xref: /trueos/sys/nlm/nlm_prot_impl.c (revision 41ac78dbf9367b7fe43a331d07447cffbd3d7115)
1 /*-
2  * Copyright (c) 2008 Isilon Inc http://www.isilon.com/
3  * Authors: Doug Rabson <dfr@rabson.org>
4  * Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org>
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 #include "opt_inet6.h"
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include <sys/param.h>
34 #include <sys/fail.h>
35 #include <sys/fcntl.h>
36 #include <sys/kernel.h>
37 #include <sys/kthread.h>
38 #include <sys/lockf.h>
39 #include <sys/malloc.h>
40 #include <sys/mount.h>
41 #if __FreeBSD_version >= 700000
42 #include <sys/priv.h>
43 #endif
44 #include <sys/proc.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/syscall.h>
48 #include <sys/sysctl.h>
49 #include <sys/sysent.h>
50 #include <sys/syslog.h>
51 #include <sys/sysproto.h>
52 #include <sys/systm.h>
53 #include <sys/taskqueue.h>
54 #include <sys/unistd.h>
55 #include <sys/vnode.h>
56 
57 #include <nfs/nfsproto.h>
58 #include <nfs/nfs_lock.h>
59 
60 #include <nlm/nlm_prot.h>
61 #include <nlm/sm_inter.h>
62 #include <nlm/nlm.h>
63 #include <rpc/rpc_com.h>
64 #include <rpc/rpcb_prot.h>
65 
66 MALLOC_DEFINE(M_NLM, "NLM", "Network Lock Manager");
67 
68 /*
69  * If a host is inactive (and holds no locks) for this amount of
70  * seconds, we consider it idle and stop tracking it.
71  */
72 #define NLM_IDLE_TIMEOUT	30
73 
74 /*
75  * We check the host list for idle every few seconds.
76  */
77 #define NLM_IDLE_PERIOD		5
78 
79 /*
80  * We only look for GRANTED_RES messages for a little while.
81  */
82 #define NLM_EXPIRE_TIMEOUT	10
83 
84 /*
85  * Support for sysctl vfs.nlm.sysid
86  */
87 static SYSCTL_NODE(_vfs, OID_AUTO, nlm, CTLFLAG_RW, NULL,
88     "Network Lock Manager");
89 static SYSCTL_NODE(_vfs_nlm, OID_AUTO, sysid, CTLFLAG_RW, NULL, "");
90 
91 /*
92  * Syscall hooks
93  */
94 static int nlm_syscall_offset = SYS_nlm_syscall;
95 static struct sysent nlm_syscall_prev_sysent;
96 #if __FreeBSD_version < 700000
97 static struct sysent nlm_syscall_sysent = {
98 	(sizeof(struct nlm_syscall_args) / sizeof(register_t)) | SYF_MPSAFE,
99 	(sy_call_t *) nlm_syscall
100 };
101 #else
102 MAKE_SYSENT(nlm_syscall);
103 #endif
104 static bool_t nlm_syscall_registered = FALSE;
105 
106 /*
107  * Debug level passed in from userland. We also support a sysctl hook
108  * so that it can be changed on a live system.
109  */
110 static int nlm_debug_level;
111 SYSCTL_INT(_debug, OID_AUTO, nlm_debug, CTLFLAG_RW, &nlm_debug_level, 0, "");
112 
113 #define NLM_DEBUG(_level, args...)			\
114 	do {						\
115 		if (nlm_debug_level >= (_level))	\
116 			log(LOG_DEBUG, args);		\
117 	} while(0)
118 #define NLM_ERR(args...)			\
119 	do {					\
120 		log(LOG_ERR, args);		\
121 	} while(0)
122 
123 /*
124  * Grace period handling. The value of nlm_grace_threshold is the
125  * value of time_uptime after which we are serving requests normally.
126  */
127 static time_t nlm_grace_threshold;
128 
129 /*
130  * We check for idle hosts if time_uptime is greater than
131  * nlm_next_idle_check,
132  */
133 static time_t nlm_next_idle_check;
134 
135 /*
136  * A flag to indicate the server is already running.
137  */
138 static int nlm_is_running;
139 
140 /*
141  * A socket to use for RPC - shared by all IPv4 RPC clients.
142  */
143 static struct socket *nlm_socket;
144 
145 #ifdef INET6
146 
147 /*
148  * A socket to use for RPC - shared by all IPv6 RPC clients.
149  */
150 static struct socket *nlm_socket6;
151 
152 #endif
153 
154 /*
155  * An RPC client handle that can be used to communicate with the local
156  * NSM.
157  */
158 static CLIENT *nlm_nsm;
159 
160 /*
161  * An AUTH handle for the server's creds.
162  */
163 static AUTH *nlm_auth;
164 
165 /*
166  * A zero timeval for sending async RPC messages.
167  */
168 struct timeval nlm_zero_tv = { 0, 0 };
169 
170 /*
171  * The local NSM state number
172  */
173 int nlm_nsm_state;
174 
175 
176 /*
177  * A lock to protect the host list and waiting lock list.
178  */
179 static struct mtx nlm_global_lock;
180 
181 /*
182  * Locks:
183  * (l)		locked by nh_lock
184  * (s)		only accessed via server RPC which is single threaded
185  * (g)		locked by nlm_global_lock
186  * (c)		const until freeing
187  * (a)		modified using atomic ops
188  */
189 
190 /*
191  * A pending client-side lock request, stored on the nlm_waiting_locks
192  * list.
193  */
194 struct nlm_waiting_lock {
195 	TAILQ_ENTRY(nlm_waiting_lock) nw_link; /* (g) */
196 	bool_t		nw_waiting;	       /* (g) */
197 	nlm4_lock	nw_lock;	       /* (c) */
198 	union nfsfh	nw_fh;		       /* (c) */
199 	struct vnode	*nw_vp;		       /* (c) */
200 };
201 TAILQ_HEAD(nlm_waiting_lock_list, nlm_waiting_lock);
202 
203 struct nlm_waiting_lock_list nlm_waiting_locks; /* (g) */
204 
205 /*
206  * A pending server-side asynchronous lock request, stored on the
207  * nh_pending list of the NLM host.
208  */
209 struct nlm_async_lock {
210 	TAILQ_ENTRY(nlm_async_lock) af_link; /* (l) host's list of locks */
211 	struct task	af_task;	/* (c) async callback details */
212 	void		*af_cookie;	/* (l) lock manager cancel token */
213 	struct vnode	*af_vp;		/* (l) vnode to lock */
214 	struct flock	af_fl;		/* (c) lock details */
215 	struct nlm_host *af_host;	/* (c) host which is locking */
216 	CLIENT		*af_rpc;	/* (c) rpc client to send message */
217 	nlm4_testargs	af_granted;	/* (c) notification details */
218 	time_t		af_expiretime;	/* (c) notification time */
219 };
220 TAILQ_HEAD(nlm_async_lock_list, nlm_async_lock);
221 
222 /*
223  * NLM host.
224  */
225 enum nlm_host_state {
226 	NLM_UNMONITORED,
227 	NLM_MONITORED,
228 	NLM_MONITOR_FAILED,
229 	NLM_RECOVERING
230 };
231 
232 struct nlm_rpc {
233 	CLIENT		*nr_client;    /* (l) RPC client handle */
234 	time_t		nr_create_time; /* (l) when client was created */
235 };
236 
237 struct nlm_host {
238 	struct mtx	nh_lock;
239 	volatile u_int	nh_refs;       /* (a) reference count */
240 	TAILQ_ENTRY(nlm_host) nh_link; /* (g) global list of hosts */
241 	char		nh_caller_name[MAXNAMELEN]; /* (c) printable name of host */
242 	uint32_t	nh_sysid;	 /* (c) our allocaed system ID */
243 	char		nh_sysid_string[10]; /* (c) string rep. of sysid */
244 	struct sockaddr_storage	nh_addr; /* (s) remote address of host */
245 	struct nlm_rpc	nh_srvrpc;	 /* (l) RPC for server replies */
246 	struct nlm_rpc	nh_clntrpc;	 /* (l) RPC for client requests */
247 	rpcvers_t	nh_vers;	 /* (s) NLM version of host */
248 	int		nh_state;	 /* (s) last seen NSM state of host */
249 	enum nlm_host_state nh_monstate; /* (l) local NSM monitoring state */
250 	time_t		nh_idle_timeout; /* (s) Time at which host is idle */
251 	struct sysctl_ctx_list nh_sysctl; /* (c) vfs.nlm.sysid nodes */
252 	uint32_t	nh_grantcookie;  /* (l) grant cookie counter */
253 	struct nlm_async_lock_list nh_pending; /* (l) pending async locks */
254 	struct nlm_async_lock_list nh_granted; /* (l) granted locks */
255 	struct nlm_async_lock_list nh_finished; /* (l) finished async locks */
256 };
257 TAILQ_HEAD(nlm_host_list, nlm_host);
258 
259 static struct nlm_host_list nlm_hosts; /* (g) */
260 static uint32_t nlm_next_sysid = 1;    /* (g) */
261 
262 static void	nlm_host_unmonitor(struct nlm_host *);
263 
264 struct nlm_grantcookie {
265 	uint32_t	ng_sysid;
266 	uint32_t	ng_cookie;
267 };
268 
269 static inline uint32_t
ng_sysid(struct netobj * src)270 ng_sysid(struct netobj *src)
271 {
272 
273 	return ((struct nlm_grantcookie *)src->n_bytes)->ng_sysid;
274 }
275 
276 static inline uint32_t
ng_cookie(struct netobj * src)277 ng_cookie(struct netobj *src)
278 {
279 
280 	return ((struct nlm_grantcookie *)src->n_bytes)->ng_cookie;
281 }
282 
283 /**********************************************************************/
284 
285 /*
286  * Initialise NLM globals.
287  */
288 static void
nlm_init(void * dummy)289 nlm_init(void *dummy)
290 {
291 	int error;
292 
293 	mtx_init(&nlm_global_lock, "nlm_global_lock", NULL, MTX_DEF);
294 	TAILQ_INIT(&nlm_waiting_locks);
295 	TAILQ_INIT(&nlm_hosts);
296 
297 	error = syscall_register(&nlm_syscall_offset, &nlm_syscall_sysent,
298 	    &nlm_syscall_prev_sysent);
299 	if (error)
300 		NLM_ERR("Can't register NLM syscall\n");
301 	else
302 		nlm_syscall_registered = TRUE;
303 }
304 SYSINIT(nlm_init, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_init, NULL);
305 
306 static void
nlm_uninit(void * dummy)307 nlm_uninit(void *dummy)
308 {
309 
310 	if (nlm_syscall_registered)
311 		syscall_deregister(&nlm_syscall_offset,
312 		    &nlm_syscall_prev_sysent);
313 }
314 SYSUNINIT(nlm_uninit, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_uninit, NULL);
315 
316 /*
317  * Create a netobj from an arbitrary source.
318  */
319 void
nlm_make_netobj(struct netobj * dst,caddr_t src,size_t srcsize,struct malloc_type * type)320 nlm_make_netobj(struct netobj *dst, caddr_t src, size_t srcsize,
321     struct malloc_type *type)
322 {
323 
324 	dst->n_len = srcsize;
325 	dst->n_bytes = malloc(srcsize, type, M_WAITOK);
326 	memcpy(dst->n_bytes, src, srcsize);
327 }
328 
329 /*
330  * Copy a struct netobj.
331  */
332 void
nlm_copy_netobj(struct netobj * dst,struct netobj * src,struct malloc_type * type)333 nlm_copy_netobj(struct netobj *dst, struct netobj *src,
334     struct malloc_type *type)
335 {
336 
337 	nlm_make_netobj(dst, src->n_bytes, src->n_len, type);
338 }
339 
340 
341 /*
342  * Create an RPC client handle for the given (address,prog,vers)
343  * triple using UDP.
