vma
->vm_ops
= &ext4_file_vm_ops
;
346 static int ext4_file_open(struct inode
* inode
, struct file
* filp
)
348 struct super_block
*sb
= inode
->i_sb
;
349 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
350 struct vfsmount
*mnt
= filp
->f_path
.mnt
;
355 if (unlikely(!(sbi
->s_mount_flags
& EXT4_MF_MNTDIR_SAMPLED
) &&
356 !(sb
->s_flags
& MS_RDONLY
))) {
357 sbi
->s_mount_flags
|= EXT4_MF_MNTDIR_SAMPLED
;
359 * Sample where the filesystem has been mounted and
360 * store it in the superblock for sysadmin convenience
361 * when trying to sort through large numbers of block
362 * devices or filesystem images.
364 memset(buf
, 0, sizeof(buf
));
366 path
.dentry
= mnt
->mnt_root
;
367 cp
= d_path(&path
, buf
, sizeof(buf
));
372 handle
= ext4_journal_start_sb(sb
, EXT4_HT_MISC
, 1);
374 return PTR_ERR(handle
);
375 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
376 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
378 ext4_journal_stop(handle
);
381 strlcpy(sbi
->s_es
->s_last_mounted
, cp
,
382 sizeof(sbi
->s_es
->s_last_mounted
));
383 ext4_handle_dirty_super(handle
, sb
);
384 ext4_journal_stop(handle
);
387 if (ext4_encrypted_inode(inode
)) {
388 ret
= ext4_get_encryption_info(inode
);
391 if (ext4_encryption_info(inode
) == NULL
)
395 * Set up the jbd2_inode if we are opening the inode for
396 * writing and the journal is present
398 if (filp
->f_mode
& FMODE_WRITE
) {
399 ret
= ext4_inode_attach_jinode(inode
);
403 return dquot_file_open(inode
, filp
);
407 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
408 * file rather than ext4_ext_walk_space() because we can introduce
409 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
410 * function. When extent status tree has been fully implemented, it will
411 * track all extent status for a file and we can directly use it to
412 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
416 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
417 * lookup page cache to check whether or not there has some data between
418 * [startoff, endoff] because, if this range contains an unwritten extent,
419 * we determine this extent as a data or a hole according to whether the
420 * page cache has data or not.
422 static int ext4_find_unwritten_pgoff(struct inode
*inode
,
424 struct ext4_map_blocks
*map
,
428 unsigned int blkbits
;
436 blkbits
= inode
->i_sb
->s_blocksize_bits
;
439 endoff
= (loff_t
)(map
->m_lblk
+ map
->m_len
) << blkbits
;
441 index
= startoff
>> PAGE_CACHE_SHIFT
;
442 end
= endoff
>> PAGE_CACHE_SHIFT
;
444 pagevec_init(&pvec
, 0);
447 unsigned long nr_pages
;
449 num
= min_t(pgoff_t
, end
- index
, PAGEVEC_SIZE
);
450 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, index
,
453 if (whence
== SEEK_DATA
)
456 BUG_ON(whence
!= SEEK_HOLE
);
458 * If this is the first time to go into the loop and
459 * offset is not beyond the end offset, it will be a
460 * hole at this offset
462 if (lastoff
== startoff
|| lastoff
< endoff
)
468 * If this is the first time to go into the loop and
469 * offset is smaller than the first page offset, it will be a
470 * hole at this offset.
472 if (lastoff
== startoff
&& whence
== SEEK_HOLE
&&
473 lastoff
< page_offset(pvec
.pages
[0])) {
478 for (i
= 0; i
< nr_pages
; i
++) {
479 struct page
*page
= pvec
.pages
[i
];
480 struct buffer_head
*bh
, *head
;
483 * If the current offset is not beyond the end of given
484 * range, it will be a hole.
486 if (lastoff
< endoff
&& whence
== SEEK_HOLE
&&
495 if (unlikely(page
->mapping
!= inode
->i_mapping
)) {
500 if (!page_has_buffers(page
)) {
505 if (page_has_buffers(page
)) {
506 lastoff
= page_offset(page
);
507 bh
= head
= page_buffers(page
);
509 if (buffer_uptodate(bh
) ||
510 buffer_unwritten(bh
)) {
511 if (whence
== SEEK_DATA
)
514 if (whence
== SEEK_HOLE
)
518 *offset
= max_t(loff_t
,
523 lastoff
+= bh
->b_size
;
524 bh
= bh
->b_this_page
;
525 } while (bh
!= head
);
528 lastoff
= page_offset(page
) + PAGE_SIZE
;
533 * The no. of pages is less than our desired, that would be a
536 if (nr_pages
< num
&& whence
== SEEK_HOLE
) {
542 index
= pvec
.pages
[i
- 1]->index
+ 1;
543 pagevec_release(&pvec
);
544 } while (index
<= end
);
547 pagevec_release(&pvec
);
552 * ext4_seek_data() retrieves the offset for SEEK_DATA.
