1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2008 Isilon Inc http://www.isilon.com/
5 * Authors: Doug Rabson <dfr@rabson.org>
6 * Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org>
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30 #include "opt_inet6.h"
31
32 #include <sys/param.h>
33 #include <sys/fail.h>
34 #include <sys/fcntl.h>
35 #include <sys/kernel.h>
36 #include <sys/kthread.h>
37 #include <sys/lockf.h>
38 #include <sys/malloc.h>
39 #include <sys/mount.h>
40 #include <sys/priv.h>
41 #include <sys/proc.h>
42 #include <sys/socket.h>
43 #include <sys/socketvar.h>
44 #include <sys/syscall.h>
45 #include <sys/sysctl.h>
46 #include <sys/sysent.h>
47 #include <sys/syslog.h>
48 #include <sys/sysproto.h>
49 #include <sys/systm.h>
50 #include <sys/taskqueue.h>
51 #include <sys/unistd.h>
52 #include <sys/vnode.h>
53
54 #include <nfs/nfsproto.h>
55 #include <nfs/nfs_lock.h>
56
57 #include <nlm/nlm_prot.h>
58 #include <nlm/sm_inter.h>
59 #include <nlm/nlm.h>
60 #include <rpc/rpc_com.h>
61 #include <rpc/rpcb_prot.h>
62
63 MALLOC_DEFINE(M_NLM, "NLM", "Network Lock Manager");
64
65 /*
66 * If a host is inactive (and holds no locks) for this amount of
67 * seconds, we consider it idle and stop tracking it.
68 */
69 #define NLM_IDLE_TIMEOUT 30
70
71 /*
72 * We check the host list for idle every few seconds.
73 */
74 #define NLM_IDLE_PERIOD 5
75
76 /*
77 * We only look for GRANTED_RES messages for a little while.
78 */
79 #define NLM_EXPIRE_TIMEOUT 10
80
81 /*
82 * Support for sysctl vfs.nlm.sysid
83 */
84 static SYSCTL_NODE(_vfs, OID_AUTO, nlm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
85 "Network Lock Manager");
86 static SYSCTL_NODE(_vfs_nlm, OID_AUTO, sysid,
87 CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
88 "");
89
90 /*
91 * Syscall hooks
92 */
93 static struct syscall_helper_data nlm_syscalls[] = {
94 SYSCALL_INIT_HELPER(nlm_syscall),
95 SYSCALL_INIT_LAST
96 };
97
98 /*
99 * Debug level passed in from userland. We also support a sysctl hook
100 * so that it can be changed on a live system.
101 */
102 static int nlm_debug_level;
103 SYSCTL_INT(_debug, OID_AUTO, nlm_debug, CTLFLAG_RW, &nlm_debug_level, 0, "");
104
105 #define NLM_DEBUG(_level, args...) \
106 do { \
107 if (nlm_debug_level >= (_level)) \
108 log(LOG_DEBUG, args); \
109 } while(0)
110 #define NLM_ERR(args...) \
111 do { \
112 log(LOG_ERR, args); \
113 } while(0)
114
115 /*
116 * Grace period handling. The value of nlm_grace_threshold is the
117 * value of time_uptime after which we are serving requests normally.
118 */
119 static time_t nlm_grace_threshold;
120
121 /*
122 * We check for idle hosts if time_uptime is greater than
123 * nlm_next_idle_check,
124 */
125 static time_t nlm_next_idle_check;
126
127 /*
128 * A flag to indicate the server is already running.
129 */
130 static int nlm_is_running;
131
132 /*
133 * A socket to use for RPC - shared by all IPv4 RPC clients.
134 */
135 static struct socket *nlm_socket;
136
137 #ifdef INET6
138
139 /*
140 * A socket to use for RPC - shared by all IPv6 RPC clients.
141 */
142 static struct socket *nlm_socket6;
143
144 #endif
145
146 /*
147 * An RPC client handle that can be used to communicate with the local
148 * NSM.
149 */
150 static CLIENT *nlm_nsm;
151
152 /*
153 * An AUTH handle for the server's creds.
154 */
155 static AUTH *nlm_auth;
156
157 /*
158 * A zero timeval for sending async RPC messages.
159 */
160 struct timeval nlm_zero_tv = { 0, 0 };
161
162 /*
163 * The local NSM state number
164 */
165 int nlm_nsm_state;
166
167 /*
168 * A lock to protect the host list and waiting lock list.
169 */
170 static struct mtx nlm_global_lock;
171
172 /*
173 * Locks:
174 * (l) locked by nh_lock
175 * (s) only accessed via server RPC which is single threaded
176 * (g) locked by nlm_global_lock
177 * (c) const until freeing
178 * (a) modified using atomic ops
179 */
180
181 /*
182 * A pending client-side lock request, stored on the nlm_waiting_locks
183 * list.
184 */
185 struct nlm_waiting_lock {
186 TAILQ_ENTRY(nlm_waiting_lock) nw_link; /* (g) */
187 bool_t nw_waiting; /* (g) */
188 nlm4_lock nw_lock; /* (c) */
189 union nfsfh nw_fh; /* (c) */
190 struct vnode *nw_vp; /* (c) */
191 };
192 TAILQ_HEAD(nlm_waiting_lock_list, nlm_waiting_lock);
193
194 struct nlm_waiting_lock_list nlm_waiting_locks; /* (g) */
195
196 /*
197 * A pending server-side asynchronous lock request, stored on the
198 * nh_pending list of the NLM host.
199 */
200 struct nlm_async_lock {
201 TAILQ_ENTRY(nlm_async_lock) af_link; /* (l) host's list of locks */
202 struct task af_task; /* (c) async callback details */
203 void *af_cookie; /* (l) lock manager cancel token */
204 struct vnode *af_vp; /* (l) vnode to lock */
205 struct flock af_fl; /* (c) lock details */
206 struct nlm_host *af_host; /* (c) host which is locking */
207 CLIENT *af_rpc; /* (c) rpc client to send message */
208 nlm4_testargs af_granted; /* (c) notification details */
209 time_t af_expiretime; /* (c) notification time */
210 };
211 TAILQ_HEAD(nlm_async_lock_list, nlm_async_lock);
212
213 /*
214 * NLM host.
215 */
216 enum nlm_host_state {
217 NLM_UNMONITORED,
218 NLM_MONITORED,
219 NLM_MONITOR_FAILED,
220 NLM_RECOVERING
221 };
222
223 struct nlm_rpc {
224 CLIENT *nr_client; /* (l) RPC client handle */
225 time_t nr_create_time; /* (l) when client was created */
226 };
227
228 struct nlm_host {
229 struct mtx nh_lock;
230 volatile u_int nh_refs; /* (a) reference count */
231 TAILQ_ENTRY(nlm_host) nh_link; /* (g) global list of hosts */
232 char nh_caller_name[MAXNAMELEN]; /* (c) printable name of host */
233 uint32_t nh_sysid; /* (c) our allocaed system ID */
234 char nh_sysid_string[10]; /* (c) string rep. of sysid */
235 struct sockaddr_storage nh_addr; /* (s) remote address of host */
236 struct nlm_rpc nh_srvrpc; /* (l) RPC for server replies */
237 struct nlm_rpc nh_clntrpc; /* (l) RPC for client requests */
238 rpcvers_t nh_vers; /* (s) NLM version of host */
239 int nh_state; /* (s) last seen NSM state of host */
240 enum nlm_host_state nh_monstate; /* (l) local NSM monitoring state */
241 time_t nh_idle_timeout; /* (s) Time at which host is idle */
242 struct sysctl_ctx_list nh_sysctl; /* (c) vfs.nlm.sysid nodes */
243 uint32_t nh_grantcookie; /* (l) grant cookie counter */
244 struct nlm_async_lock_list nh_pending; /* (l) pending async locks */
245 struct nlm_async_lock_list nh_granted; /* (l) granted locks */
246 struct nlm_async_lock_list nh_finished; /* (l) finished async locks */
247 };
248 TAILQ_HEAD(nlm_host_list, nlm_host);
249
250 static struct nlm_host_list nlm_hosts; /* (g) */
251 static uint32_t nlm_next_sysid = 1; /* (g) */
252
253 static void nlm_host_unmonitor(struct nlm_host *);
254
255 struct nlm_grantcookie {
256 uint32_t ng_sysid;
257 uint32_t ng_cookie;
258 };
259
260 static inline uint32_t
ng_sysid(struct netobj * src)261 ng_sysid(struct netobj *src)
262 {
263
264 return ((struct nlm_grantcookie *)src->n_bytes)->ng_sysid;
265 }
266
267 static inline uint32_t
ng_cookie(struct netobj * src)268 ng_cookie(struct netobj *src)
269 {
270
271 return ((struct nlm_grantcookie *)src->n_bytes)->ng_cookie;
272 }
273
274 /**********************************************************************/
275
276 /*
277 * Initialise NLM globals.
278 */
279 static int
nlm_init(void)280 nlm_init(void)
281 {
282 int error;
283
284 mtx_init(&nlm_global_lock, "nlm_global_lock", NULL, MTX_DEF);
285 TAILQ_INIT(&nlm_waiting_locks);
286 TAILQ_INIT(&nlm_hosts);
287
288 error = syscall_helper_register(nlm_syscalls, SY_THR_STATIC_KLD);
289 if (error != 0)
290 NLM_ERR("Can't register NLM syscall\n");
291 return (error);
292 }
293
294 static void
nlm_uninit(void)295 nlm_uninit(void)
296 {
297
298 syscall_helper_unregister(nlm_syscalls);
299 }
300
301 /*
302 * Create a netobj from an arbitrary source.
303 */
304 void
nlm_make_netobj(struct netobj * dst,caddr_t src,size_t srcsize,struct malloc_type * type)305 nlm_make_netobj(struct netobj *dst, caddr_t src, size_t srcsize,
306 struct malloc_type *type)
307 {
308
309 dst->n_len = srcsize;
310 dst->n_bytes = malloc(srcsize, type, M_WAITOK);
311 memcpy(dst->n_bytes, src, srcsize);
312 }
313
314 /*
315 * Copy a struct netobj.
316 */
317 void
nlm_copy_netobj(struct netobj * dst,struct netobj * src,struct malloc_type * type)318 nlm_copy_netobj(struct netobj *dst, struct netobj *src,
319 struct malloc_type *type)
320 {
321
322 nlm_make_netobj(dst, src->n_bytes, src->n_len, type);
323 }
324
325 /*
326 * Create an RPC client handle for the given (address,prog,vers)
327 * triple using UDP.
