1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
7 #include <linux/slab.h>
8 #include <linux/spinlock.h>
9 #include <linux/compat.h>
10 #include <linux/completion.h>
11 #include <linux/buffer_head.h>
12 #include <linux/pagemap.h>
13 #include <linux/uio.h>
14 #include <linux/blkdev.h>
16 #include <linux/mount.h>
18 #include <linux/gfs2_ondisk.h>
19 #include <linux/falloc.h>
20 #include <linux/swap.h>
21 #include <linux/crc32.h>
22 #include <linux/writeback.h>
23 #include <linux/uaccess.h>
24 #include <linux/dlm.h>
25 #include <linux/dlm_plock.h>
26 #include <linux/delay.h>
27 #include <linux/backing-dev.h>
28 #include <linux/fileattr.h>
46 * gfs2_llseek - seek to a location in a file
49 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
51 * SEEK_END requires the glock for the file because it references the
54 * Returns: The new offset, or errno
57 static loff_t
gfs2_llseek(struct file
*file
, loff_t offset
, int whence
)
59 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
60 struct gfs2_holder i_gh
;
65 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
68 error
= generic_file_llseek(file
, offset
, whence
);
69 gfs2_glock_dq_uninit(&i_gh
);
74 error
= gfs2_seek_data(file
, offset
);
78 error
= gfs2_seek_hole(file
, offset
);
84 * These don't reference inode->i_size and don't depend on the
85 * block mapping, so we don't need the glock.
87 error
= generic_file_llseek(file
, offset
, whence
);
97 * gfs2_readdir - Iterator for a directory
98 * @file: The directory to read from
99 * @ctx: What to feed directory entries to
104 static int gfs2_readdir(struct file
*file
, struct dir_context
*ctx
)
106 struct inode
*dir
= file
->f_mapping
->host
;
107 struct gfs2_inode
*dip
= GFS2_I(dir
);
108 struct gfs2_holder d_gh
;
111 error
= gfs2_glock_nq_init(dip
->i_gl
, LM_ST_SHARED
, 0, &d_gh
);
115 error
= gfs2_dir_read(dir
, ctx
, &file
->f_ra
);
117 gfs2_glock_dq_uninit(&d_gh
);
123 * struct fsflag_gfs2flag
125 * The FS_JOURNAL_DATA_FL flag maps to GFS2_DIF_INHERIT_JDATA for directories,
126 * and to GFS2_DIF_JDATA for non-directories.
131 } fsflag_gfs2flag
[] = {
132 {FS_SYNC_FL
, GFS2_DIF_SYNC
},
133 {FS_IMMUTABLE_FL
, GFS2_DIF_IMMUTABLE
},
134 {FS_APPEND_FL
, GFS2_DIF_APPENDONLY
},
135 {FS_NOATIME_FL
, GFS2_DIF_NOATIME
},
136 {FS_INDEX_FL
, GFS2_DIF_EXHASH
},
137 {FS_TOPDIR_FL
, GFS2_DIF_TOPDIR
},
138 {FS_JOURNAL_DATA_FL
, GFS2_DIF_JDATA
| GFS2_DIF_INHERIT_JDATA
},
141 static inline u32
gfs2_gfsflags_to_fsflags(struct inode
*inode
, u32 gfsflags
)
146 if (S_ISDIR(inode
->i_mode
))
147 gfsflags
&= ~GFS2_DIF_JDATA
;
149 gfsflags
&= ~GFS2_DIF_INHERIT_JDATA
;
151 for (i
= 0; i
< ARRAY_SIZE(fsflag_gfs2flag
); i
++)
152 if (gfsflags
& fsflag_gfs2flag
[i
].gfsflag
)
153 fsflags
|= fsflag_gfs2flag
[i
].fsflag
;
157 int gfs2_fileattr_get(struct dentry
*dentry
, struct fileattr
*fa
)
159 struct inode
*inode
= d_inode(dentry
);
160 struct gfs2_inode
*ip
= GFS2_I(inode
);
161 struct gfs2_holder gh
;
165 if (d_is_special(dentry
))
168 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, 0, &gh
);
169 error
= gfs2_glock_nq(&gh
);
173 fsflags
= gfs2_gfsflags_to_fsflags(inode
, ip
->i_diskflags
);
175 fileattr_fill_flags(fa
, fsflags
);
179 gfs2_holder_uninit(&gh
);
183 void gfs2_set_inode_flags(struct inode
*inode
)
185 struct gfs2_inode
*ip
= GFS2_I(inode
);
186 unsigned int flags
= inode
->i_flags
;
188 flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_NOSEC
);
189 if ((ip
->i_eattr
== 0) && !is_sxid(inode
->i_mode
))
191 if (ip
->i_diskflags
& GFS2_DIF_IMMUTABLE
)
192 flags
|= S_IMMUTABLE
;
193 if (ip
->i_diskflags
& GFS2_DIF_APPENDONLY
)
195 if (ip
->i_diskflags
& GFS2_DIF_NOATIME
)
197 if (ip
->i_diskflags
& GFS2_DIF_SYNC
)
199 inode
->i_flags
= flags
;
202 /* Flags that can be set by user space */
203 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
204 GFS2_DIF_IMMUTABLE| \
205 GFS2_DIF_APPENDONLY| \
209 GFS2_DIF_INHERIT_JDATA)
212 * do_gfs2_set_flags - set flags on an inode
214 * @reqflags: The flags to set
215 * @mask: Indicates which flags are valid
218 static int do_gfs2_set_flags(struct inode
*inode
, u32 reqflags
, u32 mask
)
220 struct gfs2_inode
*ip
= GFS2_I(inode
);
221 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
222 struct buffer_head
*bh
;
223 struct gfs2_holder gh
;
225 u32 new_flags
, flags
;
