xref: /linux/fs/dlm/user.c (revision 4413e16d9d21673bb5048a2e542f1aaa00015c2e)
1 /*
2  * Copyright (C) 2006-2010 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 #include <linux/slab.h>
21 
22 #include "dlm_internal.h"
23 #include "lockspace.h"
24 #include "lock.h"
25 #include "lvb_table.h"
26 #include "user.h"
27 #include "ast.h"
28 
29 static const char name_prefix[] = "dlm";
30 static const struct file_operations device_fops;
31 static atomic_t dlm_monitor_opened;
32 static int dlm_monitor_unused = 1;
33 
34 #ifdef CONFIG_COMPAT
35 
36 struct dlm_lock_params32 {
37 	__u8 mode;
38 	__u8 namelen;
39 	__u16 unused;
40 	__u32 flags;
41 	__u32 lkid;
42 	__u32 parent;
43 	__u64 xid;
44 	__u64 timeout;
45 	__u32 castparam;
46 	__u32 castaddr;
47 	__u32 bastparam;
48 	__u32 bastaddr;
49 	__u32 lksb;
50 	char lvb[DLM_USER_LVB_LEN];
51 	char name[0];
52 };
53 
54 struct dlm_write_request32 {
55 	__u32 version[3];
56 	__u8 cmd;
57 	__u8 is64bit;
58 	__u8 unused[2];
59 
60 	union  {
61 		struct dlm_lock_params32 lock;
62 		struct dlm_lspace_params lspace;
63 		struct dlm_purge_params purge;
64 	} i;
65 };
66 
67 struct dlm_lksb32 {
68 	__u32 sb_status;
69 	__u32 sb_lkid;
70 	__u8 sb_flags;
71 	__u32 sb_lvbptr;
72 };
73 
74 struct dlm_lock_result32 {
75 	__u32 version[3];
76 	__u32 length;
77 	__u32 user_astaddr;
78 	__u32 user_astparam;
79 	__u32 user_lksb;
80 	struct dlm_lksb32 lksb;
81 	__u8 bast_mode;
82 	__u8 unused[3];
83 	/* Offsets may be zero if no data is present */
84 	__u32 lvb_offset;
85 };
86 
87 static void compat_input(struct dlm_write_request *kb,
88 			 struct dlm_write_request32 *kb32,
89 			 int namelen)
90 {
91 	kb->version[0] = kb32->version[0];
92 	kb->version[1] = kb32->version[1];
93 	kb->version[2] = kb32->version[2];
94 
95 	kb->cmd = kb32->cmd;
96 	kb->is64bit = kb32->is64bit;
97 	if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
98 	    kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
99 		kb->i.lspace.flags = kb32->i.lspace.flags;
100 		kb->i.lspace.minor = kb32->i.lspace.minor;
101 		memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen);
102 	} else if (kb->cmd == DLM_USER_PURGE) {
103 		kb->i.purge.nodeid = kb32->i.purge.nodeid;
104 		kb->i.purge.pid = kb32->i.purge.pid;
105 	} else {
106 		kb->i.lock.mode = kb32->i.lock.mode;
107 		kb->i.lock.namelen = kb32->i.lock.namelen;
108 		kb->i.lock.flags = kb32->i.lock.flags;
109 		kb->i.lock.lkid = kb32->i.lock.lkid;
110 		kb->i.lock.parent = kb32->i.lock.parent;
111 		kb->i.lock.xid = kb32->i.lock.xid;
112 		kb->i.lock.timeout = kb32->i.lock.timeout;
113 		kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
114 		kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
115 		kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
116 		kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
117 		kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
118 		memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
119 		memcpy(kb->i.lock.name, kb32->i.lock.name, namelen);
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(int mode, int status)
157 {
158 	switch (status) {
159 	case -DLM_EUNLOCK:
160 		return 1;
161 	case -DLM_ECANCEL:
162 	case -ETIMEDOUT:
163 	case -EDEADLK:
164 	case -EAGAIN:
165 		if (mode == DLM_LOCK_IV)
166 			return 1;
167 		break;
168 	}
