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/ext2_fs.h>
22 #include <linux/falloc.h>
23 #include <linux/swap.h>
24 #include <linux/crc32.h>
25 #include <linux/writeback.h>
26 #include <asm/uaccess.h>
27 #include <linux/dlm.h>
28 #include <linux/dlm_plock.h>
45 * gfs2_llseek - seek to a location in a file
48 * @origin: 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 origin
)
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_unlocked(file
, offset
, origin
);
70 gfs2_glock_dq_uninit(&i_gh
);
75 error
= generic_file_llseek_unlocked(file
, offset
, origin
);
85 * gfs2_readdir - Read directory entries from a directory
86 * @file: The directory to read from
87 * @dirent: Buffer for dirents
88 * @filldir: Function used to do the copying
93 static int gfs2_readdir(struct file
*file
, void *dirent
, filldir_t filldir
)
95 struct inode
*dir
= file
->f_mapping
->host
;
96 struct gfs2_inode
*dip
= GFS2_I(dir
);
97 struct gfs2_holder d_gh
;
98 u64 offset
= file
->f_pos
;
101 gfs2_holder_init(dip
->i_gl
, LM_ST_SHARED
, 0, &d_gh
);
102 error
= gfs2_glock_nq(&d_gh
);
104 gfs2_holder_uninit(&d_gh
);
108 error
= gfs2_dir_read(dir
, &offset
, dirent
, filldir
);
110 gfs2_glock_dq_uninit(&d_gh
);
112 file
->f_pos
= offset
;
119 * @table: A table of 32 u32 flags
120 * @val: a 32 bit value to convert
122 * This function can be used to convert between fsflags values and
123 * GFS2's own flags values.
125 * Returns: the converted flags
127 static u32
fsflags_cvt(const u32
*table
, u32 val
)
139 static const u32 fsflags_to_gfs2
[32] = {
141 [4] = GFS2_DIF_IMMUTABLE
,
142 [5] = GFS2_DIF_APPENDONLY
,
143 [7] = GFS2_DIF_NOATIME
,
144 [12] = GFS2_DIF_EXHASH
,
145 [14] = GFS2_DIF_INHERIT_JDATA
,
148 static const u32 gfs2_to_fsflags
[32] = {
149 [gfs2fl_Sync
] = FS_SYNC_FL
,
150 [gfs2fl_Immutable
] = FS_IMMUTABLE_FL
,
151 [gfs2fl_AppendOnly
] = FS_APPEND_FL
,
152 [gfs2fl_NoAtime
] = FS_NOATIME_FL
,
153 [gfs2fl_ExHash
] = FS_INDEX_FL
,
154 [gfs2fl_InheritJdata
] = FS_JOURNAL_DATA_FL
,
157 static int gfs2_get_flags(struct file
*filp
, u32 __user
*ptr
)
159 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
160 struct gfs2_inode
*ip
= GFS2_I(inode
);
161 struct gfs2_holder gh
;
165 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, 0, &gh
);
166 error
= gfs2_glock_nq(&gh
);
170 fsflags
= fsflags_cvt(gfs2_to_fsflags
, ip
->i_diskflags
);
171 if (!S_ISDIR(inode
->i_mode
) && ip
->i_diskflags
& GFS2_DIF_JDATA
)
172 fsflags
|= FS_JOURNAL_DATA_FL
;
173 if (put_user(fsflags
, ptr
))
177 gfs2_holder_uninit(&gh
);
181 void gfs2_set_inode_flags(struct inode
*inode
)
183 struct gfs2_inode
*ip
= GFS2_I(inode
);
184 unsigned int flags
= inode
->i_flags
;
186 flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_NOSEC
);
187 if ((ip
->i_eattr
== 0) && !is_sxid(inode
->i_mode
))
188 inode
->i_flags
|= S_NOSEC
;
189 if (ip
->i_diskflags
& GFS2_DIF_IMMUTABLE
)
190 flags
|= S_IMMUTABLE
;
191 if (ip
->i_diskflags
& GFS2_DIF_APPENDONLY
)
193 if (ip
->i_diskflags
& GFS2_DIF_NOATIME
