1 /* $NetBSD: nlm_prot_impl.c,v 1.2 2015/12/13 18:31:09 christos Exp $ */
2 /*-
3 * Copyright (c) 2008 Isilon Inc http://www.isilon.com/
4 * Authors: Doug Rabson <dfr@rabson.org>
5 * Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org>
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 #include "opt_inet6.h"
30
31 #include <sys/cdefs.h>
32 /* __FBSDID("FreeBSD: head/sys/nlm/nlm_prot_impl.c 255333 2013-09-06 23:14:31Z rmacklem "); */
33 __RCSID("$NetBSD: nlm_prot_impl.c,v 1.2 2015/12/13 18:31:09 christos Exp $");
34
35 #include <sys/param.h>
36 #include <sys/fail.h>
37 #include <sys/fcntl.h>
38 #include <sys/kernel.h>
39 #include <sys/kthread.h>
40 #include <sys/lockf.h>
41 #include <sys/malloc.h>
42 #include <sys/mount.h>
43 #if __FreeBSD_version >= 700000
44 #include <sys/priv.h>
45 #endif
46 #include <sys/proc.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/syscall.h>
50 #include <sys/sysctl.h>
51 #include <sys/sysent.h>
52 #include <sys/syslog.h>
53 #include <sys/sysproto.h>
54 #include <sys/systm.h>
55 #include <sys/taskqueue.h>
56 #include <sys/unistd.h>
57 #include <sys/vnode.h>
58
59 #include <nfs/nfsproto.h>
60 #include <nfs/nfs_lock.h>
61
62 #include <nlm/nlm_prot.h>
63 #include <nlm/sm_inter.h>
64 #include <nlm/nlm.h>
65 #include <rpc/rpc_com.h>
66 #include <rpc/rpcb_prot.h>
67
68 MALLOC_DEFINE(M_NLM, "NLM", "Network Lock Manager");
69
70 /*
71 * If a host is inactive (and holds no locks) for this amount of
72 * seconds, we consider it idle and stop tracking it.
73 */
74 #define NLM_IDLE_TIMEOUT 30
75
76 /*
77 * We check the host list for idle every few seconds.
78 */
79 #define NLM_IDLE_PERIOD 5
80
81 /*
82 * We only look for GRANTED_RES messages for a little while.
83 */
84 #define NLM_EXPIRE_TIMEOUT 10
85
86 /*
87 * Support for sysctl vfs.nlm.sysid
88 */
89 static SYSCTL_NODE(_vfs, OID_AUTO, nlm, CTLFLAG_RW, NULL,
90 "Network Lock Manager");
91 static SYSCTL_NODE(_vfs_nlm, OID_AUTO, sysid, CTLFLAG_RW, NULL, "");
92
93 /*
94 * Syscall hooks
95 */
96 static int nlm_syscall_offset = SYS_nlm_syscall;
97 static struct sysent nlm_syscall_prev_sysent;
98 #if __FreeBSD_version < 700000
99 static struct sysent nlm_syscall_sysent = {
100 (sizeof(struct nlm_syscall_args) / sizeof(register_t)) | SYF_MPSAFE,
101 (sy_call_t *) nlm_syscall
102 };
103 #else
104 MAKE_SYSENT(nlm_syscall);
105 #endif
106 static bool_t nlm_syscall_registered = FALSE;
107
108 /*
109 * Debug level passed in from userland. We also support a sysctl hook
110 * so that it can be changed on a live system.
111 */
112 static int nlm_debug_level;
113 SYSCTL_INT(_debug, OID_AUTO, nlm_debug, CTLFLAG_RW, &nlm_debug_level, 0, "");
114
115 #define NLM_DEBUG(_level, args...) \
116 do { \
117 if (nlm_debug_level >= (_level)) \
118 log(LOG_DEBUG, args); \
119 } while(0)
120 #define NLM_ERR(args...) \
121 do { \
122 log(LOG_ERR, args); \
123 } while(0)
124
125 /*
126 * Grace period handling. The value of nlm_grace_threshold is the
127 * value of time_uptime after which we are serving requests normally.
128 */
129 static time_t nlm_grace_threshold;
130
131 /*
132 * We check for idle hosts if time_uptime is greater than
133 * nlm_next_idle_check,
134 */
135 static time_t nlm_next_idle_check;
136
137 /*
138 * A flag to indicate the server is already running.
139 */
140 static int nlm_is_running;
141
142 /*
143 * A socket to use for RPC - shared by all IPv4 RPC clients.
144 */
145 static struct socket *nlm_socket;
146
147 #ifdef INET6
148
149 /*
150 * A socket to use for RPC - shared by all IPv6 RPC clients.
151 */
152 static struct socket *nlm_socket6;
153
154 #endif
155
156 /*
157 * An RPC client handle that can be used to communicate with the local
158 * NSM.
159 */
160 static CLIENT *nlm_nsm;
161
162 /*
163 * An AUTH handle for the server's creds.
164 */
165 static AUTH *nlm_auth;
166
167 /*
168 * A zero timeval for sending async RPC messages.
169 */
170 struct timeval nlm_zero_tv = { 0, 0 };
171
172 /*
173 * The local NSM state number
174 */
175 int nlm_nsm_state;
176
177
178 /*
179 * A lock to protect the host list and waiting lock list.
180 */
181 static struct mtx nlm_global_lock;
182
183 /*
184 * Locks:
185 * (l) locked by nh_lock
186 * (s) only accessed via server RPC which is single threaded
187 * (g) locked by nlm_global_lock
188 * (c) const until freeing
189 * (a) modified using atomic ops
190 */
191
192 /*
193 * A pending client-side lock request, stored on the nlm_waiting_locks
194 * list.
195 */
196 struct nlm_waiting_lock {
197 TAILQ_ENTRY(nlm_waiting_lock) nw_link; /* (g) */
198 bool_t nw_waiting; /* (g) */
199 nlm4_lock nw_lock; /* (c) */
200 union nfsfh nw_fh; /* (c) */
201 struct vnode *nw_vp; /* (c) */
202 };
203 TAILQ_HEAD(nlm_waiting_lock_list, nlm_waiting_lock);
204
205 struct nlm_waiting_lock_list nlm_waiting_locks; /* (g) */
206
207 /*
208 * A pending server-side asynchronous lock request, stored on the
209 * nh_pending list of the NLM host.
210 */
211 struct nlm_async_lock {
212 TAILQ_ENTRY(nlm_async_lock) af_link; /* (l) host's list of locks */
213 struct task af_task; /* (c) async callback details */
214 void *af_cookie; /* (l) lock manager cancel token */
215 struct vnode *af_vp; /* (l) vnode to lock */
216 struct flock af_fl; /* (c) lock details */
217 struct nlm_host *af_host; /* (c) host which is locking */
218 CLIENT *af_rpc; /* (c) rpc client to send message */
219 nlm4_testargs af_granted; /* (c) notification details */
220 time_t af_expiretime; /* (c) notification time */
221 };
222 TAILQ_HEAD(nlm_async_lock_list, nlm_async_lock);
223
224 /*
225 * NLM host.
226 */
227 enum nlm_host_state {
228 NLM_UNMONITORED,
229 NLM_MONITORED,
230 NLM_MONITOR_FAILED,
231 NLM_RECOVERING
232 };
233
234 struct nlm_rpc {
235 CLIENT *nr_client; /* (l) RPC client handle */
236 time_t nr_create_time; /* (l) when client was created */
237 };
238
239 struct nlm_host {
240 struct mtx nh_lock;
241 volatile u_int nh_refs; /* (a) reference count */
242 TAILQ_ENTRY(nlm_host) nh_link; /* (g) global list of hosts */
243 char nh_caller_name[MAXNAMELEN]; /* (c) printable name of host */
244 uint32_t nh_sysid; /* (c) our allocaed system ID */
245 char nh_sysid_string[10]; /* (c) string rep. of sysid */
246 struct sockaddr_storage nh_addr; /* (s) remote address of host */
247 struct nlm_rpc nh_srvrpc; /* (l) RPC for server replies */
248 struct nlm_rpc nh_clntrpc; /* (l) RPC for client requests */
249 rpcvers_t nh_vers; /* (s) NLM version of host */
250 int nh_state; /* (s) last seen NSM state of host */
251 enum nlm_host_state nh_monstate; /* (l) local NSM monitoring state */
252 time_t nh_idle_timeout; /* (s) Time at which host is idle */
253 struct sysctl_ctx_list nh_sysctl; /* (c) vfs.nlm.sysid nodes */
254 uint32_t nh_grantcookie; /* (l) grant cookie counter */
255 struct nlm_async_lock_list nh_pending; /* (l) pending async locks */
256 struct nlm_async_lock_list nh_granted; /* (l) granted locks */
257 struct nlm_async_lock_list nh_finished; /* (l) finished async locks */
258 };
259 TAILQ_HEAD(nlm_host_list, nlm_host);
260
261 static struct nlm_host_list nlm_hosts; /* (g) */
262 static uint32_t nlm_next_sysid = 1; /* (g) */
263
264 static void nlm_host_unmonitor(struct nlm_host *);
265
266 struct nlm_grantcookie {
267 uint32_t ng_sysid;
268 uint32_t ng_cookie;
269 };
270
271 static inline uint32_t
ng_sysid(struct netobj * src)272 ng_sysid(struct netobj *src)
273 {
274
275 return ((struct nlm_grantcookie *)src->n_bytes)->ng_sysid;
276 }
277
278 static inline uint32_t
ng_cookie(struct netobj * src)279 ng_cookie(struct netobj *src)
280 {
281
282 return ((struct nlm_grantcookie *)src->n_bytes)->ng_cookie;
283 }
284
285 /**********************************************************************/
286
287 /*
288 * Initialise NLM globals.
289 */
290 static void
nlm_init(void * dummy)291 nlm_init(void *dummy)
292 {
293 int error;
294
295 mtx_init(&nlm_global_lock, "nlm_global_lock", NULL, MTX_DEF);
296 TAILQ_INIT(&nlm_waiting_locks);
297 TAILQ_INIT(&nlm_hosts);
298
299 error = syscall_register(&nlm_syscall_offset, &nlm_syscall_sysent,
300 &nlm_syscall_prev_sysent);
301 if (error)
302 NLM_ERR("Can't register NLM syscall\n");
303 else
304 nlm_syscall_registered = TRUE;
305 }
306 SYSINIT(nlm_init, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_init, NULL);
307
308 static void
nlm_uninit(void * dummy)309 nlm_uninit(void *dummy)
310 {
311
312 if (nlm_syscall_registered)
313 syscall_deregister(&nlm_syscall_offset,
314 &nlm_syscall_prev_sysent);
315 }
316 SYSUNINIT(nlm_uninit, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_uninit, NULL);
317
318 /*
319 * Create a netobj from an arbitrary source.
