2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/pagemap.h>
15 #include <linux/uio.h>
16 #include <linux/blkdev.h>
18 #include <linux/mount.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/falloc.h>
22 #include <linux/swap.h>
23 #include <linux/crc32.h>
24 #include <linux/writeback.h>
25 #include <asm/uaccess.h>
26 #include <linux/dlm.h>
27 #include <linux/dlm_plock.h>
28 #include <linux/aio.h>
45 * gfs2_llseek - seek to a location in a file
48 * @whence: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
50 * SEEK_END requires the glock for the file because it references the
53 * Returns: The new offset, or errno
56 static loff_t
gfs2_llseek(struct file
*file
, loff_t offset
, int whence
)
58 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
59 struct gfs2_holder i_gh
;
63 case SEEK_END
: /* These reference inode->i_size */
66 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
69 error
= generic_file_llseek(file
, offset
, whence
);
70 gfs2_glock_dq_uninit(&i_gh
);
75 error
= generic_file_llseek(file
, offset
, whence
);
85 * gfs2_readdir - Iterator for a directory
86 * @file: The directory to read from
87 * @ctx: What to feed directory entries to
92 static int gfs2_readdir(struct file
*file
, struct dir_context
*ctx
)
94 struct inode
*dir
= file
->f_mapping
->host
;
95 struct gfs2_inode
*dip
= GFS2_I(dir
);
96 struct gfs2_holder d_gh
;
99 error
= gfs2_glock_nq_init(dip
->i_gl
, LM_ST_SHARED
, 0, &d_gh
);
103 error
= gfs2_dir_read(dir
, ctx
, &file
->f_ra
);
105 gfs2_glock_dq_uninit(&d_gh
);
112 * @table: A table of 32 u32 flags
113 * @val: a 32 bit value to convert
115 * This function can be used to convert between fsflags values and
116 * GFS2's own flags values.
118 * Returns: the converted flags
120 static u32
fsflags_cvt(const u32
*table
, u32 val
)
132 static const u32 fsflags_to_gfs2
[32] = {
134 [4] = GFS2_DIF_IMMUTABLE
,
135 [5] = GFS2_DIF_APPENDONLY
,
136 [7] = GFS2_DIF_NOATIME
,
137 [12] = GFS2_DIF_EXHASH
,
138 [14] = GFS2_DIF_INHERIT_JDATA
,
139 [17] = GFS2_DIF_TOPDIR
,
142 static const u32 gfs2_to_fsflags
[32] = {
143 [gfs2fl_Sync
] = FS_SYNC_FL
,
144 [gfs2fl_Immutable
] = FS_IMMUTABLE_FL
,
145 [gfs2fl_AppendOnly
] = FS_APPEND_FL
,
146 [gfs2fl_NoAtime
] = FS_NOATIME_FL
,
147 [gfs2fl_ExHash
] = FS_INDEX_FL
,
148 [gfs2fl_TopLevel
] = FS_TOPDIR_FL
,
149 [gfs2fl_InheritJdata
] = FS_JOURNAL_DATA_FL
,
152 static int gfs2_get_flags(struct file
*filp
, u32 __user
*ptr
)
154 struct inode
*inode
= file_inode(filp
);
155 struct gfs2_inode
*ip
= GFS2_I(inode
);
156 struct gfs2_holder gh
;
160 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, 0, &gh
);
161 error
= gfs2_glock_nq(&gh
);
165 fsflags
= fsflags_cvt(gfs2_to_fsflags
, ip
->i_diskflags
);
166 if (!S_ISDIR(inode
->i_mode
) && ip
->i_diskflags
& GFS2_DIF_JDATA
)
167 fsflags
|= FS_JOURNAL_DATA_FL
;
168 if (put_user(fsflags
, ptr
))
172 gfs2_holder_uninit(&gh
);
176 void gfs2_set_inode_flags(struct inode
*inode
)
178 struct gfs2_inode
*ip
= GFS2_I(inode
);
179 unsigned int flags
= inode
->i_flags
;
181 flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_NOSEC
);
182 if ((ip
->i_eattr
== 0) && !is_sxid(inode
->i_mode
))
183 inode
->i_flags
|= S_NOSEC
;
184 if (ip
->i_diskflags
