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
29static const char name_prefix[] = "dlm";
30static const struct file_operations device_fops;
31static atomic_t dlm_monitor_opened;
32static int dlm_monitor_unused = 1;
33
34#ifdef CONFIG_COMPAT
35
36struct 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
54struct 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
67struct dlm_lksb32 {
68	__u32 sb_status;
69	__u32 sb_lkid;
70	__u8 sb_flags;
71	__u32 sb_lvbptr;
72};
73
74struct 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
87static 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
123static 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
156static 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
175void 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
236static 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	uint32_t lkid;
242	int error = -ENOMEM;
243
244	ls = dlm_find_lockspace_local(proc->lockspace);
245	if (!ls)
246		return -ENOENT;
247
248	if (!params->castaddr || !params->lksb) {
249		error = -EINVAL;
250		goto out;
251	}
252
253	ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
254	if (!ua)
255		goto out;
256	ua->proc = proc;
257	ua->user_lksb = params->lksb;
258	ua->castparam = params->castparam;
259	ua->castaddr = params->castaddr;
260	ua->bastparam = params->bastparam;
261	ua->bastaddr = params->bastaddr;
262	ua->xid = params->xid;
263
264	if (params->flags & DLM_LKF_CONVERT) {
265		error = dlm_user_convert(ls, ua,
266				         params->mode, params->flags,
267				         params->lkid, params->lvb,
268					 (unsigned long) params->timeout);
269	} else if (params->flags & DLM_LKF_ORPHAN) {
270		error = dlm_user_adopt_orphan(ls, ua,
271					 params->mode, params->flags,
272					 params->name, params->namelen,
273					 (unsigned long) params->timeout,
274					 &lkid);
275		if (!error)
276			error = lkid;
277	} else {
278		error = dlm_user_request(ls, ua,
279					 params->mode, params->flags,
280					 params->name, params->namelen,
281					 (unsigned long) params->timeout);
282		if (!error)
283			error = ua->lksb.sb_lkid;
284	}
285 out:
286	dlm_put_lockspace(ls);
287	return error;
288}
289
290static int device_user_unlock(struct dlm_user_proc *proc,
291			      struct dlm_lock_params *params)
292{
293	struct dlm_ls *ls;
294	struct dlm_user_args *ua;
295	int error = -ENOMEM;
296
297	ls = dlm_find_lockspace_local(proc->lockspace);
298	if (!ls)
299		return -ENOENT;
300
301	ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
302	if (!ua)
303		goto out;
304	ua->proc = proc;
305	ua->user_lksb = params->lksb;
306	ua->castparam = params->castparam;
307	ua->castaddr = params->castaddr;
308
309	if (params->flags & DLM_LKF_CANCEL)
310		error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
311	else
312		error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
313					params->lvb);
314 out:
315	dlm_put_lockspace(ls);
316	return error;
317}
318
319static int device_user_deadlock(struct dlm_user_proc *proc,
320				struct dlm_lock_params *params)
321{
322	struct dlm_ls *ls;
323	int error;
324
325	ls = dlm_find_lockspace_local(proc->lockspace);
326	if (!ls)
327		return -ENOENT;
328
329	error = dlm_user_deadlock(ls, params->flags, params->lkid);
330
331	dlm_put_lockspace(ls);
332	return error;
333}
334
335static int dlm_device_register(struct dlm_ls *ls, char *name)
336{
337	int error, len;
338
339	/* The device is already registered.  This happens when the
340	   lockspace is created multiple times from userspace. */
341	if (ls->ls_device.name)
342		return 0;
343
344	error = -ENOMEM;