328 */
329 static CLIENT *
nlm_get_rpc(struct sockaddr * sa,rpcprog_t prog,rpcvers_t vers)330 nlm_get_rpc(struct sockaddr *sa, rpcprog_t prog, rpcvers_t vers)
331 {
332 char *wchan = "nlmrcv";
333 struct sockaddr_storage ss;
334 struct socket *so;
335 CLIENT *rpcb;
336 struct timeval timo;
337 RPCB parms;
338 char *uaddr;
339 enum clnt_stat stat = RPC_SUCCESS;
340 int rpcvers = RPCBVERS4;
341 bool_t do_tcp = FALSE;
342 bool_t tryagain = FALSE;
343 struct portmap mapping;
344 u_short port = 0;
345 struct sockaddr_in *sin4;
346 char namebuf[INET_ADDRSTRLEN];
347 #ifdef INET6
348 struct sockaddr_in6 *sin6;
349 char namebuf6[INET6_ADDRSTRLEN];
350 #endif
351
352 /*
353 * First we need to contact the remote RPCBIND service to find
354 * the right port.
355 */
356 memcpy(&ss, sa, sa->sa_len);
357 switch (ss.ss_family) {
358 case AF_INET:
359 ((struct sockaddr_in *)&ss)->sin_port = htons(111);
360 so = nlm_socket;
361 break;
362 #ifdef INET6
363 case AF_INET6:
364 ((struct sockaddr_in6 *)&ss)->sin6_port = htons(111);
365 so = nlm_socket6;
366 break;
367 #endif
368
369 default:
370 /*
371 * Unsupported address family - fail.
372 */
373 return (NULL);
374 }
375
376 rpcb = clnt_dg_create(so, (struct sockaddr *)&ss,
377 RPCBPROG, rpcvers, 0, 0);
378 if (!rpcb)
379 return (NULL);
380
381 try_tcp:
382 parms.r_prog = prog;
383 parms.r_vers = vers;
384 if (do_tcp)
385 parms.r_netid = "tcp";
386 else
387 parms.r_netid = "udp";
388 parms.r_addr = "";
389 parms.r_owner = "";
390
391 /*
392 * Use the default timeout.
393 */
394 timo.tv_sec = 25;
395 timo.tv_usec = 0;
396 again:
397 switch (rpcvers) {
398 case RPCBVERS4:
399 case RPCBVERS:
400 /*
401 * Try RPCBIND 4 then 3.
402 */
403 uaddr = NULL;
404 stat = CLNT_CALL(rpcb, (rpcprog_t) RPCBPROC_GETADDR,
405 (xdrproc_t) xdr_rpcb, &parms,
406 (xdrproc_t) xdr_wrapstring, &uaddr, timo);
407 if (stat == RPC_SUCCESS) {
408 /*
409 * We have a reply from the remote RPCBIND - turn it
410 * into an appropriate address and make a new client
411 * that can talk to the remote NLM.
412 *
413 * XXX fixup IPv6 scope ID.
414 */
415 struct netbuf *a;
416 a = __rpc_uaddr2taddr_af(ss.ss_family, uaddr);
417 if (!a) {
418 tryagain = TRUE;
419 } else {
420 tryagain = FALSE;
421 memcpy(&ss, a->buf, a->len);
422 free(a->buf, M_RPC);
423 free(a, M_RPC);
424 xdr_free((xdrproc_t) xdr_wrapstring, &uaddr);
425 }
426 }
427 if (tryagain || stat == RPC_PROGVERSMISMATCH) {
428 if (rpcvers == RPCBVERS4)
429 rpcvers = RPCBVERS;
430 else if (rpcvers == RPCBVERS)
431 rpcvers = PMAPVERS;
432 CLNT_CONTROL(rpcb, CLSET_VERS, &rpcvers);
433 goto again;
434 }
435 break;
436 case PMAPVERS:
437 /*
438 * Try portmap.
439 */
440 mapping.pm_prog = parms.r_prog;
441 mapping.pm_vers = parms.r_vers;
442 mapping.pm_prot = do_tcp ? IPPROTO_TCP : IPPROTO_UDP;
443 mapping.pm_port = 0;
444
445 stat = CLNT_CALL(rpcb, (rpcprog_t) PMAPPROC_GETPORT,
446 (xdrproc_t) xdr_portmap, &mapping,
447 (xdrproc_t) xdr_u_short, &port, timo);
448
449 if (stat == RPC_SUCCESS) {
450 switch (ss.ss_family) {
451 case AF_INET:
452 ((struct sockaddr_in *)&ss)->sin_port =
453 htons(port);
454 break;
455
456 #ifdef INET6
457 case AF_INET6:
458 ((struct sockaddr_in6 *)&ss)->sin6_port =
459 htons(port);
460 break;
461 #endif
462 }
463 }
464 break;
465 default:
466 panic("invalid rpcvers %d", rpcvers);
467 }
468 /*
469 * We may have a positive response from the portmapper, but the NLM
470 * service was not found. Make sure we received a valid port.
471 */
472 switch (ss.ss_family) {
473 case AF_INET:
474 port = ((struct sockaddr_in *)&ss)->sin_port;
475 break;
476 #ifdef INET6
477 case AF_INET6:
478 port = ((struct sockaddr_in6 *)&ss)->sin6_port;
479 break;
480 #endif
481 }
482 if (stat != RPC_SUCCESS || !port) {
483 /*
484 * If we were able to talk to rpcbind or portmap, but the udp
485 * variant wasn't available, ask about tcp.
486 *
487 * XXX - We could also check for a TCP portmapper, but
488 * if the host is running a portmapper at all, we should be able
489 * to hail it over UDP.
490 */
491 if (stat == RPC_SUCCESS && !do_tcp) {
492 do_tcp = TRUE;
493 goto try_tcp;
494 }
495
496 /* Otherwise, bad news. */
497 switch (ss.ss_family) {
498 case AF_INET:
499 sin4 = (struct sockaddr_in *)&ss;
500 inet_ntop(ss.ss_family, &sin4->sin_addr,
501 namebuf, sizeof namebuf);
502 NLM_ERR("NLM: failed to contact remote rpcbind, "
503 "stat = %d, host = %s, port = %d\n",
504 (int) stat, namebuf, htons(port));
505 break;
506 #ifdef INET6
507 case AF_INET6:
508 sin6 = (struct sockaddr_in6 *)&ss;
509 inet_ntop(ss.ss_family, &sin6->sin6_addr,
510 namebuf6, sizeof namebuf6);
511 NLM_ERR("NLM: failed to contact remote rpcbind, "
512 "stat = %d, host = %s, port = %d\n",
513 (int) stat, namebuf6, htons(port));
514 break;
515 #endif
516 }
517 CLNT_DESTROY(rpcb);
518 return (NULL);
519 }
520
521 if (do_tcp) {
522 /*
523 * Destroy the UDP client we used to speak to rpcbind and
524 * recreate as a TCP client.
525 */
526 struct netconfig *nconf = NULL;
527
528 CLNT_DESTROY(rpcb);
529
530 switch (ss.ss_family) {
531 case AF_INET:
532 nconf = getnetconfigent("tcp");
533 break;
534 #ifdef INET6
535 case AF_INET6:
536 nconf = getnetconfigent("tcp6");
537 break;
538 #endif
539 }
540
541 rpcb = clnt_reconnect_create(nconf, (struct sockaddr *)&ss,
542 prog, vers, 0, 0);
543 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan);
544 rpcb->cl_auth = nlm_auth;
545
546 } else {
547 /*
548 * Re-use the client we used to speak to rpcbind.
549 */
550 CLNT_CONTROL(rpcb, CLSET_SVC_ADDR, &ss);
551 CLNT_CONTROL(rpcb, CLSET_PROG, &prog);
552 CLNT_CONTROL(rpcb, CLSET_VERS, &vers);
553 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan);
554 rpcb->cl_auth = nlm_auth;
555 }
556
557 return (rpcb);
558 }
559
560 /*
561 * This async callback after when an async lock request has been
562 * granted. We notify the host which initiated the request.
563 */
564 static void
nlm_lock_callback(void * arg,int pending)565 nlm_lock_callback(void *arg, int pending)
566 {
567 struct nlm_async_lock *af = (struct nlm_async_lock *) arg;
568 struct rpc_callextra ext;
569
570 NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) granted,"
571 " cookie %d:%d\n", af, af->af_host->nh_caller_name,
572 af->af_host->nh_sysid, ng_sysid(&af->af_granted.cookie),
573 ng_cookie(&af->af_granted.cookie));
574
575 /*
576 * Send the results back to the host.
577 *
578 * Note: there is a possible race here with nlm_host_notify
579 * destroying the RPC client. To avoid problems, the first
580 * thing nlm_host_notify does is to cancel pending async lock
581 * requests.
582 */
583 memset(&ext, 0, sizeof(ext));
584 ext.rc_auth = nlm_auth;
585 if (af->af_host->nh_vers == NLM_VERS4) {
586 nlm4_granted_msg_4(&af->af_granted,
587 NULL, af->af_rpc, &ext, nlm_zero_tv);
588 } else {
589 /*
590 * Back-convert to legacy protocol
591 */
592 nlm_testargs granted;
593 granted.cookie = af->af_granted.cookie;
594 granted.exclusive = af->af_granted.exclusive;
595 granted.alock.caller_name =
596 af->af_granted.alock.caller_name;
597 granted.alock.fh = af->af_granted.alock.fh;
598 granted.alock.oh = af->af_granted.alock.oh;
599 granted.alock.svid = af->af_granted.alock.svid;
600 granted.alock.l_offset =
601 af->af_granted.alock.l_offset;
602 granted.alock.l_len =
603 af->af_granted.alock.l_len;
604
605 nlm_granted_msg_1(&granted,
606 NULL, af->af_rpc, &ext, nlm_zero_tv);
607 }
608
609 /*
610 * Move this entry to the nh_granted list.
611 */
612 af->af_expiretime = time_uptime + NLM_EXPIRE_TIMEOUT;
613 mtx_lock(&af->af_host->nh_lock);
614 TAILQ_REMOVE(&af->af_host->nh_pending, af, af_link);
615 TAILQ_INSERT_TAIL(&af->af_host->nh_granted, af, af_link);
616 mtx_unlock(&af->af_host->nh_lock);
617 }
618
619 /*
620 * Free an async lock request. The request must have been removed from
621 * any list.
622 */
623 static void
nlm_free_async_lock(struct nlm_async_lock * af)624 nlm_free_async_lock(struct nlm_async_lock *af)
625 {
626 /*
627 * Free an async lock.
628 */
629 if (af->af_rpc)
630 CLNT_RELEASE(af->af_rpc);
631 xdr_free((xdrproc_t) xdr_nlm4_testargs, &af->af_granted);
632 if (af->af_vp)
633 vrele(af->af_vp);
634 free(af, M_NLM);
635 }
636
637 /*
638 * Cancel our async request - this must be called with
639 * af->nh_host->nh_lock held. This is slightly complicated by a
640 * potential race with our own callback. If we fail to cancel the
641 * lock, it must already have been granted - we make sure our async
642 * task has completed by calling taskqueue_drain in this case.
