xref: /linux/fs/dlm/user.c (revision 5499b45190237ca90dd2ac86395cf464fe1f4cc7)
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 
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 			 int namelen)
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, namelen);
100 	} else if (kb->cmd == DLM_USER_PURGE) {
101 		kb->i.purge.nodeid = kb32->i.purge.nodeid;
102 		kb->i.purge.pid = kb32->i.purge.pid;
103 	} else {
104 		kb->i.lock.mode = kb32->i.lock.mode;
105 		kb->i.lock.namelen = kb32->i.lock.namelen;
106 		kb->i.lock.flags = kb32->i.lock.flags;
107 		kb->i.lock.lkid = kb32->i.lock.lkid;
108 		kb->i.lock.parent = kb32->i.lock.parent;
109 		kb->i.lock.xid = kb32->i.lock.xid;
110 		kb->i.lock.timeout = kb32->i.lock.timeout;
111 		kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
112 		kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
113 		kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
114 		kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
115 		kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
116 		memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
117 		memcpy(kb->i.lock.name, kb32->i.lock.name, namelen);
118 	}
119 }
120 
121 static void compat_output(struct dlm_lock_result *res,
122 			  struct dlm_lock_result32 *res32)
123 {
124 	res32->version[0] = res->version[0];
125 	res32->version[1] = res->version[1];
126 	res32->version[2] = res->version[2];
127 
128 	res32->user_astaddr = (__u32)(long)res->user_astaddr;
129 	res32->user_astparam = (__u32)(long)res->user_astparam;
130 	res32->user_lksb = (__u32)(long)res->user_lksb;
131 	res32->bast_mode = res->bast_mode;
132 
133 	res32->lvb_offset = res->lvb_offset;
134 	res32->length = res->length;
135 
136 	res32->lksb.sb_status = res->lksb.sb_status;
137 	res32->lksb.sb_flags = res->lksb.sb_flags;
138 	res32->lksb.sb_lkid = res->lksb.sb_lkid;
139 	res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
140 }
141 #endif
142 
143 /* Figure out if this lock is at the end of its life and no longer
144    available for the application to use.  The lkb still exists until
145    the final ast is read.  A lock becomes EOL in three situations:
146      1. a noqueue request fails with EAGAIN
147      2. an unlock completes with EUNLOCK
148      3. a cancel of a waiting request completes with ECANCEL/EDEADLK
149    An EOL lock needs to be removed from the process's list of locks.
150    And we can't allow any new operation on an EOL lock.  This is
151    not related to the lifetime of the lkb struct which is managed
152    entirely by refcount. */
153 
154 static int lkb_is_endoflife(struct dlm_lkb *lkb, int sb_status, int type)
155 {
156 	switch (sb_status) {
157 	case -DLM_EUNLOCK:
158 		return 1;
159 	case -DLM_ECANCEL:
160 	case -ETIMEDOUT:
161 	case -EDEADLK:
162 		if (lkb->lkb_grmode == DLM_LOCK_IV)
163 			return 1;
164 		break;
165 	case -EAGAIN:
166 		if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV)
167 			return 1;
168 		break;
169 	}
170 	return 0;
171 }
172 
173 /* we could possibly check if the cancel of an orphan has resulted in the lkb
174    being removed and then remove that lkb from the orphans list and free it */
175 
176 void dlm_user_add_ast(struct dlm_lkb *lkb, int type, int mode)
177 {
178 	struct dlm_ls *ls;
179 	struct dlm_user_args *ua;
180 	struct dlm_user_proc *proc;
181 	int eol = 0, ast_type;
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 (type == AST_BAST && ua->bastaddr == NULL)
203 		goto out;
204 
205 	spin_lock(&proc->asts_spin);
206 
207 	ast_type = lkb->lkb_ast_type;
208 	lkb->lkb_ast_type |= type;
209 	if (type == AST_BAST)
210 		lkb->lkb_bastmode = mode;
211 	else
212 		lkb->lkb_castmode = mode;
