xref: /illumos-gate/usr/src/uts/common/klm/nlm_impl.c (revision 0dfe541e)
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 /*
29  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
30  * Copyright (c) 2012 by Delphix. All rights reserved.
31  */
32 
33 /*
34  * NFS LockManager, start/stop, support functions, etc.
35  * Most of the interesting code is here.
36  *
37  * Source code derived from FreeBSD nlm_prot_impl.c
38  */
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/thread.h>
43 #include <sys/fcntl.h>
44 #include <sys/flock.h>
45 #include <sys/mount.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/share.h>
49 #include <sys/socket.h>
50 #include <sys/syscall.h>
51 #include <sys/syslog.h>
52 #include <sys/systm.h>
53 #include <sys/class.h>
54 #include <sys/unistd.h>
55 #include <sys/vnode.h>
56 #include <sys/vfs.h>
57 #include <sys/queue.h>
58 #include <sys/bitmap.h>
59 #include <sys/sdt.h>
60 #include <netinet/in.h>
61 
62 #include <rpc/rpc.h>
63 #include <rpc/xdr.h>
64 #include <rpc/pmap_prot.h>
65 #include <rpc/pmap_clnt.h>
66 #include <rpc/rpcb_prot.h>
67 
68 #include <rpcsvc/nlm_prot.h>
69 #include <rpcsvc/sm_inter.h>
70 #include <rpcsvc/nsm_addr.h>
71 
72 #include <nfs/nfs.h>
73 #include <nfs/nfs_clnt.h>
74 #include <nfs/export.h>
75 #include <nfs/rnode.h>
76 #include <nfs/lm.h>
77 
78 #include "nlm_impl.h"
79 
80 struct nlm_knc {
81 	struct knetconfig	n_knc;
82 	const char		*n_netid;
83 };
84 
85 /*
86  * Number of attempts NLM tries to obtain RPC binding
87  * of local statd.
88  */
89 #define	NLM_NSM_RPCBIND_RETRIES 10
90 
91 /*
92  * Timeout (in seconds) NLM waits before making another
93  * attempt to obtain RPC binding of local statd.
94  */
95 #define	NLM_NSM_RPCBIND_TIMEOUT 5
96 
97 /*
98  * Total number of sysids in NLM sysid bitmap
99  */
100 #define	NLM_BMAP_NITEMS	(LM_SYSID_MAX + 1)
101 
102 /*
103  * Number of ulong_t words in bitmap that is used
104  * for allocation of sysid numbers.
105  */
106 #define	NLM_BMAP_WORDS  (NLM_BMAP_NITEMS / BT_NBIPUL)
107 
108 /*
109  * Given an integer x, the macro returns
110  * -1 if x is negative,
111  *  0 if x is zero
112  *  1 if x is positive
113  */
114 #define	SIGN(x) (((x) > 0) - ((x) < 0))
115 
116 #define	ARRSIZE(arr)	(sizeof (arr) / sizeof ((arr)[0]))
117 #define	NLM_KNCS	ARRSIZE(nlm_netconfigs)
118 
119 krwlock_t lm_lck;
120 
121 /*
122  * Zero timeout for asynchronous NLM RPC operations
123  */
124 static const struct timeval nlm_rpctv_zero = { 0,  0 };
125 
126 /*
127  * List of all Zone globals nlm_globals instences
128  * linked together.
129  */
130 static struct nlm_globals_list nlm_zones_list; /* (g) */
131 
132 /*
133  * NLM kmem caches
134  */
135 static struct kmem_cache *nlm_hosts_cache = NULL;
136 static struct kmem_cache *nlm_vhold_cache = NULL;
137 
138 /*
139  * A bitmap for allocation of new sysids.
140  * Sysid is a unique number between LM_SYSID
141  * and LM_SYSID_MAX. Sysid represents unique remote
142  * host that does file locks on the given host.
143  */
144 static ulong_t	nlm_sysid_bmap[NLM_BMAP_WORDS];	/* (g) */
145 static int	nlm_sysid_nidx;			/* (g) */
146 
147 /*
148  * RPC service registration for all transports
149  */
150 static SVC_CALLOUT nlm_svcs[] = {
151 	{ NLM_PROG, 4, 4, nlm_prog_4 },	/* NLM4_VERS */
152 	{ NLM_PROG, 1, 3, nlm_prog_3 }	/* NLM_VERS - NLM_VERSX */
153 };
154 
155 static SVC_CALLOUT_TABLE nlm_sct = {
156 	ARRSIZE(nlm_svcs),
157 	FALSE,
158 	nlm_svcs
159 };
160 
161 /*
162  * Static table of all netid/knetconfig network
163  * lock manager can work with. nlm_netconfigs table
164  * is used when we need to get valid knetconfig by
165  * netid and vice versa.
166  *
167  * Knetconfigs are activated either by the call from
168  * user-space lockd daemon (server side) or by taking
169  * knetconfig from NFS mountinfo (client side)
170  */
171 static struct nlm_knc nlm_netconfigs[] = { /* (g) */
172 	/* UDP */
173 	{
174 		{ NC_TPI_CLTS, NC_INET, NC_UDP, NODEV },
175 		"udp",
176 	},
177 	/* TCP */
178 	{
179 		{ NC_TPI_COTS_ORD, NC_INET, NC_TCP, NODEV },
180 		"tcp",
181 	},
182 	/* UDP over IPv6 */
183 	{
184 		{ NC_TPI_CLTS, NC_INET6, NC_UDP, NODEV },
185 		"udp6",
186 	},
187 	/* TCP over IPv6 */
188 	{
189 		{ NC_TPI_COTS_ORD, NC_INET6, NC_TCP, NODEV },
190 		"tcp6",
191 	},
192 	/* ticlts (loopback over UDP) */
193 	{
194 		{ NC_TPI_CLTS, NC_LOOPBACK, NC_NOPROTO, NODEV },
195 		"ticlts",
196 	},
197 	/* ticotsord (loopback over TCP) */
198 	{
199 		{ NC_TPI_COTS_ORD, NC_LOOPBACK, NC_NOPROTO, NODEV },
200 		"ticotsord",
201 	},
202 };
203 
204 /*
205  * NLM misc. function
206  */
207 static void nlm_copy_netbuf(struct netbuf *, struct netbuf *);
208 static int nlm_netbuf_addrs_cmp(struct netbuf *, struct netbuf *);
209 static void nlm_kmem_reclaim(void *);
210 static void nlm_pool_shutdown(void);
211 static void nlm_suspend_zone(struct nlm_globals *);
212 static void nlm_resume_zone(struct nlm_globals *);
213 static void nlm_nsm_clnt_init(CLIENT *, struct nlm_nsm *);
214 static void nlm_netbuf_to_netobj(struct netbuf *, int *, netobj *);
215 
216 /*
217  * NLM thread functions
218  */
219 static void nlm_gc(struct nlm_globals *);
220 static void nlm_reclaimer(struct nlm_host *);
221 
222 /*
223  * NLM NSM functions
224  */
225 static int nlm_init_local_knc(struct knetconfig *);
226 static int nlm_nsm_init_local(struct nlm_nsm *);
227 static int nlm_nsm_init(struct nlm_nsm *, struct knetconfig *, struct netbuf *);
228 static void nlm_nsm_fini(struct nlm_nsm *);
229 static enum clnt_stat nlm_nsm_simu_crash(struct nlm_nsm *);
230 static enum clnt_stat nlm_nsm_stat(struct nlm_nsm *, int32_t *);
231 static enum clnt_stat nlm_nsm_mon(struct nlm_nsm *, char *, uint16_t);
232 static enum clnt_stat nlm_nsm_unmon(struct nlm_nsm *, char *);
233 
234 /*
235  * NLM host functions
236  */
237 static int nlm_host_ctor(void *, void *, int);
238 static void nlm_host_dtor(void *, void *);
239 static void nlm_host_destroy(struct nlm_host *);
240 static struct nlm_host *nlm_host_create(char *, const char *,
241     struct knetconfig *, struct netbuf *);
242 static struct nlm_host *nlm_host_find_locked(struct nlm_globals *,
243     const char *, struct netbuf *, avl_index_t *);
244 static void nlm_host_unregister(struct nlm_globals *, struct nlm_host *);
245 static void nlm_host_gc_vholds(struct nlm_host *);
246 static bool_t nlm_host_has_srv_locks(struct nlm_host *);
247 static bool_t nlm_host_has_cli_locks(struct nlm_host *);
248 static bool_t nlm_host_has_locks(struct nlm_host *);
249 
250 /*
251  * NLM vhold functions
252  */
253 static int nlm_vhold_ctor(void *, void *, int);
254 static void nlm_vhold_dtor(void *, void *);
255 static void nlm_vhold_destroy(struct nlm_host *,
256     struct nlm_vhold *);
257 static bool_t nlm_vhold_busy(struct nlm_host *, struct nlm_vhold *);
258 static void nlm_vhold_clean(struct nlm_vhold *, int);
259 
260 /*
261  * NLM client/server sleeping locks/share reservation functions
262  */
263 struct nlm_slreq *nlm_slreq_find_locked(struct nlm_host *,
264     struct nlm_vhold *, struct flock64 *);
265 static struct nlm_shres *nlm_shres_create_item(struct shrlock *, vnode_t *);
266 static void nlm_shres_destroy_item(struct nlm_shres *);
267 static bool_t nlm_shres_equal(struct shrlock *, struct shrlock *);
268 
269 /*
270  * NLM initialization functions.
271  */
272 void
nlm_init(void)273 nlm_init(void)
274 {
275 	nlm_hosts_cache = kmem_cache_create("nlm_host_cache",
276 	    sizeof (struct nlm_host), 0, nlm_host_ctor, nlm_host_dtor,
277 	    nlm_kmem_reclaim, NULL, NULL, 0);
278 
279 	nlm_vhold_cache = kmem_cache_create("nlm_vhold_cache",
280 	    sizeof (struct nlm_vhold), 0, nlm_vhold_ctor, nlm_vhold_dtor,
281 	    NULL, NULL, NULL, 0);
282 
283 	nlm_rpc_init();
284 	TAILQ_INIT(&nlm_zones_list);
285 
286 	/* initialize sysids bitmap */
287 	bzero(nlm_sysid_bmap, sizeof (nlm_sysid_bmap));
288 	nlm_sysid_nidx = 1;
289 
290 	/*
291 	 * Reserv the sysid #0, because it's associated
292 	 * with local locks only. Don't let to allocate
293 	 * it for remote locks.
294 	 */
295 	BT_SET(nlm_sysid_bmap, 0);
296 }
297 
298 void
nlm_globals_register(struct nlm_globals * g)299 nlm_globals_register(struct nlm_globals *g)
300 {
301 	rw_enter(&lm_lck, RW_WRITER);
302 	TAILQ_INSERT_TAIL(&nlm_zones_list, g, nlm_link);
303 	rw_exit(&lm_lck);
304 }
305 
306 void
nlm_globals_unregister(struct nlm_globals * g)307 nlm_globals_unregister(struct nlm_globals *g)
308 {
309 	rw_enter(&lm_lck, RW_WRITER);
310 	TAILQ_REMOVE(&nlm_zones_list, g, nlm_link);
311 	rw_exit(&lm_lck);
312 }
313 
314 /* ARGSUSED */
315 static void
nlm_kmem_reclaim(void * cdrarg)316 nlm_kmem_reclaim(void *cdrarg)
317 {
318 	struct nlm_globals *g;
319 
320 	rw_enter(&lm_lck, RW_READER);
321 	TAILQ_FOREACH(g, &nlm_zones_list, nlm_link)
322 		cv_broadcast(&g->nlm_gc_sched_cv);
323 
324 	rw_exit(&lm_lck);
325 }
326 
327 /*
328  * NLM garbage collector thread (GC).
329  *
330  * NLM GC periodically checks whether there're any host objects
331  * that can be cleaned up. It also releases stale vnodes that
332  * live on the server side (under protection of vhold objects).
333  *
334  * NLM host objects are cleaned up from GC thread because
335  * operations helping us to determine whether given host has
336  * any locks can be quite expensive and it's not good to call
337  * them every time the very last reference to the host is dropped.
338  * Thus we use "lazy" approach for hosts cleanup.
339  *
340  * The work of GC is to release stale vnodes on the server side
341  * and destroy hosts that haven't any locks and any activity for
342  * some time (i.e. idle hosts).
343  */
344 static void
nlm_gc(struct nlm_globals * g)345 nlm_gc(struct nlm_globals *g)
346 {
347 	struct nlm_host *hostp;
348 	clock_t now, idle_period;
349 
350 	idle_period = SEC_TO_TICK(g->cn_idle_tmo);
351 	mutex_enter(&g->lock);
352 	for (;;) {
353 		/*
354 		 * GC thread can be explicitly scheduled from
355 		 * memory reclamation function.
356 		 */
357 		(void) cv_timedwait(&g->nlm_gc_sched_cv, &g->lock,
358 		    ddi_get_lbolt() + idle_period);
359 
360 		/*
361 		 * NLM is shutting down, time to die.
362 		 */
363 		if (g->run_status == NLM_ST_STOPPING)
364 			break;
365 
366 		now = ddi_get_lbolt();
367 		DTRACE_PROBE2(gc__start, struct nlm_globals *, g,
368 		    clock_t, now);
369 
370 		/*
371 		 * Find all obviously unused vholds and destroy them.
372 		 */
373 		for (hostp = avl_first(&g->nlm_hosts_tree); hostp != NULL;
374 		    hostp = AVL_NEXT(&g->nlm_hosts_tree, hostp)) {
375 			struct nlm_vhold *nvp;
376 
377 			mutex_enter(&hostp->nh_lock);
378 
379 			nvp = TAILQ_FIRST(&hostp->nh_vholds_list);
380 			while (nvp != NULL) {
381 				struct nlm_vhold *new_nvp;
382 
383 				new_nvp = TAILQ_NEXT(nvp, nv_link);
384 
385 				/*
386 				 * If these conditions are met, the vhold is
387 				 * obviously unused and we will destroy it.  In
388 				 * a case either v_filocks and/or v_shrlocks is
389 				 * non-NULL the vhold might still be unused by
390 				 * the host, but it is expensive to check that.
391 				 * We defer such check until the host is idle.