169 	return 0;
170 }
171 
172 /* we could possibly check if the cancel of an orphan has resulted in the lkb
173    being removed and then remove that lkb from the orphans list and free it */
174 
175 void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
176 		      int status, uint32_t sbflags, uint64_t seq)
177 {
178 	struct dlm_ls *ls;
179 	struct dlm_user_args *ua;
180 	struct dlm_user_proc *proc;
181 	int rv;
182 
183 	if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
184 		return;
185 
186 	ls = lkb->lkb_resource->res_ls;
187 	mutex_lock(&ls->ls_clear_proc_locks);
188 
189 	/* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
190 	   can't be delivered.  For ORPHAN's, dlm_clear_proc_locks() freed
191 	   lkb->ua so we can't try to use it.  This second check is necessary
192 	   for cases where a completion ast is received for an operation that
193 	   began before clear_proc_locks did its cancel/unlock. */
194 
195 	if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
196 		goto out;
197 
198 	DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
199 	ua = lkb->lkb_ua;
200 	proc = ua->proc;
201 
202 	if ((flags & DLM_CB_BAST) && ua->bastaddr == NULL)
203 		goto out;
204 
205 	if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status))
206 		lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
207 
208 	spin_lock(&proc->asts_spin);
209 
210 	rv = dlm_add_lkb_callback(lkb, flags, mode, status, sbflags, seq);
211 	if (rv < 0) {
212 		spin_unlock(&proc->asts_spin);
213 		goto out;
214 	}
215 
216 	if (list_empty(&lkb->lkb_cb_list)) {
217 		kref_get(&lkb->lkb_ref);
218 		list_add_tail(&lkb->lkb_cb_list, &proc->asts);
219 		wake_up_interruptible(&proc->wait);
220 	}
221 	spin_unlock(&proc->asts_spin);
222 
223 	if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
224 		/* N.B. spin_lock locks_spin, not asts_spin */
225 		spin_lock(&proc->locks_spin);
226 		if (!list_empty(&lkb->lkb_ownqueue)) {
227 			list_del_init(&lkb->lkb_ownqueue);
228 			dlm_put_lkb(lkb);
229 		}
230 		spin_unlock(&proc->locks_spin);
231 	}
232  out:
233 	mutex_unlock(&ls->ls_clear_proc_locks);
234 }
235 
236 static int device_user_lock(struct dlm_user_proc *proc,
237 			    struct dlm_lock_params *params)
238 {
239 	struct dlm_ls *ls;
240 	struct dlm_user_args *ua;
241 	int error = -ENOMEM;
242 
243 	ls = dlm_find_lockspace_local(proc->lockspace);
244 	if (!ls)
245 		return -ENOENT;
246 
247 	if (!params->castaddr || !params->lksb) {
248 		error = -EINVAL;
249 		goto out;
250 	}
251 
252 	ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
253 	if (!ua)
254 		goto out;
255 	ua->proc = proc;
256 	ua->user_lksb = params->lksb;
257 	ua->castparam = params->castparam;
258 	ua->castaddr = params->castaddr;
259 	ua->bastparam = params->bastparam;
260 	ua->bastaddr = params->bastaddr;
261 	ua->xid = params->xid;
262 
263 	if (params->flags & DLM_LKF_CONVERT)
264 		error = dlm_user_convert(ls, ua,
265 				         params->mode, params->flags,
266 				         params->lkid, params->lvb,
267 					 (unsigned long) params->timeout);
268 	else {
269 		error = dlm_user_request(ls, ua,
270 					 params->mode, params->flags,
271 					 params->name, params->namelen,
272 					 (unsigned long) params->timeout);
273 		if (!error)
274 			error = ua->lksb.sb_lkid;
275 	}