320 */
321 void
nlm_make_netobj(struct netobj * dst,caddr_t src,size_t srcsize,struct malloc_type * type)322 nlm_make_netobj(struct netobj *dst, caddr_t src, size_t srcsize,
323 struct malloc_type *type)
324 {
325
326 dst->n_len = srcsize;
327 dst->n_bytes = malloc(srcsize, type, M_WAITOK);
328 memcpy(dst->n_bytes, src, srcsize);
329 }
330
331 /*
332 * Copy a struct netobj.
333 */
334 void
nlm_copy_netobj(struct netobj * dst,struct netobj * src,struct malloc_type * type)335 nlm_copy_netobj(struct netobj *dst, struct netobj *src,
336 struct malloc_type *type)
337 {
338
339 nlm_make_netobj(dst, src->n_bytes, src->n_len, type);
340 }
341
342
343 /*
344 * Create an RPC client handle for the given (address,prog,vers)
345 * triple using UDP.
346 */
347 static CLIENT *
nlm_get_rpc(struct sockaddr * sa,rpcprog_t prog,rpcvers_t vers)348 nlm_get_rpc(struct sockaddr *sa, rpcprog_t prog, rpcvers_t vers)
349 {
350 char *wchan = "nlmrcv";
351 const char* protofmly;
352 struct sockaddr_storage ss;
353 struct socket *so;
354 CLIENT *rpcb;
355 struct timeval timo;
356 RPCB parms;
357 char *uaddr;
358 enum clnt_stat stat = RPC_SUCCESS;
359 int rpcvers = RPCBVERS4;
360 bool_t do_tcp = FALSE;
361 bool_t tryagain = FALSE;
362 struct portmap mapping;
363 u_short port = 0;
364
365 /*
366 * First we need to contact the remote RPCBIND service to find
367 * the right port.
368 */
369 memcpy(&ss, sa, sa->sa_len);
370 switch (ss.ss_family) {
371 case AF_INET:
372 ((struct sockaddr_in *)&ss)->sin_port = htons(111);
373 protofmly = "inet";
374 so = nlm_socket;
375 break;
376
377 #ifdef INET6
378 case AF_INET6:
379 ((struct sockaddr_in6 *)&ss)->sin6_port = htons(111);
380 protofmly = "inet6";
381 so = nlm_socket6;
382 break;
383 #endif
384
385 default:
386 /*
387 * Unsupported address family - fail.
388 */
389 return (NULL);
390 }
391
392 rpcb = clnt_dg_create(so, (struct sockaddr *)&ss,
393 RPCBPROG, rpcvers, 0, 0);
394 if (!rpcb)
395 return (NULL);
396
397 try_tcp:
398 parms.r_prog = prog;
399 parms.r_vers = vers;
400 if (do_tcp)
401 parms.r_netid = "tcp";
402 else
403 parms.r_netid = "udp";
404 parms.r_addr = "";
405 parms.r_owner = "";
406
407 /*
408 * Use the default timeout.
409 */
410 timo.tv_sec = 25;
411 timo.tv_usec = 0;
412 again:
413 switch (rpcvers) {
414 case RPCBVERS4:
415 case RPCBVERS:
416 /*
417 * Try RPCBIND 4 then 3.
418 */
419 uaddr = NULL;
420 stat = CLNT_CALL(rpcb, (rpcprog_t) RPCBPROC_GETADDR,
421 (xdrproc_t) xdr_rpcb, &parms,
422 (xdrproc_t) xdr_wrapstring, &uaddr, timo);
423 if (stat == RPC_SUCCESS) {
424 /*
425 * We have a reply from the remote RPCBIND - turn it
426 * into an appropriate address and make a new client
427 * that can talk to the remote NLM.
428 *
429 * XXX fixup IPv6 scope ID.
430 */
431 struct netbuf *a;
432 a = __rpc_uaddr2taddr_af(ss.ss_family, uaddr);
433 if (!a) {
434 tryagain = TRUE;
435 } else {
436 tryagain = FALSE;
437 memcpy(&ss, a->buf, a->len);
438 free(a->buf, M_RPC);
439 free(a, M_RPC);
440 xdr_free((xdrproc_t) xdr_wrapstring, &uaddr);
441 }
442 }
443 if (tryagain || stat == RPC_PROGVERSMISMATCH) {
444 if (rpcvers == RPCBVERS4)
445 rpcvers = RPCBVERS;
446 else if (rpcvers == RPCBVERS)
447 rpcvers = PMAPVERS;
448 CLNT_CONTROL(rpcb, CLSET_VERS, &rpcvers);
449 goto again;
450 }
451 break;
452 case PMAPVERS:
453 /*
454 * Try portmap.
455 */
456 mapping.pm_prog = parms.r_prog;
457 mapping.pm_vers = parms.r_vers;
458 mapping.pm_prot = do_tcp ? IPPROTO_TCP : IPPROTO_UDP;
459 mapping.pm_port = 0;
460
461 stat = CLNT_CALL(rpcb, (rpcprog_t) PMAPPROC_GETPORT,
462 (xdrproc_t) xdr_portmap, &mapping,
463 (xdrproc_t) xdr_u_short, &port, timo);
464
465 if (stat == RPC_SUCCESS) {
466 switch (ss.ss_family) {
467 case AF_INET:
468 ((struct sockaddr_in *)&ss)->sin_port =
469 htons(port);
470 break;
471
472 #ifdef INET6
473 case AF_INET6:
474 ((struct sockaddr_in6 *)&ss)->sin6_port =
475 htons(port);
476 break;
477 #endif
478 }
479 }
480 break;
481 default:
482 panic("invalid rpcvers %d", rpcvers);
483 }
484 /*
485 * We may have a positive response from the portmapper, but the NLM
486 * service was not found. Make sure we received a valid port.
487 */
488 switch (ss.ss_family) {
489 case AF_INET:
490 port = ((struct sockaddr_in *)&ss)->sin_port;
491 break;
492 #ifdef INET6
493 case AF_INET6:
494 port = ((struct sockaddr_in6 *)&ss)->sin6_port;
495 break;
496 #endif
497 }
498 if (stat != RPC_SUCCESS || !port) {
499 /*
500 * If we were able to talk to rpcbind or portmap, but the udp
501 * variant wasn't available, ask about tcp.
502 *
503 * XXX - We could also check for a TCP portmapper, but
504 * if the host is running a portmapper at all, we should be able
505 * to hail it over UDP.
506 */
507 if (stat == RPC_SUCCESS && !do_tcp) {
508 do_tcp = TRUE;
509 goto try_tcp;
510 }
511
512 /* Otherwise, bad news. */
513 NLM_ERR("NLM: failed to contact remote rpcbind, "
514 "stat = %d, port = %d\n", (int) stat, port);
515 CLNT_DESTROY(rpcb);
516 return (NULL);
517 }
518
519 if (do_tcp) {
520 /*
521 * Destroy the UDP client we used to speak to rpcbind and
522 * recreate as a TCP client.
523 */
524 struct netconfig *nconf = NULL;
525
526 CLNT_DESTROY(rpcb);
527
528 switch (ss.ss_family) {
529 case AF_INET:
530 nconf = getnetconfigent("tcp");
531 break;
532 #ifdef INET6
533 case AF_INET6:
534 nconf = getnetconfigent("tcp6");
535 break;
536 #endif
537 }
538
539 rpcb = clnt_reconnect_create(nconf, (struct sockaddr *)&ss,
540 prog, vers, 0, 0);
541 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan);
542 rpcb->cl_auth = nlm_auth;
543
544 } else {
545 /*
546 * Re-use the client we used to speak to rpcbind.
547 */
548 CLNT_CONTROL(rpcb, CLSET_SVC_ADDR, &ss);
549 CLNT_CONTROL(rpcb, CLSET_PROG, &prog);
550 CLNT_CONTROL(rpcb, CLSET_VERS, &vers);
551 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan);
552 rpcb->cl_auth = nlm_auth;
553 }
554
555 return (rpcb);
556 }
557
558 /*
559 * This async callback after when an async lock request has been
560 * granted. We notify the host which initiated the request.
561 */
562 static void
nlm_lock_callback(void * arg,int pending)563 nlm_lock_callback(void *arg, int pending)
564 {
565 struct nlm_async_lock *af = (struct nlm_async_lock *) arg;
566 struct rpc_callextra ext;
567
568 NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) granted,"
569 " cookie %d:%d\n", af, af->af_host->nh_caller_name,
570 af->af_host->nh_sysid, ng_sysid(&af->af_granted.cookie),
571 ng_cookie(&af->af_granted.cookie));
572
573 /*
574 * Send the results back to the host.
575 *
576 * Note: there is a possible race here with nlm_host_notify
577 * destroying the RPC client. To avoid problems, the first
578 * thing nlm_host_notify does is to cancel pending async lock
579 * requests.
580 */
581 memset(&ext, 0, sizeof(ext));
582 ext.rc_auth = nlm_auth;
583 if (af->af_host->nh_vers == NLM_VERS4) {
584 nlm4_granted_msg_4(&af->af_granted,
585 NULL, af->af_rpc, &ext, nlm_zero_tv);
586 } else {
587 /*
588 * Back-convert to legacy protocol
589 */
590 nlm_testargs granted;
591 granted.cookie = af->af_granted.cookie;
592 granted.exclusive = af->af_granted.exclusive;
593 granted.alock.caller_name =
594 af->af_granted.alock.caller_name;
595 granted.alock.fh = af->af_granted.alock.fh;
596 granted.alock.oh = af->af_granted.alock.oh;
597 granted.alock.svid = af->af_granted.alock.svid;
598 granted.alock.l_offset =
599 af->af_granted.alock.l_offset;
600 granted.alock.l_len =
601 af->af_granted.alock.l_len;
602
603 nlm_granted_msg_1(&granted,
604 NULL, af->af_rpc, &ext, nlm_zero_tv);
605 }
606
607 /*
608 * Move this entry to the nh_granted list.
609 */
610 af->af_expiretime = time_uptime + NLM_EXPIRE_TIMEOUT;
611 mtx_lock(&af->af_host->nh_lock);
612 TAILQ_REMOVE(&af->af_host->nh_pending, af, af_link);
613 TAILQ_INSERT_TAIL(&af->af_host->nh_granted, af, af_link);
614 mtx_unlock(&af->af_host->nh_lock);
615 }
616
617 /*
618 * Free an async lock request. The request must have been removed from
619 * any list.
620 */
621 static void
nlm_free_async_lock(struct nlm_async_lock * af)622 nlm_free_async_lock(struct nlm_async_lock *af)
623 {
624 /*
625 * Free an async lock.
626 */
627 if (af->af_rpc)
628 CLNT_RELEASE(af->af_rpc);
629 xdr_free((xdrproc_t) xdr_nlm4_testargs, &af->af_granted);
630 if (af->af_vp)
631 vrele(af->af_vp);
632 free(af, M_NLM);
633 }
634
635 /*
636 * Cancel our async request - this must be called with
637 * af->nh_host->nh_lock held. This is slightly complicated by a
638 * potential race with our own callback. If we fail to cancel the
639 * lock, it must already have been granted - we make sure our async
640 * task has completed by calling taskqueue_drain in this case.
