2 * Copyright (C) 2006 Red Hat, Inc. All rights reserved.
4 * This copyrighted material is made available to anyone wishing to use,
5 * modify, copy, or redistribute it subject to the terms and conditions
6 * of the GNU General Public License v.2.
9 #include <linux/miscdevice.h>
10 #include <linux/init.h>
11 #include <linux/wait.h>
12 #include <linux/module.h>
13 #include <linux/file.h>
15 #include <linux/poll.h>
16 #include <linux/signal.h>
17 #include <linux/spinlock.h>
18 #include <linux/dlm.h>
19 #include <linux/dlm_device.h>
21 #include "dlm_internal.h"
22 #include "lockspace.h"
24 #include "lvb_table.h"
26 static const char *name_prefix
="dlm";
27 static struct miscdevice ctl_device
;
28 static struct file_operations device_fops
;
32 struct dlm_lock_params32
{
45 char lvb
[DLM_USER_LVB_LEN
];
49 struct dlm_write_request32
{
56 struct dlm_lock_params32 lock
;
57 struct dlm_lspace_params lspace
;
68 struct dlm_lock_result32
{
73 struct dlm_lksb32 lksb
;
76 /* Offsets may be zero if no data is present */
80 static void compat_input(struct dlm_write_request
*kb
,
81 struct dlm_write_request32
*kb32
)
83 kb
->version
[0] = kb32
->version
[0];
84 kb
->version
[1] = kb32
->version
[1];
85 kb
->version
[2] = kb32
->version
[2];
88 kb
->is64bit
= kb32
->is64bit
;
89 if (kb
->cmd
== DLM_USER_CREATE_LOCKSPACE
||
90 kb
->cmd
== DLM_USER_REMOVE_LOCKSPACE
) {
91 kb
->i
.lspace
.flags
= kb32
->i
.lspace
.flags
;
92 kb
->i
.lspace
.minor
= kb32
->i
.lspace
.minor
;
93 strcpy(kb
->i
.lspace
.name
, kb32
->i
.lspace
.name
);
95 kb
->i
.lock
.mode
= kb32
->i
.lock
.mode
;
96 kb
->i
.lock
.namelen
= kb32
->i
.lock
.namelen
;
97 kb
->i
.lock
.flags
= kb32
->i
.lock
.flags
;
98 kb
->i
.lock
.lkid
= kb32
->i
.lock
.lkid
;
99 kb
->i
.lock
.parent
= kb32
->i
.lock
.parent
;
100 kb
->i
.lock
.castparam
= (void *)(long)kb32
->i
.lock
.castparam
;
101 kb
->i
.lock
.castaddr
= (void *)(long)kb32
->i
.lock
.castaddr
;
102 kb
->i
.lock
.bastparam
= (void *)(long)kb32
->i
.lock
.bastparam
;
103 kb
->i
.lock
.bastaddr
= (void *)(long)kb32
->i
.lock
.bastaddr
;
104 kb
->i
.lock
.lksb
= (void *)(long)kb32
->i
.lock
.lksb
;
105 memcpy(kb
->i
.lock
.lvb
, kb32
->i
.lock
.lvb
, DLM_USER_LVB_LEN
);
106 memcpy(kb
->i
.lock
.name
, kb32
->i
.lock
.name
, kb
->i
.lock
.namelen
);
110 static void compat_output(struct dlm_lock_result
*res
,
111 struct dlm_lock_result32
*res32
)
113 res32
->length
= res
->length
- (sizeof(struct dlm_lock_result
) -
114 sizeof(struct dlm_lock_result32
));
115 res32
->user_astaddr
= (__u32
)(long)res
->user_astaddr
;
116 res32
->user_astparam
= (__u32
)(long)res
->user_astparam
;
117 res32
->user_lksb
= (__u32
)(long)res
->user_lksb
;
118 res32
->bast_mode
= res
->bast_mode
;
120 res32
->lvb_offset
= res
->lvb_offset
;
121 res32
->length
= res
->length
;
123 res32
->lksb
.sb_status
= res
->lksb
.sb_status
;
124 res32
->lksb
.sb_flags
= res
->lksb
.sb_flags
;
125 res32
->lksb
.sb_lkid
= res
->lksb
.sb_lkid
;
126 res32
->lksb
.sb_lvbptr
= (__u32
)(long)res
->lksb
.sb_lvbptr
;
131 void dlm_user_add_ast(struct dlm_lkb
*lkb
, int type
)
134 struct dlm_user_args
*ua
;
135 struct dlm_user_proc
*proc
;
136 int remove_ownqueue
= 0;
138 /* dlm_clear_proc_locks() sets ORPHAN/DEAD flag on each
139 lkb before dealing with it. We need to check this
140 flag before taking ls_clear_proc_locks mutex because if
141 it's set, dlm_clear_proc_locks() holds the mutex. */