276  out:
277 	dlm_put_lockspace(ls);
278 	return error;
279 }
280 
281 static int device_user_unlock(struct dlm_user_proc *proc,
282 			      struct dlm_lock_params *params)
283 {
284 	struct dlm_ls *ls;
285 	struct dlm_user_args *ua;
286 	int error = -ENOMEM;
287 
288 	ls = dlm_find_lockspace_local(proc->lockspace);
289 	if (!ls)
290 		return -ENOENT;
291 
292 	ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
293 	if (!ua)
294 		goto out;
295 	ua->proc = proc;
296 	ua->user_lksb = params->lksb;
297 	ua->castparam = params->castparam;
298 	ua->castaddr = params->castaddr;
299 
300 	if (params->flags & DLM_LKF_CANCEL)
301 		error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
302 	else
303 		error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
304 					params->lvb);
305  out:
306 	dlm_put_lockspace(ls);
307 	return error;
308 }
309 
310 static int device_user_deadlock(struct dlm_user_proc *proc,
311 				struct dlm_lock_params *params)
312 {
313 	struct dlm_ls *ls;
314 	int error;
315 
316 	ls = dlm_find_lockspace_local(proc->lockspace);
317 	if (!ls)
318 		return -ENOENT;
319 
320 	error = dlm_user_deadlock(ls, params->flags, params->lkid);
321 
322 	dlm_put_lockspace(ls);
323 	return error;
324 }
325 
326 static int dlm_device_register(struct dlm_ls *ls, char *name)
327 {
328 	int error, len;
329 
330 	/* The device is already registered.  This happens when the
331 	   lockspace is created multiple times from userspace. */
332 	if (ls->ls_device.name)
333 		return 0;
334 
335 	error = -ENOMEM;
336 	len = strlen(name) + strlen(name_prefix) + 2;
337 	ls->ls_device.name = kzalloc(len, GFP_NOFS);
338 	if (!ls->ls_device.name)
339 		goto fail;
340 
341 	snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
342 		 name);
343 	ls->ls_device.fops = &device_fops;
344 	ls->ls_device.minor = MISC_DYNAMIC_MINOR;
345 
346 	error = misc_register(&ls->ls_device);
347 	if (error) {
348 		kfree(ls->ls_device.name);
349 	}
350 fail:
351 	return error;
352 }
353 
354 int dlm_device_deregister(struct dlm_ls *ls)
355 {
356 	int error;
357 
358 	/* The device is not registered.  This happens when the lockspace
359 	   was never used from userspace, or when device_create_lockspace()
360 	   calls dlm_release_lockspace() after the register fails. */
361 	if (!ls->ls_device.name)
362 		return 0;
363 
364 	error = misc_deregister(&ls->ls_device);
365 	if (!error)
366 		kfree(ls->ls_device.name);
367 	return error;
368 }
369 
370 static int device_user_purge(struct dlm_user_proc *proc,
371 			     struct dlm_purge_params *params)
372 {
373 	struct dlm_ls *ls;
374 	int error;
375 
376 	ls = dlm_find_lockspace_local(proc->lockspace);
377 	if (!ls)
378 		return -ENOENT;
379 
380 	error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
381 
382 	dlm_put_lockspace(ls);
383 	return error;
384 }
385 
386 static int device_create_lockspace(struct dlm_lspace_params *params)
387 {
388 	dlm_lockspace_t *lockspace;
389 	struct dlm_ls *ls;
390 	int error;
391 
392 	if (!capable(CAP_SYS_ADMIN))
393 		return -EPERM;
394 
395 	error = dlm_new_lockspace(params->name, NULL, params->flags,
396 				  DLM_USER_LVB_LEN, NULL, NULL, NULL,
397 				  &lockspace);
398 	if (error)
399 		return error;
400 
401 	ls = dlm_find_lockspace_local(lockspace);
402 	if (!ls)
403 		return -ENOENT;
