xref: /openbmc/linux/fs/lockd/svc.c (revision 3252b11f)
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 static void restart_grace(void)
108 {
109 	if (nlmsvc_ops) {
110 		cancel_delayed_work_sync(&grace_period_end);
111 		locks_end_grace(&lockd_manager);
112 		nlmsvc_invalidate_all();
113 		set_grace_period();
114 	}
115 }
116 
117 /*
118  * This is the lockd kernel thread
119  */
120 static int
121 lockd(void *vrqstp)
122 {
123 	int		err = 0, preverr = 0;
124 	struct svc_rqst *rqstp = vrqstp;
125 
126 	/* try_to_freeze() is called from svc_recv() */
127 	set_freezable();
128 
129 	/* Allow SIGKILL to tell lockd to drop all of its locks */
130 	allow_signal(SIGKILL);
131 
132 	dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
133 
134 	/*
135 	 * FIXME: it would be nice if lockd didn't spend its entire life
136 	 * running under the BKL. At the very least, it would be good to
137 	 * have someone clarify what it's intended to protect here. I've
138 	 * seen some handwavy posts about posix locking needing to be
139 	 * done under the BKL, but it's far from clear.
140 	 */
141 	lock_kernel();
142 
143 	if (!nlm_timeout)
144 		nlm_timeout = LOCKD_DFLT_TIMEO;
145 	nlmsvc_timeout = nlm_timeout * HZ;
146 
147 	set_grace_period();
148 
149 	/*
150 	 * The main request loop. We don't terminate until the last
151 	 * NFS mount or NFS daemon has gone away.
152 	 */
153 	while (!kthread_should_stop()) {
154 		long timeout = MAX_SCHEDULE_TIMEOUT;
155 		RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
156 
157 		/* update sv_maxconn if it has changed */
158 		rqstp->rq_server->sv_maxconn = nlm_max_connections;
159 
160 		if (signalled()) {
161 			flush_signals(current);
162 			restart_grace();
163 			continue;
164 		}
165 
166 		timeout = nlmsvc_retry_blocked();
167 
168 		/*
169 		 * Find a socket with data available and call its
170 		 * recvfrom routine.
171 		 */
172 		err = svc_recv(rqstp, timeout);
173 		if (err == -EAGAIN || err == -EINTR) {
174 			preverr = err;
175 			continue;
176 		}
177 		if (err < 0) {
178 			if (err != preverr) {
179 				printk(KERN_WARNING "%s: unexpected error "
180 					"from svc_recv (%d)\n", __func__, err);
181 				preverr = err;
182 			}
183 			schedule_timeout_interruptible(HZ);
184 			continue;
185 		}
186 		preverr = err;
187 
188 		dprintk("lockd: request from %s\n",
189 				svc_print_addr(rqstp, buf, sizeof(buf)));
190 
191 		svc_process(rqstp);
192 	}
193 	flush_signals(current);
194 	cancel_delayed_work_sync(&grace_period_end);
195 	locks_end_grace(&lockd_manager);
196 	if (nlmsvc_ops)
197 		nlmsvc_invalidate_all();
198 	nlm_shutdown_hosts();
199 	unlock_kernel();
200 	return 0;
201 }
202 
203 static int create_lockd_listener(struct svc_serv *serv, const char *name,
204 				 const int family, const unsigned short port)
205 {
206 	struct svc_xprt *xprt;
207 
208 	xprt = svc_find_xprt(serv, name, family, 0);
209 	if (xprt == NULL)
210 		return svc_create_xprt(serv, name, family, port,
211 						SVC_SOCK_DEFAULTS);
212 	svc_xprt_put(xprt);
213 	return 0;
214 }
215 
216 static int create_lockd_family(struct svc_serv *serv, const int family)
217 {
218 	int err;
219 
220 	err = create_lockd_listener(serv, "udp", family, nlm_udpport);
221 	if (err < 0)
222 		return err;
223 
224 	return create_lockd_listener(serv, "tcp", family, nlm_tcpport);
225 }
226 
227 /*
228  * Ensure there are active UDP and TCP listeners for lockd.
229  *
230  * Even if we have only TCP NFS mounts and/or TCP NFSDs, some
231  * local services (such as rpc.statd) still require UDP, and
232  * some NFS servers do not yet support NLM over TCP.