143 if (lkb
->lkb_flags
& (DLM_IFL_ORPHAN
| DLM_IFL_DEAD
)) {
144 /* log_print("user_add_ast skip1 %x", lkb->lkb_flags); */
148 ls
= lkb
->lkb_resource
->res_ls
;
149 mutex_lock(&ls
->ls_clear_proc_locks
);
151 /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
152 can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
153 lkb->ua so we can't try to use it. */
155 if (lkb
->lkb_flags
& (DLM_IFL_ORPHAN
| DLM_IFL_DEAD
)) {
156 /* log_print("user_add_ast skip2 %x", lkb->lkb_flags); */
160 DLM_ASSERT(lkb
->lkb_astparam
, dlm_print_lkb(lkb
););
161 ua
= (struct dlm_user_args
*)lkb
->lkb_astparam
;
164 if (type
== AST_BAST
&& ua
->bastaddr
== NULL
)
167 spin_lock(&proc
->asts_spin
);
168 if (!(lkb
->lkb_ast_type
& (AST_COMP
| AST_BAST
))) {
169 kref_get(&lkb
->lkb_ref
);
170 list_add_tail(&lkb
->lkb_astqueue
, &proc
->asts
);
171 lkb
->lkb_ast_type
|= type
;
172 wake_up_interruptible(&proc
->wait
);
175 /* noqueue requests that fail may need to be removed from the
176 proc's locks list, there should be a better way of detecting
177 this situation than checking all these things... */
179 if (type
== AST_COMP
&& lkb
->lkb_grmode
== DLM_LOCK_IV
&&
180 ua
->lksb
.sb_status
== -EAGAIN
&& !list_empty(&lkb
->lkb_ownqueue
))
183 /* We want to copy the lvb to userspace when the completion
184 ast is read if the status is 0, the lock has an lvb and
185 lvb_ops says we should. We could probably have set_lvb_lock()
186 set update_user_lvb instead and not need old_mode */
188 if ((lkb
->lkb_ast_type
& AST_COMP
) &&
189 (lkb
->lkb_lksb
->sb_status
== 0) &&
190 lkb
->lkb_lksb
->sb_lvbptr
&&
191 dlm_lvb_operations
[ua
->old_mode
+ 1][lkb
->lkb_grmode
+ 1])
192 ua
->update_user_lvb
= 1;
194 ua
->update_user_lvb
= 0;
196 spin_unlock(&proc
->asts_spin
);
198 if (remove_ownqueue
) {
199 spin_lock(&ua
->proc
->locks_spin
);
200 list_del_init(&lkb
->lkb_ownqueue
);
201 spin_unlock(&ua
->proc
->locks_spin
);
205 mutex_unlock(&ls
->ls_clear_proc_locks
);
208 static int device_user_lock(struct dlm_user_proc
*proc
,
209 struct dlm_lock_params
*params
)
212 struct dlm_user_args
*ua
;
215 ls
= dlm_find_lockspace_local(proc
->lockspace
);
219 if (!params
->castaddr
|| !params
->lksb
) {
224 ua
= kzalloc(sizeof(struct dlm_user_args
), GFP_KERNEL
);
228 ua
->user_lksb
= params
->lksb
;
229 ua
->castparam
= params
->castparam
;
230 ua
->castaddr
= params
->castaddr
;
231 ua
->bastparam
= params
->bastparam
;
232 ua
->bastaddr
= params
->bastaddr
;
234 if (params
->flags
& DLM_LKF_CONVERT
)
235 error
= dlm_user_convert(ls
, ua
,
236 params
->mode
, params
->flags
,
237 params
->lkid
, params
->lvb
);
239 error
= dlm_user_request(ls
, ua
,
240 params
->mode
, params
->flags
,
241 params
->name
, params
->namelen
,
244 error
= ua
->lksb
.sb_lkid
;
247 dlm_put_lockspace(ls
);
251 static int device_user_unlock(struct dlm_user_proc
*proc
,
252 struct dlm_lock_params
*params
)
255 struct dlm_user_args
*ua
;
258 ls
= dlm_find_lockspace_local(proc
->lockspace
);
262 ua
= kzalloc(sizeof(struct dlm_user_args
), GFP_KERNEL
);
266 ua
->user_lksb
= params
->lksb
;
267 ua
->castparam
= params
->castparam
;
268 ua
->castaddr
= params
->castaddr
;
270 if (params
->flags
& DLM_LKF_CANCEL
)
271 error
= dlm_user_cancel(ls
, ua
, params
->flags
, params
->lkid
);
273 error
= dlm_user_unlock(ls
, ua
, params
->flags
, params
->lkid
,
276 dlm_put_lockspace(ls
);