641 */
642 static int
nlm_cancel_async_lock(struct nlm_async_lock * af)643 nlm_cancel_async_lock(struct nlm_async_lock *af)
644 {
645 struct nlm_host *host = af->af_host;
646 int error;
647
648 mtx_assert(&host->nh_lock, MA_OWNED);
649
650 mtx_unlock(&host->nh_lock);
651
652 error = VOP_ADVLOCKASYNC(af->af_vp, NULL, F_CANCEL, &af->af_fl,
653 F_REMOTE, NULL, &af->af_cookie);
654
655 if (error) {
656 /*
657 * We failed to cancel - make sure our callback has
658 * completed before we continue.
659 */
660 taskqueue_drain(taskqueue_thread, &af->af_task);
661 }
662
663 mtx_lock(&host->nh_lock);
664
665 if (!error) {
666 NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) "
667 "cancelled\n", af, host->nh_caller_name, host->nh_sysid);
668
669 /*
670 * Remove from the nh_pending list and free now that
671 * we are safe from the callback.
672 */
673 TAILQ_REMOVE(&host->nh_pending, af, af_link);
674 mtx_unlock(&host->nh_lock);
675 nlm_free_async_lock(af);
676 mtx_lock(&host->nh_lock);
677 }
678
679 return (error);
680 }
681
682 static void
nlm_check_expired_locks(struct nlm_host * host)683 nlm_check_expired_locks(struct nlm_host *host)
684 {
685 struct nlm_async_lock *af;
686 time_t uptime = time_uptime;
687
688 mtx_lock(&host->nh_lock);
689 while ((af = TAILQ_FIRST(&host->nh_granted)) != NULL
690 && uptime >= af->af_expiretime) {
691 NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) expired,"
692 " cookie %d:%d\n", af, af->af_host->nh_caller_name,
693 af->af_host->nh_sysid, ng_sysid(&af->af_granted.cookie),
694 ng_cookie(&af->af_granted.cookie));
695 TAILQ_REMOVE(&host->nh_granted, af, af_link);
696 mtx_unlock(&host->nh_lock);
697 nlm_free_async_lock(af);
698 mtx_lock(&host->nh_lock);
699 }
700 while ((af = TAILQ_FIRST(&host->nh_finished)) != NULL) {
701 TAILQ_REMOVE(&host->nh_finished, af, af_link);
702 mtx_unlock(&host->nh_lock);
703 nlm_free_async_lock(af);
704 mtx_lock(&host->nh_lock);
705 }
706 mtx_unlock(&host->nh_lock);
707 }
708
709 /*
710 * Free resources used by a host. This is called after the reference
711 * count has reached zero so it doesn't need to worry about locks.
712 */
713 static void
nlm_host_destroy(struct nlm_host * host)714 nlm_host_destroy(struct nlm_host *host)
715 {
716
717 mtx_lock(&nlm_global_lock);
718 TAILQ_REMOVE(&nlm_hosts, host, nh_link);
719 mtx_unlock(&nlm_global_lock);
720
721 if (host->nh_srvrpc.nr_client)
722 CLNT_RELEASE(host->nh_srvrpc.nr_client);
723 if (host->nh_clntrpc.nr_client)
724 CLNT_RELEASE(host->nh_clntrpc.nr_client);
725 mtx_destroy(&host->nh_lock);
726 sysctl_ctx_free(&host->nh_sysctl);
727 free(host, M_NLM);
728 }
729
730 /*
731 * Thread start callback for client lock recovery
732 */
733 static void
nlm_client_recovery_start(void * arg)734 nlm_client_recovery_start(void *arg)
735 {
736 struct nlm_host *host = (struct nlm_host *) arg;
737
738 NLM_DEBUG(1, "NLM: client lock recovery for %s started\n",
739 host->nh_caller_name);
740
741 nlm_client_recovery(host);
742
743 NLM_DEBUG(1, "NLM: client lock recovery for %s completed\n",
744 host->nh_caller_name);
745
746 host->nh_monstate = NLM_MONITORED;
747 nlm_host_release(host);
748
749 kthread_exit();
750 }
751
752 /*
753 * This is called when we receive a host state change notification. We
754 * unlock any active locks owned by the host. When rpc.lockd is
755 * shutting down, this function is called with newstate set to zero
756 * which allows us to cancel any pending async locks and clear the
757 * locking state.
758 */
759 static void
nlm_host_notify(struct nlm_host * host,int newstate)760 nlm_host_notify(struct nlm_host *host, int newstate)
761 {
762 struct nlm_async_lock *af;
763
764 if (newstate) {
765 NLM_DEBUG(1, "NLM: host %s (sysid %d) rebooted, new "
766 "state is %d\n", host->nh_caller_name,
767 host->nh_sysid, newstate);
768 }
769
770 /*
771 * Cancel any pending async locks for this host.
772 */
773 mtx_lock(&host->nh_lock);
774 while ((af = TAILQ_FIRST(&host->nh_pending)) != NULL) {
775 /*
776 * nlm_cancel_async_lock will remove the entry from
777 * nh_pending and free it.
778 */
779 nlm_cancel_async_lock(af);
780 }
781 mtx_unlock(&host->nh_lock);
782 nlm_check_expired_locks(host);
783
784 /*
785 * The host just rebooted - trash its locks.
786 */
787 lf_clearremotesys(host->nh_sysid);
788 host->nh_state = newstate;
789
790 /*
791 * If we have any remote locks for this host (i.e. it
792 * represents a remote NFS server that our local NFS client
793 * has locks for), start a recovery thread.
794 */
795 if (newstate != 0
796 && host->nh_monstate != NLM_RECOVERING
797 && lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid) > 0) {
798 struct thread *td;
799 host->nh_monstate = NLM_RECOVERING;
800 refcount_acquire(&host->nh_refs);
801 kthread_add(nlm_client_recovery_start, host, curproc, &td, 0, 0,
802 "NFS lock recovery for %s", host->nh_caller_name);
803 }
804 }
805
806 /*
807 * Sysctl handler to count the number of locks for a sysid.
808 */
809 static int
nlm_host_lock_count_sysctl(SYSCTL_HANDLER_ARGS)810 nlm_host_lock_count_sysctl(SYSCTL_HANDLER_ARGS)
811 {
812 struct nlm_host *host;
813 int count;
814
815 host = oidp->oid_arg1;
816 count = lf_countlocks(host->nh_sysid);
817 return sysctl_handle_int(oidp, &count, 0, req);
818 }
819
820 /*
821 * Sysctl handler to count the number of client locks for a sysid.
822 */
823 static int
nlm_host_client_lock_count_sysctl(SYSCTL_HANDLER_ARGS)824 nlm_host_client_lock_count_sysctl(SYSCTL_HANDLER_ARGS)
825 {
826 struct nlm_host *host;
827 int count;
828
829 host = oidp->oid_arg1;
830 count = lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid);
831 return sysctl_handle_int(oidp, &count, 0, req);
832 }
833
834 /*
835 * Create a new NLM host.
836 */
837 static struct nlm_host *
nlm_create_host(const char * caller_name)838 nlm_create_host(const char* caller_name)
839 {
840 struct nlm_host *host;
841 struct sysctl_oid *oid;
842
843 mtx_assert(&nlm_global_lock, MA_OWNED);
844
845 NLM_DEBUG(1, "NLM: new host %s (sysid %d)\n",
846 caller_name, nlm_next_sysid);
847 host = malloc(sizeof(struct nlm_host), M_NLM, M_NOWAIT|M_ZERO);
848 if (!host)
849 return (NULL);
850 mtx_init(&host->nh_lock, "nh_lock", NULL, MTX_DEF);
851 host->nh_refs = 1;
852 strlcpy(host->nh_caller_name, caller_name, MAXNAMELEN);
853 host->nh_sysid = nlm_next_sysid++;
854 snprintf(host->nh_sysid_string, sizeof(host->nh_sysid_string),
855 "%d", host->nh_sysid);
856 host->nh_vers = 0;
857 host->nh_state = 0;
858 host->nh_monstate = NLM_UNMONITORED;
859 host->nh_grantcookie = 1;
860 TAILQ_INIT(&host->nh_pending);
861 TAILQ_INIT(&host->nh_granted);
862 TAILQ_INIT(&host->nh_finished);
863 TAILQ_INSERT_TAIL(&nlm_hosts, host, nh_link);
864
865 mtx_unlock(&nlm_global_lock);
866
867 sysctl_ctx_init(&host->nh_sysctl);
868 oid = SYSCTL_ADD_NODE(&host->nh_sysctl,
869 SYSCTL_STATIC_CHILDREN(_vfs_nlm_sysid),
870 OID_AUTO, host->nh_sysid_string, CTLFLAG_RD, NULL, "");
871 SYSCTL_ADD_STRING(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
872 "hostname", CTLFLAG_RD, host->nh_caller_name, 0, "");
873 SYSCTL_ADD_UINT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
874 "version", CTLFLAG_RD, &host->nh_vers, 0, "");
875 SYSCTL_ADD_UINT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
876 "monitored", CTLFLAG_RD, &host->nh_monstate, 0, "");
877 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
878 "lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0,
879 nlm_host_lock_count_sysctl, "I", "");
880 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
881 "client_lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0,
882 nlm_host_client_lock_count_sysctl, "I", "");
883
884 mtx_lock(&nlm_global_lock);
885
886 return (host);
887 }
888
889 /*
890 * Acquire the next sysid for remote locks not handled by the NLM.
891 */
892 uint32_t
nlm_acquire_next_sysid(void)893 nlm_acquire_next_sysid(void)
894 {
895 uint32_t next_sysid;
896
897 mtx_lock(&nlm_global_lock);
898 next_sysid = nlm_next_sysid++;
899 mtx_unlock(&nlm_global_lock);
900 return (next_sysid);
901 }
902
903 /*
904 * Return non-zero if the address parts of the two sockaddrs are the
905 * same.
906 */
907 static int
nlm_compare_addr(const struct sockaddr * a,const struct sockaddr * b)908 nlm_compare_addr(const struct sockaddr *a, const struct sockaddr *b)
909 {
910 const struct sockaddr_in *a4, *b4;
911 #ifdef INET6
912 const struct sockaddr_in6 *a6, *b6;
913 #endif
914
915 if (a->sa_family != b->sa_family)
916 return (FALSE);
917
918 switch (a->sa_family) {
919 case AF_INET:
920 a4 = (const struct sockaddr_in *) a;
921 b4 = (const struct sockaddr_in *) b;
922 return !memcmp(&a4->sin_addr, &b4->sin_addr,
923 sizeof(a4->sin_addr));
924 #ifdef INET6
925 case AF_INET6:
926 a6 = (const struct sockaddr_in6 *) a;
927 b6 = (const struct sockaddr_in6 *) b;
928 return !memcmp(&a6->sin6_addr, &b6->sin6_addr,
929 sizeof(a6->sin6_addr));
930 #endif
931 }
932
933 return (0);
934 }
935
936 /*
937 * Check for idle hosts and stop monitoring them. We could also free
938 * the host structure here, possibly after a larger timeout but that
939 * would require some care to avoid races with
940 * e.g. nlm_host_lock_count_sysctl.