233  *
234  * Returns zero if all listeners are available; otherwise a
235  * negative errno value is returned.
236  */
237 static int make_socks(struct svc_serv *serv)
238 {
239 	static int warned;
240 	int err;
241 
242 	err = create_lockd_family(serv, PF_INET);
243 	if (err < 0)
244 		goto out_err;
245 
246 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
247 	err = create_lockd_family(serv, PF_INET6);
248 	if (err < 0 && err != -EAFNOSUPPORT)
249 		goto out_err;
250 #endif	/* CONFIG_IPV6 || CONFIG_IPV6_MODULE */
251 
252 	warned = 0;
253 	return 0;
254 
255 out_err:
256 	if (warned++ == 0)
257 		printk(KERN_WARNING
258 			"lockd_up: makesock failed, error=%d\n", err);
259 	return err;
260 }
261 
262 /*
263  * Bring up the lockd process if it's not already up.
264  */
265 int lockd_up(void)
266 {
267 	struct svc_serv *serv;
268 	int		error = 0;
269 
270 	mutex_lock(&nlmsvc_mutex);
271 	/*
272 	 * Check whether we're already up and running.
273 	 */
274 	if (nlmsvc_rqst)
275 		goto out;
276 
277 	/*
278 	 * Sanity check: if there's no pid,
279 	 * we should be the first user ...
280 	 */
281 	if (nlmsvc_users)
282 		printk(KERN_WARNING
283 			"lockd_up: no pid, %d users??\n", nlmsvc_users);
284 
285 	error = -ENOMEM;
286 	serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL);
287 	if (!serv) {
288 		printk(KERN_WARNING "lockd_up: create service failed\n");
289 		goto out;
290 	}
291 
292 	error = make_socks(serv);
293 	if (error < 0)
294 		goto destroy_and_out;
295 
296 	/*
297 	 * Create the kernel thread and wait for it to start.
298 	 */
299 	nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0]);
300 	if (IS_ERR(nlmsvc_rqst)) {
301 		error = PTR_ERR(nlmsvc_rqst);
302 		nlmsvc_rqst = NULL;
303 		printk(KERN_WARNING
304 			"lockd_up: svc_rqst allocation failed, error=%d\n",
305 			error);
306 		goto destroy_and_out;
307 	}
308 
309 	svc_sock_update_bufs(serv);
310 	serv->sv_maxconn = nlm_max_connections;
311 
312 	nlmsvc_task = kthread_run(lockd, nlmsvc_rqst, serv->sv_name);
313 	if (IS_ERR(nlmsvc_task)) {
314 		error = PTR_ERR(nlmsvc_task);
315 		svc_exit_thread(nlmsvc_rqst);
316 		nlmsvc_task = NULL;
317 		nlmsvc_rqst = NULL;
318 		printk(KERN_WARNING
319 			"lockd_up: kthread_run failed, error=%d\n", error);
320 		goto destroy_and_out;
321 	}
322 
323 	/*
324 	 * Note: svc_serv structures have an initial use count of 1,
325 	 * so we exit through here on both success and failure.
326 	 */
327 destroy_and_out:
328 	svc_destroy(serv);
329 out:
330 	if (!error)
331 		nlmsvc_users++;
332 	mutex_unlock(&nlmsvc_mutex);
333 	return error;
334 }
335 EXPORT_SYMBOL_GPL(lockd_up);
336 
337 /*
338  * Decrement the user count and bring down lockd if we're the last.
339  */
340 void
341 lockd_down(void)
342 {
343 	mutex_lock(&nlmsvc_mutex);
344 	if (nlmsvc_users) {
345 		if (--nlmsvc_users)
346 			goto out;
347 	} else {
348 		printk(KERN_ERR "lockd_down: no users! task=%p\n",
349 			nlmsvc_task);
350 		BUG();
351 	}
352 
353 	if (!nlmsvc_task) {
354 		printk(KERN_ERR "lockd_down: no lockd running.\n");
355 		BUG();
356 	}
357 	kthread_stop(nlmsvc_task);
358 	svc_exit_thread(nlmsvc_rqst);
359 	nlmsvc_task = NULL;
360 	nlmsvc_rqst = NULL;
361 out:
362 	mutex_unlock(&nlmsvc_mutex);
363 }
364 EXPORT_SYMBOL_GPL(lockd_down);
365 
366 #ifdef CONFIG_SYSCTL
367 
368 /*
369  * Sysctl parameters (same as module parameters, different interface).