392 				 * The expensive check is done below without
393 				 * the global lock held.
394 				 */
395 				if (nvp->nv_refcnt == 0 &&
396 				    nvp->nv_vp->v_filocks == NULL &&
397 				    nvp->nv_vp->v_shrlocks == NULL) {
398 					nlm_vhold_destroy(hostp, nvp);
399 				}
400 
401 				nvp = new_nvp;
402 			}
403 
404 			mutex_exit(&hostp->nh_lock);
405 		}
406 
407 		/*
408 		 * Handle all hosts that are unused at the moment
409 		 * until we meet one with idle timeout in future.
410 		 */
411 		while ((hostp = TAILQ_FIRST(&g->nlm_idle_hosts)) != NULL) {
412 			bool_t has_locks;
413 
414 			if (hostp->nh_idle_timeout > now)
415 				break;
416 
417 			/*
418 			 * Drop global lock while doing expensive work
419 			 * on this host. We'll re-check any conditions
420 			 * that might change after retaking the global
421 			 * lock.
422 			 */
423 			mutex_exit(&g->lock);
424 			mutex_enter(&hostp->nh_lock);
425 
426 			/*
427 			 * nlm_globals lock was dropped earlier because
428 			 * garbage collecting of vholds and checking whether
429 			 * host has any locks/shares are expensive operations.
430 			 */
431 			nlm_host_gc_vholds(hostp);
432 			has_locks = nlm_host_has_locks(hostp);
433 
434 			mutex_exit(&hostp->nh_lock);
435 			mutex_enter(&g->lock);
436 
437 			/*
438 			 * While we were doing expensive operations
439 			 * outside of nlm_globals critical section,
440 			 * somebody could take the host and remove it
441 			 * from the idle list.  Whether its been
442 			 * reinserted or not, our information about
443 			 * the host is outdated, and we should take no
444 			 * further action.
445 			 */
446 			if ((hostp->nh_flags & NLM_NH_INIDLE) == 0 ||
447 			    hostp->nh_idle_timeout > now)
448 				continue;
449 
450 			/*
451 			 * If the host has locks we have to renew the
452 			 * host's timeout and put it at the end of LRU
453 			 * list.
454 			 */
455 			if (has_locks) {
456 				TAILQ_REMOVE(&g->nlm_idle_hosts,
457 				    hostp, nh_link);
458 				hostp->nh_idle_timeout = now + idle_period;
459 				TAILQ_INSERT_TAIL(&g->nlm_idle_hosts,
460 				    hostp, nh_link);
461 				continue;
462 			}
463 
464 			/*
465 			 * We're here if all the following conditions hold:
466 			 * 1) Host hasn't any locks or share reservations
467 			 * 2) Host is unused
468 			 * 3) Host wasn't touched by anyone at least for
469 			 *    g->cn_idle_tmo seconds.
470 			 *
471 			 * So, now we can destroy it.
472 			 */
473 			nlm_host_unregister(g, hostp);
474 			mutex_exit(&g->lock);
475 
476 			nlm_host_unmonitor(g, hostp);
477 			nlm_host_destroy(hostp);
478 			mutex_enter(&g->lock);
479 			if (g->run_status == NLM_ST_STOPPING)
480 				break;
481 
482 		}
483 
484 		DTRACE_PROBE(gc__end);
485 	}
486 
487 	DTRACE_PROBE1(gc__exit, struct nlm_globals *, g);
488 
489 	/* Let others know that GC has died */
490 	g->nlm_gc_thread = NULL;
491 	mutex_exit(&g->lock);
492 
493 	cv_broadcast(&g->nlm_gc_finish_cv);
494 	zthread_exit();
495 }
496 
497 /*
498  * Thread reclaim locks/shares acquired by the client side
499  * on the given server represented by hostp.
500  */
501 static void
nlm_reclaimer(struct nlm_host * hostp)502 nlm_reclaimer(struct nlm_host *hostp)
503 {
504 	struct nlm_globals *g;
505 
506 	mutex_enter(&hostp->nh_lock);
507 	hostp->nh_reclaimer = curthread;
508 	mutex_exit(&hostp->nh_lock);
509 
510 	g = zone_getspecific(nlm_zone_key, curzone);
511 	nlm_reclaim_client(g, hostp);
512 
513 	mutex_enter(&hostp->nh_lock);
514 	hostp->nh_flags &= ~NLM_NH_RECLAIM;
515 	hostp->nh_reclaimer = NULL;
516 	cv_broadcast(&hostp->nh_recl_cv);
517 	mutex_exit(&hostp->nh_lock);
518 
519 	/*
520 	 * Host was explicitly referenced before
521 	 * nlm_reclaim() was called, release it
522 	 * here.
523 	 */
524 	nlm_host_release(g, hostp);
525 	zthread_exit();
526 }
527 
528 /*
529  * Copy a struct netobj.  (see xdr.h)
530  */
531 void
nlm_copy_netobj(struct netobj * dst,struct netobj * src)532 nlm_copy_netobj(struct netobj *dst, struct netobj *src)
533 {
534 	dst->n_len = src->n_len;
535 	dst->n_bytes = kmem_alloc(src->n_len, KM_SLEEP);
536 	bcopy(src->n_bytes, dst->n_bytes, src->n_len);
537 }
538 
539 /*
540  * An NLM specificw replacement for clnt_call().
541  * nlm_clnt_call() is used by all RPC functions generated
542  * from nlm_prot.x specification. The function is aware
543  * about some pitfalls of NLM RPC procedures and has a logic
544  * that handles them properly.
545  */
546 enum clnt_stat
nlm_clnt_call(CLIENT * clnt,rpcproc_t procnum,xdrproc_t xdr_args,caddr_t argsp,xdrproc_t xdr_result,caddr_t resultp,struct timeval wait)547 nlm_clnt_call(CLIENT *clnt, rpcproc_t procnum, xdrproc_t xdr_args,
548     caddr_t argsp, xdrproc_t xdr_result, caddr_t resultp, struct timeval wait)
549 {
550 	k_sigset_t oldmask;
551 	enum clnt_stat stat;
552 	bool_t sig_blocked = FALSE;
553 
554 	/*
555 	 * If NLM RPC procnum is one of the NLM _RES procedures
556 	 * that are used to reply to asynchronous NLM RPC
557 	 * (MSG calls), explicitly set RPC timeout to zero.
558 	 * Client doesn't send a reply to RES procedures, so
559 	 * we don't need to wait anything.
560 	 *
561 	 * NOTE: we ignore NLM4_*_RES procnums because they are
562 	 * equal to NLM_*_RES numbers.
563 	 */
564 	if (procnum >= NLM_TEST_RES && procnum <= NLM_GRANTED_RES)
565 		wait = nlm_rpctv_zero;
566 
567 	/*
568 	 * We need to block signals in case of NLM_CANCEL RPC
569 	 * in order to prevent interruption of network RPC
570 	 * calls.
571 	 */
572 	if (procnum == NLM_CANCEL) {
573 		k_sigset_t newmask;
574 
575 		sigfillset(&newmask);
576 		sigreplace(&newmask, &oldmask);
577 		sig_blocked = TRUE;
578 	}
579 
580 	stat = clnt_call(clnt, procnum, xdr_args,
581 	    argsp, xdr_result, resultp, wait);
582 
583 	/*
584 	 * Restore signal mask back if signals were blocked
585 	 */
586 	if (sig_blocked)
587 		sigreplace(&oldmask, (k_sigset_t *)NULL);
588 
589 	return (stat);
590 }
591 
592 /*
593  * Suspend NLM client/server in the given zone.
594  *
595  * During suspend operation we mark those hosts
596  * that have any locks with NLM_NH_SUSPEND flags,
597  * so that they can be checked later, when resume
598  * operation occurs.
599  */
600 static void
nlm_suspend_zone(struct nlm_globals * g)601 nlm_suspend_zone(struct nlm_globals *g)
602 {
603 	struct nlm_host *hostp;
604 	struct nlm_host_list all_hosts;
605 
606 	/*
607 	 * Note that while we're doing suspend, GC thread is active
608 	 * and it can destroy some hosts while we're walking through
609 	 * the hosts tree. To prevent that and make suspend logic
610 	 * a bit more simple we put all hosts to local "all_hosts"
611 	 * list and increment reference counter of each host.
612 	 * This guaranties that no hosts will be released while
613 	 * we're doing suspend.
614 	 * NOTE: reference of each host must be dropped during
615 	 * resume operation.
616 	 */
617 	TAILQ_INIT(&all_hosts);
618 	mutex_enter(&g->lock);
619 	for (hostp = avl_first(&g->nlm_hosts_tree); hostp != NULL;
620 	    hostp = AVL_NEXT(&g->nlm_hosts_tree, hostp)) {
621 		/*
622 		 * If host is idle, remove it from idle list and
623 		 * clear idle flag. That is done to prevent GC
624 		 * from touching this host.
625 		 */
626 		if (hostp->nh_flags & NLM_NH_INIDLE) {
627 			TAILQ_REMOVE(&g->nlm_idle_hosts, hostp, nh_link);
628 			hostp->nh_flags &= ~NLM_NH_INIDLE;
629 		}
630 
631 		hostp->nh_refs++;
632 		TAILQ_INSERT_TAIL(&all_hosts, hostp, nh_link);
633 	}
634 
635 	/*
636 	 * Now we can walk through all hosts on the system
637 	 * with zone globals lock released. The fact the
638 	 * we have taken a reference to each host guaranties
639 	 * that no hosts can be destroyed during that process.
640 	 */
641 	mutex_exit(&g->lock);
642 	while ((hostp = TAILQ_FIRST(&all_hosts)) != NULL) {
643 		mutex_enter(&hostp->nh_lock);
644 		if (nlm_host_has_locks(hostp))
645 			hostp->nh_flags |= NLM_NH_SUSPEND;
646 
647 		mutex_exit(&hostp->nh_lock);
648 		TAILQ_REMOVE(&all_hosts, hostp, nh_link);
649 	}
650 }
651 
652 /*
653  * Resume NLM hosts for the given zone.
654  *
655  * nlm_resume_zone() is called after hosts were suspended
656  * (see nlm_suspend_zone) and its main purpose to check
657  * whether remote locks owned by hosts are still in consistent
658  * state. If they aren't, resume function tries to reclaim
659  * locks (for client side hosts) and clean locks (for
660  * server side hosts).
661  */
662 static void
nlm_resume_zone(struct nlm_globals * g)663 nlm_resume_zone(struct nlm_globals *g)
664 {
665 	struct nlm_host *hostp, *h_next;
666 
667 	mutex_enter(&g->lock);
668 	hostp = avl_first(&g->nlm_hosts_tree);
669 
670 	/*
671 	 * In nlm_suspend_zone() the reference counter of each
672 	 * host was incremented, so we can safely iterate through
673 	 * all hosts without worrying that any host we touch will
674 	 * be removed at the moment.
675 	 */
676 	while (hostp != NULL) {
677 		struct nlm_nsm nsm;
678 		enum clnt_stat stat;
679 		int32_t sm_state;
680 		int error;
681 		bool_t resume_failed = FALSE;
682 
683 		h_next = AVL_NEXT(&g->nlm_hosts_tree, hostp);
684 		mutex_exit(&g->lock);
685 
686 		DTRACE_PROBE1(resume__host, struct nlm_host *, hostp);
687 
688 		/*
689 		 * Suspend operation marked that the host doesn't
690 		 * have any locks. Skip it.
691 		 */
692 		if (!(hostp->nh_flags & NLM_NH_SUSPEND))
693 			goto cycle_end;
694 
695 		error = nlm_nsm_init(&nsm, &hostp->nh_knc, &hostp->nh_addr);
696 		if (error != 0) {
697 			NLM_ERR("Resume: Failed to contact to NSM of host %s "
698 			    "[error=%d]\n", hostp->nh_name, error);
699 			resume_failed = TRUE;
700 			goto cycle_end;
701 		}
702 
703 		stat = nlm_nsm_stat(&nsm, &sm_state);
704 		if (stat != RPC_SUCCESS) {
705 			NLM_ERR("Resume: Failed to call SM_STAT operation for "
706 			    "host %s [stat=%d]\n", hostp->nh_name, stat);
707 			resume_failed = TRUE;
708 			nlm_nsm_fini(&nsm);
709 			goto cycle_end;
710 		}
711 
712 		if (sm_state != hostp->nh_state) {
713 			/*
714 			 * Current SM state of the host isn't equal
715 			 * to the one host had when it was suspended.
716 			 * Probably it was rebooted. Try to reclaim
717 			 * locks if the host has any on its client side.
718 			 * Also try to clean up its server side locks
719 			 * (if the host has any).
720 			 */
721 			nlm_host_notify_client(hostp, sm_state);
722 			nlm_host_notify_server(hostp, sm_state);
723 		}
724 
725 		nlm_nsm_fini(&nsm);
726 
727 cycle_end:
728 		if (resume_failed) {
729 			/*
730 			 * Resume failed for the given host.
731 			 * Just clean up all resources it owns.
732 			 */
733 			nlm_host_notify_server(hostp, 0);
734 			nlm_client_cancel_all(g, hostp);
735 		}
736 
737 		hostp->nh_flags &= ~NLM_NH_SUSPEND;
738 		nlm_host_release(g, hostp);
739 		hostp = h_next;
740 		mutex_enter(&g->lock);
741 	}
742 
743 	mutex_exit(&g->lock);
744 }
745 
746 /*
747  * NLM functions responsible for operations on NSM handle.
748  */
749 
750 /*
751  * Initialize knetconfig that is used for communication
752  * with local statd via loopback interface.
753  */
754 static int
nlm_init_local_knc(struct knetconfig * knc)755 nlm_init_local_knc(struct knetconfig *knc)
756 {
757 	int error;
758 	vnode_t *vp;
759 
760 	bzero(knc, sizeof (*knc));
761 	error = lookupname("/dev/tcp", UIO_SYSSPACE,
762 	    FOLLOW, NULLVPP, &vp);
763 	if (error != 0)
764 		return (error);
765 
766 	knc->knc_semantics = NC_TPI_COTS;
767 	knc->knc_protofmly = NC_INET;
768 	knc->knc_proto = NC_TCP;
769 	knc->knc_rdev = vp->v_rdev;
770 	VN_RELE(vp);
771 
772 
773 	return (0);
774 }
775 
776 /*
777  * Initialize NSM handle that will be used to talk
778  * to local statd via loopback interface.