344  */
345 static CLIENT *
nlm_get_rpc(struct sockaddr * sa,rpcprog_t prog,rpcvers_t vers)346 nlm_get_rpc(struct sockaddr *sa, rpcprog_t prog, rpcvers_t vers)
347 {
348 	char *wchan = "nlmrcv";
349 	const char* protofmly;
350 	struct sockaddr_storage ss;
351 	struct socket *so;
352 	CLIENT *rpcb;
353 	struct timeval timo;
354 	RPCB parms;
355 	char *uaddr;
356 	enum clnt_stat stat = RPC_SUCCESS;
357 	int rpcvers = RPCBVERS4;
358 	bool_t do_tcp = FALSE;
359 	bool_t tryagain = FALSE;
360 	struct portmap mapping;
361 	u_short port = 0;
362 
363 	char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"];
364 
365 	/*
366 	 * First we need to contact the remote RPCBIND service to find
367 	 * the right port.
368 	 */
369 	memcpy(&ss, sa, sa->sa_len);
370 	switch (ss.ss_family) {
371 	case AF_INET:
372 		((struct sockaddr_in *)&ss)->sin_port = htons(111);
373 		protofmly = "inet";
374 		so = nlm_socket;
375 		break;
376 
377 #ifdef INET6
378 	case AF_INET6:
379 		((struct sockaddr_in6 *)&ss)->sin6_port = htons(111);
380 		protofmly = "inet6";
381 		so = nlm_socket6;
382 		break;
383 #endif
384 
385 	default:
386 		/*
387 		 * Unsupported address family - fail.
388 		 */
389 		return (NULL);
390 	}
391 
392 	rpcb = clnt_dg_create(so, (struct sockaddr *)&ss,
393 	    RPCBPROG, rpcvers, 0, 0);
394 	if (!rpcb)
395 		return (NULL);
396 
397 try_tcp:
398 	parms.r_prog = prog;
399 	parms.r_vers = vers;
400 	if (do_tcp)
401 		parms.r_netid = "tcp";
402 	else
403 		parms.r_netid = "udp";
404 	parms.r_addr = "";
405 	parms.r_owner = "";
406 
407 	/*
408 	 * Use the default timeout.
409 	 */
410 	timo.tv_sec = 25;
411 	timo.tv_usec = 0;
412 again:
413 	switch (rpcvers) {
414 	case RPCBVERS4:
415 	case RPCBVERS:
416 		/*
417 		 * Try RPCBIND 4 then 3.
418 		 */
419 		uaddr = NULL;
420 		stat = CLNT_CALL(rpcb, (rpcprog_t) RPCBPROC_GETADDR,
421 		    (xdrproc_t) xdr_rpcb, &parms,
422 		    (xdrproc_t) xdr_wrapstring, &uaddr, timo);
423 		if (stat == RPC_SUCCESS) {
424 			/*
425 			 * We have a reply from the remote RPCBIND - turn it
426 			 * into an appropriate address and make a new client
427 			 * that can talk to the remote NLM.
428 			 *
429 			 * XXX fixup IPv6 scope ID.
430 			 */
431 			struct netbuf *a;
432 			a = __rpc_uaddr2taddr_af(ss.ss_family, uaddr);
433 			if (!a) {
434 				tryagain = TRUE;
435 			} else {
436 				tryagain = FALSE;
437 				memcpy(&ss, a->buf, a->len);
438 				free(a->buf, M_RPC);
439 				free(a, M_RPC);
440 				xdr_free((xdrproc_t) xdr_wrapstring, &uaddr);
441 			}
442 		}
443 		if (tryagain || stat == RPC_PROGVERSMISMATCH) {
444 			if (rpcvers == RPCBVERS4)
445 				rpcvers = RPCBVERS;
446 			else if (rpcvers == RPCBVERS)
447 				rpcvers = PMAPVERS;
448 			CLNT_CONTROL(rpcb, CLSET_VERS, &rpcvers);
449 			goto again;
450 		}
451 		break;
452 	case PMAPVERS:
453 		/*
454 		 * Try portmap.
455 		 */
456 		mapping.pm_prog = parms.r_prog;
457 		mapping.pm_vers = parms.r_vers;
458 		mapping.pm_prot = do_tcp ? IPPROTO_TCP : IPPROTO_UDP;
459 		mapping.pm_port = 0;
460 
461 		stat = CLNT_CALL(rpcb, (rpcprog_t) PMAPPROC_GETPORT,
462 		    (xdrproc_t) xdr_portmap, &mapping,
463 		    (xdrproc_t) xdr_u_short, &port, timo);
464 
465 		if (stat == RPC_SUCCESS) {
466 			switch (ss.ss_family) {
467 			case AF_INET:
468 				((struct sockaddr_in *)&ss)->sin_port =
469 					htons(port);
470 				break;
471 
472 #ifdef INET6
473 			case AF_INET6:
474 				((struct sockaddr_in6 *)&ss)->sin6_port =
475 					htons(port);
476 				break;
477 #endif
478 			}
479 		}
480 		break;
481 	default:
482 		panic("invalid rpcvers %d", rpcvers);
483 	}
484 	/*
485 	 * We may have a positive response from the portmapper, but the NLM
486 	 * service was not found. Make sure we received a valid port.
487 	 */
488 	switch (ss.ss_family) {
489 	case AF_INET:
490 		port = ((struct sockaddr_in *)&ss)->sin_port;
491 		break;
492 #ifdef INET6
493 	case AF_INET6:
494 		port = ((struct sockaddr_in6 *)&ss)->sin6_port;
495 		break;
496 #endif
497 	}
498 	if (stat != RPC_SUCCESS || !port) {
499 		/*
500 		 * If we were able to talk to rpcbind or portmap, but the udp
501 		 * variant wasn't available, ask about tcp.
502 		 *
503 		 * XXX - We could also check for a TCP portmapper, but
504 		 * if the host is running a portmapper at all, we should be able
505 		 * to hail it over UDP.
506 		 */
507 		if (stat == RPC_SUCCESS && !do_tcp) {
508 			do_tcp = TRUE;
509 			goto try_tcp;
510 		}
511 
512 		/* Otherwise, bad news. */
513 		switch (sa->sa_family) {
514 		case AF_INET:
515 			inet_ntop(AF_INET,
516 			    &((const struct sockaddr_in *) sa)->sin_addr,
517 			    tmp, sizeof tmp);
518 			break;
519 #ifdef INET6
520 		case AF_INET6:
521 			inet_ntop(AF_INET6,
522 			    &((const struct sockaddr_in6 *) sa)->sin6_addr,
523 			    tmp, sizeof tmp);
524 			break;
525 #endif
526 		default:
527 			strlcpy(tmp, "<unknown>", sizeof(tmp));
528 		}
529 
530 		NLM_ERR("NLM: failed to contact remote rpcbind, "
531 		    "stat = %d, addr = %s, port = %d\n", (int) stat, tmp, port);
532 		CLNT_DESTROY(rpcb);
533 		return (NULL);
534 	}
535 
536 	if (do_tcp) {
537 		/*
538 		 * Destroy the UDP client we used to speak to rpcbind and
539 		 * recreate as a TCP client.
540 		 */
541 		struct netconfig *nconf = NULL;
542 
543 		CLNT_DESTROY(rpcb);
544 
545 		switch (ss.ss_family) {
546 		case AF_INET:
547 			nconf = getnetconfigent("tcp");
548 			break;
549 #ifdef INET6
550 		case AF_INET6:
551 			nconf = getnetconfigent("tcp6");
552 			break;
553 #endif
554 		}
555 
556 		rpcb = clnt_reconnect_create(nconf, (struct sockaddr *)&ss,
557 		    prog, vers, 0, 0);
558 		CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan);
559 		rpcb->cl_auth = nlm_auth;
560 
561 	} else {
562 		/*
563 		 * Re-use the client we used to speak to rpcbind.
564 		 */
565 		CLNT_CONTROL(rpcb, CLSET_SVC_ADDR, &ss);
566 		CLNT_CONTROL(rpcb, CLSET_PROG, &prog);
567 		CLNT_CONTROL(rpcb, CLSET_VERS, &vers);
568 		CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan);
569 		rpcb->cl_auth = nlm_auth;
570 	}
571 
572 	return (rpcb);
573 }
574 
575 /*
576  * This async callback after when an async lock request has been
577  * granted. We notify the host which initiated the request.
578  */
579 static void
nlm_lock_callback(void * arg,int pending)580 nlm_lock_callback(void *arg, int pending)
581 {
582 	struct nlm_async_lock *af = (struct nlm_async_lock *) arg;
583 	struct rpc_callextra ext;
584 
585 	NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) granted,"
586 	    " cookie %d:%d\n", af, af->af_host->nh_caller_name,
587 	    af->af_host->nh_sysid, ng_sysid(&af->af_granted.cookie),
588 	    ng_cookie(&af->af_granted.cookie));
589 
590 	/*
591 	 * Send the results back to the host.
592 	 *
593 	 * Note: there is a possible race here with nlm_host_notify
594 	 * destroying the RPC client. To avoid problems, the first
595 	 * thing nlm_host_notify does is to cancel pending async lock
596 	 * requests.
597 	 */
598 	memset(&ext, 0, sizeof(ext));
599 	ext.rc_auth = nlm_auth;
600 	if (af->af_host->nh_vers == NLM_VERS4) {
601 		nlm4_granted_msg_4(&af->af_granted,
602 		    NULL, af->af_rpc, &ext, nlm_zero_tv);
603 	} else {
604 		/*
605 		 * Back-convert to legacy protocol
606 		 */
607 		nlm_testargs granted;
608 		granted.cookie = af->af_granted.cookie;
609 		granted.exclusive = af->af_granted.exclusive;
610 		granted.alock.caller_name =
611 			af->af_granted.alock.caller_name;
612 		granted.alock.fh = af->af_granted.alock.fh;
613 		granted.alock.oh = af->af_granted.alock.oh;
614 		granted.alock.svid = af->af_granted.alock.svid;
615 		granted.alock.l_offset =
616 			af->af_granted.alock.l_offset;
617 		granted.alock.l_len =
618 			af->af_granted.alock.l_len;
619 
620 		nlm_granted_msg_1(&granted,
621 		    NULL, af->af_rpc, &ext, nlm_zero_tv);
622 	}
623 
624 	/*
625 	 * Move this entry to the nh_granted list.
626 	 */
627 	af->af_expiretime = time_uptime + NLM_EXPIRE_TIMEOUT;
628 	mtx_lock(&af->af_host->nh_lock);
629 	TAILQ_REMOVE(&af->af_host->nh_pending, af, af_link);
630 	TAILQ_INSERT_TAIL(&af->af_host->nh_granted, af, af_link);
631 	mtx_unlock(&af->af_host->nh_lock);
632 }
633 
634 /*
635  * Free an async lock request. The request must have been removed from
636  * any list.
637  */
638 static void
nlm_free_async_lock(struct nlm_async_lock * af)639 nlm_free_async_lock(struct nlm_async_lock *af)
640 {
641 	/*
642 	 * Free an async lock.
643 	 */
644 	if (af->af_rpc)
645 		CLNT_RELEASE(af->af_rpc);
646 	xdr_free((xdrproc_t) xdr_nlm4_testargs, &af->af_granted);
647 	if (af->af_vp)
648 		vrele(af->af_vp);
649 	free(af, M_NLM);
650 }
651 
652 /*
653  * Cancel our async request - this must be called with
654  * af->nh_host->nh_lock held. This is slightly complicated by a
655  * potential race with our own callback. If we fail to cancel the
656  * lock, it must already have been granted - we make sure our async
657  * task has completed by calling taskqueue_drain in this case.