370  */
371 
372 static ctl_table nlm_sysctls[] = {
373 	{
374 		.procname	= "nlm_grace_period",
375 		.data		= &nlm_grace_period,
376 		.maxlen		= sizeof(unsigned long),
377 		.mode		= 0644,
378 		.proc_handler	= proc_doulongvec_minmax,
379 		.extra1		= (unsigned long *) &nlm_grace_period_min,
380 		.extra2		= (unsigned long *) &nlm_grace_period_max,
381 	},
382 	{
383 		.procname	= "nlm_timeout",
384 		.data		= &nlm_timeout,
385 		.maxlen		= sizeof(unsigned long),
386 		.mode		= 0644,
387 		.proc_handler	= proc_doulongvec_minmax,
388 		.extra1		= (unsigned long *) &nlm_timeout_min,
389 		.extra2		= (unsigned long *) &nlm_timeout_max,
390 	},
391 	{
392 		.procname	= "nlm_udpport",
393 		.data		= &nlm_udpport,
394 		.maxlen		= sizeof(int),
395 		.mode		= 0644,
396 		.proc_handler	= proc_dointvec_minmax,
397 		.extra1		= (int *) &nlm_port_min,
398 		.extra2		= (int *) &nlm_port_max,
399 	},
400 	{
401 		.procname	= "nlm_tcpport",
402 		.data		= &nlm_tcpport,
403 		.maxlen		= sizeof(int),
404 		.mode		= 0644,
405 		.proc_handler	= proc_dointvec_minmax,
406 		.extra1		= (int *) &nlm_port_min,
407 		.extra2		= (int *) &nlm_port_max,
408 	},
409 	{
410 		.procname	= "nsm_use_hostnames",
411 		.data		= &nsm_use_hostnames,
412 		.maxlen		= sizeof(int),
413 		.mode		= 0644,
414 		.proc_handler	= proc_dointvec,
415 	},
416 	{
417 		.procname	= "nsm_local_state",
418 		.data		= &nsm_local_state,
419 		.maxlen		= sizeof(int),
420 		.mode		= 0644,
421 		.proc_handler	= proc_dointvec,
422 	},
423 	{ }
424 };
425 
426 static ctl_table nlm_sysctl_dir[] = {
427 	{
428 		.procname	= "nfs",
429 		.mode		= 0555,
430 		.child		= nlm_sysctls,
431 	},
432 	{ }
433 };
434 
435 static ctl_table nlm_sysctl_root[] = {
436 	{
437 		.procname	= "fs",
438 		.mode		= 0555,
439 		.child		= nlm_sysctl_dir,
440 	},
441 	{ }
442 };
443 
444 #endif	/* CONFIG_SYSCTL */
445 
446 /*
447  * Module (and sysfs) parameters.
