xref: /openbmc/linux/fs/dlm/user.c (revision 089a49b6)
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 	int error;
497 
498 #ifdef CONFIG_COMPAT
499 	if (count < sizeof(struct dlm_write_request32))
500 #else
501 	if (count < sizeof(struct dlm_write_request))
502 #endif
503 		return -EINVAL;
504 
505 	/*
506 	 * can't compare against COMPAT/dlm_write_request32 because
507 	 * we don't yet know if is64bit is zero
508 	 */
509 	if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
510 		return -EINVAL;
511 
512 	kbuf = kzalloc(count + 1, GFP_NOFS);
513 	if (!kbuf)
514 		return -ENOMEM;
515 
516 	if (copy_from_user(kbuf, buf, count)) {
517 		error = -EFAULT;
518 		goto out_free;
519 	}
520 
521 	if (check_version(kbuf)) {
522 		error = -EBADE;
523 		goto out_free;
524 	}
525 
526 #ifdef CONFIG_COMPAT
527 	if (!kbuf->is64bit) {
528 		struct dlm_write_request32 *k32buf;
529 		int namelen = 0;
530 
531 		if (count > sizeof(struct dlm_write_request32))
532 			namelen = count - sizeof(struct dlm_write_request32);
533 
534 		k32buf = (struct dlm_write_request32 *)kbuf;
535 
536 		/* add 1 after namelen so that the name string is terminated */
537 		kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
538 			       GFP_NOFS);
539 		if (!kbuf) {
540 			kfree(k32buf);
541 			return -ENOMEM;
542 		}
543 
544 		if (proc)
545 			set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
546 
547 		compat_input(kbuf, k32buf, namelen);
548 		kfree(k32buf);
549 	}
550 #endif
551 
552 	/* do we really need this? can a write happen after a close? */
553 	if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
554 	    (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
555 		error = -EINVAL;
556 		goto out_free;
557 	}
558 
559 	error = -EINVAL;
560 
561 	switch (kbuf->cmd)
562 	{
563 	case DLM_USER_LOCK:
564 		if (!proc) {
565 			log_print("no locking on control device");
566 			goto out_free;
567 		}
568 		error = device_user_lock(proc, &kbuf->i.lock);
569 		break;
570 
571 	case DLM_USER_UNLOCK:
572 		if (!proc) {
573 			log_print("no locking on control device");
574 			goto out_free;
575 		}
576 		error = device_user_unlock(proc, &kbuf->i.lock);
577 		break;
578 
579 	case DLM_USER_DEADLOCK:
580 		if (!proc) {
581 			log_print("no locking on control device");
582 			goto out_free;
583 		}
584 		error = device_user_deadlock(proc, &kbuf->i.lock);
585 		break;
586 
587 	case DLM_USER_CREATE_LOCKSPACE:
588 		if (proc) {
589 			log_print("create/remove only on control device");
590 			goto out_free;
591 		}
592 		error = device_create_lockspace(&kbuf->i.lspace);
593 		break;
594 
595 	case DLM_USER_REMOVE_LOCKSPACE:
596 		if (proc) {
597 			log_print("create/remove only on control device");
598 			goto out_free;
599 		}
600 		error = device_remove_lockspace(&kbuf->i.lspace);
601 		break;
602 
603 	case DLM_USER_PURGE:
604 		if (!proc) {
605 			log_print("no locking on control device");
606 			goto out_free;
607 		}
608 		error = device_user_purge(proc, &kbuf->i.purge);
609 		break;
610 
611 	default:
612 		log_print("Unknown command passed to DLM device : %d\n",
613 			  kbuf->cmd);
614 	}
615 
616  out_free:
617 	kfree(kbuf);
618 	return error;
619 }
620 
621 /* Every process that opens the lockspace device has its own "proc" structure
622    hanging off the open file that's used to keep track of locks owned by the
623    process and asts that need to be delivered to the process. */
624 
625 static int device_open(struct inode *inode, struct file *file)