643 */
644 static int
nlm_cancel_async_lock(struct nlm_async_lock * af)645 nlm_cancel_async_lock(struct nlm_async_lock *af)
646 {
647 struct nlm_host *host = af->af_host;
648 int error;
649
650 mtx_assert(&host->nh_lock, MA_OWNED);
651
652 mtx_unlock(&host->nh_lock);
653
654 error = VOP_ADVLOCKASYNC(af->af_vp, NULL, F_CANCEL, &af->af_fl,
655 F_REMOTE, NULL, &af->af_cookie);
656
657 if (error) {
658 /*
659 * We failed to cancel - make sure our callback has
660 * completed before we continue.
661 */
662 taskqueue_drain(taskqueue_thread, &af->af_task);
663 }
664
665 mtx_lock(&host->nh_lock);
666
667 if (!error) {
668 NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) "
669 "cancelled\n", af, host->nh_caller_name, host->nh_sysid);
670
671 /*
672 * Remove from the nh_pending list and free now that
673 * we are safe from the callback.
674 */
675 TAILQ_REMOVE(&host->nh_pending, af, af_link);
676 mtx_unlock(&host->nh_lock);
677 nlm_free_async_lock(af);
678 mtx_lock(&host->nh_lock);
679 }
680
681 return (error);
682 }
683
684 static void
nlm_check_expired_locks(struct nlm_host * host)685 nlm_check_expired_locks(struct nlm_host *host)
686 {
687 struct nlm_async_lock *af;
688 time_t uptime = time_uptime;
689
690 mtx_lock(&host->nh_lock);
691 while ((af = TAILQ_FIRST(&host->nh_granted)) != NULL
692 && uptime >= af->af_expiretime) {
693 NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) expired,"
694 " cookie %d:%d\n", af, af->af_host->nh_caller_name,
695 af->af_host->nh_sysid, ng_sysid(&af->af_granted.cookie),
696 ng_cookie(&af->af_granted.cookie));
697 TAILQ_REMOVE(&host->nh_granted, af, af_link);
698 mtx_unlock(&host->nh_lock);
699 nlm_free_async_lock(af);
700 mtx_lock(&host->nh_lock);
701 }
702 while ((af = TAILQ_FIRST(&host->nh_finished)) != NULL) {
703 TAILQ_REMOVE(&host->nh_finished, af, af_link);
704 mtx_unlock(&host->nh_lock);
705 nlm_free_async_lock(af);
706 mtx_lock(&host->nh_lock);
707 }
708 mtx_unlock(&host->nh_lock);
709 }
710
711 /*
712 * Free resources used by a host. This is called after the reference
713 * count has reached zero so it doesn't need to worry about locks.
714 */
715 static void
nlm_host_destroy(struct nlm_host * host)716 nlm_host_destroy(struct nlm_host *host)
717 {
718
719 mtx_lock(&nlm_global_lock);
720 TAILQ_REMOVE(&nlm_hosts, host, nh_link);
721 mtx_unlock(&nlm_global_lock);
722
723 if (host->nh_srvrpc.nr_client)
724 CLNT_RELEASE(host->nh_srvrpc.nr_client);
725 if (host->nh_clntrpc.nr_client)
726 CLNT_RELEASE(host->nh_clntrpc.nr_client);
727 mtx_destroy(&host->nh_lock);
728 sysctl_ctx_free(&host->nh_sysctl);
729 free(host, M_NLM);
730 }
731
732 /*
733 * Thread start callback for client lock recovery
734 */
735 static void
nlm_client_recovery_start(void * arg)736 nlm_client_recovery_start(void *arg)
737 {
738 struct nlm_host *host = (struct nlm_host *) arg;
739
740 NLM_DEBUG(1, "NLM: client lock recovery for %s started\n",
741 host->nh_caller_name);
742
743 nlm_client_recovery(host);
744
745 NLM_DEBUG(1, "NLM: client lock recovery for %s completed\n",
746 host->nh_caller_name);
747
748 host->nh_monstate = NLM_MONITORED;
749 nlm_host_release(host);
750
751 kthread_exit();
752 }
753
754 /*
755 * This is called when we receive a host state change notification. We
756 * unlock any active locks owned by the host. When rpc.lockd is
757 * shutting down, this function is called with newstate set to zero
758 * which allows us to cancel any pending async locks and clear the
759 * locking state.
760 */
761 static void
nlm_host_notify(struct nlm_host * host,int newstate)762 nlm_host_notify(struct nlm_host *host, int newstate)
763 {
764 struct nlm_async_lock *af;
765
766 if (newstate) {
767 NLM_DEBUG(1, "NLM: host %s (sysid %d) rebooted, new "
768 "state is %d\n", host->nh_caller_name,
769 host->nh_sysid, newstate);
770 }
771
772 /*
773 * Cancel any pending async locks for this host.
774 */
775 mtx_lock(&host->nh_lock);
776 while ((af = TAILQ_FIRST(&host->nh_pending)) != NULL) {
777 /*
778 * nlm_cancel_async_lock will remove the entry from
779 * nh_pending and free it.
780 */
781 nlm_cancel_async_lock(af);
782 }
783 mtx_unlock(&host->nh_lock);
784 nlm_check_expired_locks(host);
785
786 /*
787 * The host just rebooted - trash its locks.
788 */
789 lf_clearremotesys(host->nh_sysid);
790 host->nh_state = newstate;
791
792 /*
793 * If we have any remote locks for this host (i.e. it
794 * represents a remote NFS server that our local NFS client
795 * has locks for), start a recovery thread.
796 */
797 if (newstate != 0
798 && host->nh_monstate != NLM_RECOVERING
799 && lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid) > 0) {
800 struct thread *td;
801 host->nh_monstate = NLM_RECOVERING;
802 refcount_acquire(&host->nh_refs);
803 kthread_add(nlm_client_recovery_start, host, curproc, &td, 0, 0,
804 "NFS lock recovery for %s", host->nh_caller_name);
805 }
806 }
807
808 /*
809 * Sysctl handler to count the number of locks for a sysid.
810 */
811 static int
nlm_host_lock_count_sysctl(SYSCTL_HANDLER_ARGS)812 nlm_host_lock_count_sysctl(SYSCTL_HANDLER_ARGS)
813 {
814 struct nlm_host *host;
815 int count;
816
817 host = oidp->oid_arg1;
818 count = lf_countlocks(host->nh_sysid);
819 return sysctl_handle_int(oidp, &count, 0, req);
820 }
821
822 /*
823 * Sysctl handler to count the number of client locks for a sysid.
824 */
825 static int
nlm_host_client_lock_count_sysctl(SYSCTL_HANDLER_ARGS)826 nlm_host_client_lock_count_sysctl(SYSCTL_HANDLER_ARGS)
827 {
828 struct nlm_host *host;
829 int count;
830
831 host = oidp->oid_arg1;
832 count = lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid);
833 return sysctl_handle_int(oidp, &count, 0, req);
834 }
835
836 /*
837 * Create a new NLM host.
838 */
839 static struct nlm_host *
nlm_create_host(const char * caller_name)840 nlm_create_host(const char* caller_name)
841 {
842 struct nlm_host *host;
843 struct sysctl_oid *oid;
844
845 mtx_assert(&nlm_global_lock, MA_OWNED);
846
847 NLM_DEBUG(1, "NLM: new host %s (sysid %d)\n",
848 caller_name, nlm_next_sysid);
849 host = malloc(sizeof(struct nlm_host), M_NLM, M_NOWAIT|M_ZERO);
850 if (!host)
851 return (NULL);
852 mtx_init(&host->nh_lock, "nh_lock", NULL, MTX_DEF);
853 refcount_init(&host->nh_refs, 1);
854 strlcpy(host->nh_caller_name, caller_name, MAXNAMELEN);
855 host->nh_sysid = nlm_next_sysid++;
856 snprintf(host->nh_sysid_string, sizeof(host->nh_sysid_string),
857 "%d", host->nh_sysid);
858 host->nh_vers = 0;
859 host->nh_state = 0;
860 host->nh_monstate = NLM_UNMONITORED;
861 host->nh_grantcookie = 1;
862 TAILQ_INIT(&host->nh_pending);
863 TAILQ_INIT(&host->nh_granted);
864 TAILQ_INIT(&host->nh_finished);
865 TAILQ_INSERT_TAIL(&nlm_hosts, host, nh_link);
866
867 mtx_unlock(&nlm_global_lock);
868
869 sysctl_ctx_init(&host->nh_sysctl);
870 oid = SYSCTL_ADD_NODE(&host->nh_sysctl,
871 SYSCTL_STATIC_CHILDREN(_vfs_nlm_sysid),
872 OID_AUTO, host->nh_sysid_string, CTLFLAG_RD | CTLFLAG_MPSAFE,
873 NULL, "");
874 SYSCTL_ADD_STRING(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
875 "hostname", CTLFLAG_RD, host->nh_caller_name, 0, "");
876 SYSCTL_ADD_UINT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
877 "version", CTLFLAG_RD, &host->nh_vers, 0, "");
878 SYSCTL_ADD_UINT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
879 "monitored", CTLFLAG_RD, &host->nh_monstate, 0, "");
880 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
881 "lock_count", CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, host,
882 0, nlm_host_lock_count_sysctl, "I", "");
883 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
884 "client_lock_count", CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
885 host, 0, nlm_host_client_lock_count_sysctl, "I", "");
886
887 mtx_lock(&nlm_global_lock);
888
889 return (host);
890 }
891
892 /*
893 * Acquire the next sysid for remote locks not handled by the NLM.
894 */
895 uint32_t
nlm_acquire_next_sysid(void)896 nlm_acquire_next_sysid(void)
897 {
898 uint32_t next_sysid;
899
900 mtx_lock(&nlm_global_lock);
901 next_sysid = nlm_next_sysid++;
902 mtx_unlock(&nlm_global_lock);
903 return (next_sysid);
904 }
905
906 /*
907 * Return non-zero if the address parts of the two sockaddrs are the
908 * same.
909 */
910 static int
nlm_compare_addr(const struct sockaddr * a,const struct sockaddr * b)911 nlm_compare_addr(const struct sockaddr *a, const struct sockaddr *b)
912 {
913 const struct sockaddr_in *a4, *b4;
914 #ifdef INET6
915 const struct sockaddr_in6 *a6, *b6;
916 #endif
917
918 if (a->sa_family != b->sa_family)
919 return (FALSE);
920
921 switch (a->sa_family) {
922 case AF_INET:
923 a4 = (const struct sockaddr_in *) a;
924 b4 = (const struct sockaddr_in *) b;
925 return !memcmp(&a4->sin_addr, &b4->sin_addr,
926 sizeof(a4->sin_addr));
927 #ifdef INET6
928 case AF_INET6:
929 a6 = (const struct sockaddr_in6 *) a;
930 b6 = (const struct sockaddr_in6 *) b;
931 return !memcmp(&a6->sin6_addr, &b6->sin6_addr,
932 sizeof(a6->sin6_addr));
933 #endif
934 }
935
936 return (0);
937 }
938
939 /*
940 * Check for idle hosts and stop monitoring them. We could also free
941 * the host structure here, possibly after a larger timeout but that
942 * would require some care to avoid races with
943 * e.g. nlm_host_lock_count_sysctl.