)
195 if (ip
->i_diskflags
& GFS2_DIF_SYNC
)
197 inode
->i_flags
= flags
;
200 /* Flags that can be set by user space */
201 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
202 GFS2_DIF_IMMUTABLE| \
203 GFS2_DIF_APPENDONLY| \
207 GFS2_DIF_INHERIT_JDATA)
210 * gfs2_set_flags - set flags on an inode
212 * @flags: The flags to set
213 * @mask: Indicates which flags are valid
216 static int do_gfs2_set_flags(struct file
*filp
, u32 reqflags
, u32 mask
)
218 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
219 struct gfs2_inode
*ip
= GFS2_I(inode
);
220 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
221 struct buffer_head
*bh
;
222 struct gfs2_holder gh
;
224 u32 new_flags
, flags
;
226 error
= mnt_want_write(filp
->f_path
.mnt
);
230 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
235 if (!inode_owner_or_capable(inode
))
239 flags
= ip
->i_diskflags
;
240 new_flags
= (flags
& ~mask
) | (reqflags
& mask
);
241 if ((new_flags
^ flags
) == 0)
245 if ((new_flags
^ flags
) & ~GFS2_FLAGS_USER_SET
)
249 if (IS_IMMUTABLE(inode
) && (new_flags
& GFS2_DIF_IMMUTABLE
))
251 if (IS_APPEND(inode
) && (new_flags
& GFS2_DIF_APPENDONLY
))
253 if (((new_flags
^ flags
) & GFS2_DIF_IMMUTABLE
) &&
254 !capable(CAP_LINUX_IMMUTABLE
))
256 if (!IS_IMMUTABLE(inode
)) {
257 error
= gfs2_permission(inode
, MAY_WRITE
);
261 if ((flags
^ new_flags
) & GFS2_DIF_JDATA
) {
262 if (flags
& GFS2_DIF_JDATA
)
263 gfs2_log_flush(sdp
, ip
->i_gl
);
264 error
= filemap_fdatawrite(inode
->i_mapping
);
267 error
= filemap_fdatawait(inode
->i_mapping
);
271 error
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
274 error
= gfs2_meta_inode_buffer(ip
, &bh
);
277 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
278 ip
->i_diskflags
= new_flags
;
279 gfs2_dinode_out(ip
, bh
->b_data
);
281 gfs2_set_inode_flags(inode
);
282 gfs2_set_aops(inode
);
286 gfs2_glock_dq_uninit(&gh
);
288 mnt_drop_write(filp
->f_path
.mnt
);
292 static int gfs2_set_flags(struct file
*filp
, u32 __user
*ptr
)
294 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
295 u32 fsflags
, gfsflags
;
297 if (get_user(fsflags
, ptr
))
300 gfsflags
= fsflags_cvt(fsflags_to_gfs2
, fsflags
);
301 if (!S_ISDIR(inode
->i_mode
)) {
302 if (gfsflags
& GFS2_DIF_INHERIT_JDATA
)
303 gfsflags
^= (GFS2_DIF_JDATA
| GFS2_DIF_INHERIT_JDATA
);
304 return do_gfs2_set_flags(filp
, gfsflags
, ~0);
306 return do_gfs2_set_flags(filp
, gfsflags
, ~GFS2_DIF_JDATA
);
309 static long gfs2_ioctl(struct file
*filp
, unsigned int cmd
, unsigned long arg
)
312 case FS_IOC_GETFLAGS
:
313 return gfs2_get_flags(filp
, (u32 __user
*)arg
);
314 case FS_IOC_SETFLAGS
:
315 return gfs2_set_flags(filp
, (u32 __user
*)arg
);
321 * gfs2_allocate_page_backing - Use bmap to allocate blocks
322 * @page: The (locked) page to allocate backing for
324 * We try to allocate all the blocks required for the page in
325 * one go. This might fail for various reasons, so we keep
326 * trying until all the blocks to back this page are allocated.
