xref: /linux/fs/dlm/user.c (revision 9190d4a263264eabf715f5fc1827da45e3fdc247)
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 
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[];
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 = (__user void *)(long)kb32->i.lock.castparam;
114 		kb->i.lock.castaddr = (__user void *)(long)kb32->i.lock.castaddr;
115 		kb->i.lock.bastparam = (__user void *)(long)kb32->i.lock.bastparam;
116 		kb->i.lock.bastaddr = (__user void *)(long)kb32->i.lock.bastaddr;
117 		kb->i.lock.lksb = (__user 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 	memset(res32, 0, sizeof(*res32));
127 
128 	res32->version[0] = res->version[0];
129 	res32->version[1] = res->version[1];
130 	res32->version[2] = res->version[2];
131 
132 	res32->user_astaddr = (__u32)(__force long)res->user_astaddr;
133 	res32->user_astparam = (__u32)(__force long)res->user_astparam;
134 	res32->user_lksb = (__u32)(__force long)res->user_lksb;
135 	res32->bast_mode = res->bast_mode;
136 
137 	res32->lvb_offset = res->lvb_offset;
138 	res32->length = res->length;
139 
140 	res32->lksb.sb_status = res->lksb.sb_status;
141 	res32->lksb.sb_flags = res->lksb.sb_flags;
142 	res32->lksb.sb_lkid = res->lksb.sb_lkid;
143 	res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
144 }
145 #endif
146 
147 /* Figure out if this lock is at the end of its life and no longer
148    available for the application to use.  The lkb still exists until
149    the final ast is read.  A lock becomes EOL in three situations:
150      1. a noqueue request fails with EAGAIN
151      2. an unlock completes with EUNLOCK
152      3. a cancel of a waiting request completes with ECANCEL/EDEADLK
153    An EOL lock needs to be removed from the process's list of locks.
154    And we can't allow any new operation on an EOL lock.  This is
155    not related to the lifetime of the lkb struct which is managed
156    entirely by refcount. */
157 
158 static int lkb_is_endoflife(int mode, int status)
159 {
160 	switch (status) {
161 	case -DLM_EUNLOCK:
162 		return 1;
163 	case -DLM_ECANCEL:
164 	case -ETIMEDOUT:
165 	case -EDEADLK:
166 	case -EAGAIN:
167 		if (mode == DLM_LOCK_IV)
168 			return 1;
169 		break;
170 	}
171 	return 0;
172 }
173 
174 /* we could possibly check if the cancel of an orphan has resulted in the lkb
175    being removed and then remove that lkb from the orphans list and free it */
176 
177 void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
178 		      int status, uint32_t sbflags, uint64_t seq)
179 {
180 	struct dlm_ls *ls;
181 	struct dlm_user_args *ua;
182 	struct dlm_user_proc *proc;
183 	int rv;
184 
185 	if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
186 		return;
187 
188 	ls = lkb->lkb_resource->res_ls;
189 	spin_lock(&ls->ls_clear_proc_locks);
190 
191 	/* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
192 	   can't be delivered.  For ORPHAN's, dlm_clear_proc_locks() freed
193 	   lkb->ua so we can't try to use it.  This second check is necessary
194 	   for cases where a completion ast is received for an operation that
195 	   began before clear_proc_locks did its cancel/unlock. */
196 
197 	if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
198 		goto out;
199 
200 	DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
201 	ua = lkb->lkb_ua;
202 	proc = ua->proc;
203 
204 	if ((flags & DLM_CB_BAST) && ua->bastaddr == NULL)
205 		goto out;
206 
207 	if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status))
208 		lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
209 
210 	spin_lock(&proc->asts_spin);
211 
212 	rv = dlm_add_lkb_callback(lkb, flags, mode, status, sbflags, seq);