213 
214 	if (!ast_type) {
215 		kref_get(&lkb->lkb_ref);
216 		list_add_tail(&lkb->lkb_astqueue, &proc->asts);
217 		wake_up_interruptible(&proc->wait);
218 	}
219 	if (type == AST_COMP && (ast_type & AST_COMP))
220 		log_debug(ls, "ast overlap %x status %x %x",
221 			  lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags);
222 
223 	eol = lkb_is_endoflife(lkb, ua->lksb.sb_status, type);
224 	if (eol) {
225 		lkb->lkb_ast_type &= ~AST_BAST;
226 		lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
227 	}
228 
229 	/* We want to copy the lvb to userspace when the completion
230 	   ast is read if the status is 0, the lock has an lvb and
231 	   lvb_ops says we should.  We could probably have set_lvb_lock()
232 	   set update_user_lvb instead and not need old_mode */
233 
234 	if ((lkb->lkb_ast_type & AST_COMP) &&
235 	    (lkb->lkb_lksb->sb_status == 0) &&
236 	    lkb->lkb_lksb->sb_lvbptr &&
237 	    dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1])
238 		ua->update_user_lvb = 1;
239 	else
240 		ua->update_user_lvb = 0;
241 
242 	spin_unlock(&proc->asts_spin);
243 
244 	if (eol) {
245 		spin_lock(&proc->locks_spin);
246 		if (!list_empty(&lkb->lkb_ownqueue)) {
247 			list_del_init(&lkb->lkb_ownqueue);
248 			dlm_put_lkb(lkb);
249 		}
250 		spin_unlock(&proc->locks_spin);
251 	}
252  out:
253 	mutex_unlock(&ls->ls_clear_proc_locks);
254 }
255 
256 static int device_user_lock(struct dlm_user_proc *proc,
257 			    struct dlm_lock_params *params)
258 {
259 	struct dlm_ls *ls;
260 	struct dlm_user_args *ua;
261 	int error = -ENOMEM;
262 
263 	ls = dlm_find_lockspace_local(proc->lockspace);
264 	if (!ls)
265 		return -ENOENT;
266 
267 	if (!params->castaddr || !params->lksb) {
268 		error = -EINVAL;
269 		goto out;
270 	}
271 
272 	ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
273 	if (!ua)
274 		goto out;
275 	ua->proc = proc;
276 	ua->user_lksb = params->lksb;
277 	ua->castparam = params->castparam;
278 	ua->castaddr = params->castaddr;
279 	ua->bastparam = params->bastparam;
280 	ua->bastaddr = params->bastaddr;
281 	ua->xid = params->xid;
282 
283 	if (params->flags & DLM_LKF_CONVERT)
284 		error = dlm_user_convert(ls, ua,
285 				         params->mode, params->flags,
286 				         params->lkid, params->lvb,
287 					 (unsigned long) params->timeout);
288 	else {
289 		error = dlm_user_request(ls, ua,
290 					 params->mode, params->flags,
291 					 params->name, params->namelen,
292 					 (unsigned long) params->timeout);
293 		if (!error)
294 			error = ua->lksb.sb_lkid;
295 	}
296  out:
297 	dlm_put_lockspace(ls);
298 	return error;
299 }
300 
301 static int device_user_unlock(struct dlm_user_proc *proc,
302 			      struct dlm_lock_params *params)
303 {
304 	struct dlm_ls *ls;
305 	struct dlm_user_args *ua;
306 	int error = -ENOMEM;
307 
308 	ls = dlm_find_lockspace_local(proc->lockspace);
309 	if (!ls)
310 		return -ENOENT;
311 
312 	ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
313 	if (!ua)
314 		goto out;
315 	ua->proc = proc;
316 	ua->user_lksb = params->lksb;
317 	ua->castparam = params->castparam;
318 	ua->castaddr = params->castaddr;
319 
320 	if (params->flags & DLM_LKF_CANCEL)
321 		error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
322 	else
323 		error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
324 					params->lvb);
325  out:
326 	dlm_put_lockspace(ls);
327 	return error;
328 }
329 
330 static int device_user_deadlock(struct dlm_user_proc *proc,
331 				struct dlm_lock_params *params)
332 {
333 	struct dlm_ls *ls;
334 	int error;
335 
336 	ls = dlm_find_lockspace_local(proc->lockspace);
337 	if (!ls)
338 		return -ENOENT;
339 