227 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
232 flags
= ip
->i_diskflags
;
233 new_flags
= (flags
& ~mask
) | (reqflags
& mask
);
234 if ((new_flags
^ flags
) == 0)
237 if (!IS_IMMUTABLE(inode
)) {
238 error
= gfs2_permission(&init_user_ns
, inode
, MAY_WRITE
);
242 if ((flags
^ new_flags
) & GFS2_DIF_JDATA
) {
243 if (new_flags
& GFS2_DIF_JDATA
)
244 gfs2_log_flush(sdp
, ip
->i_gl
,
245 GFS2_LOG_HEAD_FLUSH_NORMAL
|
247 error
= filemap_fdatawrite(inode
->i_mapping
);
250 error
= filemap_fdatawait(inode
->i_mapping
);
253 if (new_flags
& GFS2_DIF_JDATA
)
254 gfs2_ordered_del_inode(ip
);
256 error
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
259 error
= gfs2_meta_inode_buffer(ip
, &bh
);
262 inode
->i_ctime
= current_time(inode
);
263 gfs2_trans_add_meta(ip
->i_gl
, bh
);
264 ip
->i_diskflags
= new_flags
;
265 gfs2_dinode_out(ip
, bh
->b_data
);
267 gfs2_set_inode_flags(inode
);
268 gfs2_set_aops(inode
);
272 gfs2_glock_dq_uninit(&gh
);
276 int gfs2_fileattr_set(struct mnt_idmap
*idmap
,
277 struct dentry
*dentry
, struct fileattr
*fa
)
279 struct inode
*inode
= d_inode(dentry
);
280 u32 fsflags
= fa
->flags
, gfsflags
= 0;
284 if (d_is_special(dentry
))
287 if (fileattr_has_fsx(fa
))
290 for (i
= 0; i
< ARRAY_SIZE(fsflag_gfs2flag
); i
++) {
291 if (fsflags
& fsflag_gfs2flag
[i
].fsflag
) {
292 fsflags
&= ~fsflag_gfs2flag
[i
].fsflag
;
293 gfsflags
|= fsflag_gfs2flag
[i
].gfsflag
;
296 if (fsflags
|| gfsflags
& ~GFS2_FLAGS_USER_SET
)
299 mask
= GFS2_FLAGS_USER_SET
;
300 if (S_ISDIR(inode
->i_mode
)) {
301 mask
&= ~GFS2_DIF_JDATA
;
303 /* The GFS2_DIF_TOPDIR flag is only valid for directories. */
304 if (gfsflags
& GFS2_DIF_TOPDIR
)
306 mask
&= ~(GFS2_DIF_TOPDIR
| GFS2_DIF_INHERIT_JDATA
);
309 return do_gfs2_set_flags(inode
, gfsflags
, mask
);
312 static int gfs2_getlabel(struct file
*filp
, char __user
*label
)
314 struct inode
*inode
= file_inode(filp
);
315 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
317 if (copy_to_user(label
, sdp
->sd_sb
.sb_locktable
, GFS2_LOCKNAME_LEN
))
323 static long gfs2_ioctl(struct file
*filp
, unsigned int cmd
, unsigned long arg
)
327 return gfs2_fitrim(filp
, (void __user
*)arg
);
328 case FS_IOC_GETFSLABEL
:
329 return gfs2_getlabel(filp
, (char __user
*)arg
);
336 static long gfs2_compat_ioctl(struct file
*filp
, unsigned int cmd
, unsigned long arg
)
339 /* Keep this list in sync with gfs2_ioctl */
341 case FS_IOC_GETFSLABEL
:
347 return gfs2_ioctl(filp
, cmd
, (unsigned long)compat_ptr(arg
));
350 #define gfs2_compat_ioctl NULL
354 * gfs2_size_hint - Give a hint to the size of a write request
355 * @filep: The struct file
356 * @offset: The file offset of the write
357 * @size: The length of the write
359 * When we are about to do a write, this function records the total
360 * write size in order to provide a suitable hint to the lower layers
361 * about how many blocks will be required.
365 static void gfs2_size_hint(struct file
*filep
, loff_t offset
, size_t size
)
367 struct inode
*inode
= file_inode(filep
);
368 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
369 struct gfs2_inode
*ip
= GFS2_I(inode
);
370 size_t blks
= (size
+ sdp
->sd_sb
.sb_bsize
- 1) >> sdp
->sd_sb
.sb_bsize_shift
;
371 int hint
= min_t(size_t, INT_MAX
, blks
);
373 if (hint
> atomic_read(&ip
->i_sizehint
))
374 atomic_set(&ip
->i_sizehint
, hint
);
378 * gfs2_allocate_page_backing - Allocate blocks for a write fault
379 * @page: The (locked) page to allocate backing for
380 * @length: Size of the allocation
382 * We try to allocate all the blocks required for the page in one go. This
383 * might fail for various reasons, so we keep trying until all the blocks to
384 * back this page are allocated. If some of the blocks are already allocated,
387 static int gfs2_allocate_page_backing(struct page
*page
, unsigned int length
)
389 u64 pos
= page_offset(page
);
392 struct iomap iomap
= { };
394 if (gfs2_iomap_alloc(page
->mapping
->host
, pos
, length
, &iomap
))
397 if (length
< iomap
.length
)
398 iomap
.length
= length
;
399 length
-= iomap
.length
;
401 } while (length
> 0);
407 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
408 * @vmf: The virtual memory fault containing the page to become writable
410 * When the page becomes writable, we need to ensure that we have
411 * blocks allocated on disk to back that page.