626 {
627 	struct dlm_user_proc *proc;
628 	struct dlm_ls *ls;
629 
630 	ls = dlm_find_lockspace_device(iminor(inode));
631 	if (!ls)
632 		return -ENOENT;
633 
634 	proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
635 	if (!proc) {
636 		dlm_put_lockspace(ls);
637 		return -ENOMEM;
638 	}
639 
640 	proc->lockspace = ls->ls_local_handle;
641 	INIT_LIST_HEAD(&proc->asts);
642 	INIT_LIST_HEAD(&proc->locks);
643 	INIT_LIST_HEAD(&proc->unlocking);
644 	spin_lock_init(&proc->asts_spin);
645 	spin_lock_init(&proc->locks_spin);
646 	init_waitqueue_head(&proc->wait);
647 	file->private_data = proc;
648 
649 	return 0;
650 }
651 
652 static int device_close(struct inode *inode, struct file *file)
653 {
654 	struct dlm_user_proc *proc = file->private_data;
655 	struct dlm_ls *ls;
656 
657 	ls = dlm_find_lockspace_local(proc->lockspace);
658 	if (!ls)
659 		return -ENOENT;
660 
661 	set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
662 
663 	dlm_clear_proc_locks(ls, proc);
664 
665 	/* at this point no more lkb's should exist for this lockspace,
666 	   so there's no chance of dlm_user_add_ast() being called and
667 	   looking for lkb->ua->proc */
668 
669 	kfree(proc);
670 	file->private_data = NULL;
671 
672 	dlm_put_lockspace(ls);
673 	dlm_put_lockspace(ls);  /* for the find in device_open() */
674 
675 	/* FIXME: AUTOFREE: if this ls is no longer used do
676 	   device_remove_lockspace() */
677 
678 	return 0;
679 }
680 
681 static int copy_result_to_user(struct dlm_user_args *ua, int compat,
682 			       uint32_t flags, int mode, int copy_lvb,
683 			       char __user *buf, size_t count)
684 {
685 #ifdef CONFIG_COMPAT
686 	struct dlm_lock_result32 result32;
687 #endif
688 	struct dlm_lock_result result;
689 	void *resultptr;
690 	int error=0;
691 	int len;
692 	int struct_len;
693 
694 	memset(&result, 0, sizeof(struct dlm_lock_result));
695 	result.version[0] = DLM_DEVICE_VERSION_MAJOR;
696 	result.version[1] = DLM_DEVICE_VERSION_MINOR;
697 	result.version[2] = DLM_DEVICE_VERSION_PATCH;
698 	memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
699 	result.user_lksb = ua->user_lksb;
700 
701 	/* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
702 	   in a conversion unless the conversion is successful.  See code
703 	   in dlm_user_convert() for updating ua from ua_tmp.  OpenVMS, though,
704 	   notes that a new blocking AST address and parameter are set even if
705 	   the conversion fails, so maybe we should just do that. */
706 
707 	if (flags & DLM_CB_BAST) {
708 		result.user_astaddr = ua->bastaddr;
709 		result.user_astparam = ua->bastparam;
710 		result.bast_mode = mode;
711 	} else {
712 		result.user_astaddr = ua->castaddr;
713 		result.user_astparam = ua->castparam;
714 	}
715 
716 #ifdef CONFIG_COMPAT
717 	if (compat)
718 		len = sizeof(struct dlm_lock_result32);
719 	else
720 #endif
721 		len = sizeof(struct dlm_lock_result);
722 	struct_len = len;
723 
724 	/* copy lvb to userspace if there is one, it's been updated, and
725 	   the user buffer has space for it */
726 
727 	if (copy_lvb && ua->lksb.sb_lvbptr && count >= len + DLM_USER_LVB_LEN) {
728 		if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
729 				 DLM_USER_LVB_LEN)) {
730 			error = -EFAULT;
731 			goto out;
732 		}
733 
734 		result.lvb_offset = len;
735 		len += DLM_USER_LVB_LEN;
736 	}
737 
738 	result.length = len;
739 	resultptr = &result;
740 #ifdef CONFIG_COMPAT
741 	if (compat) {
742 		compat_output(&result, &result32);
743 		resultptr = &result32;