658  */
659 static int
nlm_cancel_async_lock(struct nlm_async_lock * af)660 nlm_cancel_async_lock(struct nlm_async_lock *af)
661 {
662 	struct nlm_host *host = af->af_host;
663 	int error;
664 
665 	mtx_assert(&host->nh_lock, MA_OWNED);
666 
667 	mtx_unlock(&host->nh_lock);
668 
669 	error = VOP_ADVLOCKASYNC(af->af_vp, NULL, F_CANCEL, &af->af_fl,
670 	    F_REMOTE, NULL, &af->af_cookie);
671 
672 	if (error) {
673 		/*
674 		 * We failed to cancel - make sure our callback has
675 		 * completed before we continue.
676 		 */
677 		taskqueue_drain(taskqueue_thread, &af->af_task);
678 	}
679 
680 	mtx_lock(&host->nh_lock);
681 
682 	if (!error) {
683 		NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) "
684 		    "cancelled\n", af, host->nh_caller_name, host->nh_sysid);
685 
686 		/*
687 		 * Remove from the nh_pending list and free now that
688 		 * we are safe from the callback.
689 		 */
690 		TAILQ_REMOVE(&host->nh_pending, af, af_link);
691 		mtx_unlock(&host->nh_lock);
692 		nlm_free_async_lock(af);
693 		mtx_lock(&host->nh_lock);
694 	}
695 
696 	return (error);
697 }
698 
699 static void
nlm_check_expired_locks(struct nlm_host * host)700 nlm_check_expired_locks(struct nlm_host *host)
701 {
702 	struct nlm_async_lock *af;
703 	time_t uptime = time_uptime;
704 
705 	mtx_lock(&host->nh_lock);
706 	while ((af = TAILQ_FIRST(&host->nh_granted)) != NULL
707 	    && uptime >= af->af_expiretime) {
708 		NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) expired,"
709 		    " cookie %d:%d\n", af, af->af_host->nh_caller_name,
710 		    af->af_host->nh_sysid, ng_sysid(&af->af_granted.cookie),
711 		    ng_cookie(&af->af_granted.cookie));
712 		TAILQ_REMOVE(&host->nh_granted, af, af_link);
713 		mtx_unlock(&host->nh_lock);
714 		nlm_free_async_lock(af);
715 		mtx_lock(&host->nh_lock);
716 	}
717 	while ((af = TAILQ_FIRST(&host->nh_finished)) != NULL) {
718 		TAILQ_REMOVE(&host->nh_finished, af, af_link);
719 		mtx_unlock(&host->nh_lock);
720 		nlm_free_async_lock(af);
721 		mtx_lock(&host->nh_lock);
722 	}
723 	mtx_unlock(&host->nh_lock);
724 }
725 
726 /*
727  * Free resources used by a host. This is called after the reference
728  * count has reached zero so it doesn't need to worry about locks.
729  */
730 static void
nlm_host_destroy(struct nlm_host * host)731 nlm_host_destroy(struct nlm_host *host)
732 {
733 
734 	mtx_lock(&nlm_global_lock);
735 	TAILQ_REMOVE(&nlm_hosts, host, nh_link);
736 	mtx_unlock(&nlm_global_lock);
737 
738 	if (host->nh_srvrpc.nr_client)
739 		CLNT_RELEASE(host->nh_srvrpc.nr_client);
740 	if (host->nh_clntrpc.nr_client)
741 		CLNT_RELEASE(host->nh_clntrpc.nr_client);
742 	mtx_destroy(&host->nh_lock);
743 	sysctl_ctx_free(&host->nh_sysctl);
744 	free(host, M_NLM);
745 }
746 
747 /*
748  * Thread start callback for client lock recovery
749  */
750 static void
nlm_client_recovery_start(void * arg)751 nlm_client_recovery_start(void *arg)
752 {
753 	struct nlm_host *host = (struct nlm_host *) arg;
754 
755 	NLM_DEBUG(1, "NLM: client lock recovery for %s started\n",
756 	    host->nh_caller_name);
757 
758 	nlm_client_recovery(host);
759 
760 	NLM_DEBUG(1, "NLM: client lock recovery for %s completed\n",
761 	    host->nh_caller_name);
762 
763 	host->nh_monstate = NLM_MONITORED;
764 	nlm_host_release(host);
765 
766 	kthread_exit();
767 }
768 
769 /*
770  * This is called when we receive a host state change notification. We
771  * unlock any active locks owned by the host. When rpc.lockd is
772  * shutting down, this function is called with newstate set to zero
773  * which allows us to cancel any pending async locks and clear the
774  * locking state.
775  */
776 static void
nlm_host_notify(struct nlm_host * host,int newstate)777 nlm_host_notify(struct nlm_host *host, int newstate)
778 {
779 	struct nlm_async_lock *af;
780 
781 	if (newstate) {
782 		NLM_DEBUG(1, "NLM: host %s (sysid %d) rebooted, new "
783 		    "state is %d\n", host->nh_caller_name,
784 		    host->nh_sysid, newstate);
785 	}
786 
787 	/*
788 	 * Cancel any pending async locks for this host.
789 	 */
790 	mtx_lock(&host->nh_lock);
791 	while ((af = TAILQ_FIRST(&host->nh_pending)) != NULL) {
792 		/*
793 		 * nlm_cancel_async_lock will remove the entry from
794 		 * nh_pending and free it.
795 		 */
796 		nlm_cancel_async_lock(af);
797 	}
798 	mtx_unlock(&host->nh_lock);
799 	nlm_check_expired_locks(host);
800 
801 	/*
802 	 * The host just rebooted - trash its locks.
803 	 */
804 	lf_clearremotesys(host->nh_sysid);
805 	host->nh_state = newstate;
806 
807 	/*
808 	 * If we have any remote locks for this host (i.e. it
809 	 * represents a remote NFS server that our local NFS client
810 	 * has locks for), start a recovery thread.
811 	 */
812 	if (newstate != 0
813 	    && host->nh_monstate != NLM_RECOVERING
814 	    && lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid) > 0) {
815 		struct thread *td;
816 		host->nh_monstate = NLM_RECOVERING;
817 		refcount_acquire(&host->nh_refs);
818 		kthread_add(nlm_client_recovery_start, host, curproc, &td, 0, 0,
819 		    "NFS lock recovery for %s", host->nh_caller_name);
820 	}
821 }
822 
823 /*
824  * Sysctl handler to count the number of locks for a sysid.
825  */
826 static int
nlm_host_lock_count_sysctl(SYSCTL_HANDLER_ARGS)827 nlm_host_lock_count_sysctl(SYSCTL_HANDLER_ARGS)
828 {
829 	struct nlm_host *host;
830 	int count;
831 
832 	host = oidp->oid_arg1;
833 	count = lf_countlocks(host->nh_sysid);
834 	return sysctl_handle_int(oidp, &count, 0, req);
835 }
836 
837 /*
838  * Sysctl handler to count the number of client locks for a sysid.
839  */
840 static int
nlm_host_client_lock_count_sysctl(SYSCTL_HANDLER_ARGS)841 nlm_host_client_lock_count_sysctl(SYSCTL_HANDLER_ARGS)
842 {
843 	struct nlm_host *host;
844 	int count;
845 
846 	host = oidp->oid_arg1;
847 	count = lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid);
848 	return sysctl_handle_int(oidp, &count, 0, req);
849 }
850 
851 /*
852  * Create a new NLM host.
853  */
854 static struct nlm_host *
nlm_create_host(const char * caller_name)855 nlm_create_host(const char* caller_name)
856 {
857 	struct nlm_host *host;
858 	struct sysctl_oid *oid;
859 
860 	mtx_assert(&nlm_global_lock, MA_OWNED);
861 
862 	NLM_DEBUG(1, "NLM: new host %s (sysid %d)\n",
863 	    caller_name, nlm_next_sysid);
864 	host = malloc(sizeof(struct nlm_host), M_NLM, M_NOWAIT|M_ZERO);
865 	if (!host)
866 		return (NULL);
867 	mtx_init(&host->nh_lock, "nh_lock", NULL, MTX_DEF);
868 	host->nh_refs = 1;
869 	strlcpy(host->nh_caller_name, caller_name, MAXNAMELEN);
870 	host->nh_sysid = nlm_next_sysid++;
871 	snprintf(host->nh_sysid_string, sizeof(host->nh_sysid_string),
872 		"%d", host->nh_sysid);
873 	host->nh_vers = 0;
874 	host->nh_state = 0;
875 	host->nh_monstate = NLM_UNMONITORED;
876 	host->nh_grantcookie = 1;
877 	TAILQ_INIT(&host->nh_pending);
878 	TAILQ_INIT(&host->nh_granted);
879 	TAILQ_INIT(&host->nh_finished);
880 	TAILQ_INSERT_TAIL(&nlm_hosts, host, nh_link);
881 
882 	mtx_unlock(&nlm_global_lock);
883 
884 	sysctl_ctx_init(&host->nh_sysctl);
885 	oid = SYSCTL_ADD_NODE(&host->nh_sysctl,
886 	    SYSCTL_STATIC_CHILDREN(_vfs_nlm_sysid),
887 	    OID_AUTO, host->nh_sysid_string, CTLFLAG_RD, NULL, "");
888 	SYSCTL_ADD_STRING(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
889 	    "hostname", CTLFLAG_RD, host->nh_caller_name, 0, "");
890 	SYSCTL_ADD_UINT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
891 	    "version", CTLFLAG_RD, &host->nh_vers, 0, "");
892 	SYSCTL_ADD_UINT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
893 	    "monitored", CTLFLAG_RD, &host->nh_monstate, 0, "");
894 	SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
895 	    "lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0,
896 	    nlm_host_lock_count_sysctl, "I", "");
897 	SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
898 	    "client_lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0,
899 	    nlm_host_client_lock_count_sysctl, "I", "");
900 
901 	mtx_lock(&nlm_global_lock);
902 
903 	return (host);
904 }
905 
906 /*
907  * Acquire the next sysid for remote locks not handled by the NLM.
908  */
909 uint32_t
nlm_acquire_next_sysid(void)910 nlm_acquire_next_sysid(void)
911 {
912 	uint32_t next_sysid;
913 
914 	mtx_lock(&nlm_global_lock);
915 	next_sysid = nlm_next_sysid++;
916 	mtx_unlock(&nlm_global_lock);
917 	return (next_sysid);
918 }
919 
920 /*
921  * Return non-zero if the address parts of the two sockaddrs are the
922  * same.
923  */
924 static int
nlm_compare_addr(const struct sockaddr * a,const struct sockaddr * b)925 nlm_compare_addr(const struct sockaddr *a, const struct sockaddr *b)
926 {
927 	const struct sockaddr_in *a4, *b4;
928 #ifdef INET6
929 	const struct sockaddr_in6 *a6, *b6;
930 #endif
931 
932 	if (a->sa_family != b->sa_family)
933 		return (FALSE);
934 
935 	switch (a->sa_family) {
936 	case AF_INET:
937 		a4 = (const struct sockaddr_in *) a;
938 		b4 = (const struct sockaddr_in *) b;
939 		return !memcmp(&a4->sin_addr, &b4->sin_addr,
940 		    sizeof(a4->sin_addr));
941 #ifdef INET6
942 	case AF_INET6:
943 		a6 = (const struct sockaddr_in6 *) a;
944 		b6 = (const struct sockaddr_in6 *) b;
945 		return !memcmp(&a6->sin6_addr, &b6->sin6_addr,
946 		    sizeof(a6->sin6_addr));
947 #endif
948 	}
949 
950 	return (0);
951 }
952 
953 /*
954  * Check for idle hosts and stop monitoring them. We could also free
955  * the host structure here, possibly after a larger timeout but that
956  * would require some care to avoid races with
957  * e.g. nlm_host_lock_count_sysctl.