213 	if (rv < 0) {
214 		spin_unlock(&proc->asts_spin);
215 		goto out;
216 	}
217 
218 	if (list_empty(&lkb->lkb_cb_list)) {
219 		kref_get(&lkb->lkb_ref);
220 		list_add_tail(&lkb->lkb_cb_list, &proc->asts);
221 		wake_up_interruptible(&proc->wait);
222 	}
223 	spin_unlock(&proc->asts_spin);
224 
225 	if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
226 		/* N.B. spin_lock locks_spin, not asts_spin */
227 		spin_lock(&proc->locks_spin);
228 		if (!list_empty(&lkb->lkb_ownqueue)) {
229 			list_del_init(&lkb->lkb_ownqueue);
230 			dlm_put_lkb(lkb);
231 		}
232 		spin_unlock(&proc->locks_spin);
233 	}
234  out:
235 	spin_unlock(&ls->ls_clear_proc_locks);
236 }
237 
238 static int device_user_lock(struct dlm_user_proc *proc,
239 			    struct dlm_lock_params *params)
240 {
241 	struct dlm_ls *ls;
242 	struct dlm_user_args *ua;
243 	uint32_t lkid;
244 	int error = -ENOMEM;
245 
246 	ls = dlm_find_lockspace_local(proc->lockspace);
247 	if (!ls)
248 		return -ENOENT;
249 
250 	if (!params->castaddr || !params->lksb) {
251 		error = -EINVAL;
252 		goto out;
253 	}
254 
255 #ifdef CONFIG_DLM_DEPRECATED_API
256 	if (params->timeout)
257 		pr_warn_once("========================================================\n"
258 			     "WARNING: the lkb timeout feature is being deprecated and\n"
259 			     "         will be removed in v6.2!\n"
260 			     "========================================================\n");
261 #endif
262 
263 	ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
264 	if (!ua)
265 		goto out;
266 	ua->proc = proc;
267 	ua->user_lksb = params->lksb;
268 	ua->castparam = params->castparam;
269 	ua->castaddr = params->castaddr;
270 	ua->bastparam = params->bastparam;
271 	ua->bastaddr = params->bastaddr;
272 	ua->xid = params->xid;
273 
274 	if (params->flags & DLM_LKF_CONVERT) {
275 #ifdef CONFIG_DLM_DEPRECATED_API
276 		error = dlm_user_convert(ls, ua,
277 				         params->mode, params->flags,
278 				         params->lkid, params->lvb,
279 					 (unsigned long) params->timeout);
280 #else
281 		error = dlm_user_convert(ls, ua,
282 					 params->mode, params->flags,
283 					 params->lkid, params->lvb);
284 #endif
285 	} else if (params->flags & DLM_LKF_ORPHAN) {
286 		error = dlm_user_adopt_orphan(ls, ua,
287 					 params->mode, params->flags,
288 					 params->name, params->namelen,
289 					 &lkid);
290 		if (!error)
291 			error = lkid;
292 	} else {
293 #ifdef CONFIG_DLM_DEPRECATED_API
294 		error = dlm_user_request(ls, ua,
295 					 params->mode, params->flags,
296 					 params->name, params->namelen,
297 					 (unsigned long) params->timeout);
298 #else
299 		error = dlm_user_request(ls, ua,
300 					 params->mode, params->flags,
301 					 params->name, params->namelen);
302 #endif
303 		if (!error)
304 			error = ua->lksb.sb_lkid;
305 	}
306  out:
307 	dlm_put_lockspace(ls);
308 	return error;
309 }
310 
311 static int device_user_unlock(struct dlm_user_proc *proc,
312 			      struct dlm_lock_params *params)
313 {
314 	struct dlm_ls *ls;
315 	struct dlm_user_args *ua;
316 	int error = -ENOMEM;
317 
318 	ls = dlm_find_lockspace_local(proc->lockspace);
319 	if (!ls)
320 		return -ENOENT;
321 
322 	ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
323 	if (!ua)
324 		goto out;
325 	ua->proc = proc;
326 	ua->user_lksb = params->lksb;
327 	ua->castparam = params->castparam;
328 	ua->castaddr = params->castaddr;
329 
330 	if (params->flags & DLM_LKF_CANCEL)
331 		error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
332 	else
333 		error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
334 					params->lvb);
335  out:
336 	dlm_put_lockspace(ls);
337 	return error;
338 }
339 