& GFS2_DIF_IMMUTABLE
)
185 flags
|= S_IMMUTABLE
;
186 if (ip
->i_diskflags
& GFS2_DIF_APPENDONLY
)
188 if (ip
->i_diskflags
& GFS2_DIF_NOATIME
)
190 if (ip
->i_diskflags
& GFS2_DIF_SYNC
)
192 inode
->i_flags
= flags
;
195 /* Flags that can be set by user space */
196 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
197 GFS2_DIF_IMMUTABLE| \
198 GFS2_DIF_APPENDONLY| \
203 GFS2_DIF_INHERIT_JDATA)
206 * gfs2_set_flags - set flags on an inode
208 * @flags: The flags to set
209 * @mask: Indicates which flags are valid
212 static int do_gfs2_set_flags(struct file
*filp
, u32 reqflags
, u32 mask
)
214 struct inode
*inode
= file_inode(filp
);
215 struct gfs2_inode
*ip
= GFS2_I(inode
);
216 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
217 struct buffer_head
*bh
;
218 struct gfs2_holder gh
;
220 u32 new_flags
, flags
;
222 error
= mnt_want_write_file(filp
);
226 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
231 if (!inode_owner_or_capable(inode
))
235 flags
= ip
->i_diskflags
;
236 new_flags
= (flags
& ~mask
) | (reqflags
& mask
);
237 if ((new_flags
^ flags
) == 0)
241 if ((new_flags
^ flags
) & ~GFS2_FLAGS_USER_SET
)
245 if (IS_IMMUTABLE(inode
) && (new_flags
& GFS2_DIF_IMMUTABLE
))
247 if (IS_APPEND(inode
) && (new_flags
& GFS2_DIF_APPENDONLY
))
249 if (((new_flags
^ flags
) & GFS2_DIF_IMMUTABLE
) &&
250 !capable(CAP_LINUX_IMMUTABLE
))
252 if (!IS_IMMUTABLE(inode
)) {
253 error
= gfs2_permission(inode
, MAY_WRITE
);
257 if ((flags
^ new_flags
) & GFS2_DIF_JDATA
) {
258 if (flags
& GFS2_DIF_JDATA
)
259 gfs2_log_flush(sdp
, ip
->i_gl
);
260 error
= filemap_fdatawrite(inode
->i_mapping
);
263 error
= filemap_fdatawait(inode
->i_mapping
);
267 error
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
270 error
= gfs2_meta_inode_buffer(ip
, &bh
);
273 gfs2_trans_add_meta(ip
->i_gl
, bh
);
274 ip
->i_diskflags
= new_flags
;
275 gfs2_dinode_out(ip
, bh
->b_data
);
277 gfs2_set_inode_flags(inode
);
278 gfs2_set_aops(inode
);
282 gfs2_glock_dq_uninit(&gh
);
284 mnt_drop_write_file(filp
);
288 static int gfs2_set_flags(struct file
*filp
, u32 __user
*ptr
)
290 struct inode
*inode
= file_inode(filp
);
291 u32 fsflags
, gfsflags
;
293 if (get_user(fsflags
, ptr
))
296 gfsflags
= fsflags_cvt(fsflags_to_gfs2
, fsflags
);
297 if (!S_ISDIR(inode
->i_mode
)) {
298 gfsflags
&= ~GFS2_DIF_TOPDIR
;
299 if (gfsflags
& GFS2_DIF_INHERIT_JDATA
)
300 gfsflags
^= (GFS2_DIF_JDATA
| GFS2_DIF_INHERIT_JDATA
);
301 return do_gfs2_set_flags(filp
, gfsflags
, ~0);
303 return do_gfs2_set_flags(filp
, gfsflags
, ~GFS2_DIF_JDATA
);
306 static long gfs2_ioctl(struct file
*filp
, unsigned int cmd
, unsigned long arg
)
309 case FS_IOC_GETFLAGS
:
310 return gfs2_get_flags(filp
, (u32 __user
*)arg
);
311 case FS_IOC_SETFLAGS
:
312 return gfs2_set_flags(filp
, (u32 __user
*)arg
);
314 return gfs2_fitrim(filp
, (void __user
*)arg
);
320 * gfs2_size_hint - Give a hint to the size of a write request
321 * @file: The struct file
322 * @offset: The file offset of the write
323 * @size: The length of the write
325 * When we are about to do a write, this function records the total
326 * write size in order to provide a suitable hint to the lower layers
327 * about how many blocks will be required.