280 static int device_create_lockspace(struct dlm_lspace_params
*params
)
282 dlm_lockspace_t
*lockspace
;
286 if (!capable(CAP_SYS_ADMIN
))
289 error
= dlm_new_lockspace(params
->name
, strlen(params
->name
),
290 &lockspace
, 0, DLM_USER_LVB_LEN
);
294 ls
= dlm_find_lockspace_local(lockspace
);
299 len
= strlen(params
->name
) + strlen(name_prefix
) + 2;
300 ls
->ls_device
.name
= kzalloc(len
, GFP_KERNEL
);
301 if (!ls
->ls_device
.name
)
303 snprintf((char *)ls
->ls_device
.name
, len
, "%s_%s", name_prefix
,
305 ls
->ls_device
.fops
= &device_fops
;
306 ls
->ls_device
.minor
= MISC_DYNAMIC_MINOR
;
308 error
= misc_register(&ls
->ls_device
);
310 kfree(ls
->ls_device
.name
);
314 error
= ls
->ls_device
.minor
;
315 dlm_put_lockspace(ls
);
319 dlm_put_lockspace(ls
);
320 dlm_release_lockspace(lockspace
, 0);
324 static int device_remove_lockspace(struct dlm_lspace_params
*params
)
326 dlm_lockspace_t
*lockspace
;
328 int error
, force
= 0;
330 if (!capable(CAP_SYS_ADMIN
))
333 ls
= dlm_find_lockspace_device(params
->minor
);
337 error
= misc_deregister(&ls
->ls_device
);
339 dlm_put_lockspace(ls
);
342 kfree(ls
->ls_device
.name
);
344 if (params
->flags
& DLM_USER_LSFLG_FORCEFREE
)
347 lockspace
= ls
->ls_local_handle
;
349 /* dlm_release_lockspace waits for references to go to zero,
350 so all processes will need to close their device for the ls
351 before the release will procede */
353 dlm_put_lockspace(ls
);
354 error
= dlm_release_lockspace(lockspace
, force
);
359 /* Check the user's version matches ours */
360 static int check_version(struct dlm_write_request
*req
)
362 if (req
->version
[0] != DLM_DEVICE_VERSION_MAJOR
||
363 (req
->version
[0] == DLM_DEVICE_VERSION_MAJOR
&&
364 req
->version
[1] > DLM_DEVICE_VERSION_MINOR
)) {
366 printk(KERN_DEBUG
"dlm: process %s (%d) version mismatch "
367 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
373 DLM_DEVICE_VERSION_MAJOR
,
374 DLM_DEVICE_VERSION_MINOR
,
375 DLM_DEVICE_VERSION_PATCH
);
385 * dlm_user_request -> request_lock
386 * dlm_user_convert -> convert_lock
389 * dlm_user_unlock -> unlock_lock
390 * dlm_user_cancel -> cancel_lock
392 * device_create_lockspace
395 * device_remove_lockspace
396 * dlm_release_lockspace
399 /* a write to a lockspace device is a lock or unlock request, a write
400 to the control device is to create/remove a lockspace */
402 static ssize_t
device_write(struct file
*file
, const char __user
*buf
,
403 size_t count
, loff_t
*ppos
)
405 struct dlm_user_proc
*proc
= file
->private_data
;
406 struct dlm_write_request
*kbuf
;
407 sigset_t tmpsig
, allsigs
;
411 if (count
< sizeof(struct dlm_write_request32
))
413 if (count
< sizeof(struct dlm_write_request
))
417 kbuf
= kmalloc(count
, GFP_KERNEL
);
421 if (copy_from_user(kbuf
, buf
, count
)) {
426 if (check_version(kbuf
)) {
432 if (!kbuf
->is64bit
) {
433 struct dlm_write_request32
*k32buf
;
434 k32buf
= (struct dlm_write_request32
*)kbuf
;
435 kbuf
= kmalloc(count
+ (sizeof(struct dlm_write_request
) -
436 sizeof(struct dlm_write_request32
)), GFP_KERNEL
);
441 set_bit(DLM_PROC_FLAGS_COMPAT
, &proc
->flags
);
442 compat_input(kbuf
, k32buf
);
447 /* do we really need this? can a write happen after a close? */
448 if ((kbuf
->cmd
== DLM_USER_LOCK
|| kbuf
->cmd
== DLM_USER_UNLOCK
) &&
449 test_bit(DLM_PROC_FLAGS_CLOSING
, &proc
->flags
))
452 sigfillset(&allsigs
);
453 sigprocmask(SIG_BLOCK
, &allsigs
, &tmpsig
);
