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/fcntl.h>
35 #include <sys/kernel.h>
36 #include <sys/lockf.h>
37 #include <sys/malloc.h>
38 #include <sys/mount.h>
39 #if __FreeBSD_version >= 700000
40 #include <sys/priv.h>
41 #endif
42 #include <sys/proc.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/syscall.h>
46 #include <sys/sysctl.h>
47 #include <sys/sysent.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 <nlm/nlm_prot.h>
55 #include <nlm/sm_inter.h>
56 #include <nlm/nlm.h>
57 #include <rpc/rpc_com.h>
58 #include <rpc/rpcb_prot.h>
59
60 MALLOC_DEFINE(M_NLM, "NLM", "Network Lock Manager");
61
62 /*
63 * If a host is inactive (and holds no locks) for this amount of
64 * seconds, we consider it idle and stop tracking it.
65 */
66 #define NLM_IDLE_TIMEOUT 30
67
68 /*
69 * We check the host list for idle every few seconds.
70 */
71 #define NLM_IDLE_PERIOD 5
72
73 /*
74 * Support for sysctl vfs.nlm.sysid
75 */
76 SYSCTL_NODE(_vfs, OID_AUTO, nlm, CTLFLAG_RW, NULL, "Network Lock Manager");
77 SYSCTL_NODE(_vfs_nlm, OID_AUTO, sysid, CTLFLAG_RW, NULL, "");
78
79 /*
80 * Syscall hooks
81 */
82 static int nlm_syscall_offset = SYS_nlm_syscall;
83 static struct sysent nlm_syscall_prev_sysent;
84 #if __FreeBSD_version < 700000
85 static struct sysent nlm_syscall_sysent = {
86 (sizeof(struct nlm_syscall_args) / sizeof(register_t)) | SYF_MPSAFE,
87 (sy_call_t *) nlm_syscall
88 };
89 #else
90 MAKE_SYSENT(nlm_syscall);
91 #endif
92 static bool_t nlm_syscall_registered = FALSE;
93
94 /*
95 * Debug level passed in from userland. We also support a sysctl hook
96 * so that it can be changed on a live system.
97 */
98 static int nlm_debug_level;
99 SYSCTL_INT(_debug, OID_AUTO, nlm_debug, CTLFLAG_RW, &nlm_debug_level, 0, "");
100
101 /*
102 * Grace period handling. The value of nlm_grace_threshold is the
103 * value of time_uptime after which we are serving requests normally.
104 */
105 static time_t nlm_grace_threshold;
106
107 /*
108 * We check for idle hosts if time_uptime is greater than
109 * nlm_next_idle_check,
110 */
111 static time_t nlm_next_idle_check;
112
113 /*
114 * A socket to use for RPC - shared by all IPv4 RPC clients.
115 */
116 static struct socket *nlm_socket;
117
118 #ifdef INET6
119
120 /*
121 * A socket to use for RPC - shared by all IPv6 RPC clients.
122 */
123 static struct socket *nlm_socket6;
124
125 #endif
126
127 /*
128 * An RPC client handle that can be used to communicate with the local
129 * NSM.
130 */
131 static CLIENT *nlm_nsm;
132
133 /*
134 * An RPC client handle that can be used to communicate with the
135 * userland part of lockd.
136 */
137 static CLIENT *nlm_lockd;
138
139 /*
140 * Locks:
141 * (l) locked by nh_lock
142 * (s) only accessed via server RPC which is single threaded
143 * (c) const until freeing
144 */
145
146 /*
147 * A pending asynchronous lock request, stored on the nh_pending list
148 * of the NLM host.
149 */
150 struct nlm_async_lock {
151 TAILQ_ENTRY(nlm_async_lock) af_link; /* (l) host's list of locks */
152 struct task af_task; /* (c) async callback details */
153 void *af_cookie; /* (l) lock manager cancel token */
154 struct vnode *af_vp; /* (l) vnode to lock */
155 struct flock af_fl; /* (c) lock details */
156 struct nlm_host *af_host; /* (c) host which is locking */
157 nlm4_testargs af_granted; /* (c) notification details */
158 };
159 TAILQ_HEAD(nlm_async_lock_list, nlm_async_lock);
160
161 /*
162 * NLM host.
163 */
164 enum nlm_host_state {
165 NLM_UNMONITORED,
166 NLM_MONITORED,
167 NLM_MONITOR_FAILED
168 };
169 struct nlm_host {
170 struct mtx nh_lock;
171 TAILQ_ENTRY(nlm_host) nh_link; /* (s) global list of hosts */
172 char *nh_caller_name; /* (c) printable name of host */
173 uint32_t nh_sysid; /* (c) our allocaed system ID */
174 char nh_sysid_string[10]; /* (c) string rep. of sysid */
175 struct sockaddr_storage nh_addr; /* (s) remote address of host */
176 CLIENT *nh_rpc; /* (s) RPC handle to send to host */
177 rpcvers_t nh_vers; /* (s) NLM version of host */
178 int nh_state; /* (s) last seen NSM state of host */
179 enum nlm_host_state nh_monstate; /* (s) local NSM monitoring state */
180 time_t nh_idle_timeout; /* (s) Time at which host is idle */
181 time_t nh_rpc_create_time; /* (s) Time we create RPC client */
182 struct sysctl_ctx_list nh_sysctl; /* (c) vfs.nlm.sysid nodes */
183 struct nlm_async_lock_list nh_pending; /* (l) pending async locks */
184 struct nlm_async_lock_list nh_finished; /* (l) finished async locks */
185 };
186 TAILQ_HEAD(nlm_host_list, nlm_host);
187
188 static struct nlm_host_list nlm_hosts;
189 static uint32_t nlm_next_sysid = 1;
190
191 static void nlm_host_unmonitor(struct nlm_host *);
192
193 /**********************************************************************/
194
195 /*
196 * Initialise NLM globals.
197 */
198 static void
199 nlm_init(void *dummy)
200 {
201 int error;
202
203 TAILQ_INIT(&nlm_hosts);
204
205 error = syscall_register(&nlm_syscall_offset, &nlm_syscall_sysent,
206 &nlm_syscall_prev_sysent);
207 if (error)
208 printf("Can't register NLM syscall\n");
209 else
210 nlm_syscall_registered = TRUE;
211 }
212 SYSINIT(nlm_init, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_init, NULL);
213
214 static void
215 nlm_uninit(void *dummy)
216 {
217
218 if (nlm_syscall_registered)
219 syscall_deregister(&nlm_syscall_offset,
220 &nlm_syscall_prev_sysent);
221 }
222 SYSUNINIT(nlm_uninit, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_uninit, NULL);
223
224 /*
225 * Copy a struct netobj.
226 */
227 void
228 nlm_copy_netobj(struct netobj *dst, struct netobj *src,
229 struct malloc_type *type)
230 {
231
232 dst->n_len = src->n_len;
233 dst->n_bytes = malloc(src->n_len, type, M_WAITOK);
234 memcpy(dst->n_bytes, src->n_bytes, src->n_len);
235 }
236
237 /*
238 * Create an RPC client handle for the given (address,prog,vers)
239 * triple using UDP.
240 */
241 static CLIENT *
242 nlm_get_rpc(struct sockaddr *sa, rpcprog_t prog, rpcvers_t vers)
243 {
244 const char *wchan = "nlmrcv";
245 const char* protofmly;
246 struct sockaddr_storage ss;
247 struct socket *so;
248 CLIENT *rpcb;
249 struct timeval timo;
250 RPCB parms;
251 char *uaddr;
252 enum clnt_stat stat;
253 int rpcvers;
254
255 /*
256 * First we need to contact the remote RPCBIND service to find
257 * the right port.