340 	error = dlm_user_deadlock(ls, params->flags, params->lkid);
341 
342 	dlm_put_lockspace(ls);
343 	return error;
344 }
345 
346 static int dlm_device_register(struct dlm_ls *ls, char *name)
347 {
348 	int error, len;
349 
350 	/* The device is already registered.  This happens when the
351 	   lockspace is created multiple times from userspace. */
352 	if (ls->ls_device.name)
353 		return 0;
354 
355 	error = -ENOMEM;
356 	len = strlen(name) + strlen(name_prefix) + 2;
357 	ls->ls_device.name = kzalloc(len, GFP_NOFS);
358 	if (!ls->ls_device.name)
359 		goto fail;
360 
361 	snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
362 		 name);
363 	ls->ls_device.fops = &device_fops;
364 	ls->ls_device.minor = MISC_DYNAMIC_MINOR;
365 
366 	error = misc_register(&ls->ls_device);
367 	if (error) {
368 		kfree(ls->ls_device.name);
369 	}
370 fail:
371 	return error;
372 }
373 
374 int dlm_device_deregister(struct dlm_ls *ls)
375 {
376 	int error;
377 
378 	/* The device is not registered.  This happens when the lockspace
379 	   was never used from userspace, or when device_create_lockspace()
380 	   calls dlm_release_lockspace() after the register fails. */
381 	if (!ls->ls_device.name)
382 		return 0;
383 
384 	error = misc_deregister(&ls->ls_device);
385 	if (!error)
386 		kfree(ls->ls_device.name);
387 	return error;
388 }
389 
390 static int device_user_purge(struct dlm_user_proc *proc,
391 			     struct dlm_purge_params *params)
392 {
393 	struct dlm_ls *ls;
394 	int error;
395 
396 	ls = dlm_find_lockspace_local(proc->lockspace);
397 	if (!ls)
398 		return -ENOENT;
399 
400 	error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
401 
402 	dlm_put_lockspace(ls);
403 	return error;
404 }
405 
406 static int device_create_lockspace(struct dlm_lspace_params *params)
407 {
408 	dlm_lockspace_t *lockspace;
409 	struct dlm_ls *ls;
410 	int error;
411 
412 	if (!capable(CAP_SYS_ADMIN))
413 		return -EPERM;
414 
415 	error = dlm_new_lockspace(params->name, strlen(params->name),
416 				  &lockspace, params->flags, DLM_USER_LVB_LEN);
417 	if (error)
418 		return error;
419 
420 	ls = dlm_find_lockspace_local(lockspace);
421 	if (!ls)
422 		return -ENOENT;
423 
424 	error = dlm_device_register(ls, params->name);
425 	dlm_put_lockspace(ls);
426 
427 	if (error)
428 		dlm_release_lockspace(lockspace, 0);
429 	else
430 		error = ls->ls_device.minor;
431 
432 	return error;
433 }
434 
435 static int device_remove_lockspace(struct dlm_lspace_params *params)
436 {
437 	dlm_lockspace_t *lockspace;
438 	struct dlm_ls *ls;
439 	int error, force = 0;
440 
441 	if (!capable(CAP_SYS_ADMIN))
442 		return -EPERM;
443 
444 	ls = dlm_find_lockspace_device(params->minor);
445 	if (!ls)
446 		return -ENOENT;
447 
448 	if (params->flags & DLM_USER_LSFLG_FORCEFREE)
449 		force = 2;
450 
451 	lockspace = ls->ls_local_handle;
452 	dlm_put_lockspace(ls);
453 
454 	/* The final dlm_release_lockspace waits for references to go to
455 	   zero, so all processes will need to close their device for the
456 	   ls before the release will proceed.  release also calls the
457 	   device_deregister above.  Converting a positive return value
458 	   from release to zero means that userspace won't know when its
459 	   release was the final one, but it shouldn't need to know. */
460 
461 	error = dlm_release_lockspace(lockspace, force);
462 	if (error > 0)
463 		error = 0;
464 	return error;
465 }
466 
467 /* Check the user's version matches ours */
468 static int check_version(struct dlm_write_request *req)
469 {
470 	if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
471 	    (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
472 	     req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