941 */
942 static void
nlm_check_idle(void)943 nlm_check_idle(void)
944 {
945 struct nlm_host *host;
946
947 mtx_assert(&nlm_global_lock, MA_OWNED);
948
949 if (time_uptime <= nlm_next_idle_check)
950 return;
951
952 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD;
953
954 TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
955 if (host->nh_monstate == NLM_MONITORED
956 && time_uptime > host->nh_idle_timeout) {
957 mtx_unlock(&nlm_global_lock);
958 if (lf_countlocks(host->nh_sysid) > 0
959 || lf_countlocks(NLM_SYSID_CLIENT
960 + host->nh_sysid)) {
961 host->nh_idle_timeout =
962 time_uptime + NLM_IDLE_TIMEOUT;
963 mtx_lock(&nlm_global_lock);
964 continue;
965 }
966 nlm_host_unmonitor(host);
967 mtx_lock(&nlm_global_lock);
968 }
969 }
970 }
971
972 /*
973 * Search for an existing NLM host that matches the given name
974 * (typically the caller_name element of an nlm4_lock). If none is
975 * found, create a new host. If 'addr' is non-NULL, record the remote
976 * address of the host so that we can call it back for async
977 * responses. If 'vers' is greater than zero then record the NLM
978 * program version to use to communicate with this client.
979 */
980 struct nlm_host *
nlm_find_host_by_name(const char * name,const struct sockaddr * addr,rpcvers_t vers)981 nlm_find_host_by_name(const char *name, const struct sockaddr *addr,
982 rpcvers_t vers)
983 {
984 struct nlm_host *host;
985
986 mtx_lock(&nlm_global_lock);
987
988 /*
989 * The remote host is determined by caller_name.
990 */
991 TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
992 if (!strcmp(host->nh_caller_name, name))
993 break;
994 }
995
996 if (!host) {
997 host = nlm_create_host(name);
998 if (!host) {
999 mtx_unlock(&nlm_global_lock);
1000 return (NULL);
1001 }
1002 }
1003 refcount_acquire(&host->nh_refs);
1004
1005 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT;
1006
1007 /*
1008 * If we have an address for the host, record it so that we
1009 * can send async replies etc.
1010 */
1011 if (addr) {
1012
1013 KASSERT(addr->sa_len < sizeof(struct sockaddr_storage),
1014 ("Strange remote transport address length"));
1015
1016 /*
1017 * If we have seen an address before and we currently
1018 * have an RPC client handle, make sure the address is
1019 * the same, otherwise discard the client handle.
1020 */
1021 if (host->nh_addr.ss_len && host->nh_srvrpc.nr_client) {
1022 if (!nlm_compare_addr(
1023 (struct sockaddr *) &host->nh_addr,
1024 addr)
1025 || host->nh_vers != vers) {
1026 CLIENT *client;
1027 mtx_lock(&host->nh_lock);
1028 client = host->nh_srvrpc.nr_client;
1029 host->nh_srvrpc.nr_client = NULL;
1030 mtx_unlock(&host->nh_lock);
1031 if (client) {
1032 CLNT_RELEASE(client);
1033 }
1034 }
1035 }
1036 memcpy(&host->nh_addr, addr, addr->sa_len);
1037 host->nh_vers = vers;
1038 }
1039
1040 nlm_check_idle();
1041
1042 mtx_unlock(&nlm_global_lock);
1043
1044 return (host);
1045 }
1046
1047 /*
1048 * Search for an existing NLM host that matches the given remote
1049 * address. If none is found, create a new host with the requested
1050 * address and remember 'vers' as the NLM protocol version to use for
1051 * that host.
1052 */
1053 struct nlm_host *
nlm_find_host_by_addr(const struct sockaddr * addr,int vers)1054 nlm_find_host_by_addr(const struct sockaddr *addr, int vers)
1055 {
1056 /*
1057 * Fake up a name using inet_ntop. This buffer is
1058 * large enough for an IPv6 address.
1059 */
1060 char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"];
1061 struct nlm_host *host;
1062
1063 switch (addr->sa_family) {
1064 case AF_INET:
1065 inet_ntop(AF_INET,
1066 &((const struct sockaddr_in *) addr)->sin_addr,
1067 tmp, sizeof tmp);
1068 break;
1069 #ifdef INET6
1070 case AF_INET6:
1071 inet_ntop(AF_INET6,
1072 &((const struct sockaddr_in6 *) addr)->sin6_addr,
1073 tmp, sizeof tmp);
1074 break;
1075 #endif
1076 default:
1077 strcpy(tmp, "<unknown>");
1078 }
1079
1080
1081 mtx_lock(&nlm_global_lock);
1082
1083 /*
1084 * The remote host is determined by caller_name.
1085 */
1086 TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
1087 if (nlm_compare_addr(addr,
1088 (const struct sockaddr *) &host->nh_addr))
1089 break;
1090 }
1091
1092 if (!host) {
1093 host = nlm_create_host(tmp);
1094 if (!host) {
1095 mtx_unlock(&nlm_global_lock);
1096 return (NULL);
1097 }
1098 memcpy(&host->nh_addr, addr, addr->sa_len);
1099 host->nh_vers = vers;
1100 }
1101 refcount_acquire(&host->nh_refs);
1102
1103 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT;
1104
1105 nlm_check_idle();
1106
1107 mtx_unlock(&nlm_global_lock);
1108
1109 return (host);
1110 }
1111
1112 /*
1113 * Find the NLM host that matches the value of 'sysid'. If none
1114 * exists, return NULL.
1115 */
1116 static struct nlm_host *
nlm_find_host_by_sysid(int sysid)1117 nlm_find_host_by_sysid(int sysid)
1118 {
1119 struct nlm_host *host;
1120
1121 TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
1122 if (host->nh_sysid == sysid) {
1123 refcount_acquire(&host->nh_refs);
1124 return (host);
1125 }
1126 }
1127
1128 return (NULL);
1129 }
1130
nlm_host_release(struct nlm_host * host)1131 void nlm_host_release(struct nlm_host *host)
1132 {
1133 if (refcount_release(&host->nh_refs)) {
1134 /*
1135 * Free the host
1136 */
1137 nlm_host_destroy(host);
1138 }
1139 }
1140
1141 /*
1142 * Unregister this NLM host with the local NSM due to idleness.
1143 */
1144 static void
nlm_host_unmonitor(struct nlm_host * host)1145 nlm_host_unmonitor(struct nlm_host *host)
1146 {
1147 mon_id smmonid;
1148 sm_stat_res smstat;
1149 struct timeval timo;
1150 enum clnt_stat stat;
1151
1152 NLM_DEBUG(1, "NLM: unmonitoring %s (sysid %d)\n",
1153 host->nh_caller_name, host->nh_sysid);
1154
1155 /*
1156 * We put our assigned system ID value in the priv field to
1157 * make it simpler to find the host if we are notified of a
1158 * host restart.
1159 */
1160 smmonid.mon_name = host->nh_caller_name;
1161 smmonid.my_id.my_name = "localhost";
1162 smmonid.my_id.my_prog = NLM_PROG;
1163 smmonid.my_id.my_vers = NLM_SM;
1164 smmonid.my_id.my_proc = NLM_SM_NOTIFY;
1165
1166 timo.tv_sec = 25;
1167 timo.tv_usec = 0;
1168 stat = CLNT_CALL(nlm_nsm, SM_UNMON,
1169 (xdrproc_t) xdr_mon, &smmonid,
1170 (xdrproc_t) xdr_sm_stat, &smstat, timo);
1171
1172 if (stat != RPC_SUCCESS) {
1173 NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat);
1174 return;
1175 }
1176 if (smstat.res_stat == stat_fail) {
1177 NLM_ERR("Local NSM refuses to unmonitor %s\n",
1178 host->nh_caller_name);
1179 return;
1180 }
1181
1182 host->nh_monstate = NLM_UNMONITORED;
1183 }
1184
1185 /*
1186 * Register this NLM host with the local NSM so that we can be
1187 * notified if it reboots.
1188 */
1189 void
nlm_host_monitor(struct nlm_host * host,int state)1190 nlm_host_monitor(struct nlm_host *host, int state)
1191 {
1192 mon smmon;
1193 sm_stat_res smstat;
1194 struct timeval timo;
1195 enum clnt_stat stat;
1196
1197 if (state && !host->nh_state) {
1198 /*
1199 * This is the first time we have seen an NSM state
1200 * value for this host. We record it here to help
1201 * detect host reboots.
1202 */
1203 host->nh_state = state;
1204 NLM_DEBUG(1, "NLM: host %s (sysid %d) has NSM state %d\n",
1205 host->nh_caller_name, host->nh_sysid, state);
1206 }
1207
1208 mtx_lock(&host->nh_lock);
1209 if (host->nh_monstate != NLM_UNMONITORED) {
1210 mtx_unlock(&host->nh_lock);
1211 return;
1212 }
1213 host->nh_monstate = NLM_MONITORED;
1214 mtx_unlock(&host->nh_lock);
1215
1216 NLM_DEBUG(1, "NLM: monitoring %s (sysid %d)\n",
1217 host->nh_caller_name, host->nh_sysid);
1218
1219 /*
1220 * We put our assigned system ID value in the priv field to
1221 * make it simpler to find the host if we are notified of a
1222 * host restart.
1223 */
1224 smmon.mon_id.mon_name = host->nh_caller_name;
1225 smmon.mon_id.my_id.my_name = "localhost";
1226 smmon.mon_id.my_id.my_prog = NLM_PROG;
1227 smmon.mon_id.my_id.my_vers = NLM_SM;
1228 smmon.mon_id.my_id.my_proc = NLM_SM_NOTIFY;
1229 memcpy(smmon.priv, &host->nh_sysid, sizeof(host->nh_sysid));
1230
1231 timo.tv_sec = 25;
1232 timo.tv_usec = 0;
1233 stat = CLNT_CALL(nlm_nsm, SM_MON,
1234 (xdrproc_t) xdr_mon, &smmon,
1235 (xdrproc_t) xdr_sm_stat, &smstat, timo);
1236
1237 if (stat != RPC_SUCCESS) {
1238 NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat);
1239 return;
1240 }
1241 if (smstat.res_stat == stat_fail) {
1242 NLM_ERR("Local NSM refuses to monitor %s\n",
1243 host->nh_caller_name);
1244 mtx_lock(&host->nh_lock);
1245 host->nh_monstate = NLM_MONITOR_FAILED;
1246 mtx_unlock(&host->nh_lock);
1247 return;
1248 }
1249
1250 host->nh_monstate = NLM_MONITORED;
1251 }
1252
1253 /*
1254 * Return an RPC client handle that can be used to talk to the NLM
1255 * running on the given host.