345	len = strlen(name) + strlen(name_prefix) + 2;
346	ls->ls_device.name = kzalloc(len, GFP_NOFS);
347	if (!ls->ls_device.name)
348		goto fail;
349
350	snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
351		 name);
352	ls->ls_device.fops = &device_fops;
353	ls->ls_device.minor = MISC_DYNAMIC_MINOR;
354
355	error = misc_register(&ls->ls_device);
356	if (error) {
357		kfree(ls->ls_device.name);
358	}
359fail:
360	return error;
361}
362
363int dlm_device_deregister(struct dlm_ls *ls)
364{
365	int error;
366
367	/* The device is not registered.  This happens when the lockspace
368	   was never used from userspace, or when device_create_lockspace()
369	   calls dlm_release_lockspace() after the register fails. */
370	if (!ls->ls_device.name)
371		return 0;
372
373	error = misc_deregister(&ls->ls_device);
374	if (!error)
375		kfree(ls->ls_device.name);
376	return error;
377}
378
379static int device_user_purge(struct dlm_user_proc *proc,
380			     struct dlm_purge_params *params)
381{
382	struct dlm_ls *ls;
383	int error;
384
385	ls = dlm_find_lockspace_local(proc->lockspace);
386	if (!ls)
387		return -ENOENT;
388
389	error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
390
391	dlm_put_lockspace(ls);
392	return error;
393}
394
395static int device_create_lockspace(struct dlm_lspace_params *params)
396{
397	dlm_lockspace_t *lockspace;
398	struct dlm_ls *ls;
399	int error;
400
401	if (!capable(CAP_SYS_ADMIN))
402		return -EPERM;
403
404	error = dlm_new_lockspace(params->name, NULL, params->flags,
405				  DLM_USER_LVB_LEN, NULL, NULL, NULL,
406				  &lockspace);
407	if (error)
408		return error;
409
410	ls = dlm_find_lockspace_local(lockspace);
411	if (!ls)
412		return -ENOENT;
413
414	error = dlm_device_register(ls, params->name);
415	dlm_put_lockspace(ls);
416
417	if (error)
418		dlm_release_lockspace(lockspace, 0);
419	else
420		error = ls->ls_device.minor;
421
422	return error;
423}
424
425static int device_remove_lockspace(struct dlm_lspace_params *params)
426{
427	dlm_lockspace_t *lockspace;
428	struct dlm_ls *ls;
429	int error, force = 0;
430
431	if (!capable(CAP_SYS_ADMIN))
432		return -EPERM;
433
434	ls = dlm_find_lockspace_device(params->minor);
435	if (!ls)
436		return -ENOENT;
437
438	if (params->flags & DLM_USER_LSFLG_FORCEFREE)
439		force = 2;
440
441	lockspace = ls->ls_local_handle;
442	dlm_put_lockspace(ls);
443
444	/* The final dlm_release_lockspace waits for references to go to
445	   zero, so all processes will need to close their device for the
446	   ls before the release will proceed.  release also calls the
447	   device_deregister above.  Converting a positive return value
448	   from release to zero means that userspace won't know when its
449	   release was the final one, but it shouldn't need to know. */
450
451	error = dlm_release_lockspace(lockspace, force);
452	if (error > 0)
453		error = 0;
454	return error;
455}
456
457/* Check the user's version matches ours */
458static int check_version(struct dlm_write_request *req)
459{
460	if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
461	    (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
462	     req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
463
464		printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
465		       "user (%d.%d.%d) kernel (%d.%d.%d)\n",
466		       current->comm,
467		       task_pid_nr(current),
468		       req->version[0],
469		       req->version[1],
470		       req->version[2],
471		       DLM_DEVICE_VERSION_MAJOR,
472		       DLM_DEVICE_VERSION_MINOR,
473		       DLM_DEVICE_VERSION_PATCH);
474		return -EINVAL;
475	}
476	return 0;
477}