404 
405 	error = dlm_device_register(ls, params->name);
406 	dlm_put_lockspace(ls);
407 
408 	if (error)
409 		dlm_release_lockspace(lockspace, 0);
410 	else
411 		error = ls->ls_device.minor;
412 
413 	return error;
414 }
415 
416 static int device_remove_lockspace(struct dlm_lspace_params *params)
417 {
418 	dlm_lockspace_t *lockspace;
419 	struct dlm_ls *ls;
420 	int error, force = 0;
421 
422 	if (!capable(CAP_SYS_ADMIN))
423 		return -EPERM;
424 
425 	ls = dlm_find_lockspace_device(params->minor);
426 	if (!ls)
427 		return -ENOENT;
428 
429 	if (params->flags & DLM_USER_LSFLG_FORCEFREE)
430 		force = 2;
431 
432 	lockspace = ls->ls_local_handle;
433 	dlm_put_lockspace(ls);
434 
435 	/* The final dlm_release_lockspace waits for references to go to
436 	   zero, so all processes will need to close their device for the
437 	   ls before the release will proceed.  release also calls the
438 	   device_deregister above.  Converting a positive return value
439 	   from release to zero means that userspace won't know when its
440 	   release was the final one, but it shouldn't need to know. */
441 
442 	error = dlm_release_lockspace(lockspace, force);
443 	if (error > 0)
444 		error = 0;
445 	return error;
446 }
447 
448 /* Check the user's version matches ours */
449 static int check_version(struct dlm_write_request *req)
450 {
451 	if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
452 	    (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
453 	     req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
454 
455 		printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
456 		       "user (%d.%d.%d) kernel (%d.%d.%d)\n",
457 		       current->comm,
458 		       task_pid_nr(current),
459 		       req->version[0],
460 		       req->version[1],
461 		       req->version[2],
462 		       DLM_DEVICE_VERSION_MAJOR,
463 		       DLM_DEVICE_VERSION_MINOR,
464 		       DLM_DEVICE_VERSION_PATCH);
465 		return -EINVAL;
466 	}
467 	return 0;
468 }
469 
470 /*
471  * device_write
472  *
473  *   device_user_lock
474  *     dlm_user_request -> request_lock
475  *     dlm_user_convert -> convert_lock
476  *
477  *   device_user_unlock
478  *     dlm_user_unlock -> unlock_lock
479  *     dlm_user_cancel -> cancel_lock
480  *
481  *   device_create_lockspace
482  *     dlm_new_lockspace
483  *
484  *   device_remove_lockspace
485  *     dlm_release_lockspace
486  */
487 
488 /* a write to a lockspace device is a lock or unlock request, a write
489    to the control device is to create/remove a lockspace */
490 
491 static ssize_t device_write(struct file *file, const char __user *buf,
492 			    size_t count, loff_t *ppos)
493 {
494 	struct dlm_user_proc *proc = file->private_data;
495 	struct dlm_write_request *kbuf;
496 	sigset_t tmpsig, allsigs;
497 	int error;
498 
499 #ifdef CONFIG_COMPAT
500 	if (count < sizeof(struct dlm_write_request32))
501 #else
502 	if (count < sizeof(struct dlm_write_request))
503 #endif
504 		return -EINVAL;
505 
506 #ifdef CONFIG_COMPAT
507 	if (count > sizeof(struct dlm_write_request32) + DLM_RESNAME_MAXLEN)
508 #else
509 	if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
510 #endif
511 		return -EINVAL;
512 
513 	kbuf = kzalloc(count + 1, GFP_NOFS);
514 	if (!kbuf)
515 		return -ENOMEM;
516 
517 	if (copy_from_user(kbuf, buf, count)) {
