xref: /openbmc/linux/fs/lockd/svc.c (revision 22246614)
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_serv		*nlmsvc_serv;
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 
198 	flush_signals(current);
199 	if (nlmsvc_ops)
200 		nlmsvc_invalidate_all();
201 	nlm_shutdown_hosts();
202 
203 	unlock_kernel();
204 
205 	nlmsvc_task = NULL;
206 	nlmsvc_serv = NULL;
207 
208 	/* Exit the RPC thread */
209 	svc_exit_thread(rqstp);
210 
211 	return 0;
212 }
213 
214 /*
215  * Make any sockets that are needed but not present.
216  * If nlm_udpport or nlm_tcpport were set as module
217  * options, make those sockets unconditionally
218  */
219 static int make_socks(struct svc_serv *serv, int proto)
220 {
221 	static int warned;
222 	struct svc_xprt *xprt;
223 	int err = 0;
224 
225 	if (proto == IPPROTO_UDP || nlm_udpport) {
226 		xprt = svc_find_xprt(serv, "udp", 0, 0);
227 		if (!xprt)
228 			err = svc_create_xprt(serv, "udp", nlm_udpport,
229 					      SVC_SOCK_DEFAULTS);
230 		else
231 			svc_xprt_put(xprt);
232 	}
233 	if (err >= 0 && (proto == IPPROTO_TCP || nlm_tcpport)) {
234 		xprt = svc_find_xprt(serv, "tcp", 0, 0);
235 		if (!xprt)
236 			err = svc_create_xprt(serv, "tcp", nlm_tcpport,
237 					      SVC_SOCK_DEFAULTS);
238 		else
239 			svc_xprt_put(xprt);
240 	}
241 	if (err >= 0) {
242 		warned = 0;
243 		err = 0;
244 	} else if (warned++ == 0)
245 		printk(KERN_WARNING
246 		       "lockd_up: makesock failed, error=%d\n", err);
247 	return err;
248 }
249 
250 /*
251  * Bring up the lockd process if it's not already up.
252  */
253 int
254 lockd_up(int proto) /* Maybe add a 'family' option when IPv6 is supported ?? */
255 {
256 	struct svc_serv *serv;
257 	struct svc_rqst *rqstp;
258 	int		error = 0;
259 
260 	mutex_lock(&nlmsvc_mutex);
261 	/*
262 	 * Check whether we're already up and running.
263 	 */
264 	if (nlmsvc_serv) {
265 		if (proto)
266 			error = make_socks(nlmsvc_serv, proto);
267 		goto out;
268 	}
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 	if ((error = make_socks(serv, proto)) < 0)
286 		goto destroy_and_out;
287 
288 	/*
289 	 * Create the kernel thread and wait for it to start.
290 	 */
291 	rqstp = svc_prepare_thread(serv, &serv->sv_pools[0]);
292 	if (IS_ERR(rqstp)) {
293 		error = PTR_ERR(rqstp);
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 	nlmsvc_serv = rqstp->rq_server;
302 
303 	nlmsvc_task = kthread_run(lockd, rqstp, serv->sv_name);
304 	if (IS_ERR(nlmsvc_task)) {
305 		error = PTR_ERR(nlmsvc_task);
306 		nlmsvc_task = NULL;
307 		nlmsvc_serv = NULL;
308 		printk(KERN_WARNING
309 			"lockd_up: kthread_run failed, error=%d\n", error);
310 		svc_exit_thread(rqstp);
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(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 out:
350 	mutex_unlock(&nlmsvc_mutex);
351 }
352 EXPORT_SYMBOL(lockd_down);
353 
354 #ifdef CONFIG_SYSCTL
355 
356 /*
357  * Sysctl parameters (same as module parameters, different interface).