258 */
259 memcpy(&ss, sa, sa->sa_len);
260 switch (ss.ss_family) {
261 case AF_INET:
262 ((struct sockaddr_in *)&ss)->sin_port = htons(111);
263 protofmly = "inet";
264 so = nlm_socket;
265 break;
266
267 #ifdef INET6
268 case AF_INET6:
269 ((struct sockaddr_in6 *)&ss)->sin6_port = htons(111);
270 protofmly = "inet6";
271 so = nlm_socket6;
272 break;
273 #endif
274
275 default:
276 /*
277 * Unsupported address family - fail.
278 */
279 return (NULL);
280 }
281
282 rpcb = clnt_dg_create(so, (struct sockaddr *)&ss,
283 RPCBPROG, RPCBVERS4, 0, 0);
284 if (!rpcb)
285 return (NULL);
286
287 parms.r_prog = prog;
288 parms.r_vers = vers;
289 parms.r_netid = "udp";
290 parms.r_addr = "";
291 parms.r_owner = "";
292
293 /*
294 * Use the default timeout.
295 */
296 timo.tv_sec = 25;
297 timo.tv_usec = 0;
298 again:
299 uaddr = NULL;
300 stat = CLNT_CALL(rpcb, (rpcprog_t) RPCBPROC_GETADDR,
301 (xdrproc_t) xdr_rpcb, &parms,
302 (xdrproc_t) xdr_wrapstring, &uaddr, timo);
303 if (stat == RPC_PROGVERSMISMATCH) {
304 /*
305 * Try RPCBIND version 3 if we haven't already.
306 *
307 * XXX fall back to portmap?
308 */
309 CLNT_CONTROL(rpcb, CLGET_VERS, &rpcvers);
310 if (rpcvers == RPCBVERS4) {
311 rpcvers = RPCBVERS;
312 CLNT_CONTROL(rpcb, CLSET_VERS, &rpcvers);
313 goto again;
314 }
315 }
316
317 if (stat == RPC_SUCCESS) {
318 /*
319 * We have a reply from the remote RPCBIND - turn it into an
320 * appropriate address and make a new client that can talk to
321 * the remote NLM.
322 *
323 * XXX fixup IPv6 scope ID.
324 */
325 struct netbuf *a;
326 a = __rpc_uaddr2taddr_af(ss.ss_family, uaddr);
327 if (!a) {
328 CLNT_DESTROY(rpcb);
329 return (NULL);
330 }
331 memcpy(&ss, a->buf, a->len);
332 free(a->buf, M_RPC);
333 free(a, M_RPC);
334 xdr_free((xdrproc_t) xdr_wrapstring, &uaddr);
335 } else if (stat == RPC_PROGVERSMISMATCH) {
336 /*
337 * Try portmap.
338 */
339 struct pmap mapping;
340 u_short port;
341
342 rpcvers = PMAPVERS;
343 CLNT_CONTROL(rpcb, CLSET_VERS, &rpcvers);
344
345 mapping.pm_prog = parms.r_prog;
346 mapping.pm_vers = parms.r_vers;
347 mapping.pm_prot = IPPROTO_UDP;
348 mapping.pm_port = 0;
349
350 stat = CLNT_CALL(rpcb, (rpcprog_t) PMAPPROC_GETPORT,
351 (xdrproc_t) xdr_pmap, &mapping,
352 (xdrproc_t) xdr_u_short, &port, timo);
353
354 if (stat == RPC_SUCCESS) {
355 switch (ss.ss_family) {
356 case AF_INET:
357 ((struct sockaddr_in *)&ss)->sin_port =
358 htons(port);
359 break;
360
361 #ifdef INET6
362 case AF_INET6:
363 ((struct sockaddr_in6 *)&ss)->sin6_port =
364 htons(port);
365 break;
366 #endif
367 }
368 }
369 }
370 if (stat != RPC_SUCCESS) {
371 printf("NLM: failed to contact remote rpcbind, stat = %d\n",
372 (int) stat);
373 CLNT_DESTROY(rpcb);
374 return (NULL);
375 }
376
377 /*
378 * Re-use the client we used to speak to rpcbind.
379 */
380 CLNT_CONTROL(rpcb, CLSET_SVC_ADDR, &ss);
381 CLNT_CONTROL(rpcb, CLSET_PROG, &prog);
382 CLNT_CONTROL(rpcb, CLSET_VERS, &vers);
383 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, &wchan);
384 rpcb->cl_auth = authunix_create(curthread->td_ucred);
385
386 return (rpcb);
387 }
388
389 /*
390 * This async callback after when an async lock request has been
391 * granted. We notify the host which initiated the request.
392 */
393 static void
394 nlm_lock_callback(void *arg, int pending)
395 {
396 struct nlm_async_lock *af = (struct nlm_async_lock *) arg;
397
398 if (nlm_debug_level >= 2)
399 printf("NLM: async lock %p for %s (sysid %d) granted\n",
400 af, af->af_host->nh_caller_name,
401 af->af_host->nh_sysid);
402
403 /*
404 * Send the results back to the host.
405 *
406 * Note: there is a possible race here with nlm_host_notify
407 * destroying the RPC client. To avoid problems, the first
408 * thing nlm_host_notify does is to cancel pending async lock
409 * requests.
410 */
411 if (af->af_host->nh_vers == NLM_VERS4) {
412 nlm4_granted_msg_4(&af->af_granted,
413 NULL, af->af_host->nh_rpc);
414 } else {
415 /*
416 * Back-convert to legacy protocol
417 */
418 nlm_testargs granted;
419 granted.cookie = af->af_granted.cookie;
420 granted.exclusive = af->af_granted.exclusive;
421 granted.alock.caller_name =
422 af->af_granted.alock.caller_name;
423 granted.alock.fh = af->af_granted.alock.fh;
424 granted.alock.oh = af->af_granted.alock.oh;
425 granted.alock.svid = af->af_granted.alock.svid;
426 granted.alock.l_offset =
427 af->af_granted.alock.l_offset;
428 granted.alock.l_len =
429 af->af_granted.alock.l_len;
430
431 nlm_granted_msg_1(&granted,
432 NULL, af->af_host->nh_rpc);
433 }
434
435 /*
436 * Move this entry to the nh_finished list. Someone else will
437 * free it later - its too hard to do it here safely without
438 * racing with cancel.
439 *
440 * XXX possibly we should have a third "granted sent but not
441 * ack'ed" list so that we can re-send the granted message.
442 */
443 mtx_lock(&af->af_host->nh_lock);
444 TAILQ_REMOVE(&af->af_host->nh_pending, af, af_link);
445 TAILQ_INSERT_TAIL(&af->af_host->nh_finished, af, af_link);
446 mtx_unlock(&af->af_host->nh_lock);
447 }
448
449 /*
450 * Free an async lock request. The request must have been removed from
451 * any list.
452 */
453 static void
454 nlm_free_async_lock(struct nlm_async_lock *af)
455 {
456 /*
457 * Free an async lock.
458 */
459 xdr_free((xdrproc_t) xdr_nlm4_testargs, &af->af_granted);
460 if (af->af_vp)
461 vrele(af->af_vp);
462 free(af, M_NLM);
463 }
464
465 /*
466 * Cancel our async request - this must be called with
467 * af->nh_host->nh_lock held. This is slightly complicated by a
468 * potential race with our own callback. If we fail to cancel the
469 * lock, it must already have been granted - we make sure our async
470 * task has completed by calling taskqueue_drain in this case.
471 */
472 static int
473 nlm_cancel_async_lock(struct nlm_async_lock *af)
474 {
475 struct nlm_host *host = af->af_host;
476 int error;
477
478 mtx_assert(&host->nh_lock, MA_OWNED);
479
480 mtx_unlock(&host->nh_lock);
481
482 error = VOP_ADVLOCKASYNC(af->af_vp, NULL, F_CANCEL, &af->af_fl,
483 F_REMOTE, NULL, &af->af_cookie);
484
485 if (error) {
486 /*
487 * We failed to cancel - make sure our callback has
488 * completed before we continue.