944 */
945 static void
nlm_check_idle(void)946 nlm_check_idle(void)
947 {
948 struct nlm_host *host;
949
950 mtx_assert(&nlm_global_lock, MA_OWNED);
951
952 if (time_uptime <= nlm_next_idle_check)
953 return;
954
955 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD;
956
957 TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
958 if (host->nh_monstate == NLM_MONITORED
959 && time_uptime > host->nh_idle_timeout) {
960 mtx_unlock(&nlm_global_lock);
961 if (lf_countlocks(host->nh_sysid) > 0
962 || lf_countlocks(NLM_SYSID_CLIENT
963 + host->nh_sysid)) {
964 host->nh_idle_timeout =
965 time_uptime + NLM_IDLE_TIMEOUT;
966 mtx_lock(&nlm_global_lock);
967 continue;
968 }
969 nlm_host_unmonitor(host);
970 mtx_lock(&nlm_global_lock);
971 }
972 }
973 }
974
975 /*
976 * Search for an existing NLM host that matches the given name
977 * (typically the caller_name element of an nlm4_lock). If none is
978 * found, create a new host. If 'addr' is non-NULL, record the remote
979 * address of the host so that we can call it back for async
980 * responses. If 'vers' is greater than zero then record the NLM
981 * program version to use to communicate with this client.
982 */
983 struct nlm_host *
nlm_find_host_by_name(const char * name,const struct sockaddr * addr,rpcvers_t vers)984 nlm_find_host_by_name(const char *name, const struct sockaddr *addr,
985 rpcvers_t vers)
986 {
987 struct nlm_host *host;
988
989 mtx_lock(&nlm_global_lock);
990
991 /*
992 * The remote host is determined by caller_name.
993 */
994 TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
995 if (!strcmp(host->nh_caller_name, name))
996 break;
997 }
998
999 if (!host) {
1000 host = nlm_create_host(name);
1001 if (!host) {
1002 mtx_unlock(&nlm_global_lock);
1003 return (NULL);
1004 }
1005 }
1006 refcount_acquire(&host->nh_refs);
1007
1008 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT;
1009
1010 /*
1011 * If we have an address for the host, record it so that we
1012 * can send async replies etc.
1013 */
1014 if (addr) {
1015
1016 KASSERT(addr->sa_len < sizeof(struct sockaddr_storage),
1017 ("Strange remote transport address length"));
1018
1019 /*
1020 * If we have seen an address before and we currently
1021 * have an RPC client handle, make sure the address is
1022 * the same, otherwise discard the client handle.
1023 */
1024 if (host->nh_addr.ss_len && host->nh_srvrpc.nr_client) {
1025 if (!nlm_compare_addr(
1026 (struct sockaddr *) &host->nh_addr,
1027 addr)
1028 || host->nh_vers != vers) {
1029 CLIENT *client;
1030 mtx_lock(&host->nh_lock);
1031 client = host->nh_srvrpc.nr_client;
1032 host->nh_srvrpc.nr_client = NULL;
1033 mtx_unlock(&host->nh_lock);
1034 if (client) {
1035 CLNT_RELEASE(client);
1036 }
1037 }
1038 }
1039 memcpy(&host->nh_addr, addr, addr->sa_len);
1040 host->nh_vers = vers;
1041 }
1042
1043 nlm_check_idle();
1044
1045 mtx_unlock(&nlm_global_lock);
1046
1047 return (host);
1048 }
1049
1050 /*
1051 * Search for an existing NLM host that matches the given remote
1052 * address. If none is found, create a new host with the requested
1053 * address and remember 'vers' as the NLM protocol version to use for
1054 * that host.
1055 */
1056 struct nlm_host *
nlm_find_host_by_addr(const struct sockaddr * addr,int vers)1057 nlm_find_host_by_addr(const struct sockaddr *addr, int vers)
1058 {
1059 /*
1060 * Fake up a name using inet_ntop. This buffer is
1061 * large enough for an IPv6 address.
1062 */
1063 char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"];
1064 struct nlm_host *host;
1065
1066 switch (addr->sa_family) {
1067 case AF_INET:
1068 inet_ntop(AF_INET,
1069 &((const struct sockaddr_in *) addr)->sin_addr,
1070 tmp, sizeof tmp);
1071 break;
1072 #ifdef INET6
1073 case AF_INET6:
1074 inet_ntop(AF_INET6,
1075 &((const struct sockaddr_in6 *) addr)->sin6_addr,
1076 tmp, sizeof tmp);
1077 break;
1078 #endif
1079 default:
1080 strlcpy(tmp, "<unknown>", sizeof(tmp));
1081 }
1082
1083 mtx_lock(&nlm_global_lock);
1084
1085 /*
1086 * The remote host is determined by caller_name.
1087 */
1088 TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
1089 if (nlm_compare_addr(addr,
1090 (const struct sockaddr *) &host->nh_addr))
1091 break;
1092 }
1093
1094 if (!host) {
1095 host = nlm_create_host(tmp);
1096 if (!host) {
1097 mtx_unlock(&nlm_global_lock);
1098 return (NULL);
1099 }
1100 memcpy(&host->nh_addr, addr, addr->sa_len);
1101 host->nh_vers = vers;
1102 }
1103 refcount_acquire(&host->nh_refs);
1104
1105 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT;
1106
1107 nlm_check_idle();
1108
1109 mtx_unlock(&nlm_global_lock);
1110
1111 return (host);
1112 }
1113
1114 /*
1115 * Find the NLM host that matches the value of 'sysid'. If none
1116 * exists, return NULL.
1117 */
1118 static struct nlm_host *
nlm_find_host_by_sysid(int sysid)1119 nlm_find_host_by_sysid(int sysid)
1120 {
1121 struct nlm_host *host;
1122
1123 TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
1124 if (host->nh_sysid == sysid) {
1125 refcount_acquire(&host->nh_refs);
1126 return (host);
1127 }
1128 }
1129
1130 return (NULL);
1131 }
1132
nlm_host_release(struct nlm_host * host)1133 void nlm_host_release(struct nlm_host *host)
1134 {
1135 if (refcount_release(&host->nh_refs)) {
1136 /*
1137 * Free the host
1138 */
1139 nlm_host_destroy(host);
1140 }
1141 }
1142
1143 /*
1144 * Unregister this NLM host with the local NSM due to idleness.
1145 */
1146 static void
nlm_host_unmonitor(struct nlm_host * host)1147 nlm_host_unmonitor(struct nlm_host *host)
1148 {
1149 mon_id smmonid;
1150 sm_stat_res smstat;
1151 struct timeval timo;
1152 enum clnt_stat stat;
1153
1154 NLM_DEBUG(1, "NLM: unmonitoring %s (sysid %d)\n",
1155 host->nh_caller_name, host->nh_sysid);
1156
1157 /*
1158 * We put our assigned system ID value in the priv field to
1159 * make it simpler to find the host if we are notified of a
1160 * host restart.
1161 */
1162 smmonid.mon_name = host->nh_caller_name;
1163 smmonid.my_id.my_name = "localhost";
1164 smmonid.my_id.my_prog = NLM_PROG;
1165 smmonid.my_id.my_vers = NLM_SM;
1166 smmonid.my_id.my_proc = NLM_SM_NOTIFY;
1167
1168 timo.tv_sec = 25;
1169 timo.tv_usec = 0;
1170 stat = CLNT_CALL(nlm_nsm, SM_UNMON,
1171 (xdrproc_t) xdr_mon, &smmonid,
1172 (xdrproc_t) xdr_sm_stat, &smstat, timo);
1173
1174 if (stat != RPC_SUCCESS) {
1175 NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat);
1176 return;
1177 }
1178 if (smstat.res_stat == stat_fail) {
1179 NLM_ERR("Local NSM refuses to unmonitor %s\n",
1180 host->nh_caller_name);
1181 return;
1182 }
1183
1184 host->nh_monstate = NLM_UNMONITORED;
1185 }
1186
1187 /*
1188 * Register this NLM host with the local NSM so that we can be
1189 * notified if it reboots.
1190 */
1191 void
nlm_host_monitor(struct nlm_host * host,int state)1192 nlm_host_monitor(struct nlm_host *host, int state)
1193 {
1194 mon smmon;
1195 sm_stat_res smstat;
1196 struct timeval timo;
1197 enum clnt_stat stat;
1198
1199 if (state && !host->nh_state) {
1200 /*
1201 * This is the first time we have seen an NSM state
1202 * value for this host. We record it here to help
1203 * detect host reboots.
1204 */
1205 host->nh_state = state;
1206 NLM_DEBUG(1, "NLM: host %s (sysid %d) has NSM state %d\n",
1207 host->nh_caller_name, host->nh_sysid, state);
1208 }
1209
1210 mtx_lock(&host->nh_lock);
1211 if (host->nh_monstate != NLM_UNMONITORED) {
1212 mtx_unlock(&host->nh_lock);
1213 return;
1214 }
1215 host->nh_monstate = NLM_MONITORED;
1216 mtx_unlock(&host->nh_lock);
1217
1218 NLM_DEBUG(1, "NLM: monitoring %s (sysid %d)\n",
1219 host->nh_caller_name, host->nh_sysid);
1220
1221 /*
1222 * We put our assigned system ID value in the priv field to
1223 * make it simpler to find the host if we are notified of a
1224 * host restart.
1225 */
1226 smmon.mon_id.mon_name = host->nh_caller_name;
1227 smmon.mon_id.my_id.my_name = "localhost";
1228 smmon.mon_id.my_id.my_prog = NLM_PROG;
1229 smmon.mon_id.my_id.my_vers = NLM_SM;
1230 smmon.mon_id.my_id.my_proc = NLM_SM_NOTIFY;
1231 memcpy(smmon.priv, &host->nh_sysid, sizeof(host->nh_sysid));
1232
1233 timo.tv_sec = 25;
1234 timo.tv_usec = 0;
1235 stat = CLNT_CALL(nlm_nsm, SM_MON,
1236 (xdrproc_t) xdr_mon, &smmon,
1237 (xdrproc_t) xdr_sm_stat, &smstat, timo);
1238
1239 if (stat != RPC_SUCCESS) {
1240 NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat);
1241 return;
1242 }
1243 if (smstat.res_stat == stat_fail) {
1244 NLM_ERR("Local NSM refuses to monitor %s\n",
1245 host->nh_caller_name);
1246 mtx_lock(&host->nh_lock);
1247 host->nh_monstate = NLM_MONITOR_FAILED;
1248 mtx_unlock(&host->nh_lock);
1249 return;
1250 }
1251
1252 host->nh_monstate = NLM_MONITORED;
1253 }
1254
1255 /*
1256 * Return an RPC client handle that can be used to talk to the NLM
1257 * running on the given host.
