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
>= ALIGN(i_size_read(inode
), sb
->s_blocksize
))
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 void ext4_end_io_unwritten(struct buffer_head
*bh
, int uptodate
)
198 struct inode
*inode
= bh
->b_assoc_map
->host
;
199 /* XXX: breaks on 32-bit > 16TB. Is that even supported? */
200 loff_t offset
= (loff_t
)(uintptr_t)bh
->b_private
<< inode
->i_blkbits
;
204 WARN_ON(!buffer_unwritten(bh
));
205 err
= ext4_convert_unwritten_extents(NULL
, inode
, offset
, bh
->b_size
);
208 static int ext4_dax_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
211 handle_t
*handle
= NULL
;
212 struct inode
*inode
= file_inode(vma
->vm_file
);
213 struct super_block
*sb
= inode
->i_sb
;
214 bool write
= vmf
->flags
& FAULT_FLAG_WRITE
;
217 sb_start_pagefault(sb
);
218 file_update_time(vma
->vm_file
);
219 down_read(&EXT4_I(inode
)->i_mmap_sem
);
220 handle
= ext4_journal_start_sb(sb
, EXT4_HT_WRITE_PAGE
,
221 EXT4_DATA_TRANS_BLOCKS(sb
));
223 down_read(&EXT4_I(inode
)->i_mmap_sem
);
226 result
= VM_FAULT_SIGBUS
;
228 result
= __dax_fault(vma
, vmf
, ext4_get_block_dax
,
229 ext4_end_io_unwritten
);
233 ext4_journal_stop(handle
);
234 up_read(&EXT4_I(inode
)->i_mmap_sem
);
235 sb_end_pagefault(sb
);
237 up_read(&EXT4_I(inode
)->i_mmap_sem
);
242 static int ext4_dax_pmd_fault(struct vm_area_struct
*vma
, unsigned long addr
,
243 pmd_t
*pmd
, unsigned int flags
)
246 handle_t
*handle
= NULL
;
247 struct inode
*inode
= file_inode(vma
->vm_file
);
248 struct super_block
*sb
= inode
->i_sb
;
249 bool write
= flags
& FAULT_FLAG_WRITE
;
252 sb_start_pagefault(sb
);
253 file_update_time(vma
->vm_file
);
254 down_read(&EXT4_I(inode
)->i_mmap_sem
);
255 handle
= ext4_journal_start_sb(sb
, EXT4_HT_WRITE_PAGE
,
256 ext4_chunk_trans_blocks(inode
,
257 PMD_SIZE
/ PAGE_SIZE
));
259 down_read(&EXT4_I(inode
)->i_mmap_sem
);
262 result
= VM_FAULT_SIGBUS
;
264 result
= __dax_pmd_fault(vma
, addr
, pmd
, flags
,
265 ext4_get_block_dax
, ext4_end_io_unwritten
);
269 ext4_journal_stop(handle
);
270 up_read(&EXT4_I(inode
)->i_mmap_sem
);
271 sb_end_pagefault(sb
);
273 up_read(&EXT4_I(inode
)->i_mmap_sem
);
278 static int ext4_dax_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
281 struct inode
*inode
= file_inode(vma
->vm_file
);
283 sb_start_pagefault(inode
->i_sb
);
284 file_update_time(vma
->vm_file
);
285 down_read(&EXT4_I(inode
)->i_mmap_sem
);
286 err
= __dax_mkwrite(vma
, vmf
, ext4_get_block_dax
,
287 ext4_end_io_unwritten
);
288 up_read(&EXT4_I(inode
)->i_mmap_sem
);
289 sb_end_pagefault(inode
->i_sb
);
295 * Handle write fault for VM_MIXEDMAP mappings. Similarly to ext4_dax_mkwrite()
296 * handler we check for races agaist truncate. Note that since we cycle through
297 * i_mmap_sem, we are sure that also any hole punching that began before we
298 * were called is finished by now and so if it included part of the file we
299 * are working on, our pte will get unmapped and the check for pte_same() in
300 * wp_pfn_shared() fails. Thus fault gets retried and things work out as
303 static int ext4_dax_pfn_mkwrite(struct vm_area_struct
*vma
,
304 struct vm_fault
*vmf
)
306 struct inode
*inode