1256 */
1257 CLIENT *
nlm_host_get_rpc(struct nlm_host * host,bool_t isserver)1258 nlm_host_get_rpc(struct nlm_host *host, bool_t isserver)
1259 {
1260 struct nlm_rpc *rpc;
1261 CLIENT *client;
1262
1263 mtx_lock(&host->nh_lock);
1264
1265 if (isserver)
1266 rpc = &host->nh_srvrpc;
1267 else
1268 rpc = &host->nh_clntrpc;
1269
1270 /*
1271 * We can't hold onto RPC handles for too long - the async
1272 * call/reply protocol used by some NLM clients makes it hard
1273 * to tell when they change port numbers (e.g. after a
1274 * reboot). Note that if a client reboots while it isn't
1275 * holding any locks, it won't bother to notify us. We
1276 * expire the RPC handles after two minutes.
1277 */
1278 if (rpc->nr_client && time_uptime > rpc->nr_create_time + 2*60) {
1279 client = rpc->nr_client;
1280 rpc->nr_client = NULL;
1281 mtx_unlock(&host->nh_lock);
1282 CLNT_RELEASE(client);
1283 mtx_lock(&host->nh_lock);
1284 }
1285
1286 if (!rpc->nr_client) {
1287 mtx_unlock(&host->nh_lock);
1288 client = nlm_get_rpc((struct sockaddr *)&host->nh_addr,
1289 NLM_PROG, host->nh_vers);
1290 mtx_lock(&host->nh_lock);
1291
1292 if (client) {
1293 if (rpc->nr_client) {
1294 mtx_unlock(&host->nh_lock);
1295 CLNT_DESTROY(client);
1296 mtx_lock(&host->nh_lock);
1297 } else {
1298 rpc->nr_client = client;
1299 rpc->nr_create_time = time_uptime;
1300 }
1301 }
1302 }
1303
1304 client = rpc->nr_client;
1305 if (client)
1306 CLNT_ACQUIRE(client);
1307 mtx_unlock(&host->nh_lock);
1308
1309 return (client);
1310
1311 }
1312
nlm_host_get_sysid(struct nlm_host * host)1313 int nlm_host_get_sysid(struct nlm_host *host)
1314 {
1315
1316 return (host->nh_sysid);
1317 }
1318
1319 int
nlm_host_get_state(struct nlm_host * host)1320 nlm_host_get_state(struct nlm_host *host)
1321 {
1322
1323 return (host->nh_state);
1324 }
1325
1326 void *
nlm_register_wait_lock(struct nlm4_lock * lock,struct vnode * vp)1327 nlm_register_wait_lock(struct nlm4_lock *lock, struct vnode *vp)
1328 {
1329 struct nlm_waiting_lock *nw;
1330
1331 nw = malloc(sizeof(struct nlm_waiting_lock), M_NLM, M_WAITOK);
1332 nw->nw_lock = *lock;
1333 memcpy(&nw->nw_fh.fh_bytes, nw->nw_lock.fh.n_bytes,
1334 nw->nw_lock.fh.n_len);
1335 nw->nw_lock.fh.n_bytes = nw->nw_fh.fh_bytes;
1336 nw->nw_waiting = TRUE;
1337 nw->nw_vp = vp;
1338 mtx_lock(&nlm_global_lock);
1339 TAILQ_INSERT_TAIL(&nlm_waiting_locks, nw, nw_link);
1340 mtx_unlock(&nlm_global_lock);
1341
1342 return nw;
1343 }
1344
1345 void
nlm_deregister_wait_lock(void * handle)1346 nlm_deregister_wait_lock(void *handle)
1347 {
1348 struct nlm_waiting_lock *nw = handle;
1349
1350 mtx_lock(&nlm_global_lock);
1351 TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link);
1352 mtx_unlock(&nlm_global_lock);
1353
1354 free(nw, M_NLM);
1355 }
1356
1357 int
nlm_wait_lock(void * handle,int timo)1358 nlm_wait_lock(void *handle, int timo)
1359 {
1360 struct nlm_waiting_lock *nw = handle;
1361 int error;
1362
1363 /*
1364 * If the granted message arrived before we got here,
1365 * nw->nw_waiting will be FALSE - in that case, don't sleep.
1366 */
1367 mtx_lock(&nlm_global_lock);
1368 error = 0;
1369 if (nw->nw_waiting)
1370 error = msleep(nw, &nlm_global_lock, PCATCH, "nlmlock", timo);
1371 TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link);
1372 if (error) {
1373 /*
1374 * The granted message may arrive after the
1375 * interrupt/timeout but before we manage to lock the
1376 * mutex. Detect this by examining nw_lock.
1377 */
1378 if (!nw->nw_waiting)
1379 error = 0;
1380 } else {
1381 /*
1382 * If nlm_cancel_wait is called, then error will be
1383 * zero but nw_waiting will still be TRUE. We
1384 * translate this into EINTR.
1385 */
1386 if (nw->nw_waiting)
1387 error = EINTR;
1388 }
1389 mtx_unlock(&nlm_global_lock);
1390
1391 free(nw, M_NLM);
1392
1393 return (error);
1394 }
1395
1396 void
nlm_cancel_wait(struct vnode * vp)1397 nlm_cancel_wait(struct vnode *vp)
1398 {
1399 struct nlm_waiting_lock *nw;
1400
1401 mtx_lock(&nlm_global_lock);
1402 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
1403 if (nw->nw_vp == vp) {
1404 wakeup(nw);
1405 }
1406 }
1407 mtx_unlock(&nlm_global_lock);
1408 }
1409
1410
1411 /**********************************************************************/
1412
1413 /*
1414 * Syscall interface with userland.
1415 */
1416
1417 extern void nlm_prog_0(struct svc_req *rqstp, SVCXPRT *transp);
1418 extern void nlm_prog_1(struct svc_req *rqstp, SVCXPRT *transp);
1419 extern void nlm_prog_3(struct svc_req *rqstp, SVCXPRT *transp);
1420 extern void nlm_prog_4(struct svc_req *rqstp, SVCXPRT *transp);
1421
1422 static int
nlm_register_services(SVCPOOL * pool,int addr_count,char ** addrs)1423 nlm_register_services(SVCPOOL *pool, int addr_count, char **addrs)
1424 {
1425 static rpcvers_t versions[] = {
1426 NLM_SM, NLM_VERS, NLM_VERSX, NLM_VERS4
1427 };
1428 static void (*dispatchers[])(struct svc_req *, SVCXPRT *) = {
1429 nlm_prog_0, nlm_prog_1, nlm_prog_3, nlm_prog_4
1430 };
1431 static const int version_count = sizeof(versions) / sizeof(versions[0]);
1432
1433 SVCXPRT **xprts;
1434 char netid[16];
1435 char uaddr[128];
1436 struct netconfig *nconf;
1437 int i, j, error;
1438
1439 if (!addr_count) {
1440 NLM_ERR("NLM: no service addresses given - can't start server");
1441 return (EINVAL);
1442 }
1443
1444 xprts = malloc(addr_count * sizeof(SVCXPRT *), M_NLM, M_WAITOK|M_ZERO);
1445 for (i = 0; i < version_count; i++) {
1446 for (j = 0; j < addr_count; j++) {
1447 /*
1448 * Create transports for the first version and
1449 * then just register everything else to the
1450 * same transports.
1451 */
1452 if (i == 0) {
1453 char *up;
1454
1455 error = copyin(&addrs[2*j], &up,
1456 sizeof(char*));
1457 if (error)
1458 goto out;
1459 error = copyinstr(up, netid, sizeof(netid),
1460 NULL);
1461 if (error)
1462 goto out;
1463 error = copyin(&addrs[2*j+1], &up,
1464 sizeof(char*));
1465 if (error)
1466 goto out;
1467 error = copyinstr(up, uaddr, sizeof(uaddr),
1468 NULL);
1469 if (error)
1470 goto out;
1471 nconf = getnetconfigent(netid);
1472 if (!nconf) {
1473 NLM_ERR("Can't lookup netid %s\n",
1474 netid);
1475 error = EINVAL;
1476 goto out;
1477 }
1478 xprts[j] = svc_tp_create(pool, dispatchers[i],
1479 NLM_PROG, versions[i], uaddr, nconf);
1480 if (!xprts[j]) {
1481 NLM_ERR("NLM: unable to create "
1482 "(NLM_PROG, %d).\n", versions[i]);
1483 error = EINVAL;
1484 goto out;
1485 }
1486 freenetconfigent(nconf);
1487 } else {
1488 nconf = getnetconfigent(xprts[j]->xp_netid);
1489 rpcb_unset(NLM_PROG, versions[i], nconf);
1490 if (!svc_reg(xprts[j], NLM_PROG, versions[i],
1491 dispatchers[i], nconf)) {
1492 NLM_ERR("NLM: can't register "
1493 "(NLM_PROG, %d)\n", versions[i]);
1494 error = EINVAL;
1495 goto out;
1496 }
1497 }
1498 }
1499 }
1500 error = 0;
1501 out:
1502 for (j = 0; j < addr_count; j++) {
1503 if (xprts[j])
1504 SVC_RELEASE(xprts[j]);
1505 }
1506 free(xprts, M_NLM);
1507 return (error);
1508 }
1509
1510 /*
1511 * Main server entry point. Contacts the local NSM to get its current
1512 * state and send SM_UNMON_ALL. Registers the NLM services and then
1513 * services requests. Does not return until the server is interrupted
1514 * by a signal.