958  */
959 static void
nlm_check_idle(void)960 nlm_check_idle(void)
961 {
962 	struct nlm_host *host;
963 
964 	mtx_assert(&nlm_global_lock, MA_OWNED);
965 
966 	if (time_uptime <= nlm_next_idle_check)
967 		return;
968 
969 	nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD;
970 
971 	TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
972 		if (host->nh_monstate == NLM_MONITORED
973 		    && time_uptime > host->nh_idle_timeout) {
974 			mtx_unlock(&nlm_global_lock);
975 			if (lf_countlocks(host->nh_sysid) > 0
976 			    || lf_countlocks(NLM_SYSID_CLIENT
977 				+ host->nh_sysid)) {
978 				host->nh_idle_timeout =
979 					time_uptime + NLM_IDLE_TIMEOUT;
980 				mtx_lock(&nlm_global_lock);
981 				continue;
982 			}
983 			nlm_host_unmonitor(host);
984 			mtx_lock(&nlm_global_lock);
985 		}
986 	}
987 }
988 
989 /*
990  * Search for an existing NLM host that matches the given name
991  * (typically the caller_name element of an nlm4_lock).  If none is
992  * found, create a new host. If 'addr' is non-NULL, record the remote
993  * address of the host so that we can call it back for async
994  * responses. If 'vers' is greater than zero then record the NLM
995  * program version to use to communicate with this client.
996  */
997 struct nlm_host *
nlm_find_host_by_name(const char * name,const struct sockaddr * addr,rpcvers_t vers)998 nlm_find_host_by_name(const char *name, const struct sockaddr *addr,
999     rpcvers_t vers)
1000 {
1001 	struct nlm_host *host;
1002 
1003 	mtx_lock(&nlm_global_lock);
1004 
1005 	/*
1006 	 * The remote host is determined by caller_name.
1007 	 */
1008 	TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
1009 		if (!strcmp(host->nh_caller_name, name))
1010 			break;
1011 	}
1012 
1013 	if (!host) {
1014 		host = nlm_create_host(name);
1015 		if (!host) {
1016 			mtx_unlock(&nlm_global_lock);
1017 			return (NULL);
1018 		}
1019 	}
1020 	refcount_acquire(&host->nh_refs);
1021 
1022 	host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT;
1023 
1024 	/*
1025 	 * If we have an address for the host, record it so that we
1026 	 * can send async replies etc.
1027 	 */
1028 	if (addr) {
1029 
1030 		KASSERT(addr->sa_len < sizeof(struct sockaddr_storage),
1031 		    ("Strange remote transport address length"));
1032 
1033 		/*
1034 		 * If we have seen an address before and we currently
1035 		 * have an RPC client handle, make sure the address is
1036 		 * the same, otherwise discard the client handle.
1037 		 */
1038 		if (host->nh_addr.ss_len && host->nh_srvrpc.nr_client) {
1039 			if (!nlm_compare_addr(
1040 				    (struct sockaddr *) &host->nh_addr,
1041 				    addr)
1042 			    || host->nh_vers != vers) {
1043 				CLIENT *client;
1044 				mtx_lock(&host->nh_lock);
1045 				client = host->nh_srvrpc.nr_client;
1046 				host->nh_srvrpc.nr_client = NULL;
1047 				mtx_unlock(&host->nh_lock);
1048 				if (client) {
1049 					CLNT_RELEASE(client);
1050 				}
1051 			}
1052 		}
1053 		memcpy(&host->nh_addr, addr, addr->sa_len);
1054 		host->nh_vers = vers;
1055 	}
1056 
1057 	nlm_check_idle();
1058 
1059 	mtx_unlock(&nlm_global_lock);
1060 
1061 	return (host);
1062 }
1063 
1064 /*
1065  * Search for an existing NLM host that matches the given remote
1066  * address. If none is found, create a new host with the requested
1067  * address and remember 'vers' as the NLM protocol version to use for
1068  * that host.
1069  */
1070 struct nlm_host *
nlm_find_host_by_addr(const struct sockaddr * addr,int vers)1071 nlm_find_host_by_addr(const struct sockaddr *addr, int vers)
1072 {
1073 	/*
1074 	 * Fake up a name using inet_ntop. This buffer is
1075 	 * large enough for an IPv6 address.
1076 	 */
1077 	char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"];
1078 	struct nlm_host *host;
1079 
1080 	switch (addr->sa_family) {
1081 	case AF_INET:
1082 		inet_ntop(AF_INET,
1083 		    &((const struct sockaddr_in *) addr)->sin_addr,
1084 		    tmp, sizeof tmp);
1085 		break;
1086 #ifdef INET6
1087 	case AF_INET6:
1088 		inet_ntop(AF_INET6,
1089 		    &((const struct sockaddr_in6 *) addr)->sin6_addr,
1090 		    tmp, sizeof tmp);
1091 		break;
1092 #endif
1093 	default:
1094 		strlcpy(tmp, "<unknown>", sizeof(tmp));
1095 	}
1096 
1097 
1098 	mtx_lock(&nlm_global_lock);
1099 
1100 	/*
1101 	 * The remote host is determined by caller_name.
1102 	 */
1103 	TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
1104 		if (nlm_compare_addr(addr,
1105 			(const struct sockaddr *) &host->nh_addr))
1106 			break;
1107 	}
1108 
1109 	if (!host) {
1110 		host = nlm_create_host(tmp);
1111 		if (!host) {
1112 			mtx_unlock(&nlm_global_lock);
1113 			return (NULL);
1114 		}
1115 		memcpy(&host->nh_addr, addr, addr->sa_len);
1116 		host->nh_vers = vers;
1117 	}
1118 	refcount_acquire(&host->nh_refs);
1119 
1120 	host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT;
1121 
1122 	nlm_check_idle();
1123 
1124 	mtx_unlock(&nlm_global_lock);
1125 
1126 	return (host);
1127 }
1128 
1129 /*
1130  * Find the NLM host that matches the value of 'sysid'. If none
1131  * exists, return NULL.
1132  */
1133 static struct nlm_host *
nlm_find_host_by_sysid(int sysid)1134 nlm_find_host_by_sysid(int sysid)
1135 {
1136 	struct nlm_host *host;
1137 
1138 	TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
1139 		if (host->nh_sysid == sysid) {
1140 			refcount_acquire(&host->nh_refs);
1141 			return (host);
1142 		}
1143 	}
1144 
1145 	return (NULL);
1146 }
1147 
nlm_host_release(struct nlm_host * host)1148 void nlm_host_release(struct nlm_host *host)
1149 {
1150 	if (refcount_release(&host->nh_refs)) {
1151 		/*
1152 		 * Free the host
1153 		 */
1154 		nlm_host_destroy(host);
1155 	}
1156 }
1157 
1158 static void
nlm_host_monfail(const struct nlm_host * host,bool action)1159 nlm_host_monfail(const struct nlm_host *host, bool action)
1160 {
1161 	const char *monitor = (action ? "monitor" : "unmonitor");
1162 
1163 	char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"];
1164 	const struct sockaddr *addr = (const struct sockaddr *)&(host->nh_addr);
1165 
1166 	switch (addr->sa_family) {
1167 	case AF_INET:
1168 		inet_ntop(AF_INET,
1169 		    &((const struct sockaddr_in *) addr)->sin_addr,
1170 		    tmp, sizeof tmp);
1171 		break;
1172 #ifdef INET6
1173 	case AF_INET6:
1174 		inet_ntop(AF_INET6,
1175 		    &((const struct sockaddr_in6 *) addr)->sin6_addr,
1176 		    tmp, sizeof tmp);
1177 		break;
1178 #endif
1179 	default:
1180 		strlcpy(tmp, "<unknown>", sizeof(tmp));
1181 	}
1182 
1183 	NLM_ERR("Local NSM refuses to %s %s (%s)\n",
1184 	    monitor, host->nh_caller_name, tmp);
1185 }
1186 
1187 /*
1188  * Unregister this NLM host with the local NSM due to idleness.
1189  */
1190 static void
nlm_host_unmonitor(struct nlm_host * host)1191 nlm_host_unmonitor(struct nlm_host *host)
1192 {
1193 	mon_id smmonid;
1194 	sm_stat_res smstat;
1195 	struct timeval timo;
1196 	enum clnt_stat stat;
1197 
1198 	NLM_DEBUG(1, "NLM: unmonitoring %s (sysid %d)\n",
1199 	    host->nh_caller_name, host->nh_sysid);
1200 
1201 	/*
1202 	 * We put our assigned system ID value in the priv field to
1203 	 * make it simpler to find the host if we are notified of a
1204 	 * host restart.
1205 	 */
1206 	smmonid.mon_name = host->nh_caller_name;
1207 	smmonid.my_id.my_name = "localhost";
1208 	smmonid.my_id.my_prog = NLM_PROG;
1209 	smmonid.my_id.my_vers = NLM_SM;
1210 	smmonid.my_id.my_proc = NLM_SM_NOTIFY;
1211 
1212 	timo.tv_sec = 25;
1213 	timo.tv_usec = 0;
1214 	stat = CLNT_CALL(nlm_nsm, SM_UNMON,
1215 	    (xdrproc_t) xdr_mon, &smmonid,
1216 	    (xdrproc_t) xdr_sm_stat, &smstat, timo);
1217 
1218 	if (stat != RPC_SUCCESS) {
1219 		NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat);
1220 		return;
1221 	}
1222 	if (smstat.res_stat == stat_fail) {
1223 		nlm_host_monfail(host, false);
1224 		return;
1225 	}
1226 
1227 	host->nh_monstate = NLM_UNMONITORED;
1228 }
1229 
1230 /*
1231  * Register this NLM host with the local NSM so that we can be
1232  * notified if it reboots.
1233  */
1234 void
nlm_host_monitor(struct nlm_host * host,int state)1235 nlm_host_monitor(struct nlm_host *host, int state)
1236 {
1237 	mon smmon;
1238 	sm_stat_res smstat;
1239 	struct timeval timo;
1240 	enum clnt_stat stat;
1241 
1242 	if (state && !host->nh_state) {
1243 		/*
1244 		 * This is the first time we have seen an NSM state
1245 		 * value for this host. We record it here to help
1246 		 * detect host reboots.
1247 		 */
1248 		host->nh_state = state;
1249 		NLM_DEBUG(1, "NLM: host %s (sysid %d) has NSM state %d\n",
1250 		    host->nh_caller_name, host->nh_sysid, state);
1251 	}
1252 
1253 	mtx_lock(&host->nh_lock);
1254 	if (host->nh_monstate != NLM_UNMONITORED) {
1255 		mtx_unlock(&host->nh_lock);
1256 		return;
1257 	}
1258 	host->nh_monstate = NLM_MONITORED;
1259 	mtx_unlock(&host->nh_lock);
1260 
1261 	NLM_DEBUG(1, "NLM: monitoring %s (sysid %d)\n",
1262 	    host->nh_caller_name, host->nh_sysid);
1263 
1264 	/*
1265 	 * We put our assigned system ID value in the priv field to
1266 	 * make it simpler to find the host if we are notified of a
1267 	 * host restart.
1268 	 */
1269 	smmon.mon_id.mon_name = host->nh_caller_name;
1270 	smmon.mon_id.my_id.my_name = "localhost";
1271 	smmon.mon_id.my_id.my_prog = NLM_PROG;
1272 	smmon.mon_id.my_id.my_vers = NLM_SM;
1273 	smmon.mon_id.my_id.my_proc = NLM_SM_NOTIFY;
1274 	memcpy(smmon.priv, &host->nh_sysid, sizeof(host->nh_sysid));
1275 
1276 	timo.tv_sec = 25;
1277 	timo.tv_usec = 0;
1278 	stat = CLNT_CALL(nlm_nsm, SM_MON,
1279 	    (xdrproc_t) xdr_mon, &smmon,
1280 	    (xdrproc_t) xdr_sm_stat, &smstat, timo);
1281 
1282 	if (stat != RPC_SUCCESS) {
1283 		NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat);
1284 		return;
1285 	}
1286 	if (smstat.res_stat == stat_fail) {
1287 		nlm_host_monfail(host, true);
1288 		mtx_lock(&host->nh_lock);
1289 		host->nh_monstate = NLM_MONITOR_FAILED;
1290 		mtx_unlock(&host->nh_lock);
1291 		return;
1292 	}
1293 
1294 	host->nh_monstate = NLM_MONITORED;
1295 }
1296 
1297 /*
1298  * Return an RPC client handle that can be used to talk to the NLM
1299  * running on the given host.