779  */
780 static int
nlm_nsm_init_local(struct nlm_nsm * nsm)781 nlm_nsm_init_local(struct nlm_nsm *nsm)
782 {
783 	int error;
784 	struct knetconfig knc;
785 	struct sockaddr_in sin;
786 	struct netbuf nb;
787 
788 	error = nlm_init_local_knc(&knc);
789 	if (error != 0)
790 		return (error);
791 
792 	bzero(&sin, sizeof (sin));
793 	sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
794 	sin.sin_family = AF_INET;
795 
796 	nb.buf = (char *)&sin;
797 	nb.len = nb.maxlen = sizeof (sin);
798 
799 	return (nlm_nsm_init(nsm, &knc, &nb));
800 }
801 
802 /*
803  * Initialize NSM handle used for talking to statd
804  */
805 static int
nlm_nsm_init(struct nlm_nsm * nsm,struct knetconfig * knc,struct netbuf * nb)806 nlm_nsm_init(struct nlm_nsm *nsm, struct knetconfig *knc, struct netbuf *nb)
807 {
808 	enum clnt_stat stat;
809 	int error, retries;
810 
811 	bzero(nsm, sizeof (*nsm));
812 	nsm->ns_knc = *knc;
813 	nlm_copy_netbuf(&nsm->ns_addr, nb);
814 
815 	/*
816 	 * Try several times to get the port of statd service,
817 	 * If rpcbind_getaddr returns  RPC_PROGNOTREGISTERED,
818 	 * retry an attempt, but wait for NLM_NSM_RPCBIND_TIMEOUT
819 	 * seconds berofore.
820 	 */
821 	for (retries = 0; retries < NLM_NSM_RPCBIND_RETRIES; retries++) {
822 		stat = rpcbind_getaddr(&nsm->ns_knc, SM_PROG,
823 		    SM_VERS, &nsm->ns_addr);
824 		if (stat != RPC_SUCCESS) {
825 			if (stat == RPC_PROGNOTREGISTERED) {
826 				delay(SEC_TO_TICK(NLM_NSM_RPCBIND_TIMEOUT));
827 				continue;
828 			}
829 		}
830 
831 		break;
832 	}
833 
834 	if (stat != RPC_SUCCESS) {
835 		DTRACE_PROBE2(rpcbind__error, enum clnt_stat, stat,
836 		    int, retries);
837 		error = ENOENT;
838 		goto error;
839 	}
840 
841 	/*
842 	 * Create an RPC handle that'll be used for communication with local
843 	 * statd using the status monitor protocol.
844 	 */
845 	error = clnt_tli_kcreate(&nsm->ns_knc, &nsm->ns_addr, SM_PROG, SM_VERS,
846 	    0, NLM_RPC_RETRIES, zone_kcred(), &nsm->ns_handle);
847 	if (error != 0)
848 		goto error;
849 
850 	/*
851 	 * Create an RPC handle that'll be used for communication with the
852 	 * local statd using the address registration protocol.
853 	 */
854 	error = clnt_tli_kcreate(&nsm->ns_knc, &nsm->ns_addr, NSM_ADDR_PROGRAM,
855 	    NSM_ADDR_V1, 0, NLM_RPC_RETRIES, zone_kcred(),
856 	    &nsm->ns_addr_handle);
857 	if (error != 0)
858 		goto error;
859 
860 	sema_init(&nsm->ns_sem, 1, NULL, SEMA_DEFAULT, NULL);
861 	return (0);
862 
863 error:
864 	kmem_free(nsm->ns_addr.buf, nsm->ns_addr.maxlen);
865 	if (nsm->ns_handle) {
866 		ASSERT(nsm->ns_handle->cl_auth != NULL);
867 		auth_destroy(nsm->ns_handle->cl_auth);
868 		CLNT_DESTROY(nsm->ns_handle);
869 	}
870 
871 	return (error);
872 }
873 
874 static void
nlm_nsm_fini(struct nlm_nsm * nsm)875 nlm_nsm_fini(struct nlm_nsm *nsm)
876 {
877 	kmem_free(nsm->ns_addr.buf, nsm->ns_addr.maxlen);
878 	if (nsm->ns_addr_handle->cl_auth != NULL)
879 		auth_destroy(nsm->ns_addr_handle->cl_auth);
880 	CLNT_DESTROY(nsm->ns_addr_handle);
881 	nsm->ns_addr_handle = NULL;
882 	if (nsm->ns_handle->cl_auth != NULL)
883 		auth_destroy(nsm->ns_handle->cl_auth);
884 	CLNT_DESTROY(nsm->ns_handle);
885 	nsm->ns_handle = NULL;
886 	sema_destroy(&nsm->ns_sem);
887 }
888 
889 static enum clnt_stat
nlm_nsm_simu_crash(struct nlm_nsm * nsm)890 nlm_nsm_simu_crash(struct nlm_nsm *nsm)
891 {
892 	enum clnt_stat stat;
893 
894 	sema_p(&nsm->ns_sem);
895 	nlm_nsm_clnt_init(nsm->ns_handle, nsm);
896 	stat = sm_simu_crash_1(NULL, NULL, nsm->ns_handle);
897 	sema_v(&nsm->ns_sem);
898 
899 	return (stat);
900 }
901 
902 static enum clnt_stat
nlm_nsm_stat(struct nlm_nsm * nsm,int32_t * out_stat)903 nlm_nsm_stat(struct nlm_nsm *nsm, int32_t *out_stat)
904 {
905 	struct sm_name args;
906 	struct sm_stat_res res;
907 	enum clnt_stat stat;
908 
909 	args.mon_name = uts_nodename();
910 	bzero(&res, sizeof (res));
911 
912 	sema_p(&nsm->ns_sem);
913 	nlm_nsm_clnt_init(nsm->ns_handle, nsm);
914 	stat = sm_stat_1(&args, &res, nsm->ns_handle);
915 	sema_v(&nsm->ns_sem);
916 
917 	if (stat == RPC_SUCCESS)
918 		*out_stat = res.state;
919 
920 	return (stat);
921 }
922 
923 static enum clnt_stat
nlm_nsm_mon(struct nlm_nsm * nsm,char * hostname,uint16_t priv)924 nlm_nsm_mon(struct nlm_nsm *nsm, char *hostname, uint16_t priv)
925 {
926 	struct mon args;
927 	struct sm_stat_res res;
928 	enum clnt_stat stat;
929 
930 	bzero(&args, sizeof (args));
931 	bzero(&res, sizeof (res));
932 
933 	args.mon_id.mon_name = hostname;
934 	args.mon_id.my_id.my_name = uts_nodename();
935 	args.mon_id.my_id.my_prog = NLM_PROG;
936 	args.mon_id.my_id.my_vers = NLM_SM;
937 	args.mon_id.my_id.my_proc = NLM_SM_NOTIFY1;
938 	bcopy(&priv, args.priv, sizeof (priv));
939 
940 	sema_p(&nsm->ns_sem);
941 	nlm_nsm_clnt_init(nsm->ns_handle, nsm);
942 	stat = sm_mon_1(&args, &res, nsm->ns_handle);
943 	sema_v(&nsm->ns_sem);
944 
945 	return (stat);
946 }
947 
948 static enum clnt_stat
nlm_nsm_unmon(struct nlm_nsm * nsm,char * hostname)949 nlm_nsm_unmon(struct nlm_nsm *nsm, char *hostname)
950 {
951 	struct mon_id args;
952 	struct sm_stat res;
953 	enum clnt_stat stat;
954 
955 	bzero(&args, sizeof (args));
956 	bzero(&res, sizeof (res));
957 
958 	args.mon_name = hostname;
959 	args.my_id.my_name = uts_nodename();
960 	args.my_id.my_prog = NLM_PROG;
961 	args.my_id.my_vers = NLM_SM;
962 	args.my_id.my_proc = NLM_SM_NOTIFY1;
963 
964 	sema_p(&nsm->ns_sem);
965 	nlm_nsm_clnt_init(nsm->ns_handle, nsm);
966 	stat = sm_unmon_1(&args, &res, nsm->ns_handle);
967 	sema_v(&nsm->ns_sem);
968 
969 	return (stat);
970 }
971 
972 static enum clnt_stat
nlm_nsmaddr_reg(struct nlm_nsm * nsm,char * name,int family,netobj * address)973 nlm_nsmaddr_reg(struct nlm_nsm *nsm, char *name, int family, netobj *address)
974 {
975 	struct reg1args args = { 0 };
976 	struct reg1res res = { 0 };
977 	enum clnt_stat stat;
978 
979 	args.family = family;
980 	args.name = name;
981 	args.address = *address;
982 
983 	sema_p(&nsm->ns_sem);
984 	nlm_nsm_clnt_init(nsm->ns_addr_handle, nsm);
985 	stat = nsmaddrproc1_reg_1(&args, &res, nsm->ns_addr_handle);
986 	sema_v(&nsm->ns_sem);
987 
988 	return (stat);
989 }
990 
991 /*
992  * Get NLM vhold object corresponding to vnode "vp".
993  * If no such object was found, create a new one.
994  *
995  * The purpose of this function is to associate vhold
996  * object with given vnode, so that:
997  * 1) vnode is hold (VN_HOLD) while vhold object is alive.
998  * 2) host has a track of all vnodes it touched by lock
999  *    or share operations. These vnodes are accessible
1000  *    via collection of vhold objects.
1001  */
1002 struct nlm_vhold *
nlm_vhold_get(struct nlm_host * hostp,vnode_t * vp)1003 nlm_vhold_get(struct nlm_host *hostp, vnode_t *vp)
1004 {
1005 	struct nlm_vhold *nvp, *new_nvp = NULL;
1006 
1007 	mutex_enter(&hostp->nh_lock);
1008 	nvp = nlm_vhold_find_locked(hostp, vp);
1009 	if (nvp != NULL)
1010 		goto out;
1011 
1012 	/* nlm_vhold wasn't found, then create a new one */
1013 	mutex_exit(&hostp->nh_lock);
1014 	new_nvp = kmem_cache_alloc(nlm_vhold_cache, KM_SLEEP);
1015 
1016 	/*
1017 	 * Check if another thread has already
1018 	 * created the same nlm_vhold.
1019 	 */
1020 	mutex_enter(&hostp->nh_lock);
1021 	nvp = nlm_vhold_find_locked(hostp, vp);
1022 	if (nvp == NULL) {
1023 		nvp = new_nvp;
1024 		new_nvp = NULL;
1025 
1026 		TAILQ_INIT(&nvp->nv_slreqs);
1027 		nvp->nv_vp = vp;
1028 		nvp->nv_refcnt = 1;
1029 		VN_HOLD(nvp->nv_vp);
1030 
1031 		VERIFY(mod_hash_insert(hostp->nh_vholds_by_vp,
1032 		    (mod_hash_key_t)vp, (mod_hash_val_t)nvp) == 0);
1033 		TAILQ_INSERT_TAIL(&hostp->nh_vholds_list, nvp, nv_link);
1034 	}
1035 
1036 out:
1037 	mutex_exit(&hostp->nh_lock);
1038 	if (new_nvp != NULL)
1039 		kmem_cache_free(nlm_vhold_cache, new_nvp);
1040 
1041 	return (nvp);
1042 }
1043 
1044 /*
1045  * Drop a reference to vhold object nvp.
1046  */
1047 void
nlm_vhold_release(struct nlm_host * hostp,struct nlm_vhold * nvp)1048 nlm_vhold_release(struct nlm_host *hostp, struct nlm_vhold *nvp)
1049 {
1050 	if (nvp == NULL)
1051 		return;
1052 
1053 	mutex_enter(&hostp->nh_lock);
1054 	ASSERT(nvp->nv_refcnt > 0);
1055 	nvp->nv_refcnt--;
1056 
1057 	/*
1058 	 * If these conditions are met, the vhold is obviously unused and we
1059 	 * will destroy it.  In a case either v_filocks and/or v_shrlocks is
1060 	 * non-NULL the vhold might still be unused by the host, but it is
1061 	 * expensive to check that.  We defer such check until the host is
1062 	 * idle.  The expensive check is done in the NLM garbage collector.
1063 	 */
1064 	if (nvp->nv_refcnt == 0 &&
1065 	    nvp->nv_vp->v_filocks == NULL &&
1066 	    nvp->nv_vp->v_shrlocks == NULL) {
1067 		nlm_vhold_destroy(hostp, nvp);
1068 	}
1069 
1070 	mutex_exit(&hostp->nh_lock);
1071 }
1072 
1073 /*
1074  * Clean all locks and share reservations on the
1075  * given vhold object that were acquired by the
1076  * given sysid
1077  */
1078 static void
nlm_vhold_clean(struct nlm_vhold * nvp,int sysid)1079 nlm_vhold_clean(struct nlm_vhold *nvp, int sysid)
1080 {
1081 	cleanlocks(nvp->nv_vp, IGN_PID, sysid);
1082 	cleanshares_by_sysid(nvp->nv_vp, sysid);
1083 }
1084 
1085 static void
nlm_vhold_destroy(struct nlm_host * hostp,struct nlm_vhold * nvp)1086 nlm_vhold_destroy(struct nlm_host *hostp, struct nlm_vhold *nvp)
1087 {
1088 	ASSERT(MUTEX_HELD(&hostp->nh_lock));
1089 
1090 	ASSERT(nvp->nv_refcnt == 0);
1091 	ASSERT(TAILQ_EMPTY(&nvp->nv_slreqs));
1092 
1093 	VERIFY(mod_hash_remove(hostp->nh_vholds_by_vp,
1094 	    (mod_hash_key_t)nvp->nv_vp,
1095 	    (mod_hash_val_t)&nvp) == 0);
1096 
1097 	TAILQ_REMOVE(&hostp->nh_vholds_list, nvp, nv_link);
1098 	VN_RELE(nvp->nv_vp);
1099 	nvp->nv_vp = NULL;
1100 
1101 	kmem_cache_free(nlm_vhold_cache, nvp);
1102 }
1103 
1104 /*
1105  * Return TRUE if the given vhold is busy.