414 static vm_fault_t
gfs2_page_mkwrite(struct vm_fault
*vmf
)
416 struct page
*page
= vmf
->page
;
417 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
418 struct gfs2_inode
*ip
= GFS2_I(inode
);
419 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
420 struct gfs2_alloc_parms ap
= { .aflags
= 0, };
421 u64 offset
= page_offset(page
);
422 unsigned int data_blocks
, ind_blocks
, rblocks
;
423 vm_fault_t ret
= VM_FAULT_LOCKED
;
424 struct gfs2_holder gh
;
429 sb_start_pagefault(inode
->i_sb
);
431 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
432 err
= gfs2_glock_nq(&gh
);
434 ret
= block_page_mkwrite_return(err
);
438 /* Check page index against inode size */
439 size
= i_size_read(inode
);
440 if (offset
>= size
) {
441 ret
= VM_FAULT_SIGBUS
;
445 /* Update file times before taking page lock */
446 file_update_time(vmf
->vma
->vm_file
);
448 /* page is wholly or partially inside EOF */
449 if (size
- offset
< PAGE_SIZE
)
450 length
= size
- offset
;
454 gfs2_size_hint(vmf
->vma
->vm_file
, offset
, length
);
456 set_bit(GLF_DIRTY
, &ip
->i_gl
->gl_flags
);
457 set_bit(GIF_SW_PAGED
, &ip
->i_flags
);
460 * iomap_writepage / iomap_writepages currently don't support inline
461 * files, so always unstuff here.
464 if (!gfs2_is_stuffed(ip
) &&
465 !gfs2_write_alloc_required(ip
, offset
, length
)) {
467 if (!PageUptodate(page
) || page
->mapping
!= inode
->i_mapping
) {
468 ret
= VM_FAULT_NOPAGE
;
474 err
= gfs2_rindex_update(sdp
);
476 ret
= block_page_mkwrite_return(err
);
480 gfs2_write_calc_reserv(ip
, length
, &data_blocks
, &ind_blocks
);
481 ap
.target
= data_blocks
+ ind_blocks
;
482 err
= gfs2_quota_lock_check(ip
, &ap
);
484 ret
= block_page_mkwrite_return(err
);
487 err
= gfs2_inplace_reserve(ip
, &ap
);
489 ret
= block_page_mkwrite_return(err
);
490 goto out_quota_unlock
;
493 rblocks
= RES_DINODE
+ ind_blocks
;
494 if (gfs2_is_jdata(ip
))
495 rblocks
+= data_blocks
? data_blocks
: 1;
496 if (ind_blocks
|| data_blocks
) {
497 rblocks
+= RES_STATFS
+ RES_QUOTA
;
498 rblocks
+= gfs2_rg_blocks(ip
, data_blocks
+ ind_blocks
);
500 err
= gfs2_trans_begin(sdp
, rblocks
, 0);
502 ret
= block_page_mkwrite_return(err
);
506 /* Unstuff, if required, and allocate backing blocks for page */
507 if (gfs2_is_stuffed(ip
)) {
508 err
= gfs2_unstuff_dinode(ip
);
510 ret
= block_page_mkwrite_return(err
);
516 /* If truncated, we must retry the operation, we may have raced
517 * with the glock demotion code.
519 if (!PageUptodate(page
) || page
->mapping
!= inode
->i_mapping
) {
520 ret
= VM_FAULT_NOPAGE
;
521 goto out_page_locked
;
524 err
= gfs2_allocate_page_backing(page
, length
);
526 ret
= block_page_mkwrite_return(err
);
529 if (ret
!= VM_FAULT_LOCKED
)
534 gfs2_inplace_release(ip
);
536 gfs2_quota_unlock(ip
);
540 gfs2_holder_uninit(&gh
);
541 if (ret
== VM_FAULT_LOCKED
) {
542 set_page_dirty(page
);
543 wait_for_stable_page(page
);
545 sb_end_pagefault(inode
->i_sb
);
549 static vm_fault_t
gfs2_fault(struct vm_fault
*vmf
)
551 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
552 struct gfs2_inode
*ip
= GFS2_I(inode
);
553 struct gfs2_holder gh
;
557 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, 0, &gh
);
558 err
= gfs2_glock_nq(&gh
);
560 ret
= block_page_mkwrite_return(err
);
563 ret
= filemap_fault(vmf
);
566 gfs2_holder_uninit(&gh
);
570 static const struct vm_operations_struct gfs2_vm_ops
= {
572 .map_pages
= filemap_map_pages
,
573 .page_mkwrite
= gfs2_page_mkwrite
,
578 * @file: The file to map
579 * @vma: The VMA which described the mapping
581 * There is no need to get a lock here unless we should be updating
582 * atime. We ignore any locking errors since the only consequence is
583 * a missed atime update (which will just be deferred until later).
588 static int gfs2_mmap(struct file
*file
, struct vm_area_struct
*vma
)
590 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
592 if (!(file
->f_flags
& O_NOATIME
) &&
593 !IS_NOATIME(&ip
->i_inode
)) {
594 struct gfs2_holder i_gh
;
597 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
601 /* grab lock to update inode */
602 gfs2_glock_dq_uninit(&i_gh
);
605 vma
->vm_ops
= &gfs2_vm_ops
;
611 * gfs2_open_common - This is common to open and atomic_open
612 * @inode: The inode being opened
613 * @file: The file being opened
615 * This maybe called under a glock or not depending upon how it has
616 * been called. We must always be called under a glock for regular
617 * files, however. For other file types, it does not matter whether
618 * we hold the glock or not.
620 * Returns: Error code or 0 for success
623 int gfs2_open_common(struct inode
*inode
, struct file
*file
)
625 struct gfs2_file
*fp
;
628 if (S_ISREG(inode
->i_mode
)) {
629 ret
= generic_file_open(inode
, file
);
634 fp
= kzalloc(sizeof(struct gfs2_file
), GFP_NOFS
);
638 mutex_init(&fp
->f_fl_mutex
);
640 gfs2_assert_warn(GFS2_SB(inode
), !file
->private_data
);
641 file
->private_data
= fp
;
642 if (file
->f_mode
& FMODE_WRITE
) {
643 ret
= gfs2_qa_get(GFS2_I(inode
));
650 kfree(file
->private_data
);
651 file
->private_data
= NULL
;
656 * gfs2_open - open a file
657 * @inode: the inode to open
658 * @file: the struct file for this opening
660 * After atomic_open, this function is only used for opening files
661 * which are already cached. We must still get the glock for regular
662 * files to ensure that we have the file size uptodate for the large
663 * file check which is in the common code. That is only an issue for
664 * regular files though.