= file_inode(vma
->vm_file
);
307 struct super_block
*sb
= inode
->i_sb
;
308 int ret
= VM_FAULT_NOPAGE
;
311 sb_start_pagefault(sb
);
312 file_update_time(vma
->vm_file
);
313 down_read(&EXT4_I(inode
)->i_mmap_sem
);
314 size
= (i_size_read(inode
) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
315 if (vmf
->pgoff
>= size
)
316 ret
= VM_FAULT_SIGBUS
;
317 up_read(&EXT4_I(inode
)->i_mmap_sem
);
318 sb_end_pagefault(sb
);
323 static const struct vm_operations_struct ext4_dax_vm_ops
= {
324 .fault
= ext4_dax_fault
,
325 .pmd_fault
= ext4_dax_pmd_fault
,
326 .page_mkwrite
= ext4_dax_mkwrite
,
327 .pfn_mkwrite
= ext4_dax_pfn_mkwrite
,
330 #define ext4_dax_vm_ops ext4_file_vm_ops
333 static const struct vm_operations_struct ext4_file_vm_ops
= {
334 .fault
= ext4_filemap_fault
,
335 .map_pages
= filemap_map_pages
,
336 .page_mkwrite
= ext4_page_mkwrite
,
339 static int ext4_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
341 struct inode
*inode
= file
->f_mapping
->host
;
343 if (ext4_encrypted_inode(inode
)) {
344 int err
= ext4_get_encryption_info(inode
);
347 if (ext4_encryption_info(inode
) == NULL
)
351 if (IS_DAX(file_inode(file
))) {
352 vma
->vm_ops
= &ext4_dax_vm_ops
;
353 vma
->vm_flags
|= VM_MIXEDMAP
| VM_HUGEPAGE
;
355 vma
->vm_ops
= &ext4_file_vm_ops
;
360 static int ext4_file_open(struct inode
* inode
, struct file
* filp
)
362 struct super_block
*sb
= inode
->i_sb
;
363 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
364 struct vfsmount
*mnt
= filp
->f_path
.mnt
;
369 if (unlikely(!(sbi
->s_mount_flags
& EXT4_MF_MNTDIR_SAMPLED
) &&
370 !(sb
->s_flags
& MS_RDONLY
))) {
371 sbi
->s_mount_flags
|= EXT4_MF_MNTDIR_SAMPLED
;
373 * Sample where the filesystem has been mounted and
374 * store it in the superblock for sysadmin convenience
375 * when trying to sort through large numbers of block
376 * devices or filesystem images.
378 memset(buf
, 0, sizeof(buf
));
380 path
.dentry
= mnt
->mnt_root
;
381 cp
= d_path(&path
, buf
, sizeof(buf
));
386 handle
= ext4_journal_start_sb(sb
, EXT4_HT_MISC
, 1);
388 return PTR_ERR(handle
);
389 BUFFER_TRACE(sbi
->s_sbh
, "get_write_access");
390 err
= ext4_journal_get_write_access(handle
, sbi
->s_sbh
);
392 ext4_journal_stop(handle
);
395 strlcpy(sbi
->s_es
->s_last_mounted
, cp
,
396 sizeof(sbi
->s_es
->s_last_mounted
));
397 ext4_handle_dirty_super(handle
, sb
);
398 ext4_journal_stop(handle
);
401 if (ext4_encrypted_inode(inode
)) {
402 ret
= ext4_get_encryption_info(inode
);
405 if (ext4_encryption_info(inode
) == NULL
)
409 * Set up the jbd2_inode if we are opening the inode for
410 * writing and the journal is present
412 if (filp
->f_mode
& FMODE_WRITE
) {
413 ret
= ext4_inode_attach_jinode(inode
);
417 return dquot_file_open(inode
, filp
);
421 * Here we use ext4_map_blocks() to get a block mapping for a extent-based
422 * file rather than ext4_ext_walk_space() because we can introduce
423 * SEEK_DATA/SEEK_HOLE for block-mapped and extent-mapped file at the same
424 * function. When extent status tree has been fully implemented, it will
425 * track all extent status for a file and we can directly use it to
426 * retrieve the offset for SEEK_DATA/SEEK_HOLE.