744 	}
745 #endif
746 
747 	if (copy_to_user(buf, resultptr, struct_len))
748 		error = -EFAULT;
749 	else
750 		error = len;
751  out:
752 	return error;
753 }
754 
755 static int copy_version_to_user(char __user *buf, size_t count)
756 {
757 	struct dlm_device_version ver;
758 
759 	memset(&ver, 0, sizeof(struct dlm_device_version));
760 	ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
761 	ver.version[1] = DLM_DEVICE_VERSION_MINOR;
762 	ver.version[2] = DLM_DEVICE_VERSION_PATCH;
763 
764 	if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
765 		return -EFAULT;
766 	return sizeof(struct dlm_device_version);
767 }
768 
769 /* a read returns a single ast described in a struct dlm_lock_result */
770 
771 static ssize_t device_read(struct file *file, char __user *buf, size_t count,
772 			   loff_t *ppos)
773 {
774 	struct dlm_user_proc *proc = file->private_data;
775 	struct dlm_lkb *lkb;
776 	DECLARE_WAITQUEUE(wait, current);
777 	struct dlm_callback cb;
778 	int rv, resid, copy_lvb = 0;
779 
780 	if (count == sizeof(struct dlm_device_version)) {
781 		rv = copy_version_to_user(buf, count);
782 		return rv;
783 	}
784 
785 	if (!proc) {
786 		log_print("non-version read from control device %zu", count);
787 		return -EINVAL;
788 	}
789 
790 #ifdef CONFIG_COMPAT
791 	if (count < sizeof(struct dlm_lock_result32))
792 #else
793 	if (count < sizeof(struct dlm_lock_result))
794 #endif
795 		return -EINVAL;
796 
797  try_another:
798 
799 	/* do we really need this? can a read happen after a close? */
800 	if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
801 		return -EINVAL;
802 
803 	spin_lock(&proc->asts_spin);
804 	if (list_empty(&proc->asts)) {
805 		if (file->f_flags & O_NONBLOCK) {
806 			spin_unlock(&proc->asts_spin);
807 			return -EAGAIN;
808 		}
809 
810 		add_wait_queue(&proc->wait, &wait);
811 
812 	repeat:
813 		set_current_state(TASK_INTERRUPTIBLE);
814 		if (list_empty(&proc->asts) && !signal_pending(current)) {
815 			spin_unlock(&proc->asts_spin);
816 			schedule();
817 			spin_lock(&proc->asts_spin);
818 			goto repeat;
819 		}
820 		set_current_state(TASK_RUNNING);
821 		remove_wait_queue(&proc->wait, &wait);
822 
823 		if (signal_pending(current)) {
824 			spin_unlock(&proc->asts_spin);
825 			return -ERESTARTSYS;
826 		}
827 	}
828 
829 	/* if we empty lkb_callbacks, we don't want to unlock the spinlock
830 	   without removing lkb_cb_list; so empty lkb_cb_list is always
831 	   consistent with empty lkb_callbacks */
832 
833 	lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_cb_list);
834 
835 	rv = dlm_rem_lkb_callback(lkb->lkb_resource->res_ls, lkb, &cb, &resid);
836 	if (rv < 0) {
837 		/* this shouldn't happen; lkb should have been removed from
838 		   list when resid was zero */
839 		log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
840 		list_del_init(&lkb->lkb_cb_list);
841 		spin_unlock(&proc->asts_spin);
842 		/* removes ref for proc->asts, may cause lkb to be freed */
843 		dlm_put_lkb(lkb);
844 		goto try_another;
845 	}
846 	if (!resid)
847 		list_del_init(&lkb->lkb_cb_list);
848 	spin_unlock(&proc->asts_spin);
849 
850 	if (cb.flags & DLM_CB_SKIP) {
851 		/* removes ref for proc->asts, may cause lkb to be freed */
852 		if (!resid)
853 			dlm_put_lkb(lkb);
854 		goto try_another;
855 	}
856 
857 	if (cb.flags & DLM_CB_CAST) {
858 		int old_mode, new_mode;
859 
860 		old_mode = lkb->lkb_last_cast.mode;
861 		new_mode = cb.mode;
862 
863 		if (!cb.sb_status && lkb->lkb_lksb->sb_lvbptr &&