478
479/*
480 * device_write
481 *
482 *   device_user_lock
483 *     dlm_user_request -> request_lock
484 *     dlm_user_convert -> convert_lock
485 *
486 *   device_user_unlock
487 *     dlm_user_unlock -> unlock_lock
488 *     dlm_user_cancel -> cancel_lock
489 *
490 *   device_create_lockspace
491 *     dlm_new_lockspace
492 *
493 *   device_remove_lockspace
494 *     dlm_release_lockspace
495 */
496
497/* a write to a lockspace device is a lock or unlock request, a write
498   to the control device is to create/remove a lockspace */
499
500static ssize_t device_write(struct file *file, const char __user *buf,
501			    size_t count, loff_t *ppos)
502{
503	struct dlm_user_proc *proc = file->private_data;
504	struct dlm_write_request *kbuf;
505	int error;
506
507#ifdef CONFIG_COMPAT
508	if (count < sizeof(struct dlm_write_request32))
509#else
510	if (count < sizeof(struct dlm_write_request))
511#endif
512		return -EINVAL;
513
514	/*
515	 * can't compare against COMPAT/dlm_write_request32 because
516	 * we don't yet know if is64bit is zero
517	 */
518	if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
519		return -EINVAL;
520
521	kbuf = kzalloc(count + 1, GFP_NOFS);
522	if (!kbuf)
523		return -ENOMEM;
524
525	if (copy_from_user(kbuf, buf, count)) {
526		error = -EFAULT;
527		goto out_free;
528	}
529
530	if (check_version(kbuf)) {
531		error = -EBADE;
532		goto out_free;
533	}
534
535#ifdef CONFIG_COMPAT
536	if (!kbuf->is64bit) {
537		struct dlm_write_request32 *k32buf;
538		int namelen = 0;
539
540		if (count > sizeof(struct dlm_write_request32))
541			namelen = count - sizeof(struct dlm_write_request32);
542
543		k32buf = (struct dlm_write_request32 *)kbuf;
544
545		/* add 1 after namelen so that the name string is terminated */
546		kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
547			       GFP_NOFS);
548		if (!kbuf) {
549			kfree(k32buf);
550			return -ENOMEM;
551		}
552
553		if (proc)
554			set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
555
556		compat_input(kbuf, k32buf, namelen);
557		kfree(k32buf);
558	}
559#endif
560
561	/* do we really need this? can a write happen after a close? */
562	if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
563	    (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
564		error = -EINVAL;
565		goto out_free;
566	}
567
568	error = -EINVAL;
569
570	switch (kbuf->cmd)
571	{
572	case DLM_USER_LOCK:
573		if (!proc) {
574			log_print("no locking on control device");
575			goto out_free;
576		}
577		error = device_user_lock(proc, &kbuf->i.lock);
578		break;
579
580	case DLM_USER_UNLOCK:
581		if (!proc) {
582			log_print("no locking on control device");
583			goto out_free;
584		}
585		error = device_user_unlock(proc, &kbuf->i.lock);
586		break;
587
588	case DLM_USER_DEADLOCK:
589		if (!proc) {
590			log_print("no locking on control device");
591			goto out_free;
592		}
593		error = device_user_deadlock(proc, &kbuf->i.lock);
594		break;
595
596	case DLM_USER_CREATE_LOCKSPACE:
597		if (proc) {
598			log_print("create/remove only on control device");
599			goto out_free;
600		}
601		error = device_create_lockspace(&kbuf->i.lspace);
602		break;
603
604	case DLM_USER_REMOVE_LOCKSPACE:
605		if (proc) {
606			log_print("create/remove only on control device");
607			goto out_free;
608		}
609		error = device_remove_lockspace(&kbuf->i.lspace);
610		break;
611
612	case DLM_USER_PURGE:
613		if (!proc) {
614			log_print("no locking on control device");
615			goto out_free;
616		}
617		error = device_user_purge(proc, &kbuf->i.purge);
618		break;
619
620	default:
621		log_print("Unknown command passed to DLM device : %d\n",