473 
474 		printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
475 		       "user (%d.%d.%d) kernel (%d.%d.%d)\n",
476 		       current->comm,
477 		       task_pid_nr(current),
478 		       req->version[0],
479 		       req->version[1],
480 		       req->version[2],
481 		       DLM_DEVICE_VERSION_MAJOR,
482 		       DLM_DEVICE_VERSION_MINOR,
483 		       DLM_DEVICE_VERSION_PATCH);
484 		return -EINVAL;
485 	}
486 	return 0;
487 }
488 
489 /*
490  * device_write
491  *
492  *   device_user_lock
493  *     dlm_user_request -> request_lock
494  *     dlm_user_convert -> convert_lock
495  *
496  *   device_user_unlock
497  *     dlm_user_unlock -> unlock_lock
498  *     dlm_user_cancel -> cancel_lock
499  *
500  *   device_create_lockspace
501  *     dlm_new_lockspace
502  *
503  *   device_remove_lockspace
504  *     dlm_release_lockspace
505  */
506 
507 /* a write to a lockspace device is a lock or unlock request, a write
508    to the control device is to create/remove a lockspace */
509 
510 static ssize_t device_write(struct file *file, const char __user *buf,
511 			    size_t count, loff_t *ppos)
512 {
513 	struct dlm_user_proc *proc = file->private_data;
514 	struct dlm_write_request *kbuf;
515 	sigset_t tmpsig, allsigs;
516 	int error;
517 
518 #ifdef CONFIG_COMPAT
519 	if (count < sizeof(struct dlm_write_request32))
520 #else
521 	if (count < sizeof(struct dlm_write_request))
522 #endif
523 		return -EINVAL;
524 
525 	kbuf = kzalloc(count + 1, GFP_NOFS);
526 	if (!kbuf)
527 		return -ENOMEM;
528 
529 	if (copy_from_user(kbuf, buf, count)) {
530 		error = -EFAULT;
531 		goto out_free;
532 	}
533 
534 	if (check_version(kbuf)) {
535 		error = -EBADE;
536 		goto out_free;
537 	}
538 
539 #ifdef CONFIG_COMPAT
540 	if (!kbuf->is64bit) {
541 		struct dlm_write_request32 *k32buf;
542 		int namelen = 0;
543 
544 		if (count > sizeof(struct dlm_write_request32))
545 			namelen = count - sizeof(struct dlm_write_request32);
546 
547 		k32buf = (struct dlm_write_request32 *)kbuf;
548 
549 		/* add 1 after namelen so that the name string is terminated */
550 		kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
551 			       GFP_NOFS);
552 		if (!kbuf) {
553 			kfree(k32buf);
554 			return -ENOMEM;
555 		}
556 
557 		if (proc)
558 			set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
559 
560 		compat_input(kbuf, k32buf, namelen);
561 		kfree(k32buf);
562 	}
563 #endif
564 
565 	/* do we really need this? can a write happen after a close? */
566 	if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
567 	    (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
568 		error = -EINVAL;
569 		goto out_free;
570 	}
571 
572 	sigfillset(&allsigs);
573 	sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
574 
575 	error = -EINVAL;
576 
577 	switch (kbuf->cmd)
578 	{
579 	case DLM_USER_LOCK:
580 		if (!proc) {
581 			log_print("no locking on control device");
582 			goto out_sig;
583 		}
584 		error = device_user_lock(proc, &kbuf->i.lock);
585 		break;
586 
587 	case DLM_USER_UNLOCK:
588 		if (!proc) {
589 			log_print("no locking on control device");
590 			goto out_sig;
591 		}
592 		error = device_user_unlock(proc, &kbuf->i.lock);
593 		break;
594 
595 	case DLM_USER_DEADLOCK:
596 		if (!proc) {
597 			log_print("no locking on control device");
598 			goto out_sig;
599 		}
600 		error = device_user_deadlock(proc, &kbuf->i.lock);
601 		break;
602 
603 	case DLM_USER_CREATE_LOCKSPACE:
604 		if (proc) {
605 			log_print("create/remove only on control device");
606 			goto out_sig;
607 		}
608 		error = device_create_lockspace(&kbuf->i.lspace);
609 		break;
610 
611 	case DLM_USER_REMOVE_LOCKSPACE:
