xref: /openbmc/linux/fs/lockd/svc.c (revision 0dba7c2a)
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  * If the kernel has IPv6 support available, always listen for
57  * both AF_INET and AF_INET6 requests.
58  */
59 #if (defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)) && \
60 	defined(CONFIG_SUNRPC_REGISTER_V4)
61 static const sa_family_t	nlmsvc_family = AF_INET6;
62 #else	/* (CONFIG_IPV6 || CONFIG_IPV6_MODULE) && CONFIG_SUNRPC_REGISTER_V4 */
63 static const sa_family_t	nlmsvc_family = AF_INET;
64 #endif	/* (CONFIG_IPV6 || CONFIG_IPV6_MODULE) && CONFIG_SUNRPC_REGISTER_V4 */
65 
66 /*
67  * These can be set at insmod time (useful for NFS as root filesystem),
68  * and also changed through the sysctl interface.  -- Jamie Lokier, Aug 2003
69  */
70 static unsigned long		nlm_grace_period;
71 static unsigned long		nlm_timeout = LOCKD_DFLT_TIMEO;
72 static int			nlm_udpport, nlm_tcpport;
73 
74 /* RLIM_NOFILE defaults to 1024. That seems like a reasonable default here. */
75 static unsigned int		nlm_max_connections = 1024;
76 
77 /*
78  * Constants needed for the sysctl interface.
79  */
80 static const unsigned long	nlm_grace_period_min = 0;
81 static const unsigned long	nlm_grace_period_max = 240;
82 static const unsigned long	nlm_timeout_min = 3;
83 static const unsigned long	nlm_timeout_max = 20;
84 static const int		nlm_port_min = 0, nlm_port_max = 65535;
85 
86 #ifdef CONFIG_SYSCTL
87 static struct ctl_table_header * nlm_sysctl_table;
88 #endif
89 
90 static unsigned long get_lockd_grace_period(void)
91 {
92 	/* Note: nlm_timeout should always be nonzero */
93 	if (nlm_grace_period)
94 		return roundup(nlm_grace_period, nlm_timeout) * HZ;
95 	else
96 		return nlm_timeout * 5 * HZ;
97 }
98 
99 static struct lock_manager lockd_manager = {
100 };
101 
102 static void grace_ender(struct work_struct *not_used)
103 {
104 	locks_end_grace(&lockd_manager);
105 }
106 
107 static DECLARE_DELAYED_WORK(grace_period_end, grace_ender);
108 
109 static void set_grace_period(void)
110 {
111 	unsigned long grace_period = get_lockd_grace_period();
112 
113 	locks_start_grace(&lockd_manager);
114 	cancel_delayed_work_sync(&grace_period_end);
115 	schedule_delayed_work(&grace_period_end, grace_period);
116 }
117 
118 /*
119  * This is the lockd kernel thread
120  */
121 static int
122 lockd(void *vrqstp)
123 {
124 	int		err = 0, preverr = 0;
125 	struct svc_rqst *rqstp = vrqstp;
126 
127 	/* try_to_freeze() is called from svc_recv() */
128 	set_freezable();
129 
130 	/* Allow SIGKILL to tell lockd to drop all of its locks */
131 	allow_signal(SIGKILL);
132 
133 	dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
134 
135 	/*
136 	 * FIXME: it would be nice if lockd didn't spend its entire life
137 	 * running under the BKL. At the very least, it would be good to
138 	 * have someone clarify what it's intended to protect here. I've
139 	 * seen some handwavy posts about posix locking needing to be
140 	 * done under the BKL, but it's far from clear.
141 	 */
142 	lock_kernel();
143 
144 	if (!nlm_timeout)
145 		nlm_timeout = LOCKD_DFLT_TIMEO;
146 	nlmsvc_timeout = nlm_timeout * HZ;
147 
148 	set_grace_period();
149 
150 	/*
151 	 * The main request loop. We don't terminate until the last
152 	 * NFS mount or NFS daemon has gone away.
153 	 */
154 	while (!kthread_should_stop()) {
155 		long timeout = MAX_SCHEDULE_TIMEOUT;
156 		RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
157 
158 		/* update sv_maxconn if it has changed */
159 		rqstp->rq_server->sv_maxconn = nlm_max_connections;
160 
161 		if (signalled()) {
162 			flush_signals(current);
163 			if (nlmsvc_ops) {
164 				nlmsvc_invalidate_all();
165 				set_grace_period();
166 			}
167 			continue;
168 		}
169 
170 		timeout = nlmsvc_retry_blocked();
171 
172 		/*
173 		 * Find a socket with data available and call its
174 		 * recvfrom routine.