669 static int gfs2_open(struct inode
*inode
, struct file
*file
)
671 struct gfs2_inode
*ip
= GFS2_I(inode
);
672 struct gfs2_holder i_gh
;
674 bool need_unlock
= false;
676 if (S_ISREG(ip
->i_inode
.i_mode
)) {
677 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
684 error
= gfs2_open_common(inode
, file
);
687 gfs2_glock_dq_uninit(&i_gh
);
693 * gfs2_release - called to close a struct file
694 * @inode: the inode the struct file belongs to
695 * @file: the struct file being closed
700 static int gfs2_release(struct inode
*inode
, struct file
*file
)
702 struct gfs2_inode
*ip
= GFS2_I(inode
);
704 kfree(file
->private_data
);
705 file
->private_data
= NULL
;
707 if (file
->f_mode
& FMODE_WRITE
) {
708 if (gfs2_rs_active(&ip
->i_res
))
716 * gfs2_fsync - sync the dirty data for a file (across the cluster)
717 * @file: the file that points to the dentry
718 * @start: the start position in the file to sync
719 * @end: the end position in the file to sync
720 * @datasync: set if we can ignore timestamp changes
722 * We split the data flushing here so that we don't wait for the data
723 * until after we've also sent the metadata to disk. Note that for
724 * data=ordered, we will write & wait for the data at the log flush
725 * stage anyway, so this is unlikely to make much of a difference
726 * except in the data=writeback case.
728 * If the fdatawrite fails due to any reason except -EIO, we will
729 * continue the remainder of the fsync, although we'll still report
730 * the error at the end. This is to match filemap_write_and_wait_range()
736 static int gfs2_fsync(struct file
*file
, loff_t start
, loff_t end
,
739 struct address_space
*mapping
= file
->f_mapping
;
740 struct inode
*inode
= mapping
->host
;
741 int sync_state
= inode
->i_state
& I_DIRTY
;
742 struct gfs2_inode
*ip
= GFS2_I(inode
);
743 int ret
= 0, ret1
= 0;
745 if (mapping
->nrpages
) {
746 ret1
= filemap_fdatawrite_range(mapping
, start
, end
);
751 if (!gfs2_is_jdata(ip
))
752 sync_state
&= ~I_DIRTY_PAGES
;
754 sync_state
&= ~I_DIRTY_SYNC
;
757 ret
= sync_inode_metadata(inode
, 1);
760 if (gfs2_is_jdata(ip
))
761 ret
= file_write_and_wait(file
);
764 gfs2_ail_flush(ip
->i_gl
, 1);
767 if (mapping
->nrpages
)
768 ret
= file_fdatawait_range(file
, start
, end
);
770 return ret
? ret
: ret1
;
773 static inline bool should_fault_in_pages(struct iov_iter
*i
,
778 size_t count
= iov_iter_count(i
);
783 if (!user_backed_iter(i
))
787 offs
= offset_in_page(iocb
->ki_pos
);
788 if (*prev_count
!= count
|| !*window_size
) {
791 nr_dirtied
= max(current
->nr_dirtied_pause
-
792 current
->nr_dirtied
, 8);
793 size
= min_t(size_t, SZ_1M
, nr_dirtied
<< PAGE_SHIFT
);
797 *window_size
= size
- offs
;
801 static ssize_t
gfs2_file_direct_read(struct kiocb
*iocb
, struct iov_iter
*to
,
802 struct gfs2_holder
*gh
)
804 struct file
*file
= iocb
->ki_filp
;
805 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
806 size_t prev_count
= 0, window_size
= 0;
811 * In this function, we disable page faults when we're holding the
812 * inode glock while doing I/O. If a page fault occurs, we indicate
813 * that the inode glock may be dropped, fault in the pages manually,
816 * Unlike generic_file_read_iter, for reads, iomap_dio_rw can trigger
817 * physical as well as manual page faults, and we need to disable both
820 * For direct I/O, gfs2 takes the inode glock in deferred mode. This
821 * locking mode is compatible with other deferred holders, so multiple
822 * processes and nodes can do direct I/O to a file at the same time.
823 * There's no guarantee that reads or writes will be atomic. Any
824 * coordination among readers and writers needs to happen externally.
827 if (!iov_iter_count(to
))
828 return 0; /* skip atime */
830 gfs2_holder_init(ip
->i_gl
, LM_ST_DEFERRED
, 0, gh
);
832 ret
= gfs2_glock_nq(gh
);
837 ret
= iomap_dio_rw(iocb
, to
, &gfs2_iomap_ops
, NULL
,
838 IOMAP_DIO_PARTIAL
, NULL
, read
);
841 if (ret
<= 0 && ret
!= -EFAULT
)
843 /* No increment (+=) because iomap_dio_rw returns a cumulative value. */
847 if (should_fault_in_pages(to
, iocb
, &prev_count
, &window_size
)) {
849 window_size
-= fault_in_iov_iter_writeable(to
, window_size
);
854 if (gfs2_holder_queued(gh
))
857 gfs2_holder_uninit(gh
);
858 /* User space doesn't expect partial success. */
864 static ssize_t
gfs2_file_direct_write(struct kiocb
*iocb
, struct iov_iter
*from
,
865 struct gfs2_holder
*gh
)
867 struct file
*file
= iocb
->ki_filp
;
868 struct inode
*inode
= file
->f_mapping
->host
;
869 struct gfs2_inode
*ip
= GFS2_I(inode
);
870 size_t prev_count
= 0, window_size
= 0;
875 * In this function, we disable page faults when we're holding the
876 * inode glock while doing I/O. If a page fault occurs, we indicate
877 * that the inode glock may be dropped, fault in the pages manually,
880 * For writes, iomap_dio_rw only triggers manual page faults, so we
881 * don't need to disable physical ones.
