xref: /openbmc/linux/fs/lockd/svc.c (revision 3932b9ca)
1 /*
2  * linux/fs/lockd/svc.c
3  *
4  * This is the central lockd service.
5  *
6  * FIXME: Separate the lockd NFS server functionality from the lockd NFS
7  * 	  client functionality. Oh why didn't Sun create two separate
8  *	  services in the first place?
9  *
10  * Authors:	Olaf Kirch (okir@monad.swb.de)
11  *
12  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
13  */
14 
15 #include <linux/module.h>
16 #include <linux/init.h>
17 #include <linux/sysctl.h>
18 #include <linux/moduleparam.h>
19 
20 #include <linux/sched.h>
21 #include <linux/errno.h>
22 #include <linux/in.h>
23 #include <linux/uio.h>
24 #include <linux/smp.h>
25 #include <linux/mutex.h>
26 #include <linux/kthread.h>
27 #include <linux/freezer.h>
28 
29 #include <linux/sunrpc/types.h>
30 #include <linux/sunrpc/stats.h>
31 #include <linux/sunrpc/clnt.h>
32 #include <linux/sunrpc/svc.h>
33 #include <linux/sunrpc/svcsock.h>
34 #include <net/ip.h>
35 #include <linux/lockd/lockd.h>
36 #include <linux/nfs.h>
37 
38 #include "netns.h"
39 
40 #define NLMDBG_FACILITY		NLMDBG_SVC
41 #define LOCKD_BUFSIZE		(1024 + NLMSVC_XDRSIZE)
42 #define ALLOWED_SIGS		(sigmask(SIGKILL))
43 
44 static struct svc_program	nlmsvc_program;
45 
46 struct nlmsvc_binding *		nlmsvc_ops;
47 EXPORT_SYMBOL_GPL(nlmsvc_ops);
48 
49 static DEFINE_MUTEX(nlmsvc_mutex);
50 static unsigned int		nlmsvc_users;
51 static struct task_struct	*nlmsvc_task;
52 static struct svc_rqst		*nlmsvc_rqst;
53 unsigned long			nlmsvc_timeout;
54 
55 int lockd_net_id;
56 
57 /*
58  * These can be set at insmod time (useful for NFS as root filesystem),
59  * and also changed through the sysctl interface.  -- Jamie Lokier, Aug 2003
60  */
61 static unsigned long		nlm_grace_period;
62 static unsigned long		nlm_timeout = LOCKD_DFLT_TIMEO;
63 static int			nlm_udpport, nlm_tcpport;
64 
65 /* RLIM_NOFILE defaults to 1024. That seems like a reasonable default here. */
66 static unsigned int		nlm_max_connections = 1024;
67 
68 /*
69  * Constants needed for the sysctl interface.
70  */
71 static const unsigned long	nlm_grace_period_min = 0;
72 static const unsigned long	nlm_grace_period_max = 240;
73 static const unsigned long	nlm_timeout_min = 3;
74 static const unsigned long	nlm_timeout_max = 20;
75 static const int		nlm_port_min = 0, nlm_port_max = 65535;
76 
77 #ifdef CONFIG_SYSCTL
78 static struct ctl_table_header * nlm_sysctl_table;
79 #endif
80 
81 static unsigned long get_lockd_grace_period(void)
82 {
83 	/* Note: nlm_timeout should always be nonzero */
84 	if (nlm_grace_period)
85 		return roundup(nlm_grace_period, nlm_timeout) * HZ;
86 	else
87 		return nlm_timeout * 5 * HZ;
88 }
89 
90 static void grace_ender(struct work_struct *grace)
91 {
92 	struct delayed_work *dwork = container_of(grace, struct delayed_work,
93 						  work);
94 	struct lockd_net *ln = container_of(dwork, struct lockd_net,
95 					    grace_period_end);
96 
97 	locks_end_grace(&ln->lockd_manager);
98 }
99 
100 static void set_grace_period(struct net *net)
101 {
102 	unsigned long grace_period = get_lockd_grace_period();
103 	struct lockd_net *ln = net_generic(net, lockd_net_id);