518 		error = -EFAULT;
519 		goto out_free;
520 	}
521 
522 	if (check_version(kbuf)) {
523 		error = -EBADE;
524 		goto out_free;
525 	}
526 
527 #ifdef CONFIG_COMPAT
528 	if (!kbuf->is64bit) {
529 		struct dlm_write_request32 *k32buf;
530 		int namelen = 0;
531 
532 		if (count > sizeof(struct dlm_write_request32))
533 			namelen = count - sizeof(struct dlm_write_request32);
534 
535 		k32buf = (struct dlm_write_request32 *)kbuf;
536 
537 		/* add 1 after namelen so that the name string is terminated */
538 		kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
539 			       GFP_NOFS);
540 		if (!kbuf) {
541 			kfree(k32buf);
542 			return -ENOMEM;
543 		}
544 
545 		if (proc)
546 			set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
547 
548 		compat_input(kbuf, k32buf, namelen);
549 		kfree(k32buf);
550 	}
551 #endif
552 
553 	/* do we really need this? can a write happen after a close? */
554 	if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
555 	    (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
556 		error = -EINVAL;
557 		goto out_free;
558 	}
559 
560 	sigfillset(&allsigs);
561 	sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
562 
563 	error = -EINVAL;
564 
565 	switch (kbuf->cmd)
566 	{
567 	case DLM_USER_LOCK:
568 		if (!proc) {
569 			log_print("no locking on control device");
570 			goto out_sig;
571 		}
572 		error = device_user_lock(proc, &kbuf->i.lock);
573 		break;
574 
575 	case DLM_USER_UNLOCK:
576 		if (!proc) {
577 			log_print("no locking on control device");
578 			goto out_sig;
579 		}
580 		error = device_user_unlock(proc, &kbuf->i.lock);
581 		break;
582 
583 	case DLM_USER_DEADLOCK:
584 		if (!proc) {
585 			log_print("no locking on control device");
586 			goto out_sig;
587 		}
588 		error = device_user_deadlock(proc, &kbuf->i.lock);
589 		break;
590 
591 	case DLM_USER_CREATE_LOCKSPACE:
592 		if (proc) {
593 			log_print("create/remove only on control device");
594 			goto out_sig;
595 		}
596 		error = device_create_lockspace(&kbuf->i.lspace);
597 		break;
598 
599 	case DLM_USER_REMOVE_LOCKSPACE:
600 		if (proc) {
601 			log_print("create/remove only on control device");
602 			goto out_sig;
603 		}
604 		error = device_remove_lockspace(&kbuf->i.lspace);
605 		break;
606 
607 	case DLM_USER_PURGE:
608 		if (!proc) {
609 			log_print("no locking on control device");
610 			goto out_sig;
611 		}
612 		error = device_user_purge(proc, &kbuf->i.purge);
613 		break;
614 
615 	default:
616 		log_print("Unknown command passed to DLM device : %d\n",
617 			  kbuf->cmd);
618 	}
619 
620  out_sig:
621 	sigprocmask(SIG_SETMASK, &tmpsig, NULL);
622  out_free:
623 	kfree(kbuf);
624 	return error;
625 }
626 
627 /* Every process that opens the lockspace device has its own "proc" structure
628    hanging off the open file that's used to keep track of locks owned by the
629    process and asts that need to be delivered to the process. */
630 
631 static int device_open(struct inode *inode, struct file *file)
632 {
633 	struct dlm_user_proc *proc;
634 	struct dlm_ls *ls;
635 
636 	ls = dlm_find_lockspace_device(iminor(inode));
637 	if (!ls)
638 		return -ENOENT;
639 
640 	proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
641 	if (!proc) {
642 		dlm_put_lockspace(ls);
643 		return -ENOMEM;