327 * If some of the blocks are already allocated, thats ok too.
330 static int gfs2_allocate_page_backing(struct page
*page
)
332 struct inode
*inode
= page
->mapping
->host
;
333 struct buffer_head bh
;
334 unsigned long size
= PAGE_CACHE_SIZE
;
335 u64 lblock
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
340 gfs2_block_map(inode
, lblock
, &bh
, 1);
341 if (!buffer_mapped(&bh
))
344 lblock
+= (bh
.b_size
>> inode
->i_blkbits
);
350 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
351 * @vma: The virtual memory area
352 * @page: The page which is about to become writable
354 * When the page becomes writable, we need to ensure that we have
355 * blocks allocated on disk to back that page.
358 static int gfs2_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
360 struct page
*page
= vmf
->page
;
361 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
362 struct gfs2_inode
*ip
= GFS2_I(inode
);
363 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
364 unsigned long last_index
;
365 u64 pos
= page
->index
<< PAGE_CACHE_SHIFT
;
366 unsigned int data_blocks
, ind_blocks
, rblocks
;
367 struct gfs2_holder gh
;
368 struct gfs2_alloc
*al
;
371 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
372 ret
= gfs2_glock_nq(&gh
);
376 set_bit(GLF_DIRTY
, &ip
->i_gl
->gl_flags
);
377 set_bit(GIF_SW_PAGED
, &ip
->i_flags
);
379 if (!gfs2_write_alloc_required(ip
, pos
, PAGE_CACHE_SIZE
))
382 al
= gfs2_alloc_get(ip
);
386 ret
= gfs2_quota_lock_check(ip
);
389 gfs2_write_calc_reserv(ip
, PAGE_CACHE_SIZE
, &data_blocks
, &ind_blocks
);
390 al
->al_requested
= data_blocks
+ ind_blocks
;
391 ret
= gfs2_inplace_reserve(ip
);
393 goto out_quota_unlock
;
395 rblocks
= RES_DINODE
+ ind_blocks
;
396 if (gfs2_is_jdata(ip
))
397 rblocks
+= data_blocks
? data_blocks
: 1;
398 if (ind_blocks
|| data_blocks
) {
399 rblocks
+= RES_STATFS
+ RES_QUOTA
;
400 rblocks
+= gfs2_rg_blocks(al
);
402 ret
= gfs2_trans_begin(sdp
, rblocks
, 0);
408 last_index
= ip
->i_inode
.i_size
>> PAGE_CACHE_SHIFT
;
409 if (page
->index
> last_index
)
410 goto out_unlock_page
;
412 if (!PageUptodate(page
) || page
->mapping
!= ip
->i_inode
.i_mapping
)
413 goto out_unlock_page
;
414 if (gfs2_is_stuffed(ip
)) {
415 ret
= gfs2_unstuff_dinode(ip
, page
);
417 goto out_unlock_page
;
419 ret
= gfs2_allocate_page_backing(page
);
425 gfs2_inplace_release(ip
);
427 gfs2_quota_unlock(ip
);
433 gfs2_holder_uninit(&gh
);
437 ret
= VM_FAULT_SIGBUS
;
441 static const struct vm_operations_struct gfs2_vm_ops
= {
442 .fault
= filemap_fault
,
443 .page_mkwrite
= gfs2_page_mkwrite
,
448 * @file: The file to map
449 * @vma: The VMA which described the mapping
451 * There is no need to get a lock here unless we should be updating
452 * atime. We ignore any locking errors since the only consequence is
453 * a missed atime update (which will just be deferred until later).