331 static void gfs2_size_hint(struct file
*filep
, loff_t offset
, size_t size
)
333 struct inode
*inode
= file_inode(filep
);
334 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
335 struct gfs2_inode
*ip
= GFS2_I(inode
);
336 size_t blks
= (size
+ sdp
->sd_sb
.sb_bsize
- 1) >> sdp
->sd_sb
.sb_bsize_shift
;
337 int hint
= min_t(size_t, INT_MAX
, blks
);
339 atomic_set(&ip
->i_res
->rs_sizehint
, hint
);
343 * gfs2_allocate_page_backing - Use bmap to allocate blocks
344 * @page: The (locked) page to allocate backing for
346 * We try to allocate all the blocks required for the page in
347 * one go. This might fail for various reasons, so we keep
348 * trying until all the blocks to back this page are allocated.
349 * If some of the blocks are already allocated, thats ok too.
352 static int gfs2_allocate_page_backing(struct page
*page
)
354 struct inode
*inode
= page
->mapping
->host
;
355 struct buffer_head bh
;
356 unsigned long size
= PAGE_CACHE_SIZE
;
357 u64 lblock
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
362 gfs2_block_map(inode
, lblock
, &bh
, 1);
363 if (!buffer_mapped(&bh
))
366 lblock
+= (bh
.b_size
>> inode
->i_blkbits
);
372 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
373 * @vma: The virtual memory area
374 * @page: The page which is about to become writable
376 * When the page becomes writable, we need to ensure that we have
377 * blocks allocated on disk to back that page.
380 static int gfs2_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
382 struct page
*page
= vmf
->page
;
383 struct inode
*inode
= file_inode(vma
->vm_file
);
384 struct gfs2_inode
*ip
= GFS2_I(inode
);
385 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
386 struct gfs2_alloc_parms ap
= { .aflags
= 0, };
387 unsigned long last_index
;
388 u64 pos
= page
->index
<< PAGE_CACHE_SHIFT
;
389 unsigned int data_blocks
, ind_blocks
, rblocks
;
390 struct gfs2_holder gh
;
394 sb_start_pagefault(inode
->i_sb
);
396 /* Update file times before taking page lock */
397 file_update_time(vma
->vm_file
);
399 ret
= get_write_access(inode
);
403 ret
= gfs2_rs_alloc(ip
);
405 goto out_write_access
;
407 gfs2_size_hint(vma
->vm_file
, pos
, PAGE_CACHE_SIZE
);
409 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
410 ret
= gfs2_glock_nq(&gh
);
414 set_bit(GLF_DIRTY
, &ip
->i_gl
->gl_flags
);
415 set_bit(GIF_SW_PAGED
, &ip
->i_flags
);
417 if (!gfs2_write_alloc_required(ip
, pos
, PAGE_CACHE_SIZE
)) {
419 if (!PageUptodate(page
) || page
->mapping
!= inode
->i_mapping
) {
426 ret
= gfs2_rindex_update(sdp
);
430 ret
= gfs2_quota_lock_check(ip
);
433 gfs2_write_calc_reserv(ip
, PAGE_CACHE_SIZE
, &data_blocks
, &ind_blocks
);
434 ap
.target
= data_blocks
+ ind_blocks
;
435 ret
= gfs2_inplace_reserve(ip
, &ap
);
437 goto out_quota_unlock
;
439 rblocks
= RES_DINODE
+ ind_blocks
;
440 if (gfs2_is_jdata(ip
))
441 rblocks
+= data_blocks
? data_blocks
: 1;
442 if (ind_blocks
|| data_blocks
) {
443 rblocks
+= RES_STATFS
+ RES_QUOTA
;
444 rblocks
+= gfs2_rg_blocks(ip
, data_blocks
+ ind_blocks
);
446 ret
= gfs2_trans_begin(sdp
, rblocks
, 0);
452 size
= i_size_read(inode
);
453 last_index
= (size
- 1) >> PAGE_CACHE_SHIFT
;
454 /* Check page index against inode size */
455 if (size
== 0 || (page
->index
> last_index
))
459 /* If truncated, we must retry the operation, we may have raced
460 * with the glock demotion code.