175 		 */
176 		err = svc_recv(rqstp, timeout);
177 		if (err == -EAGAIN || err == -EINTR) {
178 			preverr = err;
179 			continue;
180 		}
181 		if (err < 0) {
182 			if (err != preverr) {
183 				printk(KERN_WARNING "%s: unexpected error "
184 					"from svc_recv (%d)\n", __func__, err);
185 				preverr = err;
186 			}
187 			schedule_timeout_interruptible(HZ);
188 			continue;
189 		}
190 		preverr = err;
191 
192 		dprintk("lockd: request from %s\n",
193 				svc_print_addr(rqstp, buf, sizeof(buf)));
194 
195 		svc_process(rqstp);
196 	}
197 	flush_signals(current);
198 	cancel_delayed_work_sync(&grace_period_end);
199 	locks_end_grace(&lockd_manager);
200 	if (nlmsvc_ops)
201 		nlmsvc_invalidate_all();
202 	nlm_shutdown_hosts();
203 	unlock_kernel();
204 	return 0;
205 }
206 
207 static int create_lockd_listener(struct svc_serv *serv, char *name,
208 				 unsigned short port)
209 {
210 	struct svc_xprt *xprt;
211 
212 	xprt = svc_find_xprt(serv, name, 0, 0);
213 	if (xprt == NULL)
214 		return svc_create_xprt(serv, name, port, SVC_SOCK_DEFAULTS);
215 
216 	svc_xprt_put(xprt);
217 	return 0;
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_listener(serv, "udp", nlm_udpport);
236 	if (err < 0)
237 		goto out_err;
238 
239 	err = create_lockd_listener(serv, "tcp", nlm_tcpport);
240 	if (err < 0)
241 		goto out_err;
242 
243 	warned = 0;
244 	return 0;
245 
246 out_err:
247 	if (warned++ == 0)
248 		printk(KERN_WARNING
249 			"lockd_up: makesock failed, error=%d\n", err);
250 	return err;
251 }
252 
253 /*
254  * Bring up the lockd process if it's not already up.
255  */
256 int lockd_up(void)
257 {
258 	struct svc_serv *serv;
259 	int		error = 0;
260 
261 	mutex_lock(&nlmsvc_mutex);
262 	/*
263 	 * Check whether we're already up and running.
264 	 */
265 	if (nlmsvc_rqst)
266 		goto out;
267 
268 	/*
269 	 * Sanity check: if there's no pid,
270 	 * we should be the first user ...
271 	 */
272 	if (nlmsvc_users)
273 		printk(KERN_WARNING
274 			"lockd_up: no pid, %d users??\n", nlmsvc_users);
275 
276 	error = -ENOMEM;
277 	serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, nlmsvc_family, NULL);
278 	if (!serv) {
279 		printk(KERN_WARNING "lockd_up: create service failed\n");
280 		goto out;
281 	}
282 
283 	error = make_socks(serv);
284 	if (error < 0)
285 		goto destroy_and_out;
286 
287 	/*
288 	 * Create the kernel thread and wait for it to start.
289 	 */
290 	nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0]);
291 	if (IS_ERR(nlmsvc_rqst)) {
292 		error = PTR_ERR(nlmsvc_rqst);
293 		nlmsvc_rqst = NULL;
294 		printk(KERN_WARNING
295 			"lockd_up: svc_rqst allocation failed, error=%d\n",
296 			error);
297 		goto destroy_and_out;
298 	}
299 
300 	svc_sock_update_bufs(serv);
301 	serv->sv_maxconn = nlm_max_connections;
302 
303 	nlmsvc_task = kthread_run(lockd, nlmsvc_rqst, serv->sv_name);
304 	if (IS_ERR(nlmsvc_task)) {
305 		error = PTR_ERR(nlmsvc_task);
306 		svc_exit_thread(nlmsvc_rqst);
307 		nlmsvc_task = NULL;
308 		nlmsvc_rqst = NULL;
309 		printk(KERN_WARNING
310 			"lockd_up: kthread_run failed, error=%d\n", error);
311 		goto destroy_and_out;
312 	}
313 
314 	/*
315 	 * Note: svc_serv structures have an initial use count of 1,
316 	 * so we exit through here on both success and failure.
317 	 */
318 destroy_and_out:
319 	svc_destroy(serv);
320 out:
321 	if (!error)
322 		nlmsvc_users++;
323 	mutex_unlock(&nlmsvc_mutex);
324 	return error;
325 }
326 EXPORT_SYMBOL_GPL(lockd_up);
327 
328 /*
329  * Decrement the user count and bring down lockd if we're the last.