622			  kbuf->cmd);
623	}
624
625 out_free:
626	kfree(kbuf);
627	return error;
628}
629
630/* Every process that opens the lockspace device has its own "proc" structure
631   hanging off the open file that's used to keep track of locks owned by the
632   process and asts that need to be delivered to the process. */
633
634static int device_open(struct inode *inode, struct file *file)
635{
636	struct dlm_user_proc *proc;
637	struct dlm_ls *ls;
638
639	ls = dlm_find_lockspace_device(iminor(inode));
640	if (!ls)
641		return -ENOENT;
642
643	proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
644	if (!proc) {
645		dlm_put_lockspace(ls);
646		return -ENOMEM;
647	}
648
649	proc->lockspace = ls->ls_local_handle;
650	INIT_LIST_HEAD(&proc->asts);
651	INIT_LIST_HEAD(&proc->locks);
652	INIT_LIST_HEAD(&proc->unlocking);
653	spin_lock_init(&proc->asts_spin);
654	spin_lock_init(&proc->locks_spin);
655	init_waitqueue_head(&proc->wait);
656	file->private_data = proc;
657
658	return 0;
659}
660
661static int device_close(struct inode *inode, struct file *file)
662{
663	struct dlm_user_proc *proc = file->private_data;
664	struct dlm_ls *ls;
665
666	ls = dlm_find_lockspace_local(proc->lockspace);
667	if (!ls)
668		return -ENOENT;
669
670	set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
671
672	dlm_clear_proc_locks(ls, proc);
673
674	/* at this point no more lkb's should exist for this lockspace,
675	   so there's no chance of dlm_user_add_ast() being called and
676	   looking for lkb->ua->proc */
677
678	kfree(proc);
679	file->private_data = NULL;
680
681	dlm_put_lockspace(ls);
682	dlm_put_lockspace(ls);  /* for the find in device_open() */
683
684	/* FIXME: AUTOFREE: if this ls is no longer used do
685	   device_remove_lockspace() */
686
687	return 0;
688}
689
690static int copy_result_to_user(struct dlm_user_args *ua, int compat,
691			       uint32_t flags, int mode, int copy_lvb,
692			       char __user *buf, size_t count)
693{
694#ifdef CONFIG_COMPAT
695	struct dlm_lock_result32 result32;
696#endif
697	struct dlm_lock_result result;
698	void *resultptr;
699	int error=0;
700	int len;
701	int struct_len;
702
703	memset(&result, 0, sizeof(struct dlm_lock_result));
704	result.version[0] = DLM_DEVICE_VERSION_MAJOR;
705	result.version[1] = DLM_DEVICE_VERSION_MINOR;
706	result.version[2] = DLM_DEVICE_VERSION_PATCH;
707	memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
708	result.user_lksb = ua->user_lksb;
709
710	/* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
711	   in a conversion unless the conversion is successful.  See code
712	   in dlm_user_convert() for updating ua from ua_tmp.  OpenVMS, though,
713	   notes that a new blocking AST address and parameter are set even if
714	   the conversion fails, so maybe we should just do that. */
715
716	if (flags & DLM_CB_BAST) {
717		result.user_astaddr = ua->bastaddr;
718		result.user_astparam = ua->bastparam;
719		result.bast_mode = mode;
720	} else {
721		result.user_astaddr = ua->castaddr;
722		result.user_astparam = ua->castparam;
723	}
724
725#ifdef CONFIG_COMPAT
726	if (compat)
727		len = sizeof(struct dlm_lock_result32);
728	else
729#endif
730		len = sizeof(struct dlm_lock_result);
731	struct_len = len;
732
733	/* copy lvb to userspace if there is one, it's been updated, and
734	   the user buffer has space for it */
735
736	if (copy_lvb && ua->lksb.sb_lvbptr && count >= len + DLM_USER_LVB_LEN) {
737		if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
738				 DLM_USER_LVB_LEN)) {
739			error = -EFAULT;
740			goto out;
741		}
742
743		result.lvb_offset = len;