448  */
449 
450 #define param_set_min_max(name, type, which_strtol, min, max)		\
451 static int param_set_##name(const char *val, struct kernel_param *kp)	\
452 {									\
453 	char *endp;							\
454 	__typeof__(type) num = which_strtol(val, &endp, 0);		\
455 	if (endp == val || *endp || num < (min) || num > (max))		\
456 		return -EINVAL;						\
457 	*((int *) kp->arg) = num;					\
458 	return 0;							\
459 }
460 
461 static inline int is_callback(u32 proc)
462 {
463 	return proc == NLMPROC_GRANTED
464 		|| proc == NLMPROC_GRANTED_MSG
465 		|| proc == NLMPROC_TEST_RES
466 		|| proc == NLMPROC_LOCK_RES
467 		|| proc == NLMPROC_CANCEL_RES
468 		|| proc == NLMPROC_UNLOCK_RES
469 		|| proc == NLMPROC_NSM_NOTIFY;
470 }
471 
472 
473 static int lockd_authenticate(struct svc_rqst *rqstp)
474 {
475 	rqstp->rq_client = NULL;
476 	switch (rqstp->rq_authop->flavour) {
477 		case RPC_AUTH_NULL:
478 		case RPC_AUTH_UNIX:
479 			if (rqstp->rq_proc == 0)
480 				return SVC_OK;
481 			if (is_callback(rqstp->rq_proc)) {
482 				/* Leave it to individual procedures to
483 				 * call nlmsvc_lookup_host(rqstp)
484 				 */
485 				return SVC_OK;
486 			}
487 			return svc_set_client(rqstp);
488 	}
489 	return SVC_DENIED;
490 }
491 
492 
493 param_set_min_max(port, int, simple_strtol, 0, 65535)
494 param_set_min_max(grace_period, unsigned long, simple_strtoul,
495 		  nlm_grace_period_min, nlm_grace_period_max)
496 param_set_min_max(timeout, unsigned long, simple_strtoul,
497 		  nlm_timeout_min, nlm_timeout_max)
498 
499 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
500 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
501 MODULE_LICENSE("GPL");
502 
503 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
504 		  &nlm_grace_period, 0644);
505 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
506 		  &nlm_timeout, 0644);
507 module_param_call(nlm_udpport, param_set_port, param_get_int,
508 		  &nlm_udpport, 0644);
509 module_param_call(nlm_tcpport, param_set_port, param_get_int,
510 		  &nlm_tcpport, 0644);
511 module_param(nsm_use_hostnames, bool, 0644);
512 module_param(nlm_max_connections, uint, 0644);
513 
514 /*
515  * Initialising and terminating the module.
516  */
517 
518 static int __init init_nlm(void)
519 {
520 #ifdef CONFIG_SYSCTL
521 	nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
522 	return nlm_sysctl_table ? 0 : -ENOMEM;
523 #else
524 	return 0;
525 #endif
526 }
527 
528 static void __exit exit_nlm(void)
529 {
530 	/* FIXME: delete all NLM clients */
531 	nlm_shutdown_hosts();
532 #ifdef CONFIG_SYSCTL
533 	unregister_sysctl_table(nlm_sysctl_table);
534 #endif
535 }
536 
537 module_init(init_nlm);
538 module_exit(exit_nlm);
539 
540 /*
541  * Define NLM program and procedures
542  */
543 static struct svc_version	nlmsvc_version1 = {
544 		.vs_vers	= 1,
545 		.vs_nproc	= 17,
546 		.vs_proc	= nlmsvc_procedures,
547 		.vs_xdrsize	= NLMSVC_XDRSIZE,
548 };
549 static struct svc_version	nlmsvc_version3 = {
550 		.vs_vers	= 3,
551 		.vs_nproc	= 24,
552 		.vs_proc	= nlmsvc_procedures,
553 		.vs_xdrsize	= NLMSVC_XDRSIZE,
554 };
555 #ifdef CONFIG_LOCKD_V4
556 static struct svc_version	nlmsvc_version4 = {
557 		.vs_vers	= 4,
558 		.vs_nproc	= 24,
559 		.vs_proc	= nlmsvc_procedures4,
560 		.vs_xdrsize	= NLMSVC_XDRSIZE,
561 };
562 #endif
563 static struct svc_version *	nlmsvc_version[] = {
564 	[1] = &nlmsvc_version1,
565 	[3] = &nlmsvc_version3,
566 #ifdef CONFIG_LOCKD_V4
567 	[4] = &nlmsvc_version4,
568 #endif
569 };
570 
571 static struct svc_stat		nlmsvc_stats;
572 
573 #define NLM_NRVERS	ARRAY_SIZE(nlmsvc_version)
574 static struct svc_program	nlmsvc_program = {
575 	.pg_prog		= NLM_PROGRAM,		/* program number */
576 	.pg_nvers		= NLM_NRVERS,		/* number of entries in nlmsvc_version */
577 	.pg_vers		= nlmsvc_version,	/* version table */
578 	.pg_name		= "lockd",		/* service name */
579 	.pg_class		= "nfsd",		/* share authentication with nfsd */
580 	.pg_stats		= &nlmsvc_stats,	/* stats table */
581 	.pg_authenticate = &lockd_authenticate	/* export authentication */
582 };
583