458 static int gfs2_mmap(struct file
*file
, struct vm_area_struct
*vma
)
460 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
462 if (!(file
->f_flags
& O_NOATIME
) &&
463 !IS_NOATIME(&ip
->i_inode
)) {
464 struct gfs2_holder i_gh
;
467 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
, &i_gh
);
468 error
= gfs2_glock_nq(&i_gh
);
471 gfs2_glock_dq(&i_gh
);
473 gfs2_holder_uninit(&i_gh
);
477 vma
->vm_ops
= &gfs2_vm_ops
;
478 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
484 * gfs2_open - open a file
485 * @inode: the inode to open
486 * @file: the struct file for this opening
491 static int gfs2_open(struct inode
*inode
, struct file
*file
)
493 struct gfs2_inode
*ip
= GFS2_I(inode
);
494 struct gfs2_holder i_gh
;
495 struct gfs2_file
*fp
;
498 fp
= kzalloc(sizeof(struct gfs2_file
), GFP_KERNEL
);
502 mutex_init(&fp
->f_fl_mutex
);
504 gfs2_assert_warn(GFS2_SB(inode
), !file
->private_data
);
505 file
->private_data
= fp
;
507 if (S_ISREG(ip
->i_inode
.i_mode
)) {
508 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
513 if (!(file
->f_flags
& O_LARGEFILE
) &&
514 i_size_read(inode
) > MAX_NON_LFS
) {
519 gfs2_glock_dq_uninit(&i_gh
);
525 gfs2_glock_dq_uninit(&i_gh
);
527 file
->private_data
= NULL
;
533 * gfs2_close - called to close a struct file
534 * @inode: the inode the struct file belongs to
535 * @file: the struct file being closed
540 static int gfs2_close(struct inode
*inode
, struct file
*file
)
542 struct gfs2_sbd
*sdp
= inode
->i_sb
->s_fs_info
;
543 struct gfs2_file
*fp
;
545 fp
= file
->private_data
;
546 file
->private_data
= NULL
;
548 if (gfs2_assert_warn(sdp
, fp
))
557 * gfs2_fsync - sync the dirty data for a file (across the cluster)
558 * @file: the file that points to the dentry
559 * @start: the start position in the file to sync
560 * @end: the end position in the file to sync
561 * @datasync: set if we can ignore timestamp changes
563 * We split the data flushing here so that we don't wait for the data
564 * until after we've also sent the metadata to disk. Note that for
565 * data=ordered, we will write & wait for the data at the log flush
566 * stage anyway, so this is unlikely to make much of a difference
567 * except in the data=writeback case.
569 * If the fdatawrite fails due to any reason except -EIO, we will
570 * continue the remainder of the fsync, although we'll still report
571 * the error at the end. This is to match filemap_write_and_wait_range()
577 static int gfs2_fsync(struct file
*file
, loff_t start
, loff_t end
,
580 struct address_space
*mapping
= file
->f_mapping
;
581 struct inode
*inode
= mapping
->host
;
582 int sync_state
= inode
->i_state
& (I_DIRTY_SYNC
|I_DIRTY_DATASYNC
);
583 struct gfs2_inode
*ip
= GFS2_I(inode
);
586 if (mapping
->nrpages
) {
587 ret1
= filemap_fdatawrite_range(mapping
, start
, end
);
593 sync_state
&= ~I_DIRTY_SYNC
;
596 mutex_lock(&inode
->i_mutex
);
597 ret
= sync_inode_metadata(inode
, 1);
599 mutex_unlock(&inode
->i_mutex
);
602 if (gfs2_is_jdata(ip
))
603 filemap_write_and_wait(mapping
);
604 gfs2_ail_flush(ip
->i_gl
);
605 mutex_unlock(&inode
->i_mutex
);
608 if (mapping
->nrpages
)
609 ret
= filemap_fdatawait_range(mapping
, start
, end
);
611 return ret
? ret
: ret1
;
615 * gfs2_file_aio_write - Perform a write to a file
616 * @iocb: The io context
617 * @iov: The data to write
618 * @nr_segs: Number of @iov segments
619 * @pos: The file position
621 * We have to do a lock/unlock here to refresh the inode size for
622 * O_APPEND writes, otherwise we can land up writing at the wrong
623 * offset. There is still a race, but provided the app is using its
624 * own file locking, this will make O_APPEND work as expected.
