xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs_server.c (revision bc1f688b4872ace323eaddbb1a6365d054e7bf56)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011 Bayard G. Bell. All rights reserved.
24  * Copyright (c) 2013 by Delphix. All rights reserved.
25  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
26  * Copyright (c) 2017 Joyent Inc
27  */
28 
29 /*
30  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
31  *	All rights reserved.
32  *	Use is subject to license terms.
33  */
34 
35 #include <sys/param.h>
36 #include <sys/types.h>
37 #include <sys/systm.h>
38 #include <sys/cred.h>
39 #include <sys/proc.h>
40 #include <sys/user.h>
41 #include <sys/buf.h>
42 #include <sys/vfs.h>
43 #include <sys/vnode.h>
44 #include <sys/pathname.h>
45 #include <sys/uio.h>
46 #include <sys/file.h>
47 #include <sys/stat.h>
48 #include <sys/errno.h>
49 #include <sys/socket.h>
50 #include <sys/sysmacros.h>
51 #include <sys/siginfo.h>
52 #include <sys/tiuser.h>
53 #include <sys/statvfs.h>
54 #include <sys/stream.h>
55 #include <sys/strsun.h>
56 #include <sys/strsubr.h>
57 #include <sys/stropts.h>
58 #include <sys/timod.h>
59 #include <sys/t_kuser.h>
60 #include <sys/kmem.h>
61 #include <sys/kstat.h>
62 #include <sys/dirent.h>
63 #include <sys/cmn_err.h>
64 #include <sys/debug.h>
65 #include <sys/unistd.h>
66 #include <sys/vtrace.h>
67 #include <sys/mode.h>
68 #include <sys/acl.h>
69 #include <sys/sdt.h>
70 #include <sys/debug.h>
71 
72 #include <rpc/types.h>
73 #include <rpc/auth.h>
74 #include <rpc/auth_unix.h>
75 #include <rpc/auth_des.h>
76 #include <rpc/svc.h>
77 #include <rpc/xdr.h>
78 #include <rpc/rpc_rdma.h>
79 
80 #include <nfs/nfs.h>
81 #include <nfs/export.h>
82 #include <nfs/nfssys.h>
83 #include <nfs/nfs_clnt.h>
84 #include <nfs/nfs_acl.h>
85 #include <nfs/nfs_log.h>
86 #include <nfs/nfs_cmd.h>
87 #include <nfs/lm.h>
88 #include <nfs/nfs_dispatch.h>
89 #include <nfs/nfs4_drc.h>
90 
91 #include <sys/modctl.h>
92 #include <sys/cladm.h>
93 #include <sys/clconf.h>
94 
95 #include <sys/tsol/label.h>
96 
97 #define	MAXHOST 32
98 const char *kinet_ntop6(uchar_t *, char *, size_t);
99 
100 /*
101  * Module linkage information.
102  */
103 
104 static struct modlmisc modlmisc = {
105 	&mod_miscops, "NFS server module"
106 };
107 
108 static struct modlinkage modlinkage = {
109 	MODREV_1, (void *)&modlmisc, NULL
110 };
111 
112 kmem_cache_t *nfs_xuio_cache;
113 int nfs_loaned_buffers = 0;
114 
115 int
116 _init(void)
117 {
118 	int status;
119 
120 	if ((status = nfs_srvinit()) != 0) {
121 		cmn_err(CE_WARN, "_init: nfs_srvinit failed");
122 		return (status);
123 	}
124 
125 	status = mod_install((struct modlinkage *)&modlinkage);
126 	if (status != 0) {
127 		/*
128 		 * Could not load module, cleanup previous
129 		 * initialization work.
130 		 */
131 		nfs_srvfini();
132 
133 		return (status);
134 	}
135 
136 	/*
137 	 * Initialise some placeholders for nfssys() calls. These have
138 	 * to be declared by the nfs module, since that handles nfssys()
139 	 * calls - also used by NFS clients - but are provided by this
140 	 * nfssrv module. These also then serve as confirmation to the
141 	 * relevant code in nfs that nfssrv has been loaded, as they're
142 	 * initially NULL.
143 	 */
144 	nfs_srv_quiesce_func = nfs_srv_quiesce_all;
145 	nfs_srv_dss_func = rfs4_dss_setpaths;
146 
147 	/* setup DSS paths here; must be done before initial server startup */
148 	rfs4_dss_paths = rfs4_dss_oldpaths = NULL;
149 
150 	/* initialize the copy reduction caches */
151 
152 	nfs_xuio_cache = kmem_cache_create("nfs_xuio_cache",
153 	    sizeof (nfs_xuio_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
154 
155 	return (status);
156 }
157 
158 int
159 _fini()
160 {
161 	return (EBUSY);
162 }
163 
164 int
165 _info(struct modinfo *modinfop)
166 {
167 	return (mod_info(&modlinkage, modinfop));
168 }
169 
170 /*
171  * PUBLICFH_CHECK() checks if the dispatch routine supports
172  * RPC_PUBLICFH_OK, if the filesystem is exported public, and if the
173  * incoming request is using the public filehandle. The check duplicates
174  * the exportmatch() call done in checkexport(), and we should consider
175  * modifying those routines to avoid the duplication. For now, we optimize
176  * by calling exportmatch() only after checking that the dispatch routine
177  * supports RPC_PUBLICFH_OK, and if the filesystem is explicitly exported
178  * public (i.e., not the placeholder).
179  */
180 #define	PUBLICFH_CHECK(disp, exi, fsid, xfid) \
181 		((disp->dis_flags & RPC_PUBLICFH_OK) && \
182 		((exi->exi_export.ex_flags & EX_PUBLIC) || \
183 		(exi == exi_public && exportmatch(exi_root, \
184 		fsid, xfid))))
185 
186 static void	nfs_srv_shutdown_all(int);
187 static void	rfs4_server_start(int);
188 static void	nullfree(void);
189 static void	rfs_dispatch(struct svc_req *, SVCXPRT *);
190 static void	acl_dispatch(struct svc_req *, SVCXPRT *);
191 static void	common_dispatch(struct svc_req *, SVCXPRT *,
192 		rpcvers_t, rpcvers_t, char *,
193 		struct rpc_disptable *);
194 static void	hanfsv4_failover(void);
195 static	int	checkauth(struct exportinfo *, struct svc_req *, cred_t *, int,
196 		bool_t, bool_t *);
197 static char	*client_name(struct svc_req *req);
198 static char	*client_addr(struct svc_req *req, char *buf);
199 extern	int	sec_svc_getcred(struct svc_req *, cred_t *cr, char **, int *);
200 extern	bool_t	sec_svc_inrootlist(int, caddr_t, int, caddr_t *);
201 
202 #define	NFSLOG_COPY_NETBUF(exi, xprt, nb)	{		\
203 	(nb)->maxlen = (xprt)->xp_rtaddr.maxlen;		\
204 	(nb)->len = (xprt)->xp_rtaddr.len;			\
205 	(nb)->buf = kmem_alloc((nb)->len, KM_SLEEP);		\
206 	bcopy((xprt)->xp_rtaddr.buf, (nb)->buf, (nb)->len);	\
207 	}
208 
209 /*
210  * Public Filehandle common nfs routines
211  */
212 static int	MCLpath(char **);
213 static void	URLparse(char *);
214 
215 /*
216  * NFS callout table.
217  * This table is used by svc_getreq() to dispatch a request with
218  * a given prog/vers pair to an appropriate service provider
219  * dispatch routine.
220  *
221  * NOTE: ordering is relied upon below when resetting the version min/max
222  * for NFS_PROGRAM.  Careful, if this is ever changed.
223  */
224 static SVC_CALLOUT __nfs_sc_clts[] = {
225 	{ NFS_PROGRAM,	   NFS_VERSMIN,	    NFS_VERSMAX,	rfs_dispatch },
226 	{ NFS_ACL_PROGRAM, NFS_ACL_VERSMIN, NFS_ACL_VERSMAX,	acl_dispatch }
227 };
228 
229 static SVC_CALLOUT_TABLE nfs_sct_clts = {
230 	sizeof (__nfs_sc_clts) / sizeof (__nfs_sc_clts[0]), FALSE,
231 	__nfs_sc_clts
232 };
233 
234 static SVC_CALLOUT __nfs_sc_cots[] = {
235 	{ NFS_PROGRAM,	   NFS_VERSMIN,	    NFS_VERSMAX,	rfs_dispatch },
236 	{ NFS_ACL_PROGRAM, NFS_ACL_VERSMIN, NFS_ACL_VERSMAX,	acl_dispatch }
237 };
238 
239 static SVC_CALLOUT_TABLE nfs_sct_cots = {
240 	sizeof (__nfs_sc_cots) / sizeof (__nfs_sc_cots[0]), FALSE, __nfs_sc_cots
241 };
242 
243 static SVC_CALLOUT __nfs_sc_rdma[] = {
244 	{ NFS_PROGRAM,	   NFS_VERSMIN,	    NFS_VERSMAX,	rfs_dispatch },
245 	{ NFS_ACL_PROGRAM, NFS_ACL_VERSMIN, NFS_ACL_VERSMAX,	acl_dispatch }
246 };
247 
248 static SVC_CALLOUT_TABLE nfs_sct_rdma = {
249 	sizeof (__nfs_sc_rdma) / sizeof (__nfs_sc_rdma[0]), FALSE, __nfs_sc_rdma
250 };
251 rpcvers_t nfs_versmin = NFS_VERSMIN_DEFAULT;
252 rpcvers_t nfs_versmax = NFS_VERSMAX_DEFAULT;
253 
254 /*
255  * Used to track the state of the server so that initialization
256  * can be done properly.
257  */
258 typedef enum {
259 	NFS_SERVER_STOPPED,	/* server state destroyed */
260 	NFS_SERVER_STOPPING,	/* server state being destroyed */
261 	NFS_SERVER_RUNNING,
262 	NFS_SERVER_QUIESCED,	/* server state preserved */
263 	NFS_SERVER_OFFLINE	/* server pool offline */
264 } nfs_server_running_t;
265 
266 static nfs_server_running_t nfs_server_upordown;
267 static kmutex_t nfs_server_upordown_lock;
268 static	kcondvar_t nfs_server_upordown_cv;
269 
270 /*
271  * DSS: distributed stable storage
272  * lists of all DSS paths: current, and before last warmstart
273  */
274 nvlist_t *rfs4_dss_paths, *rfs4_dss_oldpaths;
275 
276 int rfs4_dispatch(struct rpcdisp *, struct svc_req *, SVCXPRT *, char *);
277 bool_t rfs4_minorvers_mismatch(struct svc_req *, SVCXPRT *, void *);
278 
279 /*
280  * RDMA wait variables.
281  */
282 static kcondvar_t rdma_wait_cv;
283 static kmutex_t rdma_wait_mutex;
284 
285 /*
286  * Will be called at the point the server pool is being unregistered
287  * from the pool list. From that point onwards, the pool is waiting
288  * to be drained and as such the server state is stale and pertains
289  * to the old instantiation of the NFS server pool.
290  */
291 void
292 nfs_srv_offline(void)
293 {
294 	mutex_enter(&nfs_server_upordown_lock);
295 	if (nfs_server_upordown == NFS_SERVER_RUNNING) {
296 		nfs_server_upordown = NFS_SERVER_OFFLINE;
297 	}
298 	mutex_exit(&nfs_server_upordown_lock);
299 }
300 
301 /*
302  * Will be called at the point the server pool is being destroyed so
303  * all transports have been closed and no service threads are in
304  * existence.
305  *
306  * If we quiesce the server, we're shutting it down without destroying the
307  * server state. This allows it to warm start subsequently.
308  */
309 void
310 nfs_srv_stop_all(void)
311 {
312 	int quiesce = 0;
313 	nfs_srv_shutdown_all(quiesce);
314 }
315 
316 /*
317  * This alternative shutdown routine can be requested via nfssys()
318  */
319 void
320 nfs_srv_quiesce_all(void)
321 {
322 	int quiesce = 1;
323 	nfs_srv_shutdown_all(quiesce);
324 }
325 
326 static void
327 nfs_srv_shutdown_all(int quiesce) {
328 	mutex_enter(&nfs_server_upordown_lock);
329 	if (quiesce) {
330 		if (nfs_server_upordown == NFS_SERVER_RUNNING ||
331 			nfs_server_upordown == NFS_SERVER_OFFLINE) {
332 			nfs_server_upordown = NFS_SERVER_QUIESCED;
333 			cv_signal(&nfs_server_upordown_cv);
334 
335 			/* reset DSS state, for subsequent warm restart */
336 			rfs4_dss_numnewpaths = 0;
337 			rfs4_dss_newpaths = NULL;
338 
339 			cmn_err(CE_NOTE, "nfs_server: server is now quiesced; "
340 			    "NFSv4 state has been preserved");
341 		}
342 	} else {
343 		if (nfs_server_upordown == NFS_SERVER_OFFLINE) {
344 			nfs_server_upordown = NFS_SERVER_STOPPING;
345 			mutex_exit(&nfs_server_upordown_lock);
346 			rfs4_state_fini();
347 			rfs4_fini_drc(nfs4_drc);
348 			mutex_enter(&nfs_server_upordown_lock);
349 			nfs_server_upordown = NFS_SERVER_STOPPED;
350 			cv_signal(&nfs_server_upordown_cv);
351 		}
352 	}
353 	mutex_exit(&nfs_server_upordown_lock);
354 }
355 
356 static int
357 nfs_srv_set_sc_versions(struct file *fp, SVC_CALLOUT_TABLE **sctpp,
358 			rpcvers_t versmin, rpcvers_t versmax)
359 {
360 	struct strioctl strioc;
361 	struct T_info_ack tinfo;
362 	int		error, retval;
363 
364 	/*
365 	 * Find out what type of transport this is.
366 	 */
367 	strioc.ic_cmd = TI_GETINFO;
368 	strioc.ic_timout = -1;
369 	strioc.ic_len = sizeof (tinfo);
370 	strioc.ic_dp = (char *)&tinfo;
371 	tinfo.PRIM_type = T_INFO_REQ;
372 
373 	error = strioctl(fp->f_vnode, I_STR, (intptr_t)&strioc, 0, K_TO_K,
374 	    CRED(), &retval);
375 	if (error || retval)
376 		return (error);
377 
378 	/*
379 	 * Based on our query of the transport type...
380 	 *
381 	 * Reset the min/max versions based on the caller's request
382 	 * NOTE: This assumes that NFS_PROGRAM is first in the array!!
383 	 * And the second entry is the NFS_ACL_PROGRAM.
384 	 */
385 	switch (tinfo.SERV_type) {
386 	case T_CLTS:
387 		if (versmax == NFS_V4)
388 			return (EINVAL);
389 		__nfs_sc_clts[0].sc_versmin = versmin;
390 		__nfs_sc_clts[0].sc_versmax = versmax;
391 		__nfs_sc_clts[1].sc_versmin = versmin;
392 		__nfs_sc_clts[1].sc_versmax = versmax;
393 		*sctpp = &nfs_sct_clts;
394 		break;
395 	case T_COTS:
396 	case T_COTS_ORD:
397 		__nfs_sc_cots[0].sc_versmin = versmin;
398 		__nfs_sc_cots[0].sc_versmax = versmax;
399 		/* For the NFS_ACL program, check the max version */
400 		if (versmax > NFS_ACL_VERSMAX)
401 			versmax = NFS_ACL_VERSMAX;
402 		__nfs_sc_cots[1].sc_versmin = versmin;
403 		__nfs_sc_cots[1].sc_versmax = versmax;
404 		*sctpp = &nfs_sct_cots;
405 		break;
406 	default:
407 		error = EINVAL;
408 	}
409 
410 	return (error);
411 }
412 
413 /*
414  * NFS Server system call.