1106  * Vhold object is considered to be "busy" when
1107  * all the following conditions hold:
1108  * 1) No one uses it at the moment;
1109  * 2) It hasn't any locks;
1110  * 3) It hasn't any share reservations;
1111  */
1112 static bool_t
nlm_vhold_busy(struct nlm_host * hostp,struct nlm_vhold * nvp)1113 nlm_vhold_busy(struct nlm_host *hostp, struct nlm_vhold *nvp)
1114 {
1115 	vnode_t *vp;
1116 	int sysid;
1117 
1118 	ASSERT(MUTEX_HELD(&hostp->nh_lock));
1119 
1120 	if (nvp->nv_refcnt > 0)
1121 		return (TRUE);
1122 
1123 	vp = nvp->nv_vp;
1124 	sysid = hostp->nh_sysid;
1125 	if (flk_has_remote_locks_for_sysid(vp, sysid) ||
1126 	    shr_has_remote_shares(vp, sysid))
1127 		return (TRUE);
1128 
1129 	return (FALSE);
1130 }
1131 
1132 /* ARGSUSED */
1133 static int
nlm_vhold_ctor(void * datap,void * cdrarg,int kmflags)1134 nlm_vhold_ctor(void *datap, void *cdrarg, int kmflags)
1135 {
1136 	struct nlm_vhold *nvp = (struct nlm_vhold *)datap;
1137 
1138 	bzero(nvp, sizeof (*nvp));
1139 	return (0);
1140 }
1141 
1142 /* ARGSUSED */
1143 static void
nlm_vhold_dtor(void * datap,void * cdrarg)1144 nlm_vhold_dtor(void *datap, void *cdrarg)
1145 {
1146 	struct nlm_vhold *nvp = (struct nlm_vhold *)datap;
1147 
1148 	ASSERT(nvp->nv_refcnt == 0);
1149 	ASSERT(TAILQ_EMPTY(&nvp->nv_slreqs));
1150 	ASSERT(nvp->nv_vp == NULL);
1151 }
1152 
1153 struct nlm_vhold *
nlm_vhold_find_locked(struct nlm_host * hostp,const vnode_t * vp)1154 nlm_vhold_find_locked(struct nlm_host *hostp, const vnode_t *vp)
1155 {
1156 	struct nlm_vhold *nvp = NULL;
1157 
1158 	ASSERT(MUTEX_HELD(&hostp->nh_lock));
1159 	(void) mod_hash_find(hostp->nh_vholds_by_vp,
1160 	    (mod_hash_key_t)vp,
1161 	    (mod_hash_val_t)&nvp);
1162 
1163 	if (nvp != NULL)
1164 		nvp->nv_refcnt++;
1165 
1166 	return (nvp);
1167 }
1168 
1169 /*
1170  * NLM host functions
1171  */
1172 static void
nlm_copy_netbuf(struct netbuf * dst,struct netbuf * src)1173 nlm_copy_netbuf(struct netbuf *dst, struct netbuf *src)
1174 {
1175 	ASSERT(src->len <= src->maxlen);
1176 
1177 	dst->maxlen = src->maxlen;
1178 	dst->len = src->len;
1179 	dst->buf = kmem_zalloc(src->maxlen, KM_SLEEP);
1180 	bcopy(src->buf, dst->buf, src->len);
1181 }
1182 
1183 /* ARGSUSED */
1184 static int
nlm_host_ctor(void * datap,void * cdrarg,int kmflags)1185 nlm_host_ctor(void *datap, void *cdrarg, int kmflags)
1186 {
1187 	struct nlm_host *hostp = (struct nlm_host *)datap;
1188 
1189 	bzero(hostp, sizeof (*hostp));
1190 	return (0);
1191 }
1192 
1193 /* ARGSUSED */
1194 static void
nlm_host_dtor(void * datap,void * cdrarg)1195 nlm_host_dtor(void *datap, void *cdrarg)
1196 {
1197 	struct nlm_host *hostp = (struct nlm_host *)datap;
1198 	ASSERT(hostp->nh_refs == 0);
1199 }
1200 
1201 static void
nlm_host_unregister(struct nlm_globals * g,struct nlm_host * hostp)1202 nlm_host_unregister(struct nlm_globals *g, struct nlm_host *hostp)
1203 {
1204 	ASSERT(hostp->nh_refs == 0);
1205 	ASSERT(hostp->nh_flags & NLM_NH_INIDLE);
1206 
1207 	avl_remove(&g->nlm_hosts_tree, hostp);
1208 	VERIFY(mod_hash_remove(g->nlm_hosts_hash,
1209 	    (mod_hash_key_t)(uintptr_t)hostp->nh_sysid,
1210 	    (mod_hash_val_t)&hostp) == 0);
1211 	TAILQ_REMOVE(&g->nlm_idle_hosts, hostp, nh_link);
1212 	hostp->nh_flags &= ~NLM_NH_INIDLE;
1213 }
1214 
1215 /*
1216  * Free resources used by a host. This is called after the reference
1217  * count has reached zero so it doesn't need to worry about locks.
1218  */
1219 static void
nlm_host_destroy(struct nlm_host * hostp)1220 nlm_host_destroy(struct nlm_host *hostp)
1221 {
1222 	ASSERT(hostp->nh_name != NULL);
1223 	ASSERT(hostp->nh_netid != NULL);
1224 	ASSERT(TAILQ_EMPTY(&hostp->nh_vholds_list));
1225 
1226 	strfree(hostp->nh_name);
1227 	strfree(hostp->nh_netid);
1228 	kmem_free(hostp->nh_addr.buf, hostp->nh_addr.maxlen);
1229 
1230 	if (hostp->nh_sysid != LM_NOSYSID)
1231 		nlm_sysid_free(hostp->nh_sysid);
1232 
1233 	nlm_rpc_cache_destroy(hostp);
1234 
1235 	ASSERT(TAILQ_EMPTY(&hostp->nh_vholds_list));
1236 	mod_hash_destroy_ptrhash(hostp->nh_vholds_by_vp);
1237 
1238 	mutex_destroy(&hostp->nh_lock);
1239 	cv_destroy(&hostp->nh_rpcb_cv);
1240 	cv_destroy(&hostp->nh_recl_cv);
1241 
1242 	kmem_cache_free(nlm_hosts_cache, hostp);
1243 }
1244 
1245 /*
1246  * Cleanup SERVER-side state after a client restarts,
1247  * or becomes unresponsive, or whatever.
1248  *
1249  * We unlock any active locks owned by the host.
1250  * When rpc.lockd is shutting down,
1251  * this function is called with newstate set to zero
1252  * which allows us to cancel any pending async locks
1253  * and clear the locking state.
1254  *
1255  * When "state" is 0, we don't update host's state,
1256  * but cleanup all remote locks on the host.
1257  * It's useful to call this function for resources
1258  * cleanup.
1259  */
1260 void
nlm_host_notify_server(struct nlm_host * hostp,int32_t state)1261 nlm_host_notify_server(struct nlm_host *hostp, int32_t state)
1262 {
1263 	struct nlm_vhold *nvp;
1264 	struct nlm_slreq *slr;
1265 	struct nlm_slreq_list slreqs2free;
1266 
1267 	TAILQ_INIT(&slreqs2free);
1268 	mutex_enter(&hostp->nh_lock);
1269 	if (state != 0)
1270 		hostp->nh_state = state;
1271 
1272 	TAILQ_FOREACH(nvp, &hostp->nh_vholds_list, nv_link) {
1273 
1274 		/* cleanup sleeping requests at first */
1275 		while ((slr = TAILQ_FIRST(&nvp->nv_slreqs)) != NULL) {
1276 			TAILQ_REMOVE(&nvp->nv_slreqs, slr, nsr_link);
1277 
1278 			/*
1279 			 * Instead of freeing cancelled sleeping request
1280 			 * here, we add it to the linked list created
1281 			 * on the stack in order to do all frees outside
1282 			 * the critical section.
1283 			 */
1284 			TAILQ_INSERT_TAIL(&slreqs2free, slr, nsr_link);
1285 		}
1286 
1287 		nvp->nv_refcnt++;
1288 		mutex_exit(&hostp->nh_lock);
1289 
1290 		nlm_vhold_clean(nvp, hostp->nh_sysid);
1291 
1292 		mutex_enter(&hostp->nh_lock);
1293 		nvp->nv_refcnt--;
1294 	}
1295 
1296 	mutex_exit(&hostp->nh_lock);
1297 	while ((slr = TAILQ_FIRST(&slreqs2free)) != NULL) {
1298 		TAILQ_REMOVE(&slreqs2free, slr, nsr_link);
1299 		kmem_free(slr, sizeof (*slr));
1300 	}
1301 }
1302 
1303 /*
1304  * Cleanup CLIENT-side state after a server restarts,
1305  * or becomes unresponsive, or whatever.
1306  *
1307  * This is called by the local NFS statd when we receive a
1308  * host state change notification.  (also nlm_svc_stopping)
1309  *
1310  * Deal with a server restart.  If we are stopping the
1311  * NLM service, we'll have newstate == 0, and will just
1312  * cancel all our client-side lock requests.  Otherwise,
1313  * start the "recovery" process to reclaim any locks
1314  * we hold on this server.
1315  */
1316 void
nlm_host_notify_client(struct nlm_host * hostp,int32_t state)1317 nlm_host_notify_client(struct nlm_host *hostp, int32_t state)
1318 {
1319 	mutex_enter(&hostp->nh_lock);
1320 	hostp->nh_state = state;
1321 	if (hostp->nh_flags & NLM_NH_RECLAIM) {
1322 		/*
1323 		 * Either host's state is up to date or
1324 		 * host is already in recovery.
1325 		 */
1326 		mutex_exit(&hostp->nh_lock);
1327 		return;
1328 	}
1329 
1330 	hostp->nh_flags |= NLM_NH_RECLAIM;
1331 
1332 	/*
1333 	 * Host will be released by the recovery thread,
1334 	 * thus we need to increment refcount.
1335 	 */
1336 	hostp->nh_refs++;
1337 	mutex_exit(&hostp->nh_lock);
1338 
1339 	(void) zthread_create(NULL, 0, nlm_reclaimer,
1340 	    hostp, 0, minclsyspri);
1341 }
1342 
1343 /*
1344  * The function is called when NLM client detects that
1345  * server has entered in grace period and client needs
1346  * to wait until reclamation process (if any) does
1347  * its job.
1348  */
1349 int
nlm_host_wait_grace(struct nlm_host * hostp)1350 nlm_host_wait_grace(struct nlm_host *hostp)
1351 {
1352 	struct nlm_globals *g;
1353 	int error = 0;
1354 
1355 	g = zone_getspecific(nlm_zone_key, curzone);
1356 	mutex_enter(&hostp->nh_lock);
1357 
1358 	do {
1359 		int rc;
1360 
1361 		rc = cv_timedwait_sig(&hostp->nh_recl_cv,
1362 		    &hostp->nh_lock, ddi_get_lbolt() +
1363 		    SEC_TO_TICK(g->retrans_tmo));
1364 
1365 		if (rc == 0) {
1366 			error = EINTR;
1367 			break;
1368 		}
1369 	} while (hostp->nh_flags & NLM_NH_RECLAIM);
1370 
1371 	mutex_exit(&hostp->nh_lock);
1372 	return (error);
1373 }
1374 
1375 /*
1376  * Create a new NLM host.
1377  *
1378  * NOTE: The in-kernel RPC (kRPC) subsystem uses TLI/XTI,
1379  * which needs both a knetconfig and an address when creating
1380  * endpoints. Thus host object stores both knetconfig and
1381  * netid.
1382  */
1383 static struct nlm_host *
nlm_host_create(char * name,const char * netid,struct knetconfig * knc,struct netbuf * naddr)1384 nlm_host_create(char *name, const char *netid,
1385     struct knetconfig *knc, struct netbuf *naddr)
1386 {
1387 	struct nlm_host *host;
1388 
1389 	host = kmem_cache_alloc(nlm_hosts_cache, KM_SLEEP);
1390 
1391 	mutex_init(&host->nh_lock, NULL, MUTEX_DEFAULT, NULL);
1392 	cv_init(&host->nh_rpcb_cv, NULL, CV_DEFAULT, NULL);
1393 	cv_init(&host->nh_recl_cv, NULL, CV_DEFAULT, NULL);
1394 
1395 	host->nh_sysid = LM_NOSYSID;
1396 	host->nh_refs = 1;
1397 	host->nh_name = strdup(name);
1398 	host->nh_netid = strdup(netid);
1399 	host->nh_knc = *knc;
1400 	nlm_copy_netbuf(&host->nh_addr, naddr);
1401 
1402 	host->nh_state = 0;
1403 	host->nh_rpcb_state = NRPCB_NEED_UPDATE;
1404 	host->nh_flags = 0;
1405 
1406 	host->nh_vholds_by_vp = mod_hash_create_ptrhash("nlm vholds hash",
1407 	    32, mod_hash_null_valdtor, sizeof (vnode_t));
1408 
1409 	TAILQ_INIT(&host->nh_vholds_list);
1410 	TAILQ_INIT(&host->nh_rpchc);
1411 
1412 	return (host);
1413 }
1414 
1415 /*
1416  * Cancel all client side sleeping locks owned by given host.
1417  */
1418 void
nlm_host_cancel_slocks(struct nlm_globals * g,struct nlm_host * hostp)1419 nlm_host_cancel_slocks(struct nlm_globals *g, struct nlm_host *hostp)
1420 {
1421 	struct nlm_slock *nslp;
1422 
1423 	mutex_enter(&g->lock);
1424 	TAILQ_FOREACH(nslp, &g->nlm_slocks, nsl_link) {
1425 		if (nslp->nsl_host == hostp) {
1426 			nslp->nsl_state = NLM_SL_CANCELLED;
1427 			cv_broadcast(&nslp->nsl_cond);
1428 		}
1429 	}
1430 
1431 	mutex_exit(&g->lock);
1432 }
1433 
1434 /*
1435  * Garbage collect stale vhold objects.
1436  *
1437  * In other words check whether vnodes that are
1438  * held by vhold objects still have any locks
1439  * or shares or still in use. If they aren't,
1440  * just destroy them.