885 * Deferred lock, even if its a write, since we do no allocation on
886 * this path. All we need to change is the atime, and this lock mode
887 * ensures that other nodes have flushed their buffered read caches
888 * (i.e. their page cache entries for this inode). We do not,
889 * unfortunately, have the option of only flushing a range like the
892 gfs2_holder_init(ip
->i_gl
, LM_ST_DEFERRED
, 0, gh
);
894 ret
= gfs2_glock_nq(gh
);
897 /* Silently fall back to buffered I/O when writing beyond EOF */
898 if (iocb
->ki_pos
+ iov_iter_count(from
) > i_size_read(&ip
->i_inode
))
901 from
->nofault
= true;
902 ret
= iomap_dio_rw(iocb
, from
, &gfs2_iomap_ops
, NULL
,
903 IOMAP_DIO_PARTIAL
, NULL
, written
);
904 from
->nofault
= false;
911 /* No increment (+=) because iomap_dio_rw returns a cumulative value. */
915 if (should_fault_in_pages(from
, iocb
, &prev_count
, &window_size
)) {
917 window_size
-= fault_in_iov_iter_readable(from
, window_size
);
922 if (gfs2_holder_queued(gh
))
925 gfs2_holder_uninit(gh
);
926 /* User space doesn't expect partial success. */
932 static ssize_t
gfs2_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*to
)
934 struct gfs2_inode
*ip
;
935 struct gfs2_holder gh
;
936 size_t prev_count
= 0, window_size
= 0;
941 * In this function, we disable page faults when we're holding the
942 * inode glock while doing I/O. If a page fault occurs, we indicate
943 * that the inode glock may be dropped, fault in the pages manually,
947 if (iocb
->ki_flags
& IOCB_DIRECT
)
948 return gfs2_file_direct_read(iocb
, to
, &gh
);
951 iocb
->ki_flags
|= IOCB_NOIO
;
952 ret
= generic_file_read_iter(iocb
, to
);
953 iocb
->ki_flags
&= ~IOCB_NOIO
;
956 if (!iov_iter_count(to
))
959 } else if (ret
!= -EFAULT
) {
962 if (iocb
->ki_flags
& IOCB_NOWAIT
)
965 ip
= GFS2_I(iocb
->ki_filp
->f_mapping
->host
);
966 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, 0, &gh
);
968 ret
= gfs2_glock_nq(&gh
);
972 ret
= generic_file_read_iter(iocb
, to
);
974 if (ret
<= 0 && ret
!= -EFAULT
)
979 if (should_fault_in_pages(to
, iocb
, &prev_count
, &window_size
)) {
981 window_size
-= fault_in_iov_iter_writeable(to
, window_size
);
986 if (gfs2_holder_queued(&gh
))
989 gfs2_holder_uninit(&gh
);
990 return read
? read
: ret
;
993 static ssize_t
gfs2_file_buffered_write(struct kiocb
*iocb
,
994 struct iov_iter
*from
,
995 struct gfs2_holder
*gh
)
997 struct file
*file
= iocb
->ki_filp
;
998 struct inode
*inode
= file_inode(file
);
999 struct gfs2_inode
*ip
= GFS2_I(inode
);
1000 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1001 struct gfs2_holder
*statfs_gh
= NULL
;
1002 size_t prev_count
= 0, window_size
= 0;
1003 size_t orig_count
= iov_iter_count(from
);
1008 * In this function, we disable page faults when we're holding the
1009 * inode glock while doing I/O. If a page fault occurs, we indicate
1010 * that the inode glock may be dropped, fault in the pages manually,
1014 if (inode
== sdp
->sd_rindex
) {
1015 statfs_gh
= kmalloc(sizeof(*statfs_gh
), GFP_NOFS
);
1020 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, gh
);
1022 if (should_fault_in_pages(from
, iocb
, &prev_count
, &window_size
)) {
1023 window_size
-= fault_in_iov_iter_readable(from
, window_size
);
1028 from
->count
= min(from
->count
, window_size
);
1030 ret
= gfs2_glock_nq(gh
);
1034 if (inode
== sdp
->sd_rindex
) {
1035 struct gfs2_inode
*m_ip
= GFS2_I(sdp
->sd_statfs_inode
);
1037 ret
= gfs2_glock_nq_init(m_ip
->i_gl
, LM_ST_EXCLUSIVE
,
1038 GL_NOCACHE
, statfs_gh
);
1043 current
->backing_dev_info
= inode_to_bdi(inode
);
1044 pagefault_disable();
1045 ret
= iomap_file_buffered_write(iocb
, from
, &gfs2_iomap_ops
);
1047 current
->backing_dev_info
= NULL
;
1049 iocb
->ki_pos
+= ret
;
1053 if (inode
== sdp
->sd_rindex
)
1054 gfs2_glock_dq_uninit(statfs_gh
);
1056 if (ret
<= 0 && ret
!= -EFAULT
)
1059 from
->count
= orig_count
- written
;
1060 if (should_fault_in_pages(from
, iocb
, &prev_count
, &window_size
)) {
1065 if (gfs2_holder_queued(gh
))
1068 gfs2_holder_uninit(gh
);
1070 from
->count
= orig_count
- written
;
1071 return written
? written
: ret
;
1075 * gfs2_file_write_iter - Perform a write to a file
1076 * @iocb: The io context
1077 * @from: The data to write
1079 * We have to do a lock/unlock here to refresh the inode size for
1080 * O_APPEND writes, otherwise we can land up writing at the wrong
1081 * offset. There is still a race, but provided the app is using its
1082 * own file locking, this will make O_APPEND work as expected.
