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/crc32.h>
23 #include <linux/writeback.h>
24 #include <asm/uaccess.h>
25 #include <linux/dlm.h>
26 #include <linux/dlm_plock.h>
43 * gfs2_llseek - seek to a location in a file
46 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
48 * SEEK_END requires the glock for the file because it references the
51 * Returns: The new offset, or errno
54 static loff_t
gfs2_llseek(struct file
*file
, loff_t offset
, int origin
)
56 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
57 struct gfs2_holder i_gh
;
61 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
64 error
= generic_file_llseek_unlocked(file
, offset
, origin
);
65 gfs2_glock_dq_uninit(&i_gh
);
68 error
= generic_file_llseek_unlocked(file
, offset
, origin
);
74 * gfs2_readdir - Read directory entries from a directory
75 * @file: The directory to read from
76 * @dirent: Buffer for dirents
77 * @filldir: Function used to do the copying
82 static int gfs2_readdir(struct file
*file
, void *dirent
, filldir_t filldir
)
84 struct inode
*dir
= file
->f_mapping
->host
;
85 struct gfs2_inode
*dip
= GFS2_I(dir
);
86 struct gfs2_holder d_gh
;
87 u64 offset
= file
->f_pos
;
90 gfs2_holder_init(dip
->i_gl
, LM_ST_SHARED
, 0, &d_gh
);
91 error
= gfs2_glock_nq(&d_gh
);
93 gfs2_holder_uninit(&d_gh
);
97 error
= gfs2_dir_read(dir
, &offset
, dirent
, filldir
);
99 gfs2_glock_dq_uninit(&d_gh
);
101 file
->f_pos
= offset
;
108 * @table: A table of 32 u32 flags
109 * @val: a 32 bit value to convert
111 * This function can be used to convert between fsflags values and
112 * GFS2's own flags values.
114 * Returns: the converted flags
116 static u32
fsflags_cvt(const u32
*table
, u32 val
)
128 static const u32 fsflags_to_gfs2
[32] = {
130 [4] = GFS2_DIF_IMMUTABLE
,
131 [5] = GFS2_DIF_APPENDONLY
,
132 [7] = GFS2_DIF_NOATIME
,
133 [12] = GFS2_DIF_EXHASH
,
134 [14] = GFS2_DIF_INHERIT_JDATA
,
137 static const u32 gfs2_to_fsflags
[32] = {
138 [gfs2fl_Sync
] = FS_SYNC_FL
,
139 [gfs2fl_Immutable
] = FS_IMMUTABLE_FL
,
140 [gfs2fl_AppendOnly
] = FS_APPEND_FL
,
141 [gfs2fl_NoAtime
] = FS_NOATIME_FL
,
142 [gfs2fl_ExHash
] = FS_INDEX_FL
,
143 [gfs2fl_InheritJdata
] = FS_JOURNAL_DATA_FL
,
146 static int gfs2_get_flags(struct file
*filp
, u32 __user
*ptr
)
148 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
149 struct gfs2_inode
*ip
= GFS2_I(inode
);
150 struct gfs2_holder gh
;
154 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, 0, &gh
);
155 error
= gfs2_glock_nq(&gh
);
159 fsflags
= fsflags_cvt(gfs2_to_fsflags
, ip
->i_diskflags
);
160 if (!S_ISDIR(inode
->i_mode
) && ip
->i_diskflags
& GFS2_DIF_JDATA
)
161 fsflags
|= FS_JOURNAL_DATA_FL
;
162 if (put_user(fsflags
, ptr
))
166 gfs2_holder_uninit(&gh
);
170 void gfs2_set_inode_flags(struct inode
*inode
)
172 struct gfs2_inode
*ip
= GFS2_I(inode
);
173 unsigned int flags
= inode
->i_flags
;
175 flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
176 if (ip
->i_diskflags
& GFS2_DIF_IMMUTABLE
)
177 flags
|= S_IMMUTABLE
;
178 if (ip
->i_diskflags
& GFS2_DIF_APPENDONLY
)
180 if (ip
->i_diskflags
& GFS2_DIF_NOATIME
)
182 if (ip
->i_diskflags
& GFS2_DIF_SYNC
)
184 inode
->i_flags
= flags
;
187 /* Flags that can be set by user space */
188 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
189 GFS2_DIF_IMMUTABLE| \
190 GFS2_DIF_APPENDONLY| \
194 GFS2_DIF_INHERIT_JDATA)
197 * gfs2_set_flags - set flags on an inode
