xref: /openbmc/linux/fs/lockd/svc.c (revision abd1ec4e)
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/lockd/sm_inter.h>
39 #include <linux/nfs.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(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 int				nlmsvc_grace_period;
55 unsigned long			nlmsvc_timeout;
56 
57 /*
58  * These can be set at insmod time (useful for NFS as root filesystem),
59  * and also changed through the sysctl interface.  -- Jamie Lokier, Aug 2003
60  */
61 static unsigned long		nlm_grace_period;
62 static unsigned long		nlm_timeout = LOCKD_DFLT_TIMEO;
63 static int			nlm_udpport, nlm_tcpport;
64 int				nsm_use_hostnames = 0;
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 unsigned long get_nfs_grace_period(void)
89 {
90 	unsigned long lockdgrace = get_lockd_grace_period();
91 	unsigned long nfsdgrace = 0;
92 
93 	if (nlmsvc_ops)
94 		nfsdgrace = nlmsvc_ops->get_grace_period();
95 
96 	return max(lockdgrace, nfsdgrace);
97 }
98 EXPORT_SYMBOL(get_nfs_grace_period);
99 
100 static unsigned long set_grace_period(void)
101 {
102 	nlmsvc_grace_period = 1;
103 	return get_nfs_grace_period() + jiffies;
104 }
105 
106 static inline void clear_grace_period(void)
107 {
108 	nlmsvc_grace_period = 0;
109 }
110 
111 /*
112  * This is the lockd kernel thread
113  */
114 static int
115 lockd(void *vrqstp)
116 {
117 	int		err = 0, preverr = 0;
118 	struct svc_rqst *rqstp = vrqstp;
119 	unsigned long grace_period_expire;
120 
121 	/* try_to_freeze() is called from svc_recv() */
122 	set_freezable();
123 
124 	/* Allow SIGKILL to tell lockd to drop all of its locks */
125 	allow_signal(SIGKILL);
126 
127 	dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
128 
129 	/*
130 	 * FIXME: it would be nice if lockd didn't spend its entire life
131 	 * running under the BKL. At the very least, it would be good to
132 	 * have someone clarify what it's intended to protect here. I've
133 	 * seen some handwavy posts about posix locking needing to be
134 	 * done under the BKL, but it's far from clear.
135 	 */
136 	lock_kernel();
137 
138 	if (!nlm_timeout)
139 		nlm_timeout = LOCKD_DFLT_TIMEO;
140 	nlmsvc_timeout = nlm_timeout * HZ;
141 
142 	grace_period_expire = set_grace_period();
143 
144 	/*
145 	 * The main request loop. We don't terminate until the last
146 	 * NFS mount or NFS daemon has gone away.
147 	 */
148 	while (!kthread_should_stop()) {
149 		long timeout = MAX_SCHEDULE_TIMEOUT;
150 		RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
151 
152 		if (signalled()) {
153 			flush_signals(current);
154 			if (nlmsvc_ops) {
155 				nlmsvc_invalidate_all();
156 				grace_period_expire = set_grace_period();
157 			}
158 			continue;
159 		}
160 
161 		/*
162 		 * Retry any blocked locks that have been notified by
163 		 * the VFS. Don't do this during grace period.
164 		 * (Theoretically, there shouldn't even be blocked locks
165 		 * during grace period).
166 		 */
167 		if (!nlmsvc_grace_period) {
168 			timeout = nlmsvc_retry_blocked();
169 		} else if (time_before(grace_period_expire, jiffies))
170 			clear_grace_period();
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 	if (nlmsvc_ops)
199 		nlmsvc_invalidate_all();
200 	nlm_shutdown_hosts();
201 	unlock_kernel();
202 	return 0;
203 }
204 
205 /*
206  * Make any sockets that are needed but not present.