489 */
490 taskqueue_drain(taskqueue_thread, &af->af_task);
491 }
492
493 mtx_lock(&host->nh_lock);
494
495 if (!error) {
496 if (nlm_debug_level >= 2)
497 printf("NLM: async lock %p for %s (sysid %d) "
498 "cancelled\n",
499 af, host->nh_caller_name, host->nh_sysid);
500
501 /*
502 * Remove from the nh_pending list and free now that
503 * we are safe from the callback.
504 */
505 TAILQ_REMOVE(&host->nh_pending, af, af_link);
506 mtx_unlock(&host->nh_lock);
507 nlm_free_async_lock(af);
508 mtx_lock(&host->nh_lock);
509 }
510
511 return (error);
512 }
513
514 static void
515 nlm_free_finished_locks(struct nlm_host *host)
516 {
517 struct nlm_async_lock *af;
518
519 mtx_lock(&host->nh_lock);
520 while ((af = TAILQ_FIRST(&host->nh_finished)) != NULL) {
521 TAILQ_REMOVE(&host->nh_finished, af, af_link);
522 mtx_unlock(&host->nh_lock);
523 nlm_free_async_lock(af);
524 mtx_lock(&host->nh_lock);
525 }
526 mtx_unlock(&host->nh_lock);
527 }
528
529 /*
530 * This is called when we receive a host state change
531 * notification. We unlock any active locks owned by the host.
532 */
533 static void
534 nlm_host_notify(struct nlm_host *host, int newstate, bool_t destroy)
535 {
536 struct nlm_async_lock *af;
537
538 if (newstate) {
539 if (nlm_debug_level >= 1)
540 printf("NLM: host %s (sysid %d) rebooted, new "
541 "state is %d\n",
542 host->nh_caller_name, host->nh_sysid, newstate);
543 }
544
545 /*
546 * Cancel any pending async locks for this host.
547 */
548 mtx_lock(&host->nh_lock);
549 while ((af = TAILQ_FIRST(&host->nh_pending)) != NULL) {
550 /*
551 * nlm_cancel_async_lock will remove the entry from
552 * nh_pending and free it.
553 */
554 nlm_cancel_async_lock(af);
555 }
556 mtx_unlock(&host->nh_lock);
557 nlm_free_finished_locks(host);
558
559 /*
560 * The host just rebooted - trash its locks and forget any
561 * RPC client handle that we may have for it.
562 */
563 lf_clearremotesys(host->nh_sysid);
564 if (host->nh_rpc) {
565 AUTH_DESTROY(host->nh_rpc->cl_auth);
566 CLNT_DESTROY(host->nh_rpc);
567 host->nh_rpc = NULL;
568 }
569 host->nh_state = newstate;
570
571 /*
572 * Destroy the host if the caller believes that it won't be
573 * used again. This is safe enough - if we see the same name
574 * again, we will just create a new host.
575 */
576 if (destroy) {
577 TAILQ_REMOVE(&nlm_hosts, host, nh_link);
578 mtx_destroy(&host->nh_lock);
579 sysctl_ctx_free(&host->nh_sysctl);
580 free(host->nh_caller_name, M_NLM);
581 free(host, M_NLM);
582 }
583 }
584
585 /*
586 * Sysctl handler to count the number of locks for a sysid.
587 */
588 static int
589 nlm_host_lock_count_sysctl(SYSCTL_HANDLER_ARGS)
590 {
591 struct nlm_host *host;
592 int count;
593
594 host = oidp->oid_arg1;
595 count = lf_countlocks(host->nh_sysid);
596 return sysctl_handle_int(oidp, &count, 0, req);
597 }
598
599 /*
600 * Create a new NLM host.
601 */
602 static struct nlm_host *
603 nlm_create_host(const char* caller_name)
604 {
605 struct nlm_host *host;
606 struct sysctl_oid *oid;
607
608 if (nlm_debug_level >= 1)
609 printf("NLM: new host %s (sysid %d)\n",
610 caller_name, nlm_next_sysid);
611 host = malloc(sizeof(struct nlm_host), M_NLM, M_WAITOK|M_ZERO);
612 mtx_init(&host->nh_lock, "nh_lock", NULL, MTX_DEF);
613 host->nh_caller_name = strdup(caller_name, M_NLM);
614 host->nh_sysid = nlm_next_sysid++;
615 snprintf(host->nh_sysid_string, sizeof(host->nh_sysid_string),
616 "%d", host->nh_sysid);
617 host->nh_rpc = NULL;
618 host->nh_vers = 0;
619 host->nh_state = 0;
620 host->nh_monstate = NLM_UNMONITORED;
621 TAILQ_INIT(&host->nh_pending);
622 TAILQ_INIT(&host->nh_finished);
623 TAILQ_INSERT_TAIL(&nlm_hosts, host, nh_link);
624
625 sysctl_ctx_init(&host->nh_sysctl);
626 oid = SYSCTL_ADD_NODE(&host->nh_sysctl,
627 SYSCTL_STATIC_CHILDREN(_vfs_nlm_sysid),
628 OID_AUTO, host->nh_sysid_string, CTLFLAG_RD, NULL, "");
629 SYSCTL_ADD_STRING(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
630 "hostname", CTLFLAG_RD, host->nh_caller_name, 0, "");
631 SYSCTL_ADD_INT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
632 "version", CTLFLAG_RD, &host->nh_vers, 0, "");
633 SYSCTL_ADD_INT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
634 "monitored", CTLFLAG_RD, &host->nh_monstate, 0, "");
635 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
636 "lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0,
637 nlm_host_lock_count_sysctl, "I", "");
638
639 return (host);
640 }
641
642 /*
643 * Return non-zero if the address parts of the two sockaddrs are the
644 * same.
645 */
646 static int
647 nlm_compare_addr(const struct sockaddr *a, const struct sockaddr *b)
648 {
649 const struct sockaddr_in *a4, *b4;
650 #ifdef INET6
651 const struct sockaddr_in6 *a6, *b6;
652 #endif
653
654 if (a->sa_family != b->sa_family)
655 return (FALSE);
656
657 switch (a->sa_family) {
658 case AF_INET:
659 a4 = (const struct sockaddr_in *) a;
660 b4 = (const struct sockaddr_in *) b;
661 return !memcmp(&a4->sin_addr, &b4->sin_addr,
662 sizeof(a4->sin_addr));
663 #ifdef INET6
664 case AF_INET6:
665 a6 = (const struct sockaddr_in6 *) a;
666 b6 = (const struct sockaddr_in6 *) b;
667 return !memcmp(&a6->sin6_addr, &b6->sin6_addr,
668 sizeof(a6->sin6_addr));
669 #endif
670 }
671
672 return (0);
673 }
674
675 /*
676 * Check for idle hosts and stop monitoring them. We could also free
677 * the host structure here, possibly after a larger timeout but that
678 * would require some care to avoid races with
679 * e.g. nlm_host_lock_count_sysctl.
680 */
681 static void
682 nlm_check_idle(void)
683 {
684 struct nlm_host *host;
685
686 if (time_uptime <= nlm_next_idle_check)
687 return;
688
689 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD;
690
691 TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
692 if (host->nh_monstate == NLM_MONITORED
693 && time_uptime > host->nh_idle_timeout) {
694 if (lf_countlocks(host->nh_sysid) > 0) {
695 host->nh_idle_timeout =
696 time_uptime + NLM_IDLE_TIMEOUT;
697 continue;
698 }
699 nlm_host_unmonitor(host);
700 }
701 }
702 }
703
704 /*
705 * Search for an existing NLM host that matches the given name
706 * (typically the caller_name element of an nlm4_lock). If none is
707 * found, create a new host. If 'rqstp' is non-NULL, record the remote
708 * address of the host so that we can call it back for async
709 * responses.