104 
105 	locks_start_grace(net, &ln->lockd_manager);
106 	cancel_delayed_work_sync(&ln->grace_period_end);
107 	schedule_delayed_work(&ln->grace_period_end, grace_period);
108 }
109 
110 static void restart_grace(void)
111 {
112 	if (nlmsvc_ops) {
113 		struct net *net = &init_net;
114 		struct lockd_net *ln = net_generic(net, lockd_net_id);
115 
116 		cancel_delayed_work_sync(&ln->grace_period_end);
117 		locks_end_grace(&ln->lockd_manager);
118 		nlmsvc_invalidate_all();
119 		set_grace_period(net);
120 	}
121 }
122 
123 /*
124  * This is the lockd kernel thread
125  */
126 static int
127 lockd(void *vrqstp)
128 {
129 	int		err = 0;
130 	struct svc_rqst *rqstp = vrqstp;
131 
132 	/* try_to_freeze() is called from svc_recv() */
133 	set_freezable();
134 
135 	/* Allow SIGKILL to tell lockd to drop all of its locks */
136 	allow_signal(SIGKILL);
137 
138 	dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
139 
140 	if (!nlm_timeout)
141 		nlm_timeout = LOCKD_DFLT_TIMEO;
142 	nlmsvc_timeout = nlm_timeout * HZ;
143 
144 	/*
145 	 * The main request loop. We don't terminate until the last
146 	 * NFS mount or NFS daemon has gone away.
147 	 */
148 	while (!kthread_should_stop()) {
149 		long timeout = MAX_SCHEDULE_TIMEOUT;
150 		RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
151 
152 		/* update sv_maxconn if it has changed */
153 		rqstp->rq_server->sv_maxconn = nlm_max_connections;
154 
155 		if (signalled()) {
156 			flush_signals(current);
157 			restart_grace();
158 			continue;
159 		}
160 
161 		timeout = nlmsvc_retry_blocked();
162 
163 		/*
164 		 * Find a socket with data available and call its
165 		 * recvfrom routine.
166 		 */
167 		err = svc_recv(rqstp, timeout);
168 		if (err == -EAGAIN || err == -EINTR)
169 			continue;
170 		dprintk("lockd: request from %s\n",
171 				svc_print_addr(rqstp, buf, sizeof(buf)));
172 
173 		svc_process(rqstp);
174 	}
175 	flush_signals(current);
176 	if (nlmsvc_ops)
177 		nlmsvc_invalidate_all();
178 	nlm_shutdown_hosts();
179 	return 0;
180 }
181 
182 static int create_lockd_listener(struct svc_serv *serv, const char *name,
183 				 struct net *net, const int family,
184 				 const unsigned short port)
185 {
186 	struct svc_xprt *xprt;
187 
188 	xprt = svc_find_xprt(serv, name, net, family, 0);
189 	if (xprt == NULL)
190 		return svc_create_xprt(serv, name, net, family, port,
191 						SVC_SOCK_DEFAULTS);
192 	svc_xprt_put(xprt);
193 	return 0;
194 }
195 
196 static int create_lockd_family(struct svc_serv *serv, struct net *net,
197 			       const int family)
198 {
199 	int err;
200 
201 	err = create_lockd_listener(serv, "udp", net, family, nlm_udpport);
202 	if (err < 0)
203 		return err;
204 
205 	return create_lockd_listener(serv, "tcp", net, family, nlm_tcpport);
206 }
207 
208 /*
209  * Ensure there are active UDP and TCP listeners for lockd.
210  *
211  * Even if we have only TCP NFS mounts and/or TCP NFSDs, some
212  * local services (such as rpc.statd) still require UDP, and
213  * some NFS servers do not yet support NLM over TCP.
214  *
215  * Returns zero if all listeners are available; otherwise a
216  * negative errno value is returned.