461 log_print("no locking on control device");
464 error
= device_user_lock(proc
, &kbuf
->i
.lock
);
467 case DLM_USER_UNLOCK
:
469 log_print("no locking on control device");
472 error
= device_user_unlock(proc
, &kbuf
->i
.lock
);
475 case DLM_USER_CREATE_LOCKSPACE
:
477 log_print("create/remove only on control device");
480 error
= device_create_lockspace(&kbuf
->i
.lspace
);
483 case DLM_USER_REMOVE_LOCKSPACE
:
485 log_print("create/remove only on control device");
488 error
= device_remove_lockspace(&kbuf
->i
.lspace
);
492 log_print("Unknown command passed to DLM device : %d\n",
497 sigprocmask(SIG_SETMASK
, &tmpsig
, NULL
);
504 /* Every process that opens the lockspace device has its own "proc" structure
505 hanging off the open file that's used to keep track of locks owned by the
506 process and asts that need to be delivered to the process. */
508 static int device_open(struct inode
*inode
, struct file
*file
)
510 struct dlm_user_proc
*proc
;
513 ls
= dlm_find_lockspace_device(iminor(inode
));
517 proc
= kzalloc(sizeof(struct dlm_user_proc
), GFP_KERNEL
);
519 dlm_put_lockspace(ls
);
523 proc
->lockspace
= ls
->ls_local_handle
;
524 INIT_LIST_HEAD(&proc
->asts
);
525 INIT_LIST_HEAD(&proc
->locks
);
526 spin_lock_init(&proc
->asts_spin
);
527 spin_lock_init(&proc
->locks_spin
);
528 init_waitqueue_head(&proc
->wait
);
529 file
->private_data
= proc
;
534 static int device_close(struct inode
*inode
, struct file
*file
)
536 struct dlm_user_proc
*proc
= file
->private_data
;
538 sigset_t tmpsig
, allsigs
;
540 ls
= dlm_find_lockspace_local(proc
->lockspace
);
544 sigfillset(&allsigs
);
545 sigprocmask(SIG_BLOCK
, &allsigs
, &tmpsig
);
547 set_bit(DLM_PROC_FLAGS_CLOSING
, &proc
->flags
);
549 dlm_clear_proc_locks(ls
, proc
);
551 /* at this point no more lkb's should exist for this lockspace,
552 so there's no chance of dlm_user_add_ast() being called and
553 looking for lkb->ua->proc */
556 file
->private_data
= NULL
;
558 dlm_put_lockspace(ls
);
559 dlm_put_lockspace(ls
); /* for the find in device_open() */
561 /* FIXME: AUTOFREE: if this ls is no longer used do
562 device_remove_lockspace() */
564 sigprocmask(SIG_SETMASK
, &tmpsig
, NULL
);
570 static int copy_result_to_user(struct dlm_user_args
*ua
, int compat
, int type
,
571 int bmode
, char __user
*buf
, size_t count
)
574 struct dlm_lock_result32 result32
;
576 struct dlm_lock_result result
;
582 memset(&result
, 0, sizeof(struct dlm_lock_result
));
583 memcpy(&result
.lksb
, &ua
->lksb
, sizeof(struct dlm_lksb
));
584 result
.user_lksb
= ua
->user_lksb
;
586 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
587 in a conversion unless the conversion is successful. See code
588 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
589 notes that a new blocking AST address and parameter are set even if
590 the conversion fails, so maybe we should just do that. */
592 if (type
== AST_BAST
) {
593 result
.user_astaddr
= ua
->bastaddr
;
594 result
.user_astparam
= ua
->bastparam
;
595 result
.bast_mode
= bmode
;
597 result
.user_astaddr
= ua
->castaddr
;
598 result
.user_astparam
= ua
->castparam
;
603 len
= sizeof(struct dlm_lock_result32
);
606 len
= sizeof(struct dlm_lock_result
);
609 /* copy lvb to userspace if there is one, it's been updated, and
610 the user buffer has space for it */
612 if (ua
->update_user_lvb
&& ua
->lksb
.sb_lvbptr
&&
613 count
>= len