430 * When we retrieve the offset for SEEK_DATA/SEEK_HOLE, we would need to
431 * lookup page cache to check whether or not there has some data between
432 * [startoff, endoff] because, if this range contains an unwritten extent,
433 * we determine this extent as a data or a hole according to whether the
434 * page cache has data or not.
436 static int ext4_find_unwritten_pgoff(struct inode
*inode
,
438 struct ext4_map_blocks
*map
,
442 unsigned int blkbits
;
450 blkbits
= inode
->i_sb
->s_blocksize_bits
;
453 endoff
= (loff_t
)(map
->m_lblk
+ map
->m_len
) << blkbits
;
455 index
= startoff
>> PAGE_CACHE_SHIFT
;
456 end
= endoff
>> PAGE_CACHE_SHIFT
;
458 pagevec_init(&pvec
, 0);
461 unsigned long nr_pages
;
463 num
= min_t(pgoff_t
, end
- index
, PAGEVEC_SIZE
- 1) + 1;
464 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, index
,
469 for (i
= 0; i
< nr_pages
; i
++) {
470 struct page
*page
= pvec
.pages
[i
];
471 struct buffer_head
*bh
, *head
;
474 * If current offset is smaller than the page offset,
475 * there is a hole at this offset.
477 if (whence
== SEEK_HOLE
&& lastoff
< endoff
&&
478 lastoff
< page_offset(pvec
.pages
[i
])) {
484 if (page
->index
> end
)
489 if (unlikely(page
->mapping
!= inode
->i_mapping
)) {
494 if (!page_has_buffers(page
)) {
499 if (page_has_buffers(page
)) {
500 lastoff
= page_offset(page
);
501 bh
= head
= page_buffers(page
);
503 if (lastoff
+ bh
->b_size
<= startoff
)
505 if (buffer_uptodate(bh
) ||
506 buffer_unwritten(bh
)) {
507 if (whence
== SEEK_DATA
)
510 if (whence
== SEEK_HOLE
)
514 *offset
= max_t(loff_t
,
520 lastoff
+= bh
->b_size
;
521 bh
= bh
->b_this_page
;
522 } while (bh
!= head
);
525 lastoff
= page_offset(page
) + PAGE_SIZE
;
529 /* The no. of pages is less than our desired, we are done. */
533 index
= pvec
.pages
[i
- 1]->index
+ 1;
534 pagevec_release(&pvec
);
535 } while (index
<= end
);
537 if (whence
== SEEK_HOLE
&& lastoff
< endoff
) {
542 pagevec_release(&pvec
);
547 * ext4_seek_data() retrieves the offset for SEEK_DATA.
549 static loff_t
ext4_seek_data(struct file
*file
, loff_t offset
, loff_t maxsize
)
551 struct inode
*inode
= file
->f_mapping
->host
;
552 struct ext4_map_blocks map
;
553 struct extent_status es
;
554 ext4_lblk_t start
, last
, end
;
555 loff_t dataoff
, isize
;
559 mutex_lock(&inode
->i_mutex
);
561 isize
= i_size_read(inode
);
562 if (offset
< 0 || offset
>= isize
) {
563 mutex_unlock(&inode
->i_mutex
);
567 blkbits
= inode
->i_sb
->s_blocksize_bits
;
568 start
= offset
>> blkbits
;
570 end
= isize
>> blkbits
;
575 map
.m_len
= end
- last
+ 1;
576 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
577 if (ret
> 0 && !(map
.m_flags
& EXT4_MAP_UNWRITTEN
)) {
579 dataoff
= (loff_t
)last
<< blkbits
;
584 * If there is a delay extent at this offset,
585 * it will be as a data.
587 ext4_es_find_delayed_extent_range(inode
, last
, last
, &es
);
588 if (es
.es_len
!= 0 && in_range(last
, es
.es_lblk
, es
.es_len
)) {
590 dataoff
= (loff_t
)last
<< blkbits
;
595 * If there is a unwritten extent at this offset,
596 * it will be as a data or a hole according to page
597 * cache that has data or not.