217  */
218 static int make_socks(struct svc_serv *serv, struct net *net)
219 {
220 	static int warned;
221 	int err;
222 
223 	err = create_lockd_family(serv, net, PF_INET);
224 	if (err < 0)
225 		goto out_err;
226 
227 	err = create_lockd_family(serv, net, PF_INET6);
228 	if (err < 0 && err != -EAFNOSUPPORT)
229 		goto out_err;
230 
231 	warned = 0;
232 	return 0;
233 
234 out_err:
235 	if (warned++ == 0)
236 		printk(KERN_WARNING
237 			"lockd_up: makesock failed, error=%d\n", err);
238 	svc_shutdown_net(serv, net);
239 	return err;
240 }
241 
242 static int lockd_up_net(struct svc_serv *serv, struct net *net)
243 {
244 	struct lockd_net *ln = net_generic(net, lockd_net_id);
245 	int error;
246 
247 	if (ln->nlmsvc_users++)
248 		return 0;
249 
250 	error = svc_bind(serv, net);
251 	if (error)
252 		goto err_bind;
253 
254 	error = make_socks(serv, net);
255 	if (error < 0)
256 		goto err_bind;
257 	set_grace_period(net);
258 	dprintk("lockd_up_net: per-net data created; net=%p\n", net);
259 	return 0;
260 
261 err_bind:
262 	ln->nlmsvc_users--;
263 	return error;
264 }
265 
266 static void lockd_down_net(struct svc_serv *serv, struct net *net)
267 {
268 	struct lockd_net *ln = net_generic(net, lockd_net_id);
269 
270 	if (ln->nlmsvc_users) {
271 		if (--ln->nlmsvc_users == 0) {
272 			nlm_shutdown_hosts_net(net);
273 			cancel_delayed_work_sync(&ln->grace_period_end);
274 			locks_end_grace(&ln->lockd_manager);
275 			svc_shutdown_net(serv, net);
276 			dprintk("lockd_down_net: per-net data destroyed; net=%p\n", net);
277 		}
278 	} else {
279 		printk(KERN_ERR "lockd_down_net: no users! task=%p, net=%p\n",
280 				nlmsvc_task, net);
281 		BUG();
282 	}
283 }
284 
285 static int lockd_start_svc(struct svc_serv *serv)
286 {
287 	int error;
288 
289 	if (nlmsvc_rqst)
290 		return 0;
291 
292 	/*
293 	 * Create the kernel thread and wait for it to start.
294 	 */
295 	nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0], NUMA_NO_NODE);
296 	if (IS_ERR(nlmsvc_rqst)) {
297 		error = PTR_ERR(nlmsvc_rqst);
298 		printk(KERN_WARNING
299 			"lockd_up: svc_rqst allocation failed, error=%d\n",
300 			error);
301 		goto out_rqst;
302 	}
303 
304 	svc_sock_update_bufs(serv);
305 	serv->sv_maxconn = nlm_max_connections;
306 
307 	nlmsvc_task = kthread_run(lockd, nlmsvc_rqst, "%s", serv->sv_name);
308 	if (IS_ERR(nlmsvc_task)) {
309 		error = PTR_ERR(nlmsvc_task);
310 		printk(KERN_WARNING
311 			"lockd_up: kthread_run failed, error=%d\n", error);
312 		goto out_task;
313 	}
314 	dprintk("lockd_up: service started\n");
315 	return 0;
316 
317 out_task:
318 	svc_exit_thread(nlmsvc_rqst);
319 	nlmsvc_task = NULL;
320 out_rqst:
321 	nlmsvc_rqst = NULL;
322 	return error;
323 }
324 
325 static struct svc_serv *lockd_create_svc(void)
326 {
327 	struct svc_serv *serv;
328 
329 	/*
330 	 * Check whether we're already up and running.
331 	 */
332 	if (nlmsvc_rqst) {
333 		/*
334 		 * Note: increase service usage, because later in case of error
335 		 * svc_destroy() will be called.
336 		 */
337 		svc_get(nlmsvc_rqst->rq_server);
338 		return nlmsvc_rqst->rq_server;
339 	}
340 
341 	/*
342 	 * Sanity check: if there's no pid,
343 	 * we should be the first user ...