415  * Does all of the work of running a NFS server.
416  * uap->fd is the fd of an open transport provider
417  */
418 int
419 nfs_svc(struct nfs_svc_args *arg, model_t model)
420 {
421 	file_t *fp;
422 	SVCMASTERXPRT *xprt;
423 	int error;
424 	int readsize;
425 	char buf[KNC_STRSIZE];
426 	size_t len;
427 	STRUCT_HANDLE(nfs_svc_args, uap);
428 	struct netbuf addrmask;
429 	SVC_CALLOUT_TABLE *sctp = NULL;
430 
431 #ifdef lint
432 	model = model;		/* STRUCT macros don't always refer to it */
433 #endif
434 
435 	STRUCT_SET_HANDLE(uap, model, arg);
436 
437 	/* Check privileges in nfssys() */
438 
439 	if ((fp = getf(STRUCT_FGET(uap, fd))) == NULL)
440 		return (EBADF);
441 
442 	/*
443 	 * Set read buffer size to rsize
444 	 * and add room for RPC headers.
445 	 */
446 	readsize = nfs3tsize() + (RPC_MAXDATASIZE - NFS_MAXDATA);
447 	if (readsize < RPC_MAXDATASIZE)
448 		readsize = RPC_MAXDATASIZE;
449 
450 	error = copyinstr((const char *)STRUCT_FGETP(uap, netid), buf,
451 	    KNC_STRSIZE, &len);
452 	if (error) {
453 		releasef(STRUCT_FGET(uap, fd));
454 		return (error);
455 	}
456 
457 	addrmask.len = STRUCT_FGET(uap, addrmask.len);
458 	addrmask.maxlen = STRUCT_FGET(uap, addrmask.maxlen);
459 	addrmask.buf = kmem_alloc(addrmask.maxlen, KM_SLEEP);
460 	error = copyin(STRUCT_FGETP(uap, addrmask.buf), addrmask.buf,
461 	    addrmask.len);
462 	if (error) {
463 		releasef(STRUCT_FGET(uap, fd));
464 		kmem_free(addrmask.buf, addrmask.maxlen);
465 		return (error);
466 	}
467 
468 	nfs_versmin = STRUCT_FGET(uap, versmin);
469 	nfs_versmax = STRUCT_FGET(uap, versmax);
470 
471 	/* Double check the vers min/max ranges */
472 	if ((nfs_versmin > nfs_versmax) ||
473 	    (nfs_versmin < NFS_VERSMIN) ||
474 	    (nfs_versmax > NFS_VERSMAX)) {
475 		nfs_versmin = NFS_VERSMIN_DEFAULT;
476 		nfs_versmax = NFS_VERSMAX_DEFAULT;
477 	}
478 
479 	if (error =
480 	    nfs_srv_set_sc_versions(fp, &sctp, nfs_versmin, nfs_versmax)) {
481 		releasef(STRUCT_FGET(uap, fd));
482 		kmem_free(addrmask.buf, addrmask.maxlen);
483 		return (error);
484 	}
485 
486 	/* Initialize nfsv4 server */
487 	if (nfs_versmax == (rpcvers_t)NFS_V4)
488 		rfs4_server_start(STRUCT_FGET(uap, delegation));
489 
490 	/* Create a transport handle. */
491 	error = svc_tli_kcreate(fp, readsize, buf, &addrmask, &xprt,
492 	    sctp, NULL, NFS_SVCPOOL_ID, TRUE);
493 
494 	if (error)
495 		kmem_free(addrmask.buf, addrmask.maxlen);
496 
497 	releasef(STRUCT_FGET(uap, fd));
498 
499 	/* HA-NFSv4: save the cluster nodeid */
500 	if (cluster_bootflags & CLUSTER_BOOTED)
501 		lm_global_nlmid = clconf_get_nodeid();
502 
503 	return (error);
504 }
505 
506 static void
507 rfs4_server_start(int nfs4_srv_delegation)
508 {
509 	/*
510 	 * Determine if the server has previously been "started" and
511 	 * if not, do the per instance initialization
512 	 */
513 	mutex_enter(&nfs_server_upordown_lock);
514 
515 	if (nfs_server_upordown != NFS_SERVER_RUNNING) {
516 		/* Do we need to stop and wait on the previous server? */
517 		while (nfs_server_upordown == NFS_SERVER_STOPPING ||
518 		    nfs_server_upordown == NFS_SERVER_OFFLINE)
519 			cv_wait(&nfs_server_upordown_cv,
520 			    &nfs_server_upordown_lock);
521 
522 		if (nfs_server_upordown != NFS_SERVER_RUNNING) {
523 			(void) svc_pool_control(NFS_SVCPOOL_ID,
524 			    SVCPSET_UNREGISTER_PROC, (void *)&nfs_srv_offline);
525 			(void) svc_pool_control(NFS_SVCPOOL_ID,
526 			    SVCPSET_SHUTDOWN_PROC, (void *)&nfs_srv_stop_all);
527 
528 			/* is this an nfsd warm start? */
529 			if (nfs_server_upordown == NFS_SERVER_QUIESCED) {
530 				cmn_err(CE_NOTE, "nfs_server: "
531 				    "server was previously quiesced; "
532 				    "existing NFSv4 state will be re-used");
533 
534 				/*
535 				 * HA-NFSv4: this is also the signal
536 				 * that a Resource Group failover has
537 				 * occurred.
538 				 */
539 				if (cluster_bootflags & CLUSTER_BOOTED)
540 					hanfsv4_failover();
541 			} else {
542 				/* cold start */
543 				rfs4_state_init();
544 				nfs4_drc = rfs4_init_drc(nfs4_drc_max,
545 				    nfs4_drc_hash);
546 			}
547 
548 			/*
549 			 * Check to see if delegation is to be
550 			 * enabled at the server
551 			 */
552 			if (nfs4_srv_delegation != FALSE)
553 				rfs4_set_deleg_policy(SRV_NORMAL_DELEGATE);
554 
555 			nfs_server_upordown = NFS_SERVER_RUNNING;
556 		}
557 		cv_signal(&nfs_server_upordown_cv);
558 	}
559 	mutex_exit(&nfs_server_upordown_lock);
560 }
561 
562 /*
563  * If RDMA device available,
564  * start RDMA listener.
565  */
566 int
567 rdma_start(struct rdma_svc_args *rsa)
568 {
569 	int error;
570 	rdma_xprt_group_t started_rdma_xprts;
571 	rdma_stat stat;
572 	int svc_state = 0;
573 
574 	/* Double check the vers min/max ranges */
575 	if ((rsa->nfs_versmin > rsa->nfs_versmax) ||
576 	    (rsa->nfs_versmin < NFS_VERSMIN) ||
577 	    (rsa->nfs_versmax > NFS_VERSMAX)) {
578 		rsa->nfs_versmin = NFS_VERSMIN_DEFAULT;
579 		rsa->nfs_versmax = NFS_VERSMAX_DEFAULT;
580 	}
581 	nfs_versmin = rsa->nfs_versmin;
582 	nfs_versmax = rsa->nfs_versmax;
583 
584 	/* Set the versions in the callout table */
585 	__nfs_sc_rdma[0].sc_versmin = rsa->nfs_versmin;
586 	__nfs_sc_rdma[0].sc_versmax = rsa->nfs_versmax;
587 	/* For the NFS_ACL program, check the max version */
588 	__nfs_sc_rdma[1].sc_versmin = rsa->nfs_versmin;
589 	if (rsa->nfs_versmax > NFS_ACL_VERSMAX)
590 		__nfs_sc_rdma[1].sc_versmax = NFS_ACL_VERSMAX;
591 	else
592 		__nfs_sc_rdma[1].sc_versmax = rsa->nfs_versmax;
593 
594 	/* Initialize nfsv4 server */
595 	if (rsa->nfs_versmax == (rpcvers_t)NFS_V4)
596 		rfs4_server_start(rsa->delegation);
597 
598 	started_rdma_xprts.rtg_count = 0;
599 	started_rdma_xprts.rtg_listhead = NULL;
600 	started_rdma_xprts.rtg_poolid = rsa->poolid;
601 
602 restart:
603 	error = svc_rdma_kcreate(rsa->netid, &nfs_sct_rdma, rsa->poolid,
604 	    &started_rdma_xprts);
605 
606 	svc_state = !error;
607 
608 	while (!error) {
609 
610 		/*
611 		 * wait till either interrupted by a signal on
612 		 * nfs service stop/restart or signalled by a
613 		 * rdma plugin attach/detatch.
614 		 */
615 
616 		stat = rdma_kwait();
617 
618 		/*
619 		 * stop services if running -- either on a HCA detach event
620 		 * or if the nfs service is stopped/restarted.
621 		 */
622 
623 		if ((stat == RDMA_HCA_DETACH || stat == RDMA_INTR) &&
624 		    svc_state) {
625 			rdma_stop(&started_rdma_xprts);
626 			svc_state = 0;
627 		}
628 
629 		/*
630 		 * nfs service stop/restart, break out of the
631 		 * wait loop and return;
632 		 */
633 		if (stat == RDMA_INTR)
634 			return (0);
635 
636 		/*
637 		 * restart stopped services on a HCA attach event
638 		 * (if not already running)
639 		 */
640 
641 		if ((stat == RDMA_HCA_ATTACH) && (svc_state == 0))
642 			goto restart;
643 
644 		/*
645 		 * loop until a nfs service stop/restart
646 		 */
647 	}
648 
649 	return (error);
650 }
651 
652 /* ARGSUSED */
653 void
654 rpc_null(caddr_t *argp, caddr_t *resp, struct exportinfo *exi,
655     struct svc_req *req, cred_t *cr, bool_t ro)
656 {
657 }
658 
659 /* ARGSUSED */
660 void
661 rpc_null_v3(caddr_t *argp, caddr_t *resp, struct exportinfo *exi,
662     struct svc_req *req, cred_t *cr, bool_t ro)
663 {
664 	DTRACE_NFSV3_3(op__null__start, struct svc_req *, req,
665 	    cred_t *, cr, vnode_t *, NULL);
666 	DTRACE_NFSV3_3(op__null__done, struct svc_req *, req,
667 	    cred_t *, cr, vnode_t *, NULL);
668 }
669 
670 /* ARGSUSED */
671 static void
672 rfs_error(caddr_t *argp, caddr_t *resp, struct exportinfo *exi,
673     struct svc_req *req, cred_t *cr, bool_t ro)
674 {
675 	/* return (EOPNOTSUPP); */
676 }
677 
678 static void
679 nullfree(void)
680 {
681 }
682 
683 static char *rfscallnames_v2[] = {
684 	"RFS2_NULL",
685 	"RFS2_GETATTR",
686 	"RFS2_SETATTR",
687 	"RFS2_ROOT",
688 	"RFS2_LOOKUP",
689 	"RFS2_READLINK",
690 	"RFS2_READ",
691 	"RFS2_WRITECACHE",
692 	"RFS2_WRITE",
693 	"RFS2_CREATE",
694 	"RFS2_REMOVE",
695 	"RFS2_RENAME",
696 	"RFS2_LINK",
697 	"RFS2_SYMLINK",
698 	"RFS2_MKDIR",
699 	"RFS2_RMDIR",
700 	"RFS2_READDIR",
701 	"RFS2_STATFS"
702 };
703 
704 static struct rpcdisp rfsdisptab_v2[] = {
705 	/*
706 	 * NFS VERSION 2
707 	 */
708 
709 	/* RFS_NULL = 0 */
710 	{rpc_null,
711 	    xdr_void, NULL_xdrproc_t, 0,
712 	    xdr_void, NULL_xdrproc_t, 0,
713 	    nullfree, RPC_IDEMPOTENT,
714 	    0},
715 
716 	/* RFS_GETATTR = 1 */
717 	{rfs_getattr,
718 	    xdr_fhandle, xdr_fastfhandle, sizeof (fhandle_t),
719 	    xdr_attrstat, xdr_fastattrstat, sizeof (struct nfsattrstat),
720 	    nullfree, RPC_IDEMPOTENT|RPC_ALLOWANON|RPC_MAPRESP,
721 	    rfs_getattr_getfh},
722 
723 	/* RFS_SETATTR = 2 */
724 	{rfs_setattr,
725 	    xdr_saargs, NULL_xdrproc_t, sizeof (struct nfssaargs),
726 	    xdr_attrstat, xdr_fastattrstat, sizeof (struct nfsattrstat),
727 	    nullfree, RPC_MAPRESP,
728 	    rfs_setattr_getfh},
729 
730 	/* RFS_ROOT = 3 *** NO LONGER SUPPORTED *** */
731 	{rfs_error,
732 	    xdr_void, NULL_xdrproc_t, 0,
733 	    xdr_void, NULL_xdrproc_t, 0,
734 	    nullfree, RPC_IDEMPOTENT,
735 	    0},
736 
737 	/* RFS_LOOKUP = 4 */
738 	{rfs_lookup,
739 	    xdr_diropargs, NULL_xdrproc_t, sizeof (struct nfsdiropargs),
740 	    xdr_diropres, xdr_fastdiropres, sizeof (struct nfsdiropres),
741 	    nullfree, RPC_IDEMPOTENT|RPC_MAPRESP|RPC_PUBLICFH_OK,
742 	    rfs_lookup_getfh},
743 
744 	/* RFS_READLINK = 5 */
745 	{rfs_readlink,
746 	    xdr_fhandle, xdr_fastfhandle, sizeof (fhandle_t),
747 	    xdr_rdlnres, NULL_xdrproc_t, sizeof (struct nfsrdlnres),
748 	    rfs_rlfree, RPC_IDEMPOTENT,
749 	    rfs_readlink_getfh},
750 
751 	/* RFS_READ = 6 */
752 	{rfs_read,
753 	    xdr_readargs, NULL_xdrproc_t, sizeof (struct nfsreadargs),
754 	    xdr_rdresult, NULL_xdrproc_t, sizeof (struct nfsrdresult),
755 	    rfs_rdfree, RPC_IDEMPOTENT,
756 	    rfs_read_getfh},
757 
758 	/* RFS_WRITECACHE = 7 *** NO LONGER SUPPORTED *** */
759 	{rfs_error,
760 	    xdr_void, NULL_xdrproc_t, 0,
761 	    xdr_void, NULL_xdrproc_t, 0,
762 	    nullfree, RPC_IDEMPOTENT,
763 	    0},
764 
765 	/* RFS_WRITE = 8 */
766 	{rfs_write,
767 	    xdr_writeargs, NULL_xdrproc_t, sizeof (struct nfswriteargs),
768 	    xdr_attrstat, xdr_fastattrstat, sizeof (struct nfsattrstat),
769 	    nullfree, RPC_MAPRESP,
770 	    rfs_write_getfh},
771 
772 	/* RFS_CREATE = 9 */