340 static int device_user_deadlock(struct dlm_user_proc *proc,
341 				struct dlm_lock_params *params)
342 {
343 	struct dlm_ls *ls;
344 	int error;
345 
346 	ls = dlm_find_lockspace_local(proc->lockspace);
347 	if (!ls)
348 		return -ENOENT;
349 
350 	error = dlm_user_deadlock(ls, params->flags, params->lkid);
351 
352 	dlm_put_lockspace(ls);
353 	return error;
354 }
355 
356 static int dlm_device_register(struct dlm_ls *ls, char *name)
357 {
358 	int error, len;
359 
360 	/* The device is already registered.  This happens when the
361 	   lockspace is created multiple times from userspace. */
362 	if (ls->ls_device.name)
363 		return 0;
364 
365 	error = -ENOMEM;
366 	len = strlen(name) + strlen(name_prefix) + 2;
367 	ls->ls_device.name = kzalloc(len, GFP_NOFS);
368 	if (!ls->ls_device.name)
369 		goto fail;
370 
371 	snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
372 		 name);
373 	ls->ls_device.fops = &device_fops;
374 	ls->ls_device.minor = MISC_DYNAMIC_MINOR;
375 
376 	error = misc_register(&ls->ls_device);
377 	if (error) {
378 		kfree(ls->ls_device.name);
379 		/* this has to be set to NULL
380 		 * to avoid a double-free in dlm_device_deregister
381 		 */
382 		ls->ls_device.name = NULL;
383 	}
384 fail:
385 	return error;
386 }
387 
388 int dlm_device_deregister(struct dlm_ls *ls)
389 {
390 	/* The device is not registered.  This happens when the lockspace
391 	   was never used from userspace, or when device_create_lockspace()
392 	   calls dlm_release_lockspace() after the register fails. */
393 	if (!ls->ls_device.name)
394 		return 0;
395 
396 	misc_deregister(&ls->ls_device);
397 	kfree(ls->ls_device.name);
398 	return 0;
399 }
400 
401 static int device_user_purge(struct dlm_user_proc *proc,
402 			     struct dlm_purge_params *params)
403 {
404 	struct dlm_ls *ls;
405 	int error;
406 
407 	ls = dlm_find_lockspace_local(proc->lockspace);
408 	if (!ls)
409 		return -ENOENT;
410 
411 	error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
412 
413 	dlm_put_lockspace(ls);
414 	return error;
415 }
416 
417 static int device_create_lockspace(struct dlm_lspace_params *params)
418 {
419 	dlm_lockspace_t *lockspace;
420 	struct dlm_ls *ls;
421 	int error;
422 
423 	if (!capable(CAP_SYS_ADMIN))
424 		return -EPERM;
425 
426 	error = dlm_new_user_lockspace(params->name, dlm_config.ci_cluster_name,
427 				       params->flags, DLM_USER_LVB_LEN, NULL,
428 				       NULL, NULL, &lockspace);
429 	if (error)
430 		return error;
431 
432 	ls = dlm_find_lockspace_local(lockspace);
433 	if (!ls)
434 		return -ENOENT;
435 
436 	error = dlm_device_register(ls, params->name);
437 	dlm_put_lockspace(ls);
438 
439 	if (error)
440 		dlm_release_lockspace(lockspace, 0);
441 	else
442 		error = ls->ls_device.minor;
443 
444 	return error;
445 }
446 
447 static int device_remove_lockspace(struct dlm_lspace_params *params)
448 {
449 	dlm_lockspace_t *lockspace;
450 	struct dlm_ls *ls;
451 	int error, force = 0;
452 
453 	if (!capable(CAP_SYS_ADMIN))
454 		return -EPERM;
455 
456 	ls = dlm_find_lockspace_device(params->minor);
457 	if (!ls)
458 		return -ENOENT;
459 
460 	if (params->flags & DLM_USER_LSFLG_FORCEFREE)
461 		force = 2;
462 
463 	lockspace = ls->ls_local_handle;
464 	dlm_put_lockspace(ls);
465 
466 	/* The final dlm_release_lockspace waits for references to go to
467 	   zero, so all processes will need to close their device for the
468 	   ls before the release will proceed.  release also calls the
469 	   device_deregister above.  Converting a positive return value
470 	   from release to zero means that userspace won't know when its
471 	   release was the final one, but it shouldn't need to know. */