344 	 */
345 	if (nlmsvc_users)
346 		printk(KERN_WARNING
347 			"lockd_up: no pid, %d users??\n", nlmsvc_users);
348 
349 	serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL);
350 	if (!serv) {
351 		printk(KERN_WARNING "lockd_up: create service failed\n");
352 		return ERR_PTR(-ENOMEM);
353 	}
354 	dprintk("lockd_up: service created\n");
355 	return serv;
356 }
357 
358 /*
359  * Bring up the lockd process if it's not already up.
360  */
361 int lockd_up(struct net *net)
362 {
363 	struct svc_serv *serv;
364 	int error;
365 
366 	mutex_lock(&nlmsvc_mutex);
367 
368 	serv = lockd_create_svc();
369 	if (IS_ERR(serv)) {
370 		error = PTR_ERR(serv);
371 		goto err_create;
372 	}
373 
374 	error = lockd_up_net(serv, net);
375 	if (error < 0)
376 		goto err_net;
377 
378 	error = lockd_start_svc(serv);
379 	if (error < 0)
380 		goto err_start;
381 
382 	nlmsvc_users++;
383 	/*
384 	 * Note: svc_serv structures have an initial use count of 1,
385 	 * so we exit through here on both success and failure.
386 	 */
387 err_net:
388 	svc_destroy(serv);
389 err_create:
390 	mutex_unlock(&nlmsvc_mutex);
391 	return error;
392 
393 err_start:
394 	lockd_down_net(serv, net);
395 	goto err_net;
396 }
397 EXPORT_SYMBOL_GPL(lockd_up);
398 
399 /*
400  * Decrement the user count and bring down lockd if we're the last.
401  */
402 void
403 lockd_down(struct net *net)
404 {
405 	mutex_lock(&nlmsvc_mutex);
406 	lockd_down_net(nlmsvc_rqst->rq_server, net);
407 	if (nlmsvc_users) {
408 		if (--nlmsvc_users)
409 			goto out;
410 	} else {
411 		printk(KERN_ERR "lockd_down: no users! task=%p\n",
412 			nlmsvc_task);
413 		BUG();
414 	}
415 
416 	if (!nlmsvc_task) {
417 		printk(KERN_ERR "lockd_down: no lockd running.\n");
418 		BUG();
419 	}
420 	kthread_stop(nlmsvc_task);
421 	dprintk("lockd_down: service stopped\n");
422 	svc_exit_thread(nlmsvc_rqst);
423 	dprintk("lockd_down: service destroyed\n");
424 	nlmsvc_task = NULL;
425 	nlmsvc_rqst = NULL;
426 out:
427 	mutex_unlock(&nlmsvc_mutex);
428 }
429 EXPORT_SYMBOL_GPL(lockd_down);
430 
431 #ifdef CONFIG_SYSCTL
432 
433 /*
434  * Sysctl parameters (same as module parameters, different interface).