1441  */
1442 static void
nlm_host_gc_vholds(struct nlm_host * hostp)1443 nlm_host_gc_vholds(struct nlm_host *hostp)
1444 {
1445 	struct nlm_vhold *nvp;
1446 
1447 	ASSERT(MUTEX_HELD(&hostp->nh_lock));
1448 
1449 	nvp = TAILQ_FIRST(&hostp->nh_vholds_list);
1450 	while (nvp != NULL) {
1451 		struct nlm_vhold *nvp_tmp;
1452 
1453 		if (nlm_vhold_busy(hostp, nvp)) {
1454 			nvp = TAILQ_NEXT(nvp, nv_link);
1455 			continue;
1456 		}
1457 
1458 		nvp_tmp = TAILQ_NEXT(nvp, nv_link);
1459 		nlm_vhold_destroy(hostp, nvp);
1460 		nvp = nvp_tmp;
1461 	}
1462 }
1463 
1464 /*
1465  * Check whether the given host has any
1466  * server side locks or share reservations.
1467  */
1468 static bool_t
nlm_host_has_srv_locks(struct nlm_host * hostp)1469 nlm_host_has_srv_locks(struct nlm_host *hostp)
1470 {
1471 	/*
1472 	 * It's cheap and simple: if server has
1473 	 * any locks/shares there must be vhold
1474 	 * object storing the affected vnode.
1475 	 *
1476 	 * NOTE: We don't need to check sleeping
1477 	 * locks on the server side, because if
1478 	 * server side sleeping lock is alive,
1479 	 * there must be a vhold object corresponding
1480 	 * to target vnode.
1481 	 */
1482 	ASSERT(MUTEX_HELD(&hostp->nh_lock));
1483 	if (!TAILQ_EMPTY(&hostp->nh_vholds_list))
1484 		return (TRUE);
1485 
1486 	return (FALSE);
1487 }
1488 
1489 /*
1490  * Check whether the given host has any client side
1491  * locks or share reservations.
1492  */
1493 static bool_t
nlm_host_has_cli_locks(struct nlm_host * hostp)1494 nlm_host_has_cli_locks(struct nlm_host *hostp)
1495 {
1496 	ASSERT(MUTEX_HELD(&hostp->nh_lock));
1497 
1498 	/*
1499 	 * XXX: It's not the way I'd like to do the check,
1500 	 * because flk_sysid_has_locks() can be very
1501 	 * expensive by design. Unfortunatelly it iterates
1502 	 * through all locks on the system, doesn't matter
1503 	 * were they made on remote system via NLM or
1504 	 * on local system via reclock. To understand the
1505 	 * problem, consider that there're dozens of thousands
1506 	 * of locks that are made on some ZFS dataset. And there's
1507 	 * another dataset shared by NFS where NLM client had locks
1508 	 * some time ago, but doesn't have them now.
1509 	 * In this case flk_sysid_has_locks() will iterate
1510 	 * thrught dozens of thousands locks until it returns us
1511 	 * FALSE.
1512 	 * Oh, I hope that in shiny future somebody will make
1513 	 * local lock manager (os/flock.c) better, so that
1514 	 * it'd be more friedly to remote locks and
1515 	 * flk_sysid_has_locks() wouldn't be so expensive.
1516 	 */
1517 	if (flk_sysid_has_locks(hostp->nh_sysid |
1518 	    LM_SYSID_CLIENT, FLK_QUERY_ACTIVE))
1519 		return (TRUE);
1520 
1521 	/*
1522 	 * Check whether host has any share reservations
1523 	 * registered on the client side.
1524 	 */
1525 	if (hostp->nh_shrlist != NULL)
1526 		return (TRUE);
1527 
1528 	return (FALSE);
1529 }
1530 
1531 /*
1532  * Determine whether the given host owns any
1533  * locks or share reservations.
1534  */
1535 static bool_t
nlm_host_has_locks(struct nlm_host * hostp)1536 nlm_host_has_locks(struct nlm_host *hostp)
1537 {
1538 	if (nlm_host_has_srv_locks(hostp))
1539 		return (TRUE);
1540 
1541 	return (nlm_host_has_cli_locks(hostp));
1542 }
1543 
1544 /*
1545  * This function compares only addresses of two netbufs
1546  * that belong to NC_TCP[6] or NC_UDP[6] protofamily.
1547  * Port part of netbuf is ignored.
1548  *
1549  * Return values:
1550  *  -1: nb1's address is "smaller" than nb2's
1551  *   0: addresses are equal
1552  *   1: nb1's address is "greater" than nb2's
1553  */
1554 static int
nlm_netbuf_addrs_cmp(struct netbuf * nb1,struct netbuf * nb2)1555 nlm_netbuf_addrs_cmp(struct netbuf *nb1, struct netbuf *nb2)
1556 {
1557 	union nlm_addr {
1558 		struct sockaddr sa;
1559 		struct sockaddr_in sin;
1560 		struct sockaddr_in6 sin6;
1561 	} *na1, *na2;
1562 	int res;
1563 
1564 	/* LINTED E_BAD_PTR_CAST_ALIGN */
1565 	na1 = (union nlm_addr *)nb1->buf;
1566 	/* LINTED E_BAD_PTR_CAST_ALIGN */
1567 	na2 = (union nlm_addr *)nb2->buf;
1568 
1569 	if (na1->sa.sa_family < na2->sa.sa_family)
1570 		return (-1);
1571 	if (na1->sa.sa_family > na2->sa.sa_family)
1572 		return (1);
1573 
1574 	switch (na1->sa.sa_family) {
1575 	case AF_INET:
1576 		res = memcmp(&na1->sin.sin_addr, &na2->sin.sin_addr,
1577 		    sizeof (na1->sin.sin_addr));
1578 		break;
1579 	case AF_INET6:
1580 		res = memcmp(&na1->sin6.sin6_addr, &na2->sin6.sin6_addr,
1581 		    sizeof (na1->sin6.sin6_addr));
1582 		break;
1583 	default:
1584 		VERIFY(0);
1585 		return (0);
1586 	}
1587 
1588 	return (SIGN(res));
1589 }
1590 
1591 /*
1592  * Compare two nlm hosts.
1593  * Return values:
1594  * -1: host1 is "smaller" than host2
1595  *  0: host1 is equal to host2
1596  *  1: host1 is "greater" than host2
1597  */
1598 int
nlm_host_cmp(const void * p1,const void * p2)1599 nlm_host_cmp(const void *p1, const void *p2)
1600 {
1601 	struct nlm_host *h1 = (struct nlm_host *)p1;
1602 	struct nlm_host *h2 = (struct nlm_host *)p2;
1603 	int res;
1604 
1605 	res = strcmp(h1->nh_netid, h2->nh_netid);
1606 	if (res != 0)
1607 		return (SIGN(res));
1608 
1609 	res = nlm_netbuf_addrs_cmp(&h1->nh_addr, &h2->nh_addr);
1610 	return (res);
1611 }
1612 
1613 /*
1614  * Find the host specified by...  (see below)
1615  * If found, increment the ref count.
1616  */
1617 static struct nlm_host *
nlm_host_find_locked(struct nlm_globals * g,const char * netid,struct netbuf * naddr,avl_index_t * wherep)1618 nlm_host_find_locked(struct nlm_globals *g, const char *netid,
1619     struct netbuf *naddr, avl_index_t *wherep)
1620 {
1621 	struct nlm_host *hostp, key;
1622 	avl_index_t pos;
1623 
1624 	ASSERT(MUTEX_HELD(&g->lock));
1625 
1626 	key.nh_netid = (char *)netid;
1627 	key.nh_addr.buf = naddr->buf;
1628 	key.nh_addr.len = naddr->len;
1629 	key.nh_addr.maxlen = naddr->maxlen;
1630 
1631 	hostp = avl_find(&g->nlm_hosts_tree, &key, &pos);
1632 
1633 	if (hostp != NULL) {
1634 		/*
1635 		 * Host is inuse now. Remove it from idle
1636 		 * hosts list if needed.
1637 		 */
1638 		if (hostp->nh_flags & NLM_NH_INIDLE) {
1639 			TAILQ_REMOVE(&g->nlm_idle_hosts, hostp, nh_link);
1640 			hostp->nh_flags &= ~NLM_NH_INIDLE;
1641 		}
1642 
1643 		hostp->nh_refs++;
1644 	}
1645 	if (wherep != NULL)
1646 		*wherep = pos;
1647 
1648 	return (hostp);
1649 }
1650 
1651 /*
1652  * Find NLM host for the given name and address.
1653  */
1654 struct nlm_host *
nlm_host_find(struct nlm_globals * g,const char * netid,struct netbuf * addr)1655 nlm_host_find(struct nlm_globals *g, const char *netid,
1656     struct netbuf *addr)
1657 {
1658 	struct nlm_host *hostp = NULL;
1659 
1660 	mutex_enter(&g->lock);
1661 	if (g->run_status != NLM_ST_UP)
1662 		goto out;
1663 
1664 	hostp = nlm_host_find_locked(g, netid, addr, NULL);
1665 
1666 out:
1667 	mutex_exit(&g->lock);
1668 	return (hostp);
1669 }
1670 
1671 
1672 /*
1673  * Find or create an NLM host for the given name and address.
1674  *
1675  * The remote host is determined by all of: name, netid, address.
1676  * Note that the netid is whatever nlm_svc_add_ep() gave to
1677  * svc_tli_kcreate() for the service binding.  If any of these
1678  * are different, allocate a new host (new sysid).
1679  */
1680 struct nlm_host *
nlm_host_findcreate(struct nlm_globals * g,char * name,const char * netid,struct netbuf * addr)1681 nlm_host_findcreate(struct nlm_globals *g, char *name,
1682     const char *netid, struct netbuf *addr)
1683 {
1684 	int err;
1685 	struct nlm_host *host, *newhost = NULL;
1686 	struct knetconfig knc;
1687 	avl_index_t where;
1688 
1689 	mutex_enter(&g->lock);
1690 	if (g->run_status != NLM_ST_UP) {
1691 		mutex_exit(&g->lock);
1692 		return (NULL);
1693 	}
1694 
1695 	host = nlm_host_find_locked(g, netid, addr, NULL);
1696 	mutex_exit(&g->lock);
1697 	if (host != NULL)
1698 		return (host);
1699 
1700 	err = nlm_knc_from_netid(netid, &knc);
1701 	if (err != 0)
1702 		return (NULL);
1703 	/*
1704 	 * Do allocations (etc.) outside of mutex,
1705 	 * and then check again before inserting.
1706 	 */
1707 	newhost = nlm_host_create(name, netid, &knc, addr);
1708 	newhost->nh_sysid = nlm_sysid_alloc();
1709 	if (newhost->nh_sysid == LM_NOSYSID)
1710 		goto out;
1711 
1712 	mutex_enter(&g->lock);
1713 	host = nlm_host_find_locked(g, netid, addr, &where);
1714 	if (host == NULL) {
1715 		host = newhost;
1716 		newhost = NULL;
1717 
1718 		/*
1719 		 * Insert host to the hosts AVL tree that is
1720 		 * used to lookup by <netid, address> pair.
1721 		 */
1722 		avl_insert(&g->nlm_hosts_tree, host, where);
1723 
1724 		/*
1725 		 * Insert host to the hosts hash table that is
1726 		 * used to lookup host by sysid.
1727 		 */
1728 		VERIFY(mod_hash_insert(g->nlm_hosts_hash,
1729 		    (mod_hash_key_t)(uintptr_t)host->nh_sysid,
1730 		    (mod_hash_val_t)host) == 0);
1731 	}
1732 
1733 	mutex_exit(&g->lock);
1734 
1735 out:
1736 	if (newhost != NULL) {
1737 		/*
1738 		 * We do not need the preallocated nlm_host
1739 		 * so decrement the reference counter
1740 		 * and destroy it.
1741 		 */
1742 		newhost->nh_refs--;
1743 		nlm_host_destroy(newhost);
1744 	}
1745 
1746 	return (host);
1747 }
1748 
1749 /*
1750  * Find the NLM host that matches the value of 'sysid'.
1751  * If found, return it with a new ref,
1752  * else return NULL.
1753  */
1754 struct nlm_host *
nlm_host_find_by_sysid(struct nlm_globals * g,sysid_t sysid)1755 nlm_host_find_by_sysid(struct nlm_globals *g, sysid_t sysid)
1756 {
1757 	struct nlm_host *hostp = NULL;
1758 
1759 	mutex_enter(&g->lock);
1760 	if (g->run_status != NLM_ST_UP)
1761 		goto out;
1762 
1763 	(void) mod_hash_find(g->nlm_hosts_hash,
1764 	    (mod_hash_key_t)(uintptr_t)sysid,
1765 	    (mod_hash_val_t)&hostp);
1766 
1767 	if (hostp == NULL)
1768 		goto out;
1769 
1770 	/*
1771 	 * Host is inuse now. Remove it
1772 	 * from idle hosts list if needed.
1773 	 */
1774 	if (hostp->nh_flags & NLM_NH_INIDLE) {
1775 		TAILQ_REMOVE(&g->nlm_idle_hosts, hostp, nh_link);
1776 		hostp->nh_flags &= ~NLM_NH_INIDLE;
1777 	}
1778 
1779 	hostp->nh_refs++;
1780 
1781 out:
1782 	mutex_exit(&g->lock);
1783 	return (hostp);
1784 }
1785 
1786 /*
1787  * Release the given host.
1788  * I.e. drop a reference that was taken earlier by one of
1789  * the following functions: nlm_host_findcreate(), nlm_host_find(),
1790  * nlm_host_find_by_sysid().
1791  *
1792  * When the very last reference is dropped, host is moved to
1793  * so-called "idle state". All hosts that are in idle state
1794  * have an idle timeout. If timeout is expired, GC thread
1795  * checks whether hosts have any locks and if they heven't
1796  * any, it removes them.
1797  * NOTE: only unused hosts can be in idle state.
1798  */
1799 static void
nlm_host_release_locked(struct nlm_globals * g,struct nlm_host * hostp)1800 nlm_host_release_locked(struct nlm_globals *g, struct nlm_host *hostp)
1801 {
1802 	if (hostp == NULL)
1803 		return;
1804 
1805 	ASSERT(MUTEX_HELD(&g->lock));
1806 	ASSERT(hostp->nh_refs > 0);
1807 
1808 	hostp->nh_refs--;
1809 	if (hostp->nh_refs != 0)
1810 		return;
1811 
1812 	/*
1813 	 * The very last reference to the host was dropped,
1814 	 * thus host is unused now. Set its idle timeout
1815 	 * and move it to the idle hosts LRU list.