773 	{rfs_create,
774 	    xdr_creatargs, NULL_xdrproc_t, sizeof (struct nfscreatargs),
775 	    xdr_diropres, xdr_fastdiropres, sizeof (struct nfsdiropres),
776 	    nullfree, RPC_MAPRESP,
777 	    rfs_create_getfh},
778 
779 	/* RFS_REMOVE = 10 */
780 	{rfs_remove,
781 	    xdr_diropargs, NULL_xdrproc_t, sizeof (struct nfsdiropargs),
782 #ifdef _LITTLE_ENDIAN
783 	    xdr_enum, xdr_fastenum, sizeof (enum nfsstat),
784 #else
785 	    xdr_enum, NULL_xdrproc_t, sizeof (enum nfsstat),
786 #endif
787 	    nullfree, RPC_MAPRESP,
788 	    rfs_remove_getfh},
789 
790 	/* RFS_RENAME = 11 */
791 	{rfs_rename,
792 	    xdr_rnmargs, NULL_xdrproc_t, sizeof (struct nfsrnmargs),
793 #ifdef _LITTLE_ENDIAN
794 	    xdr_enum, xdr_fastenum, sizeof (enum nfsstat),
795 #else
796 	    xdr_enum, NULL_xdrproc_t, sizeof (enum nfsstat),
797 #endif
798 	    nullfree, RPC_MAPRESP,
799 	    rfs_rename_getfh},
800 
801 	/* RFS_LINK = 12 */
802 	{rfs_link,
803 	    xdr_linkargs, NULL_xdrproc_t, sizeof (struct nfslinkargs),
804 #ifdef _LITTLE_ENDIAN
805 	    xdr_enum, xdr_fastenum, sizeof (enum nfsstat),
806 #else
807 	    xdr_enum, NULL_xdrproc_t, sizeof (enum nfsstat),
808 #endif
809 	    nullfree, RPC_MAPRESP,
810 	    rfs_link_getfh},
811 
812 	/* RFS_SYMLINK = 13 */
813 	{rfs_symlink,
814 	    xdr_slargs, NULL_xdrproc_t, sizeof (struct nfsslargs),
815 #ifdef _LITTLE_ENDIAN
816 	    xdr_enum, xdr_fastenum, sizeof (enum nfsstat),
817 #else
818 	    xdr_enum, NULL_xdrproc_t, sizeof (enum nfsstat),
819 #endif
820 	    nullfree, RPC_MAPRESP,
821 	    rfs_symlink_getfh},
822 
823 	/* RFS_MKDIR = 14 */
824 	{rfs_mkdir,
825 	    xdr_creatargs, NULL_xdrproc_t, sizeof (struct nfscreatargs),
826 	    xdr_diropres, xdr_fastdiropres, sizeof (struct nfsdiropres),
827 	    nullfree, RPC_MAPRESP,
828 	    rfs_mkdir_getfh},
829 
830 	/* RFS_RMDIR = 15 */
831 	{rfs_rmdir,
832 	    xdr_diropargs, NULL_xdrproc_t, sizeof (struct nfsdiropargs),
833 #ifdef _LITTLE_ENDIAN
834 	    xdr_enum, xdr_fastenum, sizeof (enum nfsstat),
835 #else
836 	    xdr_enum, NULL_xdrproc_t, sizeof (enum nfsstat),
837 #endif
838 	    nullfree, RPC_MAPRESP,
839 	    rfs_rmdir_getfh},
840 
841 	/* RFS_READDIR = 16 */
842 	{rfs_readdir,
843 	    xdr_rddirargs, NULL_xdrproc_t, sizeof (struct nfsrddirargs),
844 	    xdr_putrddirres, NULL_xdrproc_t, sizeof (struct nfsrddirres),
845 	    rfs_rddirfree, RPC_IDEMPOTENT,
846 	    rfs_readdir_getfh},
847 
848 	/* RFS_STATFS = 17 */
849 	{rfs_statfs,
850 	    xdr_fhandle, xdr_fastfhandle, sizeof (fhandle_t),
851 	    xdr_statfs, xdr_faststatfs, sizeof (struct nfsstatfs),
852 	    nullfree, RPC_IDEMPOTENT|RPC_ALLOWANON|RPC_MAPRESP,
853 	    rfs_statfs_getfh},
854 };
855 
856 static char *rfscallnames_v3[] = {
857 	"RFS3_NULL",
858 	"RFS3_GETATTR",
859 	"RFS3_SETATTR",
860 	"RFS3_LOOKUP",
861 	"RFS3_ACCESS",
862 	"RFS3_READLINK",
863 	"RFS3_READ",
864 	"RFS3_WRITE",
865 	"RFS3_CREATE",
866 	"RFS3_MKDIR",
867 	"RFS3_SYMLINK",
868 	"RFS3_MKNOD",
869 	"RFS3_REMOVE",
870 	"RFS3_RMDIR",
871 	"RFS3_RENAME",
872 	"RFS3_LINK",
873 	"RFS3_READDIR",
874 	"RFS3_READDIRPLUS",
875 	"RFS3_FSSTAT",
876 	"RFS3_FSINFO",
877 	"RFS3_PATHCONF",
878 	"RFS3_COMMIT"
879 };
880 
881 static struct rpcdisp rfsdisptab_v3[] = {
882 	/*
883 	 * NFS VERSION 3
884 	 */
885 
886 	/* RFS_NULL = 0 */
887 	{rpc_null_v3,
888 	    xdr_void, NULL_xdrproc_t, 0,
889 	    xdr_void, NULL_xdrproc_t, 0,
890 	    nullfree, RPC_IDEMPOTENT,
891 	    0},
892 
893 	/* RFS3_GETATTR = 1 */
894 	{rfs3_getattr,
895 	    xdr_nfs_fh3_server, NULL_xdrproc_t, sizeof (GETATTR3args),
896 	    xdr_GETATTR3res, NULL_xdrproc_t, sizeof (GETATTR3res),
897 	    nullfree, (RPC_IDEMPOTENT | RPC_ALLOWANON),
898 	    rfs3_getattr_getfh},
899 
900 	/* RFS3_SETATTR = 2 */
901 	{rfs3_setattr,
902 	    xdr_SETATTR3args, NULL_xdrproc_t, sizeof (SETATTR3args),
903 	    xdr_SETATTR3res, NULL_xdrproc_t, sizeof (SETATTR3res),
904 	    nullfree, 0,
905 	    rfs3_setattr_getfh},
906 
907 	/* RFS3_LOOKUP = 3 */
908 	{rfs3_lookup,
909 	    xdr_diropargs3, NULL_xdrproc_t, sizeof (LOOKUP3args),
910 	    xdr_LOOKUP3res, NULL_xdrproc_t, sizeof (LOOKUP3res),
911 	    nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK),
912 	    rfs3_lookup_getfh},
913 
914 	/* RFS3_ACCESS = 4 */
915 	{rfs3_access,
916 	    xdr_ACCESS3args, NULL_xdrproc_t, sizeof (ACCESS3args),
917 	    xdr_ACCESS3res, NULL_xdrproc_t, sizeof (ACCESS3res),
918 	    nullfree, RPC_IDEMPOTENT,
919 	    rfs3_access_getfh},
920 
921 	/* RFS3_READLINK = 5 */
922 	{rfs3_readlink,
923 	    xdr_nfs_fh3_server, NULL_xdrproc_t, sizeof (READLINK3args),
924 	    xdr_READLINK3res, NULL_xdrproc_t, sizeof (READLINK3res),
925 	    rfs3_readlink_free, RPC_IDEMPOTENT,
926 	    rfs3_readlink_getfh},
927 
928 	/* RFS3_READ = 6 */
929 	{rfs3_read,
930 	    xdr_READ3args, NULL_xdrproc_t, sizeof (READ3args),
931 	    xdr_READ3res, NULL_xdrproc_t, sizeof (READ3res),
932 	    rfs3_read_free, RPC_IDEMPOTENT,
933 	    rfs3_read_getfh},
934 
935 	/* RFS3_WRITE = 7 */
936 	{rfs3_write,
937 	    xdr_WRITE3args, NULL_xdrproc_t, sizeof (WRITE3args),
938 	    xdr_WRITE3res, NULL_xdrproc_t, sizeof (WRITE3res),
939 	    nullfree, 0,
940 	    rfs3_write_getfh},
941 
942 	/* RFS3_CREATE = 8 */
943 	{rfs3_create,
944 	    xdr_CREATE3args, NULL_xdrproc_t, sizeof (CREATE3args),
945 	    xdr_CREATE3res, NULL_xdrproc_t, sizeof (CREATE3res),
946 	    nullfree, 0,
947 	    rfs3_create_getfh},
948 
949 	/* RFS3_MKDIR = 9 */
950 	{rfs3_mkdir,
951 	    xdr_MKDIR3args, NULL_xdrproc_t, sizeof (MKDIR3args),
952 	    xdr_MKDIR3res, NULL_xdrproc_t, sizeof (MKDIR3res),
953 	    nullfree, 0,
954 	    rfs3_mkdir_getfh},
955 
956 	/* RFS3_SYMLINK = 10 */
957 	{rfs3_symlink,
958 	    xdr_SYMLINK3args, NULL_xdrproc_t, sizeof (SYMLINK3args),
959 	    xdr_SYMLINK3res, NULL_xdrproc_t, sizeof (SYMLINK3res),
960 	    nullfree, 0,
961 	    rfs3_symlink_getfh},
962 
963 	/* RFS3_MKNOD = 11 */
964 	{rfs3_mknod,
965 	    xdr_MKNOD3args, NULL_xdrproc_t, sizeof (MKNOD3args),
966 	    xdr_MKNOD3res, NULL_xdrproc_t, sizeof (MKNOD3res),
967 	    nullfree, 0,
968 	    rfs3_mknod_getfh},
969 
970 	/* RFS3_REMOVE = 12 */
971 	{rfs3_remove,
972 	    xdr_diropargs3, NULL_xdrproc_t, sizeof (REMOVE3args),
973 	    xdr_REMOVE3res, NULL_xdrproc_t, sizeof (REMOVE3res),
974 	    nullfree, 0,
975 	    rfs3_remove_getfh},
976 
977 	/* RFS3_RMDIR = 13 */
978 	{rfs3_rmdir,
979 	    xdr_diropargs3, NULL_xdrproc_t, sizeof (RMDIR3args),
980 	    xdr_RMDIR3res, NULL_xdrproc_t, sizeof (RMDIR3res),
981 	    nullfree, 0,
982 	    rfs3_rmdir_getfh},
983 
984 	/* RFS3_RENAME = 14 */
985 	{rfs3_rename,
986 	    xdr_RENAME3args, NULL_xdrproc_t, sizeof (RENAME3args),
987 	    xdr_RENAME3res, NULL_xdrproc_t, sizeof (RENAME3res),
988 	    nullfree, 0,
989 	    rfs3_rename_getfh},
990 
991 	/* RFS3_LINK = 15 */
992 	{rfs3_link,
993 	    xdr_LINK3args, NULL_xdrproc_t, sizeof (LINK3args),
994 	    xdr_LINK3res, NULL_xdrproc_t, sizeof (LINK3res),
995 	    nullfree, 0,
996 	    rfs3_link_getfh},
997 
998 	/* RFS3_READDIR = 16 */
999 	{rfs3_readdir,
1000 	    xdr_READDIR3args, NULL_xdrproc_t, sizeof (READDIR3args),
1001 	    xdr_READDIR3res, NULL_xdrproc_t, sizeof (READDIR3res),
1002 	    rfs3_readdir_free, RPC_IDEMPOTENT,
1003 	    rfs3_readdir_getfh},
1004 
1005 	/* RFS3_READDIRPLUS = 17 */
1006 	{rfs3_readdirplus,
1007 	    xdr_READDIRPLUS3args, NULL_xdrproc_t, sizeof (READDIRPLUS3args),
1008 	    xdr_READDIRPLUS3res, NULL_xdrproc_t, sizeof (READDIRPLUS3res),
1009 	    rfs3_readdirplus_free, RPC_AVOIDWORK,
1010 	    rfs3_readdirplus_getfh},
1011 
1012 	/* RFS3_FSSTAT = 18 */
1013 	{rfs3_fsstat,
1014 	    xdr_nfs_fh3_server, NULL_xdrproc_t, sizeof (FSSTAT3args),
1015 	    xdr_FSSTAT3res, NULL_xdrproc_t, sizeof (FSSTAT3res),
1016 	    nullfree, RPC_IDEMPOTENT,
1017 	    rfs3_fsstat_getfh},
1018 
1019 	/* RFS3_FSINFO = 19 */
1020 	{rfs3_fsinfo,
1021 	    xdr_nfs_fh3_server, NULL_xdrproc_t, sizeof (FSINFO3args),
1022 	    xdr_FSINFO3res, NULL_xdrproc_t, sizeof (FSINFO3res),
1023 	    nullfree, RPC_IDEMPOTENT|RPC_ALLOWANON,
1024 	    rfs3_fsinfo_getfh},
1025 
1026 	/* RFS3_PATHCONF = 20 */
1027 	{rfs3_pathconf,
1028 	    xdr_nfs_fh3_server, NULL_xdrproc_t, sizeof (PATHCONF3args),
1029 	    xdr_PATHCONF3res, NULL_xdrproc_t, sizeof (PATHCONF3res),
1030 	    nullfree, RPC_IDEMPOTENT,
1031 	    rfs3_pathconf_getfh},
1032 
1033 	/* RFS3_COMMIT = 21 */
1034 	{rfs3_commit,
1035 	    xdr_COMMIT3args, NULL_xdrproc_t, sizeof (COMMIT3args),
1036 	    xdr_COMMIT3res, NULL_xdrproc_t, sizeof (COMMIT3res),
1037 	    nullfree, RPC_IDEMPOTENT,
1038 	    rfs3_commit_getfh},
1039 };
1040 
1041 static char *rfscallnames_v4[] = {
1042 	"RFS4_NULL",
1043 	"RFS4_COMPOUND",
1044 	"RFS4_NULL",
1045 	"RFS4_NULL",
1046 	"RFS4_NULL",
1047 	"RFS4_NULL",
1048 	"RFS4_NULL",
1049 	"RFS4_NULL",
1050 	"RFS4_CREATE"
1051 };
1052 
1053 static struct rpcdisp rfsdisptab_v4[] = {
1054 	/*
1055 	 * NFS VERSION 4
1056 	 */
1057 
1058 	/* RFS_NULL = 0 */
1059 	{rpc_null,
1060 	    xdr_void, NULL_xdrproc_t, 0,
1061 	    xdr_void, NULL_xdrproc_t, 0,
1062 	    nullfree, RPC_IDEMPOTENT, 0},
1063 
1064 	/* RFS4_compound = 1 */
1065 	{rfs4_compound,
1066 	    xdr_COMPOUND4args_srv, NULL_xdrproc_t, sizeof (COMPOUND4args),
1067 	    xdr_COMPOUND4res_srv, NULL_xdrproc_t, sizeof (COMPOUND4res),
1068 	    rfs4_compound_free, 0, 0},
1069 };
1070 
1071 union rfs_args {
1072 	/*
1073 	 * NFS VERSION 2
1074 	 */
1075 
1076 	/* RFS_NULL = 0 */
1077 
1078 	/* RFS_GETATTR = 1 */
1079 	fhandle_t nfs2_getattr_args;
1080 
1081 	/* RFS_SETATTR = 2 */
1082 	struct nfssaargs nfs2_setattr_args;
1083 
1084 	/* RFS_ROOT = 3 *** NO LONGER SUPPORTED *** */
1085 
1086 	/* RFS_LOOKUP = 4 */
1087 	struct nfsdiropargs nfs2_lookup_args;
1088 
1089 	/* RFS_READLINK = 5 */
1090 	fhandle_t nfs2_readlink_args;
1091 
1092 	/* RFS_READ = 6 */
1093 	struct nfsreadargs nfs2_read_args;
1094 
1095 	/* RFS_WRITECACHE = 7 *** NO LONGER SUPPORTED *** */
1096 
1097 	/* RFS_WRITE = 8 */
1098 	struct nfswriteargs nfs2_write_args;
1099 
1100 	/* RFS_CREATE = 9 */
1101 	struct nfscreatargs nfs2_create_args;
1102 
1103 	/* RFS_REMOVE = 10 */
1104 	struct nfsdiropargs nfs2_remove_args;
1105 
1106 	/* RFS_RENAME = 11 */
1107 	struct nfsrnmargs nfs2_rename_args;
1108 
1109 	/* RFS_LINK = 12 */
1110 	struct nfslinkargs nfs2_link_args;
1111 
1112 	/* RFS_SYMLINK = 13 */
