xref: /linux/fs/dlm/user.c (revision 8fa5723aa7e053d498336b48448b292fc2e0458b)
1 /*
2  * Copyright (C) 2006-2008 Red Hat, Inc.  All rights reserved.
3  *
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.
7  */
8 
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>
14 #include <linux/fs.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>
20 
21 #include "dlm_internal.h"
22 #include "lockspace.h"
23 #include "lock.h"
24 #include "lvb_table.h"
25 #include "user.h"
26 
27 static const char name_prefix[] = "dlm";
28 static const struct file_operations device_fops;
29 static atomic_t dlm_monitor_opened;
30 static int dlm_monitor_unused = 1;
31 
32 #ifdef CONFIG_COMPAT
33 
34 struct dlm_lock_params32 {
35 	__u8 mode;
36 	__u8 namelen;
37 	__u16 unused;
38 	__u32 flags;
39 	__u32 lkid;
40 	__u32 parent;
41 	__u64 xid;
42 	__u64 timeout;
43 	__u32 castparam;
44 	__u32 castaddr;
45 	__u32 bastparam;
46 	__u32 bastaddr;
47 	__u32 lksb;
48 	char lvb[DLM_USER_LVB_LEN];
49 	char name[0];
50 };
51 
52 struct dlm_write_request32 {
53 	__u32 version[3];
54 	__u8 cmd;
55 	__u8 is64bit;
56 	__u8 unused[2];
57 
58 	union  {
59 		struct dlm_lock_params32 lock;
60 		struct dlm_lspace_params lspace;
61 		struct dlm_purge_params purge;
62 	} i;
63 };
64 
65 struct dlm_lksb32 {
66 	__u32 sb_status;
67 	__u32 sb_lkid;
68 	__u8 sb_flags;
69 	__u32 sb_lvbptr;
70 };
71 
72 struct dlm_lock_result32 {
73 	__u32 version[3];
74 	__u32 length;
75 	__u32 user_astaddr;
76 	__u32 user_astparam;
77 	__u32 user_lksb;
78 	struct dlm_lksb32 lksb;
79 	__u8 bast_mode;
80 	__u8 unused[3];
81 	/* Offsets may be zero if no data is present */
82 	__u32 lvb_offset;
83 };
84 
85 static void compat_input(struct dlm_write_request *kb,
86 			 struct dlm_write_request32 *kb32,
87 			 size_t count)
88 {
89 	kb->version[0] = kb32->version[0];
90 	kb->version[1] = kb32->version[1];
91 	kb->version[2] = kb32->version[2];
92 
93 	kb->cmd = kb32->cmd;
94 	kb->is64bit = kb32->is64bit;
95 	if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
96 	    kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
97 		kb->i.lspace.flags = kb32->i.lspace.flags;
98 		kb->i.lspace.minor = kb32->i.lspace.minor;
99 		memcpy(kb->i.lspace.name, kb32->i.lspace.name, count -
100 			offsetof(struct dlm_write_request32, i.lspace.name));
101 	} else if (kb->cmd == DLM_USER_PURGE) {
102 		kb->i.purge.nodeid = kb32->i.purge.nodeid;
103 		kb->i.purge.pid = kb32->i.purge.pid;
104 	} else {
105 		kb->i.lock.mode = kb32->i.lock.mode;
106 		kb->i.lock.namelen = kb32->i.lock.namelen;
107 		kb->i.lock.flags = kb32->i.lock.flags;
108 		kb->i.lock.lkid = kb32->i.lock.lkid;
109 		kb->i.lock.parent = kb32->i.lock.parent;
110 		kb->i.lock.xid = kb32->i.lock.xid;
111 		kb->i.lock.timeout = kb32->i.lock.timeout;
112 		kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
113 		kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
114 		kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
115 		kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
116 		kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
117 		memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
118 		memcpy(kb->i.lock.name, kb32->i.lock.name, count -
119 			offsetof(struct dlm_write_request32, i.lock.name));
120 	}
121 }
122 
123 static void compat_output(struct dlm_lock_result *res,
124 			  struct dlm_lock_result32 *res32)
125 {
126 	res32->version[0] = res->version[0];
127 	res32->version[1] = res->version[1];
128 	res32->version[2] = res->version[2];
129 
130 	res32->user_astaddr = (__u32)(long)res->user_astaddr;
131 	res32->user_astparam = (__u32)(long)res->user_astparam;
132 	res32->user_lksb = (__u32)(long)res->user_lksb;
133 	res32->bast_mode = res->bast_mode;
134 
135 	res32->lvb_offset = res->lvb_offset;
136 	res32->length = res->length;
137 
138 	res32->lksb.sb_status = res->lksb.sb_status;
139 	res32->lksb.sb_flags = res->lksb.sb_flags;
140 	res32->lksb.sb_lkid = res->lksb.sb_lkid;
141 	res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
142 }
143 #endif
144 
145 /* Figure out if this lock is at the end of its life and no longer
146    available for the application to use.  The lkb still exists until
147    the final ast is read.  A lock becomes EOL in three situations:
148      1. a noqueue request fails with EAGAIN
149      2. an unlock completes with EUNLOCK
150      3. a cancel of a waiting request completes with ECANCEL/EDEADLK
151    An EOL lock needs to be removed from the process's list of locks.
152    And we can't allow any new operation on an EOL lock.  This is
153    not related to the lifetime of the lkb struct which is managed
154    entirely by refcount. */
155 
156 static int lkb_is_endoflife(struct dlm_lkb *lkb, int sb_status, int type)
157 {
158 	switch (sb_status) {
159 	case -DLM_EUNLOCK:
160 		return 1;
161 	case -DLM_ECANCEL:
162 	case -ETIMEDOUT:
163 	case -EDEADLK:
164 		if (lkb->lkb_grmode == DLM_LOCK_IV)
165 			return 1;
166 		break;
167 	case -EAGAIN:
168 		if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV)
169 			return 1;
170 		break;
171 	}
172 	return 0;
173 }
174 
175 /* we could possibly check if the cancel of an orphan has resulted in the lkb
176    being removed and then remove that lkb from the orphans list and free it */
177 
178 void dlm_user_add_ast(struct dlm_lkb *lkb, int type)
179 {
180 	struct dlm_ls *ls;
181 	struct dlm_user_args *ua;
182 	struct dlm_user_proc *proc;
183 	int eol = 0, ast_type;
184 
185 	if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
186 		return;
187 
188 	ls = lkb->lkb_resource->res_ls;
189 	mutex_lock(&ls->ls_clear_proc_locks);
190 
191 	/* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
192 	   can't be delivered.  For ORPHAN's, dlm_clear_proc_locks() freed
193 	   lkb->ua so we can't try to use it.  This second check is necessary
194 	   for cases where a completion ast is received for an operation that
195 	   began before clear_proc_locks did its cancel/unlock. */
196 
197 	if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
198 		goto out;
199 
200 	DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
201 	ua = lkb->lkb_ua;
202 	proc = ua->proc;
203 
204 	if (type == AST_BAST && ua->bastaddr == NULL)
205 		goto out;
206 
207 	spin_lock(&proc->asts_spin);
208 
209 	ast_type = lkb->lkb_ast_type;
210 	lkb->lkb_ast_type |= type;
211 
212 	if (!ast_type) {
213 		kref_get(&lkb->lkb_ref);
214 		list_add_tail(&lkb->lkb_astqueue, &proc->asts);
215 		wake_up_interruptible(&proc->wait);
216 	}
217 	if (type == AST_COMP && (ast_type & AST_COMP))
218 		log_debug(ls, "ast overlap %x status %x %x",
219 			  lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags);
220 
221 	eol = lkb_is_endoflife(lkb, ua->lksb.sb_status, type);
222 	if (eol) {
223 		lkb->lkb_ast_type &= ~AST_BAST;
224 		lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
225 	}
226 
227 	/* We want to copy the lvb to userspace when the completion
228 	   ast is read if the status is 0, the lock has an lvb and
229 	   lvb_ops says we should.  We could probably have set_lvb_lock()
230 	   set update_user_lvb instead and not need old_mode */
231 
232 	if ((lkb->lkb_ast_type & AST_COMP) &&
233 	    (lkb->lkb_lksb->sb_status == 0) &&
234 	    lkb->lkb_lksb->sb_lvbptr &&
235 	    dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1])
236 		ua->update_user_lvb = 1;
237 	else
238 		ua->update_user_lvb = 0;
239 
240 	spin_unlock(&proc->asts_spin);
241 
242 	if (eol) {
243 		spin_lock(&proc->locks_spin);
244 		if (!list_empty(&lkb->lkb_ownqueue)) {
245 			list_del_init(&lkb->lkb_ownqueue);
246 			dlm_put_lkb(lkb);
247 		}
248 		spin_unlock(&proc->locks_spin);
249 	}
250  out:
251 	mutex_unlock(&ls->ls_clear_proc_locks);
252 }
253 
254 static int device_user_lock(struct dlm_user_proc *proc,
255 			    struct dlm_lock_params *params)
256 {
257 	struct dlm_ls *ls;
258 	struct dlm_user_args *ua;
259 	int error = -ENOMEM;
260 
261 	ls = dlm_find_lockspace_local(proc->lockspace);
262 	if (!ls)
263 		return -ENOENT;
264 
265 	if (!params->castaddr || !params->lksb) {
266 		error = -EINVAL;
267 		goto out;
268 	}
269 
270 	ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
271 	if (!ua)
272 		goto out;
273 	ua->proc = proc;
274 	ua->user_lksb = params->lksb;
275 	ua->castparam = params->castparam;
276 	ua->castaddr = params->castaddr;
277 	ua->bastparam = params->bastparam;
278 	ua->bastaddr = params->bastaddr;
279 	ua->xid = params->xid;
280 
281 	if (params->flags & DLM_LKF_CONVERT)
282 		error = dlm_user_convert(ls, ua,
283 				         params->mode, params->flags,
284 				         params->lkid, params->lvb,
285 					 (unsigned long) params->timeout);
286 	else {
287 		error = dlm_user_request(ls, ua,
288 					 params->mode, params->flags,
289 					 params->name, params->namelen,
290 					 (unsigned long) params->timeout);
291 		if (!error)
292 			error = ua->lksb.sb_lkid;
293 	}
294  out:
295 	dlm_put_lockspace(ls);
296 	return error;
297 }
298 
299 static int device_user_unlock(struct dlm_user_proc *proc,
300 			      struct dlm_lock_params *params)
301 {
302 	struct dlm_ls *ls;
303 	struct dlm_user_args *ua;
304 	int error = -ENOMEM;
305 
306 	ls = dlm_find_lockspace_local(proc->lockspace);
307 	if (!ls)
308 		return -ENOENT;
309 
310 	ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL);
311 	if (!ua)
312 		goto out;
313 	ua->proc = proc;
314 	ua->user_lksb = params->lksb;
315 	ua->castparam = params->castparam;
316 	ua->castaddr = params->castaddr;
317 
318 	if (params->flags & DLM_LKF_CANCEL)
319 		error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
320 	else
321 		error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
322 					params->lvb);
323  out:
324 	dlm_put_lockspace(ls);
325 	return error;
326 }
327 
328 static int device_user_deadlock(struct dlm_user_proc *proc,
329 				struct dlm_lock_params *params)
330 {
331 	struct dlm_ls *ls;
332 	int error;
333 
334 	ls = dlm_find_lockspace_local(proc->lockspace);
335 	if (!ls)
336 		return -ENOENT;
337 
338 	error = dlm_user_deadlock(ls, params->flags, params->lkid);
339 
340 	dlm_put_lockspace(ls);
341 	return error;
342 }
343 
344 static int dlm_device_register(struct dlm_ls *ls, char *name)
345 {
346 	int error, len;
347 
348 	/* The device is already registered.  This happens when the
349 	   lockspace is created multiple times from userspace. */
350 	if (ls->ls_device.name)
351 		return 0;
352 
353 	error = -ENOMEM;
354 	len = strlen(name) + strlen(name_prefix) + 2;
355 	ls->ls_device.name = kzalloc(len, GFP_KERNEL);
356 	if (!ls->ls_device.name)
357 		goto fail;
358 
359 	snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
360 		 name);
361 	ls->ls_device.fops = &device_fops;
362 	ls->ls_device.minor = MISC_DYNAMIC_MINOR;
363 
364 	error = misc_register(&ls->ls_device);
365 	if (error) {
366 		kfree(ls->ls_device.name);
367 	}
368 fail:
369 	return error;
370 }
371 
372 int dlm_device_deregister(struct dlm_ls *ls)
373 {
374 	int error;
375 
376 	/* The device is not registered.  This happens when the lockspace
377 	   was never used from userspace, or when device_create_lockspace()