1816 	 */
1817 	hostp->nh_idle_timeout = ddi_get_lbolt() +
1818 	    SEC_TO_TICK(g->cn_idle_tmo);
1819 
1820 	ASSERT((hostp->nh_flags & NLM_NH_INIDLE) == 0);
1821 	TAILQ_INSERT_TAIL(&g->nlm_idle_hosts, hostp, nh_link);
1822 	hostp->nh_flags |= NLM_NH_INIDLE;
1823 }
1824 
1825 void
nlm_host_release(struct nlm_globals * g,struct nlm_host * hostp)1826 nlm_host_release(struct nlm_globals *g, struct nlm_host *hostp)
1827 {
1828 	if (hostp == NULL)
1829 		return;
1830 
1831 	mutex_enter(&g->lock);
1832 	nlm_host_release_locked(g, hostp);
1833 	mutex_exit(&g->lock);
1834 }
1835 
1836 /*
1837  * Unregister this NLM host (NFS client) with the local statd
1838  * due to idleness (no locks held for a while).
1839  */
1840 void
nlm_host_unmonitor(struct nlm_globals * g,struct nlm_host * host)1841 nlm_host_unmonitor(struct nlm_globals *g, struct nlm_host *host)
1842 {
1843 	enum clnt_stat stat;
1844 
1845 	VERIFY(host->nh_refs == 0);
1846 	if (!(host->nh_flags & NLM_NH_MONITORED))
1847 		return;
1848 
1849 	host->nh_flags &= ~NLM_NH_MONITORED;
1850 	stat = nlm_nsm_unmon(&g->nlm_nsm, host->nh_name);
1851 	if (stat != RPC_SUCCESS) {
1852 		NLM_WARN("NLM: Failed to contact statd, stat=%d\n", stat);
1853 		return;
1854 	}
1855 }
1856 
1857 /*
1858  * Ask the local NFS statd to begin monitoring this host.
1859  * It will call us back when that host restarts, using the
1860  * prog,vers,proc specified below, i.e. NLM_SM_NOTIFY1,
1861  * which is handled in nlm_do_notify1().
1862  */
1863 void
nlm_host_monitor(struct nlm_globals * g,struct nlm_host * host,int state)1864 nlm_host_monitor(struct nlm_globals *g, struct nlm_host *host, int state)
1865 {
1866 	int family;
1867 	netobj obj;
1868 	enum clnt_stat stat;
1869 
1870 	if (state != 0 && host->nh_state == 0) {
1871 		/*
1872 		 * This is the first time we have seen an NSM state
1873 		 * Value for this host. We record it here to help
1874 		 * detect host reboots.
1875 		 */
1876 		host->nh_state = state;
1877 	}
1878 
1879 	mutex_enter(&host->nh_lock);
1880 	if (host->nh_flags & NLM_NH_MONITORED) {
1881 		mutex_exit(&host->nh_lock);
1882 		return;
1883 	}
1884 
1885 	host->nh_flags |= NLM_NH_MONITORED;
1886 	mutex_exit(&host->nh_lock);
1887 
1888 	/*
1889 	 * Before we begin monitoring the host register the network address
1890 	 * associated with this hostname.
1891 	 */
1892 	nlm_netbuf_to_netobj(&host->nh_addr, &family, &obj);
1893 	stat = nlm_nsmaddr_reg(&g->nlm_nsm, host->nh_name, family, &obj);
1894 	if (stat != RPC_SUCCESS) {
1895 		NLM_WARN("Failed to register address, stat=%d\n", stat);
1896 		mutex_enter(&g->lock);
1897 		host->nh_flags &= ~NLM_NH_MONITORED;
1898 		mutex_exit(&g->lock);
1899 
1900 		return;
1901 	}
1902 
1903 	/*
1904 	 * Tell statd how to call us with status updates for
1905 	 * this host. Updates arrive via nlm_do_notify1().
1906 	 *
1907 	 * We put our assigned system ID value in the priv field to
1908 	 * make it simpler to find the host if we are notified of a
1909 	 * host restart.
1910 	 */
1911 	stat = nlm_nsm_mon(&g->nlm_nsm, host->nh_name, host->nh_sysid);
1912 	if (stat != RPC_SUCCESS) {
1913 		NLM_WARN("Failed to contact local NSM, stat=%d\n", stat);
1914 		mutex_enter(&g->lock);
1915 		host->nh_flags &= ~NLM_NH_MONITORED;
1916 		mutex_exit(&g->lock);
1917 
1918 		return;
1919 	}
1920 }
1921 
1922 int
nlm_host_get_state(struct nlm_host * hostp)1923 nlm_host_get_state(struct nlm_host *hostp)
1924 {
1925 
1926 	return (hostp->nh_state);
1927 }
1928 
1929 /*
1930  * NLM client/server sleeping locks
1931  */
1932 
1933 /*
1934  * Register client side sleeping lock.
1935  *
1936  * Our client code calls this to keep information
1937  * about sleeping lock somewhere. When it receives
1938  * grant callback from server or when it just
1939  * needs to remove all sleeping locks from vnode,
1940  * it uses this information for remove/apply lock
1941  * properly.
1942  */
1943 struct nlm_slock *
nlm_slock_register(struct nlm_globals * g,struct nlm_host * host,struct nlm4_lock * lock,struct vnode * vp)1944 nlm_slock_register(
1945 	struct nlm_globals *g,
1946 	struct nlm_host *host,
1947 	struct nlm4_lock *lock,
1948 	struct vnode *vp)
1949 {
1950 	struct nlm_slock *nslp;
1951 
1952 	nslp = kmem_zalloc(sizeof (*nslp), KM_SLEEP);
1953 	cv_init(&nslp->nsl_cond, NULL, CV_DEFAULT, NULL);
1954 	nslp->nsl_lock = *lock;
1955 	nlm_copy_netobj(&nslp->nsl_fh, &nslp->nsl_lock.fh);
1956 	nslp->nsl_state = NLM_SL_BLOCKED;
1957 	nslp->nsl_host = host;
1958 	nslp->nsl_vp = vp;
1959 
1960 	mutex_enter(&g->lock);
1961 	TAILQ_INSERT_TAIL(&g->nlm_slocks, nslp, nsl_link);
1962 	mutex_exit(&g->lock);
1963 
1964 	return (nslp);
1965 }
1966 
1967 /*
1968  * Remove this lock from the wait list and destroy it.
1969  */
1970 void
nlm_slock_unregister(struct nlm_globals * g,struct nlm_slock * nslp)1971 nlm_slock_unregister(struct nlm_globals *g, struct nlm_slock *nslp)
1972 {
1973 	mutex_enter(&g->lock);
1974 	TAILQ_REMOVE(&g->nlm_slocks, nslp, nsl_link);
1975 	mutex_exit(&g->lock);
1976 
1977 	kmem_free(nslp->nsl_fh.n_bytes, nslp->nsl_fh.n_len);
1978 	cv_destroy(&nslp->nsl_cond);
1979 	kmem_free(nslp, sizeof (*nslp));
1980 }
1981 
1982 /*
1983  * Wait for a granted callback or cancellation event
1984  * for a sleeping lock.
1985  *
1986  * If a signal interrupted the wait or if the lock
1987  * was cancelled, return EINTR - the caller must arrange to send
1988  * a cancellation to the server.
1989  *
1990  * If timeout occurred, return ETIMEDOUT - the caller must
1991  * resend the lock request to the server.
1992  *
1993  * On success return 0.
1994  */
1995 int
nlm_slock_wait(struct nlm_globals * g,struct nlm_slock * nslp,uint_t timeo_secs)1996 nlm_slock_wait(struct nlm_globals *g,
1997     struct nlm_slock *nslp, uint_t timeo_secs)
1998 {
1999 	clock_t timeo_ticks;
2000 	int cv_res, error;
2001 
2002 	/*
2003 	 * If the granted message arrived before we got here,
2004 	 * nslp->nsl_state will be NLM_SL_GRANTED - in that case don't sleep.
2005 	 */
2006 	cv_res = 1;
2007 	timeo_ticks = ddi_get_lbolt() + SEC_TO_TICK(timeo_secs);
2008 
2009 	mutex_enter(&g->lock);
2010 	while (nslp->nsl_state == NLM_SL_BLOCKED && cv_res > 0) {
2011 		cv_res = cv_timedwait_sig(&nslp->nsl_cond,
2012 		    &g->lock, timeo_ticks);
2013 	}
2014 
2015 	/*
2016 	 * No matter why we wake up, if the lock was
2017 	 * cancelled, let the function caller to know
2018 	 * about it by returning EINTR.
2019 	 */
2020 	if (nslp->nsl_state == NLM_SL_CANCELLED) {
2021 		error = EINTR;
2022 		goto out;
2023 	}
2024 
2025 	if (cv_res <= 0) {
2026 		/* We were woken up either by timeout or by interrupt */
2027 		error = (cv_res < 0) ? ETIMEDOUT : EINTR;
2028 
2029 		/*
2030 		 * The granted message may arrive after the
2031 		 * interrupt/timeout but before we manage to lock the
2032 		 * mutex. Detect this by examining nslp.
2033 		 */
2034 		if (nslp->nsl_state == NLM_SL_GRANTED)
2035 			error = 0;
2036 	} else { /* Awaken via cv_signal()/cv_broadcast() or didn't block */
2037 		error = 0;
2038 		VERIFY(nslp->nsl_state == NLM_SL_GRANTED);
2039 	}
2040 
2041 out:
2042 	mutex_exit(&g->lock);
2043 	return (error);
2044 }
2045 
2046 /*
2047  * Mark client side sleeping lock as granted
2048  * and wake up a process blocked on the lock.
2049  * Called from server side NLM_GRANT handler.
2050  *
2051  * If sleeping lock is found return 0, otherwise
2052  * return ENOENT.
2053  */
2054 int
nlm_slock_grant(struct nlm_globals * g,struct nlm_host * hostp,struct nlm4_lock * alock)2055 nlm_slock_grant(struct nlm_globals *g,
2056     struct nlm_host *hostp, struct nlm4_lock *alock)
2057 {
2058 	struct nlm_slock *nslp;
2059 	int error = ENOENT;
2060 
2061 	mutex_enter(&g->lock);
2062 	TAILQ_FOREACH(nslp, &g->nlm_slocks, nsl_link) {
2063 		if ((nslp->nsl_state != NLM_SL_BLOCKED) ||
2064 		    (nslp->nsl_host != hostp))
2065 			continue;
2066 
2067 		if (alock->svid		== nslp->nsl_lock.svid &&
2068 		    alock->l_offset	== nslp->nsl_lock.l_offset &&
2069 		    alock->l_len	== nslp->nsl_lock.l_len &&
2070 		    alock->fh.n_len	== nslp->nsl_lock.fh.n_len &&
2071 		    bcmp(alock->fh.n_bytes, nslp->nsl_lock.fh.n_bytes,
2072 		    nslp->nsl_lock.fh.n_len) == 0) {
2073 			nslp->nsl_state = NLM_SL_GRANTED;
2074 			cv_broadcast(&nslp->nsl_cond);
2075 			error = 0;
2076 			break;
2077 		}
2078 	}
2079 
2080 	mutex_exit(&g->lock);
2081 	return (error);
2082 }
2083 
2084 /*
2085  * Register sleeping lock request corresponding to
2086  * flp on the given vhold object.
2087  * On success function returns 0, otherwise (if
2088  * lock request with the same flp is already
2089  * registered) function returns EEXIST.
2090  */
2091 int
nlm_slreq_register(struct nlm_host * hostp,struct nlm_vhold * nvp,struct flock64 * flp)2092 nlm_slreq_register(struct nlm_host *hostp, struct nlm_vhold *nvp,
2093     struct flock64 *flp)
2094 {
2095 	struct nlm_slreq *slr, *new_slr = NULL;
2096 	int ret = EEXIST;
2097 
2098 	mutex_enter(&hostp->nh_lock);
2099 	slr = nlm_slreq_find_locked(hostp, nvp, flp);
2100 	if (slr != NULL)
2101 		goto out;
2102 
2103 	mutex_exit(&hostp->nh_lock);
2104 	new_slr = kmem_zalloc(sizeof (*slr), KM_SLEEP);
2105 	bcopy(flp, &new_slr->nsr_fl, sizeof (*flp));
2106 
2107 	mutex_enter(&hostp->nh_lock);
2108 	slr = nlm_slreq_find_locked(hostp, nvp, flp);
2109 	if (slr == NULL) {
2110 		slr = new_slr;
2111 		new_slr = NULL;
2112 		ret = 0;
2113 
2114 		TAILQ_INSERT_TAIL(&nvp->nv_slreqs, slr, nsr_link);
2115 	}
2116 
2117 out:
2118 	mutex_exit(&hostp->nh_lock);
2119 	if (new_slr != NULL)
2120 		kmem_free(new_slr, sizeof (*new_slr));
2121 
2122 	return (ret);
2123 }
2124 
2125 /*
2126  * Unregister sleeping lock request corresponding
2127  * to flp from the given vhold object.
2128  * On success function returns 0, otherwise (if
2129  * lock request corresponding to flp isn't found
2130  * on the given vhold) function returns ENOENT.
2131  */
2132 int
nlm_slreq_unregister(struct nlm_host * hostp,struct nlm_vhold * nvp,struct flock64 * flp)2133 nlm_slreq_unregister(struct nlm_host *hostp, struct nlm_vhold *nvp,
2134     struct flock64 *flp)
2135 {
2136 	struct nlm_slreq *slr;
2137 
2138 	mutex_enter(&hostp->nh_lock);
2139 	slr = nlm_slreq_find_locked(hostp, nvp, flp);
2140 	if (slr == NULL) {
2141 		mutex_exit(&hostp->nh_lock);
2142 		return (ENOENT);
2143 	}
2144 
2145 	TAILQ_REMOVE(&nvp->nv_slreqs, slr, nsr_link);
2146 	mutex_exit(&hostp->nh_lock);
2147 
2148 	kmem_free(slr, sizeof (*slr));
2149 	return (0);
2150 }
2151 
2152 /*
2153  * Find sleeping lock request on the given vhold object by flp.
