xref: /linux/fs/lockd/svc.c (revision b8bb76713ec50df2f11efee386e16f93d51e1076)
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/slab.h>
25 #include <linux/smp.h>
26 #include <linux/smp_lock.h>
27 #include <linux/mutex.h>
28 #include <linux/kthread.h>
29 #include <linux/freezer.h>
30 
31 #include <linux/sunrpc/types.h>
32 #include <linux/sunrpc/stats.h>
33 #include <linux/sunrpc/clnt.h>
34 #include <linux/sunrpc/svc.h>
35 #include <linux/sunrpc/svcsock.h>
36 #include <net/ip.h>
37 #include <linux/lockd/lockd.h>
38 #include <linux/nfs.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 /*
56  * These can be set at insmod time (useful for NFS as root filesystem),
57  * and also changed through the sysctl interface.  -- Jamie Lokier, Aug 2003
58  */
59 static unsigned long		nlm_grace_period;
60 static unsigned long		nlm_timeout = LOCKD_DFLT_TIMEO;
61 static int			nlm_udpport, nlm_tcpport;
62 
63 /* RLIM_NOFILE defaults to 1024. That seems like a reasonable default here. */
64 static unsigned int		nlm_max_connections = 1024;
65 
66 /*
67  * Constants needed for the sysctl interface.
68  */
69 static const unsigned long	nlm_grace_period_min = 0;
70 static const unsigned long	nlm_grace_period_max = 240;
71 static const unsigned long	nlm_timeout_min = 3;
72 static const unsigned long	nlm_timeout_max = 20;
73 static const int		nlm_port_min = 0, nlm_port_max = 65535;
74 
75 #ifdef CONFIG_SYSCTL
76 static struct ctl_table_header * nlm_sysctl_table;
77 #endif
78 
79 static unsigned long get_lockd_grace_period(void)
80 {
81 	/* Note: nlm_timeout should always be nonzero */
82 	if (nlm_grace_period)
83 		return roundup(nlm_grace_period, nlm_timeout) * HZ;
84 	else
85 		return nlm_timeout * 5 * HZ;
86 }
87 
88 static struct lock_manager lockd_manager = {
89 };
90 
91 static void grace_ender(struct work_struct *not_used)
92 {
93 	locks_end_grace(&lockd_manager);
94 }
95 
96 static DECLARE_DELAYED_WORK(grace_period_end, grace_ender);
97 
98 static void set_grace_period(void)
99 {
100 	unsigned long grace_period = get_lockd_grace_period();
101 
102 	locks_start_grace(&lockd_manager);
103 	cancel_delayed_work_sync(&grace_period_end);
104 	schedule_delayed_work(&grace_period_end, grace_period);
105 }
106 
107 /*
108  * This is the lockd kernel thread
109  */
110 static int
111 lockd(void *vrqstp)
112 {
113 	int		err = 0, preverr = 0;
114 	struct svc_rqst *rqstp = vrqstp;
115 
116 	/* try_to_freeze() is called from svc_recv() */
117 	set_freezable();
118 
119 	/* Allow SIGKILL to tell lockd to drop all of its locks */
120 	allow_signal(SIGKILL);
121 
122 	dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
123 
124 	/*
125 	 * FIXME: it would be nice if lockd didn't spend its entire life
126 	 * running under the BKL. At the very least, it would be good to
127 	 * have someone clarify what it's intended to protect here. I've
128 	 * seen some handwavy posts about posix locking needing to be
129 	 * done under the BKL, but it's far from clear.
130 	 */
131 	lock_kernel();
132 
133 	if (!nlm_timeout)
134 		nlm_timeout = LOCKD_DFLT_TIMEO;
135 	nlmsvc_timeout = nlm_timeout * HZ;
136 
137 	set_grace_period();
138 
139 	/*
140 	 * The main request loop. We don't terminate until the last
141 	 * NFS mount or NFS daemon has gone away.
