xref: /openbmc/linux/fs/lockd/svc.c (revision c87fb4a3)
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 #include "procfs.h"
40 
41 #define NLMDBG_FACILITY		NLMDBG_SVC
42 #define LOCKD_BUFSIZE		(1024 + NLMSVC_XDRSIZE)
43 #define ALLOWED_SIGS		(sigmask(SIGKILL))
44 
45 static struct svc_program	nlmsvc_program;
46 
47 struct nlmsvc_binding *		nlmsvc_ops;
48 EXPORT_SYMBOL_GPL(nlmsvc_ops);
49 
50 static DEFINE_MUTEX(nlmsvc_mutex);
51 static unsigned int		nlmsvc_users;
52 static struct task_struct	*nlmsvc_task;
53 static struct svc_rqst		*nlmsvc_rqst;
54 unsigned long			nlmsvc_timeout;
55 
56 int lockd_net_id;
57 
58 /*
59  * These can be set at insmod time (useful for NFS as root filesystem),
60  * and also changed through the sysctl interface.  -- Jamie Lokier, Aug 2003
61  */
62 static unsigned long		nlm_grace_period;
63 static unsigned long		nlm_timeout = LOCKD_DFLT_TIMEO;
64 static int			nlm_udpport, nlm_tcpport;
65 
66 /* RLIM_NOFILE defaults to 1024. That seems like a reasonable default here. */
67 static unsigned int		nlm_max_connections = 1024;
68 
69 /*
70  * Constants needed for the sysctl interface.
71  */
72 static const unsigned long	nlm_grace_period_min = 0;
73 static const unsigned long	nlm_grace_period_max = 240;
74 static const unsigned long	nlm_timeout_min = 3;
75 static const unsigned long	nlm_timeout_max = 20;
76 static const int		nlm_port_min = 0, nlm_port_max = 65535;
77 
78 #ifdef CONFIG_SYSCTL
79 static struct ctl_table_header * nlm_sysctl_table;
80 #endif
81 
82 static unsigned long get_lockd_grace_period(void)
83 {
84 	/* Note: nlm_timeout should always be nonzero */
85 	if (nlm_grace_period)
86 		return roundup(nlm_grace_period, nlm_timeout) * HZ;
87 	else
88 		return nlm_timeout * 5 * HZ;
89 }
90 
91 static void grace_ender(struct work_struct *grace)
92 {
93 	struct delayed_work *dwork = container_of(grace, struct delayed_work,
94 						  work);
95 	struct lockd_net *ln = container_of(dwork, struct lockd_net,
96 					    grace_period_end);
97 
98 	locks_end_grace(&ln->lockd_manager);
99 }
100 
101 static void set_grace_period(struct net *net)
102 {
103 	unsigned long grace_period = get_lockd_grace_period();
104 	struct lockd_net *ln = net_generic(net, lockd_net_id);
105 
106 	locks_start_grace(net, &ln->lockd_manager);
107 	cancel_delayed_work_sync(&ln->grace_period_end);
108 	schedule_delayed_work(&ln->grace_period_end, grace_period);
109 }
110 
111 static void restart_grace(void)
112 {
113 	if (nlmsvc_ops) {
114 		struct net *net = &init_net;
115 		struct lockd_net *ln = net_generic(net, lockd_net_id);
116 
117 		cancel_delayed_work_sync(&ln->grace_period_end);
118 		locks_end_grace(&ln->lockd_manager);
119 		nlmsvc_invalidate_all();
120 		set_grace_period(net);
121 	}
122 }
123 
124 /*
125  * This is the lockd kernel thread
126  */
127 static int
128 lockd(void *vrqstp)
129 {
130 	int		err = 0;
131 	struct svc_rqst *rqstp = vrqstp;
132 
133 	/* try_to_freeze() is called from svc_recv() */
134 	set_freezable();
135 
136 	/* Allow SIGKILL to tell lockd to drop all of its locks */
137 	allow_signal(SIGKILL);
138 
139 	dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
140 
141 	/*
142 	 * The main request loop. We don't terminate until the last
143 	 * NFS mount or NFS daemon has gone away.
144 	 */
145 	while (!kthread_should_stop()) {
146 		long timeout = MAX_SCHEDULE_TIMEOUT;
147 		RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
148 
149 		/* update sv_maxconn if it has changed */
150 		rqstp->rq_server->sv_maxconn = nlm_max_connections;
151 
152 		if (signalled()) {
153 			flush_signals(current);
154 			restart_grace();
155 			continue;
156 		}
157 
158 		timeout = nlmsvc_retry_blocked();
159 
160 		/*
161 		 * Find a socket with data available and call its
162 		 * recvfrom routine.