207  * If nlm_udpport or nlm_tcpport were set as module
208  * options, make those sockets unconditionally
209  */
210 static int make_socks(struct svc_serv *serv, int proto)
211 {
212 	static int warned;
213 	struct svc_xprt *xprt;
214 	int err = 0;
215 
216 	if (proto == IPPROTO_UDP || nlm_udpport) {
217 		xprt = svc_find_xprt(serv, "udp", 0, 0);
218 		if (!xprt)
219 			err = svc_create_xprt(serv, "udp", nlm_udpport,
220 					      SVC_SOCK_DEFAULTS);
221 		else
222 			svc_xprt_put(xprt);
223 	}
224 	if (err >= 0 && (proto == IPPROTO_TCP || nlm_tcpport)) {
225 		xprt = svc_find_xprt(serv, "tcp", 0, 0);
226 		if (!xprt)
227 			err = svc_create_xprt(serv, "tcp", nlm_tcpport,
228 					      SVC_SOCK_DEFAULTS);
229 		else
230 			svc_xprt_put(xprt);
231 	}
232 	if (err >= 0) {
233 		warned = 0;
234 		err = 0;
235 	} else if (warned++ == 0)
236 		printk(KERN_WARNING
237 		       "lockd_up: makesock failed, error=%d\n", err);
238 	return err;
239 }
240 
241 /*
242  * Bring up the lockd process if it's not already up.
243  */
244 int
245 lockd_up(int proto) /* Maybe add a 'family' option when IPv6 is supported ?? */
246 {
247 	struct svc_serv *serv;
248 	int		error = 0;
249 
250 	mutex_lock(&nlmsvc_mutex);
251 	/*
252 	 * Check whether we're already up and running.
253 	 */
254 	if (nlmsvc_rqst) {
255 		if (proto)
256 			error = make_socks(nlmsvc_rqst->rq_server, proto);
257 		goto out;
258 	}
259 
260 	/*
261 	 * Sanity check: if there's no pid,
262 	 * we should be the first user ...
263 	 */
264 	if (nlmsvc_users)
265 		printk(KERN_WARNING
266 			"lockd_up: no pid, %d users??\n", nlmsvc_users);
267 
268 	error = -ENOMEM;
269 	serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL);
270 	if (!serv) {
271 		printk(KERN_WARNING "lockd_up: create service failed\n");
272 		goto out;
273 	}
274 
275 	if ((error = make_socks(serv, proto)) < 0)
276 		goto destroy_and_out;
277 
278 	/*
279 	 * Create the kernel thread and wait for it to start.
280 	 */
281 	nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0]);
282 	if (IS_ERR(nlmsvc_rqst)) {
283 		error = PTR_ERR(nlmsvc_rqst);
284 		nlmsvc_rqst = NULL;
285 		printk(KERN_WARNING
286 			"lockd_up: svc_rqst allocation failed, error=%d\n",
287 			error);
288 		goto destroy_and_out;
289 	}
290 
291 	svc_sock_update_bufs(serv);
292 
293 	nlmsvc_task = kthread_run(lockd, nlmsvc_rqst, serv->sv_name);
294 	if (IS_ERR(nlmsvc_task)) {
295 		error = PTR_ERR(nlmsvc_task);
296 		svc_exit_thread(nlmsvc_rqst);
297 		nlmsvc_task = NULL;
298 		nlmsvc_rqst = NULL;
299 		printk(KERN_WARNING
300 			"lockd_up: kthread_run failed, error=%d\n", error);
301 		goto destroy_and_out;
302 	}
303 
304 	/*
305 	 * Note: svc_serv structures have an initial use count of 1,
306 	 * so we exit through here on both success and failure.
307 	 */
308 destroy_and_out:
309 	svc_destroy(serv);
310 out:
311 	if (!error)
312 		nlmsvc_users++;
313 	mutex_unlock(&nlmsvc_mutex);
314 	return error;
315 }
316 EXPORT_SYMBOL(lockd_up);
317 
318 /*
319  * Decrement the user count and bring down lockd if we're the last.
320  */
321 void
322 lockd_down(void)
323 {
324 	mutex_lock(&nlmsvc_mutex);
325 	if (nlmsvc_users) {
326 		if (--nlmsvc_users)
327 			goto out;
328 	} else {
329 		printk(KERN_ERR "lockd_down: no users! task=%p\n",
330 			nlmsvc_task);
331 		BUG();
332 	}
333 
334 	if (!nlmsvc_task) {
335 		printk(KERN_ERR "lockd_down: no lockd running.\n");
336 		BUG();
337 	}
338 	kthread_stop(nlmsvc_task);
339 	svc_exit_thread(nlmsvc_rqst);
340 	nlmsvc_task = NULL;
341 	nlmsvc_rqst = NULL;
342 out:
343 	mutex_unlock(&nlmsvc_mutex);
344 }
345 EXPORT_SYMBOL(lockd_down);
346 
347 #ifdef CONFIG_SYSCTL
348 
349 /*
350  * Sysctl parameters (same as module parameters, different interface).