1258 */
1259 CLIENT *
nlm_host_get_rpc(struct nlm_host * host,bool_t isserver)1260 nlm_host_get_rpc(struct nlm_host *host, bool_t isserver)
1261 {
1262 struct nlm_rpc *rpc;
1263 CLIENT *client;
1264
1265 mtx_lock(&host->nh_lock);
1266
1267 if (isserver)
1268 rpc = &host->nh_srvrpc;
1269 else
1270 rpc = &host->nh_clntrpc;
1271
1272 /*
1273 * We can't hold onto RPC handles for too long - the async
1274 * call/reply protocol used by some NLM clients makes it hard
1275 * to tell when they change port numbers (e.g. after a
1276 * reboot). Note that if a client reboots while it isn't
1277 * holding any locks, it won't bother to notify us. We
1278 * expire the RPC handles after two minutes.
1279 */
1280 if (rpc->nr_client && time_uptime > rpc->nr_create_time + 2*60) {
1281 client = rpc->nr_client;
1282 rpc->nr_client = NULL;
1283 mtx_unlock(&host->nh_lock);
1284 CLNT_RELEASE(client);
1285 mtx_lock(&host->nh_lock);
1286 }
1287
1288 if (!rpc->nr_client) {
1289 mtx_unlock(&host->nh_lock);
1290 client = nlm_get_rpc((struct sockaddr *)&host->nh_addr,
1291 NLM_PROG, host->nh_vers);
1292 mtx_lock(&host->nh_lock);
1293
1294 if (client) {
1295 if (rpc->nr_client) {
1296 mtx_unlock(&host->nh_lock);
1297 CLNT_DESTROY(client);
1298 mtx_lock(&host->nh_lock);
1299 } else {
1300 rpc->nr_client = client;
1301 rpc->nr_create_time = time_uptime;
1302 }
1303 }
1304 }
1305
1306 client = rpc->nr_client;
1307 if (client)
1308 CLNT_ACQUIRE(client);
1309 mtx_unlock(&host->nh_lock);
1310
1311 return (client);
1312
1313 }
1314
nlm_host_get_sysid(struct nlm_host * host)1315 int nlm_host_get_sysid(struct nlm_host *host)
1316 {
1317
1318 return (host->nh_sysid);
1319 }
1320
1321 int
nlm_host_get_state(struct nlm_host * host)1322 nlm_host_get_state(struct nlm_host *host)
1323 {
1324
1325 return (host->nh_state);
1326 }
1327
1328 void *
nlm_register_wait_lock(struct nlm4_lock * lock,struct vnode * vp)1329 nlm_register_wait_lock(struct nlm4_lock *lock, struct vnode *vp)
1330 {
1331 struct nlm_waiting_lock *nw;
1332
1333 nw = malloc(sizeof(struct nlm_waiting_lock), M_NLM, M_WAITOK);
1334 nw->nw_lock = *lock;
1335 memcpy(&nw->nw_fh.fh_bytes, nw->nw_lock.fh.n_bytes,
1336 nw->nw_lock.fh.n_len);
1337 nw->nw_lock.fh.n_bytes = nw->nw_fh.fh_bytes;
1338 nw->nw_waiting = TRUE;
1339 nw->nw_vp = vp;
1340 mtx_lock(&nlm_global_lock);
1341 TAILQ_INSERT_TAIL(&nlm_waiting_locks, nw, nw_link);
1342 mtx_unlock(&nlm_global_lock);
1343
1344 return nw;
1345 }
1346
1347 void
nlm_deregister_wait_lock(void * handle)1348 nlm_deregister_wait_lock(void *handle)
1349 {
1350 struct nlm_waiting_lock *nw = handle;
1351
1352 mtx_lock(&nlm_global_lock);
1353 TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link);
1354 mtx_unlock(&nlm_global_lock);
1355
1356 free(nw, M_NLM);
1357 }
1358
1359 int
nlm_wait_lock(void * handle,int timo)1360 nlm_wait_lock(void *handle, int timo)
1361 {
1362 struct nlm_waiting_lock *nw = handle;
1363 int error, stops_deferred;
1364
1365 /*
1366 * If the granted message arrived before we got here,
1367 * nw->nw_waiting will be FALSE - in that case, don't sleep.
1368 */
1369 mtx_lock(&nlm_global_lock);
1370 error = 0;
1371 if (nw->nw_waiting) {
1372 stops_deferred = sigdeferstop(SIGDEFERSTOP_ERESTART);
1373 error = msleep(nw, &nlm_global_lock, PCATCH, "nlmlock", timo);
1374 sigallowstop(stops_deferred);
1375 }
1376 TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link);
1377 if (error) {
1378 /*
1379 * The granted message may arrive after the
1380 * interrupt/timeout but before we manage to lock the
1381 * mutex. Detect this by examining nw_lock.
1382 */
1383 if (!nw->nw_waiting)
1384 error = 0;
1385 } else {
1386 /*
1387 * If nlm_cancel_wait is called, then error will be
1388 * zero but nw_waiting will still be TRUE. We
1389 * translate this into EINTR.
1390 */
1391 if (nw->nw_waiting)
1392 error = EINTR;
1393 }
1394 mtx_unlock(&nlm_global_lock);
1395
1396 free(nw, M_NLM);
1397
1398 return (error);
1399 }
1400
1401 void
nlm_cancel_wait(struct vnode * vp)1402 nlm_cancel_wait(struct vnode *vp)
1403 {
1404 struct nlm_waiting_lock *nw;
1405
1406 mtx_lock(&nlm_global_lock);
1407 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
1408 if (nw->nw_vp == vp) {
1409 wakeup(nw);
1410 }
1411 }
1412 mtx_unlock(&nlm_global_lock);
1413 }
1414
1415 /**********************************************************************/
1416
1417 /*
1418 * Syscall interface with userland.
1419 */
1420
1421 extern void nlm_prog_0(struct svc_req *rqstp, SVCXPRT *transp);
1422 extern void nlm_prog_1(struct svc_req *rqstp, SVCXPRT *transp);
1423 extern void nlm_prog_3(struct svc_req *rqstp, SVCXPRT *transp);
1424 extern void nlm_prog_4(struct svc_req *rqstp, SVCXPRT *transp);
1425
1426 static int
nlm_register_services(SVCPOOL * pool,int addr_count,char ** addrs)1427 nlm_register_services(SVCPOOL *pool, int addr_count, char **addrs)
1428 {
1429 static rpcvers_t versions[] = {
1430 NLM_SM, NLM_VERS, NLM_VERSX, NLM_VERS4
1431 };
1432 static void (*dispatchers[])(struct svc_req *, SVCXPRT *) = {
1433 nlm_prog_0, nlm_prog_1, nlm_prog_3, nlm_prog_4
1434 };
1435
1436 SVCXPRT **xprts;
1437 char netid[16];
1438 char uaddr[128];
1439 struct netconfig *nconf;
1440 int i, j, error;
1441
1442 if (!addr_count) {
1443 NLM_ERR("NLM: no service addresses given - can't start server");
1444 return (EINVAL);
1445 }
1446
1447 if (addr_count < 0 || addr_count > 256 ) {
1448 NLM_ERR("NLM: too many service addresses (%d) given, "
1449 "max 256 - can't start server\n", addr_count);
1450 return (EINVAL);
1451 }
1452
1453 xprts = malloc(addr_count * sizeof(SVCXPRT *), M_NLM, M_WAITOK|M_ZERO);
1454 for (i = 0; i < nitems(versions); i++) {
1455 for (j = 0; j < addr_count; j++) {
1456 /*
1457 * Create transports for the first version and
1458 * then just register everything else to the
1459 * same transports.
1460 */
1461 if (i == 0) {
1462 char *up;
1463
1464 error = copyin(&addrs[2*j], &up,
1465 sizeof(char*));
1466 if (error)
1467 goto out;
1468 error = copyinstr(up, netid, sizeof(netid),
1469 NULL);
1470 if (error)
1471 goto out;
1472 error = copyin(&addrs[2*j+1], &up,
1473 sizeof(char*));
1474 if (error)
1475 goto out;
1476 error = copyinstr(up, uaddr, sizeof(uaddr),
1477 NULL);
1478 if (error)
1479 goto out;
1480 nconf = getnetconfigent(netid);
1481 if (!nconf) {
1482 NLM_ERR("Can't lookup netid %s\n",
1483 netid);
1484 error = EINVAL;
1485 goto out;
1486 }
1487 xprts[j] = svc_tp_create(pool, dispatchers[i],
1488 NLM_PROG, versions[i], uaddr, nconf);
1489 if (!xprts[j]) {
1490 NLM_ERR("NLM: unable to create "
1491 "(NLM_PROG, %d).\n", versions[i]);
1492 error = EINVAL;
1493 goto out;
1494 }
1495 freenetconfigent(nconf);
1496 } else {
1497 nconf = getnetconfigent(xprts[j]->xp_netid);
1498 rpcb_unset(NLM_PROG, versions[i], nconf);
1499 if (!svc_reg(xprts[j], NLM_PROG, versions[i],
1500 dispatchers[i], nconf)) {
1501 NLM_ERR("NLM: can't register "
1502 "(NLM_PROG, %d)\n", versions[i]);
1503 error = EINVAL;
1504 goto out;
1505 }
1506 }
1507 }
1508 }
1509 error = 0;
1510 out:
1511 for (j = 0; j < addr_count; j++) {
1512 if (xprts[j])
1513 SVC_RELEASE(xprts[j]);
1514 }
1515 free(xprts, M_NLM);
1516 return (error);
1517 }
1518
1519 /*
1520 * Main server entry point. Contacts the local NSM to get its current
1521 * state and send SM_UNMON_ALL. Registers the NLM services and then
1522 * services requests. Does not return until the server is interrupted
1523 * by a signal.