199 * @flags: The flags to set
200 * @mask: Indicates which flags are valid
203 static int do_gfs2_set_flags(struct file
*filp
, u32 reqflags
, u32 mask
)
205 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
206 struct gfs2_inode
*ip
= GFS2_I(inode
);
207 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
208 struct buffer_head
*bh
;
209 struct gfs2_holder gh
;
211 u32 new_flags
, flags
;
213 error
= mnt_want_write(filp
->f_path
.mnt
);
217 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
222 if (!is_owner_or_cap(inode
))
226 flags
= ip
->i_diskflags
;
227 new_flags
= (flags
& ~mask
) | (reqflags
& mask
);
228 if ((new_flags
^ flags
) == 0)
232 if ((new_flags
^ flags
) & ~GFS2_FLAGS_USER_SET
)
236 if (IS_IMMUTABLE(inode
) && (new_flags
& GFS2_DIF_IMMUTABLE
))
238 if (IS_APPEND(inode
) && (new_flags
& GFS2_DIF_APPENDONLY
))
240 if (((new_flags
^ flags
) & GFS2_DIF_IMMUTABLE
) &&
241 !capable(CAP_LINUX_IMMUTABLE
))
243 if (!IS_IMMUTABLE(inode
)) {
244 error
= gfs2_permission(inode
, MAY_WRITE
);
248 if ((flags
^ new_flags
) & GFS2_DIF_JDATA
) {
249 if (flags
& GFS2_DIF_JDATA
)
250 gfs2_log_flush(sdp
, ip
->i_gl
);
251 error
= filemap_fdatawrite(inode
->i_mapping
);
254 error
= filemap_fdatawait(inode
->i_mapping
);
258 error
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
261 error
= gfs2_meta_inode_buffer(ip
, &bh
);
264 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
265 ip
->i_diskflags
= new_flags
;
266 gfs2_dinode_out(ip
, bh
->b_data
);
268 gfs2_set_inode_flags(inode
);
269 gfs2_set_aops(inode
);
273 gfs2_glock_dq_uninit(&gh
);
275 mnt_drop_write(filp
->f_path
.mnt
);
279 static int gfs2_set_flags(struct file
*filp
, u32 __user
*ptr
)
281 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
282 u32 fsflags
, gfsflags
;
284 if (get_user(fsflags
, ptr
))
287 gfsflags
= fsflags_cvt(fsflags_to_gfs2
, fsflags
);
288 if (!S_ISDIR(inode
->i_mode
)) {
289 if (gfsflags
& GFS2_DIF_INHERIT_JDATA
)
290 gfsflags
^= (GFS2_DIF_JDATA
| GFS2_DIF_INHERIT_JDATA
);
291 return do_gfs2_set_flags(filp
, gfsflags
, ~0);
293 return do_gfs2_set_flags(filp
, gfsflags
, ~GFS2_DIF_JDATA
);
296 static long gfs2_ioctl(struct file
*filp
, unsigned int cmd
, unsigned long arg
)
299 case FS_IOC_GETFLAGS
:
300 return gfs2_get_flags(filp
, (u32 __user
*)arg
);
301 case FS_IOC_SETFLAGS
:
302 return gfs2_set_flags(filp
, (u32 __user
*)arg
);
308 * gfs2_allocate_page_backing - Use bmap to allocate blocks
309 * @page: The (locked) page to allocate backing for
311 * We try to allocate all the blocks required for the page in
312 * one go. This might fail for various reasons, so we keep
313 * trying until all the blocks to back this page are allocated.
314 * If some of the blocks are already allocated, thats ok too.
317 static int gfs2_allocate_page_backing(struct page
*page
)
319 struct inode
*inode
= page
->mapping
->host
;
320 struct buffer_head bh
;
321 unsigned long size
= PAGE_CACHE_SIZE
;
322 u64 lblock
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
327 gfs2_block_map(inode
, lblock
, &bh
, 1);
328 if (!buffer_mapped(&bh
))
331 lblock
+= (bh
.b_size
>> inode
->i_blkbits
);
337 * gfs2_page_mkwrite - Make a shared, mmap()ed, page writable
338 * @vma: The virtual memory area
339 * @page: The page which is about to become writable
341 * When the page becomes writable, we need to ensure that we have
342 * blocks allocated on disk to back that page.