472 
473 	error = dlm_release_lockspace(lockspace, force);
474 	if (error > 0)
475 		error = 0;
476 	return error;
477 }
478 
479 /* Check the user's version matches ours */
480 static int check_version(struct dlm_write_request *req)
481 {
482 	if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
483 	    (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
484 	     req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
485 
486 		printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
487 		       "user (%d.%d.%d) kernel (%d.%d.%d)\n",
488 		       current->comm,
489 		       task_pid_nr(current),
490 		       req->version[0],
491 		       req->version[1],
492 		       req->version[2],
493 		       DLM_DEVICE_VERSION_MAJOR,
494 		       DLM_DEVICE_VERSION_MINOR,
495 		       DLM_DEVICE_VERSION_PATCH);
496 		return -EINVAL;
497 	}
498 	return 0;
499 }
500 
501 /*
502  * device_write
503  *
504  *   device_user_lock
505  *     dlm_user_request -> request_lock
506  *     dlm_user_convert -> convert_lock
507  *
508  *   device_user_unlock
509  *     dlm_user_unlock -> unlock_lock
510  *     dlm_user_cancel -> cancel_lock
511  *
512  *   device_create_lockspace
513  *     dlm_new_lockspace
514  *
515  *   device_remove_lockspace
516  *     dlm_release_lockspace
517  */
518 
519 /* a write to a lockspace device is a lock or unlock request, a write
520    to the control device is to create/remove a lockspace */
521 
522 static ssize_t device_write(struct file *file, const char __user *buf,
523 			    size_t count, loff_t *ppos)
524 {
525 	struct dlm_user_proc *proc = file->private_data;
526 	struct dlm_write_request *kbuf;
527 	int error;
528 
529 #ifdef CONFIG_COMPAT
530 	if (count < sizeof(struct dlm_write_request32))
531 #else
532 	if (count < sizeof(struct dlm_write_request))
533 #endif
534 		return -EINVAL;
535 
536 	/*
537 	 * can't compare against COMPAT/dlm_write_request32 because
538 	 * we don't yet know if is64bit is zero
539 	 */
540 	if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
541 		return -EINVAL;
542 
543 	kbuf = memdup_user_nul(buf, count);
544 	if (IS_ERR(kbuf))
545 		return PTR_ERR(kbuf);
546 
547 	if (check_version(kbuf)) {
548 		error = -EBADE;
549 		goto out_free;
550 	}
551 
552 #ifdef CONFIG_COMPAT
553 	if (!kbuf->is64bit) {
554 		struct dlm_write_request32 *k32buf;
555 		int namelen = 0;
556 
557 		if (count > sizeof(struct dlm_write_request32))
558 			namelen = count - sizeof(struct dlm_write_request32);
559 
560 		k32buf = (struct dlm_write_request32 *)kbuf;
561 
562 		/* add 1 after namelen so that the name string is terminated */
563 		kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
564 			       GFP_NOFS);
565 		if (!kbuf) {
566 			kfree(k32buf);
567 			return -ENOMEM;
568 		}
569 
570 		if (proc)
571 			set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
572 
573 		compat_input(kbuf, k32buf, namelen);
574 		kfree(k32buf);
575 	}
576 #endif
577 
578 	/* do we really need this? can a write happen after a close? */
579 	if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
580 	    (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
581 		error = -EINVAL;
582 		goto out_free;
583 	}
584 
585 	error = -EINVAL;
586 
587 	switch (kbuf->cmd)
588 	{
589 	case DLM_USER_LOCK:
590 		if (!proc) {
591 			log_print("no locking on control device");
592 			goto out_free;
593 		}
594 		error = device_user_lock(proc, &kbuf->i.lock);
595 		break;
596 
597 	case DLM_USER_UNLOCK:
598 		if (!proc) {
599 			log_print("no locking on control device");
600 			goto out_free;
601 		}
602 		error = device_user_unlock(proc, &kbuf->i.lock);
603 		break;
604 