1113 	struct nfsslargs nfs2_symlink_args;
1114 
1115 	/* RFS_MKDIR = 14 */
1116 	struct nfscreatargs nfs2_mkdir_args;
1117 
1118 	/* RFS_RMDIR = 15 */
1119 	struct nfsdiropargs nfs2_rmdir_args;
1120 
1121 	/* RFS_READDIR = 16 */
1122 	struct nfsrddirargs nfs2_readdir_args;
1123 
1124 	/* RFS_STATFS = 17 */
1125 	fhandle_t nfs2_statfs_args;
1126 
1127 	/*
1128 	 * NFS VERSION 3
1129 	 */
1130 
1131 	/* RFS_NULL = 0 */
1132 
1133 	/* RFS3_GETATTR = 1 */
1134 	GETATTR3args nfs3_getattr_args;
1135 
1136 	/* RFS3_SETATTR = 2 */
1137 	SETATTR3args nfs3_setattr_args;
1138 
1139 	/* RFS3_LOOKUP = 3 */
1140 	LOOKUP3args nfs3_lookup_args;
1141 
1142 	/* RFS3_ACCESS = 4 */
1143 	ACCESS3args nfs3_access_args;
1144 
1145 	/* RFS3_READLINK = 5 */
1146 	READLINK3args nfs3_readlink_args;
1147 
1148 	/* RFS3_READ = 6 */
1149 	READ3args nfs3_read_args;
1150 
1151 	/* RFS3_WRITE = 7 */
1152 	WRITE3args nfs3_write_args;
1153 
1154 	/* RFS3_CREATE = 8 */
1155 	CREATE3args nfs3_create_args;
1156 
1157 	/* RFS3_MKDIR = 9 */
1158 	MKDIR3args nfs3_mkdir_args;
1159 
1160 	/* RFS3_SYMLINK = 10 */
1161 	SYMLINK3args nfs3_symlink_args;
1162 
1163 	/* RFS3_MKNOD = 11 */
1164 	MKNOD3args nfs3_mknod_args;
1165 
1166 	/* RFS3_REMOVE = 12 */
1167 	REMOVE3args nfs3_remove_args;
1168 
1169 	/* RFS3_RMDIR = 13 */
1170 	RMDIR3args nfs3_rmdir_args;
1171 
1172 	/* RFS3_RENAME = 14 */
1173 	RENAME3args nfs3_rename_args;
1174 
1175 	/* RFS3_LINK = 15 */
1176 	LINK3args nfs3_link_args;
1177 
1178 	/* RFS3_READDIR = 16 */
1179 	READDIR3args nfs3_readdir_args;
1180 
1181 	/* RFS3_READDIRPLUS = 17 */
1182 	READDIRPLUS3args nfs3_readdirplus_args;
1183 
1184 	/* RFS3_FSSTAT = 18 */
1185 	FSSTAT3args nfs3_fsstat_args;
1186 
1187 	/* RFS3_FSINFO = 19 */
1188 	FSINFO3args nfs3_fsinfo_args;
1189 
1190 	/* RFS3_PATHCONF = 20 */
1191 	PATHCONF3args nfs3_pathconf_args;
1192 
1193 	/* RFS3_COMMIT = 21 */
1194 	COMMIT3args nfs3_commit_args;
1195 
1196 	/*
1197 	 * NFS VERSION 4
1198 	 */
1199 
1200 	/* RFS_NULL = 0 */
1201 
1202 	/* COMPUND = 1 */
1203 	COMPOUND4args nfs4_compound_args;
1204 };
1205 
1206 union rfs_res {
1207 	/*
1208 	 * NFS VERSION 2
1209 	 */
1210 
1211 	/* RFS_NULL = 0 */
1212 
1213 	/* RFS_GETATTR = 1 */
1214 	struct nfsattrstat nfs2_getattr_res;
1215 
1216 	/* RFS_SETATTR = 2 */
1217 	struct nfsattrstat nfs2_setattr_res;
1218 
1219 	/* RFS_ROOT = 3 *** NO LONGER SUPPORTED *** */
1220 
1221 	/* RFS_LOOKUP = 4 */
1222 	struct nfsdiropres nfs2_lookup_res;
1223 
1224 	/* RFS_READLINK = 5 */
1225 	struct nfsrdlnres nfs2_readlink_res;
1226 
1227 	/* RFS_READ = 6 */
1228 	struct nfsrdresult nfs2_read_res;
1229 
1230 	/* RFS_WRITECACHE = 7 *** NO LONGER SUPPORTED *** */
1231 
1232 	/* RFS_WRITE = 8 */
1233 	struct nfsattrstat nfs2_write_res;
1234 
1235 	/* RFS_CREATE = 9 */
1236 	struct nfsdiropres nfs2_create_res;
1237 
1238 	/* RFS_REMOVE = 10 */
1239 	enum nfsstat nfs2_remove_res;
1240 
1241 	/* RFS_RENAME = 11 */
1242 	enum nfsstat nfs2_rename_res;
1243 
1244 	/* RFS_LINK = 12 */
1245 	enum nfsstat nfs2_link_res;
1246 
1247 	/* RFS_SYMLINK = 13 */
1248 	enum nfsstat nfs2_symlink_res;
1249 
1250 	/* RFS_MKDIR = 14 */
1251 	struct nfsdiropres nfs2_mkdir_res;
1252 
1253 	/* RFS_RMDIR = 15 */
1254 	enum nfsstat nfs2_rmdir_res;
1255 
1256 	/* RFS_READDIR = 16 */
1257 	struct nfsrddirres nfs2_readdir_res;
1258 
1259 	/* RFS_STATFS = 17 */
1260 	struct nfsstatfs nfs2_statfs_res;
1261 
1262 	/*
1263 	 * NFS VERSION 3
1264 	 */
1265 
1266 	/* RFS_NULL = 0 */
1267 
1268 	/* RFS3_GETATTR = 1 */
1269 	GETATTR3res nfs3_getattr_res;
1270 
1271 	/* RFS3_SETATTR = 2 */
1272 	SETATTR3res nfs3_setattr_res;
1273 
1274 	/* RFS3_LOOKUP = 3 */
1275 	LOOKUP3res nfs3_lookup_res;
1276 
1277 	/* RFS3_ACCESS = 4 */
1278 	ACCESS3res nfs3_access_res;
1279 
1280 	/* RFS3_READLINK = 5 */
1281 	READLINK3res nfs3_readlink_res;
1282 
1283 	/* RFS3_READ = 6 */
1284 	READ3res nfs3_read_res;
1285 
1286 	/* RFS3_WRITE = 7 */
1287 	WRITE3res nfs3_write_res;
1288 
1289 	/* RFS3_CREATE = 8 */
1290 	CREATE3res nfs3_create_res;
1291 
1292 	/* RFS3_MKDIR = 9 */
1293 	MKDIR3res nfs3_mkdir_res;
1294 
1295 	/* RFS3_SYMLINK = 10 */
1296 	SYMLINK3res nfs3_symlink_res;
1297 
1298 	/* RFS3_MKNOD = 11 */
1299 	MKNOD3res nfs3_mknod_res;
1300 
1301 	/* RFS3_REMOVE = 12 */
1302 	REMOVE3res nfs3_remove_res;
1303 
1304 	/* RFS3_RMDIR = 13 */
1305 	RMDIR3res nfs3_rmdir_res;
1306 
1307 	/* RFS3_RENAME = 14 */
1308 	RENAME3res nfs3_rename_res;
1309 
1310 	/* RFS3_LINK = 15 */
1311 	LINK3res nfs3_link_res;
1312 
1313 	/* RFS3_READDIR = 16 */
1314 	READDIR3res nfs3_readdir_res;
1315 
1316 	/* RFS3_READDIRPLUS = 17 */
1317 	READDIRPLUS3res nfs3_readdirplus_res;
1318 
1319 	/* RFS3_FSSTAT = 18 */
1320 	FSSTAT3res nfs3_fsstat_res;
1321 
1322 	/* RFS3_FSINFO = 19 */
1323 	FSINFO3res nfs3_fsinfo_res;
1324 
1325 	/* RFS3_PATHCONF = 20 */
1326 	PATHCONF3res nfs3_pathconf_res;
1327 
1328 	/* RFS3_COMMIT = 21 */
1329 	COMMIT3res nfs3_commit_res;
1330 
1331 	/*
1332 	 * NFS VERSION 4
1333 	 */
1334 
1335 	/* RFS_NULL = 0 */
1336 
1337 	/* RFS4_COMPOUND = 1 */
1338 	COMPOUND4res nfs4_compound_res;
1339 
1340 };
1341 
1342 static struct rpc_disptable rfs_disptable[] = {
1343 	{sizeof (rfsdisptab_v2) / sizeof (rfsdisptab_v2[0]),
1344 	    rfscallnames_v2,
1345 	    &rfsproccnt_v2_ptr, rfsdisptab_v2},
1346 	{sizeof (rfsdisptab_v3) / sizeof (rfsdisptab_v3[0]),
1347 	    rfscallnames_v3,
1348 	    &rfsproccnt_v3_ptr, rfsdisptab_v3},
1349 	{sizeof (rfsdisptab_v4) / sizeof (rfsdisptab_v4[0]),
1350 	    rfscallnames_v4,
1351 	    &rfsproccnt_v4_ptr, rfsdisptab_v4},
1352 };
1353 
1354 /*
1355  * If nfs_portmon is set, then clients are required to use privileged
1356  * ports (ports < IPPORT_RESERVED) in order to get NFS services.
1357  *
1358  * N.B.: this attempt to carry forward the already ill-conceived notion
1359  * of privileged ports for TCP/UDP is really quite ineffectual.  Not only
1360  * is it transport-dependent, it's laughably easy to spoof.  If you're
1361  * really interested in security, you must start with secure RPC instead.
1362  */
1363 static int nfs_portmon = 0;
1364 
1365 #ifdef DEBUG
1366 static int cred_hits = 0;
1367 static int cred_misses = 0;
1368 #endif
1369 
1370 
1371 #ifdef DEBUG
1372 /*
1373  * Debug code to allow disabling of rfs_dispatch() use of
1374  * fastxdrargs() and fastxdrres() calls for testing purposes.
1375  */
1376 static int rfs_no_fast_xdrargs = 0;
1377 static int rfs_no_fast_xdrres = 0;
1378 #endif
1379 
1380 union acl_args {
1381 	/*
1382 	 * ACL VERSION 2
1383 	 */
1384 
1385 	/* ACL2_NULL = 0 */
1386 
1387 	/* ACL2_GETACL = 1 */
1388 	GETACL2args acl2_getacl_args;
1389 
1390 	/* ACL2_SETACL = 2 */
1391 	SETACL2args acl2_setacl_args;
1392 
1393 	/* ACL2_GETATTR = 3 */
1394 	GETATTR2args acl2_getattr_args;
1395 
1396 	/* ACL2_ACCESS = 4 */
1397 	ACCESS2args acl2_access_args;
1398 
1399 	/* ACL2_GETXATTRDIR = 5 */
1400 	GETXATTRDIR2args acl2_getxattrdir_args;
1401 
1402 	/*
1403 	 * ACL VERSION 3
1404 	 */
1405 
1406 	/* ACL3_NULL = 0 */
1407 
1408 	/* ACL3_GETACL = 1 */
1409 	GETACL3args acl3_getacl_args;
1410 
1411 	/* ACL3_SETACL = 2 */
1412 	SETACL3args acl3_setacl;
1413 
1414 	/* ACL3_GETXATTRDIR = 3 */
1415 	GETXATTRDIR3args acl3_getxattrdir_args;
1416 
1417 };
1418 
1419 union acl_res {
1420 	/*
1421 	 * ACL VERSION 2
1422 	 */
1423 
1424 	/* ACL2_NULL = 0 */
1425 
1426 	/* ACL2_GETACL = 1 */
1427 	GETACL2res acl2_getacl_res;
1428 
1429 	/* ACL2_SETACL = 2 */
1430 	SETACL2res acl2_setacl_res;
1431 
1432 	/* ACL2_GETATTR = 3 */
1433 	GETATTR2res acl2_getattr_res;
1434 
1435 	/* ACL2_ACCESS = 4 */
1436 	ACCESS2res acl2_access_res;
1437 
1438 	/* ACL2_GETXATTRDIR = 5 */
1439 	GETXATTRDIR2args acl2_getxattrdir_res;
1440 
1441 	/*
1442 	 * ACL VERSION 3
1443 	 */
1444 
1445 	/* ACL3_NULL = 0 */
1446 
1447 	/* ACL3_GETACL = 1 */
1448 	GETACL3res acl3_getacl_res;
1449 
1450 	/* ACL3_SETACL = 2 */
1451 	SETACL3res acl3_setacl_res;
1452 
1453 	/* ACL3_GETXATTRDIR = 3 */
1454 	GETXATTRDIR3res acl3_getxattrdir_res;
1455 
1456 };
1457 
1458 static bool_t
1459 auth_tooweak(struct svc_req *req, char *res)
1460 {
1461 
1462 	if (req->rq_vers == NFS_VERSION && req->rq_proc == RFS_LOOKUP) {
1463 		struct nfsdiropres *dr = (struct nfsdiropres *)res;
1464 		if ((enum wnfsstat)dr->dr_status == WNFSERR_CLNT_FLAVOR)
1465 			return (TRUE);
1466 	} else if (req->rq_vers == NFS_V3 && req->rq_proc == NFSPROC3_LOOKUP) {
1467 		LOOKUP3res *resp = (LOOKUP3res *)res;
1468 		if ((enum wnfsstat)resp->status == WNFSERR_CLNT_FLAVOR)
1469 			return (TRUE);
1470 	}
1471 	return (FALSE);
1472 }
1473 
1474 
1475 static void
1476 common_dispatch(struct svc_req *req, SVCXPRT *xprt, rpcvers_t min_vers,
1477 		rpcvers_t max_vers, char *pgmname,
1478 		struct rpc_disptable *disptable)
1479 {
1480 	int which;
1481 	rpcvers_t vers;
1482 	char *args;
1483 	union {
1484 			union rfs_args ra;
1485 			union acl_args aa;
1486 		} args_buf;
1487 	char *res;
1488 	union {
1489 			union rfs_res rr;
1490 			union acl_res ar;
1491 		} res_buf;
1492 	struct rpcdisp *disp = NULL;
1493 	int dis_flags = 0;
1494 	cred_t *cr;
1495 	int error = 0;
1496 	int anon_ok;
1497 	struct exportinfo *exi = NULL;
1498 	unsigned int nfslog_rec_id;
1499 	int dupstat;
1500 	struct dupreq *dr;
1501 	int authres;
1502 	bool_t publicfh_ok = FALSE;
1503 	enum_t auth_flavor;
1504 	bool_t dupcached = FALSE;
1505 	struct netbuf	nb;
1506 	bool_t logging_enabled = FALSE;
1507 	struct exportinfo *nfslog_exi = NULL;
1508 	char **procnames;
1509 	char cbuf[INET6_ADDRSTRLEN];	/* to hold both IPv4 and IPv6 addr */
1510 	bool_t ro = FALSE;
1511 
1512 	vers = req->rq_vers;
1513 
1514 	if (vers < min_vers || vers > max_vers) {
1515 		svcerr_progvers(req->rq_xprt, min_vers, max_vers);
1516 		error++;
1517 		cmn_err(CE_NOTE, "%s: bad version number %u", pgmname, vers);
1518 		goto done;
1519 	}
1520 	vers -= min_vers;
1521 
1522 	which = req->rq_proc;
1523 	if (which < 0 || which >= disptable[(int)vers].dis_nprocs) {
1524 		svcerr_noproc(req->rq_xprt);
1525 		error++;
1526 		goto done;
1527 	}
1528 
1529 	(*(disptable[(int)vers].dis_proccntp))[which].value.ui64++;
1530 
1531 	disp = &disptable[(int)vers].dis_table[which];
1532 	procnames = disptable[(int)vers].dis_procnames;
1533 
1534 	auth_flavor = req->rq_cred.oa_flavor;
1535 
1536 	/*
1537 	 * Deserialize into the args struct.