163 		 */
164 		err = svc_recv(rqstp, timeout);
165 		if (err == -EAGAIN || err == -EINTR)
166 			continue;
167 		dprintk("lockd: request from %s\n",
168 				svc_print_addr(rqstp, buf, sizeof(buf)));
169 
170 		svc_process(rqstp);
171 	}
172 	flush_signals(current);
173 	if (nlmsvc_ops)
174 		nlmsvc_invalidate_all();
175 	nlm_shutdown_hosts();
176 	return 0;
177 }
178 
179 static int create_lockd_listener(struct svc_serv *serv, const char *name,
180 				 struct net *net, const int family,
181 				 const unsigned short port)
182 {
183 	struct svc_xprt *xprt;
184 
185 	xprt = svc_find_xprt(serv, name, net, family, 0);
186 	if (xprt == NULL)
187 		return svc_create_xprt(serv, name, net, family, port,
188 						SVC_SOCK_DEFAULTS);
189 	svc_xprt_put(xprt);
190 	return 0;
191 }
192 
193 static int create_lockd_family(struct svc_serv *serv, struct net *net,
194 			       const int family)
195 {
196 	int err;
197 
198 	err = create_lockd_listener(serv, "udp", net, family, nlm_udpport);
199 	if (err < 0)
200 		return err;
201 
202 	return create_lockd_listener(serv, "tcp", net, family, nlm_tcpport);
203 }
204 
205 /*
206  * Ensure there are active UDP and TCP listeners for lockd.
207  *
208  * Even if we have only TCP NFS mounts and/or TCP NFSDs, some
209  * local services (such as rpc.statd) still require UDP, and
210  * some NFS servers do not yet support NLM over TCP.
211  *
212  * Returns zero if all listeners are available; otherwise a
213  * negative errno value is returned.
214  */
215 static int make_socks(struct svc_serv *serv, struct net *net)
216 {
217 	static int warned;
218 	int err;
219 
220 	err = create_lockd_family(serv, net, PF_INET);
221 	if (err < 0)
222 		goto out_err;
223 
224 	err = create_lockd_family(serv, net, PF_INET6);
225 	if (err < 0 && err != -EAFNOSUPPORT)
226 		goto out_err;
227 
228 	warned = 0;
229 	return 0;
230 
231 out_err:
232 	if (warned++ == 0)
233 		printk(KERN_WARNING
234 			"lockd_up: makesock failed, error=%d\n", err);
235 	svc_shutdown_net(serv, net);
236 	return err;
237 }
238 
239 static int lockd_up_net(struct svc_serv *serv, struct net *net)
240 {
241 	struct lockd_net *ln = net_generic(net, lockd_net_id);
242 	int error;
243 
244 	if (ln->nlmsvc_users++)
245 		return 0;
246 
247 	error = svc_bind(serv, net);
248 	if (error)
249 		goto err_bind;
250 
251 	error = make_socks(serv, net);
252 	if (error < 0)
253 		goto err_bind;
254 	set_grace_period(net);
255 	dprintk("lockd_up_net: per-net data created; net=%p\n", net);
256 	return 0;
257 
258 err_bind:
259 	ln->nlmsvc_users--;
260 	return error;
261 }
262 
263 static void lockd_down_net(struct svc_serv *serv, struct net *net)
264 {
265 	struct lockd_net *ln = net_generic(net, lockd_net_id);
266 
267 	if (ln->nlmsvc_users) {
268 		if (--ln->nlmsvc_users == 0) {
269 			nlm_shutdown_hosts_net(net);
270 			cancel_delayed_work_sync(&ln->grace_period_end);
271 			locks_end_grace(&ln->lockd_manager);
272 			svc_shutdown_net(serv, net);
273 			dprintk("lockd_down_net: per-net data destroyed; net=%p\n", net);
274 		}
275 	} else {
276 		printk(KERN_ERR "lockd_down_net: no users! task=%p, net=%p\n",
277 				nlmsvc_task, net);
278 		BUG();
279 	}
280 }
281 
282 static int lockd_start_svc(struct svc_serv *serv)
283 {
284 	int error;
285 
286 	if (nlmsvc_rqst)
287 		return 0;
288 
289 	/*
290 	 * Create the kernel thread and wait for it to start.
