xref: /openbmc/linux/fs/lockd/svc.c (revision d6783b2b)
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 	err = create_lockd_family(serv, PF_INET6);
247 	if (err < 0 && err != -EAFNOSUPPORT)
248 		goto out_err;
249 
250 	warned = 0;
251 	return 0;
252 
253 out_err:
254 	if (warned++ == 0)
255 		printk(KERN_WARNING
256 			"lockd_up: makesock failed, error=%d\n", err);
257 	return err;
258 }
259 
260 /*
261  * Bring up the lockd process if it's not already up.
262  */
263 int lockd_up(void)
264 {
265 	struct svc_serv *serv;
266 	int		error = 0;
267 
268 	mutex_lock(&nlmsvc_mutex);
269 	/*
270 	 * Check whether we're already up and running.
271 	 */
272 	if (nlmsvc_rqst)
273 		goto out;
274 
275 	/*
276 	 * Sanity check: if there's no pid,
277 	 * we should be the first user ...
278 	 */
279 	if (nlmsvc_users)
280 		printk(KERN_WARNING
281 			"lockd_up: no pid, %d users??\n", nlmsvc_users);
282 
283 	error = -ENOMEM;
284 	serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL);
285 	if (!serv) {
286 		printk(KERN_WARNING "lockd_up: create service failed\n");
287 		goto out;
288 	}
289 
290 	error = make_socks(serv);
291 	if (error < 0)
292 		goto destroy_and_out;
293 
294 	/*
295 	 * Create the kernel thread and wait for it to start.
296 	 */
297 	nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0]);
298 	if (IS_ERR(nlmsvc_rqst)) {
299 		error = PTR_ERR(nlmsvc_rqst);
300 		nlmsvc_rqst = NULL;
301 		printk(KERN_WARNING
302 			"lockd_up: svc_rqst allocation failed, error=%d\n",
303 			error);
304 		goto destroy_and_out;
305 	}
306 
307 	svc_sock_update_bufs(serv);
308 	serv->sv_maxconn = nlm_max_connections;
309 
310 	nlmsvc_task = kthread_run(lockd, nlmsvc_rqst, serv->sv_name);
311 	if (IS_ERR(nlmsvc_task)) {
312 		error = PTR_ERR(nlmsvc_task);
313 		svc_exit_thread(nlmsvc_rqst);
314 		nlmsvc_task = NULL;
315 		nlmsvc_rqst = NULL;
316 		printk(KERN_WARNING
317 			"lockd_up: kthread_run failed, error=%d\n", error);
318 		goto destroy_and_out;
319 	}
320 
321 	/*
322 	 * Note: svc_serv structures have an initial use count of 1,
323 	 * so we exit through here on both success and failure.
324 	 */
325 destroy_and_out:
326 	svc_destroy(serv);
327 out:
328 	if (!error)
329 		nlmsvc_users++;
330 	mutex_unlock(&nlmsvc_mutex);
331 	return error;
332 }
333 EXPORT_SYMBOL_GPL(lockd_up);
334 
335 /*
336  * Decrement the user count and bring down lockd if we're the last.
337  */
338 void
339 lockd_down(void)
340 {
341 	mutex_lock(&nlmsvc_mutex);
342 	if (nlmsvc_users) {
343 		if (--nlmsvc_users)
344 			goto out;
345 	} else {
346 		printk(KERN_ERR "lockd_down: no users! task=%p\n",
347 			nlmsvc_task);
348 		BUG();
349 	}
350 
351 	if (!nlmsvc_task) {
352 		printk(KERN_ERR "lockd_down: no lockd running.\n");
353 		BUG();
354 	}
355 	kthread_stop(nlmsvc_task);
356 	svc_exit_thread(nlmsvc_rqst);
357 	nlmsvc_task = NULL;
358 	nlmsvc_rqst = NULL;
359 out:
360 	mutex_unlock(&nlmsvc_mutex);
361 }
362 EXPORT_SYMBOL_GPL(lockd_down);
363 
364 #ifdef CONFIG_SYSCTL
365 
366 /*
367  * Sysctl parameters (same as module parameters, different interface).