612 		if (proc) {
613 			log_print("create/remove only on control device");
614 			goto out_sig;
615 		}
616 		error = device_remove_lockspace(&kbuf->i.lspace);
617 		break;
618 
619 	case DLM_USER_PURGE:
620 		if (!proc) {
621 			log_print("no locking on control device");
622 			goto out_sig;
623 		}
624 		error = device_user_purge(proc, &kbuf->i.purge);
625 		break;
626 
627 	default:
628 		log_print("Unknown command passed to DLM device : %d\n",
629 			  kbuf->cmd);
630 	}
631 
632  out_sig:
633 	sigprocmask(SIG_SETMASK, &tmpsig, NULL);
634 	recalc_sigpending();
635  out_free:
636 	kfree(kbuf);
637 	return error;
638 }
639 
640 /* Every process that opens the lockspace device has its own "proc" structure
641    hanging off the open file that's used to keep track of locks owned by the
642    process and asts that need to be delivered to the process. */
643 
644 static int device_open(struct inode *inode, struct file *file)
645 {
646 	struct dlm_user_proc *proc;
647 	struct dlm_ls *ls;
648 
649 	ls = dlm_find_lockspace_device(iminor(inode));
650 	if (!ls)
651 		return -ENOENT;
652 
653 	proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
654 	if (!proc) {
655 		dlm_put_lockspace(ls);
656 		return -ENOMEM;
657 	}
658 
659 	proc->lockspace = ls->ls_local_handle;
660 	INIT_LIST_HEAD(&proc->asts);
661 	INIT_LIST_HEAD(&proc->locks);
662 	INIT_LIST_HEAD(&proc->unlocking);
663 	spin_lock_init(&proc->asts_spin);
664 	spin_lock_init(&proc->locks_spin);
665 	init_waitqueue_head(&proc->wait);
666 	file->private_data = proc;
667 
668 	return 0;
669 }
670 
671 static int device_close(struct inode *inode, struct file *file)
672 {
673 	struct dlm_user_proc *proc = file->private_data;
674 	struct dlm_ls *ls;
675 	sigset_t tmpsig, allsigs;
676 
677 	ls = dlm_find_lockspace_local(proc->lockspace);
678 	if (!ls)
679 		return -ENOENT;
680 
681 	sigfillset(&allsigs);
682 	sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
683 
684 	set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
685 
686 	dlm_clear_proc_locks(ls, proc);
687 
688 	/* at this point no more lkb's should exist for this lockspace,
689 	   so there's no chance of dlm_user_add_ast() being called and
690 	   looking for lkb->ua->proc */
691 
692 	kfree(proc);
693 	file->private_data = NULL;
694 
695 	dlm_put_lockspace(ls);
696 	dlm_put_lockspace(ls);  /* for the find in device_open() */
697 
698 	/* FIXME: AUTOFREE: if this ls is no longer used do
699 	   device_remove_lockspace() */
700 
701 	sigprocmask(SIG_SETMASK, &tmpsig, NULL);
702 	recalc_sigpending();
703 
704 	return 0;
705 }
706 
707 static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type,
708 			       int bmode, char __user *buf, size_t count)
709 {
710 #ifdef CONFIG_COMPAT
711 	struct dlm_lock_result32 result32;
712 #endif
713 	struct dlm_lock_result result;
714 	void *resultptr;
715 	int error=0;
716 	int len;
717 	int struct_len;
718 
719 	memset(&result, 0, sizeof(struct dlm_lock_result));
720 	result.version[0] = DLM_DEVICE_VERSION_MAJOR;
721 	result.version[1] = DLM_DEVICE_VERSION_MINOR;
722 	result.version[2] = DLM_DEVICE_VERSION_PATCH;
723 	memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
724 	result.user_lksb = ua->user_lksb;
725 
726 	/* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
727 	   in a conversion unless the conversion is successful.  See code
728 	   in dlm_user_convert() for updating ua from ua_tmp.  OpenVMS, though,
729 	   notes that a new blocking AST address and parameter are set even if
730 	   the conversion fails, so maybe we should just do that. */
731 
732 	if (type == AST_BAST) {
733 		result.user_astaddr = ua->bastaddr;
734 		result.user_astparam = ua->bastparam;