1538 	 */
1539 	args = (char *)&args_buf;
1540 
1541 #ifdef DEBUG
1542 	if (rfs_no_fast_xdrargs || (auth_flavor == RPCSEC_GSS) ||
1543 	    disp->dis_fastxdrargs == NULL_xdrproc_t ||
1544 	    !SVC_GETARGS(xprt, disp->dis_fastxdrargs, (char *)&args))
1545 #else
1546 	if ((auth_flavor == RPCSEC_GSS) ||
1547 	    disp->dis_fastxdrargs == NULL_xdrproc_t ||
1548 	    !SVC_GETARGS(xprt, disp->dis_fastxdrargs, (char *)&args))
1549 #endif
1550 	{
1551 		bzero(args, disp->dis_argsz);
1552 		if (!SVC_GETARGS(xprt, disp->dis_xdrargs, args)) {
1553 			error++;
1554 			/*
1555 			 * Check if we are outside our capabilities.
1556 			 */
1557 			if (rfs4_minorvers_mismatch(req, xprt, (void *)args))
1558 				goto done;
1559 
1560 			svcerr_decode(xprt);
1561 			cmn_err(CE_NOTE,
1562 			    "Failed to decode arguments for %s version %u "
1563 			    "procedure %s client %s%s",
1564 			    pgmname, vers + min_vers, procnames[which],
1565 			    client_name(req), client_addr(req, cbuf));
1566 			goto done;
1567 		}
1568 	}
1569 
1570 	/*
1571 	 * If Version 4 use that specific dispatch function.
1572 	 */
1573 	if (req->rq_vers == 4) {
1574 		error += rfs4_dispatch(disp, req, xprt, args);
1575 		goto done;
1576 	}
1577 
1578 	dis_flags = disp->dis_flags;
1579 
1580 	/*
1581 	 * Find export information and check authentication,
1582 	 * setting the credential if everything is ok.
1583 	 */
1584 	if (disp->dis_getfh != NULL) {
1585 		void *fh;
1586 		fsid_t *fsid;
1587 		fid_t *fid, *xfid;
1588 		fhandle_t *fh2;
1589 		nfs_fh3 *fh3;
1590 
1591 		fh = (*disp->dis_getfh)(args);
1592 		switch (req->rq_vers) {
1593 		case NFS_VERSION:
1594 			fh2 = (fhandle_t *)fh;
1595 			fsid = &fh2->fh_fsid;
1596 			fid = (fid_t *)&fh2->fh_len;
1597 			xfid = (fid_t *)&fh2->fh_xlen;
1598 			break;
1599 		case NFS_V3:
1600 			fh3 = (nfs_fh3 *)fh;
1601 			fsid = &fh3->fh3_fsid;
1602 			fid = FH3TOFIDP(fh3);
1603 			xfid = FH3TOXFIDP(fh3);
1604 			break;
1605 		}
1606 
1607 		/*
1608 		 * Fix for bug 1038302 - corbin
1609 		 * There is a problem here if anonymous access is
1610 		 * disallowed.  If the current request is part of the
1611 		 * client's mount process for the requested filesystem,
1612 		 * then it will carry root (uid 0) credentials on it, and
1613 		 * will be denied by checkauth if that client does not
1614 		 * have explicit root=0 permission.  This will cause the
1615 		 * client's mount operation to fail.  As a work-around,
1616 		 * we check here to see if the request is a getattr or
1617 		 * statfs operation on the exported vnode itself, and
1618 		 * pass a flag to checkauth with the result of this test.
1619 		 *
1620 		 * The filehandle refers to the mountpoint itself if
1621 		 * the fh_data and fh_xdata portions of the filehandle
1622 		 * are equal.
1623 		 *
1624 		 * Added anon_ok argument to checkauth().
1625 		 */
1626 
1627 		if ((dis_flags & RPC_ALLOWANON) && EQFID(fid, xfid))
1628 			anon_ok = 1;
1629 		else
1630 			anon_ok = 0;
1631 
1632 		cr = xprt->xp_cred;
1633 		ASSERT(cr != NULL);
1634 #ifdef DEBUG
1635 		if (crgetref(cr) != 1) {
1636 			crfree(cr);
1637 			cr = crget();
1638 			xprt->xp_cred = cr;
1639 			cred_misses++;
1640 		} else
1641 			cred_hits++;
1642 #else
1643 		if (crgetref(cr) != 1) {
1644 			crfree(cr);
1645 			cr = crget();
1646 			xprt->xp_cred = cr;
1647 		}
1648 #endif
1649 
1650 		exi = checkexport(fsid, xfid);
1651 
1652 		if (exi != NULL) {
1653 			publicfh_ok = PUBLICFH_CHECK(disp, exi, fsid, xfid);
1654 
1655 			/*
1656 			 * Don't allow non-V4 clients access
1657 			 * to pseudo exports
1658 			 */
1659 			if (PSEUDO(exi)) {
1660 				svcerr_weakauth(xprt);
1661 				error++;
1662 				goto done;
1663 			}
1664 
1665 			authres = checkauth(exi, req, cr, anon_ok, publicfh_ok,
1666 			    &ro);
1667 			/*
1668 			 * authres >  0: authentication OK - proceed
1669 			 * authres == 0: authentication weak - return error
1670 			 * authres <  0: authentication timeout - drop
1671 			 */
1672 			if (authres <= 0) {
1673 				if (authres == 0) {
1674 					svcerr_weakauth(xprt);
1675 					error++;
1676 				}
1677 				goto done;
1678 			}
1679 		}
1680 	} else
1681 		cr = NULL;
1682 
1683 	if ((dis_flags & RPC_MAPRESP) && (auth_flavor != RPCSEC_GSS)) {
1684 		res = (char *)SVC_GETRES(xprt, disp->dis_ressz);
1685 		if (res == NULL)
1686 			res = (char *)&res_buf;
1687 	} else
1688 		res = (char *)&res_buf;
1689 
1690 	if (!(dis_flags & RPC_IDEMPOTENT)) {
1691 		dupstat = SVC_DUP_EXT(xprt, req, res, disp->dis_ressz, &dr,
1692 		    &dupcached);
1693 
1694 		switch (dupstat) {
1695 		case DUP_ERROR:
1696 			svcerr_systemerr(xprt);
1697 			error++;
1698 			goto done;
1699 			/* NOTREACHED */
1700 		case DUP_INPROGRESS:
1701 			if (res != (char *)&res_buf)
1702 				SVC_FREERES(xprt);
1703 			error++;
1704 			goto done;
1705 			/* NOTREACHED */
1706 		case DUP_NEW:
1707 		case DUP_DROP:
1708 			curthread->t_flag |= T_DONTPEND;
1709 
1710 			(*disp->dis_proc)(args, res, exi, req, cr, ro);
1711 
1712 			curthread->t_flag &= ~T_DONTPEND;
1713 			if (curthread->t_flag & T_WOULDBLOCK) {
1714 				curthread->t_flag &= ~T_WOULDBLOCK;
1715 				SVC_DUPDONE_EXT(xprt, dr, res, NULL,
1716 				    disp->dis_ressz, DUP_DROP);
1717 				if (res != (char *)&res_buf)
1718 					SVC_FREERES(xprt);
1719 				error++;
1720 				goto done;
1721 			}
1722 			if (dis_flags & RPC_AVOIDWORK) {
1723 				SVC_DUPDONE_EXT(xprt, dr, res, NULL,
1724 				    disp->dis_ressz, DUP_DROP);
1725 			} else {
1726 				SVC_DUPDONE_EXT(xprt, dr, res,
1727 				    disp->dis_resfree == nullfree ? NULL :
1728 				    disp->dis_resfree,
1729 				    disp->dis_ressz, DUP_DONE);
1730 				dupcached = TRUE;
1731 			}
1732 			break;
1733 		case DUP_DONE:
1734 			break;
1735 		}
1736 
1737 	} else {
1738 		curthread->t_flag |= T_DONTPEND;
1739 
1740 		(*disp->dis_proc)(args, res, exi, req, cr, ro);
1741 
1742 		curthread->t_flag &= ~T_DONTPEND;
1743 		if (curthread->t_flag & T_WOULDBLOCK) {
1744 			curthread->t_flag &= ~T_WOULDBLOCK;
1745 			if (res != (char *)&res_buf)
1746 				SVC_FREERES(xprt);
1747 			error++;
1748 			goto done;
1749 		}
1750 	}
1751 
1752 	if (auth_tooweak(req, res)) {
1753 		svcerr_weakauth(xprt);
1754 		error++;
1755 		goto done;
1756 	}
1757 
1758 	/*
1759 	 * Check to see if logging has been enabled on the server.
1760 	 * If so, then obtain the export info struct to be used for
1761 	 * the later writing of the log record.  This is done for
1762 	 * the case that a lookup is done across a non-logged public
1763 	 * file system.
1764 	 */
1765 	if (nfslog_buffer_list != NULL) {
1766 		nfslog_exi = nfslog_get_exi(exi, req, res, &nfslog_rec_id);
1767 		/*
1768 		 * Is logging enabled?
1769 		 */
1770 		logging_enabled = (nfslog_exi != NULL);
1771 
1772 		/*
1773 		 * Copy the netbuf for logging purposes, before it is
1774 		 * freed by svc_sendreply().
1775 		 */
1776 		if (logging_enabled) {
1777 			NFSLOG_COPY_NETBUF(nfslog_exi, xprt, &nb);
1778 			/*
1779 			 * If RPC_MAPRESP flag set (i.e. in V2 ops) the
1780 			 * res gets copied directly into the mbuf and
1781 			 * may be freed soon after the sendreply. So we
1782 			 * must copy it here to a safe place...
1783 			 */
1784 			if (res != (char *)&res_buf) {
1785 				bcopy(res, (char *)&res_buf, disp->dis_ressz);
1786 			}
1787 		}
1788 	}
1789 
1790 	/*
1791 	 * Serialize and send results struct
1792 	 */
1793 #ifdef DEBUG
1794 	if (rfs_no_fast_xdrres == 0 && res != (char *)&res_buf)
1795 #else
1796 	if (res != (char *)&res_buf)
1797 #endif
1798 	{
1799 		if (!svc_sendreply(xprt, disp->dis_fastxdrres, res)) {
1800 			cmn_err(CE_NOTE, "%s: bad sendreply", pgmname);
1801 			svcerr_systemerr(xprt);
1802 			error++;
1803 		}
1804 	} else {
1805 		if (!svc_sendreply(xprt, disp->dis_xdrres, res)) {
1806 			cmn_err(CE_NOTE, "%s: bad sendreply", pgmname);
1807 			svcerr_systemerr(xprt);
1808 			error++;
1809 		}
1810 	}
1811 
1812 	/*
1813 	 * Log if needed
1814 	 */
1815 	if (logging_enabled) {
1816 		nfslog_write_record(nfslog_exi, req, args, (char *)&res_buf,
1817 		    cr, &nb, nfslog_rec_id, NFSLOG_ONE_BUFFER);
1818 		exi_rele(nfslog_exi);
1819 		kmem_free((&nb)->buf, (&nb)->len);
1820 	}
1821 
1822 	/*
1823 	 * Free results struct. With the addition of NFS V4 we can
1824 	 * have non-idempotent procedures with functions.