710 */
711 struct nlm_host *
712 nlm_find_host_by_name(const char *name, struct svc_req *rqstp)
713 {
714 struct nlm_host *host;
715
716 nlm_check_idle();
717
718 /*
719 * The remote host is determined by caller_name.
720 */
721 TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
722 if (!strcmp(host->nh_caller_name, name))
723 break;
724 }
725
726 if (!host)
727 host = nlm_create_host(name);
728 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT;
729
730 /*
731 * If we have an RPC request, record the remote address so
732 * that can send async replies etc.
733 */
734 if (rqstp) {
735 struct netbuf *addr = &rqstp->rq_xprt->xp_rtaddr;
736
737 KASSERT(addr->len < sizeof(struct sockaddr_storage),
738 ("Strange remote transport address length"));
739
740 /*
741 * If we have seen an address before and we currently
742 * have an RPC client handle, make sure the address is
743 * the same, otherwise discard the client handle.
744 */
745 if (host->nh_addr.ss_len && host->nh_rpc) {
746 if (!nlm_compare_addr(
747 (struct sockaddr *) &host->nh_addr,
748 (struct sockaddr *) addr->buf)
749 || host->nh_vers != rqstp->rq_vers) {
750 AUTH_DESTROY(host->nh_rpc->cl_auth);
751 CLNT_DESTROY(host->nh_rpc);
752 host->nh_rpc = NULL;
753 }
754 }
755 memcpy(&host->nh_addr, addr->buf, addr->len);
756 host->nh_vers = rqstp->rq_vers;
757 }
758
759 return (host);
760 }
761
762 /*
763 * Search for an existing NLM host that matches the given remote
764 * address. If none is found, create a new host with the requested
765 * address and remember 'vers' as the NLM protocol version to use for
766 * that host.
767 */
768 struct nlm_host *
769 nlm_find_host_by_addr(const struct sockaddr *addr, int vers)
770 {
771 struct nlm_host *host;
772
773 nlm_check_idle();
774
775 /*
776 * The remote host is determined by caller_name.
777 */
778 TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
779 if (nlm_compare_addr(addr,
780 (const struct sockaddr *) &host->nh_addr))
781 break;
782 }
783
784 if (!host) {
785 /*
786 * Fake up a name using inet_ntop. This buffer is
787 * large enough for an IPv6 address.
788 */
789 char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"];
790 switch (addr->sa_family) {
791 case AF_INET:
792 __rpc_inet_ntop(AF_INET,
793 &((const struct sockaddr_in *) addr)->sin_addr,
794 tmp, sizeof tmp);
795 break;
796 #ifdef INET6
797 case AF_INET6:
798 __rpc_inet_ntop(AF_INET6,
799 &((const struct sockaddr_in6 *) addr)->sin6_addr,
800 tmp, sizeof tmp);
801 break;
802 #endif
803 default:
804 strcmp(tmp, "<unknown>");
805 }
806 host = nlm_create_host(tmp);
807 memcpy(&host->nh_addr, addr, addr->sa_len);
808 host->nh_vers = vers;
809 }
810 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT;
811
812 return (host);
813 }
814
815 /*
816 * Find the NLM host that matches the value of 'sysid'. If none
817 * exists, return NULL.
818 */
819 static struct nlm_host *
820 nlm_find_host_by_sysid(int sysid)
821 {
822 struct nlm_host *host;
823
824 TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
825 if (host->nh_sysid == sysid)
826 return (host);
827 }
828
829 return (NULL);
830 }
831
832 /*
833 * Unregister this NLM host with the local NSM due to idleness.
834 */
835 static void
836 nlm_host_unmonitor(struct nlm_host *host)
837 {
838 mon_id smmonid;
839 sm_stat_res smstat;
840 struct timeval timo;
841 enum clnt_stat stat;
842
843 if (nlm_debug_level >= 1)
844 printf("NLM: unmonitoring %s (sysid %d)\n",
845 host->nh_caller_name, host->nh_sysid);
846
847 /*
848 * We put our assigned system ID value in the priv field to
849 * make it simpler to find the host if we are notified of a
850 * host restart.
851 */
852 smmonid.mon_name = host->nh_caller_name;
853 smmonid.my_id.my_name = "localhost";
854 smmonid.my_id.my_prog = NLM_PROG;
855 smmonid.my_id.my_vers = NLM_SM;
856 smmonid.my_id.my_proc = NLM_SM_NOTIFY;
857
858 timo.tv_sec = 25;
859 timo.tv_usec = 0;
860 stat = CLNT_CALL(nlm_nsm, SM_UNMON,
861 (xdrproc_t) xdr_mon, &smmonid,
862 (xdrproc_t) xdr_sm_stat, &smstat, timo);
863
864 if (stat != RPC_SUCCESS) {
865 printf("Failed to contact local NSM - rpc error %d\n", stat);
866 return;
867 }
868 if (smstat.res_stat == stat_fail) {
869 printf("Local NSM refuses to unmonitor %s\n",
870 host->nh_caller_name);
871 return;
872 }
873
874 host->nh_monstate = NLM_UNMONITORED;
875 }
876
877 /*
878 * Register this NLM host with the local NSM so that we can be
879 * notified if it reboots.
880 */
881 static void
882 nlm_host_monitor(struct nlm_host *host, int state)
883 {
884 mon smmon;
885 sm_stat_res smstat;
886 struct timeval timo;
887 enum clnt_stat stat;
888
889 if (state && !host->nh_state) {
890 /*
891 * This is the first time we have seen an NSM state
892 * value for this host. We record it here to help
893 * detect host reboots.
894 */
895 host->nh_state = state;
896 if (nlm_debug_level >= 1)
897 printf("NLM: host %s (sysid %d) has NSM state %d\n",
898 host->nh_caller_name, host->nh_sysid, state);
899 }
900
901 if (host->nh_monstate != NLM_UNMONITORED)
902 return;
903
904 if (nlm_debug_level >= 1)
905 printf("NLM: monitoring %s (sysid %d)\n",
906 host->nh_caller_name, host->nh_sysid);
907
908 /*
909 * We put our assigned system ID value in the priv field to
910 * make it simpler to find the host if we are notified of a
911 * host restart.
912 */
913 smmon.mon_id.mon_name = host->nh_caller_name;
914 smmon.mon_id.my_id.my_name = "localhost";
915 smmon.mon_id.my_id.my_prog = NLM_PROG;
916 smmon.mon_id.my_id.my_vers = NLM_SM;
917 smmon.mon_id.my_id.my_proc = NLM_SM_NOTIFY;
918 memcpy(smmon.priv, &host->nh_sysid, sizeof(host->nh_sysid));
919
920 timo.tv_sec = 25;
921 timo.tv_usec = 0;
922 stat = CLNT_CALL(nlm_nsm, SM_MON,
923 (xdrproc_t) xdr_mon, &smmon,
924 (xdrproc_t) xdr_sm_stat, &smstat, timo);
925
926 if (stat != RPC_SUCCESS) {
927 printf("Failed to contact local NSM - rpc error %d\n", stat);
928 return;
929 }
930 if (smstat.res_stat == stat_fail) {
931 printf("Local NSM refuses to monitor %s\n",
932 host->nh_caller_name);
933 host->nh_monstate = NLM_MONITOR_FAILED;
934 return;
935 }
936
937 host->nh_monstate = NLM_MONITORED;
938 }
939
940 /*
941 * Return an RPC client handle that can be used to talk to the NLM
942 * running on the given host.
943 */
944 CLIENT *
945 nlm_host_get_rpc(struct nlm_host *host)
946 {
947 struct timeval zero;
948
949 /*
950 * We can't hold onto RPC handles for too long - the async
951 * call/reply protocol used by some NLM clients makes it hard
952 * to tell when they change port numbers (e.g. after a
953 * reboot). Note that if a client reboots while it isn't
954 * holding any locks, it won't bother to notify us. We
955 * expire the RPC handles after two minutes.