368  */
369 
370 static ctl_table nlm_sysctls[] = {
371 	{
372 		.procname	= "nlm_grace_period",
373 		.data		= &nlm_grace_period,
374 		.maxlen		= sizeof(unsigned long),
375 		.mode		= 0644,
376 		.proc_handler	= proc_doulongvec_minmax,
377 		.extra1		= (unsigned long *) &nlm_grace_period_min,
378 		.extra2		= (unsigned long *) &nlm_grace_period_max,
379 	},
380 	{
381 		.procname	= "nlm_timeout",
382 		.data		= &nlm_timeout,
383 		.maxlen		= sizeof(unsigned long),
384 		.mode		= 0644,
385 		.proc_handler	= proc_doulongvec_minmax,
386 		.extra1		= (unsigned long *) &nlm_timeout_min,
387 		.extra2		= (unsigned long *) &nlm_timeout_max,
388 	},
389 	{
390 		.procname	= "nlm_udpport",
391 		.data		= &nlm_udpport,
392 		.maxlen		= sizeof(int),
393 		.mode		= 0644,
394 		.proc_handler	= proc_dointvec_minmax,
395 		.extra1		= (int *) &nlm_port_min,
396 		.extra2		= (int *) &nlm_port_max,
397 	},
398 	{
399 		.procname	= "nlm_tcpport",
400 		.data		= &nlm_tcpport,
401 		.maxlen		= sizeof(int),
402 		.mode		= 0644,
403 		.proc_handler	= proc_dointvec_minmax,
404 		.extra1		= (int *) &nlm_port_min,
405 		.extra2		= (int *) &nlm_port_max,
406 	},
407 	{
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 		.procname	= "nsm_local_state",
416 		.data		= &nsm_local_state,
417 		.maxlen		= sizeof(int),
418 		.mode		= 0644,
419 		.proc_handler	= proc_dointvec,
420 	},
421 	{ }
422 };
423 
424 static ctl_table nlm_sysctl_dir[] = {
425 	{
426 		.procname	= "nfs",
427 		.mode		= 0555,
428 		.child		= nlm_sysctls,
429 	},
430 	{ }
431 };
432 
433 static ctl_table nlm_sysctl_root[] = {
434 	{
435 		.procname	= "fs",
436 		.mode		= 0555,
437 		.child		= nlm_sysctl_dir,
438 	},
439 	{ }
440 };
441 
442 #endif	/* CONFIG_SYSCTL */
443 
444 /*
445  * Module (and sysfs) parameters.
446  */
447 
448 #define param_set_min_max(name, type, which_strtol, min, max)		\
449 static int param_set_##name(const char *val, struct kernel_param *kp)	\
450 {									\
451 	char *endp;							\
452 	__typeof__(type) num = which_strtol(val, &endp, 0);		\
453 	if (endp == val || *endp || num < (min) || num > (max))		\
454 		return -EINVAL;						\
455 	*((int *) kp->arg) = num;					\
456 	return 0;							\
457 }
458 
459 static inline int is_callback(u32 proc)
460 {
461 	return proc == NLMPROC_GRANTED
462 		|| proc == NLMPROC_GRANTED_MSG
463 		|| proc == NLMPROC_TEST_RES
464 		|| proc == NLMPROC_LOCK_RES
465 		|| proc == NLMPROC_CANCEL_RES
466 		|| proc == NLMPROC_UNLOCK_RES
467 		|| proc == NLMPROC_NSM_NOTIFY;
468 }
469 
470 
471 static int lockd_authenticate(struct svc_rqst *rqstp)
472 {
473 	rqstp->rq_client = NULL;
474 	switch (rqstp->rq_authop->flavour) {
475 		case RPC_AUTH_NULL:
476 		case RPC_AUTH_UNIX:
477 			if (rqstp->rq_proc == 0)
478 				return SVC_OK;
479 			if (is_callback(rqstp->rq_proc)) {
480 				/* Leave it to individual procedures to
481 				 * call nlmsvc_lookup_host(rqstp)
482 				 */
483 				return SVC_OK;