435  */
436 
437 static struct ctl_table nlm_sysctls[] = {
438 	{
439 		.procname	= "nlm_grace_period",
440 		.data		= &nlm_grace_period,
441 		.maxlen		= sizeof(unsigned long),
442 		.mode		= 0644,
443 		.proc_handler	= proc_doulongvec_minmax,
444 		.extra1		= (unsigned long *) &nlm_grace_period_min,
445 		.extra2		= (unsigned long *) &nlm_grace_period_max,
446 	},
447 	{
448 		.procname	= "nlm_timeout",
449 		.data		= &nlm_timeout,
450 		.maxlen		= sizeof(unsigned long),
451 		.mode		= 0644,
452 		.proc_handler	= proc_doulongvec_minmax,
453 		.extra1		= (unsigned long *) &nlm_timeout_min,
454 		.extra2		= (unsigned long *) &nlm_timeout_max,
455 	},
456 	{
457 		.procname	= "nlm_udpport",
458 		.data		= &nlm_udpport,
459 		.maxlen		= sizeof(int),
460 		.mode		= 0644,
461 		.proc_handler	= proc_dointvec_minmax,
462 		.extra1		= (int *) &nlm_port_min,
463 		.extra2		= (int *) &nlm_port_max,
464 	},
465 	{
466 		.procname	= "nlm_tcpport",
467 		.data		= &nlm_tcpport,
468 		.maxlen		= sizeof(int),
469 		.mode		= 0644,
470 		.proc_handler	= proc_dointvec_minmax,
471 		.extra1		= (int *) &nlm_port_min,
472 		.extra2		= (int *) &nlm_port_max,
473 	},
474 	{
475 		.procname	= "nsm_use_hostnames",
476 		.data		= &nsm_use_hostnames,
477 		.maxlen		= sizeof(int),
478 		.mode		= 0644,
479 		.proc_handler	= proc_dointvec,
480 	},
481 	{
482 		.procname	= "nsm_local_state",
483 		.data		= &nsm_local_state,
484 		.maxlen		= sizeof(int),
485 		.mode		= 0644,
486 		.proc_handler	= proc_dointvec,
487 	},
488 	{ }
489 };
490 
491 static struct ctl_table nlm_sysctl_dir[] = {
492 	{
493 		.procname	= "nfs",
494 		.mode		= 0555,
495 		.child		= nlm_sysctls,
496 	},
497 	{ }
498 };
499 
500 static struct ctl_table nlm_sysctl_root[] = {
501 	{
502 		.procname	= "fs",
503 		.mode		= 0555,
504 		.child		= nlm_sysctl_dir,
505 	},
506 	{ }
507 };
508 
509 #endif	/* CONFIG_SYSCTL */
510 
511 /*
512  * Module (and sysfs) parameters.
513  */
514 
515 #define param_set_min_max(name, type, which_strtol, min, max)		\
516 static int param_set_##name(const char *val, struct kernel_param *kp)	\
517 {									\
518 	char *endp;							\
519 	__typeof__(type) num = which_strtol(val, &endp, 0);		\
520 	if (endp == val || *endp || num < (min) || num > (max))		\
521 		return -EINVAL;						\
522 	*((type *) kp->arg) = num;					\
523 	return 0;							\
524 }
525 
526 static inline int is_callback(u32 proc)
527 {
528 	return proc == NLMPROC_GRANTED
529 		|| proc == NLMPROC_GRANTED_MSG
530 		|| proc == NLMPROC_TEST_RES
531 		|| proc == NLMPROC_LOCK_RES
532 		|| proc == NLMPROC_CANCEL_RES
533 		|| proc == NLMPROC_UNLOCK_RES
534 		|| proc == NLMPROC_NSM_NOTIFY;
535 }
536 
537 
538 static int lockd_authenticate(struct svc_rqst *rqstp)
539 {
540 	rqstp->rq_client = NULL;
541 	switch (rqstp->rq_authop->flavour) {
542 		case RPC_AUTH_NULL:
543 		case RPC_AUTH_UNIX:
544 			if (rqstp->rq_proc == 0)
545 				return SVC_OK;
546 			if (is_callback(rqstp->rq_proc)) {
547 				/* Leave it to individual procedures to
548 				 * call nlmsvc_lookup_host(rqstp)
549 				 */
550 				return SVC_OK;
551 			}
552 			return svc_set_client(rqstp);
553 	}
554 	return SVC_DENIED;
555 }
556 
557 
558 param_set_min_max(port, int, simple_strtol, 0, 65535)
559 param_set_min_max(grace_period, unsigned long, simple_strtoul,
560 		  nlm_grace_period_min, nlm_grace_period_max)
561 param_set_min_max(timeout, unsigned long, simple_strtoul,
562 		  nlm_timeout_min, nlm_timeout_max)
563 
564 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
565 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
566 MODULE_LICENSE("GPL");
567 
568 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
569 		  &nlm_grace_period, 0644);
570 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
571 		  &nlm_timeout, 0644);
572 module_param_call(nlm_udpport, param_set_port, param_get_int,
573 		  &nlm_udpport, 0644);
574 module_param_call(nlm_tcpport, param_set_port, param_get_int,
575 		  &nlm_tcpport, 0644);
576 module_param(nsm_use_hostnames, bool, 0644);
577 module_param(nlm_max_connections, uint, 0644);
578 
579 static int lockd_init_net(struct net *net)
580 {
581 	struct lockd_net *ln = net_generic(net, lockd_net_id);
582 
583 	INIT_DELAYED_WORK(&ln->grace_period_end, grace_ender);
584 	INIT_LIST_HEAD(&ln->grace_list);
585 	spin_lock_init(&ln->nsm_clnt_lock);
586 	return 0;
587 }
588 
589 static void lockd_exit_net(struct net *net)
590 {
591 }
592 
593 static struct pernet_operations lockd_net_ops = {
594 	.init = lockd_init_net,
595 	.exit = lockd_exit_net,
596 	.id = &lockd_net_id,
597 	.size = sizeof(struct lockd_net),
598 };
599 
600 
601 /*
602  * Initialising and terminating the module.