956 */
957 if (host->nh_rpc && time_uptime > host->nh_rpc_create_time + 2*60) {
958 CLIENT *client;
959 client = host->nh_rpc;
960 host->nh_rpc = NULL;
961 CLNT_DESTROY(client);
962 }
963
964 if (host->nh_rpc)
965 return (host->nh_rpc);
966
967 /*
968 * Set the send timeout to zero - we only use this rpc handle
969 * for sending async replies which have no return value.
970 */
971 host->nh_rpc = nlm_get_rpc((struct sockaddr *)&host->nh_addr,
972 NLM_PROG, host->nh_vers);
973
974 if (host->nh_rpc) {
975 zero.tv_sec = 0;
976 zero.tv_usec = 0;
977 CLNT_CONTROL(host->nh_rpc, CLSET_TIMEOUT, &zero);
978
979 host->nh_rpc_create_time = time_uptime;
980 }
981
982 return (host->nh_rpc);
983 }
984
985 /**********************************************************************/
986
987 /*
988 * Syscall interface with userland.
989 */
990
991 extern void nlm_prog_0(struct svc_req *rqstp, SVCXPRT *transp);
992 extern void nlm_prog_1(struct svc_req *rqstp, SVCXPRT *transp);
993 extern void nlm_prog_3(struct svc_req *rqstp, SVCXPRT *transp);
994 extern void nlm_prog_4(struct svc_req *rqstp, SVCXPRT *transp);
995
996 static int
997 nlm_register_services(SVCPOOL *pool, int addr_count, char **addrs)
998 {
999 static rpcvers_t versions[] = {
1000 NLM_SM, NLM_VERS, NLM_VERSX, NLM_VERS4
1001 };
1002 static void (*dispatchers[])(struct svc_req *, SVCXPRT *) = {
1003 nlm_prog_0, nlm_prog_1, nlm_prog_3, nlm_prog_4
1004 };
1005 static const int version_count = sizeof(versions) / sizeof(versions[0]);
1006
1007 SVCXPRT **xprts;
1008 char netid[16];
1009 char uaddr[128];
1010 struct netconfig *nconf;
1011 int i, j, error;
1012
1013 if (!addr_count) {
1014 printf("NLM: no service addresses given - can't start server");
1015 return (EINVAL);
1016 }
1017
1018 xprts = malloc(addr_count * sizeof(SVCXPRT *), M_NLM, M_WAITOK);
1019 for (i = 0; i < version_count; i++) {
1020 for (j = 0; j < addr_count; j++) {
1021 /*
1022 * Create transports for the first version and
1023 * then just register everything else to the
1024 * same transports.
1025 */
1026 if (i == 0) {
1027 char *up;
1028
1029 error = copyin(&addrs[2*j], &up,
1030 sizeof(char*));
1031 if (error)
1032 goto out;
1033 error = copyinstr(up, netid, sizeof(netid),
1034 NULL);
1035 if (error)
1036 goto out;
1037 error = copyin(&addrs[2*j+1], &up,
1038 sizeof(char*));
1039 if (error)
1040 goto out;
1041 error = copyinstr(up, uaddr, sizeof(uaddr),
1042 NULL);
1043 if (error)
1044 goto out;
1045 nconf = getnetconfigent(netid);
1046 if (!nconf) {
1047 printf("Can't lookup netid %s\n",
1048 netid);
1049 error = EINVAL;
1050 goto out;
1051 }
1052 xprts[j] = svc_tp_create(pool, dispatchers[i],
1053 NLM_PROG, versions[i], uaddr, nconf);
1054 if (!xprts[j]) {
1055 printf("NLM: unable to create "
1056 "(NLM_PROG, %d).\n", versions[i]);
1057 error = EINVAL;
1058 goto out;
1059 }
1060 freenetconfigent(nconf);
1061 } else {
1062 nconf = getnetconfigent(xprts[j]->xp_netid);
1063 rpcb_unset(NLM_PROG, versions[i], nconf);
1064 if (!svc_reg(xprts[j], NLM_PROG, versions[i],
1065 dispatchers[i], nconf)) {
1066 printf("NLM: can't register "
1067 "(NLM_PROG, %d)\n", versions[i]);
1068 error = EINVAL;
1069 goto out;
1070 }
1071 }
1072 }
1073 }
1074 error = 0;
1075 out:
1076 free(xprts, M_NLM);
1077 return (error);
1078 }
1079
1080 /*
1081 * Main server entry point. Contacts the local NSM to get its current
1082 * state and send SM_UNMON_ALL. Registers the NLM services and then
1083 * services requests. Does not return until the server is interrupted
1084 * by a signal.
1085 */
1086 static int
1087 nlm_server_main(int addr_count, char **addrs)
1088 {
1089 struct thread *td = curthread;
1090 int error;
1091 SVCPOOL *pool = NULL;
1092 struct sockopt opt;
1093 int portlow;
1094 #ifdef INET6
1095 struct sockaddr_in6 sin6;
1096 #endif
1097 struct sockaddr_in sin;
1098 my_id id;
1099 sm_stat smstat;
1100 struct timeval timo;
1101 enum clnt_stat stat;
1102 struct nlm_host *host;
1103
1104 if (nlm_socket) {
1105 printf("NLM: can't start server - it appears to be running already\n");
1106 return (EPERM);
1107 }
1108
1109 memset(&opt, 0, sizeof(opt));
1110
1111 nlm_socket = NULL;
1112 error = socreate(AF_INET, &nlm_socket, SOCK_DGRAM, 0,
1113 td->td_ucred, td);
1114 if (error) {
1115 printf("NLM: can't create IPv4 socket - error %d\n", error);
1116 return (error);
1117 }
1118 opt.sopt_dir = SOPT_SET;
1119 opt.sopt_level = IPPROTO_IP;
1120 opt.sopt_name = IP_PORTRANGE;
1121 portlow = IP_PORTRANGE_LOW;
1122 opt.sopt_val = &portlow;
1123 opt.sopt_valsize = sizeof(portlow);
1124 sosetopt(nlm_socket, &opt);
1125
1126 #ifdef INET6
1127 nlm_socket6 = NULL;
1128 error = socreate(AF_INET6, &nlm_socket6, SOCK_DGRAM, 0,
1129 td->td_ucred, td);
1130 if (error) {
1131 printf("NLM: can't create IPv6 socket - error %d\n", error);
1132 return (error);
1133 }
1134 opt.sopt_dir = SOPT_SET;
1135 opt.sopt_level = IPPROTO_IPV6;
1136 opt.sopt_name = IPV6_PORTRANGE;
1137 portlow = IPV6_PORTRANGE_LOW;
1138 opt.sopt_val = &portlow;
1139 opt.sopt_valsize = sizeof(portlow);
1140 sosetopt(nlm_socket6, &opt);
1141 #endif
1142
1143 #ifdef INET6
1144 memset(&sin6, 0, sizeof(sin6));
1145 sin6.sin6_len = sizeof(sin6);
1146 sin6.sin6_family = AF_INET6;
1147 sin6.sin6_addr = in6addr_loopback;
1148 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin6, SM_PROG, SM_VERS);
1149 if (!nlm_nsm) {
1150 #endif
1151 memset(&sin, 0, sizeof(sin));
1152 sin.sin_len = sizeof(sin);
1153 sin.sin_family = AF_INET;
1154 sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1155 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin, SM_PROG,
1156 SM_VERS);
1157 #ifdef INET6
1158 }
1159 #endif
1160
1161 if (!nlm_nsm) {
1162 printf("Can't start NLM - unable to contact NSM\n");
1163 error = EINVAL;
1164 goto out;
1165 }
1166
1167 pool = svcpool_create();
1168
1169 error = nlm_register_services(pool, addr_count, addrs);
1170 if (error)
1171 goto out;
1172
1173 memset(&id, 0, sizeof(id));
1174 id.my_name = "NFS NLM";
1175
1176 timo.tv_sec = 25;
1177 timo.tv_usec = 0;
1178 stat = CLNT_CALL(nlm_nsm, SM_UNMON_ALL,
1179 (xdrproc_t) xdr_my_id, &id,
1180 (xdrproc_t) xdr_sm_stat, &smstat, timo);
1181
1182 if (stat != RPC_SUCCESS) {
1183 struct rpc_err err;
1184
1185 CLNT_GETERR(nlm_nsm, &err);
1186 printf("NLM: unexpected error contacting NSM, stat=%d, errno=%d\n",
1187 stat, err.re_errno);
1188 error = EINVAL;
1189 goto out;
1190 }
1191
1192 if (nlm_debug_level >= 1)
1193 printf("NLM: local NSM state is %d\n", smstat.state);
1194
1195 svc_run(pool);
1196 error = 0;
1197
1198 out:
1199 if (pool)
1200 svcpool_destroy(pool);
1201
1202 /*
1203 * Trash all the existing state so that if the server
1204 * restarts, it gets a clean slate.
