xref: /linux/fs/dlm/user.c (revision 1e0731c05c985deb68a97fa44c1adcd3305dda90)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2006-2010 Red Hat, Inc.  All rights reserved.
4  */
5 
6 #include <linux/miscdevice.h>
7 #include <linux/init.h>
8 #include <linux/wait.h>
9 #include <linux/file.h>
10 #include <linux/fs.h>
11 #include <linux/poll.h>
12 #include <linux/signal.h>
13 #include <linux/spinlock.h>
14 #include <linux/dlm.h>
15 #include <linux/dlm_device.h>
16 #include <linux/slab.h>
17 #include <linux/sched/signal.h>
18 
19 #include <trace/events/dlm.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 #include "ast.h"
27 #include "config.h"
28 #include "memory.h"
29 
30 static const char name_prefix[] = "dlm";
31 static const struct file_operations device_fops;
32 static atomic_t dlm_monitor_opened;
33 static int dlm_monitor_unused = 1;
34 
35 #ifdef CONFIG_COMPAT
36 
37 struct dlm_lock_params32 {
38 	__u8 mode;
39 	__u8 namelen;
40 	__u16 unused;
41 	__u32 flags;
42 	__u32 lkid;
43 	__u32 parent;
44 	__u64 xid;
45 	__u64 timeout;
46 	__u32 castparam;
47 	__u32 castaddr;
48 	__u32 bastparam;
49 	__u32 bastaddr;
50 	__u32 lksb;
51 	char lvb[DLM_USER_LVB_LEN];
52 	char name[];
53 };
54 
55 struct dlm_write_request32 {
56 	__u32 version[3];
57 	__u8 cmd;
58 	__u8 is64bit;
59 	__u8 unused[2];
60 
61 	union  {
62 		struct dlm_lock_params32 lock;
63 		struct dlm_lspace_params lspace;
64 		struct dlm_purge_params purge;
65 	} i;
66 };
67 
68 struct dlm_lksb32 {
69 	__u32 sb_status;
70 	__u32 sb_lkid;
71 	__u8 sb_flags;
72 	__u32 sb_lvbptr;
73 };
74 
75 struct dlm_lock_result32 {
76 	__u32 version[3];
77 	__u32 length;
78 	__u32 user_astaddr;
79 	__u32 user_astparam;
80 	__u32 user_lksb;
81 	struct dlm_lksb32 lksb;
82 	__u8 bast_mode;
83 	__u8 unused[3];
84 	/* Offsets may be zero if no data is present */
85 	__u32 lvb_offset;
86 };
87 
88 static void compat_input(struct dlm_write_request *kb,
89 			 struct dlm_write_request32 *kb32,
90 			 int namelen)
91 {
92 	kb->version[0] = kb32->version[0];
93 	kb->version[1] = kb32->version[1];
94 	kb->version[2] = kb32->version[2];
95 
96 	kb->cmd = kb32->cmd;
97 	kb->is64bit = kb32->is64bit;
98 	if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
99 	    kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
100 		kb->i.lspace.flags = kb32->i.lspace.flags;
101 		kb->i.lspace.minor = kb32->i.lspace.minor;
102 		memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen);
103 	} else if (kb->cmd == DLM_USER_PURGE) {
104 		kb->i.purge.nodeid = kb32->i.purge.nodeid;
105 		kb->i.purge.pid = kb32->i.purge.pid;
106 	} else {
107 		kb->i.lock.mode = kb32->i.lock.mode;
108 		kb->i.lock.namelen = kb32->i.lock.namelen;
109 		kb->i.lock.flags = kb32->i.lock.flags;
110 		kb->i.lock.lkid = kb32->i.lock.lkid;
111 		kb->i.lock.parent = kb32->i.lock.parent;
112 		kb->i.lock.xid = kb32->i.lock.xid;
113 		kb->i.lock.timeout = kb32->i.lock.timeout;
114 		kb->i.lock.castparam = (__user void *)(long)kb32->i.lock.castparam;
115 		kb->i.lock.castaddr = (__user void *)(long)kb32->i.lock.castaddr;
116 		kb->i.lock.bastparam = (__user void *)(long)kb32->i.lock.bastparam;
117 		kb->i.lock.bastaddr = (__user void *)(long)kb32->i.lock.bastaddr;
118 		kb->i.lock.lksb = (__user void *)(long)kb32->i.lock.lksb;
119 		memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
120 		memcpy(kb->i.lock.name, kb32->i.lock.name, namelen);
121 	}
122 }
123 
124 static void compat_output(struct dlm_lock_result *res,
125 			  struct dlm_lock_result32 *res32)
126 {
127 	memset(res32, 0, sizeof(*res32));
128 
129 	res32->version[0] = res->version[0];
130 	res32->version[1] = res->version[1];
131 	res32->version[2] = res->version[2];
132 
133 	res32->user_astaddr = (__u32)(__force long)res->user_astaddr;