1515 */
1516 static int
nlm_server_main(int addr_count,char ** addrs)1517 nlm_server_main(int addr_count, char **addrs)
1518 {
1519 struct thread *td = curthread;
1520 int error;
1521 SVCPOOL *pool = NULL;
1522 struct sockopt opt;
1523 int portlow;
1524 #ifdef INET6
1525 struct sockaddr_in6 sin6;
1526 #endif
1527 struct sockaddr_in sin;
1528 my_id id;
1529 sm_stat smstat;
1530 struct timeval timo;
1531 enum clnt_stat stat;
1532 struct nlm_host *host, *nhost;
1533 struct nlm_waiting_lock *nw;
1534 vop_advlock_t *old_nfs_advlock;
1535 vop_reclaim_t *old_nfs_reclaim;
1536
1537 if (nlm_is_running != 0) {
1538 NLM_ERR("NLM: can't start server - "
1539 "it appears to be running already\n");
1540 return (EPERM);
1541 }
1542
1543 if (nlm_socket == NULL) {
1544 memset(&opt, 0, sizeof(opt));
1545
1546 error = socreate(AF_INET, &nlm_socket, SOCK_DGRAM, 0,
1547 td->td_ucred, td);
1548 if (error) {
1549 NLM_ERR("NLM: can't create IPv4 socket - error %d\n",
1550 error);
1551 return (error);
1552 }
1553 opt.sopt_dir = SOPT_SET;
1554 opt.sopt_level = IPPROTO_IP;
1555 opt.sopt_name = IP_PORTRANGE;
1556 portlow = IP_PORTRANGE_LOW;
1557 opt.sopt_val = &portlow;
1558 opt.sopt_valsize = sizeof(portlow);
1559 sosetopt(nlm_socket, &opt);
1560
1561 #ifdef INET6
1562 nlm_socket6 = NULL;
1563 error = socreate(AF_INET6, &nlm_socket6, SOCK_DGRAM, 0,
1564 td->td_ucred, td);
1565 if (error) {
1566 NLM_ERR("NLM: can't create IPv6 socket - error %d\n",
1567 error);
1568 soclose(nlm_socket);
1569 nlm_socket = NULL;
1570 return (error);
1571 }
1572 opt.sopt_dir = SOPT_SET;
1573 opt.sopt_level = IPPROTO_IPV6;
1574 opt.sopt_name = IPV6_PORTRANGE;
1575 portlow = IPV6_PORTRANGE_LOW;
1576 opt.sopt_val = &portlow;
1577 opt.sopt_valsize = sizeof(portlow);
1578 sosetopt(nlm_socket6, &opt);
1579 #endif
1580 }
1581
1582 nlm_auth = authunix_create(curthread->td_ucred);
1583
1584 #ifdef INET6
1585 memset(&sin6, 0, sizeof(sin6));
1586 sin6.sin6_len = sizeof(sin6);
1587 sin6.sin6_family = AF_INET6;
1588 sin6.sin6_addr = in6addr_loopback;
1589 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin6, SM_PROG, SM_VERS);
1590 if (!nlm_nsm) {
1591 #endif
1592 memset(&sin, 0, sizeof(sin));
1593 sin.sin_len = sizeof(sin);
1594 sin.sin_family = AF_INET;
1595 sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1596 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin, SM_PROG,
1597 SM_VERS);
1598 #ifdef INET6
1599 }
1600 #endif
1601
1602 if (!nlm_nsm) {
1603 NLM_ERR("Can't start NLM - unable to contact NSM\n");
1604 error = EINVAL;
1605 goto out;
1606 }
1607
1608 pool = svcpool_create("NLM", NULL);
1609
1610 error = nlm_register_services(pool, addr_count, addrs);
1611 if (error)
1612 goto out;
1613
1614 memset(&id, 0, sizeof(id));
1615 id.my_name = "NFS NLM";
1616
1617 timo.tv_sec = 25;
1618 timo.tv_usec = 0;
1619 stat = CLNT_CALL(nlm_nsm, SM_UNMON_ALL,
1620 (xdrproc_t) xdr_my_id, &id,
1621 (xdrproc_t) xdr_sm_stat, &smstat, timo);
1622
1623 if (stat != RPC_SUCCESS) {
1624 struct rpc_err err;
1625
1626 CLNT_GETERR(nlm_nsm, &err);
1627 NLM_ERR("NLM: unexpected error contacting NSM, "
1628 "stat=%d, errno=%d\n", stat, err.re_errno);
1629 error = EINVAL;
1630 goto out;
1631 }
1632 nlm_is_running = 1;
1633
1634 NLM_DEBUG(1, "NLM: local NSM state is %d\n", smstat.state);
1635 nlm_nsm_state = smstat.state;
1636
1637 old_nfs_advlock = nfs_advlock_p;
1638 nfs_advlock_p = nlm_advlock;
1639 old_nfs_reclaim = nfs_reclaim_p;
1640 nfs_reclaim_p = nlm_reclaim;
1641
1642 svc_run(pool);
1643 error = 0;
1644
1645 nfs_advlock_p = old_nfs_advlock;
1646 nfs_reclaim_p = old_nfs_reclaim;
1647
1648 out:
1649 nlm_is_running = 0;
1650 if (pool)
1651 svcpool_destroy(pool);
1652
1653 /*
1654 * We are finished communicating with the NSM.
1655 */
1656 if (nlm_nsm) {
1657 CLNT_RELEASE(nlm_nsm);
1658 nlm_nsm = NULL;
1659 }
1660
1661 /*
1662 * Trash all the existing state so that if the server
1663 * restarts, it gets a clean slate. This is complicated by the
1664 * possibility that there may be other threads trying to make
1665 * client locking requests.
1666 *
1667 * First we fake a client reboot notification which will
1668 * cancel any pending async locks and purge remote lock state
1669 * from the local lock manager. We release the reference from
1670 * nlm_hosts to the host (which may remove it from the list
1671 * and free it). After this phase, the only entries in the
1672 * nlm_host list should be from other threads performing
1673 * client lock requests.
1674 */
1675 mtx_lock(&nlm_global_lock);
1676 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
1677 wakeup(nw);
1678 }
1679 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) {
1680 mtx_unlock(&nlm_global_lock);
1681 nlm_host_notify(host, 0);
1682 nlm_host_release(host);
1683 mtx_lock(&nlm_global_lock);
1684 }
1685 mtx_unlock(&nlm_global_lock);
1686
1687 AUTH_DESTROY(nlm_auth);
1688
1689 return (error);
1690 }
1691
1692 int
sys_nlm_syscall(struct thread * td,struct nlm_syscall_args * uap)1693 sys_nlm_syscall(struct thread *td, struct nlm_syscall_args *uap)
1694 {
1695 int error;
1696
1697 #if __FreeBSD_version >= 700000
1698 error = priv_check(td, PRIV_NFS_LOCKD);
1699 #else
1700 error = suser(td);
1701 #endif
1702 if (error)
1703 return (error);
1704
1705 nlm_debug_level = uap->debug_level;
1706 nlm_grace_threshold = time_uptime + uap->grace_period;
1707 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD;
1708
1709 return nlm_server_main(uap->addr_count, uap->addrs);
1710 }
1711
1712 /**********************************************************************/
1713
1714 /*
1715 * NLM implementation details, called from the RPC stubs.
1716 */
1717
1718
1719 void
nlm_sm_notify(struct nlm_sm_status * argp)1720 nlm_sm_notify(struct nlm_sm_status *argp)
1721 {
1722 uint32_t sysid;
1723 struct nlm_host *host;
1724
1725 NLM_DEBUG(3, "nlm_sm_notify(): mon_name = %s\n", argp->mon_name);
1726 memcpy(&sysid, &argp->priv, sizeof(sysid));
1727 host = nlm_find_host_by_sysid(sysid);
1728 if (host) {
1729 nlm_host_notify(host, argp->state);
1730 nlm_host_release(host);
1731 }
1732 }
1733
1734 static void
nlm_convert_to_fhandle_t(fhandle_t * fhp,struct netobj * p)1735 nlm_convert_to_fhandle_t(fhandle_t *fhp, struct netobj *p)
1736 {
1737 memcpy(fhp, p->n_bytes, sizeof(fhandle_t));
1738 }
1739
1740 struct vfs_state {
1741 struct mount *vs_mp;
1742 struct vnode *vs_vp;
1743 int vs_vnlocked;
1744 };
1745
1746 static int
nlm_get_vfs_state(struct nlm_host * host,struct svc_req * rqstp,fhandle_t * fhp,struct vfs_state * vs,accmode_t accmode)1747 nlm_get_vfs_state(struct nlm_host *host, struct svc_req *rqstp,
1748 fhandle_t *fhp, struct vfs_state *vs, accmode_t accmode)
1749 {
1750 int error, exflags;
1751 struct ucred *cred = NULL, *credanon = NULL;
1752
1753 memset(vs, 0, sizeof(*vs));
1754
1755 vs->vs_mp = vfs_getvfs(&fhp->fh_fsid);
1756 if (!vs->vs_mp) {
1757 return (ESTALE);
1758 }
1759
1760 /* accmode == 0 means don't check, since it is an unlock. */
1761 if (accmode != 0) {
1762 error = VFS_CHECKEXP(vs->vs_mp,
1763 (struct sockaddr *)&host->nh_addr, &exflags, &credanon,
1764 NULL, NULL);
1765 if (error)
1766 goto out;
1767
1768 if (exflags & MNT_EXRDONLY ||
1769 (vs->vs_mp->mnt_flag & MNT_RDONLY)) {
1770 error = EROFS;
1771 goto out;
1772 }
1773 }
1774
1775 error = VFS_FHTOVP(vs->vs_mp, &fhp->fh_fid, LK_EXCLUSIVE, &vs->vs_vp);
1776 if (error)
1777 goto out;
1778 vs->vs_vnlocked = TRUE;
1779
1780 if (accmode != 0) {
1781 if (!svc_getcred(rqstp, &cred, NULL)) {
1782 error = EINVAL;
1783 goto out;
1784 }
1785 if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
1786 crfree(cred);
1787 cred = credanon;
1788 credanon = NULL;
1789 }
1790
1791 /*
1792 * Check cred.
1793 */
1794 error = VOP_ACCESS(vs->vs_vp, accmode, cred, curthread);
1795 /*
1796 * If this failed and accmode != VWRITE, try again with
1797 * VWRITE to maintain backwards compatibility with the
1798 * old code that always used VWRITE.