1205 */
1206 while ((host = TAILQ_FIRST(&nlm_hosts)) != NULL) {
1207 nlm_host_notify(host, 0, TRUE);
1208 }
1209 if (nlm_nsm) {
1210 AUTH_DESTROY(nlm_nsm->cl_auth);
1211 CLNT_DESTROY(nlm_nsm);
1212 nlm_nsm = NULL;
1213 }
1214 if (nlm_lockd) {
1215 AUTH_DESTROY(nlm_lockd->cl_auth);
1216 CLNT_DESTROY(nlm_lockd);
1217 nlm_lockd = NULL;
1218 }
1219
1220 soclose(nlm_socket);
1221 nlm_socket = NULL;
1222 #ifdef INET6
1223 soclose(nlm_socket6);
1224 nlm_socket6 = NULL;
1225 #endif
1226
1227 return (error);
1228 }
1229
1230 int
1231 nlm_syscall(struct thread *td, struct nlm_syscall_args *uap)
1232 {
1233 int error;
1234
1235 #if __FreeBSD_version >= 700000
1236 error = priv_check(td, PRIV_NFS_LOCKD);
1237 #else
1238 error = suser(td);
1239 #endif
1240 if (error)
1241 return (error);
1242
1243 nlm_debug_level = uap->debug_level;
1244 nlm_grace_threshold = time_uptime + uap->grace_period;
1245 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD;
1246
1247 return nlm_server_main(uap->addr_count, uap->addrs);
1248 }
1249
1250 /**********************************************************************/
1251
1252 /*
1253 * NLM implementation details, called from the RPC stubs.
1254 */
1255
1256
1257 void
1258 nlm_sm_notify(struct nlm_sm_status *argp)
1259 {
1260 uint32_t sysid;
1261 struct nlm_host *host;
1262
1263 if (nlm_debug_level >= 3)
1264 printf("nlm_sm_notify(): mon_name = %s\n", argp->mon_name);
1265 memcpy(&sysid, &argp->priv, sizeof(sysid));
1266 host = nlm_find_host_by_sysid(sysid);
1267 if (host)
1268 nlm_host_notify(host, argp->state, FALSE);
1269 }
1270
1271 static void
1272 nlm_convert_to_fhandle_t(fhandle_t *fhp, struct netobj *p)
1273 {
1274 memcpy(fhp, p->n_bytes, sizeof(fhandle_t));
1275 }
1276
1277 struct vfs_state {
1278 struct mount *vs_mp;
1279 struct vnode *vs_vp;
1280 int vs_vfslocked;
1281 int vs_vnlocked;
1282 };
1283
1284 static int
1285 nlm_get_vfs_state(struct nlm_host *host, struct svc_req *rqstp,
1286 fhandle_t *fhp, struct vfs_state *vs)
1287 {
1288 int error, exflags, freecred;
1289 struct ucred *cred = NULL, *credanon;
1290
1291 memset(vs, 0, sizeof(*vs));
1292 freecred = FALSE;
1293
1294 vs->vs_mp = vfs_getvfs(&fhp->fh_fsid);
1295 if (!vs->vs_mp) {
1296 return (ESTALE);
1297 }
1298 vs->vs_vfslocked = VFS_LOCK_GIANT(vs->vs_mp);
1299
1300 error = VFS_CHECKEXP(vs->vs_mp, (struct sockaddr *)&host->nh_addr,
1301 &exflags, &credanon);
1302 if (error)
1303 goto out;
1304
1305 if (exflags & MNT_EXRDONLY || (vs->vs_mp->mnt_flag & MNT_RDONLY)) {
1306 error = EROFS;
1307 goto out;
1308 }
1309
1310 error = VFS_FHTOVP(vs->vs_mp, &fhp->fh_fid, &vs->vs_vp);
1311 if (error)
1312 goto out;
1313 vs->vs_vnlocked = TRUE;
1314
1315 cred = crget();
1316 freecred = TRUE;
1317 if (!svc_getcred(rqstp, cred, NULL)) {
1318 error = EINVAL;
1319 goto out;
1320 }
1321 if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
1322 crfree(cred);
1323 cred = credanon;
1324 freecred = FALSE;
1325 }
1326
1327 /*
1328 * Check cred.
1329 */
1330 error = VOP_ACCESS(vs->vs_vp, VWRITE, cred, curthread);
1331 if (error)
1332 goto out;
1333
1334 #if __FreeBSD_version < 800011
1335 VOP_UNLOCK(vs->vs_vp, 0, curthread);
1336 #else
1337 VOP_UNLOCK(vs->vs_vp, 0);
1338 #endif
1339 vs->vs_vnlocked = FALSE;
1340
1341 out:
1342 if (freecred)
1343 crfree(cred);
1344
1345 return (error);
1346 }
1347
1348 static void
1349 nlm_release_vfs_state(struct vfs_state *vs)
1350 {
1351
1352 if (vs->vs_vp) {
1353 if (vs->vs_vnlocked)
1354 vput(vs->vs_vp);
1355 else
1356 vrele(vs->vs_vp);
1357 }
1358 if (vs->vs_mp)
1359 vfs_rel(vs->vs_mp);
1360 VFS_UNLOCK_GIANT(vs->vs_vfslocked);
1361 }
1362
1363 static nlm4_stats
1364 nlm_convert_error(int error)
1365 {
1366
1367 if (error == ESTALE)
1368 return nlm4_stale_fh;
1369 else if (error == EROFS)
1370 return nlm4_rofs;
1371 else
1372 return nlm4_failed;
1373 }
1374
1375 struct nlm_host *
1376 nlm_do_test(nlm4_testargs *argp, nlm4_testres *result, struct svc_req *rqstp)
1377 {
1378 fhandle_t fh;
1379 struct vfs_state vs;
1380 struct nlm_host *host, *bhost;
1381 int error, sysid;
1382 struct flock fl;
1383
1384 memset(result, 0, sizeof(*result));
1385
1386 host = nlm_find_host_by_name(argp->alock.caller_name, rqstp);
1387 if (!host) {
1388 result->stat.stat = nlm4_denied_nolocks;
1389 return (NULL);
1390 }
1391
1392 if (nlm_debug_level >= 3)
1393 printf("nlm_do_test(): caller_name = %s (sysid = %d)\n",
1394 host->nh_caller_name, host->nh_sysid);
1395
1396 nlm_free_finished_locks(host);
1397 sysid = host->nh_sysid;
1398
1399 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
1400 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
1401
1402 if (time_uptime < nlm_grace_threshold) {
1403 result->stat.stat = nlm4_denied_grace_period;
1404 return (host);
1405 }
1406
1407 error = nlm_get_vfs_state(host, rqstp, &fh, &vs);
1408 if (error) {
1409 result->stat.stat = nlm_convert_error(error);
1410 goto out;
1411 }
1412
1413 fl.l_start = argp->alock.l_offset;
1414 fl.l_len = argp->alock.l_len;
1415 fl.l_pid = argp->alock.svid;
1416 fl.l_sysid = sysid;
1417 fl.l_whence = SEEK_SET;
1418 if (argp->exclusive)
1419 fl.l_type = F_WRLCK;
1420 else
1421 fl.l_type = F_RDLCK;
1422 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_GETLK, &fl, F_REMOTE);
1423 if (error) {
1424 result->stat.stat = nlm4_failed;
1425 goto out;
1426 }
1427
1428 if (fl.l_type == F_UNLCK) {
1429 result->stat.stat = nlm4_granted;
1430 } else {
1431 result->stat.stat = nlm4_denied;
1432 result->stat.nlm4_testrply_u.holder.exclusive =
1433 (fl.l_type == F_WRLCK);
1434 result->stat.nlm4_testrply_u.holder.svid = fl.l_pid;
1435 bhost = nlm_find_host_by_sysid(fl.l_sysid);
1436 if (bhost) {
1437 /*
1438 * We don't have any useful way of recording
1439 * the value of oh used in the original lock
1440 * request. Ideally, the test reply would have
1441 * a space for the owning host's name allowing
1442 * our caller's NLM to keep track.