462 if (!PageUptodate(page
) || page
->mapping
!= inode
->i_mapping
)
465 /* Unstuff, if required, and allocate backing blocks for page */
467 if (gfs2_is_stuffed(ip
))
468 ret
= gfs2_unstuff_dinode(ip
, page
);
470 ret
= gfs2_allocate_page_backing(page
);
477 gfs2_inplace_release(ip
);
479 gfs2_quota_unlock(ip
);
483 gfs2_holder_uninit(&gh
);
485 set_page_dirty(page
);
486 wait_for_stable_page(page
);
489 put_write_access(inode
);
491 sb_end_pagefault(inode
->i_sb
);
492 return block_page_mkwrite_return(ret
);
495 static const struct vm_operations_struct gfs2_vm_ops
= {
496 .fault
= filemap_fault
,
497 .page_mkwrite
= gfs2_page_mkwrite
,
498 .remap_pages
= generic_file_remap_pages
,
503 * @file: The file to map
504 * @vma: The VMA which described the mapping
506 * There is no need to get a lock here unless we should be updating
507 * atime. We ignore any locking errors since the only consequence is
508 * a missed atime update (which will just be deferred until later).
513 static int gfs2_mmap(struct file
*file
, struct vm_area_struct
*vma
)
515 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
517 if (!(file
->f_flags
& O_NOATIME
) &&
518 !IS_NOATIME(&ip
->i_inode
)) {
519 struct gfs2_holder i_gh
;
522 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
526 /* grab lock to update inode */
527 gfs2_glock_dq_uninit(&i_gh
);
530 vma
->vm_ops
= &gfs2_vm_ops
;
536 * gfs2_open_common - This is common to open and atomic_open
537 * @inode: The inode being opened
538 * @file: The file being opened
540 * This maybe called under a glock or not depending upon how it has
541 * been called. We must always be called under a glock for regular
542 * files, however. For other file types, it does not matter whether
543 * we hold the glock or not.
545 * Returns: Error code or 0 for success
548 int gfs2_open_common(struct inode
*inode
, struct file
*file
)
550 struct gfs2_file
*fp
;
553 if (S_ISREG(inode
->i_mode
)) {
554 ret
= generic_file_open(inode
, file
);
559 fp
= kzalloc(sizeof(struct gfs2_file
), GFP_NOFS
);
563 mutex_init(&fp
->f_fl_mutex
);
565 gfs2_assert_warn(GFS2_SB(inode
), !file
->private_data
);
566 file
->private_data
= fp
;
571 * gfs2_open - open a file
572 * @inode: the inode to open
573 * @file: the struct file for this opening
575 * After atomic_open, this function is only used for opening files
576 * which are already cached. We must still get the glock for regular
577 * files to ensure that we have the file size uptodate for the large
578 * file check which is in the common code. That is only an issue for
579 * regular files though.