628 static ssize_t
gfs2_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
629 unsigned long nr_segs
, loff_t pos
)
631 struct file
*file
= iocb
->ki_filp
;
633 if (file
->f_flags
& O_APPEND
) {
634 struct dentry
*dentry
= file
->f_dentry
;
635 struct gfs2_inode
*ip
= GFS2_I(dentry
->d_inode
);
636 struct gfs2_holder gh
;
639 ret
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, 0, &gh
);
642 gfs2_glock_dq_uninit(&gh
);
645 return generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
648 static int empty_write_end(struct page
*page
, unsigned from
,
649 unsigned to
, int mode
)
651 struct inode
*inode
= page
->mapping
->host
;
652 struct gfs2_inode
*ip
= GFS2_I(inode
);
653 struct buffer_head
*bh
;
654 unsigned offset
, blksize
= 1 << inode
->i_blkbits
;
655 pgoff_t end_index
= i_size_read(inode
) >> PAGE_CACHE_SHIFT
;
657 zero_user(page
, from
, to
-from
);
658 mark_page_accessed(page
);
660 if (page
->index
< end_index
|| !(mode
& FALLOC_FL_KEEP_SIZE
)) {
661 if (!gfs2_is_writeback(ip
))
662 gfs2_page_add_databufs(ip
, page
, from
, to
);
664 block_commit_write(page
, from
, to
);
669 bh
= page_buffers(page
);
670 while (offset
< to
) {
671 if (offset
>= from
) {
672 set_buffer_uptodate(bh
);
673 mark_buffer_dirty(bh
);
674 clear_buffer_new(bh
);
675 write_dirty_buffer(bh
, WRITE
);
678 bh
= bh
->b_this_page
;
682 bh
= page_buffers(page
);
683 while (offset
< to
) {
684 if (offset
>= from
) {
686 if (!buffer_uptodate(bh
))
690 bh
= bh
->b_this_page
;
695 static int needs_empty_write(sector_t block
, struct inode
*inode
)
698 struct buffer_head bh_map
= { .b_state
= 0, .b_blocknr
= 0 };
700 bh_map
.b_size
= 1 << inode
->i_blkbits
;
701 error
= gfs2_block_map(inode
, block
, &bh_map
, 0);
704 return !buffer_mapped(&bh_map
);
707 static int write_empty_blocks(struct page
*page
, unsigned from
, unsigned to
,
710 struct inode
*inode
= page
->mapping
->host
;
711 unsigned start
, end
, next
, blksize
;
712 sector_t block
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
715 blksize
= 1 << inode
->i_blkbits
;
717 while (next
< from
) {
724 ret
= needs_empty_write(block
, inode
);
725 if (unlikely(ret
< 0))
729 ret
= __block_write_begin(page
, start
, end
- start
,
733 ret
= empty_write_end(page
, start
, end
, mode
);
746 ret
= __block_write_begin(page
, start
, end
- start
, gfs2_block_map
);
749 ret
= empty_write_end(page
, start
, end
, mode
);
757 static int fallocate_chunk(struct inode
*inode
, loff_t offset
, loff_t len
,
760 struct gfs2_inode
*ip
= GFS2_I(inode
);
761 struct buffer_head
*dibh
;
763 u64 start
= offset
>> PAGE_CACHE_SHIFT
;
764 unsigned int start_offset
= offset
& ~PAGE_CACHE_MASK
;
765 u64 end
= (offset
+ len
- 1) >> PAGE_CACHE_SHIFT
;
768 unsigned int end_offset