378 	   calls dlm_release_lockspace() after the register fails. */
379 	if (!ls->ls_device.name)
380 		return 0;
381 
382 	error = misc_deregister(&ls->ls_device);
383 	if (!error)
384 		kfree(ls->ls_device.name);
385 	return error;
386 }
387 
388 static int device_user_purge(struct dlm_user_proc *proc,
389 			     struct dlm_purge_params *params)
390 {
391 	struct dlm_ls *ls;
392 	int error;
393 
394 	ls = dlm_find_lockspace_local(proc->lockspace);
395 	if (!ls)
396 		return -ENOENT;
397 
398 	error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
399 
400 	dlm_put_lockspace(ls);
401 	return error;
402 }
403 
404 static int device_create_lockspace(struct dlm_lspace_params *params)
405 {
406 	dlm_lockspace_t *lockspace;
407 	struct dlm_ls *ls;
408 	int error;
409 
410 	if (!capable(CAP_SYS_ADMIN))
411 		return -EPERM;
412 
413 	error = dlm_new_lockspace(params->name, strlen(params->name),
414 				  &lockspace, params->flags, DLM_USER_LVB_LEN);
415 	if (error)
416 		return error;
417 
418 	ls = dlm_find_lockspace_local(lockspace);
419 	if (!ls)
420 		return -ENOENT;
421 
422 	error = dlm_device_register(ls, params->name);
423 	dlm_put_lockspace(ls);
424 
425 	if (error)
426 		dlm_release_lockspace(lockspace, 0);
427 	else
428 		error = ls->ls_device.minor;
429 
430 	return error;
431 }
432 
433 static int device_remove_lockspace(struct dlm_lspace_params *params)
434 {
435 	dlm_lockspace_t *lockspace;
436 	struct dlm_ls *ls;
437 	int error, force = 0;
438 
439 	if (!capable(CAP_SYS_ADMIN))
440 		return -EPERM;
441 
442 	ls = dlm_find_lockspace_device(params->minor);
443 	if (!ls)
444 		return -ENOENT;
445 
446 	if (params->flags & DLM_USER_LSFLG_FORCEFREE)
447 		force = 2;
448 
449 	lockspace = ls->ls_local_handle;
450 	dlm_put_lockspace(ls);
451 
452 	/* The final dlm_release_lockspace waits for references to go to
453 	   zero, so all processes will need to close their device for the
454 	   ls before the release will proceed.  release also calls the
455 	   device_deregister above.  Converting a positive return value
456 	   from release to zero means that userspace won't know when its
457 	   release was the final one, but it shouldn't need to know. */
458 
459 	error = dlm_release_lockspace(lockspace, force);
460 	if (error > 0)
461 		error = 0;
462 	return error;
463 }
464 
465 /* Check the user's version matches ours */
466 static int check_version(struct dlm_write_request *req)
467 {
468 	if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
469 	    (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
470 	     req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
471 
472 		printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
473 		       "user (%d.%d.%d) kernel (%d.%d.%d)\n",
474 		       current->comm,
475 		       task_pid_nr(current),
476 		       req->version[0],
477 		       req->version[1],
478 		       req->version[2],
479 		       DLM_DEVICE_VERSION_MAJOR,
480 		       DLM_DEVICE_VERSION_MINOR,
481 		       DLM_DEVICE_VERSION_PATCH);
482 		return -EINVAL;
483 	}
484 	return 0;
485 }
486 
487 /*
488  * device_write
489  *
490  *   device_user_lock
491  *     dlm_user_request -> request_lock
492  *     dlm_user_convert -> convert_lock
493  *
494  *   device_user_unlock
495  *     dlm_user_unlock -> unlock_lock
496  *     dlm_user_cancel -> cancel_lock
497  *
498  *   device_create_lockspace
499  *     dlm_new_lockspace
500  *
501  *   device_remove_lockspace