1300  */
1301 CLIENT *
nlm_host_get_rpc(struct nlm_host * host,bool_t isserver)1302 nlm_host_get_rpc(struct nlm_host *host, bool_t isserver)
1303 {
1304 	struct nlm_rpc *rpc;
1305 	CLIENT *client;
1306 
1307 	mtx_lock(&host->nh_lock);
1308 
1309 	if (isserver)
1310 		rpc = &host->nh_srvrpc;
1311 	else
1312 		rpc = &host->nh_clntrpc;
1313 
1314 	/*
1315 	 * We can't hold onto RPC handles for too long - the async
1316 	 * call/reply protocol used by some NLM clients makes it hard
1317 	 * to tell when they change port numbers (e.g. after a
1318 	 * reboot). Note that if a client reboots while it isn't
1319 	 * holding any locks, it won't bother to notify us. We
1320 	 * expire the RPC handles after two minutes.
1321 	 */
1322 	if (rpc->nr_client && time_uptime > rpc->nr_create_time + 2*60) {
1323 		client = rpc->nr_client;
1324 		rpc->nr_client = NULL;
1325 		mtx_unlock(&host->nh_lock);
1326 		CLNT_RELEASE(client);
1327 		mtx_lock(&host->nh_lock);
1328 	}
1329 
1330 	if (!rpc->nr_client) {
1331 		mtx_unlock(&host->nh_lock);
1332 		client = nlm_get_rpc((struct sockaddr *)&host->nh_addr,
1333 		    NLM_PROG, host->nh_vers);
1334 		mtx_lock(&host->nh_lock);
1335 
1336 		if (client) {
1337 			if (rpc->nr_client) {
1338 				mtx_unlock(&host->nh_lock);
1339 				CLNT_DESTROY(client);
1340 				mtx_lock(&host->nh_lock);
1341 			} else {
1342 				rpc->nr_client = client;
1343 				rpc->nr_create_time = time_uptime;
1344 			}
1345 		}
1346 	}
1347 
1348 	client = rpc->nr_client;
1349 	if (client)
1350 		CLNT_ACQUIRE(client);
1351 	mtx_unlock(&host->nh_lock);
1352 
1353 	return (client);
1354 
1355 }
1356 
nlm_host_get_sysid(struct nlm_host * host)1357 int nlm_host_get_sysid(struct nlm_host *host)
1358 {
1359 
1360 	return (host->nh_sysid);
1361 }
1362 
1363 int
nlm_host_get_state(struct nlm_host * host)1364 nlm_host_get_state(struct nlm_host *host)
1365 {
1366 
1367 	return (host->nh_state);
1368 }
1369 
1370 void *
nlm_register_wait_lock(struct nlm4_lock * lock,struct vnode * vp)1371 nlm_register_wait_lock(struct nlm4_lock *lock, struct vnode *vp)
1372 {
1373 	struct nlm_waiting_lock *nw;
1374 
1375 	nw = malloc(sizeof(struct nlm_waiting_lock), M_NLM, M_WAITOK);
1376 	nw->nw_lock = *lock;
1377 	memcpy(&nw->nw_fh.fh_bytes, nw->nw_lock.fh.n_bytes,
1378 	    nw->nw_lock.fh.n_len);
1379 	nw->nw_lock.fh.n_bytes = nw->nw_fh.fh_bytes;
1380 	nw->nw_waiting = TRUE;
1381 	nw->nw_vp = vp;
1382 	mtx_lock(&nlm_global_lock);
1383 	TAILQ_INSERT_TAIL(&nlm_waiting_locks, nw, nw_link);
1384 	mtx_unlock(&nlm_global_lock);
1385 
1386 	return nw;
1387 }
1388 
1389 void
nlm_deregister_wait_lock(void * handle)1390 nlm_deregister_wait_lock(void *handle)
1391 {
1392 	struct nlm_waiting_lock *nw = handle;
1393 
1394 	mtx_lock(&nlm_global_lock);
1395 	TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link);
1396 	mtx_unlock(&nlm_global_lock);
1397 
1398 	free(nw, M_NLM);
1399 }
1400 
1401 int
nlm_wait_lock(void * handle,int timo)1402 nlm_wait_lock(void *handle, int timo)
1403 {
1404 	struct nlm_waiting_lock *nw = handle;
1405 	int error;
1406 
1407 	/*
1408 	 * If the granted message arrived before we got here,
1409 	 * nw->nw_waiting will be FALSE - in that case, don't sleep.
1410 	 */
1411 	mtx_lock(&nlm_global_lock);
1412 	error = 0;
1413 	if (nw->nw_waiting)
1414 		error = msleep(nw, &nlm_global_lock, PCATCH, "nlmlock", timo);
1415 	TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link);
1416 	if (error) {
1417 		/*
1418 		 * The granted message may arrive after the
1419 		 * interrupt/timeout but before we manage to lock the
1420 		 * mutex. Detect this by examining nw_lock.
1421 		 */
1422 		if (!nw->nw_waiting)
1423 			error = 0;
1424 	} else {
1425 		/*
1426 		 * If nlm_cancel_wait is called, then error will be
1427 		 * zero but nw_waiting will still be TRUE. We
1428 		 * translate this into EINTR.
1429 		 */
1430 		if (nw->nw_waiting)
1431 			error = EINTR;
1432 	}
1433 	mtx_unlock(&nlm_global_lock);
1434 
1435 	free(nw, M_NLM);
1436 
1437 	return (error);
1438 }
1439 
1440 void
nlm_cancel_wait(struct vnode * vp)1441 nlm_cancel_wait(struct vnode *vp)
1442 {
1443 	struct nlm_waiting_lock *nw;
1444 
1445 	mtx_lock(&nlm_global_lock);
1446 	TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
1447 		if (nw->nw_vp == vp) {
1448 			wakeup(nw);
1449 		}
1450 	}
1451 	mtx_unlock(&nlm_global_lock);
1452 }
1453 
1454 
1455 /**********************************************************************/
1456 
1457 /*
1458  * Syscall interface with userland.
1459  */
1460 
1461 extern void nlm_prog_0(struct svc_req *rqstp, SVCXPRT *transp);
1462 extern void nlm_prog_1(struct svc_req *rqstp, SVCXPRT *transp);
1463 extern void nlm_prog_3(struct svc_req *rqstp, SVCXPRT *transp);
1464 extern void nlm_prog_4(struct svc_req *rqstp, SVCXPRT *transp);
1465 
1466 static int
nlm_register_services(SVCPOOL * pool,int addr_count,char ** addrs)1467 nlm_register_services(SVCPOOL *pool, int addr_count, char **addrs)
1468 {
1469 	static rpcvers_t versions[] = {
1470 		NLM_SM, NLM_VERS, NLM_VERSX, NLM_VERS4
1471 	};
1472 	static void (*dispatchers[])(struct svc_req *, SVCXPRT *) = {
1473 		nlm_prog_0, nlm_prog_1, nlm_prog_3, nlm_prog_4
1474 	};
1475 	static const int version_count = sizeof(versions) / sizeof(versions[0]);
1476 
1477 	SVCXPRT **xprts;
1478 	char netid[16];
1479 	char uaddr[128];
1480 	struct netconfig *nconf;
1481 	int i, j, error;
1482 
1483 	if (!addr_count) {
1484 		NLM_ERR("NLM: no service addresses given - can't start server");
1485 		return (EINVAL);
1486 	}
1487 
1488 	xprts = malloc(addr_count * sizeof(SVCXPRT *), M_NLM, M_WAITOK|M_ZERO);
1489 	for (i = 0; i < version_count; i++) {
1490 		for (j = 0; j < addr_count; j++) {
1491 			/*
1492 			 * Create transports for the first version and
1493 			 * then just register everything else to the
1494 			 * same transports.
1495 			 */
1496 			if (i == 0) {
1497 				char *up;
1498 
1499 				error = copyin(&addrs[2*j], &up,
1500 				    sizeof(char*));
1501 				if (error)
1502 					goto out;
1503 				error = copyinstr(up, netid, sizeof(netid),
1504 				    NULL);
1505 				if (error)
1506 					goto out;
1507 				error = copyin(&addrs[2*j+1], &up,
1508 				    sizeof(char*));
1509 				if (error)
1510 					goto out;
1511 				error = copyinstr(up, uaddr, sizeof(uaddr),
1512 				    NULL);
1513 				if (error)
1514 					goto out;
1515 				nconf = getnetconfigent(netid);
1516 				if (!nconf) {
1517 					NLM_ERR("Can't lookup netid %s\n",
1518 					    netid);
1519 					error = EINVAL;
1520 					goto out;
1521 				}
1522 				xprts[j] = svc_tp_create(pool, dispatchers[i],
1523 				    NLM_PROG, versions[i], uaddr, nconf);
1524 				if (!xprts[j]) {
1525 					NLM_ERR("NLM: unable to create "
1526 					    "(NLM_PROG, %d).\n", versions[i]);
1527 					error = EINVAL;
1528 					goto out;
1529 				}
1530 				freenetconfigent(nconf);
1531 			} else {
1532 				nconf = getnetconfigent(xprts[j]->xp_netid);
1533 				rpcb_unset(NLM_PROG, versions[i], nconf);
1534 				if (!svc_reg(xprts[j], NLM_PROG, versions[i],
1535 					dispatchers[i], nconf)) {
1536 					NLM_ERR("NLM: can't register "
1537 					    "(NLM_PROG, %d)\n", versions[i]);
1538 					error = EINVAL;
1539 					goto out;
1540 				}
1541 			}
1542 		}
1543 	}
1544 	error = 0;
1545 out:
1546 	for (j = 0; j < addr_count; j++) {
1547 		if (xprts[j])
1548 			SVC_RELEASE(xprts[j]);
1549 	}
1550 	free(xprts, M_NLM);
1551 	return (error);
1552 }
1553 
1554 /*
1555  * Main server entry point. Contacts the local NSM to get its current
1556  * state and send SM_UNMON_ALL. Registers the NLM services and then
1557  * services requests. Does not return until the server is interrupted
1558  * by a signal.