134 	res32->user_astparam = (__u32)(__force long)res->user_astparam;
135 	res32->user_lksb = (__u32)(__force long)res->user_lksb;
136 	res32->bast_mode = res->bast_mode;
137 
138 	res32->lvb_offset = res->lvb_offset;
139 	res32->length = res->length;
140 
141 	res32->lksb.sb_status = res->lksb.sb_status;
142 	res32->lksb.sb_flags = res->lksb.sb_flags;
143 	res32->lksb.sb_lkid = res->lksb.sb_lkid;
144 	res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
145 }
146 #endif
147 
148 /* should held proc->asts_spin lock */
149 void dlm_purge_lkb_callbacks(struct dlm_lkb *lkb)
150 {
151 	struct dlm_callback *cb, *safe;
152 
153 	list_for_each_entry_safe(cb, safe, &lkb->lkb_callbacks, list) {
154 		list_del(&cb->list);
155 		kref_put(&cb->ref, dlm_release_callback);
156 	}
157 
158 	clear_bit(DLM_IFL_CB_PENDING_BIT, &lkb->lkb_iflags);
159 
160 	/* invalidate */
161 	dlm_callback_set_last_ptr(&lkb->lkb_last_cast, NULL);
162 	dlm_callback_set_last_ptr(&lkb->lkb_last_cb, NULL);
163 	lkb->lkb_last_bast_mode = -1;
164 }
165 
166 /* Figure out if this lock is at the end of its life and no longer
167    available for the application to use.  The lkb still exists until
168    the final ast is read.  A lock becomes EOL in three situations:
169      1. a noqueue request fails with EAGAIN
170      2. an unlock completes with EUNLOCK
171      3. a cancel of a waiting request completes with ECANCEL/EDEADLK
172    An EOL lock needs to be removed from the process's list of locks.
173    And we can't allow any new operation on an EOL lock.  This is
174    not related to the lifetime of the lkb struct which is managed
175    entirely by refcount. */
176 
177 static int lkb_is_endoflife(int mode, int status)
178 {
179 	switch (status) {
180 	case -DLM_EUNLOCK:
181 		return 1;
182 	case -DLM_ECANCEL:
183 	case -ETIMEDOUT:
184 	case -EDEADLK:
185 	case -EAGAIN:
186 		if (mode == DLM_LOCK_IV)
187 			return 1;
188 		break;
189 	}
190 	return 0;
191 }
192 
193 /* we could possibly check if the cancel of an orphan has resulted in the lkb
194    being removed and then remove that lkb from the orphans list and free it */
195 
196 void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
197 		      int status, uint32_t sbflags)
198 {
199 	struct dlm_ls *ls;
200 	struct dlm_user_args *ua;
201 	struct dlm_user_proc *proc;
202 	int rv;
203 
204 	if (test_bit(DLM_DFL_ORPHAN_BIT, &lkb->lkb_dflags) ||
205 	    test_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags))
206 		return;
207 
208 	ls = lkb->lkb_resource->res_ls;
209 	spin_lock(&ls->ls_clear_proc_locks);
210 
211 	/* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
212 	   can't be delivered.  For ORPHAN's, dlm_clear_proc_locks() freed
213 	   lkb->ua so we can't try to use it.  This second check is necessary
214 	   for cases where a completion ast is received for an operation that
215 	   began before clear_proc_locks did its cancel/unlock. */
216 
217 	if (test_bit(DLM_DFL_ORPHAN_BIT, &lkb->lkb_dflags) ||
218 	    test_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags))
219 		goto out;
220 
221 	DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
222 	ua = lkb->lkb_ua;
223 	proc = ua->proc;
224 
225 	if ((flags & DLM_CB_BAST) && ua->bastaddr == NULL)
226 		goto out;
227 
228 	if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status))
229 		set_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags);
230 
231 	spin_lock(&proc->asts_spin);
232 
233 	rv = dlm_enqueue_lkb_callback(lkb, flags, mode, status, sbflags);
234 	switch (rv) {
235 	case DLM_ENQUEUE_CALLBACK_FAILURE:
236 		spin_unlock(&proc->asts_spin);
237 		WARN_ON_ONCE(1);
238 		goto out;
239 	case DLM_ENQUEUE_CALLBACK_NEED_SCHED:
240 		kref_get(&lkb->lkb_ref);
241 		list_add_tail(&lkb->lkb_cb_list, &proc->asts);
