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/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/pagemap.h>
16 #include <linux/uio.h>
17 #include <linux/blkdev.h>
19 #include <linux/smp_lock.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/ext2_fs.h>
23 #include <linux/crc32.h>
24 #include <linux/lm_interface.h>
25 #include <linux/writeback.h>
26 #include <asm/uaccess.h>
47 * Most fields left uninitialised to catch anybody who tries to
48 * use them. f_flags set to prevent file_accessed() from touching
49 * any other part of this. Its use is purely as a flag so that we
50 * know (in readpage()) whether or not do to locking.
52 struct file gfs2_internal_file_sentinel
= {
53 .f_flags
= O_NOATIME
|O_RDONLY
,
56 static int gfs2_read_actor(read_descriptor_t
*desc
, struct page
*page
,
57 unsigned long offset
, unsigned long size
)
60 unsigned long count
= desc
->count
;
66 memcpy(desc
->arg
.data
, kaddr
+ offset
, size
);
69 desc
->count
= count
- size
;
70 desc
->written
+= size
;
71 desc
->arg
.buf
+= size
;
75 int gfs2_internal_read(struct gfs2_inode
*ip
, struct file_ra_state
*ra_state
,
76 char *buf
, loff_t
*pos
, unsigned size
)
78 struct inode
*inode
= &ip
->i_inode
;
79 read_descriptor_t desc
;
84 do_generic_mapping_read(inode
->i_mapping
, ra_state
,
85 &gfs2_internal_file_sentinel
, pos
, &desc
,
87 return desc
.written
? desc
.written
: desc
.error
;
91 * gfs2_llseek - seek to a location in a file
94 * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END)
96 * SEEK_END requires the glock for the file because it references the
99 * Returns: The new offset, or errno
102 static loff_t
gfs2_llseek(struct file
*file
, loff_t offset
, int origin
)
104 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
105 struct gfs2_holder i_gh
;
109 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
112 error
= remote_llseek(file
, offset
, origin
);
113 gfs2_glock_dq_uninit(&i_gh
);
116 error
= remote_llseek(file
, offset
, origin
);
122 * gfs2_readdir - Read directory entries from a directory
123 * @file: The directory to read from
124 * @dirent: Buffer for dirents
125 * @filldir: Function used to do the copying
130 static int gfs2_readdir(struct file
*file
, void *dirent
, filldir_t filldir
)
132 struct inode
*dir
= file
->f_mapping
->host
;
133 struct gfs2_inode
*dip
= GFS2_I(dir
);
134 struct gfs2_holder d_gh
;
135 u64 offset
= file
->f_pos
;
138 gfs2_holder_init(dip
->i_gl
, LM_ST_SHARED
, GL_ATIME
, &d_gh
);
139 error
= gfs2_glock_nq_atime(&d_gh
);
141 gfs2_holder_uninit(&d_gh
);
145 error
= gfs2_dir_read(dir
, &offset
, dirent
, filldir
);
147 gfs2_glock_dq_uninit(&d_gh
);
149 file
->f_pos
= offset
;
156 * @table: A table of 32 u32 flags
157 * @val: a 32 bit value to convert
159 * This function can be used to convert between fsflags values and
160 * GFS2's own flags values.