1443 *
1444 * As far as I can see, Solaris uses an eight
1445 * byte structure for oh which contains a four
1446 * byte pid encoded in local byte order and
1447 * the first four bytes of the host
1448 * name. Linux uses a variable length string
1449 * 'pid@hostname' in ascii but doesn't even
1450 * return that in test replies.
1451 *
1452 * For the moment, return nothing in oh
1453 * (already zero'ed above).
1454 */
1455 }
1456 result->stat.nlm4_testrply_u.holder.l_offset = fl.l_start;
1457 result->stat.nlm4_testrply_u.holder.l_len = fl.l_len;
1458 }
1459
1460 out:
1461 nlm_release_vfs_state(&vs);
1462 return (host);
1463 }
1464
1465 struct nlm_host *
1466 nlm_do_lock(nlm4_lockargs *argp, nlm4_res *result, struct svc_req *rqstp,
1467 bool_t monitor)
1468 {
1469 fhandle_t fh;
1470 struct vfs_state vs;
1471 struct nlm_host *host;
1472 int error, sysid;
1473 struct flock fl;
1474
1475 memset(result, 0, sizeof(*result));
1476
1477 host = nlm_find_host_by_name(argp->alock.caller_name, rqstp);
1478 if (!host) {
1479 result->stat.stat = nlm4_denied_nolocks;
1480 return (NULL);
1481 }
1482
1483 if (nlm_debug_level >= 3)
1484 printf("nlm_do_lock(): caller_name = %s (sysid = %d)\n",
1485 host->nh_caller_name, host->nh_sysid);
1486
1487 if (monitor && host->nh_state && argp->state
1488 && host->nh_state != argp->state) {
1489 /*
1490 * The host rebooted without telling us. Trash its
1491 * locks.
1492 */
1493 nlm_host_notify(host, argp->state, FALSE);
1494 }
1495
1496 nlm_free_finished_locks(host);
1497 sysid = host->nh_sysid;
1498
1499 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
1500 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
1501
1502 if (time_uptime < nlm_grace_threshold && !argp->reclaim) {
1503 result->stat.stat = nlm4_denied_grace_period;
1504 return (host);
1505 }
1506
1507 error = nlm_get_vfs_state(host, rqstp, &fh, &vs);
1508 if (error) {
1509 result->stat.stat = nlm_convert_error(error);
1510 goto out;
1511 }
1512
1513 fl.l_start = argp->alock.l_offset;
1514 fl.l_len = argp->alock.l_len;
1515 fl.l_pid = argp->alock.svid;
1516 fl.l_sysid = sysid;
1517 fl.l_whence = SEEK_SET;
1518 if (argp->exclusive)
1519 fl.l_type = F_WRLCK;
1520 else
1521 fl.l_type = F_RDLCK;
1522 if (argp->block) {
1523 struct nlm_async_lock *af;
1524
1525 /*
1526 * First, make sure we can contact the host's NLM.
1527 */
1528 if (!nlm_host_get_rpc(host)) {
1529 result->stat.stat = nlm4_failed;
1530 goto out;
1531 }
1532
1533 /*
1534 * First we need to check and see if there is an
1535 * existing blocked lock that matches. This could be a
1536 * badly behaved client or an RPC re-send. If we find
1537 * one, just return nlm4_blocked.
1538 */
1539 mtx_lock(&host->nh_lock);
1540 TAILQ_FOREACH(af, &host->nh_pending, af_link) {
1541 if (af->af_fl.l_start == fl.l_start
1542 && af->af_fl.l_len == fl.l_len
1543 && af->af_fl.l_pid == fl.l_pid
1544 && af->af_fl.l_type == fl.l_type) {
1545 break;
1546 }
1547 }
1548 mtx_unlock(&host->nh_lock);
1549 if (af) {
1550 result->stat.stat = nlm4_blocked;
1551 goto out;
1552 }
1553
1554 af = malloc(sizeof(struct nlm_async_lock), M_NLM,
1555 M_WAITOK|M_ZERO);
1556 TASK_INIT(&af->af_task, 0, nlm_lock_callback, af);
1557 af->af_vp = vs.vs_vp;
1558 af->af_fl = fl;
1559 af->af_host = host;
1560 /*
1561 * We use M_RPC here so that we can xdr_free the thing
1562 * later.
1563 */
1564 af->af_granted.exclusive = argp->exclusive;
1565 af->af_granted.alock.caller_name =
1566 strdup(argp->alock.caller_name, M_RPC);
1567 nlm_copy_netobj(&af->af_granted.alock.fh,
1568 &argp->alock.fh, M_RPC);
1569 nlm_copy_netobj(&af->af_granted.alock.oh,
1570 &argp->alock.oh, M_RPC);
1571 af->af_granted.alock.svid = argp->alock.svid;
1572 af->af_granted.alock.l_offset = argp->alock.l_offset;
1573 af->af_granted.alock.l_len = argp->alock.l_len;
1574
1575 /*
1576 * Put the entry on the pending list before calling
1577 * VOP_ADVLOCKASYNC. We do this in case the lock
1578 * request was blocked (returning EINPROGRESS) but
1579 * then granted before we manage to run again. The
1580 * client may receive the granted message before we
1581 * send our blocked reply but thats their problem.
1582 */
1583 mtx_lock(&host->nh_lock);
1584 TAILQ_INSERT_TAIL(&host->nh_pending, af, af_link);
1585 mtx_unlock(&host->nh_lock);
1586
1587 error = VOP_ADVLOCKASYNC(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE,
1588 &af->af_task, &af->af_cookie);
1589
1590 /*
1591 * If the lock completed synchronously, just free the
1592 * tracking structure now.
1593 */
1594 if (error != EINPROGRESS) {
1595 mtx_lock(&host->nh_lock);
1596 TAILQ_REMOVE(&host->nh_pending, af, af_link);
1597 mtx_unlock(&host->nh_lock);
1598 xdr_free((xdrproc_t) xdr_nlm4_testargs,
1599 &af->af_granted);
1600 free(af, M_NLM);
1601 } else {
1602 if (nlm_debug_level >= 2)
1603 printf("NLM: pending async lock %p for %s "
1604 "(sysid %d)\n",
1605 af, host->nh_caller_name, sysid);
1606 /*
1607 * Don't vrele the vnode just yet - this must
1608 * wait until either the async callback
1609 * happens or the lock is cancelled.