1799 */
1800 if (error != 0 && accmode != VWRITE)
1801 error = VOP_ACCESS(vs->vs_vp, VWRITE, cred, curthread);
1802 if (error)
1803 goto out;
1804 }
1805
1806 #if __FreeBSD_version < 800011
1807 VOP_UNLOCK(vs->vs_vp, 0, curthread);
1808 #else
1809 VOP_UNLOCK(vs->vs_vp, 0);
1810 #endif
1811 vs->vs_vnlocked = FALSE;
1812
1813 out:
1814 if (cred)
1815 crfree(cred);
1816 if (credanon)
1817 crfree(credanon);
1818
1819 return (error);
1820 }
1821
1822 static void
nlm_release_vfs_state(struct vfs_state * vs)1823 nlm_release_vfs_state(struct vfs_state *vs)
1824 {
1825
1826 if (vs->vs_vp) {
1827 if (vs->vs_vnlocked)
1828 vput(vs->vs_vp);
1829 else
1830 vrele(vs->vs_vp);
1831 }
1832 if (vs->vs_mp)
1833 vfs_rel(vs->vs_mp);
1834 }
1835
1836 static nlm4_stats
nlm_convert_error(int error)1837 nlm_convert_error(int error)
1838 {
1839
1840 if (error == ESTALE)
1841 return nlm4_stale_fh;
1842 else if (error == EROFS)
1843 return nlm4_rofs;
1844 else
1845 return nlm4_failed;
1846 }
1847
1848 int
nlm_do_test(nlm4_testargs * argp,nlm4_testres * result,struct svc_req * rqstp,CLIENT ** rpcp)1849 nlm_do_test(nlm4_testargs *argp, nlm4_testres *result, struct svc_req *rqstp,
1850 CLIENT **rpcp)
1851 {
1852 fhandle_t fh;
1853 struct vfs_state vs;
1854 struct nlm_host *host, *bhost;
1855 int error, sysid;
1856 struct flock fl;
1857 accmode_t accmode;
1858
1859 memset(result, 0, sizeof(*result));
1860 memset(&vs, 0, sizeof(vs));
1861
1862 host = nlm_find_host_by_name(argp->alock.caller_name,
1863 svc_getrpccaller(rqstp), rqstp->rq_vers);
1864 if (!host) {
1865 result->stat.stat = nlm4_denied_nolocks;
1866 return (ENOMEM);
1867 }
1868
1869 NLM_DEBUG(3, "nlm_do_test(): caller_name = %s (sysid = %d)\n",
1870 host->nh_caller_name, host->nh_sysid);
1871
1872 nlm_check_expired_locks(host);
1873 sysid = host->nh_sysid;
1874
1875 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
1876 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
1877
1878 if (time_uptime < nlm_grace_threshold) {
1879 result->stat.stat = nlm4_denied_grace_period;
1880 goto out;
1881 }
1882
1883 accmode = argp->exclusive ? VWRITE : VREAD;
1884 error = nlm_get_vfs_state(host, rqstp, &fh, &vs, accmode);
1885 if (error) {
1886 result->stat.stat = nlm_convert_error(error);
1887 goto out;
1888 }
1889
1890 fl.l_start = argp->alock.l_offset;
1891 fl.l_len = argp->alock.l_len;
1892 fl.l_pid = argp->alock.svid;
1893 fl.l_sysid = sysid;
1894 fl.l_whence = SEEK_SET;
1895 if (argp->exclusive)
1896 fl.l_type = F_WRLCK;
1897 else
1898 fl.l_type = F_RDLCK;
1899 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_GETLK, &fl, F_REMOTE);
1900 if (error) {
1901 result->stat.stat = nlm4_failed;
1902 goto out;
1903 }
1904
1905 if (fl.l_type == F_UNLCK) {
1906 result->stat.stat = nlm4_granted;
1907 } else {
1908 result->stat.stat = nlm4_denied;
1909 result->stat.nlm4_testrply_u.holder.exclusive =
1910 (fl.l_type == F_WRLCK);
1911 result->stat.nlm4_testrply_u.holder.svid = fl.l_pid;
1912 bhost = nlm_find_host_by_sysid(fl.l_sysid);
1913 if (bhost) {
1914 /*
1915 * We don't have any useful way of recording
1916 * the value of oh used in the original lock
1917 * request. Ideally, the test reply would have
1918 * a space for the owning host's name allowing
1919 * our caller's NLM to keep track.
1920 *
1921 * As far as I can see, Solaris uses an eight
1922 * byte structure for oh which contains a four
1923 * byte pid encoded in local byte order and
1924 * the first four bytes of the host
1925 * name. Linux uses a variable length string
1926 * 'pid@hostname' in ascii but doesn't even
1927 * return that in test replies.
1928 *
1929 * For the moment, return nothing in oh
1930 * (already zero'ed above).
1931 */
1932 nlm_host_release(bhost);
1933 }
1934 result->stat.nlm4_testrply_u.holder.l_offset = fl.l_start;
1935 result->stat.nlm4_testrply_u.holder.l_len = fl.l_len;
1936 }
1937
1938 out:
1939 nlm_release_vfs_state(&vs);
1940 if (rpcp)
1941 *rpcp = nlm_host_get_rpc(host, TRUE);
1942 nlm_host_release(host);
1943 return (0);
1944 }
1945
1946 int
nlm_do_lock(nlm4_lockargs * argp,nlm4_res * result,struct svc_req * rqstp,bool_t monitor,CLIENT ** rpcp)1947 nlm_do_lock(nlm4_lockargs *argp, nlm4_res *result, struct svc_req *rqstp,
1948 bool_t monitor, CLIENT **rpcp)
1949 {
1950 fhandle_t fh;
1951 struct vfs_state vs;
1952 struct nlm_host *host;
1953 int error, sysid;
1954 struct flock fl;
1955 accmode_t accmode;
1956
1957 memset(result, 0, sizeof(*result));
1958 memset(&vs, 0, sizeof(vs));
1959
1960 host = nlm_find_host_by_name(argp->alock.caller_name,
1961 svc_getrpccaller(rqstp), rqstp->rq_vers);
1962 if (!host) {
1963 result->stat.stat = nlm4_denied_nolocks;
1964 return (ENOMEM);
1965 }
1966
1967 NLM_DEBUG(3, "nlm_do_lock(): caller_name = %s (sysid = %d)\n",
1968 host->nh_caller_name, host->nh_sysid);
1969
1970 if (monitor && host->nh_state && argp->state
1971 && host->nh_state != argp->state) {
1972 /*
1973 * The host rebooted without telling us. Trash its
1974 * locks.
1975 */
1976 nlm_host_notify(host, argp->state);
1977 }
1978
1979 nlm_check_expired_locks(host);
1980 sysid = host->nh_sysid;
1981
1982 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
1983 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
1984
1985 if (time_uptime < nlm_grace_threshold && !argp->reclaim) {
1986 result->stat.stat = nlm4_denied_grace_period;
1987 goto out;
1988 }
1989
1990 accmode = argp->exclusive ? VWRITE : VREAD;
1991 error = nlm_get_vfs_state(host, rqstp, &fh, &vs, accmode);
1992 if (error) {
1993 result->stat.stat = nlm_convert_error(error);
1994 goto out;
1995 }
1996
1997 fl.l_start = argp->alock.l_offset;
1998 fl.l_len = argp->alock.l_len;
1999 fl.l_pid = argp->alock.svid;
2000 fl.l_sysid = sysid;
2001 fl.l_whence = SEEK_SET;
2002 if (argp->exclusive)
2003 fl.l_type = F_WRLCK;
2004 else
2005 fl.l_type = F_RDLCK;
2006 if (argp->block) {
2007 struct nlm_async_lock *af;
2008 CLIENT *client;
2009 struct nlm_grantcookie cookie;
2010
2011 /*
2012 * First, make sure we can contact the host's NLM.
2013 */
2014 client = nlm_host_get_rpc(host, TRUE);
2015 if (!client) {
2016 result->stat.stat = nlm4_failed;
2017 goto out;
2018 }
2019
2020 /*
2021 * First we need to check and see if there is an
2022 * existing blocked lock that matches. This could be a
2023 * badly behaved client or an RPC re-send. If we find
2024 * one, just return nlm4_blocked.
2025 */
2026 mtx_lock(&host->nh_lock);
2027 TAILQ_FOREACH(af, &host->nh_pending, af_link) {
2028 if (af->af_fl.l_start == fl.l_start
2029 && af->af_fl.l_len == fl.l_len
2030 && af->af_fl.l_pid == fl.l_pid
2031 && af->af_fl.l_type == fl.l_type) {
2032 break;
2033 }
2034 }
2035 if (!af) {
2036 cookie.ng_sysid = host->nh_sysid;
2037 cookie.ng_cookie = host->nh_grantcookie++;
2038 }
2039 mtx_unlock(&host->nh_lock);
2040 if (af) {
2041 CLNT_RELEASE(client);
2042 result->stat.stat = nlm4_blocked;
2043 goto out;
2044 }
2045
2046 af = malloc(sizeof(struct nlm_async_lock), M_NLM,
2047 M_WAITOK|M_ZERO);
2048 TASK_INIT(&af->af_task, 0, nlm_lock_callback, af);
2049 af->af_vp = vs.vs_vp;
2050 af->af_fl = fl;
2051 af->af_host = host;
2052 af->af_rpc = client;
2053 /*
2054 * We use M_RPC here so that we can xdr_free the thing
2055 * later.
2056 */
2057 nlm_make_netobj(&af->af_granted.cookie,
2058 (caddr_t)&cookie, sizeof(cookie), M_RPC);
2059 af->af_granted.exclusive = argp->exclusive;
2060 af->af_granted.alock.caller_name =
2061 strdup(argp->alock.caller_name, M_RPC);
2062 nlm_copy_netobj(&af->af_granted.alock.fh,
2063 &argp->alock.fh, M_RPC);
2064 nlm_copy_netobj(&af->af_granted.alock.oh,
2065 &argp->alock.oh, M_RPC);
2066 af->af_granted.alock.svid = argp->alock.svid;
2067 af->af_granted.alock.l_offset = argp->alock.l_offset;
2068 af->af_granted.alock.l_len = argp->alock.l_len;
2069
2070 /*
2071 * Put the entry on the pending list before calling
2072 * VOP_ADVLOCKASYNC. We do this in case the lock
2073 * request was blocked (returning EINPROGRESS) but
2074 * then granted before we manage to run again. The
2075 * client may receive the granted message before we
2076 * send our blocked reply but thats their problem.
2077 */
2078 mtx_lock(&host->nh_lock);
2079 TAILQ_INSERT_TAIL(&host->nh_pending, af, af_link);
2080 mtx_unlock(&host->nh_lock);
2081
2082 error = VOP_ADVLOCKASYNC(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE,
2083 &af->af_task, &af->af_cookie);
2084
2085 /*
2086 * If the lock completed synchronously, just free the
2087 * tracking structure now.
2088 */
2089 if (error != EINPROGRESS) {
2090 CLNT_RELEASE(af->af_rpc);
2091 mtx_lock(&host->nh_lock);
2092 TAILQ_REMOVE(&host->nh_pending, af, af_link);
2093 mtx_unlock(&host->nh_lock);
2094 xdr_free((xdrproc_t) xdr_nlm4_testargs,
2095 &af->af_granted);
2096 free(af, M_NLM);
2097 } else {
2098 NLM_DEBUG(2, "NLM: pending async lock %p for %s "
2099 "(sysid %d)\n", af, host->nh_caller_name, sysid);
2100 /*
2101 * Don't vrele the vnode just yet - this must
2102 * wait until either the async callback
2103 * happens or the lock is cancelled.