= (offset
+ len
) & ~PAGE_CACHE_MASK
;
769 unsigned int from
, to
;
772 end_offset
= PAGE_CACHE_SIZE
;
774 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
778 gfs2_trans_add_bh(ip
->i_gl
, dibh
, 1);
780 if (gfs2_is_stuffed(ip
)) {
781 error
= gfs2_unstuff_dinode(ip
, NULL
);
787 offset
= start
<< PAGE_CACHE_SHIFT
;
789 to
= PAGE_CACHE_SIZE
;
790 while (curr
<= end
) {
791 page
= grab_cache_page_write_begin(inode
->i_mapping
, curr
,
793 if (unlikely(!page
)) {
800 error
= write_empty_blocks(page
, from
, to
, mode
);
801 if (!error
&& offset
+ to
> inode
->i_size
&&
802 !(mode
& FALLOC_FL_KEEP_SIZE
)) {
803 i_size_write(inode
, offset
+ to
);
806 page_cache_release(page
);
810 offset
+= PAGE_CACHE_SIZE
;
814 mark_inode_dirty(inode
);
822 static void calc_max_reserv(struct gfs2_inode
*ip
, loff_t max
, loff_t
*len
,
823 unsigned int *data_blocks
, unsigned int *ind_blocks
)
825 const struct gfs2_sbd
*sdp
= GFS2_SB(&ip
->i_inode
);
826 unsigned int max_blocks
= ip
->i_alloc
->al_rgd
->rd_free_clone
;
827 unsigned int tmp
, max_data
= max_blocks
- 3 * (sdp
->sd_max_height
- 1);
829 for (tmp
= max_data
; tmp
> sdp
->sd_diptrs
;) {
830 tmp
= DIV_ROUND_UP(tmp
, sdp
->sd_inptrs
);
833 /* This calculation isn't the exact reverse of gfs2_write_calc_reserve,
834 so it might end up with fewer data blocks */
835 if (max_data
<= *data_blocks
)
837 *data_blocks
= max_data
;
838 *ind_blocks
= max_blocks
- max_data
;
839 *len
= ((loff_t
)max_data
- 3) << sdp
->sd_sb
.sb_bsize_shift
;
842 gfs2_write_calc_reserv(ip
, max
, data_blocks
, ind_blocks
);
846 static long gfs2_fallocate(struct file
*file
, int mode
, loff_t offset
,
849 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
850 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
851 struct gfs2_inode
*ip
= GFS2_I(inode
);
852 unsigned int data_blocks
= 0, ind_blocks
= 0, rblocks
;
853 loff_t bytes
, max_bytes
;
854 struct gfs2_alloc
*al
;
856 loff_t bsize_mask
= ~((loff_t
)sdp
->sd_sb
.sb_bsize
- 1);
857 loff_t next
= (offset
+ len
- 1) >> sdp
->sd_sb
.sb_bsize_shift
;
858 next
= (next
+ 1) << sdp
->sd_sb
.sb_bsize_shift
;
860 /* We only support the FALLOC_FL_KEEP_SIZE mode */
861 if (mode
& ~FALLOC_FL_KEEP_SIZE
)
864 offset
&= bsize_mask
;
867 bytes
= sdp
->sd_max_rg_data
* sdp
->sd_sb
.sb_bsize
/ 2;
872 bytes
= sdp
->sd_sb
.sb_bsize
;
874 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &ip
->i_gh
);
875 error
= gfs2_glock_nq(&ip
->i_gh
);
879 if (!gfs2_write_alloc_required(ip
, offset
, len
))
885 al
= gfs2_alloc_get(ip
);
891 error
= gfs2_quota_lock_check(ip
);
896 gfs2_write_calc_reserv(ip
, bytes
, &data_blocks
, &ind_blocks
);
898 al
->al_requested
= data_blocks
+ ind_blocks
;
899 error
= gfs2_inplace_reserve(ip
);
901 if (error