584 static int gfs2_open(struct inode
*inode
, struct file
*file
)
586 struct gfs2_inode
*ip
= GFS2_I(inode
);
587 struct gfs2_holder i_gh
;
589 bool need_unlock
= false;
591 if (S_ISREG(ip
->i_inode
.i_mode
)) {
592 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
599 error
= gfs2_open_common(inode
, file
);
602 gfs2_glock_dq_uninit(&i_gh
);
608 * gfs2_release - called to close a struct file
609 * @inode: the inode the struct file belongs to
610 * @file: the struct file being closed
615 static int gfs2_release(struct inode
*inode
, struct file
*file
)
617 struct gfs2_inode
*ip
= GFS2_I(inode
);
619 kfree(file
->private_data
);
620 file
->private_data
= NULL
;
622 if (!(file
->f_mode
& FMODE_WRITE
))
625 gfs2_rs_delete(ip
, &inode
->i_writecount
);
630 * gfs2_fsync - sync the dirty data for a file (across the cluster)
631 * @file: the file that points to the dentry
632 * @start: the start position in the file to sync
633 * @end: the end position in the file to sync
634 * @datasync: set if we can ignore timestamp changes
636 * We split the data flushing here so that we don't wait for the data
637 * until after we've also sent the metadata to disk. Note that for
638 * data=ordered, we will write & wait for the data at the log flush
639 * stage anyway, so this is unlikely to make much of a difference
640 * except in the data=writeback case.
642 * If the fdatawrite fails due to any reason except -EIO, we will
643 * continue the remainder of the fsync, although we'll still report
644 * the error at the end. This is to match filemap_write_and_wait_range()
650 static int gfs2_fsync(struct file
*file
, loff_t start
, loff_t end
,
653 struct address_space
*mapping
= file
->f_mapping
;
654 struct inode
*inode
= mapping
->host
;
655 int sync_state
= inode
->i_state
& I_DIRTY
;
656 struct gfs2_inode
*ip
= GFS2_I(inode
);
657 int ret
= 0, ret1
= 0;
659 if (mapping
->nrpages
) {
660 ret1
= filemap_fdatawrite_range(mapping
, start
, end
);
665 if (!gfs2_is_jdata(ip
))
666 sync_state
&= ~I_DIRTY_PAGES
;
668 sync_state
&= ~I_DIRTY_SYNC
;
671 ret
= sync_inode_metadata(inode
, 1);
674 if (gfs2_is_jdata(ip
))
675 filemap_write_and_wait(mapping
);
676 gfs2_ail_flush(ip
->i_gl
, 1);
679 if (mapping
->nrpages
)
680 ret
= filemap_fdatawait_range(mapping
, start
, end
);
682 return ret
? ret
: ret1
;
686 * gfs2_file_aio_write - Perform a write to a file
687 * @iocb: The io context
688 * @iov: The data to write
689 * @nr_segs: Number of @iov segments
690 * @pos: The file position
692 * We have to do a lock/unlock here to refresh the inode size for
693 * O_APPEND writes, otherwise we can land up writing at the wrong
694 * offset. There is still a race, but provided the app is using its
695 * own file locking, this will make O_APPEND work as expected.