291 	 */
292 	nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0], NUMA_NO_NODE);
293 	if (IS_ERR(nlmsvc_rqst)) {
294 		error = PTR_ERR(nlmsvc_rqst);
295 		printk(KERN_WARNING
296 			"lockd_up: svc_rqst allocation failed, error=%d\n",
297 			error);
298 		goto out_rqst;
299 	}
300 
301 	svc_sock_update_bufs(serv);
302 	serv->sv_maxconn = nlm_max_connections;
303 
304 	nlmsvc_task = kthread_create(lockd, nlmsvc_rqst, "%s", serv->sv_name);
305 	if (IS_ERR(nlmsvc_task)) {
306 		error = PTR_ERR(nlmsvc_task);
307 		printk(KERN_WARNING
308 			"lockd_up: kthread_run failed, error=%d\n", error);
309 		goto out_task;
310 	}
311 	nlmsvc_rqst->rq_task = nlmsvc_task;
312 	wake_up_process(nlmsvc_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_ops lockd_sv_ops = {
326 	.svo_shutdown		= svc_rpcb_cleanup,
327 	.svo_enqueue_xprt	= svc_xprt_do_enqueue,
328 };
329 
330 static struct svc_serv *lockd_create_svc(void)
331 {
332 	struct svc_serv *serv;
333 
334 	/*
335 	 * Check whether we're already up and running.
336 	 */
337 	if (nlmsvc_rqst) {
338 		/*
339 		 * Note: increase service usage, because later in case of error
340 		 * svc_destroy() will be called.
341 		 */
342 		svc_get(nlmsvc_rqst->rq_server);
343 		return nlmsvc_rqst->rq_server;
344 	}
345 
346 	/*
347 	 * Sanity check: if there's no pid,
348 	 * we should be the first user ...
349 	 */
350 	if (nlmsvc_users)
351 		printk(KERN_WARNING
352 			"lockd_up: no pid, %d users??\n", nlmsvc_users);
353 
354 	if (!nlm_timeout)
355 		nlm_timeout = LOCKD_DFLT_TIMEO;
356 	nlmsvc_timeout = nlm_timeout * HZ;
357 
358 	serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, &lockd_sv_ops);
359 	if (!serv) {
360 		printk(KERN_WARNING "lockd_up: create service failed\n");
361 		return ERR_PTR(-ENOMEM);
362 	}
363 	dprintk("lockd_up: service created\n");
364 	return serv;
365 }
366 
367 /*
368  * Bring up the lockd process if it's not already up.
369  */
370 int lockd_up(struct net *net)
371 {
372 	struct svc_serv *serv;
373 	int error;
374 
375 	mutex_lock(&nlmsvc_mutex);
376 
377 	serv = lockd_create_svc();
378 	if (IS_ERR(serv)) {
379 		error = PTR_ERR(serv);
380 		goto err_create;
381 	}
382 
383 	error = lockd_up_net(serv, net);
384 	if (error < 0)
385 		goto err_net;
386 
387 	error = lockd_start_svc(serv);
388 	if (error < 0)
389 		goto err_start;
390 
391 	nlmsvc_users++;
392 	/*
393 	 * Note: svc_serv structures have an initial use count of 1,
394 	 * so we exit through here on both success and failure.
395 	 */
396 err_net:
397 	svc_destroy(serv);
398 err_create:
399 	mutex_unlock(&nlmsvc_mutex);
400 	return error;
401 
402 err_start:
403 	lockd_down_net(serv, net);
404 	goto err_net;
405 }
406 EXPORT_SYMBOL_GPL(lockd_up);
407 
408 /*
409  * Decrement the user count and bring down lockd if we're the last.
410  */
411 void
412 lockd_down(struct net *net)
413 {
414 	mutex_lock(&nlmsvc_mutex);
415 	lockd_down_net(nlmsvc_rqst->rq_server, net);
416 	if (nlmsvc_users) {
417 		if (--nlmsvc_users)
418 			goto out;
419 	} else {
420 		printk(KERN_ERR "lockd_down: no users! task=%p\n",
421 			nlmsvc_task);
422 		BUG();
423 	}
424 
425 	if (!nlmsvc_task) {
426 		printk(KERN_ERR "lockd_down: no lockd running.\n");
427 		BUG();
428 	}
429 	kthread_stop(nlmsvc_task);
430 	dprintk("lockd_down: service stopped\n");
431 	svc_exit_thread(nlmsvc_rqst);
432 	dprintk("lockd_down: service destroyed\n");
433 	nlmsvc_task = NULL;
434 	nlmsvc_rqst = NULL;
435 out:
436 	mutex_unlock(&nlmsvc_mutex);
437 }
438 EXPORT_SYMBOL_GPL(lockd_down);
439 
440 #ifdef CONFIG_SYSCTL
441 
442 /*
443  * Sysctl parameters (same as module parameters, different interface).