1086 static ssize_t
gfs2_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*from
)
1088 struct file
*file
= iocb
->ki_filp
;
1089 struct inode
*inode
= file_inode(file
);
1090 struct gfs2_inode
*ip
= GFS2_I(inode
);
1091 struct gfs2_holder gh
;
1094 gfs2_size_hint(file
, iocb
->ki_pos
, iov_iter_count(from
));
1096 if (iocb
->ki_flags
& IOCB_APPEND
) {
1097 ret
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, 0, &gh
);
1100 gfs2_glock_dq_uninit(&gh
);
1104 ret
= generic_write_checks(iocb
, from
);
1108 ret
= file_remove_privs(file
);
1112 ret
= file_update_time(file
);
1116 if (iocb
->ki_flags
& IOCB_DIRECT
) {
1117 struct address_space
*mapping
= file
->f_mapping
;
1118 ssize_t buffered
, ret2
;
1120 ret
= gfs2_file_direct_write(iocb
, from
, &gh
);
1121 if (ret
< 0 || !iov_iter_count(from
))
1124 iocb
->ki_flags
|= IOCB_DSYNC
;
1125 buffered
= gfs2_file_buffered_write(iocb
, from
, &gh
);
1126 if (unlikely(buffered
<= 0)) {
1133 * We need to ensure that the page cache pages are written to
1134 * disk and invalidated to preserve the expected O_DIRECT
1135 * semantics. If the writeback or invalidate fails, only report
1136 * the direct I/O range as we don't know if the buffered pages
1139 ret2
= generic_write_sync(iocb
, buffered
);
1140 invalidate_mapping_pages(mapping
,
1141 (iocb
->ki_pos
- buffered
) >> PAGE_SHIFT
,
1142 (iocb
->ki_pos
- 1) >> PAGE_SHIFT
);
1143 if (!ret
|| ret2
> 0)
1146 ret
= gfs2_file_buffered_write(iocb
, from
, &gh
);
1147 if (likely(ret
> 0))
1148 ret
= generic_write_sync(iocb
, ret
);
1152 inode_unlock(inode
);
1156 static int fallocate_chunk(struct inode
*inode
, loff_t offset
, loff_t len
,
1159 struct super_block
*sb
= inode
->i_sb
;
1160 struct gfs2_inode
*ip
= GFS2_I(inode
);
1161 loff_t end
= offset
+ len
;
1162 struct buffer_head
*dibh
;
1165 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
1166 if (unlikely(error
))
1169 gfs2_trans_add_meta(ip
->i_gl
, dibh
);
1171 if (gfs2_is_stuffed(ip
)) {
1172 error
= gfs2_unstuff_dinode(ip
);
1173 if (unlikely(error
))
1177 while (offset
< end
) {
1178 struct iomap iomap
= { };
1180 error
= gfs2_iomap_alloc(inode
, offset
, end
- offset
, &iomap
);
1183 offset
= iomap
.offset
+ iomap
.length
;
1184 if (!(iomap
.flags
& IOMAP_F_NEW
))
1186 error
= sb_issue_zeroout(sb
, iomap
.addr
>> inode
->i_blkbits
,
1187 iomap
.length
>> inode
->i_blkbits
,
1190 fs_err(GFS2_SB(inode
), "Failed to zero data buffers\n");
1200 * calc_max_reserv() - Reverse of write_calc_reserv. Given a number of
1201 * blocks, determine how many bytes can be written.
1202 * @ip: The inode in question.
1203 * @len: Max cap of bytes. What we return in *len must be <= this.
1204 * @data_blocks: Compute and return the number of data blocks needed
1205 * @ind_blocks: Compute and return the number of indirect blocks needed
1206 * @max_blocks: The total blocks available to work with.
1208 * Returns: void, but @len, @data_blocks and @ind_blocks are filled in.
1210 static void calc_max_reserv(struct gfs2_inode
*ip
, loff_t
*len
,
1211 unsigned int *data_blocks
, unsigned int *ind_blocks
,
1212 unsigned int max_blocks
)
1215 const struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
1216 unsigned int tmp
, max_data
= max_blocks
- 3 * (sdp
->sd_max_height
- 1);
1218 for (tmp
= max_data
; tmp
> sdp
->sd_diptrs
;) {
1219 tmp
= DIV_ROUND_UP(tmp
, sdp
->sd_inptrs
);
1223 *data_blocks
= max_data
;
1224 *ind_blocks
= max_blocks
- max_data
;
1225 *len
= ((loff_t
)max_data
- 3) << sdp
->sd_sb
.sb_bsize_shift
;
1228 gfs2_write_calc_reserv(ip
, max
, data_blocks
, ind_blocks
);
1232 static long __gfs2_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
1234 struct inode
*inode
= file_inode(file
);
1235 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1236 struct gfs2_inode
*ip
= GFS2_I(inode
);
1237 struct gfs2_alloc_parms ap
= { .aflags
= 0, };
1238 unsigned int data_blocks
= 0, ind_blocks
= 0, rblocks
;
1239 loff_t bytes
, max_bytes
, max_blks
;
1241 const loff_t pos
= offset
;
1242 const loff_t count
= len
;
1243 loff_t bsize_mask
= ~((loff_t
)sdp
->sd_sb
.sb_bsize
- 1);
1244 loff_t next
= (offset
+ len
- 1) >> sdp
->sd_sb
.sb_bsize_shift
;
1245 loff_t max_chunk_size
= UINT_MAX
& bsize_mask
;
1247 next
= (next
+ 1) << sdp
->sd_sb
.sb_bsize_shift
;
1249 offset
&= bsize_mask
;
1251 len
= next
- offset
;
1252 bytes
= sdp
->sd_max_rg_data
* sdp
->sd_sb
.sb_bsize
/ 2;
1255 bytes
&= bsize_mask
;
1257 bytes
= sdp
->sd_sb