1825 	 */
1826 	if (disp->dis_resfree != nullfree && dupcached == FALSE) {
1827 		(*disp->dis_resfree)(res);
1828 	}
1829 
1830 done:
1831 	/*
1832 	 * Free arguments struct
1833 	 */
1834 	if (disp) {
1835 		if (!SVC_FREEARGS(xprt, disp->dis_xdrargs, args)) {
1836 			cmn_err(CE_NOTE, "%s: bad freeargs", pgmname);
1837 			error++;
1838 		}
1839 	} else {
1840 		if (!SVC_FREEARGS(xprt, (xdrproc_t)0, (caddr_t)0)) {
1841 			cmn_err(CE_NOTE, "%s: bad freeargs", pgmname);
1842 			error++;
1843 		}
1844 	}
1845 
1846 	if (exi != NULL)
1847 		exi_rele(exi);
1848 
1849 	global_svstat_ptr[req->rq_vers][NFS_BADCALLS].value.ui64 += error;
1850 
1851 	global_svstat_ptr[req->rq_vers][NFS_CALLS].value.ui64++;
1852 }
1853 
1854 static void
1855 rfs_dispatch(struct svc_req *req, SVCXPRT *xprt)
1856 {
1857 	common_dispatch(req, xprt, NFS_VERSMIN, NFS_VERSMAX,
1858 	    "NFS", rfs_disptable);
1859 }
1860 
1861 static char *aclcallnames_v2[] = {
1862 	"ACL2_NULL",
1863 	"ACL2_GETACL",
1864 	"ACL2_SETACL",
1865 	"ACL2_GETATTR",
1866 	"ACL2_ACCESS",
1867 	"ACL2_GETXATTRDIR"
1868 };
1869 
1870 static struct rpcdisp acldisptab_v2[] = {
1871 	/*
1872 	 * ACL VERSION 2
1873 	 */
1874 
1875 	/* ACL2_NULL = 0 */
1876 	{rpc_null,
1877 	    xdr_void, NULL_xdrproc_t, 0,
1878 	    xdr_void, NULL_xdrproc_t, 0,
1879 	    nullfree, RPC_IDEMPOTENT,
1880 	    0},
1881 
1882 	/* ACL2_GETACL = 1 */
1883 	{acl2_getacl,
1884 	    xdr_GETACL2args, xdr_fastGETACL2args, sizeof (GETACL2args),
1885 	    xdr_GETACL2res, NULL_xdrproc_t, sizeof (GETACL2res),
1886 	    acl2_getacl_free, RPC_IDEMPOTENT,
1887 	    acl2_getacl_getfh},
1888 
1889 	/* ACL2_SETACL = 2 */
1890 	{acl2_setacl,
1891 	    xdr_SETACL2args, NULL_xdrproc_t, sizeof (SETACL2args),
1892 #ifdef _LITTLE_ENDIAN
1893 	    xdr_SETACL2res, xdr_fastSETACL2res, sizeof (SETACL2res),
1894 #else
1895 	    xdr_SETACL2res, NULL_xdrproc_t, sizeof (SETACL2res),
1896 #endif
1897 	    nullfree, RPC_MAPRESP,
1898 	    acl2_setacl_getfh},
1899 
1900 	/* ACL2_GETATTR = 3 */
1901 	{acl2_getattr,
1902 	    xdr_GETATTR2args, xdr_fastGETATTR2args, sizeof (GETATTR2args),
1903 #ifdef _LITTLE_ENDIAN
1904 	    xdr_GETATTR2res, xdr_fastGETATTR2res, sizeof (GETATTR2res),
1905 #else
1906 	    xdr_GETATTR2res, NULL_xdrproc_t, sizeof (GETATTR2res),
1907 #endif
1908 	    nullfree, RPC_IDEMPOTENT|RPC_ALLOWANON|RPC_MAPRESP,
1909 	    acl2_getattr_getfh},
1910 
1911 	/* ACL2_ACCESS = 4 */
1912 	{acl2_access,
1913 	    xdr_ACCESS2args, xdr_fastACCESS2args, sizeof (ACCESS2args),
1914 #ifdef _LITTLE_ENDIAN
1915 	    xdr_ACCESS2res, xdr_fastACCESS2res, sizeof (ACCESS2res),
1916 #else
1917 	    xdr_ACCESS2res, NULL_xdrproc_t, sizeof (ACCESS2res),
1918 #endif
1919 	    nullfree, RPC_IDEMPOTENT|RPC_MAPRESP,
1920 	    acl2_access_getfh},
1921 
1922 	/* ACL2_GETXATTRDIR = 5 */
1923 	{acl2_getxattrdir,
1924 	    xdr_GETXATTRDIR2args, NULL_xdrproc_t, sizeof (GETXATTRDIR2args),
1925 	    xdr_GETXATTRDIR2res, NULL_xdrproc_t, sizeof (GETXATTRDIR2res),
1926 	    nullfree, RPC_IDEMPOTENT,
1927 	    acl2_getxattrdir_getfh},
1928 };
1929 
1930 static char *aclcallnames_v3[] = {
1931 	"ACL3_NULL",
1932 	"ACL3_GETACL",
1933 	"ACL3_SETACL",
1934 	"ACL3_GETXATTRDIR"
1935 };
1936 
1937 static struct rpcdisp acldisptab_v3[] = {
1938 	/*
1939 	 * ACL VERSION 3
1940 	 */
1941 
1942 	/* ACL3_NULL = 0 */
1943 	{rpc_null,
1944 	    xdr_void, NULL_xdrproc_t, 0,
1945 	    xdr_void, NULL_xdrproc_t, 0,
1946 	    nullfree, RPC_IDEMPOTENT,
1947 	    0},
1948 
1949 	/* ACL3_GETACL = 1 */
1950 	{acl3_getacl,
1951 	    xdr_GETACL3args, NULL_xdrproc_t, sizeof (GETACL3args),
1952 	    xdr_GETACL3res, NULL_xdrproc_t, sizeof (GETACL3res),
1953 	    acl3_getacl_free, RPC_IDEMPOTENT,
1954 	    acl3_getacl_getfh},
1955 
1956 	/* ACL3_SETACL = 2 */
1957 	{acl3_setacl,
1958 	    xdr_SETACL3args, NULL_xdrproc_t, sizeof (SETACL3args),
1959 	    xdr_SETACL3res, NULL_xdrproc_t, sizeof (SETACL3res),
1960 	    nullfree, 0,
1961 	    acl3_setacl_getfh},
1962 
1963 	/* ACL3_GETXATTRDIR = 3 */
1964 	{acl3_getxattrdir,
1965 	    xdr_GETXATTRDIR3args, NULL_xdrproc_t, sizeof (GETXATTRDIR3args),
1966 	    xdr_GETXATTRDIR3res, NULL_xdrproc_t, sizeof (GETXATTRDIR3res),
1967 	    nullfree, RPC_IDEMPOTENT,
1968 	    acl3_getxattrdir_getfh},
1969 };
1970 
1971 static struct rpc_disptable acl_disptable[] = {
1972 	{sizeof (acldisptab_v2) / sizeof (acldisptab_v2[0]),
1973 		aclcallnames_v2,
1974 		&aclproccnt_v2_ptr, acldisptab_v2},
1975 	{sizeof (acldisptab_v3) / sizeof (acldisptab_v3[0]),
1976 		aclcallnames_v3,
1977 		&aclproccnt_v3_ptr, acldisptab_v3},
1978 };
1979 
1980 static void
1981 acl_dispatch(struct svc_req *req, SVCXPRT *xprt)
1982 {
1983 	common_dispatch(req, xprt, NFS_ACL_VERSMIN, NFS_ACL_VERSMAX,
1984 	    "ACL", acl_disptable);
1985 }
1986 
1987 int
1988 checkwin(int flavor, int window, struct svc_req *req)
1989 {
1990 	struct authdes_cred *adc;
1991 
1992 	switch (flavor) {
1993 	case AUTH_DES:
1994 		adc = (struct authdes_cred *)req->rq_clntcred;
1995 		CTASSERT(sizeof (struct authdes_cred) <= RQCRED_SIZE);
1996 		if (adc->adc_fullname.window > window)
1997 			return (0);
1998 		break;
1999 
2000 	default:
2001 		break;
2002 	}
2003 	return (1);
2004 }
2005 
2006 
2007 /*
2008  * checkauth() will check the access permission against the export
2009  * information.  Then map root uid/gid to appropriate uid/gid.
2010  *
2011  * This routine is used by NFS V3 and V2 code.
2012  */
2013 static int
2014 checkauth(struct exportinfo *exi, struct svc_req *req, cred_t *cr, int anon_ok,
2015     bool_t publicfh_ok, bool_t *ro)
2016 {
2017 	int i, nfsflavor, rpcflavor, stat, access;
2018 	struct secinfo *secp;
2019 	caddr_t principal;
2020 	char buf[INET6_ADDRSTRLEN]; /* to hold both IPv4 and IPv6 addr */
2021 	int anon_res = 0;
2022 
2023 	uid_t uid;
2024 	gid_t gid;
2025 	uint_t ngids;
2026 	gid_t *gids;
2027 
2028 	/*
2029 	 * Check for privileged port number
2030 	 * N.B.:  this assumes that we know the format of a netbuf.
2031 	 */
2032 	if (nfs_portmon) {
2033 		struct sockaddr *ca;
2034 		ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2035 
2036 		if (ca == NULL)
2037 			return (0);
2038 
2039 		if ((ca->sa_family == AF_INET &&
2040 		    ntohs(((struct sockaddr_in *)ca)->sin_port) >=
2041 		    IPPORT_RESERVED) ||
2042 		    (ca->sa_family == AF_INET6 &&
2043 		    ntohs(((struct sockaddr_in6 *)ca)->sin6_port) >=
2044 		    IPPORT_RESERVED)) {
2045 			cmn_err(CE_NOTE,
2046 			    "nfs_server: client %s%ssent NFS request from "
2047 			    "unprivileged port",
2048 			    client_name(req), client_addr(req, buf));
2049 			return (0);
2050 		}
2051 	}
2052 
2053 	/*
2054 	 *  return 1 on success or 0 on failure
2055 	 */
2056 	stat = sec_svc_getcred(req, cr, &principal, &nfsflavor);
2057 
2058 	/*
2059 	 * A failed AUTH_UNIX sec_svc_getcred() implies we couldn't set
2060 	 * the credentials; below we map that to anonymous.
2061 	 */
2062 	if (!stat && nfsflavor != AUTH_UNIX) {
2063 		cmn_err(CE_NOTE,
2064 		    "nfs_server: couldn't get unix cred for %s",
2065 		    client_name(req));
2066 		return (0);
2067 	}
2068 
2069 	/*
2070 	 * Short circuit checkauth() on operations that support the
2071 	 * public filehandle, and if the request for that operation
2072 	 * is using the public filehandle. Note that we must call
2073 	 * sec_svc_getcred() first so that xp_cookie is set to the
2074 	 * right value. Normally xp_cookie is just the RPC flavor
2075 	 * of the the request, but in the case of RPCSEC_GSS it
2076 	 * could be a pseudo flavor.
2077 	 */
2078 	if (publicfh_ok)
2079 		return (1);
2080 
2081 	rpcflavor = req->rq_cred.oa_flavor;
2082 	/*
2083 	 * Check if the auth flavor is valid for this export
2084 	 */
2085 	access = nfsauth_access(exi, req, cr, &uid, &gid, &ngids, &gids);
2086 	if (access & NFSAUTH_DROP)
2087 		return (-1);	/* drop the request */
2088 
2089 	if (access & NFSAUTH_RO)
2090 		*ro = TRUE;
2091 
2092 	if (access & NFSAUTH_DENIED) {
2093 		/*
2094 		 * If anon_ok == 1 and we got NFSAUTH_DENIED, it was
2095 		 * probably due to the flavor not matching during
2096 		 * the mount attempt. So map the flavor to AUTH_NONE
2097 		 * so that the credentials get mapped to the anonymous
2098 		 * user.
2099 		 */
2100 		if (anon_ok == 1)
2101 			rpcflavor = AUTH_NONE;
2102 		else
2103 			return (0);	/* deny access */
2104 
2105 	} else if (access & NFSAUTH_MAPNONE) {
2106 		/*
2107 		 * Access was granted even though the flavor mismatched
2108 		 * because AUTH_NONE was one of the exported flavors.
2109 		 */
2110 		rpcflavor = AUTH_NONE;
2111 
2112 	} else if (access & NFSAUTH_WRONGSEC) {
2113 		/*
2114 		 * NFSAUTH_WRONGSEC is used for NFSv4. If we get here,
2115 		 * it means a client ignored the list of allowed flavors
2116 		 * returned via the MOUNT protocol. So we just disallow it!
2117 		 */
2118 		return (0);
2119 	}
2120 
2121 	if (rpcflavor != AUTH_SYS)
2122 		kmem_free(gids, ngids * sizeof (gid_t));
2123 
2124 	switch (rpcflavor) {
2125 	case AUTH_NONE:
2126 		anon_res = crsetugid(cr, exi->exi_export.ex_anon,
2127 		    exi->exi_export.ex_anon);
2128 		(void) crsetgroups(cr, 0, NULL);
2129 		break;
2130 
2131 	case AUTH_UNIX:
2132 		if (!stat || crgetuid(cr) == 0 && !(access & NFSAUTH_UIDMAP)) {
2133 			anon_res = crsetugid(cr, exi->exi_export.ex_anon,
2134 			    exi->exi_export.ex_anon);
2135 			(void) crsetgroups(cr, 0, NULL);
2136 		} else if (crgetuid(cr) == 0 && access & NFSAUTH_ROOT) {
2137 			/*
2138 			 * It is root, so apply rootid to get real UID
2139 			 * Find the secinfo structure.  We should be able
2140 			 * to find it by the time we reach here.
2141 			 * nfsauth_access() has done the checking.
2142 			 */
2143 			secp = NULL;
2144 			for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2145 				struct secinfo *sptr;
2146 				sptr = &exi->exi_export.ex_secinfo[i];
2147 				if (sptr->s_secinfo.sc_nfsnum == nfsflavor) {
2148 					secp = sptr;
2149 					break;
2150 				}
2151 			}
2152 			if (secp != NULL) {
2153 				(void) crsetugid(cr, secp->s_rootid,
2154 				    secp->s_rootid);
2155 				(void) crsetgroups(cr, 0, NULL);
2156 			}
2157 		} else if (crgetuid(cr) != uid || crgetgid(cr) != gid) {
2158 			if (crsetugid(cr, uid, gid) != 0)
2159 				anon_res = crsetugid(cr,
2160 				    exi->exi_export.ex_anon,
2161 				    exi->exi_export.ex_anon);
2162 			(void) crsetgroups(cr, 0, NULL);
2163 		} else if (access & NFSAUTH_GROUPS) {
2164 			(void) crsetgroups(cr, ngids, gids);
2165 		}
2166 
2167 		kmem_free(gids, ngids * sizeof (gid_t));
2168 
2169 		break;
2170 
2171 	case AUTH_DES:
2172 	case RPCSEC_GSS:
2173 		/*
2174 		 *  Find the secinfo structure.  We should be able
2175 		 *  to find it by the time we reach here.
2176 		 *  nfsauth_access() has done the checking.
2177 		 */
2178 		secp = NULL;
2179 		for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2180 			if (exi->exi_export.ex_secinfo[i].s_secinfo.sc_nfsnum ==
2181 			    nfsflavor) {
2182 				secp = &exi->exi_export.ex_secinfo[i];
2183 				break;
2184 			}
2185 		}
2186 
2187 		if (!secp) {
2188 			cmn_err(CE_NOTE, "nfs_server: client %s%shad "
2189 			    "no secinfo data for flavor %d",
2190 			    client_name(req), client_addr(req, buf),
2191 			    nfsflavor);
2192 			return (0);
2193 		}
2194 
2195 		if (!checkwin(rpcflavor, secp->s_window, req)) {
2196 			cmn_err(CE_NOTE,
2197 			    "nfs_server: client %s%sused invalid "
2198 			    "auth window value",
2199 			    client_name(req), client_addr(req, buf));
2200 			return (0);
2201 		}
2202 
2203 		/*
2204 		 * Map root principals listed in the share's root= list to root,
2205 		 * and map any others principals that were mapped to root by RPC
2206 		 * to anon.
2207 		 */
2208 		if (principal && sec_svc_inrootlist(rpcflavor, principal,
2209 		    secp->s_rootcnt, secp->s_rootnames)) {
2210 			if (crgetuid(cr) == 0 && secp->s_rootid == 0)
2211 				return (1);
2212 
2213 
2214 			(void) crsetugid(cr, secp->s_rootid, secp->s_rootid);
2215 
2216 			/*
2217 			 * NOTE: If and when kernel-land privilege tracing is
2218 			 * added this may have to be replaced with code that
2219 			 * retrieves root's supplementary groups (e.g., using
2220 			 * kgss_get_group_info().  In the meantime principals
2221 			 * mapped to uid 0 get all privileges, so setting cr's
2222 			 * supplementary groups for them does nothing.
2223 			 */
2224 			(void) crsetgroups(cr, 0, NULL);
2225 
2226 			return (1);
2227 		}
2228 
2229 		/*
2230 		 * Not a root princ, or not in root list, map UID 0/nobody to
2231 		 * the anon ID for the share.  (RPC sets cr's UIDs and GIDs to
2232 		 * UID_NOBODY and GID_NOBODY, respectively.)
2233 		 */
2234 		if (crgetuid(cr) != 0 &&
2235 		    (crgetuid(cr) != UID_NOBODY || crgetgid(cr) != GID_NOBODY))
2236 			return (1);
2237 
2238 		anon_res = crsetugid(cr, exi->exi_export.ex_anon,
2239 		    exi->exi_export.ex_anon);
2240 		(void) crsetgroups(cr, 0, NULL);
2241 		break;
2242 	default:
2243 		return (0);
2244 	} /* switch on rpcflavor */
2245 
2246 	/*
2247 	 * Even if anon access is disallowed via ex_anon == -1, we allow
2248 	 * this access if anon_ok is set.  So set creds to the default
2249 	 * "nobody" id.
2250 	 */
2251 	if (anon_res != 0) {
2252 		if (anon_ok == 0) {
2253 			cmn_err(CE_NOTE,
2254 			    "nfs_server: client %s%ssent wrong "
2255 			    "authentication for %s",
2256 			    client_name(req), client_addr(req, buf),
2257 			    exi->exi_export.ex_path ?