444  */
445 
446 static struct ctl_table nlm_sysctls[] = {
447 	{
448 		.procname	= "nlm_grace_period",
449 		.data		= &nlm_grace_period,
450 		.maxlen		= sizeof(unsigned long),
451 		.mode		= 0644,
452 		.proc_handler	= proc_doulongvec_minmax,
453 		.extra1		= (unsigned long *) &nlm_grace_period_min,
454 		.extra2		= (unsigned long *) &nlm_grace_period_max,
455 	},
456 	{
457 		.procname	= "nlm_timeout",
458 		.data		= &nlm_timeout,
459 		.maxlen		= sizeof(unsigned long),
460 		.mode		= 0644,
461 		.proc_handler	= proc_doulongvec_minmax,
462 		.extra1		= (unsigned long *) &nlm_timeout_min,
463 		.extra2		= (unsigned long *) &nlm_timeout_max,
464 	},
465 	{
466 		.procname	= "nlm_udpport",
467 		.data		= &nlm_udpport,
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	= "nlm_tcpport",
476 		.data		= &nlm_tcpport,
477 		.maxlen		= sizeof(int),
478 		.mode		= 0644,
479 		.proc_handler	= proc_dointvec_minmax,
480 		.extra1		= (int *) &nlm_port_min,
481 		.extra2		= (int *) &nlm_port_max,
482 	},
483 	{
484 		.procname	= "nsm_use_hostnames",
485 		.data		= &nsm_use_hostnames,
486 		.maxlen		= sizeof(int),
487 		.mode		= 0644,
488 		.proc_handler	= proc_dointvec,
489 	},
490 	{
491 		.procname	= "nsm_local_state",
492 		.data		= &nsm_local_state,
493 		.maxlen		= sizeof(int),
494 		.mode		= 0644,
495 		.proc_handler	= proc_dointvec,
496 	},
497 	{ }
498 };
499 
500 static struct ctl_table nlm_sysctl_dir[] = {
501 	{
502 		.procname	= "nfs",
503 		.mode		= 0555,
504 		.child		= nlm_sysctls,
505 	},
506 	{ }
507 };
508 
509 static struct ctl_table nlm_sysctl_root[] = {
510 	{
511 		.procname	= "fs",
512 		.mode		= 0555,
513 		.child		= nlm_sysctl_dir,
514 	},
515 	{ }
516 };
517 
518 #endif	/* CONFIG_SYSCTL */
519 
520 /*
521  * Module (and sysfs) parameters.
522  */
523 
524 #define param_set_min_max(name, type, which_strtol, min, max)		\
525 static int param_set_##name(const char *val, struct kernel_param *kp)	\
526 {									\
527 	char *endp;							\
528 	__typeof__(type) num = which_strtol(val, &endp, 0);		\
529 	if (endp == val || *endp || num < (min) || num > (max))		\
530 		return -EINVAL;						\
531 	*((type *) kp->arg) = num;					\
532 	return 0;							\
533 }
534 
535 static inline int is_callback(u32 proc)
536 {
537 	return proc == NLMPROC_GRANTED
538 		|| proc == NLMPROC_GRANTED_MSG
539 		|| proc == NLMPROC_TEST_RES
540 		|| proc == NLMPROC_LOCK_RES
541 		|| proc == NLMPROC_CANCEL_RES
542 		|| proc == NLMPROC_UNLOCK_RES
543 		|| proc == NLMPROC_NSM_NOTIFY;
544 }
545 
546 
547 static int lockd_authenticate(struct svc_rqst *rqstp)
548 {
549 	rqstp->rq_client = NULL;
550 	switch (rqstp->rq_authop->flavour) {
551 		case RPC_AUTH_NULL:
552 		case RPC_AUTH_UNIX:
553 			if (rqstp->rq_proc == 0)
554 				return SVC_OK;
555 			if (is_callback(rqstp->rq_proc)) {
556 				/* Leave it to individual procedures to
557 				 * call nlmsvc_lookup_host(rqstp)
558 				 */
559 				return SVC_OK;
560 			}
561 			return svc_set_client(rqstp);
562 	}
563 	return SVC_DENIED;
564 }
565 
566 
567 param_set_min_max(port, int, simple_strtol, 0, 65535)
568 param_set_min_max(grace_period, unsigned long, simple_strtoul,
569 		  nlm_grace_period_min, nlm_grace_period_max)
570 param_set_min_max(timeout, unsigned long, simple_strtoul,
571 		  nlm_timeout_min, nlm_timeout_max)
572 
573 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
574 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
575 MODULE_LICENSE("GPL");
576 
577 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
578 		  &nlm_grace_period, 0644);
579 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
580 		  &nlm_timeout, 0644);
581 module_param_call(nlm_udpport, param_set_port, param_get_int,
582 		  &nlm_udpport, 0644);
583 module_param_call(nlm_tcpport, param_set_port, param_get_int,
584 		  &nlm_tcpport, 0644);
585 module_param(nsm_use_hostnames, bool, 0644);
586 module_param(nlm_max_connections, uint, 0644);
587 
588 static int lockd_init_net(struct net *net)
589 {
590 	struct lockd_net *ln = net_generic(net, lockd_net_id);
591 
592 	INIT_DELAYED_WORK(&ln->grace_period_end, grace_ender);
593 	INIT_LIST_HEAD(&ln->lockd_manager.list);
594 	ln->lockd_manager.block_opens = false;
595 	spin_lock_init(&ln->nsm_clnt_lock);
596 	return 0;
597 }
598 
599 static void lockd_exit_net(struct net *net)
600 {
601 }
602 
603 static struct pernet_operations lockd_net_ops = {
604 	.init = lockd_init_net,
605 	.exit = lockd_exit_net,
606 	.id = &lockd_net_id,
607 	.size = sizeof(struct lockd_net),
608 };
609 
610 
611 /*
612  * Initialising and terminating the module.