142 	 */
143 	while (!kthread_should_stop()) {
144 		long timeout = MAX_SCHEDULE_TIMEOUT;
145 		RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
146 
147 		/* update sv_maxconn if it has changed */
148 		rqstp->rq_server->sv_maxconn = nlm_max_connections;
149 
150 		if (signalled()) {
151 			flush_signals(current);
152 			if (nlmsvc_ops) {
153 				nlmsvc_invalidate_all();
154 				set_grace_period();
155 			}
156 			continue;
157 		}
158 
159 		timeout = nlmsvc_retry_blocked();
160 
161 		/*
162 		 * Find a socket with data available and call its
163 		 * recvfrom routine.
164 		 */
165 		err = svc_recv(rqstp, timeout);
166 		if (err == -EAGAIN || err == -EINTR) {
167 			preverr = err;
168 			continue;
169 		}
170 		if (err < 0) {
171 			if (err != preverr) {
172 				printk(KERN_WARNING "%s: unexpected error "
173 					"from svc_recv (%d)\n", __func__, err);
174 				preverr = err;
175 			}
176 			schedule_timeout_interruptible(HZ);
177 			continue;
178 		}
179 		preverr = err;
180 
181 		dprintk("lockd: request from %s\n",
182 				svc_print_addr(rqstp, buf, sizeof(buf)));
183 
184 		svc_process(rqstp);
185 	}
186 	flush_signals(current);
187 	cancel_delayed_work_sync(&grace_period_end);
188 	locks_end_grace(&lockd_manager);
189 	if (nlmsvc_ops)
190 		nlmsvc_invalidate_all();
191 	nlm_shutdown_hosts();
192 	unlock_kernel();
193 	return 0;
194 }
195 
196 static int create_lockd_listener(struct svc_serv *serv, const char *name,
197 				 const int family, const unsigned short port)
198 {
199 	struct svc_xprt *xprt;
200 
201 	xprt = svc_find_xprt(serv, name, family, 0);
202 	if (xprt == NULL)
203 		return svc_create_xprt(serv, name, family, port,
204 						SVC_SOCK_DEFAULTS);
205 	svc_xprt_put(xprt);
206 	return 0;
207 }
208 
209 static int create_lockd_family(struct svc_serv *serv, const int family)
210 {
211 	int err;
212 
213 	err = create_lockd_listener(serv, "udp", family, nlm_udpport);
214 	if (err < 0)
215 		return err;
216 
217 	return create_lockd_listener(serv, "tcp", family, nlm_tcpport);
218 }
219 
220 /*
221  * Ensure there are active UDP and TCP listeners for lockd.
222  *
223  * Even if we have only TCP NFS mounts and/or TCP NFSDs, some
224  * local services (such as rpc.statd) still require UDP, and
225  * some NFS servers do not yet support NLM over TCP.
226  *
227  * Returns zero if all listeners are available; otherwise a
228  * negative errno value is returned.
229  */
230 static int make_socks(struct svc_serv *serv)
231 {
232 	static int warned;
233 	int err;
234 
235 	err = create_lockd_family(serv, PF_INET);
236 	if (err < 0)
237 		goto out_err;
238 
239 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
240 	err = create_lockd_family(serv, PF_INET6);
241 	if (err < 0 && err != -EAFNOSUPPORT)
242 		goto out_err;
243 #endif	/* CONFIG_IPV6 || CONFIG_IPV6_MODULE */
244 
245 	warned = 0;
246 	return 0;
247 
248 out_err:
249 	if (warned++ == 0)
250 		printk(KERN_WARNING
251 			"lockd_up: makesock failed, error=%d\n", err);
252 	return err;
253 }
254 
255 /*
256  * Bring up the lockd process if it's not already up.
257  */
258 int lockd_up(void)
259 {
260 	struct svc_serv *serv;
261 	int		error = 0;
262 
263 	mutex_lock(&nlmsvc_mutex);
264 	/*
265 	 * Check whether we're already up and running.