2258 			    exi->exi_export.ex_path : "?");
2259 			return (0);
2260 		}
2261 
2262 		if (crsetugid(cr, UID_NOBODY, GID_NOBODY) != 0)
2263 			return (0);
2264 	}
2265 
2266 	return (1);
2267 }
2268 
2269 /*
2270  * returns 0 on failure, -1 on a drop, -2 on wrong security flavor,
2271  * and 1 on success
2272  */
2273 int
2274 checkauth4(struct compound_state *cs, struct svc_req *req)
2275 {
2276 	int i, rpcflavor, access;
2277 	struct secinfo *secp;
2278 	char buf[MAXHOST + 1];
2279 	int anon_res = 0, nfsflavor;
2280 	struct exportinfo *exi;
2281 	cred_t	*cr;
2282 	caddr_t	principal;
2283 
2284 	uid_t uid;
2285 	gid_t gid;
2286 	uint_t ngids;
2287 	gid_t *gids;
2288 
2289 	exi = cs->exi;
2290 	cr = cs->cr;
2291 	principal = cs->principal;
2292 	nfsflavor = cs->nfsflavor;
2293 
2294 	ASSERT(cr != NULL);
2295 
2296 	rpcflavor = req->rq_cred.oa_flavor;
2297 	cs->access &= ~CS_ACCESS_LIMITED;
2298 
2299 	/*
2300 	 * Check for privileged port number
2301 	 * N.B.:  this assumes that we know the format of a netbuf.
2302 	 */
2303 	if (nfs_portmon) {
2304 		struct sockaddr *ca;
2305 		ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2306 
2307 		if (ca == NULL)
2308 			return (0);
2309 
2310 		if ((ca->sa_family == AF_INET &&
2311 		    ntohs(((struct sockaddr_in *)ca)->sin_port) >=
2312 		    IPPORT_RESERVED) ||
2313 		    (ca->sa_family == AF_INET6 &&
2314 		    ntohs(((struct sockaddr_in6 *)ca)->sin6_port) >=
2315 		    IPPORT_RESERVED)) {
2316 			cmn_err(CE_NOTE,
2317 			    "nfs_server: client %s%ssent NFSv4 request from "
2318 			    "unprivileged port",
2319 			    client_name(req), client_addr(req, buf));
2320 			return (0);
2321 		}
2322 	}
2323 
2324 	/*
2325 	 * Check the access right per auth flavor on the vnode of
2326 	 * this export for the given request.
2327 	 */
2328 	access = nfsauth4_access(cs->exi, cs->vp, req, cr, &uid, &gid, &ngids,
2329 	    &gids);
2330 
2331 	if (access & NFSAUTH_WRONGSEC)
2332 		return (-2);	/* no access for this security flavor */
2333 
2334 	if (access & NFSAUTH_DROP)
2335 		return (-1);	/* drop the request */
2336 
2337 	if (access & NFSAUTH_DENIED) {
2338 
2339 		if (exi->exi_export.ex_seccnt > 0)
2340 			return (0);	/* deny access */
2341 
2342 	} else if (access & NFSAUTH_LIMITED) {
2343 
2344 		cs->access |= CS_ACCESS_LIMITED;
2345 
2346 	} else if (access & NFSAUTH_MAPNONE) {
2347 		/*
2348 		 * Access was granted even though the flavor mismatched
2349 		 * because AUTH_NONE was one of the exported flavors.
2350 		 */
2351 		rpcflavor = AUTH_NONE;
2352 	}
2353 
2354 	/*
2355 	 * XXX probably need to redo some of it for nfsv4?
2356 	 * return 1 on success or 0 on failure
2357 	 */
2358 
2359 	if (rpcflavor != AUTH_SYS)
2360 		kmem_free(gids, ngids * sizeof (gid_t));
2361 
2362 	switch (rpcflavor) {
2363 	case AUTH_NONE:
2364 		anon_res = crsetugid(cr, exi->exi_export.ex_anon,
2365 		    exi->exi_export.ex_anon);
2366 		(void) crsetgroups(cr, 0, NULL);
2367 		break;
2368 
2369 	case AUTH_UNIX:
2370 		if (crgetuid(cr) == 0 && !(access & NFSAUTH_UIDMAP)) {
2371 			anon_res = crsetugid(cr, exi->exi_export.ex_anon,
2372 			    exi->exi_export.ex_anon);
2373 			(void) crsetgroups(cr, 0, NULL);
2374 		} else if (crgetuid(cr) == 0 && access & NFSAUTH_ROOT) {
2375 			/*
2376 			 * It is root, so apply rootid to get real UID
2377 			 * Find the secinfo structure.  We should be able
2378 			 * to find it by the time we reach here.
2379 			 * nfsauth_access() has done the checking.
2380 			 */
2381 			secp = NULL;
2382 			for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2383 				struct secinfo *sptr;
2384 				sptr = &exi->exi_export.ex_secinfo[i];
2385 				if (sptr->s_secinfo.sc_nfsnum == nfsflavor) {
2386 					secp = &exi->exi_export.ex_secinfo[i];
2387 					break;
2388 				}
2389 			}
2390 			if (secp != NULL) {
2391 				(void) crsetugid(cr, secp->s_rootid,
2392 				    secp->s_rootid);
2393 				(void) crsetgroups(cr, 0, NULL);
2394 			}
2395 		} else if (crgetuid(cr) != uid || crgetgid(cr) != gid) {
2396 			if (crsetugid(cr, uid, gid) != 0)
2397 				anon_res = crsetugid(cr,
2398 				    exi->exi_export.ex_anon,
2399 				    exi->exi_export.ex_anon);
2400 			(void) crsetgroups(cr, 0, NULL);
2401 		} if (access & NFSAUTH_GROUPS) {
2402 			(void) crsetgroups(cr, ngids, gids);
2403 		}
2404 
2405 		kmem_free(gids, ngids * sizeof (gid_t));
2406 
2407 		break;
2408 
2409 	default:
2410 		/*
2411 		 *  Find the secinfo structure.  We should be able
2412 		 *  to find it by the time we reach here.
2413 		 *  nfsauth_access() has done the checking.
2414 		 */
2415 		secp = NULL;
2416 		for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2417 			if (exi->exi_export.ex_secinfo[i].s_secinfo.sc_nfsnum ==
2418 			    nfsflavor) {
2419 				secp = &exi->exi_export.ex_secinfo[i];
2420 				break;
2421 			}
2422 		}
2423 
2424 		if (!secp) {
2425 			cmn_err(CE_NOTE, "nfs_server: client %s%shad "
2426 			    "no secinfo data for flavor %d",
2427 			    client_name(req), client_addr(req, buf),
2428 			    nfsflavor);
2429 			return (0);
2430 		}
2431 
2432 		if (!checkwin(rpcflavor, secp->s_window, req)) {
2433 			cmn_err(CE_NOTE,
2434 			    "nfs_server: client %s%sused invalid "
2435 			    "auth window value",
2436 			    client_name(req), client_addr(req, buf));
2437 			return (0);
2438 		}
2439 
2440 		/*
2441 		 * Map root principals listed in the share's root= list to root,
2442 		 * and map any others principals that were mapped to root by RPC
2443 		 * to anon. If not going to anon, set to rootid (root_mapping).
2444 		 */
2445 		if (principal && sec_svc_inrootlist(rpcflavor, principal,
2446 		    secp->s_rootcnt, secp->s_rootnames)) {
2447 			if (crgetuid(cr) == 0 && secp->s_rootid == 0)
2448 				return (1);
2449 
2450 			(void) crsetugid(cr, secp->s_rootid, secp->s_rootid);
2451 
2452 			/*
2453 			 * NOTE: If and when kernel-land privilege tracing is
2454 			 * added this may have to be replaced with code that
2455 			 * retrieves root's supplementary groups (e.g., using
2456 			 * kgss_get_group_info().  In the meantime principals
2457 			 * mapped to uid 0 get all privileges, so setting cr's
2458 			 * supplementary groups for them does nothing.
2459 			 */
2460 			(void) crsetgroups(cr, 0, NULL);
2461 
2462 			return (1);
2463 		}
2464 
2465 		/*
2466 		 * Not a root princ, or not in root list, map UID 0/nobody to
2467 		 * the anon ID for the share.  (RPC sets cr's UIDs and GIDs to
2468 		 * UID_NOBODY and GID_NOBODY, respectively.)
2469 		 */
2470 		if (crgetuid(cr) != 0 &&
2471 		    (crgetuid(cr) != UID_NOBODY || crgetgid(cr) != GID_NOBODY))
2472 			return (1);
2473 
2474 		anon_res = crsetugid(cr, exi->exi_export.ex_anon,
2475 		    exi->exi_export.ex_anon);
2476 		(void) crsetgroups(cr, 0, NULL);
2477 		break;
2478 	} /* switch on rpcflavor */
2479 
2480 	/*
2481 	 * Even if anon access is disallowed via ex_anon == -1, we allow
2482 	 * this access if anon_ok is set.  So set creds to the default
2483 	 * "nobody" id.
2484 	 */
2485 
2486 	if (anon_res != 0) {
2487 		cmn_err(CE_NOTE,
2488 		    "nfs_server: client %s%ssent wrong "
2489 		    "authentication for %s",
2490 		    client_name(req), client_addr(req, buf),
2491 		    exi->exi_export.ex_path ?
2492 		    exi->exi_export.ex_path : "?");
2493 		return (0);
2494 	}
2495 
2496 	return (1);
2497 }
2498 
2499 
2500 static char *
2501 client_name(struct svc_req *req)
2502 {
2503 	char *hostname = NULL;
2504 
2505 	/*
2506 	 * If it's a Unix cred then use the
2507 	 * hostname from the credential.
2508 	 */
2509 	if (req->rq_cred.oa_flavor == AUTH_UNIX) {
2510 		hostname = ((struct authunix_parms *)
2511 		    req->rq_clntcred)->aup_machname;
2512 	}
2513 	if (hostname == NULL)
2514 		hostname = "";
2515 
2516 	return (hostname);
2517 }
2518 
2519 static char *
2520 client_addr(struct svc_req *req, char *buf)
2521 {
2522 	struct sockaddr *ca;
2523 	uchar_t *b;
2524 	char *frontspace = "";
2525 
2526 	/*
2527 	 * We assume we are called in tandem with client_name and the
2528 	 * format string looks like "...client %s%sblah blah..."
2529 	 *
2530 	 * If it's a Unix cred then client_name returned
2531 	 * a host name, so we need insert a space between host name
2532 	 * and IP address.
2533 	 */
2534 	if (req->rq_cred.oa_flavor == AUTH_UNIX)
2535 		frontspace = " ";
2536 
2537 	/*
2538 	 * Convert the caller's IP address to a dotted string
2539 	 */
2540 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2541 
2542 	if (ca->sa_family == AF_INET) {
2543 		b = (uchar_t *)&((struct sockaddr_in *)ca)->sin_addr;
2544 		(void) sprintf(buf, "%s(%d.%d.%d.%d) ", frontspace,
2545 		    b[0] & 0xFF, b[1] & 0xFF, b[2] & 0xFF, b[3] & 0xFF);
2546 	} else if (ca->sa_family == AF_INET6) {
2547 		struct sockaddr_in6 *sin6;
2548 		sin6 = (struct sockaddr_in6 *)ca;
2549 		(void) kinet_ntop6((uchar_t *)&sin6->sin6_addr,
2550 		    buf, INET6_ADDRSTRLEN);
2551 
2552 	} else {
2553 
2554 		/*
2555 		 * No IP address to print. If there was a host name
2556 		 * printed, then we print a space.
2557 		 */
2558 		(void) sprintf(buf, frontspace);
2559 	}
2560 
2561 	return (buf);
2562 }
2563 
2564 /*
2565  * NFS Server initialization routine.  This routine should only be called
2566  * once.  It performs the following tasks:
2567  *	- Call sub-initialization routines (localize access to variables)
2568  *	- Initialize all locks
2569  *	- initialize the version 3 write verifier
2570  */
2571 int
2572 nfs_srvinit(void)
2573 {
2574 	int error;
2575 
2576 	error = nfs_exportinit();
2577 	if (error != 0)
2578 		return (error);
2579 	error = rfs4_srvrinit();
2580 	if (error != 0) {
2581 		nfs_exportfini();
2582 		return (error);
2583 	}
2584 	rfs_srvrinit();
2585 	rfs3_srvrinit();
2586 	nfsauth_init();
2587 
2588 	/* Init the stuff to control start/stop */
2589 	nfs_server_upordown = NFS_SERVER_STOPPED;
2590 	mutex_init(&nfs_server_upordown_lock, NULL, MUTEX_DEFAULT, NULL);
2591 	cv_init(&nfs_server_upordown_cv, NULL, CV_DEFAULT, NULL);
2592 	mutex_init(&rdma_wait_mutex, NULL, MUTEX_DEFAULT, NULL);
2593 	cv_init(&rdma_wait_cv, NULL, CV_DEFAULT, NULL);
2594 
2595 	return (0);
2596 }
2597 
2598 /*
2599  * NFS Server finalization routine. This routine is called to cleanup the
2600  * initialization work previously performed if the NFS server module could
2601  * not be loaded correctly.
2602  */
2603 void
2604 nfs_srvfini(void)
2605 {
2606 	nfsauth_fini();
2607 	rfs3_srvrfini();
2608 	rfs_srvrfini();
2609 	nfs_exportfini();
2610 
2611 	mutex_destroy(&nfs_server_upordown_lock);
2612 	cv_destroy(&nfs_server_upordown_cv);
2613 	mutex_destroy(&rdma_wait_mutex);
2614 	cv_destroy(&rdma_wait_cv);
2615 }
2616 
2617 /*
2618  * Set up an iovec array of up to cnt pointers.
2619  */
2620 
2621 void
2622 mblk_to_iov(mblk_t *m, int cnt, struct iovec *iovp)
2623 {
2624 	while (m != NULL && cnt-- > 0) {
2625 		iovp->iov_base = (caddr_t)m->b_rptr;
2626 		iovp->iov_len = (m->b_wptr - m->b_rptr);
2627 		iovp++;
2628 		m = m->b_cont;
2629 	}
2630 }
2631 
2632 /*
2633  * Common code between NFS Version 2 and NFS Version 3 for the public
2634  * filehandle multicomponent lookups.
2635  */
2636 
2637 /*
2638  * Public filehandle evaluation of a multi-component lookup, following
2639  * symbolic links, if necessary. This may result in a vnode in another
2640  * filesystem, which is OK as long as the other filesystem is exported.
2641  *
2642  * Note that the exi will be set either to NULL or a new reference to the
2643  * exportinfo struct that corresponds to the vnode of the multi-component path.
2644  * It is the callers responsibility to release this reference.
2645  */
2646 int
2647 rfs_publicfh_mclookup(char *p, vnode_t *dvp, cred_t *cr, vnode_t **vpp,
2648     struct exportinfo **exi, struct sec_ol *sec)
2649 {
2650 	int pathflag;
2651 	vnode_t *mc_dvp = NULL;
2652 	vnode_t *realvp;
2653 	int error;
2654 
2655 	*exi = NULL;
2656 
2657 	/*
2658 	 * check if the given path is a url or native path. Since p is
2659 	 * modified by MCLpath(), it may be empty after returning from
2660 	 * there, and should be checked.
2661 	 */
2662 	if ((pathflag = MCLpath(&p)) == -1)
2663 		return (EIO);
2664 
2665 	/*
2666 	 * If pathflag is SECURITY_QUERY, turn the SEC_QUERY bit
2667 	 * on in sec->sec_flags. This bit will later serve as an
2668 	 * indication in makefh_ol() or makefh3_ol() to overload the
2669 	 * filehandle to contain the sec modes used by the server for
2670 	 * the path.