345 static int gfs2_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
347 struct page
*page
= vmf
->page
;
348 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
349 struct gfs2_inode
*ip
= GFS2_I(inode
);
350 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
351 unsigned long last_index
;
352 u64 pos
= page
->index
<< PAGE_CACHE_SHIFT
;
353 unsigned int data_blocks
, ind_blocks
, rblocks
;
354 int alloc_required
= 0;
355 struct gfs2_holder gh
;
356 struct gfs2_alloc
*al
;
359 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
360 ret
= gfs2_glock_nq(&gh
);
364 set_bit(GLF_DIRTY
, &ip
->i_gl
->gl_flags
);
365 set_bit(GIF_SW_PAGED
, &ip
->i_flags
);
367 ret
= gfs2_write_alloc_required(ip
, pos
, PAGE_CACHE_SIZE
, &alloc_required
);
368 if (ret
|| !alloc_required
)
371 al
= gfs2_alloc_get(ip
);
375 ret
= gfs2_quota_lock_check(ip
);
378 gfs2_write_calc_reserv(ip
, PAGE_CACHE_SIZE
, &data_blocks
, &ind_blocks
);
379 al
->al_requested
= data_blocks
+ ind_blocks
;
380 ret
= gfs2_inplace_reserve(ip
);
382 goto out_quota_unlock
;
384 rblocks
= RES_DINODE
+ ind_blocks
;
385 if (gfs2_is_jdata(ip
))
386 rblocks
+= data_blocks
? data_blocks
: 1;
387 if (ind_blocks
|| data_blocks
)
388 rblocks
+= RES_STATFS
+ RES_QUOTA
;
389 ret
= gfs2_trans_begin(sdp
, rblocks
, 0);
395 last_index
= ip
->i_inode
.i_size
>> PAGE_CACHE_SHIFT
;
396 if (page
->index
> last_index
)
397 goto out_unlock_page
;
399 if (!PageUptodate(page
) || page
->mapping
!= ip
->i_inode
.i_mapping
)
400 goto out_unlock_page
;
401 if (gfs2_is_stuffed(ip
)) {
402 ret
= gfs2_unstuff_dinode(ip
, page
);
404 goto out_unlock_page
;
406 ret
= gfs2_allocate_page_backing(page
);
412 gfs2_inplace_release(ip
);
414 gfs2_quota_unlock(ip
);
420 gfs2_holder_uninit(&gh
);
424 ret
= VM_FAULT_SIGBUS
;
428 static const struct vm_operations_struct gfs2_vm_ops
= {
429 .fault
= filemap_fault
,
430 .page_mkwrite
= gfs2_page_mkwrite
,
435 * @file: The file to map
436 * @vma: The VMA which described the mapping
438 * There is no need to get a lock here unless we should be updating
439 * atime. We ignore any locking errors since the only consequence is
440 * a missed atime update (which will just be deferred until later).
445 static int gfs2_mmap(struct file
*file
, struct vm_area_struct
*vma
)
447 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
449 if (!(file
->f_flags
& O_NOATIME
)) {
450 struct gfs2_holder i_gh
;
453 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &i_gh
);
454 error
= gfs2_glock_nq(&i_gh
);
457 gfs2_glock_dq_uninit(&i_gh
);
459 vma
->vm_ops
= &gfs2_vm_ops
;
460 vma
->vm_flags
|= VM_CAN_NONLINEAR
;
466 * gfs2_open - open a file
467 * @inode: the inode to open
468 * @file: the struct file for this opening
473 static int gfs2_open(struct inode
*inode
, struct file
*file
)
475 struct gfs2_inode
*ip
= GFS2_I(inode
);
476 struct gfs2_holder i_gh
;
477 struct gfs2_file
*fp
;
480 fp
= kzalloc(sizeof(struct gfs2_file
), GFP_KERNEL
);
484 mutex_init(&fp
->f_fl_mutex
);
486 gfs2_assert_warn(GFS2_SB(inode
), !file
->private_data
);
487 file
->private_data
= fp
;
489 if (S_ISREG(ip
->i_inode
.i_mode
)) {
490 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
495 if (!(file
->f_flags
& O_LARGEFILE
) &&
496 ip
->i_disksize
> MAX_NON_LFS
) {
501 gfs2_glock_dq_uninit(&i_gh
);
507 gfs2_glock_dq_uninit(&i_gh
);
509 file
->private_data
= NULL
;
515 * gfs2_close - called to close a struct file
516 * @inode: the inode the struct file belongs to
517 * @file: the struct file being closed
522 static int gfs2_close(struct inode
*inode
, struct file
*file
)
524 struct gfs2_sbd
*sdp
= inode
->i_sb
->s_fs_info
;
525 struct gfs2_file
*fp
;
527 fp
= file
->private_data
;
528 file
->private_data
= NULL
;
530 if (gfs2_assert_warn(sdp
, fp
))
539 * gfs2_fsync - sync the dirty data for a file (across the cluster)
540 * @file: the file that points to the dentry (we ignore this)
541 * @dentry: the dentry that points to the inode to sync
543 * The VFS will flush "normal" data for us. We only need to worry
544 * about metadata here. For journaled data, we just do a log flush
545 * as we can't avoid it. Otherwise we can just bale out if datasync
546 * is set. For stuffed inodes we must flush the log in order to
547 * ensure that all data is on disk.