.sb_bsize
;
1259 gfs2_size_hint(file
, offset
, len
);
1261 gfs2_write_calc_reserv(ip
, PAGE_SIZE
, &data_blocks
, &ind_blocks
);
1262 ap
.min_target
= data_blocks
+ ind_blocks
;
1267 if (!gfs2_write_alloc_required(ip
, offset
, bytes
)) {
1273 /* We need to determine how many bytes we can actually
1274 * fallocate without exceeding quota or going over the
1275 * end of the fs. We start off optimistically by assuming
1276 * we can write max_bytes */
1277 max_bytes
= (len
> max_chunk_size
) ? max_chunk_size
: len
;
1279 /* Since max_bytes is most likely a theoretical max, we
1280 * calculate a more realistic 'bytes' to serve as a good
1281 * starting point for the number of bytes we may be able
1283 gfs2_write_calc_reserv(ip
, bytes
, &data_blocks
, &ind_blocks
);
1284 ap
.target
= data_blocks
+ ind_blocks
;
1286 error
= gfs2_quota_lock_check(ip
, &ap
);
1289 /* ap.allowed tells us how many blocks quota will allow
1290 * us to write. Check if this reduces max_blks */
1291 max_blks
= UINT_MAX
;
1293 max_blks
= ap
.allowed
;
1295 error
= gfs2_inplace_reserve(ip
, &ap
);
1299 /* check if the selected rgrp limits our max_blks further */
1300 if (ip
->i_res
.rs_reserved
< max_blks
)
1301 max_blks
= ip
->i_res
.rs_reserved
;
1303 /* Almost done. Calculate bytes that can be written using
1304 * max_blks. We also recompute max_bytes, data_blocks and
1306 calc_max_reserv(ip
, &max_bytes
, &data_blocks
,
1307 &ind_blocks
, max_blks
);
1309 rblocks
= RES_DINODE
+ ind_blocks
+ RES_STATFS
+ RES_QUOTA
+
1310 RES_RG_HDR
+ gfs2_rg_blocks(ip
, data_blocks
+ ind_blocks
);
1311 if (gfs2_is_jdata(ip
))
1312 rblocks
+= data_blocks
? data_blocks
: 1;
1314 error
= gfs2_trans_begin(sdp
, rblocks
,
1315 PAGE_SIZE
>> inode
->i_blkbits
);
1317 goto out_trans_fail
;
1319 error
= fallocate_chunk(inode
, offset
, max_bytes
, mode
);
1320 gfs2_trans_end(sdp
);
1323 goto out_trans_fail
;
1326 offset
+= max_bytes
;
1327 gfs2_inplace_release(ip
);
1328 gfs2_quota_unlock(ip
);
1331 if (!(mode
& FALLOC_FL_KEEP_SIZE
) && (pos
+ count
) > inode
->i_size
)
1332 i_size_write(inode
, pos
+ count
);
1333 file_update_time(file
);
1334 mark_inode_dirty(inode
);
1336 if ((file
->f_flags
& O_DSYNC
) || IS_SYNC(file
->f_mapping
->host
))
1337 return vfs_fsync_range(file
, pos
, pos
+ count
- 1,
1338 (file
->f_flags
& __O_SYNC
) ? 0 : 1);
1342 gfs2_inplace_release(ip
);
1344 gfs2_quota_unlock(ip
);
1348 static long gfs2_fallocate(struct file
*file
, int mode
, loff_t offset
, loff_t len
)
1350 struct inode
*inode
= file_inode(file
);
1351 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
1352 struct gfs2_inode
*ip
= GFS2_I(inode
);
1353 struct gfs2_holder gh
;
1356 if (mode
& ~(FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
))
1358 /* fallocate is needed by gfs2_grow to reserve space in the rindex */
1359 if (gfs2_is_jdata(ip
) && inode
!= sdp
->sd_rindex
)
1364 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
1365 ret
= gfs2_glock_nq(&gh
);
1369 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
1370 (offset
+ len
) > inode
->i_size
) {
1371 ret
= inode_newsize_ok(inode
, offset
+ len
);
1376 ret
= get_write_access(inode
);
1380 if (mode
& FALLOC_FL_PUNCH_HOLE
) {
1381 ret
= __gfs2_punch_hole(file
, offset
, len
);
1383 ret
= __gfs2_fallocate(file
, mode
, offset
, len
);
1385 gfs2_rs_deltree(&ip
->i_res
);
1388 put_write_access(inode
);
1392 gfs2_holder_uninit(&gh
);
1393 inode_unlock(inode
);
1397 static ssize_t
gfs2_file_splice_write(struct pipe_inode_info
*pipe
,
1398 struct file
*out
, loff_t
*ppos
,
1399 size_t len
, unsigned int flags
)
1403 gfs2_size_hint(out
, *ppos
, len
);
1405 ret
= iter_file_splice_write(pipe
, out
, ppos
, len
, flags
);
1409 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1412 * gfs2_lock - acquire/release a posix lock on a file
1413 * @file: the file pointer
1414 * @cmd: either modify or retrieve lock state, possibly wait
1415 * @fl: type and range of lock
1420 static int gfs2_lock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1422 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
1423 struct gfs2_sbd
*sdp
= GFS2_SB(file
->f_mapping
->host
);
1424 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
1426 if (!(fl
->fl_flags
& FL_POSIX
))
1428 if (cmd
== F_CANCELLK
) {
1431 fl
->fl_type
= F_UNLCK
;
1433 if (unlikely(gfs2_withdrawn(sdp
))) {
1434 if (fl
->fl_type
== F_UNLCK
)
1435 locks_lock_file_wait(file
, fl
);
1439 return dlm_posix_get(ls
->ls_dlm