242 		wake_up_interruptible(&proc->wait);
243 		break;
244 	case DLM_ENQUEUE_CALLBACK_SUCCESS:
245 		break;
246 	default:
247 		WARN_ON_ONCE(1);
248 		break;
249 	}
250 	spin_unlock(&proc->asts_spin);
251 
252 	if (test_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags)) {
253 		/* N.B. spin_lock locks_spin, not asts_spin */
254 		spin_lock(&proc->locks_spin);
255 		if (!list_empty(&lkb->lkb_ownqueue)) {
256 			list_del_init(&lkb->lkb_ownqueue);
257 			dlm_put_lkb(lkb);
258 		}
259 		spin_unlock(&proc->locks_spin);
260 	}
261  out:
262 	spin_unlock(&ls->ls_clear_proc_locks);
263 }
264 
265 static int device_user_lock(struct dlm_user_proc *proc,
266 			    struct dlm_lock_params *params)
267 {
268 	struct dlm_ls *ls;
269 	struct dlm_user_args *ua;
270 	uint32_t lkid;
271 	int error = -ENOMEM;
272 
273 	ls = dlm_find_lockspace_local(proc->lockspace);
274 	if (!ls)
275 		return -ENOENT;
276 
277 	if (!params->castaddr || !params->lksb) {
278 		error = -EINVAL;
279 		goto out;
280 	}
281 
282 	ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
283 	if (!ua)
284 		goto out;
285 	ua->proc = proc;
286 	ua->user_lksb = params->lksb;
287 	ua->castparam = params->castparam;
288 	ua->castaddr = params->castaddr;
289 	ua->bastparam = params->bastparam;
290 	ua->bastaddr = params->bastaddr;
291 	ua->xid = params->xid;
292 
293 	if (params->flags & DLM_LKF_CONVERT) {
294 		error = dlm_user_convert(ls, ua,
295 					 params->mode, params->flags,
296 					 params->lkid, params->lvb);
297 	} else if (params->flags & DLM_LKF_ORPHAN) {
298 		error = dlm_user_adopt_orphan(ls, ua,
299 					 params->mode, params->flags,
300 					 params->name, params->namelen,
301 					 &lkid);
302 		if (!error)
303 			error = lkid;
304 	} else {
305 		error = dlm_user_request(ls, ua,
306 					 params->mode, params->flags,
307 					 params->name, params->namelen);
308 		if (!error)
309 			error = ua->lksb.sb_lkid;
310 	}
311  out:
312 	dlm_put_lockspace(ls);
313 	return error;
314 }
315 
316 static int device_user_unlock(struct dlm_user_proc *proc,
317 			      struct dlm_lock_params *params)
318 {
319 	struct dlm_ls *ls;
320 	struct dlm_user_args *ua;
321 	int error = -ENOMEM;
322 
323 	ls = dlm_find_lockspace_local(proc->lockspace);
324 	if (!ls)
325 		return -ENOENT;
326 
327 	ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
328 	if (!ua)
329 		goto out;
330 	ua->proc = proc;
331 	ua->user_lksb = params->lksb;
332 	ua->castparam = params->castparam;
333 	ua->castaddr = params->castaddr;
334 
335 	if (params->flags & DLM_LKF_CANCEL)
336 		error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
337 	else
338 		error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
339 					params->lvb);
340  out:
341 	dlm_put_lockspace(ls);
342 	return error;
343 }
344 
345 static int device_user_deadlock(struct dlm_user_proc *proc,
346 				struct dlm_lock_params *params)
347 {
348 	struct dlm_ls *ls;
349 	int error;
350 
351 	ls = dlm_find_lockspace_local(proc->lockspace);
352 	if (!ls)
353 		return -ENOENT;
354 
355 	error = dlm_user_deadlock(ls, params->flags, params->lkid);
356 
357 	dlm_put_lockspace(ls);
358 	return error;
359 }
360 
361 static int dlm_device_register(struct dlm_ls *ls, char *name)
362 {
363 	int error, len;
364 
365 	/* The device is already registered.  This happens when the
366 	   lockspace is created multiple times from userspace. */
367 	if (ls->ls_device.name)
368 		return 0;
369 
370 	error = -ENOMEM;
371 	len = strlen(name) + strlen(name_prefix) + 2;
372 	ls->ls_device.name = kzalloc(len, GFP_NOFS);
373 	if (!ls->ls_device.name)
374 		goto fail;
375 
376 	snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
377 		 name);
378 	ls->ls_device.fops = &device_fops;
379 	ls->ls_device.minor = MISC_DYNAMIC_MINOR;