644 	}
645 
646 	proc->lockspace = ls->ls_local_handle;
647 	INIT_LIST_HEAD(&proc->asts);
648 	INIT_LIST_HEAD(&proc->locks);
649 	INIT_LIST_HEAD(&proc->unlocking);
650 	spin_lock_init(&proc->asts_spin);
651 	spin_lock_init(&proc->locks_spin);
652 	init_waitqueue_head(&proc->wait);
653 	file->private_data = proc;
654 
655 	return 0;
656 }
657 
658 static int device_close(struct inode *inode, struct file *file)
659 {
660 	struct dlm_user_proc *proc = file->private_data;
661 	struct dlm_ls *ls;
662 	sigset_t tmpsig, allsigs;
663 
664 	ls = dlm_find_lockspace_local(proc->lockspace);
665 	if (!ls)
666 		return -ENOENT;
667 
668 	sigfillset(&allsigs);
669 	sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
670 
671 	set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
672 
673 	dlm_clear_proc_locks(ls, proc);
674 
675 	/* at this point no more lkb's should exist for this lockspace,
676 	   so there's no chance of dlm_user_add_ast() being called and
677 	   looking for lkb->ua->proc */
678 
679 	kfree(proc);
680 	file->private_data = NULL;
681 
682 	dlm_put_lockspace(ls);
683 	dlm_put_lockspace(ls);  /* for the find in device_open() */
684 
685 	/* FIXME: AUTOFREE: if this ls is no longer used do
686 	   device_remove_lockspace() */
687 
688 	sigprocmask(SIG_SETMASK, &tmpsig, NULL);
689 	recalc_sigpending();
690 
691 	return 0;
692 }
693 
694 static int copy_result_to_user(struct dlm_user_args *ua, int compat,
695 			       uint32_t flags, int mode, int copy_lvb,
696 			       char __user *buf, size_t count)
697 {
698 #ifdef CONFIG_COMPAT
699 	struct dlm_lock_result32 result32;
700 #endif
701 	struct dlm_lock_result result;
702 	void *resultptr;
703 	int error=0;
704 	int len;
705 	int struct_len;
706 
707 	memset(&result, 0, sizeof(struct dlm_lock_result));
708 	result.version[0] = DLM_DEVICE_VERSION_MAJOR;
709 	result.version[1] = DLM_DEVICE_VERSION_MINOR;
710 	result.version[2] = DLM_DEVICE_VERSION_PATCH;
711 	memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
712 	result.user_lksb = ua->user_lksb;
713 
714 	/* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
715 	   in a conversion unless the conversion is successful.  See code
716 	   in dlm_user_convert() for updating ua from ua_tmp.  OpenVMS, though,
717 	   notes that a new blocking AST address and parameter are set even if
718 	   the conversion fails, so maybe we should just do that. */
719 
720 	if (flags & DLM_CB_BAST) {
721 		result.user_astaddr = ua->bastaddr;
722 		result.user_astparam = ua->bastparam;
723 		result.bast_mode = mode;
724 	} else {
725 		result.user_astaddr = ua->castaddr;
726 		result.user_astparam = ua->castparam;
727 	}
728 
729 #ifdef CONFIG_COMPAT
730 	if (compat)
731 		len = sizeof(struct dlm_lock_result32);
732 	else
733 #endif
734 		len = sizeof(struct dlm_lock_result);
735 	struct_len = len;
736 
737 	/* copy lvb to userspace if there is one, it's been updated, and
738 	   the user buffer has space for it */
739 
740 	if (copy_lvb && ua->lksb.sb_lvbptr && count >= len + DLM_USER_LVB_LEN) {
741 		if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
742 				 DLM_USER_LVB_LEN)) {
743 			error = -EFAULT;
744 			goto out;
745 		}
746 
747 		result.lvb_offset = len;
748 		len += DLM_USER_LVB_LEN;
749 	}
750 
751 	result.length = len;
752 	resultptr = &result;