1559  */
1560 static int
nlm_server_main(int addr_count,char ** addrs)1561 nlm_server_main(int addr_count, char **addrs)
1562 {
1563 	struct thread *td = curthread;
1564 	int error;
1565 	SVCPOOL *pool = NULL;
1566 	struct sockopt opt;
1567 	int portlow;
1568 #ifdef INET6
1569 	struct sockaddr_in6 sin6;
1570 #endif
1571 	struct sockaddr_in sin;
1572 	my_id id;
1573 	sm_stat smstat;
1574 	struct timeval timo;
1575 	enum clnt_stat stat;
1576 	struct nlm_host *host, *nhost;
1577 	struct nlm_waiting_lock *nw;
1578 	vop_advlock_t *old_nfs_advlock;
1579 	vop_reclaim_t *old_nfs_reclaim;
1580 
1581 	if (nlm_is_running != 0) {
1582 		NLM_ERR("NLM: can't start server - "
1583 		    "it appears to be running already\n");
1584 		return (EPERM);
1585 	}
1586 
1587 	if (nlm_socket == NULL) {
1588 		memset(&opt, 0, sizeof(opt));
1589 
1590 		error = socreate(AF_INET, &nlm_socket, SOCK_DGRAM, 0,
1591 		    td->td_ucred, td);
1592 		if (error) {
1593 			NLM_ERR("NLM: can't create IPv4 socket - error %d\n",
1594 			    error);
1595 			return (error);
1596 		}
1597 		opt.sopt_dir = SOPT_SET;
1598 		opt.sopt_level = IPPROTO_IP;
1599 		opt.sopt_name = IP_PORTRANGE;
1600 		portlow = IP_PORTRANGE_LOW;
1601 		opt.sopt_val = &portlow;
1602 		opt.sopt_valsize = sizeof(portlow);
1603 		sosetopt(nlm_socket, &opt);
1604 
1605 #ifdef INET6
1606 		nlm_socket6 = NULL;
1607 		error = socreate(AF_INET6, &nlm_socket6, SOCK_DGRAM, 0,
1608 		    td->td_ucred, td);
1609 		if (error) {
1610 			NLM_ERR("NLM: can't create IPv6 socket - error %d\n",
1611 			    error);
1612 			soclose(nlm_socket);
1613 			nlm_socket = NULL;
1614 			return (error);
1615 		}
1616 		opt.sopt_dir = SOPT_SET;
1617 		opt.sopt_level = IPPROTO_IPV6;
1618 		opt.sopt_name = IPV6_PORTRANGE;
1619 		portlow = IPV6_PORTRANGE_LOW;
1620 		opt.sopt_val = &portlow;
1621 		opt.sopt_valsize = sizeof(portlow);
1622 		sosetopt(nlm_socket6, &opt);
1623 #endif
1624 	}
1625 
1626 	nlm_auth = authunix_create(curthread->td_ucred);
1627 
1628 #ifdef INET6
1629 	memset(&sin6, 0, sizeof(sin6));
1630 	sin6.sin6_len = sizeof(sin6);
1631 	sin6.sin6_family = AF_INET6;
1632 	sin6.sin6_addr = in6addr_loopback;
1633 	nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin6, SM_PROG, SM_VERS);
1634 	if (!nlm_nsm) {
1635 #endif
1636 		memset(&sin, 0, sizeof(sin));
1637 		sin.sin_len = sizeof(sin);
1638 		sin.sin_family = AF_INET;
1639 		sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1640 		nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin, SM_PROG,
1641 		    SM_VERS);
1642 #ifdef INET6
1643 	}
1644 #endif
1645 
1646 	if (!nlm_nsm) {
1647 		NLM_ERR("Can't start NLM - unable to contact NSM\n");
1648 		error = EINVAL;
1649 		goto out;
1650 	}
1651 
1652 	pool = svcpool_create("NLM", NULL);
1653 
1654 	error = nlm_register_services(pool, addr_count, addrs);
1655 	if (error)
1656 		goto out;
1657 
1658 	memset(&id, 0, sizeof(id));
1659 	id.my_name = "NFS NLM";
1660 
1661 	timo.tv_sec = 25;
1662 	timo.tv_usec = 0;
1663 	stat = CLNT_CALL(nlm_nsm, SM_UNMON_ALL,
1664 	    (xdrproc_t) xdr_my_id, &id,
1665 	    (xdrproc_t) xdr_sm_stat, &smstat, timo);
1666 
1667 	if (stat != RPC_SUCCESS) {
1668 		struct rpc_err err;
1669 
1670 		CLNT_GETERR(nlm_nsm, &err);
1671 		NLM_ERR("NLM: unexpected error contacting NSM, "
1672 		    "stat=%d, errno=%d\n", stat, err.re_errno);
1673 		error = EINVAL;
1674 		goto out;
1675 	}
1676 	nlm_is_running = 1;
1677 
1678 	NLM_DEBUG(1, "NLM: local NSM state is %d\n", smstat.state);
1679 	nlm_nsm_state = smstat.state;
1680 
1681 	old_nfs_advlock = nfs_advlock_p;
1682 	nfs_advlock_p = nlm_advlock;
1683 	old_nfs_reclaim = nfs_reclaim_p;
1684 	nfs_reclaim_p = nlm_reclaim;
1685 
1686 	svc_run(pool);
1687 	error = 0;
1688 
1689 	nfs_advlock_p = old_nfs_advlock;
1690 	nfs_reclaim_p = old_nfs_reclaim;
1691 
1692 out:
1693 	nlm_is_running = 0;
1694 	if (pool)
1695 		svcpool_destroy(pool);
1696 
1697 	/*
1698 	 * We are finished communicating with the NSM.
1699 	 */
1700 	if (nlm_nsm) {
1701 		CLNT_RELEASE(nlm_nsm);
1702 		nlm_nsm = NULL;
1703 	}
1704 
1705 	/*
1706 	 * Trash all the existing state so that if the server
1707 	 * restarts, it gets a clean slate. This is complicated by the
1708 	 * possibility that there may be other threads trying to make
1709 	 * client locking requests.
1710 	 *
1711 	 * First we fake a client reboot notification which will
1712 	 * cancel any pending async locks and purge remote lock state
1713 	 * from the local lock manager. We release the reference from
1714 	 * nlm_hosts to the host (which may remove it from the list
1715 	 * and free it). After this phase, the only entries in the
1716 	 * nlm_host list should be from other threads performing
1717 	 * client lock requests.
1718 	 */
1719 	mtx_lock(&nlm_global_lock);
1720 	TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
1721 		wakeup(nw);
1722 	}
1723 	TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) {
1724 		mtx_unlock(&nlm_global_lock);
1725 		nlm_host_notify(host, 0);
1726 		nlm_host_release(host);
1727 		mtx_lock(&nlm_global_lock);
1728 	}
1729 	mtx_unlock(&nlm_global_lock);
1730 
1731 	AUTH_DESTROY(nlm_auth);
1732 
1733 	return (error);
1734 }
1735 
1736 int
sys_nlm_syscall(struct thread * td,struct nlm_syscall_args * uap)1737 sys_nlm_syscall(struct thread *td, struct nlm_syscall_args *uap)
1738 {
1739 	int error;
1740 
1741 #if __FreeBSD_version >= 700000
1742 	error = priv_check(td, PRIV_NFS_LOCKD);
1743 #else
1744 	error = suser(td);
1745 #endif
1746 	if (error)
1747 		return (error);
1748 
1749 	nlm_debug_level = uap->debug_level;
1750 	nlm_grace_threshold = time_uptime + uap->grace_period;
1751 	nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD;
1752 
1753 	return nlm_server_main(uap->addr_count, uap->addrs);
1754 }
1755 
1756 /**********************************************************************/
1757 
1758 /*
1759  * NLM implementation details, called from the RPC stubs.
1760  */
1761 
1762 
1763 void
nlm_sm_notify(struct nlm_sm_status * argp)1764 nlm_sm_notify(struct nlm_sm_status *argp)
1765 {
1766 	uint32_t sysid;
1767 	struct nlm_host *host;
1768 
1769 	NLM_DEBUG(3, "nlm_sm_notify(): mon_name = %s\n", argp->mon_name);
1770 	memcpy(&sysid, &argp->priv, sizeof(sysid));
1771 	host = nlm_find_host_by_sysid(sysid);
1772 	if (host) {
1773 		nlm_host_notify(host, argp->state);
1774 		nlm_host_release(host);
1775 	}
1776 }
1777 
1778 static void
nlm_convert_to_fhandle_t(fhandle_t * fhp,struct netobj * p)1779 nlm_convert_to_fhandle_t(fhandle_t *fhp, struct netobj *p)
1780 {
1781 	memcpy(fhp, p->n_bytes, sizeof(fhandle_t));
1782 }
1783 
1784 struct vfs_state {
1785 	struct mount	*vs_mp;
1786 	struct vnode	*vs_vp;
1787 	int		vs_vnlocked;
1788 };
1789 
1790 static int
nlm_get_vfs_state(struct nlm_host * host,struct svc_req * rqstp,fhandle_t * fhp,struct vfs_state * vs,accmode_t accmode)1791 nlm_get_vfs_state(struct nlm_host *host, struct svc_req *rqstp,
1792     fhandle_t *fhp, struct vfs_state *vs, accmode_t accmode)
1793 {
1794 	int error, exflags;
1795 	struct ucred *cred = NULL, *credanon = NULL;
1796 
1797 	memset(vs, 0, sizeof(*vs));
1798 
1799 	vs->vs_mp = vfs_getvfs(&fhp->fh_fsid);
1800 	if (!vs->vs_mp) {
1801 		return (ESTALE);
1802 	}
1803 
1804 	/* accmode == 0 means don't check, since it is an unlock. */
1805 	if (accmode != 0) {
1806 		error = VFS_CHECKEXP(vs->vs_mp,
1807 		    (struct sockaddr *)&host->nh_addr, &exflags, &credanon,
1808 		    NULL, NULL);
1809 		if (error)
1810 			goto out;
1811 
1812 		if (exflags & MNT_EXRDONLY ||
1813 		    (vs->vs_mp->mnt_flag & MNT_RDONLY)) {
1814 			error = EROFS;
1815 			goto out;
1816 		}
1817 	}
1818 
1819 	error = VFS_FHTOVP(vs->vs_mp, &fhp->fh_fid, LK_EXCLUSIVE, &vs->vs_vp);
1820 	if (error)
1821 		goto out;
1822 	vs->vs_vnlocked = TRUE;
1823 
1824 	if (accmode != 0) {
1825 		if (!svc_getcred(rqstp, &cred, NULL)) {
1826 			error = EINVAL;
1827 			goto out;
1828 		}
1829 		if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
1830 			crfree(cred);
1831 			cred = credanon;
1832 			credanon = NULL;
1833 		}
1834 
1835 		/*
1836 		 * Check cred.
1837 		 */
1838 		error = VOP_ACCESS(vs->vs_vp, accmode, cred, curthread);
1839 		/*
1840 		 * If this failed and accmode != VWRITE, try again with
1841 		 * VWRITE to maintain backwards compatibility with the
1842 		 * old code that always used VWRITE.