380 
381 	error = misc_register(&ls->ls_device);
382 	if (error) {
383 		kfree(ls->ls_device.name);
384 		/* this has to be set to NULL
385 		 * to avoid a double-free in dlm_device_deregister
386 		 */
387 		ls->ls_device.name = NULL;
388 	}
389 fail:
390 	return error;
391 }
392 
393 int dlm_device_deregister(struct dlm_ls *ls)
394 {
395 	/* The device is not registered.  This happens when the lockspace
396 	   was never used from userspace, or when device_create_lockspace()
397 	   calls dlm_release_lockspace() after the register fails. */
398 	if (!ls->ls_device.name)
399 		return 0;
400 
401 	misc_deregister(&ls->ls_device);
402 	kfree(ls->ls_device.name);
403 	return 0;
404 }
405 
406 static int device_user_purge(struct dlm_user_proc *proc,
407 			     struct dlm_purge_params *params)
408 {
409 	struct dlm_ls *ls;
410 	int error;
411 
412 	ls = dlm_find_lockspace_local(proc->lockspace);
413 	if (!ls)
414 		return -ENOENT;
415 
416 	error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
417 
418 	dlm_put_lockspace(ls);
419 	return error;
420 }
421 
422 static int device_create_lockspace(struct dlm_lspace_params *params)
423 {
424 	dlm_lockspace_t *lockspace;
425 	struct dlm_ls *ls;
426 	int error;
427 
428 	if (!capable(CAP_SYS_ADMIN))
429 		return -EPERM;
430 
431 	error = dlm_new_user_lockspace(params->name, dlm_config.ci_cluster_name,
432 				       params->flags, DLM_USER_LVB_LEN, NULL,
433 				       NULL, NULL, &lockspace);
434 	if (error)
435 		return error;
436 
437 	ls = dlm_find_lockspace_local(lockspace);
438 	if (!ls)
439 		return -ENOENT;
440 
441 	error = dlm_device_register(ls, params->name);
442 	dlm_put_lockspace(ls);
443 
444 	if (error)
445 		dlm_release_lockspace(lockspace, 0);
446 	else
447 		error = ls->ls_device.minor;
448 
449 	return error;
450 }
451 
452 static int device_remove_lockspace(struct dlm_lspace_params *params)
453 {
454 	dlm_lockspace_t *lockspace;
455 	struct dlm_ls *ls;
456 	int error, force = 0;
457 
458 	if (!capable(CAP_SYS_ADMIN))
459 		return -EPERM;
460 
461 	ls = dlm_find_lockspace_device(params->minor);
462 	if (!ls)
463 		return -ENOENT;
464 
465 	if (params->flags & DLM_USER_LSFLG_FORCEFREE)
466 		force = 2;
467 
468 	lockspace = ls->ls_local_handle;
469 	dlm_put_lockspace(ls);
470 
471 	/* The final dlm_release_lockspace waits for references to go to
472 	   zero, so all processes will need to close their device for the
473 	   ls before the release will proceed.  release also calls the
474 	   device_deregister above.  Converting a positive return value
475 	   from release to zero means that userspace won't know when its
476 	   release was the final one, but it shouldn't need to know. */
477 
478 	error = dlm_release_lockspace(lockspace, force);
479 	if (error > 0)
480 		error = 0;
481 	return error;
482 }
483 
484 /* Check the user's version matches ours */
485 static int check_version(struct dlm_write_request *req)
486 {
487 	if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
488 	    (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
489 	     req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
490 
491 		printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
492 		       "user (%d.%d.%d) kernel (%d.%d.%d)\n",
493 		       current->comm,
494 		       task_pid_nr(current),
495 		       req->version[0],
496 		       req->version[1],
497 		       req->version[2],
498 		       DLM_DEVICE_VERSION_MAJOR,
499 		       DLM_DEVICE_VERSION_MINOR,
500 		       DLM_DEVICE_VERSION_PATCH);
501 		return -EINVAL;
502 	}
503 	return 0;
504 }
505 
506 /*
507  * device_write
508  *
509  *   device_user_lock
510  *     dlm_user_request -> request_lock
511  *     dlm_user_convert -> convert_lock
512  *
513  *   device_user_unlock