744		len += DLM_USER_LVB_LEN;
745	}
746
747	result.length = len;
748	resultptr = &result;
749#ifdef CONFIG_COMPAT
750	if (compat) {
751		compat_output(&result, &result32);
752		resultptr = &result32;
753	}
754#endif
755
756	if (copy_to_user(buf, resultptr, struct_len))
757		error = -EFAULT;
758	else
759		error = len;
760 out:
761	return error;
762}
763
764static int copy_version_to_user(char __user *buf, size_t count)
765{
766	struct dlm_device_version ver;
767
768	memset(&ver, 0, sizeof(struct dlm_device_version));
769	ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
770	ver.version[1] = DLM_DEVICE_VERSION_MINOR;
771	ver.version[2] = DLM_DEVICE_VERSION_PATCH;
772
773	if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
774		return -EFAULT;
775	return sizeof(struct dlm_device_version);
776}
777
778/* a read returns a single ast described in a struct dlm_lock_result */
779
780static ssize_t device_read(struct file *file, char __user *buf, size_t count,
781			   loff_t *ppos)
782{
783	struct dlm_user_proc *proc = file->private_data;
784	struct dlm_lkb *lkb;
785	DECLARE_WAITQUEUE(wait, current);
786	struct dlm_callback cb;
787	int rv, resid, copy_lvb = 0;
788
789	if (count == sizeof(struct dlm_device_version)) {
790		rv = copy_version_to_user(buf, count);
791		return rv;
792	}
793
794	if (!proc) {
795		log_print("non-version read from control device %zu", count);
796		return -EINVAL;
797	}
798
799#ifdef CONFIG_COMPAT
800	if (count < sizeof(struct dlm_lock_result32))
801#else
802	if (count < sizeof(struct dlm_lock_result))
803#endif
804		return -EINVAL;
805
806 try_another:
807
808	/* do we really need this? can a read happen after a close? */
809	if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
810		return -EINVAL;
811
812	spin_lock(&proc->asts_spin);
813	if (list_empty(&proc->asts)) {
814		if (file->f_flags & O_NONBLOCK) {
815			spin_unlock(&proc->asts_spin);
816			return -EAGAIN;
817		}
818
819		add_wait_queue(&proc->wait, &wait);
820
821	repeat:
822		set_current_state(TASK_INTERRUPTIBLE);
823		if (list_empty(&proc->asts) && !signal_pending(current)) {
824			spin_unlock(&proc->asts_spin);
825			schedule();
826			spin_lock(&proc->asts_spin);
827			goto repeat;
828		}
829		set_current_state(TASK_RUNNING);
830		remove_wait_queue(&proc->wait, &wait);
831
832		if (signal_pending(current)) {
833			spin_unlock(&proc->asts_spin);
834			return -ERESTARTSYS;
835		}
836	}
837
838	/* if we empty lkb_callbacks, we don't want to unlock the spinlock
839	   without removing lkb_cb_list; so empty lkb_cb_list is always
840	   consistent with empty lkb_callbacks */
841
842	lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_cb_list);
843
844	rv = dlm_rem_lkb_callback(lkb->lkb_resource->res_ls, lkb, &cb, &resid);
845	if (rv < 0) {
846		/* this shouldn't happen; lkb should have been removed from
847		   list when resid was zero */
848		log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
849		list_del_init(&lkb->lkb_cb_list);
850		spin_unlock(&proc->asts_spin);
851		/* removes ref for proc->asts, may cause lkb to be freed */
852		dlm_put_lkb(lkb);
853		goto try_another;
854	}
855	if (!resid)
856		list_del_init(&lkb->lkb_cb_list);
857	spin_unlock(&proc->asts_spin);
858
859	if (cb.flags & DLM_CB_SKIP) {
860		/* removes ref for proc->asts, may cause lkb to be freed */
861		if (!resid)
862			dlm_put_lkb(lkb);
863		goto try_another;
864	}
865
866	if (cb.flags & DLM_CB_CAST) {
867		int old_mode, new_mode;
868
869		old_mode = lkb->lkb_last_cast.mode;
870		new_mode = cb.mode;
871
872		if (!cb.sb_status && lkb->lkb_lksb->sb_lvbptr &&