2154  */
2155 struct nlm_slreq *
nlm_slreq_find_locked(struct nlm_host * hostp,struct nlm_vhold * nvp,struct flock64 * flp)2156 nlm_slreq_find_locked(struct nlm_host *hostp, struct nlm_vhold *nvp,
2157     struct flock64 *flp)
2158 {
2159 	struct nlm_slreq *slr = NULL;
2160 
2161 	ASSERT(MUTEX_HELD(&hostp->nh_lock));
2162 	TAILQ_FOREACH(slr, &nvp->nv_slreqs, nsr_link) {
2163 		if (slr->nsr_fl.l_start		== flp->l_start	&&
2164 		    slr->nsr_fl.l_len		== flp->l_len	&&
2165 		    slr->nsr_fl.l_pid		== flp->l_pid	&&
2166 		    slr->nsr_fl.l_type		== flp->l_type)
2167 			break;
2168 	}
2169 
2170 	return (slr);
2171 }
2172 
2173 /*
2174  * NLM tracks active share reservations made on the client side.
2175  * It needs to have a track of share reservations for two purposes
2176  * 1) to determine if nlm_host is busy (if it has active locks and/or
2177  *    share reservations, it is)
2178  * 2) to recover active share reservations when NLM server reports
2179  *    that it has rebooted.
2180  *
2181  * Unfortunately Illumos local share reservations manager (see os/share.c)
2182  * doesn't have an ability to lookup all reservations on the system
2183  * by sysid (like local lock manager) or get all reservations by sysid.
2184  * It tracks reservations per vnode and is able to get/looup them
2185  * on particular vnode. It's not what NLM needs. Thus it has that ugly
2186  * share reservations tracking scheme.
2187  */
2188 
2189 void
nlm_shres_track(struct nlm_host * hostp,vnode_t * vp,struct shrlock * shrp)2190 nlm_shres_track(struct nlm_host *hostp, vnode_t *vp, struct shrlock *shrp)
2191 {
2192 	struct nlm_shres *nsp, *nsp_new;
2193 
2194 	/*
2195 	 * NFS code must fill the s_owner, so that
2196 	 * s_own_len is never 0.
2197 	 */
2198 	ASSERT(shrp->s_own_len > 0);
2199 	nsp_new = nlm_shres_create_item(shrp, vp);
2200 
2201 	mutex_enter(&hostp->nh_lock);
2202 	for (nsp = hostp->nh_shrlist; nsp != NULL; nsp = nsp->ns_next)
2203 		if (nsp->ns_vp == vp && nlm_shres_equal(shrp, nsp->ns_shr))
2204 			break;
2205 
2206 	if (nsp != NULL) {
2207 		/*
2208 		 * Found a duplicate. Do nothing.
2209 		 */
2210 
2211 		goto out;
2212 	}
2213 
2214 	nsp = nsp_new;
2215 	nsp_new = NULL;
2216 	nsp->ns_next = hostp->nh_shrlist;
2217 	hostp->nh_shrlist = nsp;
2218 
2219 out:
2220 	mutex_exit(&hostp->nh_lock);
2221 	if (nsp_new != NULL)
2222 		nlm_shres_destroy_item(nsp_new);
2223 }
2224 
2225 void
nlm_shres_untrack(struct nlm_host * hostp,vnode_t * vp,struct shrlock * shrp)2226 nlm_shres_untrack(struct nlm_host *hostp, vnode_t *vp, struct shrlock *shrp)
2227 {
2228 	struct nlm_shres *nsp, *nsp_prev = NULL;
2229 
2230 	mutex_enter(&hostp->nh_lock);
2231 	nsp = hostp->nh_shrlist;
2232 	while (nsp != NULL) {
2233 		if (nsp->ns_vp == vp && nlm_shres_equal(shrp, nsp->ns_shr)) {
2234 			struct nlm_shres *nsp_del;
2235 
2236 			nsp_del = nsp;
2237 			nsp = nsp->ns_next;
2238 			if (nsp_prev != NULL)
2239 				nsp_prev->ns_next = nsp;
2240 			else
2241 				hostp->nh_shrlist = nsp;
2242 
2243 			nlm_shres_destroy_item(nsp_del);
2244 			continue;
2245 		}
2246 
2247 		nsp_prev = nsp;
2248 		nsp = nsp->ns_next;
2249 	}
2250 
2251 	mutex_exit(&hostp->nh_lock);
2252 }
2253 
2254 /*
2255  * Get a _copy_ of the list of all active share reservations
2256  * made by the given host.
2257  * NOTE: the list function returns _must_ be released using
2258  *       nlm_free_shrlist().
2259  */
2260 struct nlm_shres *
nlm_get_active_shres(struct nlm_host * hostp)2261 nlm_get_active_shres(struct nlm_host *hostp)
2262 {
2263 	struct nlm_shres *nsp, *nslist = NULL;
2264 
2265 	mutex_enter(&hostp->nh_lock);
2266 	for (nsp = hostp->nh_shrlist; nsp != NULL; nsp = nsp->ns_next) {
2267 		struct nlm_shres *nsp_new;
2268 
2269 		nsp_new = nlm_shres_create_item(nsp->ns_shr, nsp->ns_vp);
2270 		nsp_new->ns_next = nslist;
2271 		nslist = nsp_new;
2272 	}
2273 
2274 	mutex_exit(&hostp->nh_lock);
2275 	return (nslist);
2276 }
2277 
2278 /*
2279  * Free memory allocated for the active share reservations
2280  * list created by nlm_get_active_shres() function.
2281  */
2282 void
nlm_free_shrlist(struct nlm_shres * nslist)2283 nlm_free_shrlist(struct nlm_shres *nslist)
2284 {
2285 	struct nlm_shres *nsp;
2286 
2287 	while (nslist != NULL) {
2288 		nsp =  nslist;
2289 		nslist = nslist->ns_next;
2290 
2291 		nlm_shres_destroy_item(nsp);
2292 	}
2293 }
2294 
2295 static bool_t
nlm_shres_equal(struct shrlock * shrp1,struct shrlock * shrp2)2296 nlm_shres_equal(struct shrlock *shrp1, struct shrlock *shrp2)
2297 {
2298 	if (shrp1->s_sysid	== shrp2->s_sysid	&&
2299 	    shrp1->s_pid	== shrp2->s_pid		&&
2300 	    shrp1->s_own_len	== shrp2->s_own_len	&&
2301 	    bcmp(shrp1->s_owner, shrp2->s_owner,
2302 	    shrp1->s_own_len) == 0)
2303 		return (TRUE);
2304 
2305 	return (FALSE);
2306 }
2307 
2308 static struct nlm_shres *
nlm_shres_create_item(struct shrlock * shrp,vnode_t * vp)2309 nlm_shres_create_item(struct shrlock *shrp, vnode_t *vp)
2310 {
2311 	struct nlm_shres *nsp;
2312 
2313 	nsp = kmem_alloc(sizeof (*nsp), KM_SLEEP);
2314 	nsp->ns_shr = kmem_alloc(sizeof (*shrp), KM_SLEEP);
2315 	bcopy(shrp, nsp->ns_shr, sizeof (*shrp));
2316 	nsp->ns_shr->s_owner = kmem_alloc(shrp->s_own_len, KM_SLEEP);
2317 	bcopy(shrp->s_owner, nsp->ns_shr->s_owner, shrp->s_own_len);
2318 	nsp->ns_vp = vp;
2319 
2320 	return (nsp);
2321 }
2322 
2323 static void
nlm_shres_destroy_item(struct nlm_shres * nsp)2324 nlm_shres_destroy_item(struct nlm_shres *nsp)
2325 {
2326 	kmem_free(nsp->ns_shr->s_owner,
2327 	    nsp->ns_shr->s_own_len);
2328 	kmem_free(nsp->ns_shr, sizeof (struct shrlock));
2329 	kmem_free(nsp, sizeof (*nsp));
2330 }
2331 
2332 /*
2333  * Called by klmmod.c when lockd adds a network endpoint
2334  * on which we should begin RPC services.
2335  */
2336 int
nlm_svc_add_ep(struct file * fp,const char * netid,struct knetconfig * knc)2337 nlm_svc_add_ep(struct file *fp, const char *netid, struct knetconfig *knc)
2338 {
2339 	SVCMASTERXPRT *xprt = NULL;
2340 	int error;
2341 
2342 	error = svc_tli_kcreate(fp, 0, (char *)netid, NULL, &xprt,
2343 	    &nlm_sct, NULL, NLM_SVCPOOL_ID, FALSE);
2344 	if (error != 0)
2345 		return (error);
2346 
2347 	(void) nlm_knc_to_netid(knc);
2348 	return (0);
2349 }
2350 
2351 /*
2352  * Start NLM service.
2353  */
2354 int
nlm_svc_starting(struct nlm_globals * g,struct file * fp,const char * netid,struct knetconfig * knc)2355 nlm_svc_starting(struct nlm_globals *g, struct file *fp,
2356     const char *netid, struct knetconfig *knc)
2357 {
2358 	int error;
2359 	enum clnt_stat stat;
2360 
2361 	VERIFY(g->run_status == NLM_ST_STARTING);
2362 	VERIFY(g->nlm_gc_thread == NULL);
2363 
2364 	error = nlm_nsm_init_local(&g->nlm_nsm);
2365 	if (error != 0) {
2366 		NLM_ERR("Failed to initialize NSM handler "
2367 		    "(error=%d)\n", error);
2368 		g->run_status = NLM_ST_DOWN;
2369 		return (error);
2370 	}
2371 
2372 	error = EIO;
2373 
2374 	/*
2375 	 * Create an NLM garbage collector thread that will
2376 	 * clean up stale vholds and hosts objects.
2377 	 */
2378 	g->nlm_gc_thread = zthread_create(NULL, 0, nlm_gc,
2379 	    g, 0, minclsyspri);
2380 
2381 	/*
2382 	 * Send SIMU_CRASH to local statd to report that
2383 	 * NLM started, so that statd can report other hosts
2384 	 * about NLM state change.
2385 	 */
2386 
2387 	stat = nlm_nsm_simu_crash(&g->nlm_nsm);
2388 	if (stat != RPC_SUCCESS) {
2389 		NLM_ERR("Failed to connect to local statd "
2390 		    "(rpcerr=%d)\n", stat);
2391 		goto shutdown_lm;
2392 	}
2393 
2394 	stat = nlm_nsm_stat(&g->nlm_nsm, &g->nsm_state);
2395 	if (stat != RPC_SUCCESS) {
2396 		NLM_ERR("Failed to get the status of local statd "
2397 		    "(rpcerr=%d)\n", stat);
2398 		goto shutdown_lm;
2399 	}
2400 
2401 	g->grace_threshold = ddi_get_lbolt() +
2402 	    SEC_TO_TICK(g->grace_period);
2403 
2404 	/* Register endpoint used for communications with local NLM */
2405 	error = nlm_svc_add_ep(fp, netid, knc);
2406 	if (error != 0)
2407 		goto shutdown_lm;
2408 
2409 	(void) svc_pool_control(NLM_SVCPOOL_ID,
2410 	    SVCPSET_SHUTDOWN_PROC, (void *)nlm_pool_shutdown);
2411 	g->run_status = NLM_ST_UP;
2412 	return (0);
2413 
2414 shutdown_lm:
2415 	mutex_enter(&g->lock);
2416 	g->run_status = NLM_ST_STOPPING;
2417 	mutex_exit(&g->lock);
2418 
2419 	nlm_svc_stopping(g);
2420 	return (error);
2421 }
2422 
2423 /*
2424  * Called when the server pool is destroyed, so that
2425  * all transports are closed and no any server threads
2426  * exist.
2427  *
2428  * Just call lm_shutdown() to shut NLM down properly.
2429  */
2430 static void
nlm_pool_shutdown(void)2431 nlm_pool_shutdown(void)
2432 {
2433 	(void) lm_shutdown();
2434 }
2435 
2436 /*
2437  * Stop NLM service, cleanup all resources
2438  * NLM owns at the moment.
2439  *
2440  * NOTE: NFS code can call NLM while it's
2441  * stopping or even if it's shut down. Any attempt
2442  * to lock file either on client or on the server
2443  * will fail if NLM isn't in NLM_ST_UP state.
2444  */
2445 void
nlm_svc_stopping(struct nlm_globals * g)2446 nlm_svc_stopping(struct nlm_globals *g)
2447 {
2448 	mutex_enter(&g->lock);
2449 	ASSERT(g->run_status == NLM_ST_STOPPING);
2450 
2451 	/*
2452 	 * Ask NLM GC thread to exit and wait until it dies.
2453 	 */
2454 	cv_signal(&g->nlm_gc_sched_cv);
2455 	while (g->nlm_gc_thread != NULL)
2456 		cv_wait(&g->nlm_gc_finish_cv, &g->lock);
2457 
2458 	mutex_exit(&g->lock);
2459 
2460 	/*
2461 	 * Cleanup locks owned by NLM hosts.
2462 	 * NOTE: New hosts won't be created while
2463 	 * NLM is stopping.
2464 	 */
2465 	while (!avl_is_empty(&g->nlm_hosts_tree)) {
2466 		struct nlm_host *hostp;
2467 		int busy_hosts = 0;
2468 
2469 		/*
2470 		 * Iterate through all NLM hosts in the system
2471 		 * and drop the locks they own by force.
2472 		 */
2473 		hostp = avl_first(&g->nlm_hosts_tree);
2474 		while (hostp != NULL) {
2475 			/* Cleanup all client and server side locks */
2476 			nlm_client_cancel_all(g, hostp);
2477 			nlm_host_notify_server(hostp, 0);
2478 
2479 			mutex_enter(&hostp->nh_lock);
2480 			nlm_host_gc_vholds(hostp);
2481 			if (hostp->nh_refs > 0 || nlm_host_has_locks(hostp)) {
2482 				/*
2483 				 * Oh, it seems the host is still busy, let
2484 				 * it some time to release and go to the
2485 				 * next one.