1524 */
1525 static int
nlm_server_main(int addr_count,char ** addrs)1526 nlm_server_main(int addr_count, char **addrs)
1527 {
1528 struct thread *td = curthread;
1529 int error;
1530 SVCPOOL *pool = NULL;
1531 struct sockopt opt;
1532 int portlow;
1533 #ifdef INET6
1534 struct sockaddr_in6 sin6;
1535 #endif
1536 struct sockaddr_in sin;
1537 my_id id;
1538 sm_stat smstat;
1539 struct timeval timo;
1540 enum clnt_stat stat;
1541 struct nlm_host *host, *nhost;
1542 struct nlm_waiting_lock *nw;
1543 vop_advlock_t *old_nfs_advlock;
1544 vop_reclaim_t *old_nfs_reclaim;
1545
1546 if (nlm_is_running != 0) {
1547 NLM_ERR("NLM: can't start server - "
1548 "it appears to be running already\n");
1549 return (EPERM);
1550 }
1551
1552 if (nlm_socket == NULL) {
1553 memset(&opt, 0, sizeof(opt));
1554
1555 error = socreate(AF_INET, &nlm_socket, SOCK_DGRAM, 0,
1556 td->td_ucred, td);
1557 if (error) {
1558 NLM_ERR("NLM: can't create IPv4 socket - error %d\n",
1559 error);
1560 return (error);
1561 }
1562 opt.sopt_dir = SOPT_SET;
1563 opt.sopt_level = IPPROTO_IP;
1564 opt.sopt_name = IP_PORTRANGE;
1565 portlow = IP_PORTRANGE_LOW;
1566 opt.sopt_val = &portlow;
1567 opt.sopt_valsize = sizeof(portlow);
1568 sosetopt(nlm_socket, &opt);
1569
1570 #ifdef INET6
1571 nlm_socket6 = NULL;
1572 error = socreate(AF_INET6, &nlm_socket6, SOCK_DGRAM, 0,
1573 td->td_ucred, td);
1574 if (error) {
1575 NLM_ERR("NLM: can't create IPv6 socket - error %d\n",
1576 error);
1577 soclose(nlm_socket);
1578 nlm_socket = NULL;
1579 return (error);
1580 }
1581 opt.sopt_dir = SOPT_SET;
1582 opt.sopt_level = IPPROTO_IPV6;
1583 opt.sopt_name = IPV6_PORTRANGE;
1584 portlow = IPV6_PORTRANGE_LOW;
1585 opt.sopt_val = &portlow;
1586 opt.sopt_valsize = sizeof(portlow);
1587 sosetopt(nlm_socket6, &opt);
1588 #endif
1589 }
1590
1591 nlm_auth = authunix_create(curthread->td_ucred);
1592
1593 #ifdef INET6
1594 memset(&sin6, 0, sizeof(sin6));
1595 sin6.sin6_len = sizeof(sin6);
1596 sin6.sin6_family = AF_INET6;
1597 sin6.sin6_addr = in6addr_loopback;
1598 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin6, SM_PROG, SM_VERS);
1599 if (!nlm_nsm) {
1600 #endif
1601 memset(&sin, 0, sizeof(sin));
1602 sin.sin_len = sizeof(sin);
1603 sin.sin_family = AF_INET;
1604 sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1605 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin, SM_PROG,
1606 SM_VERS);
1607 #ifdef INET6
1608 }
1609 #endif
1610
1611 if (!nlm_nsm) {
1612 NLM_ERR("Can't start NLM - unable to contact NSM\n");
1613 error = EINVAL;
1614 goto out;
1615 }
1616
1617 pool = svcpool_create("NLM", NULL);
1618
1619 error = nlm_register_services(pool, addr_count, addrs);
1620 if (error)
1621 goto out;
1622
1623 memset(&id, 0, sizeof(id));
1624 id.my_name = "NFS NLM";
1625
1626 timo.tv_sec = 25;
1627 timo.tv_usec = 0;
1628 stat = CLNT_CALL(nlm_nsm, SM_UNMON_ALL,
1629 (xdrproc_t) xdr_my_id, &id,
1630 (xdrproc_t) xdr_sm_stat, &smstat, timo);
1631
1632 if (stat != RPC_SUCCESS) {
1633 struct rpc_err err;
1634
1635 CLNT_GETERR(nlm_nsm, &err);
1636 NLM_ERR("NLM: unexpected error contacting NSM, "
1637 "stat=%d, errno=%d\n", stat, err.re_errno);
1638 error = EINVAL;
1639 goto out;
1640 }
1641 nlm_is_running = 1;
1642
1643 NLM_DEBUG(1, "NLM: local NSM state is %d\n", smstat.state);
1644 nlm_nsm_state = smstat.state;
1645
1646 old_nfs_advlock = nfs_advlock_p;
1647 nfs_advlock_p = nlm_advlock;
1648 old_nfs_reclaim = nfs_reclaim_p;
1649 nfs_reclaim_p = nlm_reclaim;
1650
1651 svc_run(pool);
1652 error = 0;
1653
1654 nfs_advlock_p = old_nfs_advlock;
1655 nfs_reclaim_p = old_nfs_reclaim;
1656
1657 out:
1658 nlm_is_running = 0;
1659 if (pool)
1660 svcpool_destroy(pool);
1661
1662 /*
1663 * We are finished communicating with the NSM.
1664 */
1665 if (nlm_nsm) {
1666 CLNT_RELEASE(nlm_nsm);
1667 nlm_nsm = NULL;
1668 }
1669
1670 /*
1671 * Trash all the existing state so that if the server
1672 * restarts, it gets a clean slate. This is complicated by the
1673 * possibility that there may be other threads trying to make
1674 * client locking requests.
1675 *
1676 * First we fake a client reboot notification which will
1677 * cancel any pending async locks and purge remote lock state
1678 * from the local lock manager. We release the reference from
1679 * nlm_hosts to the host (which may remove it from the list
1680 * and free it). After this phase, the only entries in the
1681 * nlm_host list should be from other threads performing
1682 * client lock requests.
1683 */
1684 mtx_lock(&nlm_global_lock);
1685 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
1686 wakeup(nw);
1687 }
1688 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) {
1689 mtx_unlock(&nlm_global_lock);
1690 nlm_host_notify(host, 0);
1691 nlm_host_release(host);
1692 mtx_lock(&nlm_global_lock);
1693 }
1694 mtx_unlock(&nlm_global_lock);
1695
1696 AUTH_DESTROY(nlm_auth);
1697
1698 return (error);
1699 }
1700
1701 int
sys_nlm_syscall(struct thread * td,struct nlm_syscall_args * uap)1702 sys_nlm_syscall(struct thread *td, struct nlm_syscall_args *uap)
1703 {
1704 int error;
1705
1706 error = priv_check(td, PRIV_NFS_LOCKD);
1707 if (error)
1708 return (error);
1709
1710 nlm_debug_level = uap->debug_level;
1711 nlm_grace_threshold = time_uptime + uap->grace_period;
1712 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD;
1713
1714 return nlm_server_main(uap->addr_count, uap->addrs);
1715 }
1716
1717 /**********************************************************************/
1718
1719 /*
1720 * NLM implementation details, called from the RPC stubs.
1721 */
1722
1723 void
nlm_sm_notify(struct nlm_sm_status * argp)1724 nlm_sm_notify(struct nlm_sm_status *argp)
1725 {
1726 uint32_t sysid;
1727 struct nlm_host *host;
1728
1729 NLM_DEBUG(3, "nlm_sm_notify(): mon_name = %s\n", argp->mon_name);
1730 memcpy(&sysid, &argp->priv, sizeof(sysid));
1731 host = nlm_find_host_by_sysid(sysid);
1732 if (host) {
1733 nlm_host_notify(host, argp->state);
1734 nlm_host_release(host);
1735 }
1736 }
1737
1738 static void
nlm_convert_to_fhandle_t(fhandle_t * fhp,struct netobj * p)1739 nlm_convert_to_fhandle_t(fhandle_t *fhp, struct netobj *p)
1740 {
1741 memcpy(fhp, p->n_bytes, sizeof(fhandle_t));
1742 }
1743
1744 struct vfs_state {
1745 struct mount *vs_mp;
1746 struct vnode *vs_vp;
1747 int vs_vnlocked;
1748 };
1749
1750 static int
nlm_get_vfs_state(struct nlm_host * host,struct svc_req * rqstp,fhandle_t * fhp,struct vfs_state * vs,accmode_t accmode)1751 nlm_get_vfs_state(struct nlm_host *host, struct svc_req *rqstp,
1752 fhandle_t *fhp, struct vfs_state *vs, accmode_t accmode)
1753 {
1754 int error;
1755 uint64_t exflags;
1756 struct ucred *cred = NULL, *credanon = NULL;
1757
1758 memset(vs, 0, sizeof(*vs));
1759
1760 vs->vs_mp = vfs_getvfs(&fhp->fh_fsid);
1761 if (!vs->vs_mp) {
1762 return (ESTALE);
1763 }
1764
1765 /* accmode == 0 means don't check, since it is an unlock. */
1766 if (accmode != 0) {
1767 error = VFS_CHECKEXP(vs->vs_mp,
1768 (struct sockaddr *)&host->nh_addr, &exflags, &credanon,
1769 NULL, NULL);
1770 if (error)
1771 goto out;
1772
1773 if (exflags & MNT_EXRDONLY ||
1774 (vs->vs_mp->mnt_flag & MNT_RDONLY)) {
1775 error = EROFS;
1776 goto out;
1777 }
1778 }
1779
1780 error = VFS_FHTOVP(vs->vs_mp, &fhp->fh_fid, LK_EXCLUSIVE, &vs->vs_vp);
1781 if (error)
1782 goto out;
1783 vs->vs_vnlocked = TRUE;
1784
1785 if (accmode != 0) {
1786 if (!svc_getcred(rqstp, &cred, NULL)) {
1787 error = EINVAL;
1788 goto out;
1789 }
1790 if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
1791 crfree(cred);
1792 cred = credanon;
1793 credanon = NULL;
1794 }
1795
1796 /*
1797 * Check cred.
1798 */
1799 error = VOP_ACCESS(vs->vs_vp, accmode, cred, curthread);
1800 /*
1801 * If this failed and accmode != VWRITE, try again with
1802 * VWRITE to maintain backwards compatibility with the
1803 * old code that always used VWRITE.