1610 */
1611 vs.vs_vp = NULL;
1612 }
1613 } else {
1614 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE);
1615 }
1616
1617 if (error) {
1618 if (error == EINPROGRESS) {
1619 result->stat.stat = nlm4_blocked;
1620 } else if (error == EDEADLK) {
1621 result->stat.stat = nlm4_deadlck;
1622 } else if (error == EAGAIN) {
1623 result->stat.stat = nlm4_denied;
1624 } else {
1625 result->stat.stat = nlm4_failed;
1626 }
1627 } else {
1628 if (monitor)
1629 nlm_host_monitor(host, argp->state);
1630 result->stat.stat = nlm4_granted;
1631 }
1632
1633 out:
1634 nlm_release_vfs_state(&vs);
1635
1636 return (host);
1637 }
1638
1639 struct nlm_host *
1640 nlm_do_cancel(nlm4_cancargs *argp, nlm4_res *result, struct svc_req *rqstp)
1641 {
1642 fhandle_t fh;
1643 struct vfs_state vs;
1644 struct nlm_host *host;
1645 int error, sysid;
1646 struct flock fl;
1647 struct nlm_async_lock *af;
1648
1649 memset(result, 0, sizeof(*result));
1650
1651 host = nlm_find_host_by_name(argp->alock.caller_name, rqstp);
1652 if (!host) {
1653 result->stat.stat = nlm4_denied_nolocks;
1654 return (NULL);
1655 }
1656
1657 if (nlm_debug_level >= 3)
1658 printf("nlm_do_cancel(): caller_name = %s (sysid = %d)\n",
1659 host->nh_caller_name, host->nh_sysid);
1660
1661 nlm_free_finished_locks(host);
1662 sysid = host->nh_sysid;
1663
1664 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
1665 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
1666
1667 if (time_uptime < nlm_grace_threshold) {
1668 result->stat.stat = nlm4_denied_grace_period;
1669 return (host);
1670 }
1671
1672 error = nlm_get_vfs_state(host, rqstp, &fh, &vs);
1673 if (error) {
1674 result->stat.stat = nlm_convert_error(error);
1675 goto out;
1676 }
1677
1678 fl.l_start = argp->alock.l_offset;
1679 fl.l_len = argp->alock.l_len;
1680 fl.l_pid = argp->alock.svid;
1681 fl.l_sysid = sysid;
1682 fl.l_whence = SEEK_SET;
1683 if (argp->exclusive)
1684 fl.l_type = F_WRLCK;
1685 else
1686 fl.l_type = F_RDLCK;
1687
1688 /*
1689 * First we need to try and find the async lock request - if
1690 * there isn't one, we give up and return nlm4_denied.
1691 */
1692 mtx_lock(&host->nh_lock);
1693
1694 TAILQ_FOREACH(af, &host->nh_pending, af_link) {
1695 if (af->af_fl.l_start == fl.l_start
1696 && af->af_fl.l_len == fl.l_len
1697 && af->af_fl.l_pid == fl.l_pid
1698 && af->af_fl.l_type == fl.l_type) {
1699 break;
1700 }
1701 }
1702
1703 if (!af) {
1704 mtx_unlock(&host->nh_lock);
1705 result->stat.stat = nlm4_denied;
1706 goto out;
1707 }
1708
1709 error = nlm_cancel_async_lock(af);
1710
1711 if (error) {
1712 result->stat.stat = nlm4_denied;
1713 } else {
1714 result->stat.stat = nlm4_granted;
1715 }
1716
1717 mtx_unlock(&host->nh_lock);
1718
1719 out:
1720 nlm_release_vfs_state(&vs);
1721
1722 return (host);
1723 }
1724
1725 struct nlm_host *
1726 nlm_do_unlock(nlm4_unlockargs *argp, nlm4_res *result, struct svc_req *rqstp)
1727 {
1728 fhandle_t fh;
1729 struct vfs_state vs;
1730 struct nlm_host *host;
1731 int error, sysid;
1732 struct flock fl;
1733
1734 memset(result, 0, sizeof(*result));
1735
1736 host = nlm_find_host_by_name(argp->alock.caller_name, rqstp);
1737 if (!host) {
1738 result->stat.stat = nlm4_denied_nolocks;
1739 return (NULL);
1740 }
1741
1742 if (nlm_debug_level >= 3)
1743 printf("nlm_do_unlock(): caller_name = %s (sysid = %d)\n",
1744 host->nh_caller_name, host->nh_sysid);
1745
1746 nlm_free_finished_locks(host);
1747 sysid = host->nh_sysid;
1748
1749 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
1750 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
1751
1752 if (time_uptime < nlm_grace_threshold) {
1753 result->stat.stat = nlm4_denied_grace_period;
1754 return (host);
1755 }
1756
1757 error = nlm_get_vfs_state(host, rqstp, &fh, &vs);
1758 if (error) {
1759 result->stat.stat = nlm_convert_error(error);
1760 goto out;
1761 }
1762
1763 fl.l_start = argp->alock.l_offset;
1764 fl.l_len = argp->alock.l_len;
1765 fl.l_pid = argp->alock.svid;
1766 fl.l_sysid = sysid;
1767 fl.l_whence = SEEK_SET;
1768 fl.l_type = F_UNLCK;
1769 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_UNLCK, &fl, F_REMOTE);
1770
1771 /*
1772 * Ignore the error - there is no result code for failure,
1773 * only for grace period.
1774 */
1775 result->stat.stat = nlm4_granted;
1776
1777 out:
1778 nlm_release_vfs_state(&vs);
1779
1780 return (host);
1781 }
1782
1783 void
1784 nlm_do_free_all(nlm4_notify *argp)
1785 {
1786 struct nlm_host *host, *thost;
1787
1788 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, thost) {
1789 if (!strcmp(host->nh_caller_name, argp->name))
1790 nlm_host_notify(host, argp->state, FALSE);
1791 }
1792 }
1793
1794 #define _PATH_RPCLOCKDSOCK "/var/run/rpclockd.sock"
1795
1796 /*
1797 * Make a connection to the userland lockd - we push anything we can't
1798 * handle out to userland.
1799 */
1800 CLIENT *
1801 nlm_user_lockd(void)
1802 {
1803 struct sockaddr_un sun;
1804 struct netconfig *nconf;
1805 struct timeval zero;
1806
1807 if (nlm_lockd)
1808 return (nlm_lockd);
1809
1810 sun.sun_family = AF_LOCAL;
1811 strcpy(sun.sun_path, _PATH_RPCLOCKDSOCK);
1812 sun.sun_len = SUN_LEN(&sun);
1813
1814 nconf = getnetconfigent("local");
1815 nlm_lockd = clnt_reconnect_create(nconf, (struct sockaddr *) &sun,
1816 NLM_PROG, NLM_VERS4, RPC_MAXDATASIZE, RPC_MAXDATASIZE);
1817
1818 /*
1819 * Set the send timeout to zero - we only use this rpc handle
1820 * for sending async replies which have no return value.
1821 */
1822 zero.tv_sec = 0;
1823 zero.tv_usec = 0;
1824 CLNT_CONTROL(nlm_lockd, CLSET_TIMEOUT, &zero);
1825
1826 return (nlm_lockd);
1827 }
1828
1829 /*
1830 * Kernel module glue
1831 */
1832 static int
1833 nfslockd_modevent(module_t mod, int type, void *data)
1834 {
1835
1836 return (0);
1837 }
1838 static moduledata_t nfslockd_mod = {
1839 "nfslockd",
1840 nfslockd_modevent,
1841 NULL,
1842 };
1843 DECLARE_MODULE(nfslockd, nfslockd_mod, SI_SUB_VFS, SI_ORDER_ANY);
1844
1845 /* So that loader and kldload(2) can find us, wherever we are.. */
1846 MODULE_DEPEND(nfslockd, krpc, 1, 1, 1);
1847 MODULE_VERSION(nfslockd, 1);
Cache object: 5f545b66f372a0a255e318c31599ab1c
|