266 	 */
267 	if (nlmsvc_rqst)
268 		goto out;
269 
270 	/*
271 	 * Sanity check: if there's no pid,
272 	 * we should be the first user ...
273 	 */
274 	if (nlmsvc_users)
275 		printk(KERN_WARNING
276 			"lockd_up: no pid, %d users??\n", nlmsvc_users);
277 
278 	error = -ENOMEM;
279 	serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL);
280 	if (!serv) {
281 		printk(KERN_WARNING "lockd_up: create service failed\n");
282 		goto out;
283 	}
284 
285 	error = make_socks(serv);
286 	if (error < 0)
287 		goto destroy_and_out;
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]);
293 	if (IS_ERR(nlmsvc_rqst)) {
294 		error = PTR_ERR(nlmsvc_rqst);
295 		nlmsvc_rqst = NULL;
296 		printk(KERN_WARNING
297 			"lockd_up: svc_rqst allocation failed, error=%d\n",
298 			error);
299 		goto destroy_and_out;
300 	}
301 
302 	svc_sock_update_bufs(serv);
303 	serv->sv_maxconn = nlm_max_connections;
304 
305 	nlmsvc_task = kthread_run(lockd, nlmsvc_rqst, serv->sv_name);
306 	if (IS_ERR(nlmsvc_task)) {
307 		error = PTR_ERR(nlmsvc_task);
308 		svc_exit_thread(nlmsvc_rqst);
309 		nlmsvc_task = NULL;
310 		nlmsvc_rqst = NULL;
311 		printk(KERN_WARNING
312 			"lockd_up: kthread_run failed, error=%d\n", error);
313 		goto destroy_and_out;
314 	}
315 
316 	/*
317 	 * Note: svc_serv structures have an initial use count of 1,
318 	 * so we exit through here on both success and failure.
319 	 */
320 destroy_and_out:
321 	svc_destroy(serv);
322 out:
323 	if (!error)
324 		nlmsvc_users++;
325 	mutex_unlock(&nlmsvc_mutex);
326 	return error;
327 }
328 EXPORT_SYMBOL_GPL(lockd_up);
329 
330 /*
331  * Decrement the user count and bring down lockd if we're the last.
332  */
333 void
334 lockd_down(void)
335 {
336 	mutex_lock(&nlmsvc_mutex);
337 	if (nlmsvc_users) {
338 		if (--nlmsvc_users)
339 			goto out;
340 	} else {
341 		printk(KERN_ERR "lockd_down: no users! task=%p\n",
342 			nlmsvc_task);
343 		BUG();
344 	}
345 
346 	if (!nlmsvc_task) {
347 		printk(KERN_ERR "lockd_down: no lockd running.\n");
348 		BUG();
349 	}
350 	kthread_stop(nlmsvc_task);
351 	svc_exit_thread(nlmsvc_rqst);
352 	nlmsvc_task = NULL;
353 	nlmsvc_rqst = NULL;
354 out:
355 	mutex_unlock(&nlmsvc_mutex);
356 }
357 EXPORT_SYMBOL_GPL(lockd_down);
358 
359 #ifdef CONFIG_SYSCTL
360 
361 /*
362  * Sysctl parameters (same as module parameters, different interface).