1843 		 */
1844 		if (error != 0 && accmode != VWRITE)
1845 			error = VOP_ACCESS(vs->vs_vp, VWRITE, cred, curthread);
1846 		if (error)
1847 			goto out;
1848 	}
1849 
1850 #if __FreeBSD_version < 800011
1851 	VOP_UNLOCK(vs->vs_vp, 0, curthread);
1852 #else
1853 	VOP_UNLOCK(vs->vs_vp, 0);
1854 #endif
1855 	vs->vs_vnlocked = FALSE;
1856 
1857 out:
1858 	if (cred)
1859 		crfree(cred);
1860 	if (credanon)
1861 		crfree(credanon);
1862 
1863 	return (error);
1864 }
1865 
1866 static void
nlm_release_vfs_state(struct vfs_state * vs)1867 nlm_release_vfs_state(struct vfs_state *vs)
1868 {
1869 
1870 	if (vs->vs_vp) {
1871 		if (vs->vs_vnlocked)
1872 			vput(vs->vs_vp);
1873 		else
1874 			vrele(vs->vs_vp);
1875 	}
1876 	if (vs->vs_mp)
1877 		vfs_rel(vs->vs_mp);
1878 }
1879 
1880 static nlm4_stats
nlm_convert_error(int error)1881 nlm_convert_error(int error)
1882 {
1883 
1884 	if (error == ESTALE)
1885 		return nlm4_stale_fh;
1886 	else if (error == EROFS)
1887 		return nlm4_rofs;
1888 	else
1889 		return nlm4_failed;
1890 }
1891 
1892 int
nlm_do_test(nlm4_testargs * argp,nlm4_testres * result,struct svc_req * rqstp,CLIENT ** rpcp)1893 nlm_do_test(nlm4_testargs *argp, nlm4_testres *result, struct svc_req *rqstp,
1894 	CLIENT **rpcp)
1895 {
1896 	fhandle_t fh;
1897 	struct vfs_state vs;
1898 	struct nlm_host *host, *bhost;
1899 	int error, sysid;
1900 	struct flock fl;
1901 	accmode_t accmode;
1902 
1903 	memset(result, 0, sizeof(*result));
1904 	memset(&vs, 0, sizeof(vs));
1905 
1906 	host = nlm_find_host_by_name(argp->alock.caller_name,
1907 	    svc_getrpccaller(rqstp), rqstp->rq_vers);
1908 	if (!host) {
1909 		result->stat.stat = nlm4_denied_nolocks;
1910 		return (ENOMEM);
1911 	}
1912 
1913 	NLM_DEBUG(3, "nlm_do_test(): caller_name = %s (sysid = %d)\n",
1914 	    host->nh_caller_name, host->nh_sysid);
1915 
1916 	nlm_check_expired_locks(host);
1917 	sysid = host->nh_sysid;
1918 
1919 	nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
1920 	nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
1921 
1922 	if (time_uptime < nlm_grace_threshold) {
1923 		result->stat.stat = nlm4_denied_grace_period;
1924 		goto out;
1925 	}
1926 
1927 	accmode = argp->exclusive ? VWRITE : VREAD;
1928 	error = nlm_get_vfs_state(host, rqstp, &fh, &vs, accmode);
1929 	if (error) {
1930 		result->stat.stat = nlm_convert_error(error);
1931 		goto out;
1932 	}
1933 
1934 	fl.l_start = argp->alock.l_offset;
1935 	fl.l_len = argp->alock.l_len;
1936 	fl.l_pid = argp->alock.svid;
1937 	fl.l_sysid = sysid;
1938 	fl.l_whence = SEEK_SET;
1939 	if (argp->exclusive)
1940 		fl.l_type = F_WRLCK;
1941 	else
1942 		fl.l_type = F_RDLCK;
1943 	error = VOP_ADVLOCK(vs.vs_vp, NULL, F_GETLK, &fl, F_REMOTE);
1944 	if (error) {
1945 		result->stat.stat = nlm4_failed;
1946 		goto out;
1947 	}
1948 
1949 	if (fl.l_type == F_UNLCK) {
1950 		result->stat.stat = nlm4_granted;
1951 	} else {
1952 		result->stat.stat = nlm4_denied;
1953 		result->stat.nlm4_testrply_u.holder.exclusive =
1954 			(fl.l_type == F_WRLCK);
1955 		result->stat.nlm4_testrply_u.holder.svid = fl.l_pid;
1956 		bhost = nlm_find_host_by_sysid(fl.l_sysid);
1957 		if (bhost) {
1958 			/*
1959 			 * We don't have any useful way of recording
1960 			 * the value of oh used in the original lock
1961 			 * request. Ideally, the test reply would have
1962 			 * a space for the owning host's name allowing
1963 			 * our caller's NLM to keep track.
1964 			 *
1965 			 * As far as I can see, Solaris uses an eight
1966 			 * byte structure for oh which contains a four
1967 			 * byte pid encoded in local byte order and
1968 			 * the first four bytes of the host
1969 			 * name. Linux uses a variable length string
1970 			 * 'pid@hostname' in ascii but doesn't even
1971 			 * return that in test replies.
1972 			 *
1973 			 * For the moment, return nothing in oh
1974 			 * (already zero'ed above).
1975 			 */
1976 			nlm_host_release(bhost);
1977 		}
1978 		result->stat.nlm4_testrply_u.holder.l_offset = fl.l_start;
1979 		result->stat.nlm4_testrply_u.holder.l_len = fl.l_len;
1980 	}
1981 
1982 out:
1983 	nlm_release_vfs_state(&vs);
1984 	if (rpcp)
1985 		*rpcp = nlm_host_get_rpc(host, TRUE);
1986 	nlm_host_release(host);
1987 	return (0);
1988 }
1989 
1990 int
nlm_do_lock(nlm4_lockargs * argp,nlm4_res * result,struct svc_req * rqstp,bool_t monitor,CLIENT ** rpcp)1991 nlm_do_lock(nlm4_lockargs *argp, nlm4_res *result, struct svc_req *rqstp,
1992     bool_t monitor, CLIENT **rpcp)
1993 {
1994 	fhandle_t fh;
1995 	struct vfs_state vs;
1996 	struct nlm_host *host;
1997 	int error, sysid;
1998 	struct flock fl;
1999 	accmode_t accmode;
2000 
2001 	memset(result, 0, sizeof(*result));
2002 	memset(&vs, 0, sizeof(vs));
2003 
2004 	host = nlm_find_host_by_name(argp->alock.caller_name,
2005 	    svc_getrpccaller(rqstp), rqstp->rq_vers);
2006 	if (!host) {
2007 		result->stat.stat = nlm4_denied_nolocks;
2008 		return (ENOMEM);
2009 	}
2010 
2011 	NLM_DEBUG(3, "nlm_do_lock(): caller_name = %s (sysid = %d)\n",
2012 	    host->nh_caller_name, host->nh_sysid);
2013 
2014 	if (monitor && host->nh_state && argp->state
2015 	    && host->nh_state != argp->state) {
2016 		/*
2017 		 * The host rebooted without telling us. Trash its
2018 		 * locks.
2019 		 */
2020 		nlm_host_notify(host, argp->state);
2021 	}
2022 
2023 	nlm_check_expired_locks(host);
2024 	sysid = host->nh_sysid;
2025 
2026 	nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
2027 	nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2028 
2029 	if (time_uptime < nlm_grace_threshold && !argp->reclaim) {
2030 		result->stat.stat = nlm4_denied_grace_period;
2031 		goto out;
2032 	}
2033 
2034 	accmode = argp->exclusive ? VWRITE : VREAD;
2035 	error = nlm_get_vfs_state(host, rqstp, &fh, &vs, accmode);
2036 	if (error) {
2037 		result->stat.stat = nlm_convert_error(error);
2038 		goto out;
2039 	}
2040 
2041 	fl.l_start = argp->alock.l_offset;
2042 	fl.l_len = argp->alock.l_len;
2043 	fl.l_pid = argp->alock.svid;
2044 	fl.l_sysid = sysid;
2045 	fl.l_whence = SEEK_SET;
2046 	if (argp->exclusive)
2047 		fl.l_type = F_WRLCK;
2048 	else
2049 		fl.l_type = F_RDLCK;
2050 	if (argp->block) {
2051 		struct nlm_async_lock *af;
2052 		CLIENT *client;
2053 		struct nlm_grantcookie cookie;
2054 
2055 		/*
2056 		 * First, make sure we can contact the host's NLM.
2057 		 */
2058 		client = nlm_host_get_rpc(host, TRUE);
2059 		if (!client) {
2060 			result->stat.stat = nlm4_failed;
2061 			goto out;
2062 		}
2063 
2064 		/*
2065 		 * First we need to check and see if there is an
2066 		 * existing blocked lock that matches. This could be a
2067 		 * badly behaved client or an RPC re-send. If we find
2068 		 * one, just return nlm4_blocked.
2069 		 */
2070 		mtx_lock(&host->nh_lock);
2071 		TAILQ_FOREACH(af, &host->nh_pending, af_link) {
2072 			if (af->af_fl.l_start == fl.l_start
2073 			    && af->af_fl.l_len == fl.l_len
2074 			    && af->af_fl.l_pid == fl.l_pid
2075 			    && af->af_fl.l_type == fl.l_type) {
2076 				break;
2077 			}
2078 		}
2079 		if (!af) {
2080 			cookie.ng_sysid = host->nh_sysid;
2081 			cookie.ng_cookie = host->nh_grantcookie++;
2082 		}
2083 		mtx_unlock(&host->nh_lock);
2084 		if (af) {
2085 			CLNT_RELEASE(client);
2086 			result->stat.stat = nlm4_blocked;
2087 			goto out;
2088 		}
2089 
2090 		af = malloc(sizeof(struct nlm_async_lock), M_NLM,
2091 		    M_WAITOK|M_ZERO);
2092 		TASK_INIT(&af->af_task, 0, nlm_lock_callback, af);
2093 		af->af_vp = vs.vs_vp;
2094 		af->af_fl = fl;
2095 		af->af_host = host;
2096 		af->af_rpc = client;
2097 		/*
2098 		 * We use M_RPC here so that we can xdr_free the thing
2099 		 * later.
2100 		 */
2101 		nlm_make_netobj(&af->af_granted.cookie,
2102 		    (caddr_t)&cookie, sizeof(cookie), M_RPC);
2103 		af->af_granted.exclusive = argp->exclusive;
2104 		af->af_granted.alock.caller_name =
2105 			strdup(argp->alock.caller_name, M_RPC);
2106 		nlm_copy_netobj(&af->af_granted.alock.fh,
2107 		    &argp->alock.fh, M_RPC);
2108 		nlm_copy_netobj(&af->af_granted.alock.oh,
2109 		    &argp->alock.oh, M_RPC);
2110 		af->af_granted.alock.svid = argp->alock.svid;
2111 		af->af_granted.alock.l_offset = argp->alock.l_offset;
2112 		af->af_granted.alock.l_len = argp->alock.l_len;
2113 
2114 		/*
2115 		 * Put the entry on the pending list before calling
2116 		 * VOP_ADVLOCKASYNC. We do this in case the lock
2117 		 * request was blocked (returning EINPROGRESS) but
2118 		 * then granted before we manage to run again. The
2119 		 * client may receive the granted message before we
2120 		 * send our blocked reply but thats their problem.
2121 		 */
2122 		mtx_lock(&host->nh_lock);
2123 		TAILQ_INSERT_TAIL(&host->nh_pending, af, af_link);
2124 		mtx_unlock(&host->nh_lock);
2125 
2126 		error = VOP_ADVLOCKASYNC(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE,
2127 		    &af->af_task, &af->af_cookie);
2128 
2129 		/*
2130 		 * If the lock completed synchronously, just free the
2131 		 * tracking structure now.
2132 		 */
2133 		if (error != EINPROGRESS) {
2134 			CLNT_RELEASE(af->af_rpc);
2135 			mtx_lock(&host->nh_lock);
2136 			TAILQ_REMOVE(&host->nh_pending, af, af_link);
2137 			mtx_unlock(&host->nh_lock);
2138 			xdr_free((xdrproc_t) xdr_nlm4_testargs,
2139 			    &af->af_granted);
2140 			free(af, M_NLM);
2141 		} else {
2142 			NLM_DEBUG(2, "NLM: pending async lock %p for %s "
2143 			    "(sysid %d)\n", af, host->nh_caller_name, sysid);
2144 			/*
2145 			 * Don't vrele the vnode just yet - this must
2146 			 * wait until either the async callback
2147 			 * happens or the lock is cancelled.