599 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
601 unwritten
= ext4_find_unwritten_pgoff(inode
, SEEK_DATA
,
608 dataoff
= (loff_t
)last
<< blkbits
;
609 } while (last
<= end
);
611 mutex_unlock(&inode
->i_mutex
);
616 return vfs_setpos(file
, dataoff
, maxsize
);
620 * ext4_seek_hole() retrieves the offset for SEEK_HOLE.
622 static loff_t
ext4_seek_hole(struct file
*file
, loff_t offset
, loff_t maxsize
)
624 struct inode
*inode
= file
->f_mapping
->host
;
625 struct ext4_map_blocks map
;
626 struct extent_status es
;
627 ext4_lblk_t start
, last
, end
;
628 loff_t holeoff
, isize
;
632 mutex_lock(&inode
->i_mutex
);
634 isize
= i_size_read(inode
);
635 if (offset
< 0 || offset
>= isize
) {
636 mutex_unlock(&inode
->i_mutex
);
640 blkbits
= inode
->i_sb
->s_blocksize_bits
;
641 start
= offset
>> blkbits
;
643 end
= isize
>> blkbits
;
648 map
.m_len
= end
- last
+ 1;
649 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
650 if (ret
> 0 && !(map
.m_flags
& EXT4_MAP_UNWRITTEN
)) {
652 holeoff
= (loff_t
)last
<< blkbits
;
657 * If there is a delay extent at this offset,
658 * we will skip this extent.
660 ext4_es_find_delayed_extent_range(inode
, last
, last
, &es
);
661 if (es
.es_len
!= 0 && in_range(last
, es
.es_lblk
, es
.es_len
)) {
662 last
= es
.es_lblk
+ es
.es_len
;
663 holeoff
= (loff_t
)last
<< blkbits
;
668 * If there is a unwritten extent at this offset,
669 * it will be as a data or a hole according to page
670 * cache that has data or not.
672 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
674 unwritten
= ext4_find_unwritten_pgoff(inode
, SEEK_HOLE
,
678 holeoff
= (loff_t
)last
<< blkbits
;
685 } while (last
<= end
);
687 mutex_unlock(&inode
->i_mutex
);
692 return vfs_setpos(file
, holeoff
, maxsize
);
696 * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
697 * by calling generic_file_llseek_size() with the appropriate maxbytes
700 loff_t
ext4_llseek(struct file
*file
, loff_t offset
, int whence
)
702 struct inode
*inode
= file
->f_mapping
->host
;
705 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
706 maxbytes
= EXT4_SB(inode
->i_sb
)->s_bitmap_maxbytes
;
708 maxbytes
= inode
->i_sb
->s_maxbytes
;
714 return generic_file_llseek_size(file
, offset
, whence
,
715 maxbytes
, i_size_read(inode
));
717 return ext4_seek_data(file
, offset
, maxbytes
);
719 return ext4_seek_hole(file
, offset
, maxbytes
);
725 const struct file_operations ext4_file_operations
= {
726 .llseek
= ext4_llseek
,
727 .read_iter
= generic_file_read_iter
,
728 .write_iter
= ext4_file_write_iter
,
729 .unlocked_ioctl
= ext4_ioctl
,
731 .compat_ioctl
= ext4_compat_ioctl
,
733 .mmap
= ext4_file_mmap
,
734 .open
= ext4_file_open
,
735 .release
= ext4_release_file
,
736 .fsync
= ext4_sync_file
,
737 .splice_read
= generic_file_splice_read
,
738 .splice_write
= iter_file_splice_write
,
739 .fallocate
= ext4_fallocate
,
742 const struct inode_operations ext4_file_inode_operations
= {
743 .setattr
= ext4_setattr
,
744 .getattr
= ext4_getattr
,
745 .setxattr
= generic_setxattr
,
746 .getxattr
= generic_getxattr
,
747 .listxattr
= ext4_listxattr
,
748 .removexattr
= generic_removexattr
,
749 .get_acl
= ext4_get_acl
,
750 .set_acl
= ext4_set_acl
,
751 .fiemap
= ext4_fiemap
,