605 	case DLM_USER_DEADLOCK:
606 		if (!proc) {
607 			log_print("no locking on control device");
608 			goto out_free;
609 		}
610 		error = device_user_deadlock(proc, &kbuf->i.lock);
611 		break;
612 
613 	case DLM_USER_CREATE_LOCKSPACE:
614 		if (proc) {
615 			log_print("create/remove only on control device");
616 			goto out_free;
617 		}
618 		error = device_create_lockspace(&kbuf->i.lspace);
619 		break;
620 
621 	case DLM_USER_REMOVE_LOCKSPACE:
622 		if (proc) {
623 			log_print("create/remove only on control device");
624 			goto out_free;
625 		}
626 		error = device_remove_lockspace(&kbuf->i.lspace);
627 		break;
628 
629 	case DLM_USER_PURGE:
630 		if (!proc) {
631 			log_print("no locking on control device");
632 			goto out_free;
633 		}
634 		error = device_user_purge(proc, &kbuf->i.purge);
635 		break;
636 
637 	default:
638 		log_print("Unknown command passed to DLM device : %d\n",
639 			  kbuf->cmd);
640 	}
641 
642  out_free:
643 	kfree(kbuf);
644 	return error;
645 }
646 
647 /* Every process that opens the lockspace device has its own "proc" structure
648    hanging off the open file that's used to keep track of locks owned by the
649    process and asts that need to be delivered to the process. */
650 
651 static int device_open(struct inode *inode, struct file *file)
652 {
653 	struct dlm_user_proc *proc;
654 	struct dlm_ls *ls;
655 
656 	ls = dlm_find_lockspace_device(iminor(inode));
657 	if (!ls)
658 		return -ENOENT;
659 
660 	proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
661 	if (!proc) {
662 		dlm_put_lockspace(ls);
663 		return -ENOMEM;
664 	}
665 
666 	proc->lockspace = ls->ls_local_handle;
667 	INIT_LIST_HEAD(&proc->asts);
668 	INIT_LIST_HEAD(&proc->locks);
669 	INIT_LIST_HEAD(&proc->unlocking);
670 	spin_lock_init(&proc->asts_spin);
671 	spin_lock_init(&proc->locks_spin);
672 	init_waitqueue_head(&proc->wait);
673 	file->private_data = proc;
674 
675 	return 0;
676 }
677 
678 static int device_close(struct inode *inode, struct file *file)
679 {
680 	struct dlm_user_proc *proc = file->private_data;
681 	struct dlm_ls *ls;
682 
683 	ls = dlm_find_lockspace_local(proc->lockspace);
684 	if (!ls)
685 		return -ENOENT;
686 
687 	set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
688 
689 	dlm_clear_proc_locks(ls, proc);
690 
691 	/* at this point no more lkb's should exist for this lockspace,
692 	   so there's no chance of dlm_user_add_ast() being called and
693 	   looking for lkb->ua->proc */
694 
695 	kfree(proc);
696 	file->private_data = NULL;
697 
698 	dlm_put_lockspace(ls);
699 	dlm_put_lockspace(ls);  /* for the find in device_open() */
700 
701 	/* FIXME: AUTOFREE: if this ls is no longer used do
702 	   device_remove_lockspace() */
703 
704 	return 0;
705 }
706 
707 static int copy_result_to_user(struct dlm_user_args *ua, int compat,
708 			       uint32_t flags, int mode, int copy_lvb,
709 			       char __user *buf, size_t count)
710 {
711 #ifdef CONFIG_COMPAT
712 	struct dlm_lock_result32 result32;
713 #endif
714 	struct dlm_lock_result result;
715 	void *resultptr;
716 	int error=0;
717 	int len;
718 	int struct_len;
719 
720 	memset(&result, 0, sizeof(struct dlm_lock_result));
721 	result.version[0] = DLM_DEVICE_VERSION_MAJOR;
722 	result.version[1] = DLM_DEVICE_VERSION_MINOR;
723 	result.version[2] = DLM_DEVICE_VERSION_PATCH;
724 	memcpy(&result.lksb, &ua->lksb, offsetof(struct dlm_lksb, sb_lvbptr));
725 	result.user_lksb = ua->user_lksb;
726 
727 	/* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
728 	   in a conversion unless the conversion is successful.  See code
729 	   in dlm_user_convert() for updating ua from ua_tmp.  OpenVMS, though,