+ DLM_USER_LVB_LEN
) {
614 if (copy_to_user(buf
+len
, ua
->lksb
.sb_lvbptr
,
620 result
.lvb_offset
= len
;
621 len
+= DLM_USER_LVB_LEN
;
628 compat_output(&result
, &result32
);
629 resultptr
= &result32
;
633 if (copy_to_user(buf
, resultptr
, struct_len
))
641 /* a read returns a single ast described in a struct dlm_lock_result */
643 static ssize_t
device_read(struct file
*file
, char __user
*buf
, size_t count
,
646 struct dlm_user_proc
*proc
= file
->private_data
;
648 struct dlm_user_args
*ua
;
649 DECLARE_WAITQUEUE(wait
, current
);
650 int error
, type
=0, bmode
=0, removed
= 0;
653 if (count
< sizeof(struct dlm_lock_result32
))
655 if (count
< sizeof(struct dlm_lock_result
))
659 /* do we really need this? can a read happen after a close? */
660 if (test_bit(DLM_PROC_FLAGS_CLOSING
, &proc
->flags
))
663 spin_lock(&proc
->asts_spin
);
664 if (list_empty(&proc
->asts
)) {
665 if (file
->f_flags
& O_NONBLOCK
) {
666 spin_unlock(&proc
->asts_spin
);
670 add_wait_queue(&proc
->wait
, &wait
);
673 set_current_state(TASK_INTERRUPTIBLE
);
674 if (list_empty(&proc
->asts
) && !signal_pending(current
)) {
675 spin_unlock(&proc
->asts_spin
);
677 spin_lock(&proc
->asts_spin
);
680 set_current_state(TASK_RUNNING
);
681 remove_wait_queue(&proc
->wait
, &wait
);
683 if (signal_pending(current
)) {
684 spin_unlock(&proc
->asts_spin
);
689 if (list_empty(&proc
->asts
)) {
690 spin_unlock(&proc
->asts_spin
);
694 /* there may be both completion and blocking asts to return for
695 the lkb, don't remove lkb from asts list unless no asts remain */
697 lkb
= list_entry(proc
->asts
.next
, struct dlm_lkb
, lkb_astqueue
);
699 if (lkb
->lkb_ast_type
& AST_COMP
) {
700 lkb
->lkb_ast_type
&= ~AST_COMP
;
702 } else if (lkb
->lkb_ast_type
& AST_BAST
) {
703 lkb
->lkb_ast_type
&= ~AST_BAST
;
705 bmode
= lkb
->lkb_bastmode
;
708 if (!lkb
->lkb_ast_type
) {
709 list_del(&lkb
->lkb_astqueue
);
712 spin_unlock(&proc
->asts_spin
);
714 ua
= (struct dlm_user_args
*)lkb
->lkb_astparam
;
715 error
= copy_result_to_user(ua
,
716 test_bit(DLM_PROC_FLAGS_COMPAT
, &proc
->flags
),
717 type
, bmode
, buf
, count
);
719 /* removes reference for the proc->asts lists added by
720 dlm_user_add_ast() and may result in the lkb being freed */
727 static unsigned int device_poll(struct file
*file
, poll_table
*wait
)
729 struct dlm_user_proc
*proc
= file
->private_data
;
731 poll_wait(file
, &proc
->wait
, wait
);
733 spin_lock(&proc
->asts_spin
);
734 if (!list_empty(&proc
->asts
)) {
735 spin_unlock(&proc
->asts_spin
);
736 return POLLIN
| POLLRDNORM
;
738 spin_unlock(&proc
->asts_spin
);
742 static int ctl_device_open(struct inode
*inode
, struct file
*file
)
744 file
->private_data
= NULL
;
748 static int ctl_device_close(struct inode
*inode
, struct file
*file
)
753 static struct file_operations device_fops
= {
755 .release
= device_close
,
757 .write
= device_write
,
759 .owner
= THIS_MODULE
,
762 static struct file_operations ctl_device_fops
= {
763 .open
= ctl_device_open
,
764 .release
= ctl_device_close
,
765 .write
= device_write
,
766 .owner
= THIS_MODULE
,
769 int dlm_user_init(void)
773 ctl_device
.name
= "dlm-control";
774 ctl_device
.fops
= &ctl_device_fops
;
775 ctl_device
.minor
= MISC_DYNAMIC_MINOR
;
777 error
= misc_register(&ctl_device
);
779 log_print("misc_register failed for control device");
784 void dlm_user_exit(void)
786 misc_deregister(&ctl_device
);