363  */
364 
365 static ctl_table nlm_sysctls[] = {
366 	{
367 		.ctl_name	= CTL_UNNUMBERED,
368 		.procname	= "nlm_grace_period",
369 		.data		= &nlm_grace_period,
370 		.maxlen		= sizeof(unsigned long),
371 		.mode		= 0644,
372 		.proc_handler	= &proc_doulongvec_minmax,
373 		.extra1		= (unsigned long *) &nlm_grace_period_min,
374 		.extra2		= (unsigned long *) &nlm_grace_period_max,
375 	},
376 	{
377 		.ctl_name	= CTL_UNNUMBERED,
378 		.procname	= "nlm_timeout",
379 		.data		= &nlm_timeout,
380 		.maxlen		= sizeof(unsigned long),
381 		.mode		= 0644,
382 		.proc_handler	= &proc_doulongvec_minmax,
383 		.extra1		= (unsigned long *) &nlm_timeout_min,
384 		.extra2		= (unsigned long *) &nlm_timeout_max,
385 	},
386 	{
387 		.ctl_name	= CTL_UNNUMBERED,
388 		.procname	= "nlm_udpport",
389 		.data		= &nlm_udpport,
390 		.maxlen		= sizeof(int),
391 		.mode		= 0644,
392 		.proc_handler	= &proc_dointvec_minmax,
393 		.extra1		= (int *) &nlm_port_min,
394 		.extra2		= (int *) &nlm_port_max,
395 	},
396 	{
397 		.ctl_name	= CTL_UNNUMBERED,
398 		.procname	= "nlm_tcpport",
399 		.data		= &nlm_tcpport,
400 		.maxlen		= sizeof(int),
401 		.mode		= 0644,
402 		.proc_handler	= &proc_dointvec_minmax,
403 		.extra1		= (int *) &nlm_port_min,
404 		.extra2		= (int *) &nlm_port_max,
405 	},
406 	{
407 		.ctl_name	= CTL_UNNUMBERED,
408 		.procname	= "nsm_use_hostnames",
409 		.data		= &nsm_use_hostnames,
410 		.maxlen		= sizeof(int),
411 		.mode		= 0644,
412 		.proc_handler	= &proc_dointvec,
413 	},
414 	{
415 		.ctl_name	= CTL_UNNUMBERED,
416 		.procname	= "nsm_local_state",
417 		.data		= &nsm_local_state,
418 		.maxlen		= sizeof(int),
419 		.mode		= 0644,
420 		.proc_handler	= &proc_dointvec,
421 	},
422 	{ .ctl_name = 0 }
423 };
424 
425 static ctl_table nlm_sysctl_dir[] = {
426 	{
427 		.ctl_name	= CTL_UNNUMBERED,
428 		.procname	= "nfs",
429 		.mode		= 0555,
430 		.child		= nlm_sysctls,
431 	},
432 	{ .ctl_name = 0 }
433 };
434 
435 static ctl_table nlm_sysctl_root[] = {
436 	{
437 		.ctl_name	= CTL_FS,
438 		.procname	= "fs",
439 		.mode		= 0555,
440 		.child		= nlm_sysctl_dir,
441 	},
442 	{ .ctl_name = 0 }
443 };
444 
445 #endif	/* CONFIG_SYSCTL */
446 
447 /*
448  * Module (and sysfs) parameters.
449  */
450 
451 #define param_set_min_max(name, type, which_strtol, min, max)		\
452 static int param_set_##name(const char *val, struct kernel_param *kp)	\
453 {									\
454 	char *endp;							\
455 	__typeof__(type) num = which_strtol(val, &endp, 0);		\
456 	if (endp == val || *endp || num < (min) || num > (max))		\
457 		return -EINVAL;						\
458 	*((int *) kp->arg) = num;					\
459 	return 0;							\
460 }
461 
462 static inline int is_callback(u32 proc)
463 {
464 	return proc == NLMPROC_GRANTED
465 		|| proc == NLMPROC_GRANTED_MSG
466 		|| proc == NLMPROC_TEST_RES
467 		|| proc == NLMPROC_LOCK_RES
468 		|| proc == NLMPROC_CANCEL_RES
469 		|| proc == NLMPROC_UNLOCK_RES
470 		|| proc == NLMPROC_NSM_NOTIFY;
471 }
472 
473 
474 static int lockd_authenticate(struct svc_rqst *rqstp)
475 {
476 	rqstp->rq_client = NULL;
477 	switch (rqstp->rq_authop->flavour) {
478 		case RPC_AUTH_NULL:
479 		case RPC_AUTH_UNIX:
480 			if (rqstp->rq_proc == 0)
481 				return SVC_OK;
482 			if (is_callback(rqstp->rq_proc)) {
483 				/* Leave it to individual procedures to
484 				 * call nlmsvc_lookup_host(rqstp)
485 				 */
486 				return SVC_OK;
487 			}
488 			return svc_set_client(rqstp);
489 	}
490 	return SVC_DENIED;
491 }
492 
493 
494 param_set_min_max(port, int, simple_strtol, 0, 65535)
495 param_set_min_max(grace_period, unsigned long, simple_strtoul,
496 		  nlm_grace_period_min, nlm_grace_period_max)
497 param_set_min_max(timeout, unsigned long, simple_strtoul,
498 		  nlm_timeout_min, nlm_timeout_max)
499 
500 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
501 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
502 MODULE_LICENSE("GPL");
503 
504 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
505 		  &nlm_grace_period, 0644);
506 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
507 		  &nlm_timeout, 0644);
508 module_param_call(nlm_udpport, param_set_port, param_get_int,
509 		  &nlm_udpport, 0644);
510 module_param_call(nlm_tcpport, param_set_port, param_get_int,
511 		  &nlm_tcpport, 0644);
512 module_param(nsm_use_hostnames, bool, 0644);
513 module_param(nlm_max_connections, uint, 0644);
514 
515 /*
516  * Initialising and terminating the module.
517  */
518 
519 static int __init init_nlm(void)
520 {
521 #ifdef CONFIG_SYSCTL
522 	nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
523 	return nlm_sysctl_table ? 0 : -ENOMEM;
524 #else
525 	return 0;
526 #endif
527 }
528 
529 static void __exit exit_nlm(void)
530 {
531 	/* FIXME: delete all NLM clients */
532 	nlm_shutdown_hosts();
533 #ifdef CONFIG_SYSCTL
534 	unregister_sysctl_table(nlm_sysctl_table);
535 #endif
536 }
537 
538 module_init(init_nlm);
539 module_exit(exit_nlm);
540 
541 /*
542  * Define NLM program and procedures
543  */
544 static struct svc_version	nlmsvc_version1 = {
545 		.vs_vers	= 1,
546 		.vs_nproc	= 17,
547 		.vs_proc	= nlmsvc_procedures,
548 		.vs_xdrsize	= NLMSVC_XDRSIZE,
549 };
550 static struct svc_version	nlmsvc_version3 = {
551 		.vs_vers	= 3,
552 		.vs_nproc	= 24,
553 		.vs_proc	= nlmsvc_procedures,
554 		.vs_xdrsize	= NLMSVC_XDRSIZE,
555 };
556 #ifdef CONFIG_LOCKD_V4
557 static struct svc_version	nlmsvc_version4 = {
558 		.vs_vers	= 4,
559 		.vs_nproc	= 24,
560 		.vs_proc	= nlmsvc_procedures4,
561 		.vs_xdrsize	= NLMSVC_XDRSIZE,
562 };
563 #endif
564 static struct svc_version *	nlmsvc_version[] = {
565 	[1] = &nlmsvc_version1,
566 	[3] = &nlmsvc_version3,
567 #ifdef CONFIG_LOCKD_V4
568 	[4] = &nlmsvc_version4,
569 #endif
570 };
571 
572 static struct svc_stat		nlmsvc_stats;
573 
574 #define NLM_NRVERS	ARRAY_SIZE(nlmsvc_version)
575 static struct svc_program	nlmsvc_program = {
576 	.pg_prog		= NLM_PROGRAM,		/* program number */
577 	.pg_nvers		= NLM_NRVERS,		/* number of entries in nlmsvc_version */
578 	.pg_vers		= nlmsvc_version,	/* version table */
579 	.pg_name		= "lockd",		/* service name */
580 	.pg_class		= "nfsd",		/* share authentication with nfsd */
581 	.pg_stats		= &nlmsvc_stats,	/* stats table */
582 	.pg_authenticate = &lockd_authenticate	/* export authentication */
583 };
584