330  */
331 void
332 lockd_down(void)
333 {
334 	mutex_lock(&nlmsvc_mutex);
335 	if (nlmsvc_users) {
336 		if (--nlmsvc_users)
337 			goto out;
338 	} else {
339 		printk(KERN_ERR "lockd_down: no users! task=%p\n",
340 			nlmsvc_task);
341 		BUG();
342 	}
343 
344 	if (!nlmsvc_task) {
345 		printk(KERN_ERR "lockd_down: no lockd running.\n");
346 		BUG();
347 	}
348 	kthread_stop(nlmsvc_task);
349 	svc_exit_thread(nlmsvc_rqst);
350 	nlmsvc_task = NULL;
351 	nlmsvc_rqst = NULL;
352 out:
353 	mutex_unlock(&nlmsvc_mutex);
354 }
355 EXPORT_SYMBOL_GPL(lockd_down);
356 
357 #ifdef CONFIG_SYSCTL
358 
359 /*
360  * Sysctl parameters (same as module parameters, different interface).
361  */
362 
363 static ctl_table nlm_sysctls[] = {
364 	{
365 		.ctl_name	= CTL_UNNUMBERED,
366 		.procname	= "nlm_grace_period",
367 		.data		= &nlm_grace_period,
368 		.maxlen		= sizeof(unsigned long),
369 		.mode		= 0644,
370 		.proc_handler	= &proc_doulongvec_minmax,
371 		.extra1		= (unsigned long *) &nlm_grace_period_min,
372 		.extra2		= (unsigned long *) &nlm_grace_period_max,
373 	},
374 	{
375 		.ctl_name	= CTL_UNNUMBERED,
376 		.procname	= "nlm_timeout",
377 		.data		= &nlm_timeout,
378 		.maxlen		= sizeof(unsigned long),
379 		.mode		= 0644,
380 		.proc_handler	= &proc_doulongvec_minmax,
381 		.extra1		= (unsigned long *) &nlm_timeout_min,
382 		.extra2		= (unsigned long *) &nlm_timeout_max,
383 	},
384 	{
385 		.ctl_name	= CTL_UNNUMBERED,
386 		.procname	= "nlm_udpport",
387 		.data		= &nlm_udpport,
388 		.maxlen		= sizeof(int),
389 		.mode		= 0644,
390 		.proc_handler	= &proc_dointvec_minmax,
391 		.extra1		= (int *) &nlm_port_min,
392 		.extra2		= (int *) &nlm_port_max,
393 	},
394 	{
395 		.ctl_name	= CTL_UNNUMBERED,
396 		.procname	= "nlm_tcpport",
397 		.data		= &nlm_tcpport,
398 		.maxlen		= sizeof(int),
399 		.mode		= 0644,
400 		.proc_handler	= &proc_dointvec_minmax,
401 		.extra1		= (int *) &nlm_port_min,
402 		.extra2		= (int *) &nlm_port_max,
403 	},
404 	{
405 		.ctl_name	= CTL_UNNUMBERED,
406 		.procname	= "nsm_use_hostnames",
407 		.data		= &nsm_use_hostnames,
408 		.maxlen		= sizeof(int),
409 		.mode		= 0644,
410 		.proc_handler	= &proc_dointvec,
411 	},
412 	{
413 		.ctl_name	= CTL_UNNUMBERED,
414 		.procname	= "nsm_local_state",
415 		.data		= &nsm_local_state,
416 		.maxlen		= sizeof(int),
417 		.mode		= 0644,
418 		.proc_handler	= &proc_dointvec,
419 	},
420 	{ .ctl_name = 0 }
421 };
422 
423 static ctl_table nlm_sysctl_dir[] = {
424 	{
425 		.ctl_name	= CTL_UNNUMBERED,
426 		.procname	= "nfs",
427 		.mode		= 0555,
428 		.child		= nlm_sysctls,
429 	},
430 	{ .ctl_name = 0 }
431 };
432 
433 static ctl_table nlm_sysctl_root[] = {
434 	{
435 		.ctl_name	= CTL_FS,
436 		.procname	= "fs",
437 		.mode		= 0555,
438 		.child		= nlm_sysctl_dir,
439 	},
440 	{ .ctl_name = 0 }
441 };
442 
443 #endif	/* CONFIG_SYSCTL */
444 
445 /*
446  * Module (and sysfs) parameters.