351  */
352 
353 static ctl_table nlm_sysctls[] = {
354 	{
355 		.ctl_name	= CTL_UNNUMBERED,
356 		.procname	= "nlm_grace_period",
357 		.data		= &nlm_grace_period,
358 		.maxlen		= sizeof(unsigned long),
359 		.mode		= 0644,
360 		.proc_handler	= &proc_doulongvec_minmax,
361 		.extra1		= (unsigned long *) &nlm_grace_period_min,
362 		.extra2		= (unsigned long *) &nlm_grace_period_max,
363 	},
364 	{
365 		.ctl_name	= CTL_UNNUMBERED,
366 		.procname	= "nlm_timeout",
367 		.data		= &nlm_timeout,
368 		.maxlen		= sizeof(unsigned long),
369 		.mode		= 0644,
370 		.proc_handler	= &proc_doulongvec_minmax,
371 		.extra1		= (unsigned long *) &nlm_timeout_min,
372 		.extra2		= (unsigned long *) &nlm_timeout_max,
373 	},
374 	{
375 		.ctl_name	= CTL_UNNUMBERED,
376 		.procname	= "nlm_udpport",
377 		.data		= &nlm_udpport,
378 		.maxlen		= sizeof(int),
379 		.mode		= 0644,
380 		.proc_handler	= &proc_dointvec_minmax,
381 		.extra1		= (int *) &nlm_port_min,
382 		.extra2		= (int *) &nlm_port_max,
383 	},
384 	{
385 		.ctl_name	= CTL_UNNUMBERED,
386 		.procname	= "nlm_tcpport",
387 		.data		= &nlm_tcpport,
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	= "nsm_use_hostnames",
397 		.data		= &nsm_use_hostnames,
398 		.maxlen		= sizeof(int),
399 		.mode		= 0644,
400 		.proc_handler	= &proc_dointvec,
401 	},
402 	{
403 		.ctl_name	= CTL_UNNUMBERED,
404 		.procname	= "nsm_local_state",
405 		.data		= &nsm_local_state,
406 		.maxlen		= sizeof(int),
407 		.mode		= 0644,
408 		.proc_handler	= &proc_dointvec,
409 	},
410 	{ .ctl_name = 0 }
411 };
412 
413 static ctl_table nlm_sysctl_dir[] = {
414 	{
415 		.ctl_name	= CTL_UNNUMBERED,
416 		.procname	= "nfs",
417 		.mode		= 0555,
418 		.child		= nlm_sysctls,
419 	},
420 	{ .ctl_name = 0 }
421 };
422 
423 static ctl_table nlm_sysctl_root[] = {
424 	{
425 		.ctl_name	= CTL_FS,
426 		.procname	= "fs",
427 		.mode		= 0555,
428 		.child		= nlm_sysctl_dir,
429 	},
430 	{ .ctl_name = 0 }
431 };
432 
433 #endif	/* CONFIG_SYSCTL */
434 
435 /*
436  * Module (and sysfs) parameters.