699 static ssize_t
gfs2_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
700 unsigned long nr_segs
, loff_t pos
)
702 struct file
*file
= iocb
->ki_filp
;
703 size_t writesize
= iov_length(iov
, nr_segs
);
704 struct gfs2_inode
*ip
= GFS2_I(file_inode(file
));
707 ret
= gfs2_rs_alloc(ip
);
711 gfs2_size_hint(file
, pos
, writesize
);
713 if (file
->f_flags
& O_APPEND
) {
714 struct gfs2_holder gh
;
716 ret
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, 0, &gh
);
719 gfs2_glock_dq_uninit(&gh
);
722 return generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
725 static int fallocate_chunk(struct inode
*inode
, loff_t offset
, loff_t len
,
728 struct gfs2_inode
*ip
= GFS2_I(inode
);
729 struct buffer_head
*dibh
;
732 unsigned int nr_blks
;
733 sector_t lblock
= offset
>> inode
->i_blkbits
;
735 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
739 gfs2_trans_add_meta(ip
->i_gl
, dibh
);
741 if (gfs2_is_stuffed(ip
)) {
742 error
= gfs2_unstuff_dinode(ip
, NULL
);
748 struct buffer_head bh_map
= { .b_state
= 0, .b_blocknr
= 0 };
750 set_buffer_zeronew(&bh_map
);
752 error
= gfs2_block_map(inode
, lblock
, &bh_map
, 1);
755 len
-= bh_map
.b_size
;
756 nr_blks
= bh_map
.b_size
>> inode
->i_blkbits
;
758 if (!buffer_new(&bh_map
))
760 if (unlikely(!buffer_zeronew(&bh_map
))) {
765 if (offset
+ size
> inode
->i_size
&& !(mode
& FALLOC_FL_KEEP_SIZE
))
766 i_size_write(inode
, offset
+ size
);
768 mark_inode_dirty(inode
);
775 static void calc_max_reserv(struct gfs2_inode
*ip
, loff_t max
, loff_t
*len
,
776 unsigned int *data_blocks
, unsigned int *ind_blocks
)
778 const struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
779 unsigned int max_blocks
= ip
->i_rgd
->rd_free_clone
;
780 unsigned int tmp
, max_data
= max_blocks
- 3 * (sdp
->sd_max_height
- 1);
782 for (tmp
= max_data
; tmp
> sdp
->sd_diptrs
;) {
783 tmp
= DIV_ROUND_UP(tmp
, sdp
->sd_inptrs
);
786 /* This calculation isn't the exact reverse of gfs2_write_calc_reserve,
787 so it might end up with fewer data blocks */
788 if (max_data
<= *data_blocks
)
790 *data_blocks
= max_data
;
791 *ind_blocks
= max_blocks
- max_data
;
792 *len
= ((loff_t
)max_data
- 3) << sdp
->sd_sb
.sb_bsize_shift
;
795 gfs2_write_calc_reserv(ip
, max
, data_blocks
, ind_blocks
);
799 static long gfs2_fallocate(struct file
*file
, int mode
, loff_t offset
,
802 struct inode
*inode
= file_inode(file
);
803 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
804 struct gfs2_inode
*ip
= GFS2_I(inode
);
805 struct gfs2_alloc_parms ap
= { .aflags
= 0, };
806 unsigned int data_blocks
= 0, ind_blocks
= 0, rblocks
;
807 loff_t bytes
, max_bytes
;
809 const loff_t pos
= offset
;
810 const loff_t count
= len
;
811 loff_t bsize_mask
= ~((loff_t
)sdp
->sd_sb
.sb_bsize
- 1);
812 loff_t next
= (offset
+ len
- 1) >> sdp
->sd_sb
.sb_bsize_shift
;
813 loff_t max_chunk_size
= UINT_MAX
& bsize_mask
;
814 next
= (next
+ 1) << sdp
->sd_sb
.sb_bsize_shift
;
816 /* We only support the FALLOC_FL_KEEP_SIZE mode */
817 if (mode
& ~FALLOC_FL_KEEP_SIZE
)
820 offset
&= bsize_mask
;
823 bytes