2148 			 */
2149 			vs.vs_vp = NULL;
2150 		}
2151 	} else {
2152 		error = VOP_ADVLOCK(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE);
2153 	}
2154 
2155 	if (error) {
2156 		if (error == EINPROGRESS) {
2157 			result->stat.stat = nlm4_blocked;
2158 		} else if (error == EDEADLK) {
2159 			result->stat.stat = nlm4_deadlck;
2160 		} else if (error == EAGAIN) {
2161 			result->stat.stat = nlm4_denied;
2162 		} else {
2163 			result->stat.stat = nlm4_failed;
2164 		}
2165 	} else {
2166 		if (monitor)
2167 			nlm_host_monitor(host, argp->state);
2168 		result->stat.stat = nlm4_granted;
2169 	}
2170 
2171 out:
2172 	nlm_release_vfs_state(&vs);
2173 	if (rpcp)
2174 		*rpcp = nlm_host_get_rpc(host, TRUE);
2175 	nlm_host_release(host);
2176 	return (0);
2177 }
2178 
2179 int
nlm_do_cancel(nlm4_cancargs * argp,nlm4_res * result,struct svc_req * rqstp,CLIENT ** rpcp)2180 nlm_do_cancel(nlm4_cancargs *argp, nlm4_res *result, struct svc_req *rqstp,
2181     CLIENT **rpcp)
2182 {
2183 	fhandle_t fh;
2184 	struct vfs_state vs;
2185 	struct nlm_host *host;
2186 	int error, sysid;
2187 	struct flock fl;
2188 	struct nlm_async_lock *af;
2189 
2190 	memset(result, 0, sizeof(*result));
2191 	memset(&vs, 0, sizeof(vs));
2192 
2193 	host = nlm_find_host_by_name(argp->alock.caller_name,
2194 	    svc_getrpccaller(rqstp), rqstp->rq_vers);
2195 	if (!host) {
2196 		result->stat.stat = nlm4_denied_nolocks;
2197 		return (ENOMEM);
2198 	}
2199 
2200 	NLM_DEBUG(3, "nlm_do_cancel(): caller_name = %s (sysid = %d)\n",
2201 	    host->nh_caller_name, host->nh_sysid);
2202 
2203 	nlm_check_expired_locks(host);
2204 	sysid = host->nh_sysid;
2205 
2206 	nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
2207 	nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2208 
2209 	if (time_uptime < nlm_grace_threshold) {
2210 		result->stat.stat = nlm4_denied_grace_period;
2211 		goto out;
2212 	}
2213 
2214 	error = nlm_get_vfs_state(host, rqstp, &fh, &vs, (accmode_t)0);
2215 	if (error) {
2216 		result->stat.stat = nlm_convert_error(error);
2217 		goto out;
2218 	}
2219 
2220 	fl.l_start = argp->alock.l_offset;
2221 	fl.l_len = argp->alock.l_len;
2222 	fl.l_pid = argp->alock.svid;
2223 	fl.l_sysid = sysid;
2224 	fl.l_whence = SEEK_SET;
2225 	if (argp->exclusive)
2226 		fl.l_type = F_WRLCK;
2227 	else
2228 		fl.l_type = F_RDLCK;
2229 
2230 	/*
2231 	 * First we need to try and find the async lock request - if
2232 	 * there isn't one, we give up and return nlm4_denied.
2233 	 */
2234 	mtx_lock(&host->nh_lock);
2235 
2236 	TAILQ_FOREACH(af, &host->nh_pending, af_link) {
2237 		if (af->af_fl.l_start == fl.l_start
2238 		    && af->af_fl.l_len == fl.l_len
2239 		    && af->af_fl.l_pid == fl.l_pid
2240 		    && af->af_fl.l_type == fl.l_type) {
2241 			break;
2242 		}
2243 	}
2244 
2245 	if (!af) {
2246 		mtx_unlock(&host->nh_lock);
2247 		result->stat.stat = nlm4_denied;
2248 		goto out;
2249 	}
2250 
2251 	error = nlm_cancel_async_lock(af);
2252 
2253 	if (error) {
2254 		result->stat.stat = nlm4_denied;
2255 	} else {
2256 		result->stat.stat = nlm4_granted;
2257 	}
2258 
2259 	mtx_unlock(&host->nh_lock);
2260 
2261 out:
2262 	nlm_release_vfs_state(&vs);
2263 	if (rpcp)
2264 		*rpcp = nlm_host_get_rpc(host, TRUE);
2265 	nlm_host_release(host);
2266 	return (0);
2267 }
2268 
2269 int
nlm_do_unlock(nlm4_unlockargs * argp,nlm4_res * result,struct svc_req * rqstp,CLIENT ** rpcp)2270 nlm_do_unlock(nlm4_unlockargs *argp, nlm4_res *result, struct svc_req *rqstp,
2271     CLIENT **rpcp)
2272 {
2273 	fhandle_t fh;
2274 	struct vfs_state vs;
2275 	struct nlm_host *host;
2276 	int error, sysid;
2277 	struct flock fl;
2278 
2279 	memset(result, 0, sizeof(*result));
2280 	memset(&vs, 0, sizeof(vs));
2281 
2282 	host = nlm_find_host_by_name(argp->alock.caller_name,
2283 	    svc_getrpccaller(rqstp), rqstp->rq_vers);
2284 	if (!host) {
2285 		result->stat.stat = nlm4_denied_nolocks;
2286 		return (ENOMEM);
2287 	}
2288 
2289 	NLM_DEBUG(3, "nlm_do_unlock(): caller_name = %s (sysid = %d)\n",
2290 	    host->nh_caller_name, host->nh_sysid);
2291 
2292 	nlm_check_expired_locks(host);
2293 	sysid = host->nh_sysid;
2294 
2295 	nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
2296 	nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2297 
2298 	if (time_uptime < nlm_grace_threshold) {
2299 		result->stat.stat = nlm4_denied_grace_period;
2300 		goto out;
2301 	}
2302 
2303 	error = nlm_get_vfs_state(host, rqstp, &fh, &vs, (accmode_t)0);
2304 	if (error) {
2305 		result->stat.stat = nlm_convert_error(error);
2306 		goto out;
2307 	}
2308 
2309 	fl.l_start = argp->alock.l_offset;
2310 	fl.l_len = argp->alock.l_len;
2311 	fl.l_pid = argp->alock.svid;
2312 	fl.l_sysid = sysid;
2313 	fl.l_whence = SEEK_SET;
2314 	fl.l_type = F_UNLCK;
2315 	error = VOP_ADVLOCK(vs.vs_vp, NULL, F_UNLCK, &fl, F_REMOTE);
2316 
2317 	/*
2318 	 * Ignore the error - there is no result code for failure,
2319 	 * only for grace period.
2320 	 */
2321 	result->stat.stat = nlm4_granted;
2322 
2323 out:
2324 	nlm_release_vfs_state(&vs);
2325 	if (rpcp)
2326 		*rpcp = nlm_host_get_rpc(host, TRUE);
2327 	nlm_host_release(host);
2328 	return (0);
2329 }
2330 
2331 int
nlm_do_granted(nlm4_testargs * argp,nlm4_res * result,struct svc_req * rqstp,CLIENT ** rpcp)2332 nlm_do_granted(nlm4_testargs *argp, nlm4_res *result, struct svc_req *rqstp,
2333 
2334     CLIENT **rpcp)
2335 {
2336 	struct nlm_host *host;
2337 	struct nlm_waiting_lock *nw;
2338 
2339 	memset(result, 0, sizeof(*result));
2340 
2341 	host = nlm_find_host_by_addr(svc_getrpccaller(rqstp), rqstp->rq_vers);
2342 	if (!host) {
2343 		result->stat.stat = nlm4_denied_nolocks;
2344 		return (ENOMEM);
2345 	}
2346 
2347 	nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2348 	result->stat.stat = nlm4_denied;
2349 	KFAIL_POINT_CODE(DEBUG_FP, nlm_deny_grant, goto out);
2350 
2351 	mtx_lock(&nlm_global_lock);
2352 	TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
2353 		if (!nw->nw_waiting)
2354 			continue;
2355 		if (argp->alock.svid == nw->nw_lock.svid
2356 		    && argp->alock.l_offset == nw->nw_lock.l_offset
2357 		    && argp->alock.l_len == nw->nw_lock.l_len
2358 		    && argp->alock.fh.n_len == nw->nw_lock.fh.n_len
2359 		    && !memcmp(argp->alock.fh.n_bytes, nw->nw_lock.fh.n_bytes,
2360 			nw->nw_lock.fh.n_len)) {
2361 			nw->nw_waiting = FALSE;
2362 			wakeup(nw);
2363 			result->stat.stat = nlm4_granted;
2364 			break;
2365 		}
2366 	}
2367 	mtx_unlock(&nlm_global_lock);
2368 
2369 out:
2370 	if (rpcp)
2371 		*rpcp = nlm_host_get_rpc(host, TRUE);
2372 	nlm_host_release(host);
2373 	return (0);
2374 }
2375 
2376 void
nlm_do_granted_res(nlm4_res * argp,struct svc_req * rqstp)2377 nlm_do_granted_res(nlm4_res *argp, struct svc_req *rqstp)
2378 {
2379 	struct nlm_host *host = NULL;
2380 	struct nlm_async_lock *af = NULL;
2381 	int error;
2382 
2383 	if (argp->cookie.n_len != sizeof(struct nlm_grantcookie)) {
2384 		NLM_DEBUG(1, "NLM: bogus grant cookie");
2385 		goto out;
2386 	}
2387 
2388 	host = nlm_find_host_by_sysid(ng_sysid(&argp->cookie));
2389 	if (!host) {
2390 		NLM_DEBUG(1, "NLM: Unknown host rejected our grant");
2391 		goto out;
2392 	}
2393 
2394 	mtx_lock(&host->nh_lock);
2395 	TAILQ_FOREACH(af, &host->nh_granted, af_link)
2396 	    if (ng_cookie(&argp->cookie) ==
2397 		ng_cookie(&af->af_granted.cookie))
2398 		    break;
2399 	if (af)
2400 		TAILQ_REMOVE(&host->nh_granted, af, af_link);
2401 	mtx_unlock(&host->nh_lock);
2402 
2403 	if (!af) {
2404 		NLM_DEBUG(1, "NLM: host %s (sysid %d) replied to our grant "
2405 		    "with unrecognized cookie %d:%d", host->nh_caller_name,
2406 		    host->nh_sysid, ng_sysid(&argp->cookie),
2407 		    ng_cookie(&argp->cookie));
2408 		goto out;
2409 	}
2410 
2411 	if (argp->stat.stat != nlm4_granted) {
2412 		af->af_fl.l_type = F_UNLCK;
2413 		error = VOP_ADVLOCK(af->af_vp, NULL, F_UNLCK, &af->af_fl, F_REMOTE);
2414 		if (error) {
2415 			NLM_DEBUG(1, "NLM: host %s (sysid %d) rejected our grant "
2416 			    "and we failed to unlock (%d)", host->nh_caller_name,
2417 			    host->nh_sysid, error);
2418 			goto out;
2419 		}
2420 
2421 		NLM_DEBUG(5, "NLM: async lock %p rejected by host %s (sysid %d)",
2422 		    af, host->nh_caller_name, host->nh_sysid);
2423 	} else {
2424 		NLM_DEBUG(5, "NLM: async lock %p accepted by host %s (sysid %d)",
2425 		    af, host->nh_caller_name, host->nh_sysid);
2426 	}
2427 
2428  out:
2429 	if (af)
2430 		nlm_free_async_lock(af);
2431 	if (host)
2432 		nlm_host_release(host);
2433 }
2434 
2435 void
nlm_do_free_all(nlm4_notify * argp)2436 nlm_do_free_all(nlm4_notify *argp)
2437 {
2438 	struct nlm_host *host, *thost;
2439 
2440 	TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, thost) {
2441 		if (!strcmp(host->nh_caller_name, argp->name))
2442 			nlm_host_notify(host, argp->state);
2443 	}
2444 }
2445 
2446 /*
2447  * Kernel module glue
2448  */
2449 static int
nfslockd_modevent(module_t mod,int type,void * data)2450 nfslockd_modevent(module_t mod, int type, void *data)
2451 {
2452 
2453 	switch (type) {
2454 	case MOD_LOAD:
2455 		return (0);
2456 	case MOD_UNLOAD:
2457 		/* The NLM module cannot be safely unloaded. */
2458 		/* FALLTHROUGH */
2459 	default:
2460 		return (EOPNOTSUPP);
2461 	}
2462 }
2463 static moduledata_t nfslockd_mod = {
2464 	"nfslockd",
2465 	nfslockd_modevent,
2466 	NULL,
2467 };
2468 DECLARE_MODULE(nfslockd, nfslockd_mod, SI_SUB_VFS, SI_ORDER_ANY);
2469 
2470 /* So that loader and kldload(2) can find us, wherever we are.. */
2471 MODULE_DEPEND(nfslockd, krpc, 1, 1, 1);
2472 MODULE_DEPEND(nfslockd, nfslock, 1, 1, 1);
2473 MODULE_VERSION(nfslockd, 1);
2474