554 static loff_t
ext4_seek_data(struct file
*file
, loff_t offset
, loff_t maxsize
)
556 struct inode
*inode
= file
->f_mapping
->host
;
557 struct ext4_map_blocks map
;
558 struct extent_status es
;
559 ext4_lblk_t start
, last
, end
;
560 loff_t dataoff
, isize
;
564 mutex_lock(&inode
->i_mutex
);
566 isize
= i_size_read(inode
);
567 if (offset
>= isize
) {
568 mutex_unlock(&inode
->i_mutex
);
572 blkbits
= inode
->i_sb
->s_blocksize_bits
;
573 start
= offset
>> blkbits
;
575 end
= isize
>> blkbits
;
580 map
.m_len
= end
- last
+ 1;
581 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
582 if (ret
> 0 && !(map
.m_flags
& EXT4_MAP_UNWRITTEN
)) {
584 dataoff
= (loff_t
)last
<< blkbits
;
589 * If there is a delay extent at this offset,
590 * it will be as a data.
592 ext4_es_find_delayed_extent_range(inode
, last
, last
, &es
);
593 if (es
.es_len
!= 0 && in_range(last
, es
.es_lblk
, es
.es_len
)) {
595 dataoff
= (loff_t
)last
<< blkbits
;
600 * If there is a unwritten extent at this offset,
601 * it will be as a data or a hole according to page
602 * cache that has data or not.
604 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
606 unwritten
= ext4_find_unwritten_pgoff(inode
, SEEK_DATA
,
613 dataoff
= (loff_t
)last
<< blkbits
;
614 } while (last
<= end
);
616 mutex_unlock(&inode
->i_mutex
);
621 return vfs_setpos(file
, dataoff
, maxsize
);
625 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
627 static loff_t
ext4_seek_hole(struct file
*file
, loff_t offset
, loff_t maxsize
)
629 struct inode
*inode
= file
->f_mapping
->host
;
630 struct ext4_map_blocks map
;
631 struct extent_status es
;
632 ext4_lblk_t start
, last
, end
;
633 loff_t holeoff
, isize
;
637 mutex_lock(&inode
->i_mutex
);
639 isize
= i_size_read(inode
);
640 if (offset
>= isize
) {
641 mutex_unlock(&inode
->i_mutex
);
645 blkbits
= inode
->i_sb
->s_blocksize_bits
;
646 start
= offset
>> blkbits
;
648 end
= isize
>> blkbits
;
653 map
.m_len
= end
- last
+ 1;
654 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
655 if (ret
> 0 && !(map
.m_flags
& EXT4_MAP_UNWRITTEN
)) {
657 holeoff
= (loff_t
)last
<< blkbits
;
662 * If there is a delay extent at this offset,
663 * we will skip this extent.
665 ext4_es_find_delayed_extent_range(inode
, last
, last
, &es
);
666 if (es
.es_len
!= 0 && in_range(last
, es
.es_lblk
, es
.es_len
)) {
667 last
= es
.es_lblk
+ es
.es_len
;
668 holeoff
= (loff_t
)last
<< blkbits
;
673 * If there is a unwritten extent at this offset,
674 * it will be as a data or a hole according to page
675 * cache that has data or not.
677 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
679 unwritten
= ext4_find_unwritten_pgoff(inode
, SEEK_HOLE
,
683 holeoff
= (loff_t
)last
<< blkbits
;
690 } while (last
<= end
);
692 mutex_unlock(&inode
->i_mutex
);
697 return vfs_setpos(file
, holeoff
, maxsize
);
701 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
702 * by calling generic_file_llseek_size() with the appropriate maxbytes
705 loff_t
ext4_llseek(struct file
*file
, loff_t offset
, int whence
)
707 struct inode
*inode
= file
->f_mapping
->host
;
710 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
711 maxbytes
= EXT4_SB(inode
->i_sb
)->s_bitmap_maxbytes
;
713 maxbytes
= inode
->i_sb
->s_maxbytes
;
719 return generic_file_llseek_size(file
, offset
, whence
,
720 maxbytes
, i_size_read(inode
));
722 return ext4_seek_data(file
, offset
, maxbytes
);
724 return ext4_seek_hole(file
, offset
, maxbytes
);
730 const struct file_operations ext4_file_operations
= {
731 .llseek
= ext4_llseek
,
732 .read_iter
= generic_file_read_iter
,
733 .write_iter
= ext4_file_write_iter
,
734 .unlocked_ioctl
= ext4_ioctl
,
736 .compat_ioctl
= ext4_compat_ioctl
,
738 .mmap
= ext4_file_mmap
,
739 .open
= ext4_file_open
,
740 .release
= ext4_release_file
,
741 .fsync
= ext4_sync_file
,
742 .splice_read
= generic_file_splice_read
,
743 .splice_write
= iter_file_splice_write
,
744 .fallocate
= ext4_fallocate
,
747 const struct inode_operations ext4_file_inode_operations
= {
748 .setattr
= ext4_setattr
,
749 .getattr
= ext4_getattr
,
750 .setxattr
= generic_setxattr
,
751 .getxattr
= generic_getxattr
,
752 .listxattr
= ext4_listxattr
,
753 .removexattr
= generic_removexattr
,
754 .get_acl
= ext4_get_acl
,
755 .set_acl
= ext4_set_acl
,
756 .fiemap
= ext4_fiemap
,