549 * The call to write_inode_now() is there to write back metadata and
550 * the inode itself. It does also try and write the data, but thats
551 * (hopefully) a no-op due to the VFS having already called filemap_fdatawrite()
557 static int gfs2_fsync(struct file
*file
, int datasync
)
559 struct inode
*inode
= file
->f_mapping
->host
;
560 int sync_state
= inode
->i_state
& (I_DIRTY_SYNC
|I_DIRTY_DATASYNC
);
563 if (gfs2_is_jdata(GFS2_I(inode
))) {
564 gfs2_log_flush(GFS2_SB(inode
), GFS2_I(inode
)->i_gl
);
568 if (sync_state
!= 0) {
570 ret
= write_inode_now(inode
, 0);
572 if (gfs2_is_stuffed(GFS2_I(inode
)))
573 gfs2_log_flush(GFS2_SB(inode
), GFS2_I(inode
)->i_gl
);
580 * gfs2_file_aio_write - Perform a write to a file
581 * @iocb: The io context
582 * @iov: The data to write
583 * @nr_segs: Number of @iov segments
584 * @pos: The file position
586 * We have to do a lock/unlock here to refresh the inode size for
587 * O_APPEND writes, otherwise we can land up writing at the wrong
588 * offset. There is still a race, but provided the app is using its
589 * own file locking, this will make O_APPEND work as expected.
593 static ssize_t
gfs2_file_aio_write(struct kiocb
*iocb
, const struct iovec
*iov
,
594 unsigned long nr_segs
, loff_t pos
)
596 struct file
*file
= iocb
->ki_filp
;
598 if (file
->f_flags
& O_APPEND
) {
599 struct dentry
*dentry
= file
->f_dentry
;
600 struct gfs2_inode
*ip
= GFS2_I(dentry
->d_inode
);
601 struct gfs2_holder gh
;
604 ret
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, 0, &gh
);
607 gfs2_glock_dq_uninit(&gh
);
610 return generic_file_aio_write(iocb
, iov
, nr_segs
, pos
);
613 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
616 * gfs2_setlease - acquire/release a file lease
617 * @file: the file pointer
621 * We don't currently have a way to enforce a lease across the whole
622 * cluster; until we do, disable leases (by just returning -EINVAL),
623 * unless the administrator has requested purely local locking.