502  *     dlm_release_lockspace
503  */
504 
505 /* a write to a lockspace device is a lock or unlock request, a write
506    to the control device is to create/remove a lockspace */
507 
508 static ssize_t device_write(struct file *file, const char __user *buf,
509 			    size_t count, loff_t *ppos)
510 {
511 	struct dlm_user_proc *proc = file->private_data;
512 	struct dlm_write_request *kbuf;
513 	sigset_t tmpsig, allsigs;
514 	int error;
515 
516 #ifdef CONFIG_COMPAT
517 	if (count < sizeof(struct dlm_write_request32))
518 #else
519 	if (count < sizeof(struct dlm_write_request))
520 #endif
521 		return -EINVAL;
522 
523 	kbuf = kzalloc(count + 1, GFP_KERNEL);
524 	if (!kbuf)
525 		return -ENOMEM;
526 
527 	if (copy_from_user(kbuf, buf, count)) {
528 		error = -EFAULT;
529 		goto out_free;
530 	}
531 
532 	if (check_version(kbuf)) {
533 		error = -EBADE;
534 		goto out_free;
535 	}
536 
537 #ifdef CONFIG_COMPAT
538 	if (!kbuf->is64bit) {
539 		struct dlm_write_request32 *k32buf;
540 		k32buf = (struct dlm_write_request32 *)kbuf;
541 		kbuf = kmalloc(count + 1 + (sizeof(struct dlm_write_request) -
542 			       sizeof(struct dlm_write_request32)), GFP_KERNEL);
543 		if (!kbuf) {
544 			kfree(k32buf);
545 			return -ENOMEM;
546 		}
547 
548 		if (proc)
549 			set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
550 		compat_input(kbuf, k32buf, count + 1);
551 		kfree(k32buf);
552 	}
553 #endif
554 
555 	/* do we really need this? can a write happen after a close? */
556 	if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
557 	    (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
558 		error = -EINVAL;
559 		goto out_free;
560 	}
561 
562 	sigfillset(&allsigs);
563 	sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
564 
565 	error = -EINVAL;
566 
567 	switch (kbuf->cmd)
568 	{
569 	case DLM_USER_LOCK:
570 		if (!proc) {
571 			log_print("no locking on control device");
572 			goto out_sig;
573 		}
574 		error = device_user_lock(proc, &kbuf->i.lock);
575 		break;
576 
577 	case DLM_USER_UNLOCK:
578 		if (!proc) {
579 			log_print("no locking on control device");
580 			goto out_sig;
581 		}
582 		error = device_user_unlock(proc, &kbuf->i.lock);
583 		break;
584 
585 	case DLM_USER_DEADLOCK:
586 		if (!proc) {
587 			log_print("no locking on control device");
588 			goto out_sig;
589 		}
590 		error = device_user_deadlock(proc, &kbuf->i.lock);
591 		break;
592 
593 	case DLM_USER_CREATE_LOCKSPACE:
594 		if (proc) {
595 			log_print("create/remove only on control device");
596 			goto out_sig;
597 		}
598 		error = device_create_lockspace(&kbuf->i.lspace);
599 		break;
600 
601 	case DLM_USER_REMOVE_LOCKSPACE:
602 		if (proc) {
603 			log_print("create/remove only on control device");
604 			goto out_sig;
605 		}
606 		error = device_remove_lockspace(&kbuf->i.lspace);
607 		break;
608 
609 	case DLM_USER_PURGE:
610 		if (!proc) {
611 			log_print("no locking on control device");
612 			goto out_sig;
613 		}
614 		error = device_user_purge(proc, &kbuf->i.purge);
615 		break;
616 
617 	default:
618 		log_print("Unknown command passed to DLM device : %d\n",
619 			  kbuf->cmd);
620 	}
621 
622  out_sig:
623 	sigprocmask(SIG_SETMASK, &tmpsig, NULL);
624 	recalc_sigpending();
625  out_free:
626 	kfree(kbuf);
627 	return error;
628 }
629 
630 /* Every process that opens the lockspace device has its own "proc" structure