753 #ifdef CONFIG_COMPAT
754 	if (compat) {
755 		compat_output(&result, &result32);
756 		resultptr = &result32;
757 	}
758 #endif
759 
760 	if (copy_to_user(buf, resultptr, struct_len))
761 		error = -EFAULT;
762 	else
763 		error = len;
764  out:
765 	return error;
766 }
767 
768 static int copy_version_to_user(char __user *buf, size_t count)
769 {
770 	struct dlm_device_version ver;
771 
772 	memset(&ver, 0, sizeof(struct dlm_device_version));
773 	ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
774 	ver.version[1] = DLM_DEVICE_VERSION_MINOR;
775 	ver.version[2] = DLM_DEVICE_VERSION_PATCH;
776 
777 	if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
778 		return -EFAULT;
779 	return sizeof(struct dlm_device_version);
780 }
781 
782 /* a read returns a single ast described in a struct dlm_lock_result */
783 
784 static ssize_t device_read(struct file *file, char __user *buf, size_t count,
785 			   loff_t *ppos)
786 {
787 	struct dlm_user_proc *proc = file->private_data;
788 	struct dlm_lkb *lkb;
789 	DECLARE_WAITQUEUE(wait, current);
790 	struct dlm_callback cb;
791 	int rv, resid, copy_lvb = 0;
792 
793 	if (count == sizeof(struct dlm_device_version)) {
794 		rv = copy_version_to_user(buf, count);
795 		return rv;
796 	}
797 
798 	if (!proc) {
799 		log_print("non-version read from control device %zu", count);
800 		return -EINVAL;
801 	}
802 
803 #ifdef CONFIG_COMPAT
804 	if (count < sizeof(struct dlm_lock_result32))
805 #else
806 	if (count < sizeof(struct dlm_lock_result))
807 #endif
808 		return -EINVAL;
809 
810  try_another:
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 	/* if we empty lkb_callbacks, we don't want to unlock the spinlock
843 	   without removing lkb_cb_list; so empty lkb_cb_list is always
844 	   consistent with empty lkb_callbacks */
845 
846 	lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_cb_list);
847 
848 	rv = dlm_rem_lkb_callback(lkb->lkb_resource->res_ls, lkb, &cb, &resid);
849 	if (rv < 0) {
850 		/* this shouldn't happen; lkb should have been removed from
851 		   list when resid was zero */
852 		log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
853 		list_del_init(&lkb->lkb_cb_list);
854 		spin_unlock(&proc->asts_spin);
855 		/* removes ref for proc->asts, may cause lkb to be freed */
856 		dlm_put_lkb(lkb);
857 		goto try_another;
858 	}
859 	if (!resid)
860 		list_del_init(&lkb->lkb_cb_list);
861 	spin_unlock(&proc->asts_spin);
862 
863 	if (cb.flags & DLM_CB_SKIP) {
864 		/* removes ref for proc->asts, may cause lkb to be freed */
865 		if (!resid)
866 			dlm_put_lkb(lkb);
867 		goto try_another;
868 	}
869 
870 	if (cb.flags & DLM_CB_CAST) {
871 		int old_mode, new_mode;
872 
873 		old_mode = lkb->lkb_last_cast.mode;
874 		new_mode = cb.mode;
875 
876 		if (!cb.sb_status && lkb->lkb_lksb->sb_lvbptr &&
877 		    dlm_lvb_operations[old_mode + 1][new_mode + 1])
878 			copy_lvb = 1;
879 
880 		lkb->lkb_lksb->sb_status = cb.sb_status;
881 		lkb->lkb_lksb->sb_flags = cb.sb_flags;
882 	}
883 
884 	rv = copy_result_to_user(lkb->lkb_ua,
885 				 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
886 				 cb.flags, cb.mode, copy_lvb, buf, count);
887 
888 	/* removes ref for proc->asts, may cause lkb to be freed */
889 	if (!resid)
890 		dlm_put_lkb(lkb);
891 
892 	return rv;