162 * Returns: the converted flags
164 static u32
fsflags_cvt(const u32
*table
, u32 val
)
176 static const u32 fsflags_to_gfs2
[32] = {
178 [4] = GFS2_DIF_IMMUTABLE
,
179 [5] = GFS2_DIF_APPENDONLY
,
180 [7] = GFS2_DIF_NOATIME
,
181 [12] = GFS2_DIF_EXHASH
,
182 [14] = GFS2_DIF_JDATA
,
183 [20] = GFS2_DIF_DIRECTIO
,
186 static const u32 gfs2_to_fsflags
[32] = {
187 [gfs2fl_Sync
] = FS_SYNC_FL
,
188 [gfs2fl_Immutable
] = FS_IMMUTABLE_FL
,
189 [gfs2fl_AppendOnly
] = FS_APPEND_FL
,
190 [gfs2fl_NoAtime
] = FS_NOATIME_FL
,
191 [gfs2fl_ExHash
] = FS_INDEX_FL
,
192 [gfs2fl_Jdata
] = FS_JOURNAL_DATA_FL
,
193 [gfs2fl_Directio
] = FS_DIRECTIO_FL
,
194 [gfs2fl_InheritDirectio
] = FS_DIRECTIO_FL
,
195 [gfs2fl_InheritJdata
] = FS_JOURNAL_DATA_FL
,
198 static int gfs2_get_flags(struct file
*filp
, u32 __user
*ptr
)
200 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
201 struct gfs2_inode
*ip
= GFS2_I(inode
);
202 struct gfs2_holder gh
;
206 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, GL_ATIME
, &gh
);
207 error
= gfs2_glock_nq_atime(&gh
);
211 fsflags
= fsflags_cvt(gfs2_to_fsflags
, ip
->i_di
.di_flags
);
212 if (put_user(fsflags
, ptr
))
215 gfs2_glock_dq_m(1, &gh
);
216 gfs2_holder_uninit(&gh
);
220 void gfs2_set_inode_flags(struct inode
*inode
)
222 struct gfs2_inode
*ip
= GFS2_I(inode
);
223 struct gfs2_dinode_host
*di
= &ip
->i_di
;
224 unsigned int flags
= inode
->i_flags
;
226 flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
227 if (di
->di_flags
& GFS2_DIF_IMMUTABLE
)
228 flags
|= S_IMMUTABLE
;
229 if (di
->di_flags
& GFS2_DIF_APPENDONLY
)
231 if (di
->di_flags
& GFS2_DIF_NOATIME
)
233 if (di
->di_flags
& GFS2_DIF_SYNC
)
235 inode
->i_flags
= flags
;
238 /* Flags that can be set by user space */
239 #define GFS2_FLAGS_USER_SET (GFS2_DIF_JDATA| \
241 GFS2_DIF_IMMUTABLE| \
242 GFS2_DIF_APPENDONLY| \
246 GFS2_DIF_INHERIT_DIRECTIO| \
247 GFS2_DIF_INHERIT_JDATA)
250 * gfs2_set_flags - set flags on an inode
252 * @flags: The flags to set
253 * @mask: Indicates which flags are valid
256 static int do_gfs2_set_flags(struct file
*filp
, u32 reqflags
, u32 mask
)
258 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
259 struct gfs2_inode
*ip
= GFS2_I(inode
);
260 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
261 struct buffer_head
*bh
;
262 struct gfs2_holder gh
;
264 u32 new_flags
, flags
;
266 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0, &gh
);
270 flags
= ip
->i_di
.di_flags
;
271 new_flags
= (flags
& ~mask
) | (reqflags
& mask
);
272 if ((new_flags
^ flags
) == 0)
275 if (S_ISDIR(inode
->i_mode
)) {
276 if ((new_flags
^ flags
) & GFS2_DIF_JDATA
)
277 new_flags
^= (GFS2_DIF_JDATA
|GFS2_DIF_INHERIT_JDATA
);
278 if ((new_flags
^ flags
) & GFS2_DIF_DIRECTIO
)
279 new_flags
^= (GFS2_DIF_DIRECTIO
|GFS2_DIF_INHERIT_DIRECTIO
);
283 if ((new_flags
^ flags
) & ~GFS2_FLAGS_USER_SET
)
287 if (IS_IMMUTABLE(inode
) && (new_flags
& GFS2_DIF_IMMUTABLE
))
289 if (IS_APPEND(inode
) && (new_flags
& GFS2_DIF_APPENDONLY
))
291 if (((new_flags
^ flags
) & GFS2_DIF_IMMUTABLE
) &&
292 !capable(CAP_LINUX_IMMUTABLE
))
294 if (!IS_IMMUTABLE(inode
)) {
295 error
= permission(inode
, MAY_WRITE
, NULL
);
300 error
= gfs2_trans_begin(sdp
, RES_DINODE
, 0);
303 error
= gfs2_meta_inode_buffer(ip
, &bh
);
306 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
307 ip
->i_di
.di_flags
= new_flags
;
308 gfs2_dinode_out(ip
, bh
->b_data
);
310 gfs2_set_inode_flags(inode
);
314 gfs2_glock_dq_uninit(&gh
);
318 static int gfs2_set_flags(struct file
*filp
, u32 __user
*ptr
)
320 u32 fsflags
, gfsflags
;
321 if (get_user(fsflags
, ptr