631    hanging off the open file that's used to keep track of locks owned by the
632    process and asts that need to be delivered to the process. */
633 
634 static int device_open(struct inode *inode, struct file *file)
635 {
636 	struct dlm_user_proc *proc;
637 	struct dlm_ls *ls;
638 
639 	ls = dlm_find_lockspace_device(iminor(inode));
640 	if (!ls)
641 		return -ENOENT;
642 
643 	proc = kzalloc(sizeof(struct dlm_user_proc), GFP_KERNEL);
644 	if (!proc) {
645 		dlm_put_lockspace(ls);
646 		return -ENOMEM;
647 	}
648 
649 	proc->lockspace = ls->ls_local_handle;
650 	INIT_LIST_HEAD(&proc->asts);
651 	INIT_LIST_HEAD(&proc->locks);
652 	INIT_LIST_HEAD(&proc->unlocking);
653 	spin_lock_init(&proc->asts_spin);
654 	spin_lock_init(&proc->locks_spin);
655 	init_waitqueue_head(&proc->wait);
656 	file->private_data = proc;
657 
658 	return 0;
659 }
660 
661 static int device_close(struct inode *inode, struct file *file)
662 {
663 	struct dlm_user_proc *proc = file->private_data;
664 	struct dlm_ls *ls;
665 	sigset_t tmpsig, allsigs;
666 
667 	ls = dlm_find_lockspace_local(proc->lockspace);
668 	if (!ls)
669 		return -ENOENT;
670 
671 	sigfillset(&allsigs);
672 	sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
673 
674 	set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
675 
676 	dlm_clear_proc_locks(ls, proc);
677 
678 	/* at this point no more lkb's should exist for this lockspace,
679 	   so there's no chance of dlm_user_add_ast() being called and
680 	   looking for lkb->ua->proc */
681 
682 	kfree(proc);
683 	file->private_data = NULL;
684 
685 	dlm_put_lockspace(ls);
686 	dlm_put_lockspace(ls);  /* for the find in device_open() */
687 
688 	/* FIXME: AUTOFREE: if this ls is no longer used do
689 	   device_remove_lockspace() */
690 
691 	sigprocmask(SIG_SETMASK, &tmpsig, NULL);
692 	recalc_sigpending();
693 
694 	return 0;
695 }
696 
697 static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type,
698 			       int bmode, char __user *buf, size_t count)
699 {
700 #ifdef CONFIG_COMPAT
701 	struct dlm_lock_result32 result32;
702 #endif
703 	struct dlm_lock_result result;
704 	void *resultptr;
705 	int error=0;
706 	int len;
707 	int struct_len;
708 
709 	memset(&result, 0, sizeof(struct dlm_lock_result));
710 	result.version[0] = DLM_DEVICE_VERSION_MAJOR;
711 	result.version[1] = DLM_DEVICE_VERSION_MINOR;
712 	result.version[2] = DLM_DEVICE_VERSION_PATCH;
713 	memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
714 	result.user_lksb = ua->user_lksb;
715 
716 	/* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
717 	   in a conversion unless the conversion is successful.  See code
718 	   in dlm_user_convert() for updating ua from ua_tmp.  OpenVMS, though,
719 	   notes that a new blocking AST address and parameter are set even if
720 	   the conversion fails, so maybe we should just do that. */
721 
722 	if (type == AST_BAST) {
723 		result.user_astaddr = ua->bastaddr;
724 		result.user_astparam = ua->bastparam;
725 		result.bast_mode = bmode;
726 	} else {
727 		result.user_astaddr = ua->castaddr;
728 		result.user_astparam = ua->castparam;
729 	}
730 
731 #ifdef CONFIG_COMPAT
732 	if (compat)
733 		len = sizeof(struct dlm_lock_result32);
734 	else
735 #endif
736 		len = sizeof(struct dlm_lock_result);
737 	struct_len = len;
738 
739 	/* copy lvb to userspace if there is one, it's been updated, and
740 	   the user buffer has space for it */
741 
742 	if (ua->update_user_lvb && ua->lksb.sb_lvbptr &&