730 	   notes that a new blocking AST address and parameter are set even if
731 	   the conversion fails, so maybe we should just do that. */
732 
733 	if (flags & DLM_CB_BAST) {
734 		result.user_astaddr = ua->bastaddr;
735 		result.user_astparam = ua->bastparam;
736 		result.bast_mode = mode;
737 	} else {
738 		result.user_astaddr = ua->castaddr;
739 		result.user_astparam = ua->castparam;
740 	}
741 
742 #ifdef CONFIG_COMPAT
743 	if (compat)
744 		len = sizeof(struct dlm_lock_result32);
745 	else
746 #endif
747 		len = sizeof(struct dlm_lock_result);
748 	struct_len = len;
749 
750 	/* copy lvb to userspace if there is one, it's been updated, and
751 	   the user buffer has space for it */
752 
753 	if (copy_lvb && ua->lksb.sb_lvbptr && 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 	struct dlm_callback cb;
804 	int rv, resid, copy_lvb = 0;
805 	int old_mode, new_mode;
806 
807 	if (count == sizeof(struct dlm_device_version)) {
808 		rv = copy_version_to_user(buf, count);
809 		return rv;
810 	}
811 
812 	if (!proc) {
813 		log_print("non-version read from control device %zu", count);
814 		return -EINVAL;
815 	}
816 
817 #ifdef CONFIG_COMPAT
818 	if (count < sizeof(struct dlm_lock_result32))
819 #else
820 	if (count < sizeof(struct dlm_lock_result))
821 #endif
822 		return -EINVAL;
823 
824  try_another:
825 
826 	/* do we really need this? can a read happen after a close? */
827 	if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
828 		return -EINVAL;
829 
830 	spin_lock(&proc->asts_spin);
831 	if (list_empty(&proc->asts)) {
832 		if (file->f_flags & O_NONBLOCK) {
833 			spin_unlock(&proc->asts_spin);
834 			return -EAGAIN;
835 		}
836 
837 		add_wait_queue(&proc->wait, &wait);
838 
839 	repeat:
840 		set_current_state(TASK_INTERRUPTIBLE);
841 		if (list_empty(&proc->asts) && !signal_pending(current)) {
842 			spin_unlock(&proc->asts_spin);
843 			schedule();
844 			spin_lock(&proc->asts_spin);
845 			goto repeat;
846 		}
847 		set_current_state(TASK_RUNNING);
848 		remove_wait_queue(&proc->wait, &wait);
849 
850 		if (signal_pending(current)) {
851 			spin_unlock(&proc->asts_spin);
852 			return -ERESTARTSYS;
853 		}
854 	}
855 
856 	/* if we empty lkb_callbacks, we don't want to unlock the spinlock
857 	   without removing lkb_cb_list; so empty lkb_cb_list is always
858 	   consistent with empty lkb_callbacks */
859 
860 	lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_cb_list);
861 
862 	/* rem_lkb_callback sets a new lkb_last_cast */
863 	old_mode = lkb->lkb_last_cast.mode;
864 
865 	rv = dlm_rem_lkb_callback(lkb->lkb_resource->res_ls, lkb, &cb, &resid);
866 	if (rv < 0) {
867 		/* this shouldn't happen; lkb should have been removed from
868 		   list when resid was zero */
869 		log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
870 		list_del_init(&lkb->lkb_cb_list);
871 		spin_unlock(&proc->asts_spin);
872 		/* removes ref for proc->asts, may cause lkb to be freed */
873 		dlm_put_lkb(lkb);
874 		goto try_another;
875 	}
876 	if (!resid)
877 		list_del_init(&lkb->lkb_cb_list);
878 	spin_unlock(&proc->asts_spin);
879 
880 	if (cb.flags & DLM_CB_SKIP) {
881 		/* removes ref for proc->asts, may cause lkb to be freed */
882 		if (!resid)
883 			dlm_put_lkb(lkb);
884 		goto try_another;
885 	}
886 
887 	if (cb.flags & DLM_CB_BAST) {
888 		trace_dlm_bast(lkb->lkb_resource->res_ls, lkb, cb.mode);
889 	} else if (cb.flags & DLM_CB_CAST) {
890 		new_mode = cb.mode;
891 
892 		if (!cb.sb_status && lkb->lkb_lksb->sb_lvbptr &&
893 		    dlm_lvb_operations[old_mode + 1][new_mode + 1])
894 			copy_lvb = 1;
895 
896 		lkb->lkb_lksb->sb_status = cb.sb_status;
897 		lkb->lkb_lksb->sb_flags = cb.sb_flags;