= sdp
->sd_max_rg_data
* sdp
->sd_sb
.sb_bsize
/ 2;
828 bytes
= sdp
->sd_sb
.sb_bsize
;
830 error
= gfs2_rs_alloc(ip
);
834 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &ip
->i_gh
);
835 error
= gfs2_glock_nq(&ip
->i_gh
);
839 gfs2_size_hint(file
, offset
, len
);
844 if (!gfs2_write_alloc_required(ip
, offset
, bytes
)) {
849 error
= gfs2_quota_lock_check(ip
);
854 gfs2_write_calc_reserv(ip
, bytes
, &data_blocks
, &ind_blocks
);
856 ap
.target
= data_blocks
+ ind_blocks
;
857 error
= gfs2_inplace_reserve(ip
, &ap
);
859 if (error
== -ENOSPC
&& bytes
> sdp
->sd_sb
.sb_bsize
) {
863 bytes
= sdp
->sd_sb
.sb_bsize
;
869 calc_max_reserv(ip
, (len
> max_chunk_size
)? max_chunk_size
: len
,
870 &max_bytes
, &data_blocks
, &ind_blocks
);
872 rblocks
= RES_DINODE
+ ind_blocks
+ RES_STATFS
+ RES_QUOTA
+
873 RES_RG_HDR
+ gfs2_rg_blocks(ip
, data_blocks
+ ind_blocks
);
874 if (gfs2_is_jdata(ip
))
875 rblocks
+= data_blocks
? data_blocks
: 1;
877 error
= gfs2_trans_begin(sdp
, rblocks
,
878 PAGE_CACHE_SIZE
/sdp
->sd_sb
.sb_bsize
);
882 error
= fallocate_chunk(inode
, offset
, max_bytes
, mode
);
890 gfs2_inplace_release(ip
);
891 gfs2_quota_unlock(ip
);
895 error
= generic_write_sync(file
, pos
, count
);
899 gfs2_inplace_release(ip
);
901 gfs2_quota_unlock(ip
);
903 gfs2_glock_dq(&ip
->i_gh
);
905 gfs2_holder_uninit(&ip
->i_gh
);
909 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
912 * gfs2_setlease - acquire/release a file lease
913 * @file: the file pointer
917 * We don't currently have a way to enforce a lease across the whole
918 * cluster; until we do, disable leases (by just returning -EINVAL),
919 * unless the administrator has requested purely local locking.
921 * Locking: called under i_lock
926 static int gfs2_setlease(struct file
*file
, long arg
, struct file_lock
**fl
)
932 * gfs2_lock - acquire/release a posix lock on a file
933 * @file: the file pointer
934 * @cmd: either modify or retrieve lock state, possibly wait
935 * @fl: type and range of lock
940 static int gfs2_lock(struct file
*file
, int cmd
, struct file_lock
*fl
)
942 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
943 struct gfs2_sbd
*sdp
= GFS2_SB(file
->f_mapping
->host
);
944 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
946 if (!(fl
->fl_flags
& FL_POSIX
))
948 if (__mandatory_lock(&ip
->i_inode
) && fl
->fl_type
!= F_UNLCK
)
951 if (cmd
== F_CANCELLK
) {
954 fl
->fl_type
= F_UNLCK
;
956 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
))) {
957 if (fl
->fl_type
== F_UNLCK
)
958 posix_lock_file_wait(file
, fl
);
962 return dlm_posix_get(ls
->ls_dlm
, ip
->i_no_addr
, file
, fl
);
963 else if (fl
->fl_type
== F_UNLCK
)
964 return dlm_posix_unlock(ls
->ls_dlm
, ip
->i_no_addr
, file
, fl
);
966 return dlm_posix_lock(ls
->ls_dlm
, ip
->i_no_addr
, file
, cmd
, fl
);
969 static int do_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
971 struct gfs2_file
*fp
= file
->private_data
;
972 struct gfs2_holder
*fl_gh
= &fp
->f_fl_gh
;
973 struct gfs2_inode
*ip
= GFS2_I(file_inode(file
));
974 struct gfs2_glock
*gl
;
979 state