743 	    count >= len + DLM_USER_LVB_LEN) {
744 		if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
745 				 DLM_USER_LVB_LEN)) {
746 			error = -EFAULT;
747 			goto out;
748 		}
749 
750 		result.lvb_offset = len;
751 		len += DLM_USER_LVB_LEN;
752 	}
753 
754 	result.length = len;
755 	resultptr = &result;
756 #ifdef CONFIG_COMPAT
757 	if (compat) {
758 		compat_output(&result, &result32);
759 		resultptr = &result32;
760 	}
761 #endif
762 
763 	if (copy_to_user(buf, resultptr, struct_len))
764 		error = -EFAULT;
765 	else
766 		error = len;
767  out:
768 	return error;
769 }
770 
771 static int copy_version_to_user(char __user *buf, size_t count)
772 {
773 	struct dlm_device_version ver;
774 
775 	memset(&ver, 0, sizeof(struct dlm_device_version));
776 	ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
777 	ver.version[1] = DLM_DEVICE_VERSION_MINOR;
778 	ver.version[2] = DLM_DEVICE_VERSION_PATCH;
779 
780 	if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
781 		return -EFAULT;
782 	return sizeof(struct dlm_device_version);
783 }
784 
785 /* a read returns a single ast described in a struct dlm_lock_result */
786 
787 static ssize_t device_read(struct file *file, char __user *buf, size_t count,
788 			   loff_t *ppos)
789 {
790 	struct dlm_user_proc *proc = file->private_data;
791 	struct dlm_lkb *lkb;
792 	DECLARE_WAITQUEUE(wait, current);
793 	int error, type=0, bmode=0, removed = 0;
794 
795 	if (count == sizeof(struct dlm_device_version)) {
796 		error = copy_version_to_user(buf, count);
797 		return error;
798 	}
799 
800 	if (!proc) {
801 		log_print("non-version read from control device %zu", count);
802 		return -EINVAL;
803 	}
804 
805 #ifdef CONFIG_COMPAT
806 	if (count < sizeof(struct dlm_lock_result32))
807 #else
808 	if (count < sizeof(struct dlm_lock_result))
809 #endif
810 		return -EINVAL;
811 
812 	/* do we really need this? can a read happen after a close? */
813 	if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
814 		return -EINVAL;
815 
816 	spin_lock(&proc->asts_spin);
817 	if (list_empty(&proc->asts)) {
818 		if (file->f_flags & O_NONBLOCK) {
819 			spin_unlock(&proc->asts_spin);
820 			return -EAGAIN;
821 		}
822 
823 		add_wait_queue(&proc->wait, &wait);
824 
825 	repeat:
826 		set_current_state(TASK_INTERRUPTIBLE);
827 		if (list_empty(&proc->asts) && !signal_pending(current)) {
828 			spin_unlock(&proc->asts_spin);
829 			schedule();
830 			spin_lock(&proc->asts_spin);
831 			goto repeat;
832 		}
833 		set_current_state(TASK_RUNNING);
834 		remove_wait_queue(&proc->wait, &wait);
835 
836 		if (signal_pending(current)) {
837 			spin_unlock(&proc->asts_spin);
838 			return -ERESTARTSYS;
839 		}
840 	}
841 
842 	/* there may be both completion and blocking asts to return for
843 	   the lkb, don't remove lkb from asts list unless no asts remain */
844 
845 	lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue);
846 
847 	if (lkb->lkb_ast_type & AST_COMP) {
848 		lkb->lkb_ast_type &= ~AST_COMP;
849 		type = AST_COMP;
850 	} else if (lkb->lkb_ast_type & AST_BAST) {
851 		lkb->lkb_ast_type &= ~AST_BAST;
852 		type = AST_BAST;
853 		bmode = lkb->lkb_bastmode;
854 	}
855 
856 	if (!lkb->lkb_ast_type) {
857 		list_del(&lkb->lkb_astqueue);
858 		removed = 1;
859 	}
860 	spin_unlock(&proc->asts_spin);
861 
862 	error = copy_result_to_user(lkb->lkb_ua,
863 			 	test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