== -ENOSPC
&& bytes
> sdp
->sd_sb
.sb_bsize
) {
905 bytes
= sdp
->sd_sb
.sb_bsize
;
911 calc_max_reserv(ip
, len
, &max_bytes
, &data_blocks
, &ind_blocks
);
912 al
->al_requested
= data_blocks
+ ind_blocks
;
914 rblocks
= RES_DINODE
+ ind_blocks
+ RES_STATFS
+ RES_QUOTA
+
915 RES_RG_HDR
+ gfs2_rg_blocks(al
);
916 if (gfs2_is_jdata(ip
))
917 rblocks
+= data_blocks
? data_blocks
: 1;
919 error
= gfs2_trans_begin(sdp
, rblocks
,
920 PAGE_CACHE_SIZE
/sdp
->sd_sb
.sb_bsize
);
924 error
= fallocate_chunk(inode
, offset
, max_bytes
, mode
);
932 gfs2_inplace_release(ip
);
933 gfs2_quota_unlock(ip
);
939 gfs2_inplace_release(ip
);
941 gfs2_quota_unlock(ip
);
945 gfs2_glock_dq(&ip
->i_gh
);
947 gfs2_holder_uninit(&ip
->i_gh
);
951 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
954 * gfs2_setlease - acquire/release a file lease
955 * @file: the file pointer
959 * We don't currently have a way to enforce a lease across the whole
960 * cluster; until we do, disable leases (by just returning -EINVAL),
961 * unless the administrator has requested purely local locking.
963 * Locking: called under lock_flocks
968 static int gfs2_setlease(struct file
*file
, long arg
, struct file_lock
**fl
)
974 * gfs2_lock - acquire/release a posix lock on a file
975 * @file: the file pointer
976 * @cmd: either modify or retrieve lock state, possibly wait
977 * @fl: type and range of lock
982 static int gfs2_lock(struct file
*file
, int cmd
, struct file_lock
*fl
)
984 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
985 struct gfs2_sbd
*sdp
= GFS2_SB(file
->f_mapping
->host
);
986 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
988 if (!(fl
->fl_flags
& FL_POSIX
))
990 if (__mandatory_lock(&ip
->i_inode
) && fl
->fl_type
!= F_UNLCK
)
993 if (cmd
== F_CANCELLK
) {
996 fl
->fl_type
= F_UNLCK
;
998 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
1001 return dlm_posix_get(ls
->ls_dlm
, ip
->i_no_addr
, file
, fl
);
1002 else if (fl
->fl_type
== F_UNLCK
)
1003 return dlm_posix_unlock(ls
->ls_dlm
, ip
->i_no_addr
, file
, fl
);
1005 return dlm_posix_lock(ls
->ls_dlm
, ip
->i_no_addr
, file
, cmd
, fl
);
1008 static int do_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1010 struct gfs2_file
*fp
= file
->private_data
;
1011 struct gfs2_holder
*fl_gh
= &fp
->f_fl_gh
;
1012 struct gfs2_inode
*ip
= GFS2_I(file
->f_path
.dentry
->d_inode
);
1013 struct gfs2_glock
*gl
;
1018 state
= (fl
->fl_type
== F_WRLCK
) ? LM_ST_EXCLUSIVE
: LM_ST_SHARED
;
1019 flags
= (IS_SETLKW(cmd
) ? 0 : LM_FLAG_TRY
) | GL_EXACT
| GL_NOCACHE
;
1021 mutex_lock(&fp
->f_fl_mutex
);
1025 if (fl_gh
->gh_state
== state
)
1027 flock_lock_file_wait(file
,
1028 &(struct file_lock
){.fl_type
= F_UNLCK
});
1029 gfs2_glock_dq_wait(fl_gh
);