735 		result.bast_mode = bmode;
736 	} else {
737 		result.user_astaddr = ua->castaddr;
738 		result.user_astparam = ua->castparam;
739 	}
740 
741 #ifdef CONFIG_COMPAT
742 	if (compat)
743 		len = sizeof(struct dlm_lock_result32);
744 	else
745 #endif
746 		len = sizeof(struct dlm_lock_result);
747 	struct_len = len;
748 
749 	/* copy lvb to userspace if there is one, it's been updated, and
750 	   the user buffer has space for it */
751 
752 	if (ua->update_user_lvb && ua->lksb.sb_lvbptr &&
753 	    count >= len + DLM_USER_LVB_LEN) {
754 		if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
755 				 DLM_USER_LVB_LEN)) {
756 			error = -EFAULT;
757 			goto out;
758 		}
759 
760 		result.lvb_offset = len;
761 		len += DLM_USER_LVB_LEN;
762 	}
763 
764 	result.length = len;
765 	resultptr = &result;
766 #ifdef CONFIG_COMPAT
767 	if (compat) {
768 		compat_output(&result, &result32);
769 		resultptr = &result32;
770 	}
771 #endif
772 
773 	if (copy_to_user(buf, resultptr, struct_len))
774 		error = -EFAULT;
775 	else
776 		error = len;
777  out:
778 	return error;
779 }
780 
781 static int copy_version_to_user(char __user *buf, size_t count)
782 {
783 	struct dlm_device_version ver;
784 
785 	memset(&ver, 0, sizeof(struct dlm_device_version));
786 	ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
787 	ver.version[1] = DLM_DEVICE_VERSION_MINOR;
788 	ver.version[2] = DLM_DEVICE_VERSION_PATCH;
789 
790 	if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
791 		return -EFAULT;
792 	return sizeof(struct dlm_device_version);
793 }
794 
795 /* a read returns a single ast described in a struct dlm_lock_result */
796 
797 static ssize_t device_read(struct file *file, char __user *buf, size_t count,
798 			   loff_t *ppos)
799 {
800 	struct dlm_user_proc *proc = file->private_data;
801 	struct dlm_lkb *lkb;
802 	DECLARE_WAITQUEUE(wait, current);
803 	int error, type=0, bmode=0, removed = 0;
804 
805 	if (count == sizeof(struct dlm_device_version)) {
806 		error = copy_version_to_user(buf, count);
807 		return error;
808 	}
809 
810 	if (!proc) {
811 		log_print("non-version read from control device %zu", count);
812 		return -EINVAL;
813 	}
814 
815 #ifdef CONFIG_COMPAT
816 	if (count < sizeof(struct dlm_lock_result32))
817 #else
818 	if (count < sizeof(struct dlm_lock_result))
819 #endif
820 		return -EINVAL;
821 
822 	/* do we really need this? can a read happen after a close? */
823 	if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
824 		return -EINVAL;
825 
826 	spin_lock(&proc->asts_spin);
827 	if (list_empty(&proc->asts)) {
828 		if (file->f_flags & O_NONBLOCK) {
829 			spin_unlock(&proc->asts_spin);
830 			return -EAGAIN;
831 		}
832 
833 		add_wait_queue(&proc->wait, &wait);
834 
835 	repeat:
836 		set_current_state(TASK_INTERRUPTIBLE);
837 		if (list_empty(&proc->asts) && !signal_pending(current)) {
838 			spin_unlock(&proc->asts_spin);
839 			schedule();
840 			spin_lock(&proc->asts_spin);
841 			goto repeat;
842 		}
843 		set_current_state(TASK_RUNNING);
844 		remove_wait_queue(&proc->wait, &wait);
845 
846 		if (signal_pending(current)) {
847 			spin_unlock(&proc->asts_spin);
848 			return -ERESTARTSYS;
849 		}
850 	}
851 
852 	/* there may be both completion and blocking asts to return for
853 	   the lkb, don't remove lkb from asts list unless no asts remain */
854 
855 	lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue);
856 
857 	if (lkb->lkb_ast_type & AST_COMP) {
858 		lkb->lkb_ast_type &= ~AST_COMP;
859 		type = AST_COMP;
860 	} else if (lkb->lkb_ast_type & AST_BAST) {
861 		lkb->lkb_ast_type &= ~AST_BAST;
862 		type = AST_BAST;
863 		bmode = lkb->lkb_bastmode;
864 	}
865 