1804 */
1805 if (error != 0 && accmode != VWRITE)
1806 error = VOP_ACCESS(vs->vs_vp, VWRITE, cred, curthread);
1807 if (error)
1808 goto out;
1809 }
1810
1811 VOP_UNLOCK(vs->vs_vp);
1812 vs->vs_vnlocked = FALSE;
1813
1814 out:
1815 if (cred)
1816 crfree(cred);
1817 if (credanon)
1818 crfree(credanon);
1819
1820 return (error);
1821 }
1822
1823 static void
nlm_release_vfs_state(struct vfs_state * vs)1824 nlm_release_vfs_state(struct vfs_state *vs)
1825 {
1826
1827 if (vs->vs_vp) {
1828 if (vs->vs_vnlocked)
1829 vput(vs->vs_vp);
1830 else
1831 vrele(vs->vs_vp);
1832 }
1833 if (vs->vs_mp)
1834 vfs_rel(vs->vs_mp);
1835 }
1836
1837 static nlm4_stats
nlm_convert_error(int error)1838 nlm_convert_error(int error)
1839 {
1840
1841 if (error == ESTALE)
1842 return nlm4_stale_fh;
1843 else if (error == EROFS)
1844 return nlm4_rofs;
1845 else
1846 return nlm4_failed;
1847 }
1848
1849 int
nlm_do_test(nlm4_testargs * argp,nlm4_testres * result,struct svc_req * rqstp,CLIENT ** rpcp)1850 nlm_do_test(nlm4_testargs *argp, nlm4_testres *result, struct svc_req *rqstp,
1851 CLIENT **rpcp)
1852 {
1853 fhandle_t fh;
1854 struct vfs_state vs;
1855 struct nlm_host *host, *bhost;
1856 int error, sysid;
1857 struct flock fl;
1858 accmode_t accmode;
1859
1860 memset(result, 0, sizeof(*result));
1861 memset(&vs, 0, sizeof(vs));
1862
1863 host = nlm_find_host_by_name(argp->alock.caller_name,
1864 svc_getrpccaller(rqstp), rqstp->rq_vers);
1865 if (!host) {
1866 result->stat.stat = nlm4_denied_nolocks;
1867 return (ENOMEM);
1868 }
1869
1870 NLM_DEBUG(3, "nlm_do_test(): caller_name = %s (sysid = %d)\n",
1871 host->nh_caller_name, host->nh_sysid);
1872
1873 nlm_check_expired_locks(host);
1874 sysid = host->nh_sysid;
1875
1876 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
1877 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
1878
1879 if (time_uptime < nlm_grace_threshold) {
1880 result->stat.stat = nlm4_denied_grace_period;
1881 goto out;
1882 }
1883
1884 accmode = argp->exclusive ? VWRITE : VREAD;
1885 error = nlm_get_vfs_state(host, rqstp, &fh, &vs, accmode);
1886 if (error) {
1887 result->stat.stat = nlm_convert_error(error);
1888 goto out;
1889 }
1890
1891 fl.l_start = argp->alock.l_offset;
1892 fl.l_len = argp->alock.l_len;
1893 fl.l_pid = argp->alock.svid;
1894 fl.l_sysid = sysid;
1895 fl.l_whence = SEEK_SET;
1896 if (argp->exclusive)
1897 fl.l_type = F_WRLCK;
1898 else
1899 fl.l_type = F_RDLCK;
1900 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_GETLK, &fl, F_REMOTE);
1901 if (error) {
1902 result->stat.stat = nlm4_failed;
1903 goto out;
1904 }
1905
1906 if (fl.l_type == F_UNLCK) {
1907 result->stat.stat = nlm4_granted;
1908 } else {
1909 result->stat.stat = nlm4_denied;
1910 result->stat.nlm4_testrply_u.holder.exclusive =
1911 (fl.l_type == F_WRLCK);
1912 result->stat.nlm4_testrply_u.holder.svid = fl.l_pid;
1913 bhost = nlm_find_host_by_sysid(fl.l_sysid);
1914 if (bhost) {
1915 /*
1916 * We don't have any useful way of recording
1917 * the value of oh used in the original lock
1918 * request. Ideally, the test reply would have
1919 * a space for the owning host's name allowing
1920 * our caller's NLM to keep track.
1921 *
1922 * As far as I can see, Solaris uses an eight
1923 * byte structure for oh which contains a four
1924 * byte pid encoded in local byte order and
1925 * the first four bytes of the host
1926 * name. Linux uses a variable length string
1927 * 'pid@hostname' in ascii but doesn't even
1928 * return that in test replies.
1929 *
1930 * For the moment, return nothing in oh
1931 * (already zero'ed above).
1932 */
1933 nlm_host_release(bhost);
1934 }
1935 result->stat.nlm4_testrply_u.holder.l_offset = fl.l_start;
1936 result->stat.nlm4_testrply_u.holder.l_len = fl.l_len;
1937 }
1938
1939 out:
1940 nlm_release_vfs_state(&vs);
1941 if (rpcp)
1942 *rpcp = nlm_host_get_rpc(host, TRUE);
1943 nlm_host_release(host);
1944 return (0);
1945 }
1946
1947 int
nlm_do_lock(nlm4_lockargs * argp,nlm4_res * result,struct svc_req * rqstp,bool_t monitor,CLIENT ** rpcp)1948 nlm_do_lock(nlm4_lockargs *argp, nlm4_res *result, struct svc_req *rqstp,
1949 bool_t monitor, CLIENT **rpcp)
1950 {
1951 fhandle_t fh;
1952 struct vfs_state vs;
1953 struct nlm_host *host;
1954 int error, sysid;
1955 struct flock fl;
1956 accmode_t accmode;
1957
1958 memset(result, 0, sizeof(*result));
1959 memset(&vs, 0, sizeof(vs));
1960
1961 host = nlm_find_host_by_name(argp->alock.caller_name,
1962 svc_getrpccaller(rqstp), rqstp->rq_vers);
1963 if (!host) {
1964 result->stat.stat = nlm4_denied_nolocks;
1965 return (ENOMEM);
1966 }
1967
1968 NLM_DEBUG(3, "nlm_do_lock(): caller_name = %s (sysid = %d)\n",
1969 host->nh_caller_name, host->nh_sysid);
1970
1971 if (monitor && host->nh_state && argp->state
1972 && host->nh_state != argp->state) {
1973 /*
1974 * The host rebooted without telling us. Trash its
1975 * locks.
1976 */
1977 nlm_host_notify(host, argp->state);
1978 }
1979
1980 nlm_check_expired_locks(host);
1981 sysid = host->nh_sysid;
1982
1983 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
1984 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
1985
1986 if (time_uptime < nlm_grace_threshold && !argp->reclaim) {
1987 result->stat.stat = nlm4_denied_grace_period;
1988 goto out;
1989 }
1990
1991 accmode = argp->exclusive ? VWRITE : VREAD;
1992 error = nlm_get_vfs_state(host, rqstp, &fh, &vs, accmode);
1993 if (error) {
1994 result->stat.stat = nlm_convert_error(error);
1995 goto out;
1996 }
1997
1998 fl.l_start = argp->alock.l_offset;
1999 fl.l_len = argp->alock.l_len;
2000 fl.l_pid = argp->alock.svid;
2001 fl.l_sysid = sysid;
2002 fl.l_whence = SEEK_SET;
2003 if (argp->exclusive)
2004 fl.l_type = F_WRLCK;
2005 else
2006 fl.l_type = F_RDLCK;
2007 if (argp->block) {
2008 struct nlm_async_lock *af;
2009 CLIENT *client;
2010 struct nlm_grantcookie cookie;
2011
2012 /*
2013 * First, make sure we can contact the host's NLM.
2014 */
2015 client = nlm_host_get_rpc(host, TRUE);
2016 if (!client) {
2017 result->stat.stat = nlm4_failed;
2018 goto out;
2019 }
2020
2021 /*
2022 * First we need to check and see if there is an
2023 * existing blocked lock that matches. This could be a
2024 * badly behaved client or an RPC re-send. If we find
2025 * one, just return nlm4_blocked.
2026 */
2027 mtx_lock(&host->nh_lock);
2028 TAILQ_FOREACH(af, &host->nh_pending, af_link) {
2029 if (af->af_fl.l_start == fl.l_start
2030 && af->af_fl.l_len == fl.l_len
2031 && af->af_fl.l_pid == fl.l_pid
2032 && af->af_fl.l_type == fl.l_type) {
2033 break;
2034 }
2035 }
2036 if (!af) {
2037 cookie.ng_sysid = host->nh_sysid;
2038 cookie.ng_cookie = host->nh_grantcookie++;
2039 }
2040 mtx_unlock(&host->nh_lock);
2041 if (af) {
2042 CLNT_RELEASE(client);
2043 result->stat.stat = nlm4_blocked;
2044 goto out;
2045 }
2046
2047 af = malloc(sizeof(struct nlm_async_lock), M_NLM,
2048 M_WAITOK|M_ZERO);
2049 TASK_INIT(&af->af_task, 0, nlm_lock_callback, af);
2050 af->af_vp = vs.vs_vp;
2051 af->af_fl = fl;
2052 af->af_host = host;
2053 af->af_rpc = client;
2054 /*
2055 * We use M_RPC here so that we can xdr_free the thing
2056 * later.
2057 */
2058 nlm_make_netobj(&af->af_granted.cookie,
2059 (caddr_t)&cookie, sizeof(cookie), M_RPC);
2060 af->af_granted.exclusive = argp->exclusive;
2061 af->af_granted.alock.caller_name =
2062 strdup(argp->alock.caller_name, M_RPC);
2063 nlm_copy_netobj(&af->af_granted.alock.fh,
2064 &argp->alock.fh, M_RPC);
2065 nlm_copy_netobj(&af->af_granted.alock.oh,
2066 &argp->alock.oh, M_RPC);
2067 af->af_granted.alock.svid = argp->alock.svid;
2068 af->af_granted.alock.l_offset = argp->alock.l_offset;
2069 af->af_granted.alock.l_len = argp->alock.l_len;
2070
2071 /*
2072 * Put the entry on the pending list before calling
2073 * VOP_ADVLOCKASYNC. We do this in case the lock
2074 * request was blocked (returning EINPROGRESS) but
2075 * then granted before we manage to run again. The
2076 * client may receive the granted message before we
2077 * send our blocked reply but thats their problem.
2078 */
2079 mtx_lock(&host->nh_lock);
2080 TAILQ_INSERT_TAIL(&host->nh_pending, af, af_link);
2081 mtx_unlock(&host->nh_lock);
2082
2083 error = VOP_ADVLOCKASYNC(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE,
2084 &af->af_task, &af->af_cookie);
2085
2086 /*
2087 * If the lock completed synchronously, just free the
2088 * tracking structure now.
2089 */
2090 if (error != EINPROGRESS) {
2091 CLNT_RELEASE(af->af_rpc);
2092 mtx_lock(&host->nh_lock);
2093 TAILQ_REMOVE(&host->nh_pending, af, af_link);
2094 mtx_unlock(&host->nh_lock);
2095 xdr_free((xdrproc_t) xdr_nlm4_testargs,
2096 &af->af_granted);
2097 free(af, M_NLM);
2098 } else {
2099 NLM_DEBUG(2, "NLM: pending async lock %p for %s "
2100 "(sysid %d)\n", af, host->nh_caller_name, sysid);
2101 /*
2102 * Don't vrele the vnode just yet - this must
2103 * wait until either the async callback
2104 * happens or the lock is cancelled.