873		    dlm_lvb_operations[old_mode + 1][new_mode + 1])
874			copy_lvb = 1;
875
876		lkb->lkb_lksb->sb_status = cb.sb_status;
877		lkb->lkb_lksb->sb_flags = cb.sb_flags;
878	}
879
880	rv = copy_result_to_user(lkb->lkb_ua,
881				 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
882				 cb.flags, cb.mode, copy_lvb, buf, count);
883
884	/* removes ref for proc->asts, may cause lkb to be freed */
885	if (!resid)
886		dlm_put_lkb(lkb);
887
888	return rv;
889}
890
891static unsigned int device_poll(struct file *file, poll_table *wait)
892{
893	struct dlm_user_proc *proc = file->private_data;
894
895	poll_wait(file, &proc->wait, wait);
896
897	spin_lock(&proc->asts_spin);
898	if (!list_empty(&proc->asts)) {
899		spin_unlock(&proc->asts_spin);
900		return POLLIN | POLLRDNORM;
901	}
902	spin_unlock(&proc->asts_spin);
903	return 0;
904}
905
906int dlm_user_daemon_available(void)
907{
908	/* dlm_controld hasn't started (or, has started, but not
909	   properly populated configfs) */
910
911	if (!dlm_our_nodeid())
912		return 0;
913
914	/* This is to deal with versions of dlm_controld that don't
915	   know about the monitor device.  We assume that if the
916	   dlm_controld was started (above), but the monitor device
917	   was never opened, that it's an old version.  dlm_controld
918	   should open the monitor device before populating configfs. */
919
920	if (dlm_monitor_unused)
921		return 1;
922
923	return atomic_read(&dlm_monitor_opened) ? 1 : 0;
924}
925
926static int ctl_device_open(struct inode *inode, struct file *file)
927{
928	file->private_data = NULL;
929	return 0;
930}
931
932static int ctl_device_close(struct inode *inode, struct file *file)
933{
934	return 0;
935}
936
937static int monitor_device_open(struct inode *inode, struct file *file)
938{
939	atomic_inc(&dlm_monitor_opened);
940	dlm_monitor_unused = 0;
941	return 0;
942}
943
944static int monitor_device_close(struct inode *inode, struct file *file)
945{
946	if (atomic_dec_and_test(&dlm_monitor_opened))
947		dlm_stop_lockspaces();
948	return 0;
949}
950
951static const struct file_operations device_fops = {
952	.open    = device_open,
953	.release = device_close,
954	.read    = device_read,
955	.write   = device_write,
956	.poll    = device_poll,
957	.owner   = THIS_MODULE,
958	.llseek  = noop_llseek,
959};
960
961static const struct file_operations ctl_device_fops = {
962	.open    = ctl_device_open,
963	.release = ctl_device_close,
964	.read    = device_read,
965	.write   = device_write,
966	.owner   = THIS_MODULE,
967	.llseek  = noop_llseek,
968};
969
970static struct miscdevice ctl_device = {
971	.name  = "dlm-control",
972	.fops  = &ctl_device_fops,
973	.minor = MISC_DYNAMIC_MINOR,
974};
975
976static const struct file_operations monitor_device_fops = {
977	.open    = monitor_device_open,
978	.release = monitor_device_close,
979	.owner   = THIS_MODULE,
980	.llseek  = noop_llseek,
981};
982
983static struct miscdevice monitor_device = {
984	.name  = "dlm-monitor",
985	.fops  = &monitor_device_fops,
986	.minor = MISC_DYNAMIC_MINOR,
987};
988
989int __init dlm_user_init(void)
990{
991	int error;
992
993	atomic_set(&dlm_monitor_opened, 0);
994
995	error = misc_register(&ctl_device);
996	if (error) {
997		log_print("misc_register failed for control device");
998		goto out;
999	}
1000
1001	error = misc_register(&monitor_device);
1002	if (error) {
1003		log_print("misc_register failed for monitor device");
1004		misc_deregister(&ctl_device);
1005	}
1006 out:
1007	return error;
1008}
1009
1010void dlm_user_exit(void)
1011{
1012	misc_deregister(&ctl_device);
1013	misc_deregister(&monitor_device);
1014}
1015
1016