437  */
438 
439 #define param_set_min_max(name, type, which_strtol, min, max)		\
440 static int param_set_##name(const char *val, struct kernel_param *kp)	\
441 {									\
442 	char *endp;							\
443 	__typeof__(type) num = which_strtol(val, &endp, 0);		\
444 	if (endp == val || *endp || num < (min) || num > (max))		\
445 		return -EINVAL;						\
446 	*((int *) kp->arg) = num;					\
447 	return 0;							\
448 }
449 
450 static inline int is_callback(u32 proc)
451 {
452 	return proc == NLMPROC_GRANTED
453 		|| proc == NLMPROC_GRANTED_MSG
454 		|| proc == NLMPROC_TEST_RES
455 		|| proc == NLMPROC_LOCK_RES
456 		|| proc == NLMPROC_CANCEL_RES
457 		|| proc == NLMPROC_UNLOCK_RES
458 		|| proc == NLMPROC_NSM_NOTIFY;
459 }
460 
461 
462 static int lockd_authenticate(struct svc_rqst *rqstp)
463 {
464 	rqstp->rq_client = NULL;
465 	switch (rqstp->rq_authop->flavour) {
466 		case RPC_AUTH_NULL:
467 		case RPC_AUTH_UNIX:
468 			if (rqstp->rq_proc == 0)
469 				return SVC_OK;
470 			if (is_callback(rqstp->rq_proc)) {
471 				/* Leave it to individual procedures to
472 				 * call nlmsvc_lookup_host(rqstp)
473 				 */
474 				return SVC_OK;
475 			}
476 			return svc_set_client(rqstp);
477 	}
478 	return SVC_DENIED;
479 }
480 
481 
482 param_set_min_max(port, int, simple_strtol, 0, 65535)
483 param_set_min_max(grace_period, unsigned long, simple_strtoul,
484 		  nlm_grace_period_min, nlm_grace_period_max)
485 param_set_min_max(timeout, unsigned long, simple_strtoul,
486 		  nlm_timeout_min, nlm_timeout_max)
487 
488 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
489 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
490 MODULE_LICENSE("GPL");
491 
492 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
493 		  &nlm_grace_period, 0644);
494 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
495 		  &nlm_timeout, 0644);
496 module_param_call(nlm_udpport, param_set_port, param_get_int,
497 		  &nlm_udpport, 0644);
498 module_param_call(nlm_tcpport, param_set_port, param_get_int,
499 		  &nlm_tcpport, 0644);
500 module_param(nsm_use_hostnames, bool, 0644);
501 
502 /*
503  * Initialising and terminating the module.
504  */
505 
506 static int __init init_nlm(void)
507 {
508 #ifdef CONFIG_SYSCTL
509 	nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
510 	return nlm_sysctl_table ? 0 : -ENOMEM;
511 #else
512 	return 0;
513 #endif
514 }
515 
516 static void __exit exit_nlm(void)
517 {
518 	/* FIXME: delete all NLM clients */
519 	nlm_shutdown_hosts();
520 #ifdef CONFIG_SYSCTL
521 	unregister_sysctl_table(nlm_sysctl_table);
522 #endif
523 }
524 
525 module_init(init_nlm);
526 module_exit(exit_nlm);
527 
528 /*
529  * Define NLM program and procedures
530  */
531 static struct svc_version	nlmsvc_version1 = {
532 		.vs_vers	= 1,
533 		.vs_nproc	= 17,
534 		.vs_proc	= nlmsvc_procedures,
535 		.vs_xdrsize	= NLMSVC_XDRSIZE,
536 };
537 static struct svc_version	nlmsvc_version3 = {
538 		.vs_vers	= 3,
539 		.vs_nproc	= 24,
540 		.vs_proc	= nlmsvc_procedures,
541 		.vs_xdrsize	= NLMSVC_XDRSIZE,
542 };
543 #ifdef CONFIG_LOCKD_V4
544 static struct svc_version	nlmsvc_version4 = {
545 		.vs_vers	= 4,
546 		.vs_nproc	= 24,
547 		.vs_proc	= nlmsvc_procedures4,
548 		.vs_xdrsize	= NLMSVC_XDRSIZE,
549 };
550 #endif
551 static struct svc_version *	nlmsvc_version[] = {
552 	[1] = &nlmsvc_version1,
553 	[3] = &nlmsvc_version3,
554 #ifdef CONFIG_LOCKD_V4
555 	[4] = &nlmsvc_version4,
556 #endif
557 };
558 
559 static struct svc_stat		nlmsvc_stats;
560 
561 #define NLM_NRVERS	ARRAY_SIZE(nlmsvc_version)
562 static struct svc_program	nlmsvc_program = {
563 	.pg_prog		= NLM_PROGRAM,		/* program number */
564 	.pg_nvers		= NLM_NRVERS,		/* number of entries in nlmsvc_version */
565 	.pg_vers		= nlmsvc_version,	/* version table */
566 	.pg_name		= "lockd",		/* service name */
567 	.pg_class		= "nfsd",		/* share authentication with nfsd */
568 	.pg_stats		= &nlmsvc_stats,	/* stats table */
569 	.pg_authenticate = &lockd_authenticate	/* export authentication */
570 };
571