1030 gfs2_holder_reinit(state
, flags
, fl_gh
);
1032 error
= gfs2_glock_get(GFS2_SB(&ip
->i_inode
), ip
->i_no_addr
,
1033 &gfs2_flock_glops
, CREATE
, &gl
);
1036 gfs2_holder_init(gl
, state
, flags
, fl_gh
);
1039 error
= gfs2_glock_nq(fl_gh
);
1041 gfs2_holder_uninit(fl_gh
);
1042 if (error
== GLR_TRYFAILED
)
1045 error
= flock_lock_file_wait(file
, fl
);
1046 gfs2_assert_warn(GFS2_SB(&ip
->i_inode
), !error
);
1050 mutex_unlock(&fp
->f_fl_mutex
);
1054 static void do_unflock(struct file
*file
, struct file_lock
*fl
)
1056 struct gfs2_file
*fp
= file
->private_data
;
1057 struct gfs2_holder
*fl_gh
= &fp
->f_fl_gh
;
1059 mutex_lock(&fp
->f_fl_mutex
);
1060 flock_lock_file_wait(file
, fl
);
1062 gfs2_glock_dq_wait(fl_gh
);
1063 gfs2_holder_uninit(fl_gh
);
1065 mutex_unlock(&fp
->f_fl_mutex
);
1069 * gfs2_flock - acquire/release a flock lock on a file
1070 * @file: the file pointer
1071 * @cmd: either modify or retrieve lock state, possibly wait
1072 * @fl: type and range of lock
1077 static int gfs2_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
1079 if (!(fl
->fl_flags
& FL_FLOCK
))
1081 if (fl
->fl_type
& LOCK_MAND
)
1084 if (fl
->fl_type
== F_UNLCK
) {
1085 do_unflock(file
, fl
);
1088 return do_flock(file
, cmd
, fl
);
1092 const struct file_operations gfs2_file_fops
= {
1093 .llseek
= gfs2_llseek
,
1094 .read
= do_sync_read
,
1095 .aio_read
= generic_file_aio_read
,
1096 .write
= do_sync_write
,
1097 .aio_write
= gfs2_file_aio_write
,
1098 .unlocked_ioctl
= gfs2_ioctl
,
1101 .release
= gfs2_close
,
1102 .fsync
= gfs2_fsync
,
1104 .flock
= gfs2_flock
,
1105 .splice_read
= generic_file_splice_read
,
1106 .splice_write
= generic_file_splice_write
,
1107 .setlease
= gfs2_setlease
,
1108 .fallocate
= gfs2_fallocate
,
1111 const struct file_operations gfs2_dir_fops
= {
1112 .readdir
= gfs2_readdir
,
1113 .unlocked_ioctl
= gfs2_ioctl
,
1115 .release
= gfs2_close
,
1116 .fsync
= gfs2_fsync
,
1118 .flock
= gfs2_flock
,
1119 .llseek
= default_llseek
,
1122 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
1124 const struct file_operations gfs2_file_fops_nolock
= {
1125 .llseek
= gfs2_llseek
,
1126 .read
= do_sync_read
,
1127 .aio_read
= generic_file_aio_read
,
1128 .write
= do_sync_write
,
1129 .aio_write
= gfs2_file_aio_write
,
1130 .unlocked_ioctl
= gfs2_ioctl
,
1133 .release
= gfs2_close
,
1134 .fsync
= gfs2_fsync
,
1135 .splice_read
= generic_file_splice_read
,
1136 .splice_write
= generic_file_splice_write
,
1137 .setlease
= generic_setlease
,
1138 .fallocate
= gfs2_fallocate
,
1141 const struct file_operations gfs2_dir_fops_nolock
= {
1142 .readdir
= gfs2_readdir
,
1143 .unlocked_ioctl
= gfs2_ioctl
,
1145 .release
= gfs2_close
,
1146 .fsync
= gfs2_fsync
,
1147 .llseek
= default_llseek
,