514  *     dlm_user_unlock -> unlock_lock
515  *     dlm_user_cancel -> cancel_lock
516  *
517  *   device_create_lockspace
518  *     dlm_new_lockspace
519  *
520  *   device_remove_lockspace
521  *     dlm_release_lockspace
522  */
523 
524 /* a write to a lockspace device is a lock or unlock request, a write
525    to the control device is to create/remove a lockspace */
526 
527 static ssize_t device_write(struct file *file, const char __user *buf,
528 			    size_t count, loff_t *ppos)
529 {
530 	struct dlm_user_proc *proc = file->private_data;
531 	struct dlm_write_request *kbuf;
532 	int error;
533 
534 #ifdef CONFIG_COMPAT
535 	if (count < sizeof(struct dlm_write_request32))
536 #else
537 	if (count < sizeof(struct dlm_write_request))
538 #endif
539 		return -EINVAL;
540 
541 	/*
542 	 * can't compare against COMPAT/dlm_write_request32 because
543 	 * we don't yet know if is64bit is zero
544 	 */
545 	if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
546 		return -EINVAL;
547 
548 	kbuf = memdup_user_nul(buf, count);
549 	if (IS_ERR(kbuf))
550 		return PTR_ERR(kbuf);
551 
552 	if (check_version(kbuf)) {
553 		error = -EBADE;
554 		goto out_free;
555 	}
556 
557 #ifdef CONFIG_COMPAT
558 	if (!kbuf->is64bit) {
559 		struct dlm_write_request32 *k32buf;
560 		int namelen = 0;
561 
562 		if (count > sizeof(struct dlm_write_request32))
563 			namelen = count - sizeof(struct dlm_write_request32);
564 
565 		k32buf = (struct dlm_write_request32 *)kbuf;
566 
567 		/* add 1 after namelen so that the name string is terminated */
568 		kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
569 			       GFP_NOFS);
570 		if (!kbuf) {
571 			kfree(k32buf);
572 			return -ENOMEM;
573 		}
574 
575 		if (proc)
576 			set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
577 
578 		compat_input(kbuf, k32buf, namelen);
579 		kfree(k32buf);
580 	}
581 #endif
582 
583 	/* do we really need this? can a write happen after a close? */
584 	if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
585 	    (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
586 		error = -EINVAL;
587 		goto out_free;
588 	}
589 
590 	error = -EINVAL;
591 
592 	switch (kbuf->cmd)
593 	{
594 	case DLM_USER_LOCK:
595 		if (!proc) {
596 			log_print("no locking on control device");
597 			goto out_free;
598 		}
599 		error = device_user_lock(proc, &kbuf->i.lock);
600 		break;
601 
602 	case DLM_USER_UNLOCK:
603 		if (!proc) {
604 			log_print("no locking on control device");
605 			goto out_free;
606 		}
607 		error = device_user_unlock(proc, &kbuf->i.lock);
608 		break;
609 
610 	case DLM_USER_DEADLOCK:
611 		if (!proc) {
612 			log_print("no locking on control device");
613 			goto out_free;
614 		}
615 		error = device_user_deadlock(proc, &kbuf->i.lock);
616 		break;
617 
618 	case DLM_USER_CREATE_LOCKSPACE:
619 		if (proc) {
620 			log_print("create/remove only on control device");
621 			goto out_free;
622 		}
623 		error = device_create_lockspace(&kbuf->i.lspace);
624 		break;
625 
626 	case DLM_USER_REMOVE_LOCKSPACE:
627 		if (proc) {
628 			log_print("create/remove only on control device");
629 			goto out_free;
630 		}
631 		error = device_remove_lockspace(&kbuf->i.lspace);
632 		break;
633 
634 	case DLM_USER_PURGE:
635 		if (!proc) {
636 			log_print("no locking on control device");
637 			goto out_free;
638 		}
639 		error = device_user_purge(proc, &kbuf->i.purge);
640 		break;
641 
642 	default:
643 		log_print("Unknown command passed to DLM device : %d\n",
644 			  kbuf->cmd);
645 	}
646 
647  out_free:
648 	kfree(kbuf);
649 	return error;
650 }
651 
652 /* Every process that opens the lockspace device has its own "proc" structure
653    hanging off the open file that's used to keep track of locks owned by the