866 	if (!lkb->lkb_ast_type) {
867 		list_del(&lkb->lkb_astqueue);
868 		removed = 1;
869 	}
870 	spin_unlock(&proc->asts_spin);
871 
872 	error = copy_result_to_user(lkb->lkb_ua,
873 			 	test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
874 				type, bmode, buf, count);
875 
876 	/* removes reference for the proc->asts lists added by
877 	   dlm_user_add_ast() and may result in the lkb being freed */
878 	if (removed)
879 		dlm_put_lkb(lkb);
880 
881 	return error;
882 }
883 
884 static unsigned int device_poll(struct file *file, poll_table *wait)
885 {
886 	struct dlm_user_proc *proc = file->private_data;
887 
888 	poll_wait(file, &proc->wait, wait);
889 
890 	spin_lock(&proc->asts_spin);
891 	if (!list_empty(&proc->asts)) {
892 		spin_unlock(&proc->asts_spin);
893 		return POLLIN | POLLRDNORM;
894 	}
895 	spin_unlock(&proc->asts_spin);
896 	return 0;
897 }
898 
899 int dlm_user_daemon_available(void)
900 {
901 	/* dlm_controld hasn't started (or, has started, but not
902 	   properly populated configfs) */
903 
904 	if (!dlm_our_nodeid())
905 		return 0;
906 
907 	/* This is to deal with versions of dlm_controld that don't
908 	   know about the monitor device.  We assume that if the
909 	   dlm_controld was started (above), but the monitor device
910 	   was never opened, that it's an old version.  dlm_controld
911 	   should open the monitor device before populating configfs. */
912 
913 	if (dlm_monitor_unused)
914 		return 1;
915 
916 	return atomic_read(&dlm_monitor_opened) ? 1 : 0;
917 }
918 
919 static int ctl_device_open(struct inode *inode, struct file *file)
920 {
921 	file->private_data = NULL;
922 	return 0;
923 }
924 
925 static int ctl_device_close(struct inode *inode, struct file *file)
926 {
927 	return 0;
928 }
929 
930 static int monitor_device_open(struct inode *inode, struct file *file)
931 {
932 	atomic_inc(&dlm_monitor_opened);
933 	dlm_monitor_unused = 0;
934 	return 0;
935 }
936 
937 static int monitor_device_close(struct inode *inode, struct file *file)
938 {
939 	if (atomic_dec_and_test(&dlm_monitor_opened))
940 		dlm_stop_lockspaces();
941 	return 0;
942 }
943 
944 static const struct file_operations device_fops = {
945 	.open    = device_open,
946 	.release = device_close,
947 	.read    = device_read,
948 	.write   = device_write,
949 	.poll    = device_poll,
950 	.owner   = THIS_MODULE,
951 };
952 
953 static const struct file_operations ctl_device_fops = {
954 	.open    = ctl_device_open,
955 	.release = ctl_device_close,
956 	.read    = device_read,
957 	.write   = device_write,
958 	.owner   = THIS_MODULE,
959 };
960 
961 static struct miscdevice ctl_device = {
962 	.name  = "dlm-control",
963 	.fops  = &ctl_device_fops,
964 	.minor = MISC_DYNAMIC_MINOR,
965 };
966 
967 static const struct file_operations monitor_device_fops = {
968 	.open    = monitor_device_open,
969 	.release = monitor_device_close,
970 	.owner   = THIS_MODULE,
971 };
972 
973 static struct miscdevice monitor_device = {
974 	.name  = "dlm-monitor",
975 	.fops  = &monitor_device_fops,
976 	.minor = MISC_DYNAMIC_MINOR,
977 };
978 
979 int __init dlm_user_init(void)
980 {
981 	int error;
982 
983 	atomic_set(&dlm_monitor_opened, 0);
984 
985 	error = misc_register(&ctl_device);
986 	if (error) {
987 		log_print("misc_register failed for control device");
988 		goto out;
989 	}
990 
991 	error = misc_register(&monitor_device);
992 	if (error) {
993 		log_print("misc_register failed for monitor device");
994 		misc_deregister(&ctl_device);
995 	}
996  out:
997 	return error;
998 }
999 
1000 void dlm_user_exit(void)
1001 {
1002 	misc_deregister(&ctl_device);
1003 	misc_deregister(&monitor_device);
1004 }
1005 
1006