893 }
894 
895 static unsigned int device_poll(struct file *file, poll_table *wait)
896 {
897 	struct dlm_user_proc *proc = file->private_data;
898 
899 	poll_wait(file, &proc->wait, wait);
900 
901 	spin_lock(&proc->asts_spin);
902 	if (!list_empty(&proc->asts)) {
903 		spin_unlock(&proc->asts_spin);
904 		return POLLIN | POLLRDNORM;
905 	}
906 	spin_unlock(&proc->asts_spin);
907 	return 0;
908 }
909 
910 int dlm_user_daemon_available(void)
911 {
912 	/* dlm_controld hasn't started (or, has started, but not
913 	   properly populated configfs) */
914 
915 	if (!dlm_our_nodeid())
916 		return 0;
917 
918 	/* This is to deal with versions of dlm_controld that don't
919 	   know about the monitor device.  We assume that if the
920 	   dlm_controld was started (above), but the monitor device
921 	   was never opened, that it's an old version.  dlm_controld
922 	   should open the monitor device before populating configfs. */
923 
924 	if (dlm_monitor_unused)
925 		return 1;
926 
927 	return atomic_read(&dlm_monitor_opened) ? 1 : 0;
928 }
929 
930 static int ctl_device_open(struct inode *inode, struct file *file)
931 {
932 	file->private_data = NULL;
933 	return 0;
934 }
935 
936 static int ctl_device_close(struct inode *inode, struct file *file)
937 {
938 	return 0;
939 }
940 
941 static int monitor_device_open(struct inode *inode, struct file *file)
942 {
943 	atomic_inc(&dlm_monitor_opened);
944 	dlm_monitor_unused = 0;
945 	return 0;
946 }
947 
948 static int monitor_device_close(struct inode *inode, struct file *file)
949 {
950 	if (atomic_dec_and_test(&dlm_monitor_opened))
951 		dlm_stop_lockspaces();
952 	return 0;
953 }
954 
955 static const struct file_operations device_fops = {
956 	.open    = device_open,
957 	.release = device_close,
958 	.read    = device_read,
959 	.write   = device_write,
960 	.poll    = device_poll,
961 	.owner   = THIS_MODULE,
962 	.llseek  = noop_llseek,
963 };
964 
965 static const struct file_operations ctl_device_fops = {
966 	.open    = ctl_device_open,
967 	.release = ctl_device_close,
968 	.read    = device_read,
969 	.write   = device_write,
970 	.owner   = THIS_MODULE,
971 	.llseek  = noop_llseek,
972 };
973 
974 static struct miscdevice ctl_device = {
975 	.name  = "dlm-control",
976 	.fops  = &ctl_device_fops,
977 	.minor = MISC_DYNAMIC_MINOR,
978 };
979 
980 static const struct file_operations monitor_device_fops = {
981 	.open    = monitor_device_open,
982 	.release = monitor_device_close,
983 	.owner   = THIS_MODULE,
984 	.llseek  = noop_llseek,
985 };
986 
987 static struct miscdevice monitor_device = {
988 	.name  = "dlm-monitor",
989 	.fops  = &monitor_device_fops,
990 	.minor = MISC_DYNAMIC_MINOR,
991 };
992 
993 int __init dlm_user_init(void)
994 {
995 	int error;
996 
997 	atomic_set(&dlm_monitor_opened, 0);
998 
999 	error = misc_register(&ctl_device);
1000 	if (error) {
1001 		log_print("misc_register failed for control device");
1002 		goto out;
1003 	}
1004 
1005 	error = misc_register(&monitor_device);
1006 	if (error) {
1007 		log_print("misc_register failed for monitor device");
1008 		misc_deregister(&ctl_device);
1009 	}
1010  out:
1011 	return error;
1012 }
1013 
1014 void dlm_user_exit(void)
1015 {
1016 	misc_deregister(&ctl_device);
1017 	misc_deregister(&monitor_device);
1018 }
1019 
1020