864 				type, bmode, buf, count);
865 
866 	/* removes reference for the proc->asts lists added by
867 	   dlm_user_add_ast() and may result in the lkb being freed */
868 	if (removed)
869 		dlm_put_lkb(lkb);
870 
871 	return error;
872 }
873 
874 static unsigned int device_poll(struct file *file, poll_table *wait)
875 {
876 	struct dlm_user_proc *proc = file->private_data;
877 
878 	poll_wait(file, &proc->wait, wait);
879 
880 	spin_lock(&proc->asts_spin);
881 	if (!list_empty(&proc->asts)) {
882 		spin_unlock(&proc->asts_spin);
883 		return POLLIN | POLLRDNORM;
884 	}
885 	spin_unlock(&proc->asts_spin);
886 	return 0;
887 }
888 
889 int dlm_user_daemon_available(void)
890 {
891 	/* dlm_controld hasn't started (or, has started, but not
892 	   properly populated configfs) */
893 
894 	if (!dlm_our_nodeid())
895 		return 0;
896 
897 	/* This is to deal with versions of dlm_controld that don't
898 	   know about the monitor device.  We assume that if the
899 	   dlm_controld was started (above), but the monitor device
900 	   was never opened, that it's an old version.  dlm_controld
901 	   should open the monitor device before populating configfs. */
902 
903 	if (dlm_monitor_unused)
904 		return 1;
905 
906 	return atomic_read(&dlm_monitor_opened) ? 1 : 0;
907 }
908 
909 static int ctl_device_open(struct inode *inode, struct file *file)
910 {
911 	file->private_data = NULL;
912 	return 0;
913 }
914 
915 static int ctl_device_close(struct inode *inode, struct file *file)
916 {
917 	return 0;
918 }
919 
920 static int monitor_device_open(struct inode *inode, struct file *file)
921 {
922 	atomic_inc(&dlm_monitor_opened);
923 	dlm_monitor_unused = 0;
924 	return 0;
925 }
926 
927 static int monitor_device_close(struct inode *inode, struct file *file)
928 {
929 	if (atomic_dec_and_test(&dlm_monitor_opened))
930 		dlm_stop_lockspaces();
931 	return 0;
932 }
933 
934 static const struct file_operations device_fops = {
935 	.open    = device_open,
936 	.release = device_close,
937 	.read    = device_read,
938 	.write   = device_write,
939 	.poll    = device_poll,
940 	.owner   = THIS_MODULE,
941 };
942 
943 static const struct file_operations ctl_device_fops = {
944 	.open    = ctl_device_open,
945 	.release = ctl_device_close,
946 	.read    = device_read,
947 	.write   = device_write,
948 	.owner   = THIS_MODULE,
949 };
950 
951 static struct miscdevice ctl_device = {
952 	.name  = "dlm-control",
953 	.fops  = &ctl_device_fops,
954 	.minor = MISC_DYNAMIC_MINOR,
955 };
956 
957 static const struct file_operations monitor_device_fops = {
958 	.open    = monitor_device_open,
959 	.release = monitor_device_close,
960 	.owner   = THIS_MODULE,
961 };
962 
963 static struct miscdevice monitor_device = {
964 	.name  = "dlm-monitor",
965 	.fops  = &monitor_device_fops,
966 	.minor = MISC_DYNAMIC_MINOR,
967 };
968 
969 int __init dlm_user_init(void)
970 {
971 	int error;
972 
973 	atomic_set(&dlm_monitor_opened, 0);
974 
975 	error = misc_register(&ctl_device);
976 	if (error) {
977 		log_print("misc_register failed for control device");
978 		goto out;
979 	}
980 
981 	error = misc_register(&monitor_device);
982 	if (error) {
983 		log_print("misc_register failed for monitor device");
984 		misc_deregister(&ctl_device);
985 	}
986  out:
987 	return error;
988 }
989 
990 void dlm_user_exit(void)
991 {
992 	misc_deregister(&ctl_device);
993 	misc_deregister(&monitor_device);
994 }
995 
996