654    process and asts that need to be delivered to the process. */
655 
656 static int device_open(struct inode *inode, struct file *file)
657 {
658 	struct dlm_user_proc *proc;
659 	struct dlm_ls *ls;
660 
661 	ls = dlm_find_lockspace_device(iminor(inode));
662 	if (!ls)
663 		return -ENOENT;
664 
665 	proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
666 	if (!proc) {
667 		dlm_put_lockspace(ls);
668 		return -ENOMEM;
669 	}
670 
671 	proc->lockspace = ls->ls_local_handle;
672 	INIT_LIST_HEAD(&proc->asts);
673 	INIT_LIST_HEAD(&proc->locks);
674 	INIT_LIST_HEAD(&proc->unlocking);
675 	spin_lock_init(&proc->asts_spin);
676 	spin_lock_init(&proc->locks_spin);
677 	init_waitqueue_head(&proc->wait);
678 	file->private_data = proc;
679 
680 	return 0;
681 }
682 
683 static int device_close(struct inode *inode, struct file *file)
684 {
685 	struct dlm_user_proc *proc = file->private_data;
686 	struct dlm_ls *ls;
687 
688 	ls = dlm_find_lockspace_local(proc->lockspace);
689 	if (!ls)
690 		return -ENOENT;
691 
692 	set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
693 
694 	dlm_clear_proc_locks(ls, proc);
695 
696 	/* at this point no more lkb's should exist for this lockspace,
697 	   so there's no chance of dlm_user_add_ast() being called and
698 	   looking for lkb->ua->proc */
699 
700 	kfree(proc);
701 	file->private_data = NULL;
702 
703 	dlm_put_lockspace(ls);
704 	dlm_put_lockspace(ls);  /* for the find in device_open() */
705 
706 	/* FIXME: AUTOFREE: if this ls is no longer used do
707 	   device_remove_lockspace() */
708 
709 	return 0;
710 }
711 
712 static int copy_result_to_user(struct dlm_user_args *ua, int compat,
713 			       uint32_t flags, int mode, int copy_lvb,
714 			       char __user *buf, size_t count)
715 {
716 #ifdef CONFIG_COMPAT
717 	struct dlm_lock_result32 result32;
718 #endif
719 	struct dlm_lock_result result;
720 	void *resultptr;
721 	int error=0;
722 	int len;
723 	int struct_len;
724 
725 	memset(&result, 0, sizeof(struct dlm_lock_result));
726 	result.version[0] = DLM_DEVICE_VERSION_MAJOR;
727 	result.version[1] = DLM_DEVICE_VERSION_MINOR;
728 	result.version[2] = DLM_DEVICE_VERSION_PATCH;
729 	memcpy(&result.lksb, &ua->lksb, offsetof(struct dlm_lksb, sb_lvbptr));
730 	result.user_lksb = ua->user_lksb;
731 
732 	/* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
733 	   in a conversion unless the conversion is successful.  See code
734 	   in dlm_user_convert() for updating ua from ua_tmp.  OpenVMS, though,
735 	   notes that a new blocking AST address and parameter are set even if
736 	   the conversion fails, so maybe we should just do that. */
737 
738 	if (flags & DLM_CB_BAST) {
739 		result.user_astaddr = ua->bastaddr;
740 		result.user_astparam = ua->bastparam;
741 		result.bast_mode = mode;
742 	} else {
743 		result.user_astaddr = ua->castaddr;
744 		result.user_astparam = ua->castparam;
745 	}
746 
747 #ifdef CONFIG_COMPAT
748 	if (compat)
749 		len = sizeof(struct dlm_lock_result32);
750 	else
751 #endif
752 		len = sizeof(struct dlm_lock_result);
753 	struct_len = len;
754 
755 	/* copy lvb to userspace if there is one, it's been updated, and
756 	   the user buffer has space for it */
757 
758 	if (copy_lvb && ua->lksb.sb_lvbptr && count >= len + DLM_USER_LVB_LEN) {
759 		if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
760 				 DLM_USER_LVB_LEN)) {
761 			error = -EFAULT;
762 			goto out;
763 		}
764 
765 		result.lvb_offset = len;
766 		len += DLM_USER_LVB_LEN;
767 	}
768 
769 	result.length = len;
770 	resultptr = &result;
771 #ifdef CONFIG_COMPAT
772 	if (compat) {
773 		compat_output(&result, &result32);
774 		resultptr = &result32;
775 	}
776 #endif
777 
778 	if (copy_to_user(buf, resultptr, struct_len))
779 		error = -EFAULT;
780 	else
781 		error = len;