447  */
448 
449 #define param_set_min_max(name, type, which_strtol, min, max)		\
450 static int param_set_##name(const char *val, struct kernel_param *kp)	\
451 {									\
452 	char *endp;							\
453 	__typeof__(type) num = which_strtol(val, &endp, 0);		\
454 	if (endp == val || *endp || num < (min) || num > (max))		\
455 		return -EINVAL;						\
456 	*((int *) kp->arg) = num;					\
457 	return 0;							\
458 }
459 
460 static inline int is_callback(u32 proc)
461 {
462 	return proc == NLMPROC_GRANTED
463 		|| proc == NLMPROC_GRANTED_MSG
464 		|| proc == NLMPROC_TEST_RES
465 		|| proc == NLMPROC_LOCK_RES
466 		|| proc == NLMPROC_CANCEL_RES
467 		|| proc == NLMPROC_UNLOCK_RES
468 		|| proc == NLMPROC_NSM_NOTIFY;
469 }
470 
471 
472 static int lockd_authenticate(struct svc_rqst *rqstp)
473 {
474 	rqstp->rq_client = NULL;
475 	switch (rqstp->rq_authop->flavour) {
476 		case RPC_AUTH_NULL:
477 		case RPC_AUTH_UNIX:
478 			if (rqstp->rq_proc == 0)
479 				return SVC_OK;
480 			if (is_callback(rqstp->rq_proc)) {
481 				/* Leave it to individual procedures to
482 				 * call nlmsvc_lookup_host(rqstp)
483 				 */
484 				return SVC_OK;
485 			}
486 			return svc_set_client(rqstp);
487 	}
488 	return SVC_DENIED;
489 }
490 
491 
492 param_set_min_max(port, int, simple_strtol, 0, 65535)
493 param_set_min_max(grace_period, unsigned long, simple_strtoul,
494 		  nlm_grace_period_min, nlm_grace_period_max)
495 param_set_min_max(timeout, unsigned long, simple_strtoul,
496 		  nlm_timeout_min, nlm_timeout_max)
497 
498 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
499 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
500 MODULE_LICENSE("GPL");
501 
502 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
503 		  &nlm_grace_period, 0644);
504 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
505 		  &nlm_timeout, 0644);
506 module_param_call(nlm_udpport, param_set_port, param_get_int,
507 		  &nlm_udpport, 0644);
508 module_param_call(nlm_tcpport, param_set_port, param_get_int,
509 		  &nlm_tcpport, 0644);
510 module_param(nsm_use_hostnames, bool, 0644);
511 module_param(nlm_max_connections, uint, 0644);
512 
513 /*
514  * Initialising and terminating the module.
515  */
516 
517 static int __init init_nlm(void)
518 {
519 #ifdef CONFIG_SYSCTL
520 	nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
521 	return nlm_sysctl_table ? 0 : -ENOMEM;
522 #else
523 	return 0;
524 #endif
525 }
526 
527 static void __exit exit_nlm(void)
528 {
529 	/* FIXME: delete all NLM clients */
530 	nlm_shutdown_hosts();
531 #ifdef CONFIG_SYSCTL
532 	unregister_sysctl_table(nlm_sysctl_table);
533 #endif
534 }
535 
536 module_init(init_nlm);
537 module_exit(exit_nlm);
538 
539 /*
540  * Define NLM program and procedures
541  */
542 static struct svc_version	nlmsvc_version1 = {
543 		.vs_vers	= 1,
544 		.vs_nproc	= 17,
545 		.vs_proc	= nlmsvc_procedures,
546 		.vs_xdrsize	= NLMSVC_XDRSIZE,
547 };
548 static struct svc_version	nlmsvc_version3 = {
549 		.vs_vers	= 3,
550 		.vs_nproc	= 24,
551 		.vs_proc	= nlmsvc_procedures,
552 		.vs_xdrsize	= NLMSVC_XDRSIZE,
553 };
554 #ifdef CONFIG_LOCKD_V4
555 static struct svc_version	nlmsvc_version4 = {
556 		.vs_vers	= 4,
557 		.vs_nproc	= 24,
558 		.vs_proc	= nlmsvc_procedures4,
559 		.vs_xdrsize	= NLMSVC_XDRSIZE,
560 };
561 #endif
562 static struct svc_version *	nlmsvc_version[] = {
563 	[1] = &nlmsvc_version1,
564 	[3] = &nlmsvc_version3,
565 #ifdef CONFIG_LOCKD_V4
566 	[4] = &nlmsvc_version4,
567 #endif
568 };
569 
570 static struct svc_stat		nlmsvc_stats;
571 
572 #define NLM_NRVERS	ARRAY_SIZE(nlmsvc_version)
573 static struct svc_program	nlmsvc_program = {
574 	.pg_prog		= NLM_PROGRAM,		/* program number */
575 	.pg_nvers		= NLM_NRVERS,		/* number of entries in nlmsvc_version */
576 	.pg_vers		= nlmsvc_version,	/* version table */
577 	.pg_name		= "lockd",		/* service name */
578 	.pg_class		= "nfsd",		/* share authentication with nfsd */
579 	.pg_stats		= &nlmsvc_stats,	/* stats table */
580 	.pg_authenticate = &lockd_authenticate	/* export authentication */
581 };
582