))
323 gfsflags
= fsflags_cvt(fsflags_to_gfs2
, fsflags
);
324 return do_gfs2_set_flags(filp
, gfsflags
, ~0);
327 static long gfs2_ioctl(struct file
*filp
, unsigned int cmd
, unsigned long arg
)
330 case FS_IOC_GETFLAGS
:
331 return gfs2_get_flags(filp
, (u32 __user
*)arg
);
332 case FS_IOC_SETFLAGS
:
333 return gfs2_set_flags(filp
, (u32 __user
*)arg
);
341 * @file: The file to map
342 * @vma: The VMA which described the mapping
344 * Returns: 0 or error code
347 static int gfs2_mmap(struct file
*file
, struct vm_area_struct
*vma
)
349 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
350 struct gfs2_holder i_gh
;
353 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, GL_ATIME
, &i_gh
);
354 error
= gfs2_glock_nq_atime(&i_gh
);
356 gfs2_holder_uninit(&i_gh
);
360 /* This is VM_MAYWRITE instead of VM_WRITE because a call
361 to mprotect() can turn on VM_WRITE later. */
363 if ((vma
->vm_flags
& (VM_MAYSHARE
| VM_MAYWRITE
)) ==
364 (VM_MAYSHARE
| VM_MAYWRITE
))
365 vma
->vm_ops
= &gfs2_vm_ops_sharewrite
;
367 vma
->vm_ops
= &gfs2_vm_ops_private
;
369 gfs2_glock_dq_uninit(&i_gh
);
375 * gfs2_open - open a file
376 * @inode: the inode to open
377 * @file: the struct file for this opening
382 static int gfs2_open(struct inode
*inode
, struct file
*file
)
384 struct gfs2_inode
*ip
= GFS2_I(inode
);
385 struct gfs2_holder i_gh
;
386 struct gfs2_file
*fp
;
389 fp
= kzalloc(sizeof(struct gfs2_file
), GFP_KERNEL
);
393 mutex_init(&fp
->f_fl_mutex
);
395 gfs2_assert_warn(GFS2_SB(inode
), !file
->private_data
);
396 file
->private_data
= fp
;
398 if (S_ISREG(ip
->i_inode
.i_mode
)) {
399 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
,
404 if (!(file
->f_flags
& O_LARGEFILE
) &&
405 ip
->i_di
.di_size
> MAX_NON_LFS
) {
410 /* Listen to the Direct I/O flag */
412 if (ip
->i_di
.di_flags
& GFS2_DIF_DIRECTIO
)
413 file
->f_flags
|= O_DIRECT
;
415 gfs2_glock_dq_uninit(&i_gh
);
421 gfs2_glock_dq_uninit(&i_gh
);
423 file
->private_data
= NULL
;
429 * gfs2_close - called to close a struct file
430 * @inode: the inode the struct file belongs to
431 * @file: the struct file being closed
436 static int gfs2_close(struct inode
*inode
, struct file
*file
)
438 struct gfs2_sbd
*sdp
= inode
->i_sb
->s_fs_info
;
439 struct gfs2_file
*fp
;
441 fp
= file
->private_data
;
442 file
->private_data
= NULL
;
444 if (gfs2_assert_warn(sdp
, fp
))
453 * gfs2_fsync - sync the dirty data for a file (across the cluster)
454 * @file: the file that points to the dentry (we ignore this)
455 * @dentry: the dentry that points to the inode to sync
457 * The VFS will flush "normal" data for us. We only need to worry
458 * about metadata here. For journaled data, we just do a log flush
459 * as we can't avoid it. Otherwise we can just bale out if datasync
460 * is set. For stuffed inodes we must flush the log in order to
461 * ensure that all data is on disk.
463 * The call to write_inode_now() is there to write back metadata and
464 * the inode itself. It does also try and write the data, but thats
465 * (hopefully) a no-op due to the VFS having already called filemap_fdatawrite()
471 static int gfs2_fsync(struct file
*file
, struct dentry
*dentry
, int datasync
)
473 struct inode
*inode
= dentry
->d_inode
;
474 int sync_state
= inode
->i_state
& (I_DIRTY_SYNC
|I_DIRTY_DATASYNC
);
477 if (gfs2_is_jdata(GFS2_I(inode
))) {
478 gfs2_log_flush(GFS2_SB(inode
), GFS2_I(inode
)->i_gl
);
482 if (sync_state
!= 0) {
484 ret
= write_inode_now(inode
, 0);
486 if (gfs2_is_stuffed(GFS2_I(inode
)))
487 gfs2_log_flush(GFS2_SB(inode
), GFS2_I(inode
)->i_gl
);