2105 */
2106 vs.vs_vp = NULL;
2107 }
2108 } else {
2109 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE);
2110 }
2111
2112 if (error) {
2113 if (error == EINPROGRESS) {
2114 result->stat.stat = nlm4_blocked;
2115 } else if (error == EDEADLK) {
2116 result->stat.stat = nlm4_deadlck;
2117 } else if (error == EAGAIN) {
2118 result->stat.stat = nlm4_denied;
2119 } else {
2120 result->stat.stat = nlm4_failed;
2121 }
2122 } else {
2123 if (monitor)
2124 nlm_host_monitor(host, argp->state);
2125 result->stat.stat = nlm4_granted;
2126 }
2127
2128 out:
2129 nlm_release_vfs_state(&vs);
2130 if (rpcp)
2131 *rpcp = nlm_host_get_rpc(host, TRUE);
2132 nlm_host_release(host);
2133 return (0);
2134 }
2135
2136 int
nlm_do_cancel(nlm4_cancargs * argp,nlm4_res * result,struct svc_req * rqstp,CLIENT ** rpcp)2137 nlm_do_cancel(nlm4_cancargs *argp, nlm4_res *result, struct svc_req *rqstp,
2138 CLIENT **rpcp)
2139 {
2140 fhandle_t fh;
2141 struct vfs_state vs;
2142 struct nlm_host *host;
2143 int error, sysid;
2144 struct flock fl;
2145 struct nlm_async_lock *af;
2146
2147 memset(result, 0, sizeof(*result));
2148 memset(&vs, 0, sizeof(vs));
2149
2150 host = nlm_find_host_by_name(argp->alock.caller_name,
2151 svc_getrpccaller(rqstp), rqstp->rq_vers);
2152 if (!host) {
2153 result->stat.stat = nlm4_denied_nolocks;
2154 return (ENOMEM);
2155 }
2156
2157 NLM_DEBUG(3, "nlm_do_cancel(): caller_name = %s (sysid = %d)\n",
2158 host->nh_caller_name, host->nh_sysid);
2159
2160 nlm_check_expired_locks(host);
2161 sysid = host->nh_sysid;
2162
2163 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
2164 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2165
2166 if (time_uptime < nlm_grace_threshold) {
2167 result->stat.stat = nlm4_denied_grace_period;
2168 goto out;
2169 }
2170
2171 error = nlm_get_vfs_state(host, rqstp, &fh, &vs, (accmode_t)0);
2172 if (error) {
2173 result->stat.stat = nlm_convert_error(error);
2174 goto out;
2175 }
2176
2177 fl.l_start = argp->alock.l_offset;
2178 fl.l_len = argp->alock.l_len;
2179 fl.l_pid = argp->alock.svid;
2180 fl.l_sysid = sysid;
2181 fl.l_whence = SEEK_SET;
2182 if (argp->exclusive)
2183 fl.l_type = F_WRLCK;
2184 else
2185 fl.l_type = F_RDLCK;
2186
2187 /*
2188 * First we need to try and find the async lock request - if
2189 * there isn't one, we give up and return nlm4_denied.
2190 */
2191 mtx_lock(&host->nh_lock);
2192
2193 TAILQ_FOREACH(af, &host->nh_pending, af_link) {
2194 if (af->af_fl.l_start == fl.l_start
2195 && af->af_fl.l_len == fl.l_len
2196 && af->af_fl.l_pid == fl.l_pid
2197 && af->af_fl.l_type == fl.l_type) {
2198 break;
2199 }
2200 }
2201
2202 if (!af) {
2203 mtx_unlock(&host->nh_lock);
2204 result->stat.stat = nlm4_denied;
2205 goto out;
2206 }
2207
2208 error = nlm_cancel_async_lock(af);
2209
2210 if (error) {
2211 result->stat.stat = nlm4_denied;
2212 } else {
2213 result->stat.stat = nlm4_granted;
2214 }
2215
2216 mtx_unlock(&host->nh_lock);
2217
2218 out:
2219 nlm_release_vfs_state(&vs);
2220 if (rpcp)
2221 *rpcp = nlm_host_get_rpc(host, TRUE);
2222 nlm_host_release(host);
2223 return (0);
2224 }
2225
2226 int
nlm_do_unlock(nlm4_unlockargs * argp,nlm4_res * result,struct svc_req * rqstp,CLIENT ** rpcp)2227 nlm_do_unlock(nlm4_unlockargs *argp, nlm4_res *result, struct svc_req *rqstp,
2228 CLIENT **rpcp)
2229 {
2230 fhandle_t fh;
2231 struct vfs_state vs;
2232 struct nlm_host *host;
2233 int error, sysid;
2234 struct flock fl;
2235
2236 memset(result, 0, sizeof(*result));
2237 memset(&vs, 0, sizeof(vs));
2238
2239 host = nlm_find_host_by_name(argp->alock.caller_name,
2240 svc_getrpccaller(rqstp), rqstp->rq_vers);
2241 if (!host) {
2242 result->stat.stat = nlm4_denied_nolocks;
2243 return (ENOMEM);
2244 }
2245
2246 NLM_DEBUG(3, "nlm_do_unlock(): caller_name = %s (sysid = %d)\n",
2247 host->nh_caller_name, host->nh_sysid);
2248
2249 nlm_check_expired_locks(host);
2250 sysid = host->nh_sysid;
2251
2252 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
2253 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2254
2255 if (time_uptime < nlm_grace_threshold) {
2256 result->stat.stat = nlm4_denied_grace_period;
2257 goto out;
2258 }
2259
2260 error = nlm_get_vfs_state(host, rqstp, &fh, &vs, (accmode_t)0);
2261 if (error) {
2262 result->stat.stat = nlm_convert_error(error);
2263 goto out;
2264 }
2265
2266 fl.l_start = argp->alock.l_offset;
2267 fl.l_len = argp->alock.l_len;
2268 fl.l_pid = argp->alock.svid;
2269 fl.l_sysid = sysid;
2270 fl.l_whence = SEEK_SET;
2271 fl.l_type = F_UNLCK;
2272 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_UNLCK, &fl, F_REMOTE);
2273
2274 /*
2275 * Ignore the error - there is no result code for failure,
2276 * only for grace period.
2277 */
2278 result->stat.stat = nlm4_granted;
2279
2280 out:
2281 nlm_release_vfs_state(&vs);
2282 if (rpcp)
2283 *rpcp = nlm_host_get_rpc(host, TRUE);
2284 nlm_host_release(host);
2285 return (0);
2286 }
2287
2288 int
nlm_do_granted(nlm4_testargs * argp,nlm4_res * result,struct svc_req * rqstp,CLIENT ** rpcp)2289 nlm_do_granted(nlm4_testargs *argp, nlm4_res *result, struct svc_req *rqstp,
2290
2291 CLIENT **rpcp)
2292 {
2293 struct nlm_host *host;
2294 struct nlm_waiting_lock *nw;
2295
2296 memset(result, 0, sizeof(*result));
2297
2298 host = nlm_find_host_by_addr(svc_getrpccaller(rqstp), rqstp->rq_vers);
2299 if (!host) {
2300 result->stat.stat = nlm4_denied_nolocks;
2301 return (ENOMEM);
2302 }
2303
2304 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2305 result->stat.stat = nlm4_denied;
2306 KFAIL_POINT_CODE(DEBUG_FP, nlm_deny_grant, goto out);
2307
2308 mtx_lock(&nlm_global_lock);
2309 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
2310 if (!nw->nw_waiting)
2311 continue;
2312 if (argp->alock.svid == nw->nw_lock.svid
2313 && argp->alock.l_offset == nw->nw_lock.l_offset
2314 && argp->alock.l_len == nw->nw_lock.l_len
2315 && argp->alock.fh.n_len == nw->nw_lock.fh.n_len
2316 && !memcmp(argp->alock.fh.n_bytes, nw->nw_lock.fh.n_bytes,
2317 nw->nw_lock.fh.n_len)) {
2318 nw->nw_waiting = FALSE;
2319 wakeup(nw);
2320 result->stat.stat = nlm4_granted;
2321 break;
2322 }
2323 }
2324 mtx_unlock(&nlm_global_lock);
2325
2326 out:
2327 if (rpcp)
2328 *rpcp = nlm_host_get_rpc(host, TRUE);
2329 nlm_host_release(host);
2330 return (0);
2331 }
2332
2333 void
nlm_do_granted_res(nlm4_res * argp,struct svc_req * rqstp)2334 nlm_do_granted_res(nlm4_res *argp, struct svc_req *rqstp)
2335 {
2336 struct nlm_host *host = NULL;
2337 struct nlm_async_lock *af = NULL;
2338 int error;
2339
2340 if (argp->cookie.n_len != sizeof(struct nlm_grantcookie)) {
2341 NLM_DEBUG(1, "NLM: bogus grant cookie");
2342 goto out;
2343 }
2344
2345 host = nlm_find_host_by_sysid(ng_sysid(&argp->cookie));
2346 if (!host) {
2347 NLM_DEBUG(1, "NLM: Unknown host rejected our grant");
2348 goto out;
2349 }
2350
2351 mtx_lock(&host->nh_lock);
2352 TAILQ_FOREACH(af, &host->nh_granted, af_link)
2353 if (ng_cookie(&argp->cookie) ==
2354 ng_cookie(&af->af_granted.cookie))
2355 break;
2356 if (af)
2357 TAILQ_REMOVE(&host->nh_granted, af, af_link);
2358 mtx_unlock(&host->nh_lock);
2359
2360 if (!af) {
2361 NLM_DEBUG(1, "NLM: host %s (sysid %d) replied to our grant "
2362 "with unrecognized cookie %d:%d", host->nh_caller_name,
2363 host->nh_sysid, ng_sysid(&argp->cookie),
2364 ng_cookie(&argp->cookie));
2365 goto out;
2366 }
2367
2368 if (argp->stat.stat != nlm4_granted) {
2369 af->af_fl.l_type = F_UNLCK;
2370 error = VOP_ADVLOCK(af->af_vp, NULL, F_UNLCK, &af->af_fl, F_REMOTE);
2371 if (error) {
2372 NLM_DEBUG(1, "NLM: host %s (sysid %d) rejected our grant "
2373 "and we failed to unlock (%d)", host->nh_caller_name,
2374 host->nh_sysid, error);
2375 goto out;
2376 }
2377
2378 NLM_DEBUG(5, "NLM: async lock %p rejected by host %s (sysid %d)",
2379 af, host->nh_caller_name, host->nh_sysid);
2380 } else {
2381 NLM_DEBUG(5, "NLM: async lock %p accepted by host %s (sysid %d)",
2382 af, host->nh_caller_name, host->nh_sysid);
2383 }
2384
2385 out:
2386 if (af)
2387 nlm_free_async_lock(af);
2388 if (host)
2389 nlm_host_release(host);
2390 }
2391
2392 void
nlm_do_free_all(nlm4_notify * argp)2393 nlm_do_free_all(nlm4_notify *argp)
2394 {
2395 struct nlm_host *host, *thost;
2396
2397 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, thost) {
2398 if (!strcmp(host->nh_caller_name, argp->name))
2399 nlm_host_notify(host, argp->state);
2400 }
2401 }
2402
2403 /*
2404 * Kernel module glue
2405 */
2406 static int
nfslockd_modevent(module_t mod,int type,void * data)2407 nfslockd_modevent(module_t mod, int type, void *data)
2408 {
2409
2410 switch (type) {
2411 case MOD_LOAD:
2412 return (nlm_init());
2413
2414 case MOD_UNLOAD:
2415 nlm_uninit();
2416 /* The NLM module cannot be safely unloaded. */
2417 /* FALLTHROUGH */
2418 default:
2419 return (EOPNOTSUPP);
2420 }
2421 }
2422 static moduledata_t nfslockd_mod = {
2423 "nfslockd",
2424 nfslockd_modevent,
2425 NULL,
2426 };
2427 DECLARE_MODULE(nfslockd, nfslockd_mod, SI_SUB_VFS, SI_ORDER_ANY);
2428
2429 /* So that loader and kldload(2) can find us, wherever we are.. */
2430 MODULE_DEPEND(nfslockd, xdr, 1, 1, 1);
2431 MODULE_DEPEND(nfslockd, krpc, 1, 1, 1);
2432 MODULE_DEPEND(nfslockd, nfscommon, 1, 1, 1);
2433 MODULE_VERSION(nfslockd, 1);
2434