613  */
614 
615 static int __init init_nlm(void)
616 {
617 	int err;
618 
619 #ifdef CONFIG_SYSCTL
620 	err = -ENOMEM;
621 	nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
622 	if (nlm_sysctl_table == NULL)
623 		goto err_sysctl;
624 #endif
625 	err = register_pernet_subsys(&lockd_net_ops);
626 	if (err)
627 		goto err_pernet;
628 
629 	err = lockd_create_procfs();
630 	if (err)
631 		goto err_procfs;
632 
633 	return 0;
634 
635 err_procfs:
636 	unregister_pernet_subsys(&lockd_net_ops);
637 err_pernet:
638 #ifdef CONFIG_SYSCTL
639 	unregister_sysctl_table(nlm_sysctl_table);
640 err_sysctl:
641 #endif
642 	return err;
643 }
644 
645 static void __exit exit_nlm(void)
646 {
647 	/* FIXME: delete all NLM clients */
648 	nlm_shutdown_hosts();
649 	lockd_remove_procfs();
650 	unregister_pernet_subsys(&lockd_net_ops);
651 #ifdef CONFIG_SYSCTL
652 	unregister_sysctl_table(nlm_sysctl_table);
653 #endif
654 }
655 
656 module_init(init_nlm);
657 module_exit(exit_nlm);
658 
659 /*
660  * Define NLM program and procedures
661  */
662 static struct svc_version	nlmsvc_version1 = {
663 		.vs_vers	= 1,
664 		.vs_nproc	= 17,
665 		.vs_proc	= nlmsvc_procedures,
666 		.vs_xdrsize	= NLMSVC_XDRSIZE,
667 };
668 static struct svc_version	nlmsvc_version3 = {
669 		.vs_vers	= 3,
670 		.vs_nproc	= 24,
671 		.vs_proc	= nlmsvc_procedures,
672 		.vs_xdrsize	= NLMSVC_XDRSIZE,
673 };
674 #ifdef CONFIG_LOCKD_V4
675 static struct svc_version	nlmsvc_version4 = {
676 		.vs_vers	= 4,
677 		.vs_nproc	= 24,
678 		.vs_proc	= nlmsvc_procedures4,
679 		.vs_xdrsize	= NLMSVC_XDRSIZE,
680 };
681 #endif
682 static struct svc_version *	nlmsvc_version[] = {
683 	[1] = &nlmsvc_version1,
684 	[3] = &nlmsvc_version3,
685 #ifdef CONFIG_LOCKD_V4
686 	[4] = &nlmsvc_version4,
687 #endif
688 };
689 
690 static struct svc_stat		nlmsvc_stats;
691 
692 #define NLM_NRVERS	ARRAY_SIZE(nlmsvc_version)
693 static struct svc_program	nlmsvc_program = {
694 	.pg_prog		= NLM_PROGRAM,		/* program number */
695 	.pg_nvers		= NLM_NRVERS,		/* number of entries in nlmsvc_version */
696 	.pg_vers		= nlmsvc_version,	/* version table */
697 	.pg_name		= "lockd",		/* service name */
698 	.pg_class		= "nfsd",		/* share authentication with nfsd */
699 	.pg_stats		= &nlmsvc_stats,	/* stats table */
700 	.pg_authenticate = &lockd_authenticate	/* export authentication */
701 };
702