= (fl
->fl_type
== F_WRLCK
) ? LM_ST_EXCLUSIVE
: LM_ST_SHARED
;
980 flags
= (IS_SETLKW(cmd
) ? 0 : LM_FLAG_TRY
) | GL_EXACT
| GL_NOCACHE
;
982 mutex_lock(&fp
->f_fl_mutex
);
986 if (fl_gh
->gh_state
== state
)
988 flock_lock_file_wait(file
,
989 &(struct file_lock
){.fl_type
= F_UNLCK
});
990 gfs2_glock_dq_wait(fl_gh
);
991 gfs2_holder_reinit(state
, flags
, fl_gh
);
993 error
= gfs2_glock_get(GFS2_SB(&ip
->i_inode
), ip
->i_no_addr
,
994 &gfs2_flock_glops
, CREATE
, &gl
);
997 gfs2_holder_init(gl
, state
, flags
, fl_gh
);
1000 error
= gfs2_glock_nq(fl_gh
);
1002 gfs2_holder_uninit(fl_gh
);
1003 if (error
== GLR_TRYFAILED
)
1006 error
= flock_lock_file_wait(file
, fl
);
1007 gfs2_assert_warn(GFS2_SB(&ip
->i_inode
), !error
);
1011 mutex_unlock(&fp
->f_fl_mutex
);
1015 static void do_unflock(struct file
*file
, struct file_lock
*fl
)
1017 struct gfs2_file
*fp
= file
->private_data
;
1018 struct gfs2_holder
*fl_gh
= &fp
->f_fl_gh
;
1020 mutex_lock(&fp
->f_fl_mutex
);
1021 flock_lock_file_wait(file
, fl
);
1023 gfs2_glock_dq_wait(fl_gh
);
1024 gfs2_holder_uninit(fl_gh
);
1026 mutex_unlock(&fp
->f_fl_mutex
);
1030 * gfs2_flock - acquire/release a flock lock on a file
1031 * @file: the file pointer
1032 * @cmd: either modify or retrieve lock state, possibly wait
1033 * @fl: type and range of lock
1038 static int gfs2_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1040 if (!(fl
->fl_flags
& FL_FLOCK
))
1042 if (fl
->fl_type
& LOCK_MAND
)
1045 if (fl
->fl_type
== F_UNLCK
) {
1046 do_unflock(file
, fl
);
1049 return do_flock(file
, cmd
, fl
);
1053 const struct file_operations gfs2_file_fops
= {
1054 .llseek
= gfs2_llseek
,
1055 .read
= do_sync_read
,
1056 .aio_read
= generic_file_aio_read
,
1057 .write
= do_sync_write
,
1058 .aio_write
= gfs2_file_aio_write
,
1059 .unlocked_ioctl
= gfs2_ioctl
,
1062 .release
= gfs2_release
,
1063 .fsync
= gfs2_fsync
,
1065 .flock
= gfs2_flock
,
1066 .splice_read
= generic_file_splice_read
,
1067 .splice_write
= generic_file_splice_write
,
1068 .setlease
= gfs2_setlease
,
1069 .fallocate
= gfs2_fallocate
,
1072 const struct file_operations gfs2_dir_fops
= {
1073 .iterate
= gfs2_readdir
,
1074 .unlocked_ioctl
= gfs2_ioctl
,
1076 .release
= gfs2_release
,
1077 .fsync
= gfs2_fsync
,
1079 .flock
= gfs2_flock
,
1080 .llseek
= default_llseek
,
1083 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1085 const struct file_operations gfs2_file_fops_nolock
= {
1086 .llseek
= gfs2_llseek
,
1087 .read
= do_sync_read
,
1088 .aio_read
= generic_file_aio_read
,
1089 .write
= do_sync_write
,
1090 .aio_write
= gfs2_file_aio_write
,
1091 .unlocked_ioctl
= gfs2_ioctl
,
1094 .release
= gfs2_release
,
1095 .fsync
= gfs2_fsync
,
1096 .splice_read
= generic_file_splice_read
,
1097 .splice_write
= generic_file_splice_write
,
1098 .setlease
= generic_setlease
,
1099 .fallocate
= gfs2_fallocate
,
1102 const struct file_operations gfs2_dir_fops_nolock
= {
1103 .iterate
= gfs2_readdir
,
1104 .unlocked_ioctl
= gfs2_ioctl
,
1106 .release
= gfs2_release
,
1107 .fsync
= gfs2_fsync
,
1108 .llseek
= default_llseek
,