782  out:
783 	return error;
784 }
785 
786 static int copy_version_to_user(char __user *buf, size_t count)
787 {
788 	struct dlm_device_version ver;
789 
790 	memset(&ver, 0, sizeof(struct dlm_device_version));
791 	ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
792 	ver.version[1] = DLM_DEVICE_VERSION_MINOR;
793 	ver.version[2] = DLM_DEVICE_VERSION_PATCH;
794 
795 	if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
796 		return -EFAULT;
797 	return sizeof(struct dlm_device_version);
798 }
799 
800 /* a read returns a single ast described in a struct dlm_lock_result */
801 
802 static ssize_t device_read(struct file *file, char __user *buf, size_t count,
803 			   loff_t *ppos)
804 {
805 	struct dlm_user_proc *proc = file->private_data;
806 	struct dlm_lkb *lkb;
807 	DECLARE_WAITQUEUE(wait, current);
808 	struct dlm_callback *cb;
809 	int rv, copy_lvb = 0;
810 	int old_mode, new_mode;
811 
812 	if (count == sizeof(struct dlm_device_version)) {
813 		rv = copy_version_to_user(buf, count);
814 		return rv;
815 	}
816 
817 	if (!proc) {
818 		log_print("non-version read from control device %zu", count);
819 		return -EINVAL;
820 	}
821 
822 #ifdef CONFIG_COMPAT
823 	if (count < sizeof(struct dlm_lock_result32))
824 #else
825 	if (count < sizeof(struct dlm_lock_result))
826 #endif
827 		return -EINVAL;
828 
829  try_another:
830 
831 	/* do we really need this? can a read happen after a close? */
832 	if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
833 		return -EINVAL;
834 
835 	spin_lock(&proc->asts_spin);
836 	if (list_empty(&proc->asts)) {
837 		if (file->f_flags & O_NONBLOCK) {
838 			spin_unlock(&proc->asts_spin);
839 			return -EAGAIN;
840 		}
841 
842 		add_wait_queue(&proc->wait, &wait);
843 
844 	repeat:
845 		set_current_state(TASK_INTERRUPTIBLE);
846 		if (list_empty(&proc->asts) && !signal_pending(current)) {
847 			spin_unlock(&proc->asts_spin);
848 			schedule();
849 			spin_lock(&proc->asts_spin);
850 			goto repeat;
851 		}
852 		set_current_state(TASK_RUNNING);
853 		remove_wait_queue(&proc->wait, &wait);
854 
855 		if (signal_pending(current)) {
856 			spin_unlock(&proc->asts_spin);
857 			return -ERESTARTSYS;
858 		}
859 	}
860 
861 	/* if we empty lkb_callbacks, we don't want to unlock the spinlock
862 	   without removing lkb_cb_list; so empty lkb_cb_list is always
863 	   consistent with empty lkb_callbacks */
864 
865 	lkb = list_first_entry(&proc->asts, struct dlm_lkb, lkb_cb_list);
866 
867 	/* rem_lkb_callback sets a new lkb_last_cast */
868 	old_mode = lkb->lkb_last_cast->mode;
869 
870 	rv = dlm_dequeue_lkb_callback(lkb, &cb);
871 	switch (rv) {
872 	case DLM_DEQUEUE_CALLBACK_EMPTY:
873 		/* this shouldn't happen; lkb should have been removed from
874 		 * list when last item was dequeued
875 		 */
876 		log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
877 		list_del_init(&lkb->lkb_cb_list);
878 		spin_unlock(&proc->asts_spin);
879 		/* removes ref for proc->asts, may cause lkb to be freed */
880 		dlm_put_lkb(lkb);
881 		WARN_ON_ONCE(1);
882 		goto try_another;
883 	case DLM_DEQUEUE_CALLBACK_LAST:
884 		list_del_init(&lkb->lkb_cb_list);
885 		clear_bit(DLM_IFL_CB_PENDING_BIT, &lkb->lkb_iflags);
886 		break;
887 	case DLM_DEQUEUE_CALLBACK_SUCCESS:
888 		break;
889 	default:
890 		WARN_ON_ONCE(1);
891 		break;
892 	}
893 	spin_unlock(&proc->asts_spin);
894 
895 	if (cb->flags & DLM_CB_BAST) {
896 		trace_dlm_bast(lkb->lkb_resource->res_ls, lkb, cb->mode);
897 	} else if (cb->flags & DLM_CB_CAST) {
898 		new_mode = cb->mode;
899 
900 		if (!cb->sb_status && lkb->lkb_lksb->sb_lvbptr &&
901 		    dlm_lvb_operations[old_mode + 1][new_mode + 1])
902 			copy_lvb = 1;
903 
904 		lkb->lkb_lksb->sb_status = cb->sb_status;
905 		lkb->lkb_lksb->sb_flags = cb->sb_flags;