2486 				 */
2487 
2488 				mutex_exit(&hostp->nh_lock);
2489 				hostp = AVL_NEXT(&g->nlm_hosts_tree, hostp);
2490 				busy_hosts++;
2491 				continue;
2492 			}
2493 
2494 			mutex_exit(&hostp->nh_lock);
2495 			hostp = AVL_NEXT(&g->nlm_hosts_tree, hostp);
2496 		}
2497 
2498 		/*
2499 		 * All hosts go to nlm_idle_hosts list after
2500 		 * all locks they own are cleaned up and last refereces
2501 		 * were dropped. Just destroy all hosts in nlm_idle_hosts
2502 		 * list, they can not be removed from there while we're
2503 		 * in stopping state.
2504 		 */
2505 		while ((hostp = TAILQ_FIRST(&g->nlm_idle_hosts)) != NULL) {
2506 			nlm_host_unregister(g, hostp);
2507 			nlm_host_destroy(hostp);
2508 		}
2509 
2510 		if (busy_hosts > 0) {
2511 			/*
2512 			 * There're some hosts that weren't cleaned
2513 			 * up. Probably they're in resource cleanup
2514 			 * process. Give them some time to do drop
2515 			 * references.
2516 			 */
2517 			delay(MSEC_TO_TICK(500));
2518 		}
2519 	}
2520 
2521 	ASSERT(TAILQ_EMPTY(&g->nlm_slocks));
2522 
2523 	nlm_nsm_fini(&g->nlm_nsm);
2524 	g->lockd_pid = 0;
2525 	g->run_status = NLM_ST_DOWN;
2526 }
2527 
2528 /*
2529  * Returns TRUE if the given vnode has
2530  * any active or sleeping locks.
2531  */
2532 int
nlm_vp_active(const vnode_t * vp)2533 nlm_vp_active(const vnode_t *vp)
2534 {
2535 	struct nlm_globals *g;
2536 	struct nlm_host *hostp;
2537 	struct nlm_vhold *nvp;
2538 	int active = 0;
2539 
2540 	g = zone_getspecific(nlm_zone_key, curzone);
2541 
2542 	/*
2543 	 * Server side NLM has locks on the given vnode
2544 	 * if there exist a vhold object that holds
2545 	 * the given vnode "vp" in one of NLM hosts.
2546 	 */
2547 	mutex_enter(&g->lock);
2548 	hostp = avl_first(&g->nlm_hosts_tree);
2549 	while (hostp != NULL) {
2550 		mutex_enter(&hostp->nh_lock);
2551 		nvp = nlm_vhold_find_locked(hostp, vp);
2552 		mutex_exit(&hostp->nh_lock);
2553 		if (nvp != NULL) {
2554 			active = 1;
2555 			break;
2556 		}
2557 
2558 		hostp = AVL_NEXT(&g->nlm_hosts_tree, hostp);
2559 	}
2560 
2561 	mutex_exit(&g->lock);
2562 	return (active);
2563 }
2564 
2565 /*
2566  * Called right before NFS export is going to
2567  * dissapear. The function finds all vnodes
2568  * belonging to the given export and cleans
2569  * all remote locks and share reservations
2570  * on them.
2571  */
2572 void
nlm_zone_unexport(struct nlm_globals * g,struct exportinfo * exi)2573 nlm_zone_unexport(struct nlm_globals *g, struct exportinfo *exi)
2574 {
2575 	struct nlm_host *hostp;
2576 
2577 	mutex_enter(&g->lock);
2578 	if (g->run_status != NLM_ST_UP) {
2579 		/* nothing to do */
2580 		mutex_exit(&g->lock);
2581 		return;
2582 	}
2583 
2584 	hostp = avl_first(&g->nlm_hosts_tree);
2585 	while (hostp != NULL) {
2586 		struct nlm_vhold *nvp;
2587 
2588 		if (hostp->nh_flags & NLM_NH_INIDLE) {
2589 			TAILQ_REMOVE(&g->nlm_idle_hosts, hostp, nh_link);
2590 			hostp->nh_flags &= ~NLM_NH_INIDLE;
2591 		}
2592 		hostp->nh_refs++;
2593 
2594 		mutex_exit(&g->lock);
2595 
2596 		mutex_enter(&hostp->nh_lock);
2597 		TAILQ_FOREACH(nvp, &hostp->nh_vholds_list, nv_link) {
2598 			vnode_t *vp;
2599 
2600 			nvp->nv_refcnt++;
2601 			mutex_exit(&hostp->nh_lock);
2602 
2603 			vp = nvp->nv_vp;
2604 
2605 			if (!EQFSID(&exi->exi_fsid, &vp->v_vfsp->vfs_fsid))
2606 				goto next_iter;
2607 
2608 			/*
2609 			 * Ok, it we found out that vnode vp is under
2610 			 * control by the exportinfo exi, now we need
2611 			 * to drop all locks from this vnode, let's
2612 			 * do it.
2613 			 */
2614 			nlm_vhold_clean(nvp, hostp->nh_sysid);
2615 
2616 		next_iter:
2617 			mutex_enter(&hostp->nh_lock);
2618 			nvp->nv_refcnt--;
2619 		}
2620 		mutex_exit(&hostp->nh_lock);
2621 
2622 		mutex_enter(&g->lock);
2623 		nlm_host_release_locked(g, hostp);
2624 
2625 		hostp = AVL_NEXT(&g->nlm_hosts_tree, hostp);
2626 	}
2627 
2628 	mutex_exit(&g->lock);
2629 }
2630 
2631 void
nlm_unexport(struct exportinfo * exi)2632 nlm_unexport(struct exportinfo *exi)
2633 {
2634 	struct nlm_globals *g;
2635 
2636 	rw_enter(&lm_lck, RW_READER);
2637 	TAILQ_FOREACH(g, &nlm_zones_list, nlm_link) {
2638 		if (g->nlm_zoneid == exi->exi_zoneid) {
2639 			/*
2640 			 * NOTE: If we want to drop lm_lock before
2641 			 * calling nlm_zone_unexport(), we should break,
2642 			 * and have a post-rw_exit() snippit like:
2643 			 *	if (g != NULL)
2644 			 *		nlm_zone_unexport(g, exi);
2645 			 */
2646 			nlm_zone_unexport(g, exi);
2647 			break; /* Only going to match once! */
2648 		}
2649 	}
2650 	rw_exit(&lm_lck);
2651 }
2652 
2653 /*
2654  * Allocate new unique sysid.
2655  * In case of failure (no available sysids)
2656  * return LM_NOSYSID.
2657  */
2658 sysid_t
nlm_sysid_alloc(void)2659 nlm_sysid_alloc(void)
2660 {
2661 	sysid_t ret_sysid = LM_NOSYSID;
2662 
2663 	rw_enter(&lm_lck, RW_WRITER);
2664 	if (nlm_sysid_nidx > LM_SYSID_MAX)
2665 		nlm_sysid_nidx = LM_SYSID;
2666 
2667 	if (!BT_TEST(nlm_sysid_bmap, nlm_sysid_nidx)) {
2668 		BT_SET(nlm_sysid_bmap, nlm_sysid_nidx);
2669 		ret_sysid = nlm_sysid_nidx++;
2670 	} else {
2671 		index_t id;
2672 
2673 		id = bt_availbit(nlm_sysid_bmap, NLM_BMAP_NITEMS);
2674 		if (id > 0) {
2675 			nlm_sysid_nidx = id + 1;
2676 			ret_sysid = id;
2677 			BT_SET(nlm_sysid_bmap, id);
2678 		}
2679 	}
2680 
2681 	rw_exit(&lm_lck);
2682 	return (ret_sysid);
2683 }
2684 
2685 void
nlm_sysid_free(sysid_t sysid)2686 nlm_sysid_free(sysid_t sysid)
2687 {
2688 	ASSERT(sysid >= LM_SYSID && sysid <= LM_SYSID_MAX);
2689 
2690 	rw_enter(&lm_lck, RW_WRITER);
2691 	ASSERT(BT_TEST(nlm_sysid_bmap, sysid));
2692 	BT_CLEAR(nlm_sysid_bmap, sysid);
2693 	rw_exit(&lm_lck);
2694 }
2695 
2696 /*
2697  * Return true if the request came from a local caller.
2698  * By necessity, this "knows" the netid names invented
2699  * in lm_svc() and nlm_netid_from_knetconfig().
2700  */
2701 bool_t
nlm_caller_is_local(SVCXPRT * transp)2702 nlm_caller_is_local(SVCXPRT *transp)
2703 {
2704 	char *netid;
2705 	struct netbuf *rtaddr;
2706 
2707 	netid = svc_getnetid(transp);
2708 	rtaddr = svc_getrpccaller(transp);
2709 
2710 	if (netid == NULL)
2711 		return (FALSE);
2712 
2713 	if (strcmp(netid, "ticlts") == 0 ||
2714 	    strcmp(netid, "ticotsord") == 0)
2715 		return (TRUE);
2716 
2717 	if (strcmp(netid, "tcp") == 0 || strcmp(netid, "udp") == 0) {
2718 		struct sockaddr_in *sin = (void *)rtaddr->buf;
2719 		if (sin->sin_addr.s_addr == htonl(INADDR_LOOPBACK))
2720 			return (TRUE);
2721 	}
2722 	if (strcmp(netid, "tcp6") == 0 || strcmp(netid, "udp6") == 0) {
2723 		struct sockaddr_in6 *sin6 = (void *)rtaddr->buf;
2724 		if (IN6_IS_ADDR_LOOPBACK(&sin6->sin6_addr))
2725 			return (TRUE);
2726 	}
2727 
2728 	return (FALSE); /* unknown transport */
2729 }
2730 
2731 /*
2732  * Get netid string correspondig to the given knetconfig.
2733  * If not done already, save knc->knc_rdev in our table.
2734  */
2735 const char *
nlm_knc_to_netid(struct knetconfig * knc)2736 nlm_knc_to_netid(struct knetconfig *knc)
2737 {
2738 	int i;
2739 	dev_t rdev;
2740 	struct nlm_knc *nc;
2741 	const char *netid = NULL;
2742 
2743 	rw_enter(&lm_lck, RW_READER);
2744 	for (i = 0; i < NLM_KNCS; i++) {
2745 		nc = &nlm_netconfigs[i];
2746 
2747 		if (nc->n_knc.knc_semantics == knc->knc_semantics &&
2748 		    strcmp(nc->n_knc.knc_protofmly,
2749 		    knc->knc_protofmly) == 0) {
2750 			netid = nc->n_netid;
2751 			rdev = nc->n_knc.knc_rdev;
2752 			break;
2753 		}
2754 	}
2755 	rw_exit(&lm_lck);
2756 
2757 	if (netid != NULL && rdev == NODEV) {
2758 		rw_enter(&lm_lck, RW_WRITER);
2759 		if (nc->n_knc.knc_rdev == NODEV)
2760 			nc->n_knc.knc_rdev = knc->knc_rdev;
2761 		rw_exit(&lm_lck);
2762 	}
2763 
2764 	return (netid);
2765 }
2766 
2767 /*
2768  * Get a knetconfig corresponding to the given netid.
2769  * If there's no knetconfig for this netid, ENOENT
2770  * is returned.
2771  */
2772 int
nlm_knc_from_netid(const char * netid,struct knetconfig * knc)2773 nlm_knc_from_netid(const char *netid, struct knetconfig *knc)
2774 {
2775 	int i, ret;
2776 
2777 	ret = ENOENT;
2778 	for (i = 0; i < NLM_KNCS; i++) {
2779 		struct nlm_knc *nknc;
2780 
2781 		nknc = &nlm_netconfigs[i];
2782 		if (strcmp(netid, nknc->n_netid) == 0 &&
2783 		    nknc->n_knc.knc_rdev != NODEV) {
2784 			*knc = nknc->n_knc;
2785 			ret = 0;
2786 			break;
2787 		}
2788 	}
2789 
2790 	return (ret);
2791 }
2792 
2793 void
nlm_cprsuspend(void)2794 nlm_cprsuspend(void)
2795 {
2796 	struct nlm_globals *g;
2797 
2798 	rw_enter(&lm_lck, RW_READER);
2799 	TAILQ_FOREACH(g, &nlm_zones_list, nlm_link)
2800 		nlm_suspend_zone(g);
2801 
2802 	rw_exit(&lm_lck);
2803 }
2804 
2805 void
nlm_cprresume(void)2806 nlm_cprresume(void)
2807 {
2808 	struct nlm_globals *g;
2809 
2810 	rw_enter(&lm_lck, RW_READER);
2811 	TAILQ_FOREACH(g, &nlm_zones_list, nlm_link)
2812 		nlm_resume_zone(g);
2813 
2814 	rw_exit(&lm_lck);
2815 }
2816 
2817 static void
nlm_nsm_clnt_init(CLIENT * clnt,struct nlm_nsm * nsm)2818 nlm_nsm_clnt_init(CLIENT *clnt, struct nlm_nsm *nsm)
2819 {
2820 	(void) clnt_tli_kinit(clnt, &nsm->ns_knc, &nsm->ns_addr, 0,
2821 	    NLM_RPC_RETRIES, zone_kcred());
2822 }
2823 
2824 static void
nlm_netbuf_to_netobj(struct netbuf * addr,int * family,netobj * obj)2825 nlm_netbuf_to_netobj(struct netbuf *addr, int *family, netobj *obj)
2826 {
2827 	/* LINTED pointer alignment */
2828 	struct sockaddr *sa = (struct sockaddr *)addr->buf;
2829 
2830 	*family = sa->sa_family;
2831 
2832 	switch (sa->sa_family) {
2833 	case AF_INET: {
2834 		/* LINTED pointer alignment */
2835 		struct sockaddr_in *sin = (struct sockaddr_in *)sa;
2836 
2837 		obj->n_len = sizeof (sin->sin_addr);
2838 		obj->n_bytes = (char *)&sin->sin_addr;
2839 		break;
2840 	}
2841 
2842 	case AF_INET6: {
2843 		/* LINTED pointer alignment */
2844 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
2845 
2846 		obj->n_len = sizeof (sin6->sin6_addr);
2847 		obj->n_bytes = (char *)&sin6->sin6_addr;
2848 		break;
2849 	}
2850 
2851 	default:
2852 		VERIFY(0);
2853 		break;
2854 	}
2855 }
2856