864 		    dlm_lvb_operations[old_mode + 1][new_mode + 1])
865 			copy_lvb = 1;
866 
867 		lkb->lkb_lksb->sb_status = cb.sb_status;
868 		lkb->lkb_lksb->sb_flags = cb.sb_flags;
869 	}
870 
871 	rv = copy_result_to_user(lkb->lkb_ua,
872 				 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
873 				 cb.flags, cb.mode, copy_lvb, buf, count);
874 
875 	/* removes ref for proc->asts, may cause lkb to be freed */
876 	if (!resid)
877 		dlm_put_lkb(lkb);
878 
879 	return rv;
880 }
881 
882 static unsigned int device_poll(struct file *file, poll_table *wait)
883 {
884 	struct dlm_user_proc *proc = file->private_data;
885 
886 	poll_wait(file, &proc->wait, wait);
887 
888 	spin_lock(&proc->asts_spin);
889 	if (!list_empty(&proc->asts)) {
890 		spin_unlock(&proc->asts_spin);
891 		return POLLIN | POLLRDNORM;
892 	}
893 	spin_unlock(&proc->asts_spin);
894 	return 0;
895 }
896 
897 int dlm_user_daemon_available(void)
898 {
899 	/* dlm_controld hasn't started (or, has started, but not
900 	   properly populated configfs) */
901 
902 	if (!dlm_our_nodeid())
903 		return 0;
904 
905 	/* This is to deal with versions of dlm_controld that don't
906 	   know about the monitor device.  We assume that if the
907 	   dlm_controld was started (above), but the monitor device
908 	   was never opened, that it's an old version.  dlm_controld
909 	   should open the monitor device before populating configfs. */
910 
911 	if (dlm_monitor_unused)
912 		return 1;
913 
914 	return atomic_read(&dlm_monitor_opened) ? 1 : 0;
915 }
916 
917 static int ctl_device_open(struct inode *inode, struct file *file)
918 {
919 	file->private_data = NULL;
920 	return 0;
921 }
922 
923 static int ctl_device_close(struct inode *inode, struct file *file)
924 {
925 	return 0;
926 }
927 
928 static int monitor_device_open(struct inode *inode, struct file *file)
929 {
930 	atomic_inc(&dlm_monitor_opened);
931 	dlm_monitor_unused = 0;
932 	return 0;
933 }
934 
935 static int monitor_device_close(struct inode *inode, struct file *file)
936 {
937 	if (atomic_dec_and_test(&dlm_monitor_opened))
938 		dlm_stop_lockspaces();
939 	return 0;
940 }
941 
942 static const struct file_operations device_fops = {
943 	.open    = device_open,
944 	.release = device_close,
945 	.read    = device_read,
946 	.write   = device_write,
947 	.poll    = device_poll,
948 	.owner   = THIS_MODULE,
949 	.llseek  = noop_llseek,
950 };
951 
952 static const struct file_operations ctl_device_fops = {
953 	.open    = ctl_device_open,
954 	.release = ctl_device_close,
955 	.read    = device_read,
956 	.write   = device_write,
957 	.owner   = THIS_MODULE,
958 	.llseek  = noop_llseek,
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 	.llseek  = noop_llseek,
972 };
973 
974 static struct miscdevice monitor_device = {
975 	.name  = "dlm-monitor",
976 	.fops  = &monitor_device_fops,
977 	.minor = MISC_DYNAMIC_MINOR,
978 };
979 
980 int __init dlm_user_init(void)
981 {
982 	int error;
983 
984 	atomic_set(&dlm_monitor_opened, 0);
985 
986 	error = misc_register(&ctl_device);
987 	if (error) {
988 		log_print("misc_register failed for control device");
989 		goto out;
990 	}
991 
992 	error = misc_register(&monitor_device);
993 	if (error) {
994 		log_print("misc_register failed for monitor device");
995 		misc_deregister(&ctl_device);
996 	}
997  out:
998 	return error;
999 }
1000 
1001 void dlm_user_exit(void)
1002 {
1003 	misc_deregister(&ctl_device);
1004 	misc_deregister(&monitor_device);
1005 }
1006 
1007