898 		trace_dlm_ast(lkb->lkb_resource->res_ls, lkb);
899 	}
900 
901 	rv = copy_result_to_user(lkb->lkb_ua,
902 				 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
903 				 cb.flags, cb.mode, copy_lvb, buf, count);
904 
905 	/* removes ref for proc->asts, may cause lkb to be freed */
906 	if (!resid)
907 		dlm_put_lkb(lkb);
908 
909 	return rv;
910 }
911 
912 static __poll_t device_poll(struct file *file, poll_table *wait)
913 {
914 	struct dlm_user_proc *proc = file->private_data;
915 
916 	poll_wait(file, &proc->wait, wait);
917 
918 	spin_lock(&proc->asts_spin);
919 	if (!list_empty(&proc->asts)) {
920 		spin_unlock(&proc->asts_spin);
921 		return EPOLLIN | EPOLLRDNORM;
922 	}
923 	spin_unlock(&proc->asts_spin);
924 	return 0;
925 }
926 
927 int dlm_user_daemon_available(void)
928 {
929 	/* dlm_controld hasn't started (or, has started, but not
930 	   properly populated configfs) */
931 
932 	if (!dlm_our_nodeid())
933 		return 0;
934 
935 	/* This is to deal with versions of dlm_controld that don't
936 	   know about the monitor device.  We assume that if the
937 	   dlm_controld was started (above), but the monitor device
938 	   was never opened, that it's an old version.  dlm_controld
939 	   should open the monitor device before populating configfs. */
940 
941 	if (dlm_monitor_unused)
942 		return 1;
943 
944 	return atomic_read(&dlm_monitor_opened) ? 1 : 0;
945 }
946 
947 static int ctl_device_open(struct inode *inode, struct file *file)
948 {
949 	file->private_data = NULL;
950 	return 0;
951 }
952 
953 static int ctl_device_close(struct inode *inode, struct file *file)
954 {
955 	return 0;
956 }
957 
958 static int monitor_device_open(struct inode *inode, struct file *file)
959 {
960 	atomic_inc(&dlm_monitor_opened);
961 	dlm_monitor_unused = 0;
962 	return 0;
963 }
964 
965 static int monitor_device_close(struct inode *inode, struct file *file)
966 {
967 	if (atomic_dec_and_test(&dlm_monitor_opened))
968 		dlm_stop_lockspaces();
969 	return 0;
970 }
971 
972 static const struct file_operations device_fops = {
973 	.open    = device_open,
974 	.release = device_close,
975 	.read    = device_read,
976 	.write   = device_write,
977 	.poll    = device_poll,
978 	.owner   = THIS_MODULE,
979 	.llseek  = noop_llseek,
980 };
981 
982 static const struct file_operations ctl_device_fops = {
983 	.open    = ctl_device_open,
984 	.release = ctl_device_close,
985 	.read    = device_read,
986 	.write   = device_write,
987 	.owner   = THIS_MODULE,
988 	.llseek  = noop_llseek,
989 };
990 
991 static struct miscdevice ctl_device = {
992 	.name  = "dlm-control",
993 	.fops  = &ctl_device_fops,
994 	.minor = MISC_DYNAMIC_MINOR,
995 };
996 
997 static const struct file_operations monitor_device_fops = {
998 	.open    = monitor_device_open,
999 	.release = monitor_device_close,
1000 	.owner   = THIS_MODULE,
1001 	.llseek  = noop_llseek,
1002 };
1003 
1004 static struct miscdevice monitor_device = {
1005 	.name  = "dlm-monitor",
1006 	.fops  = &monitor_device_fops,
1007 	.minor = MISC_DYNAMIC_MINOR,
1008 };
1009 
1010 int __init dlm_user_init(void)
1011 {
1012 	int error;
1013 
1014 	atomic_set(&dlm_monitor_opened, 0);
1015 
1016 	error = misc_register(&ctl_device);
1017 	if (error) {
1018 		log_print("misc_register failed for control device");
1019 		goto out;
1020 	}
1021 
1022 	error = misc_register(&monitor_device);
1023 	if (error) {
1024 		log_print("misc_register failed for monitor device");
1025 		misc_deregister(&ctl_device);
1026 	}
1027  out:
1028 	return error;
1029 }
1030 
1031 void dlm_user_exit(void)
1032 {
1033 	misc_deregister(&ctl_device);
1034 	misc_deregister(&monitor_device);
1035 }
1036 
1037