628 static int gfs2_setlease(struct file
*file
, long arg
, struct file_lock
**fl
)
634 * gfs2_lock - acquire/release a posix lock on a file
635 * @file: the file pointer
636 * @cmd: either modify or retrieve lock state, possibly wait
637 * @fl: type and range of lock
642 static int gfs2_lock(struct file
*file
, int cmd
, struct file_lock
*fl
)
644 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
645 struct gfs2_sbd
*sdp
= GFS2_SB(file
->f_mapping
->host
);
646 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
648 if (!(fl
->fl_flags
& FL_POSIX
))
650 if (__mandatory_lock(&ip
->i_inode
) && fl
->fl_type
!= F_UNLCK
)
653 if (cmd
== F_CANCELLK
) {
656 fl
->fl_type
= F_UNLCK
;
658 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
661 return dlm_posix_get(ls
->ls_dlm
, ip
->i_no_addr
, file
, fl
);
662 else if (fl
->fl_type
== F_UNLCK
)
663 return dlm_posix_unlock(ls
->ls_dlm
, ip
->i_no_addr
, file
, fl
);
665 return dlm_posix_lock(ls
->ls_dlm
, ip
->i_no_addr
, file
, cmd
, fl
);
668 static int do_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
670 struct gfs2_file
*fp
= file
->private_data
;
671 struct gfs2_holder
*fl_gh
= &fp
->f_fl_gh
;
672 struct gfs2_inode
*ip
= GFS2_I(file
->f_path
.dentry
->d_inode
);
673 struct gfs2_glock
*gl
;
678 state
= (fl
->fl_type
== F_WRLCK
) ? LM_ST_EXCLUSIVE
: LM_ST_SHARED
;
679 flags
= (IS_SETLKW(cmd
) ? 0 : LM_FLAG_TRY
) | GL_EXACT
| GL_NOCACHE
;
681 mutex_lock(&fp
->f_fl_mutex
);
685 if (fl_gh
->gh_state
== state
)
687 flock_lock_file_wait(file
,
688 &(struct file_lock
){.fl_type
= F_UNLCK
});
689 gfs2_glock_dq_wait(fl_gh
);
690 gfs2_holder_reinit(state
, flags
, fl_gh
);
692 error
= gfs2_glock_get(GFS2_SB(&ip
->i_inode
), ip
->i_no_addr
,
693 &gfs2_flock_glops
, CREATE
, &gl
);
696 gfs2_holder_init(gl
, state
, flags
, fl_gh
);
699 error
= gfs2_glock_nq(fl_gh
);
701 gfs2_holder_uninit(fl_gh
);
702 if (error
== GLR_TRYFAILED
)
705 error
= flock_lock_file_wait(file
, fl
);
706 gfs2_assert_warn(GFS2_SB(&ip
->i_inode
), !error
);
710 mutex_unlock(&fp
->f_fl_mutex
);
714 static void do_unflock(struct file
*file
, struct file_lock
*fl
)
716 struct gfs2_file
*fp
= file
->private_data
;
717 struct gfs2_holder
*fl_gh
= &fp
->f_fl_gh
;
719 mutex_lock(&fp
->f_fl_mutex
);
720 flock_lock_file_wait(file
, fl
);
722 gfs2_glock_dq_uninit(fl_gh
);
723 mutex_unlock(&fp
->f_fl_mutex
);
727 * gfs2_flock - acquire/release a flock lock on a file
728 * @file: the file pointer
729 * @cmd: either modify or retrieve lock state, possibly wait
730 * @fl: type and range of lock
735 static int gfs2_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
737 if (!(fl
->fl_flags
& FL_FLOCK
))
739 if (fl
->fl_type
& LOCK_MAND
)
742 if (fl
->fl_type
== F_UNLCK
) {
743 do_unflock(file
, fl
);
746 return do_flock(file
, cmd
, fl
);
750 const struct file_operations gfs2_file_fops
= {
751 .llseek
= gfs2_llseek
,
752 .read
= do_sync_read
,
753 .aio_read
= generic_file_aio_read
,
754 .write
= do_sync_write
,
755 .aio_write
= gfs2_file_aio_write
,
756 .unlocked_ioctl
= gfs2_ioctl
,
759 .release
= gfs2_close
,
763 .splice_read
= generic_file_splice_read
,
764 .splice_write
= generic_file_splice_write
,
765 .setlease
= gfs2_setlease
,
768 const struct file_operations gfs2_dir_fops
= {
769 .readdir
= gfs2_readdir
,
770 .unlocked_ioctl
= gfs2_ioctl
,
772 .release
= gfs2_close
,
778 #endif /* CONFIG_GFS2_FS_LOCKING_DLM */
780 const struct file_operations gfs2_file_fops_nolock
= {
781 .llseek
= gfs2_llseek
,
782 .read
= do_sync_read
,
783 .aio_read
= generic_file_aio_read
,
784 .write
= do_sync_write
,
785 .aio_write
= gfs2_file_aio_write
,
786 .unlocked_ioctl
= gfs2_ioctl
,
789 .release
= gfs2_close
,
791 .splice_read
= generic_file_splice_read
,
792 .splice_write
= generic_file_splice_write
,
793 .setlease
= generic_setlease
,
796 const struct file_operations gfs2_dir_fops_nolock
= {
797 .readdir
= gfs2_readdir
,
798 .unlocked_ioctl
= gfs2_ioctl
,
800 .release
= gfs2_close
,