358  */
359 
360 static ctl_table nlm_sysctls[] = {
361 	{
362 		.ctl_name	= CTL_UNNUMBERED,
363 		.procname	= "nlm_grace_period",
364 		.data		= &nlm_grace_period,
365 		.maxlen		= sizeof(unsigned long),
366 		.mode		= 0644,
367 		.proc_handler	= &proc_doulongvec_minmax,
368 		.extra1		= (unsigned long *) &nlm_grace_period_min,
369 		.extra2		= (unsigned long *) &nlm_grace_period_max,
370 	},
371 	{
372 		.ctl_name	= CTL_UNNUMBERED,
373 		.procname	= "nlm_timeout",
374 		.data		= &nlm_timeout,
375 		.maxlen		= sizeof(unsigned long),
376 		.mode		= 0644,
377 		.proc_handler	= &proc_doulongvec_minmax,
378 		.extra1		= (unsigned long *) &nlm_timeout_min,
379 		.extra2		= (unsigned long *) &nlm_timeout_max,
380 	},
381 	{
382 		.ctl_name	= CTL_UNNUMBERED,
383 		.procname	= "nlm_udpport",
384 		.data		= &nlm_udpport,
385 		.maxlen		= sizeof(int),
386 		.mode		= 0644,
387 		.proc_handler	= &proc_dointvec_minmax,
388 		.extra1		= (int *) &nlm_port_min,
389 		.extra2		= (int *) &nlm_port_max,
390 	},
391 	{
392 		.ctl_name	= CTL_UNNUMBERED,
393 		.procname	= "nlm_tcpport",
394 		.data		= &nlm_tcpport,
395 		.maxlen		= sizeof(int),
396 		.mode		= 0644,
397 		.proc_handler	= &proc_dointvec_minmax,
398 		.extra1		= (int *) &nlm_port_min,
399 		.extra2		= (int *) &nlm_port_max,
400 	},
401 	{
402 		.ctl_name	= CTL_UNNUMBERED,
403 		.procname	= "nsm_use_hostnames",
404 		.data		= &nsm_use_hostnames,
405 		.maxlen		= sizeof(int),
406 		.mode		= 0644,
407 		.proc_handler	= &proc_dointvec,
408 	},
409 	{
410 		.ctl_name	= CTL_UNNUMBERED,
411 		.procname	= "nsm_local_state",
412 		.data		= &nsm_local_state,
413 		.maxlen		= sizeof(int),
414 		.mode		= 0644,
415 		.proc_handler	= &proc_dointvec,
416 	},
417 	{ .ctl_name = 0 }
418 };
419 
420 static ctl_table nlm_sysctl_dir[] = {
421 	{
422 		.ctl_name	= CTL_UNNUMBERED,
423 		.procname	= "nfs",
424 		.mode		= 0555,
425 		.child		= nlm_sysctls,
426 	},
427 	{ .ctl_name = 0 }
428 };
429 
430 static ctl_table nlm_sysctl_root[] = {
431 	{
432 		.ctl_name	= CTL_FS,
433 		.procname	= "fs",
434 		.mode		= 0555,
435 		.child		= nlm_sysctl_dir,
436 	},
437 	{ .ctl_name = 0 }
438 };
439 
440 #endif	/* CONFIG_SYSCTL */
441 
442 /*
443  * Module (and sysfs) parameters.
444  */
445 
446 #define param_set_min_max(name, type, which_strtol, min, max)		\
447 static int param_set_##name(const char *val, struct kernel_param *kp)	\
448 {									\
449 	char *endp;							\
450 	__typeof__(type) num = which_strtol(val, &endp, 0);		\
451 	if (endp == val || *endp || num < (min) || num > (max))		\
452 		return -EINVAL;						\
453 	*((int *) kp->arg) = num;					\
454 	return 0;							\
455 }
456 
457 static inline int is_callback(u32 proc)
458 {
459 	return proc == NLMPROC_GRANTED
460 		|| proc == NLMPROC_GRANTED_MSG
461 		|| proc == NLMPROC_TEST_RES
462 		|| proc == NLMPROC_LOCK_RES
463 		|| proc == NLMPROC_CANCEL_RES
464 		|| proc == NLMPROC_UNLOCK_RES
465 		|| proc == NLMPROC_NSM_NOTIFY;
466 }
467 
468 
469 static int lockd_authenticate(struct svc_rqst *rqstp)
470 {
471 	rqstp->rq_client = NULL;
472 	switch (rqstp->rq_authop->flavour) {
473 		case RPC_AUTH_NULL:
474 		case RPC_AUTH_UNIX:
475 			if (rqstp->rq_proc == 0)
476 				return SVC_OK;
477 			if (is_callback(rqstp->rq_proc)) {
478 				/* Leave it to individual procedures to
479 				 * call nlmsvc_lookup_host(rqstp)
480 				 */
481 				return SVC_OK;
482 			}
483 			return svc_set_client(rqstp);
484 	}
485 	return SVC_DENIED;
486 }
487 
488 
489 param_set_min_max(port, int, simple_strtol, 0, 65535)
490 param_set_min_max(grace_period, unsigned long, simple_strtoul,
491 		  nlm_grace_period_min, nlm_grace_period_max)
492 param_set_min_max(timeout, unsigned long, simple_strtoul,
493 		  nlm_timeout_min, nlm_timeout_max)
494 
495 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
496 MODULE_DESCRIPTION("NFS file locking service version " LOCKD_VERSION ".");
497 MODULE_LICENSE("GPL");
498 
499 module_param_call(nlm_grace_period, param_set_grace_period, param_get_ulong,
500 		  &nlm_grace_period, 0644);
501 module_param_call(nlm_timeout, param_set_timeout, param_get_ulong,
502 		  &nlm_timeout, 0644);
503 module_param_call(nlm_udpport, param_set_port, param_get_int,
504 		  &nlm_udpport, 0644);
505 module_param_call(nlm_tcpport, param_set_port, param_get_int,
506 		  &nlm_tcpport, 0644);
507 module_param(nsm_use_hostnames, bool, 0644);
508 
509 /*
510  * Initialising and terminating the module.
511  */
512 
513 static int __init init_nlm(void)
514 {
515 #ifdef CONFIG_SYSCTL
516 	nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
517 	return nlm_sysctl_table ? 0 : -ENOMEM;
518 #else
519 	return 0;
520 #endif
521 }
522 
523 static void __exit exit_nlm(void)
524 {
525 	/* FIXME: delete all NLM clients */
526 	nlm_shutdown_hosts();
527 #ifdef CONFIG_SYSCTL
528 	unregister_sysctl_table(nlm_sysctl_table);
529 #endif
530 }
531 
532 module_init(init_nlm);
533 module_exit(exit_nlm);
534 
535 /*
536  * Define NLM program and procedures
537  */
538 static struct svc_version	nlmsvc_version1 = {
539 		.vs_vers	= 1,
540 		.vs_nproc	= 17,
541 		.vs_proc	= nlmsvc_procedures,
542 		.vs_xdrsize	= NLMSVC_XDRSIZE,
543 };
544 static struct svc_version	nlmsvc_version3 = {
545 		.vs_vers	= 3,
546 		.vs_nproc	= 24,
547 		.vs_proc	= nlmsvc_procedures,
548 		.vs_xdrsize	= NLMSVC_XDRSIZE,
549 };
550 #ifdef CONFIG_LOCKD_V4
551 static struct svc_version	nlmsvc_version4 = {
552 		.vs_vers	= 4,
553 		.vs_nproc	= 24,
554 		.vs_proc	= nlmsvc_procedures4,
555 		.vs_xdrsize	= NLMSVC_XDRSIZE,
556 };
557 #endif
558 static struct svc_version *	nlmsvc_version[] = {
559 	[1] = &nlmsvc_version1,
560 	[3] = &nlmsvc_version3,
561 #ifdef CONFIG_LOCKD_V4
562 	[4] = &nlmsvc_version4,
563 #endif
564 };
565 
566 static struct svc_stat		nlmsvc_stats;
567 
568 #define NLM_NRVERS	ARRAY_SIZE(nlmsvc_version)
569 static struct svc_program	nlmsvc_program = {
570 	.pg_prog		= NLM_PROGRAM,		/* program number */
571 	.pg_nvers		= NLM_NRVERS,		/* number of entries in nlmsvc_version */
572 	.pg_vers		= nlmsvc_version,	/* version table */
573 	.pg_name		= "lockd",		/* service name */
574 	.pg_class		= "nfsd",		/* share authentication with nfsd */
575 	.pg_stats		= &nlmsvc_stats,	/* stats table */
576 	.pg_authenticate = &lockd_authenticate	/* export authentication */
577 };
578