, ip
->i_no_addr
, file
, fl
);
1440 else if (fl
->fl_type
== F_UNLCK
)
1441 return dlm_posix_unlock(ls
->ls_dlm
, ip
->i_no_addr
, file
, fl
);
1443 return dlm_posix_lock(ls
->ls_dlm
, ip
->i_no_addr
, file
, cmd
, fl
);
1446 static void __flock_holder_uninit(struct file
*file
, struct gfs2_holder
*fl_gh
)
1448 struct gfs2_glock
*gl
= gfs2_glock_hold(fl_gh
->gh_gl
);
1451 * Make sure gfs2_glock_put() won't sleep under the file->f_lock
1455 spin_lock(&file
->f_lock
);
1456 gfs2_holder_uninit(fl_gh
);
1457 spin_unlock(&file
->f_lock
);
1461 static int do_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1463 struct gfs2_file
*fp
= file
->private_data
;
1464 struct gfs2_holder
*fl_gh
= &fp
->f_fl_gh
;
1465 struct gfs2_inode
*ip
= GFS2_I(file_inode(file
));
1466 struct gfs2_glock
*gl
;
1472 state
= (fl
->fl_type
== F_WRLCK
) ? LM_ST_EXCLUSIVE
: LM_ST_SHARED
;
1473 flags
= GL_EXACT
| GL_NOPID
;
1474 if (!IS_SETLKW(cmd
))
1475 flags
|= LM_FLAG_TRY_1CB
;
1477 mutex_lock(&fp
->f_fl_mutex
);
1479 if (gfs2_holder_initialized(fl_gh
)) {
1480 struct file_lock request
;
1481 if (fl_gh
->gh_state
== state
)
1483 locks_init_lock(&request
);
1484 request
.fl_type
= F_UNLCK
;
1485 request
.fl_flags
= FL_FLOCK
;
1486 locks_lock_file_wait(file
, &request
);
1487 gfs2_glock_dq(fl_gh
);
1488 gfs2_holder_reinit(state
, flags
, fl_gh
);
1490 error
= gfs2_glock_get(GFS2_SB(&ip
->i_inode
), ip
->i_no_addr
,
1491 &gfs2_flock_glops
, CREATE
, &gl
);
1494 spin_lock(&file
->f_lock
);
1495 gfs2_holder_init(gl
, state
, flags
, fl_gh
);
1496 spin_unlock(&file
->f_lock
);
1499 for (sleeptime
= 1; sleeptime
<= 4; sleeptime
<<= 1) {
1500 error
= gfs2_glock_nq(fl_gh
);
1501 if (error
!= GLR_TRYFAILED
)
1503 fl_gh
->gh_flags
&= ~LM_FLAG_TRY_1CB
;
1504 fl_gh
->gh_flags
|= LM_FLAG_TRY
;
1508 __flock_holder_uninit(file
, fl_gh
);
1509 if (error
== GLR_TRYFAILED
)
1512 error
= locks_lock_file_wait(file
, fl
);
1513 gfs2_assert_warn(GFS2_SB(&ip
->i_inode
), !error
);
1517 mutex_unlock(&fp
->f_fl_mutex
);
1521 static void do_unflock(struct file
*file
, struct file_lock
*fl
)
1523 struct gfs2_file
*fp
= file
->private_data
;
1524 struct gfs2_holder
*fl_gh
= &fp
->f_fl_gh
;
1526 mutex_lock(&fp
->f_fl_mutex
);
1527 locks_lock_file_wait(file
, fl
);
1528 if (gfs2_holder_initialized(fl_gh
)) {
1529 gfs2_glock_dq(fl_gh
);
1530 __flock_holder_uninit(file
, fl_gh
);
1532 mutex_unlock(&fp
->f_fl_mutex
);
1536 * gfs2_flock - acquire/release a flock lock on a file
1537 * @file: the file pointer
1538 * @cmd: either modify or retrieve lock state, possibly wait
1539 * @fl: type and range of lock
1544 static int gfs2_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1546 if (!(fl
->fl_flags
& FL_FLOCK
))
1549 if (fl
->fl_type
== F_UNLCK
) {
1550 do_unflock(file
, fl
);
1553 return do_flock(file
, cmd
, fl
);
1557 const struct file_operations gfs2_file_fops
= {
1558 .llseek
= gfs2_llseek
,
1559 .read_iter
= gfs2_file_read_iter
,
1560 .write_iter
= gfs2_file_write_iter
,
1561 .iopoll
= iocb_bio_iopoll
,
1562 .unlocked_ioctl
= gfs2_ioctl
,
1563 .compat_ioctl
= gfs2_compat_ioctl
,
1566 .release
= gfs2_release
,
1567 .fsync
= gfs2_fsync
,
1569 .flock
= gfs2_flock
,
1570 .splice_read
= generic_file_splice_read
,
1571 .splice_write
= gfs2_file_splice_write
,
1572 .setlease
= simple_nosetlease
,
1573 .fallocate
= gfs2_fallocate
,
1576 const struct file_operations gfs2_dir_fops
= {
1577 .iterate_shared
= gfs2_readdir
,
1578 .unlocked_ioctl
= gfs2_ioctl
,
1579 .compat_ioctl
= gfs2_compat_ioctl
,
1581 .release
= gfs2_release
,
1582 .fsync
= gfs2_fsync
,
1584 .flock
= gfs2_flock
,
1585 .llseek
= default_llseek
,
1588 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1590 const struct file_operations gfs2_file_fops_nolock
= {
1591 .llseek
= gfs2_llseek
,
1592 .read_iter
= gfs2_file_read_iter
,
1593 .write_iter
= gfs2_file_write_iter
,
1594 .iopoll
= iocb_bio_iopoll
,
1595 .unlocked_ioctl
= gfs2_ioctl
,
1596 .compat_ioctl
= gfs2_compat_ioctl
,
1599 .release
= gfs2_release
,
1600 .fsync
= gfs2_fsync
,
1601 .splice_read
= generic_file_splice_read
,
1602 .splice_write
= gfs2_file_splice_write
,
1603 .setlease
= generic_setlease
,
1604 .fallocate
= gfs2_fallocate
,
1607 const struct file_operations gfs2_dir_fops_nolock
= {
1608 .iterate_shared
= gfs2_readdir
,
1609 .unlocked_ioctl
= gfs2_ioctl
,
1610 .compat_ioctl
= gfs2_compat_ioctl
,
1612 .release
= gfs2_release
,
1613 .fsync
= gfs2_fsync
,
1614 .llseek
= default_llseek
,