906 		trace_dlm_ast(lkb->lkb_resource->res_ls, lkb);
907 	}
908 
909 	rv = copy_result_to_user(lkb->lkb_ua,
910 				 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
911 				 cb->flags, cb->mode, copy_lvb, buf, count);
912 
913 	kref_put(&cb->ref, dlm_release_callback);
914 
915 	/* removes ref for proc->asts, may cause lkb to be freed */
916 	if (rv == DLM_DEQUEUE_CALLBACK_LAST)
917 		dlm_put_lkb(lkb);
918 
919 	return rv;
920 }
921 
922 static __poll_t device_poll(struct file *file, poll_table *wait)
923 {
924 	struct dlm_user_proc *proc = file->private_data;
925 
926 	poll_wait(file, &proc->wait, wait);
927 
928 	spin_lock(&proc->asts_spin);
929 	if (!list_empty(&proc->asts)) {
930 		spin_unlock(&proc->asts_spin);
931 		return EPOLLIN | EPOLLRDNORM;
932 	}
933 	spin_unlock(&proc->asts_spin);
934 	return 0;
935 }
936 
937 int dlm_user_daemon_available(void)
938 {
939 	/* dlm_controld hasn't started (or, has started, but not
940 	   properly populated configfs) */
941 
942 	if (!dlm_our_nodeid())
943 		return 0;
944 
945 	/* This is to deal with versions of dlm_controld that don't
946 	   know about the monitor device.  We assume that if the
947 	   dlm_controld was started (above), but the monitor device
948 	   was never opened, that it's an old version.  dlm_controld
949 	   should open the monitor device before populating configfs. */
950 
951 	if (dlm_monitor_unused)
952 		return 1;
953 
954 	return atomic_read(&dlm_monitor_opened) ? 1 : 0;
955 }
956 
957 static int ctl_device_open(struct inode *inode, struct file *file)
958 {
959 	file->private_data = NULL;
960 	return 0;
961 }
962 
963 static int ctl_device_close(struct inode *inode, struct file *file)
964 {
965 	return 0;
966 }
967 
968 static int monitor_device_open(struct inode *inode, struct file *file)
969 {
970 	atomic_inc(&dlm_monitor_opened);
971 	dlm_monitor_unused = 0;
972 	return 0;
973 }
974 
975 static int monitor_device_close(struct inode *inode, struct file *file)
976 {
977 	if (atomic_dec_and_test(&dlm_monitor_opened))
978 		dlm_stop_lockspaces();
979 	return 0;
980 }
981 
982 static const struct file_operations device_fops = {
983 	.open    = device_open,
984 	.release = device_close,
985 	.read    = device_read,
986 	.write   = device_write,
987 	.poll    = device_poll,
988 	.owner   = THIS_MODULE,
989 	.llseek  = noop_llseek,
990 };
991 
992 static const struct file_operations ctl_device_fops = {
993 	.open    = ctl_device_open,
994 	.release = ctl_device_close,
995 	.read    = device_read,
996 	.write   = device_write,
997 	.owner   = THIS_MODULE,
998 	.llseek  = noop_llseek,
999 };
1000 
1001 static struct miscdevice ctl_device = {
1002 	.name  = "dlm-control",
1003 	.fops  = &ctl_device_fops,
1004 	.minor = MISC_DYNAMIC_MINOR,
1005 };
1006 
1007 static const struct file_operations monitor_device_fops = {
1008 	.open    = monitor_device_open,
1009 	.release = monitor_device_close,
1010 	.owner   = THIS_MODULE,
1011 	.llseek  = noop_llseek,
1012 };
1013 
1014 static struct miscdevice monitor_device = {
1015 	.name  = "dlm-monitor",
1016 	.fops  = &monitor_device_fops,
1017 	.minor = MISC_DYNAMIC_MINOR,
1018 };
1019 
1020 int __init dlm_user_init(void)
1021 {
1022 	int error;
1023 
1024 	atomic_set(&dlm_monitor_opened, 0);
1025 
1026 	error = misc_register(&ctl_device);
1027 	if (error) {
1028 		log_print("misc_register failed for control device");
1029 		goto out;
1030 	}
1031 
1032 	error = misc_register(&monitor_device);
1033 	if (error) {
1034 		log_print("misc_register failed for monitor device");
1035 		misc_deregister(&ctl_device);
1036 	}
1037  out:
1038 	return error;
1039 }
1040 
1041 void dlm_user_exit(void)
1042 {
1043 	misc_deregister(&ctl_device);
1044 	misc_deregister(&monitor_device);
1045 }
1046 
1047