484 			}
485 			return svc_set_client(rqstp);
486 	}
487 	return SVC_DENIED;
488 }
489 
490 
491 param_set_min_max(port, int, simple_strtol, 0, 65535)
492 param_set_min_max(grace_period, unsigned long, simple_strtoul,
493 		  nlm_grace_period_min, nlm_grace_period_max)
494 param_set_min_max(timeout, unsigned long, simple_strtoul,
495 		  nlm_timeout_min, nlm_timeout_max)
496 
497 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
498 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
499 MODULE_LICENSE("GPL");
500 
501 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
502 		  &nlm_grace_period, 0644);
503 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
504 		  &nlm_timeout, 0644);
505 module_param_call(nlm_udpport, param_set_port, param_get_int,
506 		  &nlm_udpport, 0644);
507 module_param_call(nlm_tcpport, param_set_port, param_get_int,
508 		  &nlm_tcpport, 0644);
509 module_param(nsm_use_hostnames, bool, 0644);
510 module_param(nlm_max_connections, uint, 0644);
511 
512 /*
513  * Initialising and terminating the module.
514  */
515 
516 static int __init init_nlm(void)
517 {
518 #ifdef CONFIG_SYSCTL
519 	nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
520 	return nlm_sysctl_table ? 0 : -ENOMEM;
521 #else
522 	return 0;
523 #endif
524 }
525 
526 static void __exit exit_nlm(void)
527 {
528 	/* FIXME: delete all NLM clients */
529 	nlm_shutdown_hosts();
530 #ifdef CONFIG_SYSCTL
531 	unregister_sysctl_table(nlm_sysctl_table);
532 #endif
533 }
534 
535 module_init(init_nlm);
536 module_exit(exit_nlm);
537 
538 /*
539  * Define NLM program and procedures
540  */
541 static struct svc_version	nlmsvc_version1 = {
542 		.vs_vers	= 1,
543 		.vs_nproc	= 17,
544 		.vs_proc	= nlmsvc_procedures,
545 		.vs_xdrsize	= NLMSVC_XDRSIZE,
546 };
547 static struct svc_version	nlmsvc_version3 = {
548 		.vs_vers	= 3,
549 		.vs_nproc	= 24,
550 		.vs_proc	= nlmsvc_procedures,
551 		.vs_xdrsize	= NLMSVC_XDRSIZE,
552 };
553 #ifdef CONFIG_LOCKD_V4
554 static struct svc_version	nlmsvc_version4 = {
555 		.vs_vers	= 4,
556 		.vs_nproc	= 24,
557 		.vs_proc	= nlmsvc_procedures4,
558 		.vs_xdrsize	= NLMSVC_XDRSIZE,
559 };
560 #endif
561 static struct svc_version *	nlmsvc_version[] = {
562 	[1] = &nlmsvc_version1,
563 	[3] = &nlmsvc_version3,
564 #ifdef CONFIG_LOCKD_V4
565 	[4] = &nlmsvc_version4,
566 #endif
567 };
568 
569 static struct svc_stat		nlmsvc_stats;
570 
571 #define NLM_NRVERS	ARRAY_SIZE(nlmsvc_version)
572 static struct svc_program	nlmsvc_program = {
573 	.pg_prog		= NLM_PROGRAM,		/* program number */
574 	.pg_nvers		= NLM_NRVERS,		/* number of entries in nlmsvc_version */
575 	.pg_vers		= nlmsvc_version,	/* version table */
576 	.pg_name		= "lockd",		/* service name */
577 	.pg_class		= "nfsd",		/* share authentication with nfsd */
578 	.pg_stats		= &nlmsvc_stats,	/* stats table */
579 	.pg_authenticate = &lockd_authenticate	/* export authentication */
580 };
581