2104 */
2105 vs.vs_vp = NULL;
2106 }
2107 } else {
2108 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE);
2109 }
2110
2111 if (error) {
2112 if (error == EINPROGRESS) {
2113 result->stat.stat = nlm4_blocked;
2114 } else if (error == EDEADLK) {
2115 result->stat.stat = nlm4_deadlck;
2116 } else if (error == EAGAIN) {
2117 result->stat.stat = nlm4_denied;
2118 } else {
2119 result->stat.stat = nlm4_failed;
2120 }
2121 } else {
2122 if (monitor)
2123 nlm_host_monitor(host, argp->state);
2124 result->stat.stat = nlm4_granted;
2125 }
2126
2127 out:
2128 nlm_release_vfs_state(&vs);
2129 if (rpcp)
2130 *rpcp = nlm_host_get_rpc(host, TRUE);
2131 nlm_host_release(host);
2132 return (0);
2133 }
2134
2135 int
nlm_do_cancel(nlm4_cancargs * argp,nlm4_res * result,struct svc_req * rqstp,CLIENT ** rpcp)2136 nlm_do_cancel(nlm4_cancargs *argp, nlm4_res *result, struct svc_req *rqstp,
2137 CLIENT **rpcp)
2138 {
2139 fhandle_t fh;
2140 struct vfs_state vs;
2141 struct nlm_host *host;
2142 int error, sysid;
2143 struct flock fl;
2144 struct nlm_async_lock *af;
2145
2146 memset(result, 0, sizeof(*result));
2147 memset(&vs, 0, sizeof(vs));
2148
2149 host = nlm_find_host_by_name(argp->alock.caller_name,
2150 svc_getrpccaller(rqstp), rqstp->rq_vers);
2151 if (!host) {
2152 result->stat.stat = nlm4_denied_nolocks;
2153 return (ENOMEM);
2154 }
2155
2156 NLM_DEBUG(3, "nlm_do_cancel(): caller_name = %s (sysid = %d)\n",
2157 host->nh_caller_name, host->nh_sysid);
2158
2159 nlm_check_expired_locks(host);
2160 sysid = host->nh_sysid;
2161
2162 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
2163 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2164
2165 if (time_uptime < nlm_grace_threshold) {
2166 result->stat.stat = nlm4_denied_grace_period;
2167 goto out;
2168 }
2169
2170 error = nlm_get_vfs_state(host, rqstp, &fh, &vs, (accmode_t)0);
2171 if (error) {
2172 result->stat.stat = nlm_convert_error(error);
2173 goto out;
2174 }
2175
2176 fl.l_start = argp->alock.l_offset;
2177 fl.l_len = argp->alock.l_len;
2178 fl.l_pid = argp->alock.svid;
2179 fl.l_sysid = sysid;
2180 fl.l_whence = SEEK_SET;
2181 if (argp->exclusive)
2182 fl.l_type = F_WRLCK;
2183 else
2184 fl.l_type = F_RDLCK;
2185
2186 /*
2187 * First we need to try and find the async lock request - if
2188 * there isn't one, we give up and return nlm4_denied.
2189 */
2190 mtx_lock(&host->nh_lock);
2191
2192 TAILQ_FOREACH(af, &host->nh_pending, af_link) {
2193 if (af->af_fl.l_start == fl.l_start
2194 && af->af_fl.l_len == fl.l_len
2195 && af->af_fl.l_pid == fl.l_pid
2196 && af->af_fl.l_type == fl.l_type) {
2197 break;
2198 }
2199 }
2200
2201 if (!af) {
2202 mtx_unlock(&host->nh_lock);
2203 result->stat.stat = nlm4_denied;
2204 goto out;
2205 }
2206
2207 error = nlm_cancel_async_lock(af);
2208
2209 if (error) {
2210 result->stat.stat = nlm4_denied;
2211 } else {
2212 result->stat.stat = nlm4_granted;
2213 }
2214
2215 mtx_unlock(&host->nh_lock);
2216
2217 out:
2218 nlm_release_vfs_state(&vs);
2219 if (rpcp)
2220 *rpcp = nlm_host_get_rpc(host, TRUE);
2221 nlm_host_release(host);
2222 return (0);
2223 }
2224
2225 int
nlm_do_unlock(nlm4_unlockargs * argp,nlm4_res * result,struct svc_req * rqstp,CLIENT ** rpcp)2226 nlm_do_unlock(nlm4_unlockargs *argp, nlm4_res *result, struct svc_req *rqstp,
2227 CLIENT **rpcp)
2228 {
2229 fhandle_t fh;
2230 struct vfs_state vs;
2231 struct nlm_host *host;
2232 int error, sysid;
2233 struct flock fl;
2234
2235 memset(result, 0, sizeof(*result));
2236 memset(&vs, 0, sizeof(vs));
2237
2238 host = nlm_find_host_by_name(argp->alock.caller_name,
2239 svc_getrpccaller(rqstp), rqstp->rq_vers);
2240 if (!host) {
2241 result->stat.stat = nlm4_denied_nolocks;
2242 return (ENOMEM);
2243 }
2244
2245 NLM_DEBUG(3, "nlm_do_unlock(): caller_name = %s (sysid = %d)\n",
2246 host->nh_caller_name, host->nh_sysid);
2247
2248 nlm_check_expired_locks(host);
2249 sysid = host->nh_sysid;
2250
2251 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
2252 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2253
2254 if (time_uptime < nlm_grace_threshold) {
2255 result->stat.stat = nlm4_denied_grace_period;
2256 goto out;
2257 }
2258
2259 error = nlm_get_vfs_state(host, rqstp, &fh, &vs, (accmode_t)0);
2260 if (error) {
2261 result->stat.stat = nlm_convert_error(error);
2262 goto out;
2263 }
2264
2265 fl.l_start = argp->alock.l_offset;
2266 fl.l_len = argp->alock.l_len;
2267 fl.l_pid = argp->alock.svid;
2268 fl.l_sysid = sysid;
2269 fl.l_whence = SEEK_SET;
2270 fl.l_type = F_UNLCK;
2271 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_UNLCK, &fl, F_REMOTE);
2272
2273 /*
2274 * Ignore the error - there is no result code for failure,
2275 * only for grace period.
2276 */
2277 result->stat.stat = nlm4_granted;
2278
2279 out:
2280 nlm_release_vfs_state(&vs);
2281 if (rpcp)
2282 *rpcp = nlm_host_get_rpc(host, TRUE);
2283 nlm_host_release(host);
2284 return (0);
2285 }
2286
2287 int
nlm_do_granted(nlm4_testargs * argp,nlm4_res * result,struct svc_req * rqstp,CLIENT ** rpcp)2288 nlm_do_granted(nlm4_testargs *argp, nlm4_res *result, struct svc_req *rqstp,
2289
2290 CLIENT **rpcp)
2291 {
2292 struct nlm_host *host;
2293 struct nlm_waiting_lock *nw;
2294
2295 memset(result, 0, sizeof(*result));
2296
2297 host = nlm_find_host_by_addr(svc_getrpccaller(rqstp), rqstp->rq_vers);
2298 if (!host) {
2299 result->stat.stat = nlm4_denied_nolocks;
2300 return (ENOMEM);
2301 }
2302
2303 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
2304 result->stat.stat = nlm4_denied;
2305 KFAIL_POINT_CODE(DEBUG_FP, nlm_deny_grant, goto out);
2306
2307 mtx_lock(&nlm_global_lock);
2308 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
2309 if (!nw->nw_waiting)
2310 continue;
2311 if (argp->alock.svid == nw->nw_lock.svid
2312 && argp->alock.l_offset == nw->nw_lock.l_offset
2313 && argp->alock.l_len == nw->nw_lock.l_len
2314 && argp->alock.fh.n_len == nw->nw_lock.fh.n_len
2315 && !memcmp(argp->alock.fh.n_bytes, nw->nw_lock.fh.n_bytes,
2316 nw->nw_lock.fh.n_len)) {
2317 nw->nw_waiting = FALSE;
2318 wakeup(nw);
2319 result->stat.stat = nlm4_granted;
2320 break;
2321 }
2322 }
2323 mtx_unlock(&nlm_global_lock);
2324
2325 out:
2326 if (rpcp)
2327 *rpcp = nlm_host_get_rpc(host, TRUE);
2328 nlm_host_release(host);
2329 return (0);
2330 }
2331
2332 void
nlm_do_granted_res(nlm4_res * argp,struct svc_req * rqstp)2333 nlm_do_granted_res(nlm4_res *argp, struct svc_req *rqstp)
2334 {
2335 struct nlm_host *host = NULL;
2336 struct nlm_async_lock *af = NULL;
2337 int error;
2338
2339 if (argp->cookie.n_len != sizeof(struct nlm_grantcookie)) {
2340 NLM_DEBUG(1, "NLM: bogus grant cookie");
2341 goto out;
2342 }
2343
2344 host = nlm_find_host_by_sysid(ng_sysid(&argp->cookie));
2345 if (!host) {
2346 NLM_DEBUG(1, "NLM: Unknown host rejected our grant");
2347 goto out;
2348 }
2349
2350 mtx_lock(&host->nh_lock);
2351 TAILQ_FOREACH(af, &host->nh_granted, af_link)
2352 if (ng_cookie(&argp->cookie) ==
2353 ng_cookie(&af->af_granted.cookie))
2354 break;
2355 if (af)
2356 TAILQ_REMOVE(&host->nh_granted, af, af_link);
2357 mtx_unlock(&host->nh_lock);
2358
2359 if (!af) {
2360 NLM_DEBUG(1, "NLM: host %s (sysid %d) replied to our grant "
2361 "with unrecognized cookie %d:%d", host->nh_caller_name,
2362 host->nh_sysid, ng_sysid(&argp->cookie),
2363 ng_cookie(&argp->cookie));
2364 goto out;
2365 }
2366
2367 if (argp->stat.stat != nlm4_granted) {
2368 af->af_fl.l_type = F_UNLCK;
2369 error = VOP_ADVLOCK(af->af_vp, NULL, F_UNLCK, &af->af_fl, F_REMOTE);
2370 if (error) {
2371 NLM_DEBUG(1, "NLM: host %s (sysid %d) rejected our grant "
2372 "and we failed to unlock (%d)", host->nh_caller_name,
2373 host->nh_sysid, error);
2374 goto out;
2375 }
2376
2377 NLM_DEBUG(5, "NLM: async lock %p rejected by host %s (sysid %d)",
2378 af, host->nh_caller_name, host->nh_sysid);
2379 } else {
2380 NLM_DEBUG(5, "NLM: async lock %p accepted by host %s (sysid %d)",
2381 af, host->nh_caller_name, host->nh_sysid);
2382 }
2383
2384 out:
2385 if (af)
2386 nlm_free_async_lock(af);
2387 if (host)
2388 nlm_host_release(host);
2389 }
2390
2391 void
nlm_do_free_all(nlm4_notify * argp)2392 nlm_do_free_all(nlm4_notify *argp)
2393 {
2394 struct nlm_host *host, *thost;
2395
2396 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, thost) {
2397 if (!strcmp(host->nh_caller_name, argp->name))
2398 nlm_host_notify(host, argp->state);
2399 }
2400 }
2401
2402 /*
2403 * Kernel module glue
2404 */
2405 static int
nfslockd_modevent(module_t mod,int type,void * data)2406 nfslockd_modevent(module_t mod, int type, void *data)
2407 {
2408
2409 switch (type) {
2410 case MOD_LOAD:
2411 return (0);
2412 case MOD_UNLOAD:
2413 /* The NLM module cannot be safely unloaded. */
2414 /* FALLTHROUGH */
2415 default:
2416 return (EOPNOTSUPP);
2417 }
2418 }
2419 static moduledata_t nfslockd_mod = {
2420 "nfslockd",
2421 nfslockd_modevent,
2422 NULL,
2423 };
2424 DECLARE_MODULE(nfslockd, nfslockd_mod, SI_SUB_VFS, SI_ORDER_ANY);
2425
2426 /* So that loader and kldload(2) can find us, wherever we are.. */
2427 MODULE_DEPEND(nfslockd, krpc, 1, 1, 1);
2428 MODULE_DEPEND(nfslockd, nfslock, 1, 1, 1);
2429 MODULE_VERSION(nfslockd, 1);
2430