603  */
604 
605 static int __init init_nlm(void)
606 {
607 	int err;
608 
609 #ifdef CONFIG_SYSCTL
610 	err = -ENOMEM;
611 	nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
612 	if (nlm_sysctl_table == NULL)
613 		goto err_sysctl;
614 #endif
615 	err = register_pernet_subsys(&lockd_net_ops);
616 	if (err)
617 		goto err_pernet;
618 	return 0;
619 
620 err_pernet:
621 #ifdef CONFIG_SYSCTL
622 	unregister_sysctl_table(nlm_sysctl_table);
623 err_sysctl:
624 #endif
625 	return err;
626 }
627 
628 static void __exit exit_nlm(void)
629 {
630 	/* FIXME: delete all NLM clients */
631 	nlm_shutdown_hosts();
632 	unregister_pernet_subsys(&lockd_net_ops);
633 #ifdef CONFIG_SYSCTL
634 	unregister_sysctl_table(nlm_sysctl_table);
635 #endif
636 }
637 
638 module_init(init_nlm);
639 module_exit(exit_nlm);
640 
641 /*
642  * Define NLM program and procedures
643  */
644 static struct svc_version	nlmsvc_version1 = {
645 		.vs_vers	= 1,
646 		.vs_nproc	= 17,
647 		.vs_proc	= nlmsvc_procedures,
648 		.vs_xdrsize	= NLMSVC_XDRSIZE,
649 };
650 static struct svc_version	nlmsvc_version3 = {
651 		.vs_vers	= 3,
652 		.vs_nproc	= 24,
653 		.vs_proc	= nlmsvc_procedures,
654 		.vs_xdrsize	= NLMSVC_XDRSIZE,
655 };
656 #ifdef CONFIG_LOCKD_V4
657 static struct svc_version	nlmsvc_version4 = {
658 		.vs_vers	= 4,
659 		.vs_nproc	= 24,
660 		.vs_proc	= nlmsvc_procedures4,
661 		.vs_xdrsize	= NLMSVC_XDRSIZE,
662 };
663 #endif
664 static struct svc_version *	nlmsvc_version[] = {
665 	[1] = &nlmsvc_version1,
666 	[3] = &nlmsvc_version3,
667 #ifdef CONFIG_LOCKD_V4
668 	[4] = &nlmsvc_version4,
669 #endif
670 };
671 
672 static struct svc_stat		nlmsvc_stats;
673 
674 #define NLM_NRVERS	ARRAY_SIZE(nlmsvc_version)
675 static struct svc_program	nlmsvc_program = {
676 	.pg_prog		= NLM_PROGRAM,		/* program number */
677 	.pg_nvers		= NLM_NRVERS,		/* number of entries in nlmsvc_version */
678 	.pg_vers		= nlmsvc_version,	/* version table */
679 	.pg_name		= "lockd",		/* service name */
680 	.pg_class		= "nfsd",		/* share authentication with nfsd */
681 	.pg_stats		= &nlmsvc_stats,	/* stats table */
682 	.pg_authenticate = &lockd_authenticate	/* export authentication */
683 };
684