494 * gfs2_lock - acquire/release a posix lock on a file
495 * @file: the file pointer
496 * @cmd: either modify or retrieve lock state, possibly wait
497 * @fl: type and range of lock
502 static int gfs2_lock(struct file
*file
, int cmd
, struct file_lock
*fl
)
504 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
505 struct gfs2_sbd
*sdp
= GFS2_SB(file
->f_mapping
->host
);
506 struct lm_lockname name
=
507 { .ln_number
= ip
->i_num
.no_addr
,
508 .ln_type
= LM_TYPE_PLOCK
};
510 if (!(fl
->fl_flags
& FL_POSIX
))
512 if ((ip
->i_inode
.i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
)
515 if (sdp
->sd_args
.ar_localflocks
) {
517 struct file_lock tmp
;
519 ret
= posix_test_lock(file
, fl
, &tmp
);
520 fl
->fl_type
= F_UNLCK
;
522 memcpy(fl
, &tmp
, sizeof(struct file_lock
));
525 return posix_lock_file_wait(file
, fl
);
530 return gfs2_lm_plock_get(sdp
, &name
, file
, fl
);
531 else if (fl
->fl_type
== F_UNLCK
)
532 return gfs2_lm_punlock(sdp
, &name
, file
, fl
);
534 return gfs2_lm_plock(sdp
, &name
, file
, cmd
, fl
);
537 static int do_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
539 struct gfs2_file
*fp
= file
->private_data
;
540 struct gfs2_holder
*fl_gh
= &fp
->f_fl_gh
;
541 struct gfs2_inode
*ip
= GFS2_I(file
->f_path
.dentry
->d_inode
);
542 struct gfs2_glock
*gl
;
547 state
= (fl
->fl_type
== F_WRLCK
) ? LM_ST_EXCLUSIVE
: LM_ST_SHARED
;
548 flags
= (IS_SETLKW(cmd
) ? 0 : LM_FLAG_TRY
) | GL_EXACT
| GL_NOCACHE
;
550 mutex_lock(&fp
->f_fl_mutex
);
554 if (fl_gh
->gh_state
== state
)
557 flock_lock_file_wait(file
,
558 &(struct file_lock
){.fl_type
= F_UNLCK
});
559 gfs2_glock_dq_uninit(fl_gh
);
561 error
= gfs2_glock_get(GFS2_SB(&ip
->i_inode
),
562 ip
->i_num
.no_addr
, &gfs2_flock_glops
,
568 gfs2_holder_init(gl
, state
, flags
, fl_gh
);
571 error
= gfs2_glock_nq(fl_gh
);
573 gfs2_holder_uninit(fl_gh
);
574 if (error
== GLR_TRYFAILED
)
577 error
= flock_lock_file_wait(file
, fl
);
578 gfs2_assert_warn(GFS2_SB(&ip
->i_inode
), !error
);
582 mutex_unlock(&fp
->f_fl_mutex
);
586 static void do_unflock(struct file
*file
, struct file_lock
*fl
)
588 struct gfs2_file
*fp
= file
->private_data
;
589 struct gfs2_holder
*fl_gh
= &fp
->f_fl_gh
;
591 mutex_lock(&fp
->f_fl_mutex
);
592 flock_lock_file_wait(file
, fl
);
594 gfs2_glock_dq_uninit(fl_gh
);
595 mutex_unlock(&fp
->f_fl_mutex
);
599 * gfs2_flock - acquire/release a flock lock on a file
600 * @file: the file pointer
601 * @cmd: either modify or retrieve lock state, possibly wait
602 * @fl: type and range of lock
607 static int gfs2_flock(struct file
*file
, int cmd
, struct file_lock
*fl
)
609 struct gfs2_inode
*ip
= GFS2_I(file
->f_mapping
->host
);
610 struct gfs2_sbd
*sdp
= GFS2_SB(file
->f_mapping
->host
);
612 if (!(fl
->fl_flags
& FL_FLOCK
))
614 if ((ip
->i_inode
.i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
)
617 if (sdp
->sd_args
.ar_localflocks
)
618 return flock_lock_file_wait(file
, fl
);
620 if (fl
->fl_type
== F_UNLCK
) {
621 do_unflock(file
, fl
);
624 return do_flock(file
, cmd
, fl
);
628 const struct file_operations gfs2_file_fops
= {
629 .llseek
= gfs2_llseek
,
630 .read
= do_sync_read
,
631 .aio_read
= generic_file_aio_read
,
632 .write
= do_sync_write
,
633 .aio_write
= generic_file_aio_write
,
634 .unlocked_ioctl
= gfs2_ioctl
,
637 .release
= gfs2_close
,
640 .sendfile
= generic_file_sendfile
,
642 .splice_read
= generic_file_splice_read
,
643 .splice_write
= generic_file_splice_write
,
646 const struct file_operations gfs2_dir_fops
= {
647 .readdir
= gfs2_readdir
,
648 .unlocked_ioctl
= gfs2_ioctl
,
650 .release
= gfs2_close
,