2671 	 */
2672 	if (pathflag == SECURITY_QUERY) {
2673 		if ((sec->sec_index = (uint_t)(*p)) > 0) {
2674 			sec->sec_flags |= SEC_QUERY;
2675 			p++;
2676 			if ((pathflag = MCLpath(&p)) == -1)
2677 				return (EIO);
2678 		} else {
2679 			cmn_err(CE_NOTE,
2680 			    "nfs_server: invalid security index %d, "
2681 			    "violating WebNFS SNEGO protocol.", sec->sec_index);
2682 			return (EIO);
2683 		}
2684 	}
2685 
2686 	if (p[0] == '\0') {
2687 		error = ENOENT;
2688 		goto publicfh_done;
2689 	}
2690 
2691 	error = rfs_pathname(p, &mc_dvp, vpp, dvp, cr, pathflag);
2692 
2693 	/*
2694 	 * If name resolves to "/" we get EINVAL since we asked for
2695 	 * the vnode of the directory that the file is in. Try again
2696 	 * with NULL directory vnode.
2697 	 */
2698 	if (error == EINVAL) {
2699 		error = rfs_pathname(p, NULL, vpp, dvp, cr, pathflag);
2700 		if (!error) {
2701 			ASSERT(*vpp != NULL);
2702 			if ((*vpp)->v_type == VDIR) {
2703 				VN_HOLD(*vpp);
2704 				mc_dvp = *vpp;
2705 			} else {
2706 				/*
2707 				 * This should not happen, the filesystem is
2708 				 * in an inconsistent state. Fail the lookup
2709 				 * at this point.
2710 				 */
2711 				VN_RELE(*vpp);
2712 				error = EINVAL;
2713 			}
2714 		}
2715 	}
2716 
2717 	if (error)
2718 		goto publicfh_done;
2719 
2720 	if (*vpp == NULL) {
2721 		error = ENOENT;
2722 		goto publicfh_done;
2723 	}
2724 
2725 	ASSERT(mc_dvp != NULL);
2726 	ASSERT(*vpp != NULL);
2727 
2728 	if ((*vpp)->v_type == VDIR) {
2729 		do {
2730 			/*
2731 			 * *vpp may be an AutoFS node, so we perform
2732 			 * a VOP_ACCESS() to trigger the mount of the intended
2733 			 * filesystem, so we can perform the lookup in the
2734 			 * intended filesystem.
2735 			 */
2736 			(void) VOP_ACCESS(*vpp, 0, 0, cr, NULL);
2737 
2738 			/*
2739 			 * If vnode is covered, get the
2740 			 * the topmost vnode.
2741 			 */
2742 			if (vn_mountedvfs(*vpp) != NULL) {
2743 				error = traverse(vpp);
2744 				if (error) {
2745 					VN_RELE(*vpp);
2746 					goto publicfh_done;
2747 				}
2748 			}
2749 
2750 			if (VOP_REALVP(*vpp, &realvp, NULL) == 0 &&
2751 			    realvp != *vpp) {
2752 				/*
2753 				 * If realvp is different from *vpp
2754 				 * then release our reference on *vpp, so that
2755 				 * the export access check be performed on the
2756 				 * real filesystem instead.
2757 				 */
2758 				VN_HOLD(realvp);
2759 				VN_RELE(*vpp);
2760 				*vpp = realvp;
2761 			} else {
2762 				break;
2763 			}
2764 		/* LINTED */
2765 		} while (TRUE);
2766 
2767 		/*
2768 		 * Let nfs_vptexi() figure what the real parent is.
2769 		 */
2770 		VN_RELE(mc_dvp);
2771 		mc_dvp = NULL;
2772 
2773 	} else {
2774 		/*
2775 		 * If vnode is covered, get the
2776 		 * the topmost vnode.
2777 		 */
2778 		if (vn_mountedvfs(mc_dvp) != NULL) {
2779 			error = traverse(&mc_dvp);
2780 			if (error) {
2781 				VN_RELE(*vpp);
2782 				goto publicfh_done;
2783 			}
2784 		}
2785 
2786 		if (VOP_REALVP(mc_dvp, &realvp, NULL) == 0 &&
2787 		    realvp != mc_dvp) {
2788 			/*
2789 			 * *vpp is a file, obtain realvp of the parent
2790 			 * directory vnode.
2791 			 */
2792 			VN_HOLD(realvp);
2793 			VN_RELE(mc_dvp);
2794 			mc_dvp = realvp;
2795 		}
2796 	}
2797 
2798 	/*
2799 	 * The pathname may take us from the public filesystem to another.
2800 	 * If that's the case then just set the exportinfo to the new export
2801 	 * and build filehandle for it. Thanks to per-access checking there's
2802 	 * no security issues with doing this. If the client is not allowed
2803 	 * access to this new export then it will get an access error when it
2804 	 * tries to use the filehandle
2805 	 */
2806 	if (error = nfs_check_vpexi(mc_dvp, *vpp, kcred, exi)) {
2807 		VN_RELE(*vpp);
2808 		goto publicfh_done;
2809 	}
2810 
2811 	/*
2812 	 * Not allowed access to pseudo exports.
2813 	 */
2814 	if (PSEUDO(*exi)) {
2815 		error = ENOENT;
2816 		VN_RELE(*vpp);
2817 		goto publicfh_done;
2818 	}
2819 
2820 	/*
2821 	 * Do a lookup for the index file. We know the index option doesn't
2822 	 * allow paths through handling in the share command, so mc_dvp will
2823 	 * be the parent for the index file vnode, if its present. Use
2824 	 * temporary pointers to preserve and reuse the vnode pointers of the
2825 	 * original directory in case there's no index file. Note that the
2826 	 * index file is a native path, and should not be interpreted by
2827 	 * the URL parser in rfs_pathname()
2828 	 */
2829 	if (((*exi)->exi_export.ex_flags & EX_INDEX) &&
2830 	    ((*vpp)->v_type == VDIR) && (pathflag == URLPATH)) {
2831 		vnode_t *tvp, *tmc_dvp;	/* temporary vnode pointers */
2832 
2833 		tmc_dvp = mc_dvp;
2834 		mc_dvp = tvp = *vpp;
2835 
2836 		error = rfs_pathname((*exi)->exi_export.ex_index, NULL, vpp,
2837 		    mc_dvp, cr, NATIVEPATH);
2838 
2839 		if (error == ENOENT) {
2840 			*vpp = tvp;
2841 			mc_dvp = tmc_dvp;
2842 			error = 0;
2843 		} else {	/* ok or error other than ENOENT */
2844 			if (tmc_dvp)
2845 				VN_RELE(tmc_dvp);
2846 			if (error)
2847 				goto publicfh_done;
2848 
2849 			/*
2850 			 * Found a valid vp for index "filename". Sanity check
2851 			 * for odd case where a directory is provided as index
2852 			 * option argument and leads us to another filesystem
2853 			 */
2854 
2855 			/* Release the reference on the old exi value */
2856 			ASSERT(*exi != NULL);
2857 			exi_rele(*exi);
2858 
2859 			if (error = nfs_check_vpexi(mc_dvp, *vpp, kcred, exi)) {
2860 				VN_RELE(*vpp);
2861 				goto publicfh_done;
2862 			}
2863 		}
2864 	}
2865 
2866 publicfh_done:
2867 	if (mc_dvp)
2868 		VN_RELE(mc_dvp);
2869 
2870 	return (error);
2871 }
2872 
2873 /*
2874  * Evaluate a multi-component path
2875  */
2876 int
2877 rfs_pathname(
2878 	char *path,			/* pathname to evaluate */
2879 	vnode_t **dirvpp,		/* ret for ptr to parent dir vnode */
2880 	vnode_t **compvpp,		/* ret for ptr to component vnode */
2881 	vnode_t *startdvp,		/* starting vnode */
2882 	cred_t *cr,			/* user's credential */
2883 	int pathflag)			/* flag to identify path, e.g. URL */
2884 {
2885 	char namebuf[TYPICALMAXPATHLEN];
2886 	struct pathname pn;
2887 	int error;
2888 
2889 	/*
2890 	 * If pathname starts with '/', then set startdvp to root.
2891 	 */
2892 	if (*path == '/') {
2893 		while (*path == '/')
2894 			path++;
2895 
2896 		startdvp = rootdir;
2897 	}
2898 
2899 	error = pn_get_buf(path, UIO_SYSSPACE, &pn, namebuf, sizeof (namebuf));
2900 	if (error == 0) {
2901 		/*
2902 		 * Call the URL parser for URL paths to modify the original
2903 		 * string to handle any '%' encoded characters that exist.
2904 		 * Done here to avoid an extra bcopy in the lookup.
2905 		 * We need to be careful about pathlen's. We know that
2906 		 * rfs_pathname() is called with a non-empty path. However,
2907 		 * it could be emptied due to the path simply being all /'s,
2908 		 * which is valid to proceed with the lookup, or due to the
2909 		 * URL parser finding an encoded null character at the
2910 		 * beginning of path which should not proceed with the lookup.
2911 		 */
2912 		if (pn.pn_pathlen != 0 && pathflag == URLPATH) {
2913 			URLparse(pn.pn_path);
2914 			if ((pn.pn_pathlen = strlen(pn.pn_path)) == 0)
2915 				return (ENOENT);
2916 		}
2917 		VN_HOLD(startdvp);
2918 		error = lookuppnvp(&pn, NULL, NO_FOLLOW, dirvpp, compvpp,
2919 		    rootdir, startdvp, cr);
2920 	}
2921 	if (error == ENAMETOOLONG) {
2922 		/*
2923 		 * This thread used a pathname > TYPICALMAXPATHLEN bytes long.
2924 		 */
2925 		if (error = pn_get(path, UIO_SYSSPACE, &pn))
2926 			return (error);
2927 		if (pn.pn_pathlen != 0 && pathflag == URLPATH) {
2928 			URLparse(pn.pn_path);
2929 			if ((pn.pn_pathlen = strlen(pn.pn_path)) == 0) {
2930 				pn_free(&pn);
2931 				return (ENOENT);
2932 			}
2933 		}
2934 		VN_HOLD(startdvp);
2935 		error = lookuppnvp(&pn, NULL, NO_FOLLOW, dirvpp, compvpp,
2936 		    rootdir, startdvp, cr);
2937 		pn_free(&pn);
2938 	}
2939 
2940 	return (error);
2941 }
2942 
2943 /*
2944  * Adapt the multicomponent lookup path depending on the pathtype
2945  */
2946 static int
2947 MCLpath(char **path)
2948 {
2949 	unsigned char c = (unsigned char)**path;
2950 
2951 	/*
2952 	 * If the MCL path is between 0x20 and 0x7E (graphic printable
2953 	 * character of the US-ASCII coded character set), its a URL path,
2954 	 * per RFC 1738.
2955 	 */
2956 	if (c >= 0x20 && c <= 0x7E)
2957 		return (URLPATH);
2958 
2959 	/*
2960 	 * If the first octet of the MCL path is not an ASCII character
2961 	 * then it must be interpreted as a tag value that describes the
2962 	 * format of the remaining octets of the MCL path.
2963 	 *
2964 	 * If the first octet of the MCL path is 0x81 it is a query
2965 	 * for the security info.
2966 	 */
2967 	switch (c) {
2968 	case 0x80:	/* native path, i.e. MCL via mount protocol */
2969 		(*path)++;
2970 		return (NATIVEPATH);
2971 	case 0x81:	/* security query */
2972 		(*path)++;
2973 		return (SECURITY_QUERY);
2974 	default:
2975 		return (-1);
2976 	}
2977 }
2978 
2979 #define	fromhex(c)  ((c >= '0' && c <= '9') ? (c - '0') : \
2980 			((c >= 'A' && c <= 'F') ? (c - 'A' + 10) :\
2981 			((c >= 'a' && c <= 'f') ? (c - 'a' + 10) : 0)))
2982 
2983 /*
2984  * The implementation of URLparse guarantees that the final string will
2985  * fit in the original one. Replaces '%' occurrences followed by 2 characters
2986  * with its corresponding hexadecimal character.
2987  */
2988 static void
2989 URLparse(char *str)
2990 {
2991 	char *p, *q;
2992 
2993 	p = q = str;
2994 	while (*p) {
2995 		*q = *p;
2996 		if (*p++ == '%') {
2997 			if (*p) {
2998 				*q = fromhex(*p) * 16;
2999 				p++;
3000 				if (*p) {
3001 					*q += fromhex(*p);
3002 					p++;
3003 				}
3004 			}
3005 		}
3006 		q++;
3007 	}
3008 	*q = '\0';
3009 }
3010 
3011 
3012 /*
3013  * Get the export information for the lookup vnode, and verify its
3014  * useable.
3015  */
3016 int
3017 nfs_check_vpexi(vnode_t *mc_dvp, vnode_t *vp, cred_t *cr,
3018     struct exportinfo **exi)
3019 {
3020 	int walk;
3021 	int error = 0;
3022 
3023 	*exi = nfs_vptoexi(mc_dvp, vp, cr, &walk, NULL, FALSE);
3024 	if (*exi == NULL)
3025 		error = EACCES;
3026 	else {
3027 		/*
3028 		 * If nosub is set for this export then
3029 		 * a lookup relative to the public fh
3030 		 * must not terminate below the
3031 		 * exported directory.
3032 		 */
3033 		if ((*exi)->exi_export.ex_flags & EX_NOSUB && walk > 0)
3034 			error = EACCES;
3035 	}
3036 
3037 	return (error);
3038 }
3039 
3040 /*
3041  * Do the main work of handling HA-NFSv4 Resource Group failover on
3042  * Sun Cluster.
3043  * We need to detect whether any RG admin paths have been added or removed,
3044  * and adjust resources accordingly.
3045  * Currently we're using a very inefficient algorithm, ~ 2 * O(n**2). In
3046  * order to scale, the list and array of paths need to be held in more
3047  * suitable data structures.
3048  */
3049 static void
3050 hanfsv4_failover(void)
3051 {
3052 	int i, start_grace, numadded_paths = 0;
3053 	char **added_paths = NULL;
3054 	rfs4_dss_path_t *dss_path;
3055 
3056 	/*
3057 	 * Note: currently, rfs4_dss_pathlist cannot be NULL, since
3058 	 * it will always include an entry for NFS4_DSS_VAR_DIR. If we
3059 	 * make the latter dynamically specified too, the following will
3060 	 * need to be adjusted.
3061 	 */
3062 
3063 	/*
3064 	 * First, look for removed paths: RGs that have been failed-over
3065 	 * away from this node.
3066 	 * Walk the "currently-serving" rfs4_dss_pathlist and, for each
3067 	 * path, check if it is on the "passed-in" rfs4_dss_newpaths array
3068 	 * from nfsd. If not, that RG path has been removed.
3069 	 *
3070 	 * Note that nfsd has sorted rfs4_dss_newpaths for us, and removed
3071 	 * any duplicates.
3072 	 */
3073 	dss_path = rfs4_dss_pathlist;
3074 	do {
3075 		int found = 0;
3076 		char *path = dss_path->path;
3077 
3078 		/* used only for non-HA so may not be removed */
3079 		if (strcmp(path, NFS4_DSS_VAR_DIR) == 0) {
3080 			dss_path = dss_path->next;
3081 			continue;
3082 		}
3083 
3084 		for (i = 0; i < rfs4_dss_numnewpaths; i++) {
3085 			int cmpret;
3086 			char *newpath = rfs4_dss_newpaths[i];
3087 
3088 			/*
3089 			 * Since nfsd has sorted rfs4_dss_newpaths for us,
3090 			 * once the return from strcmp is negative we know
3091 			 * we've passed the point where "path" should be,
3092 			 * and can stop searching: "path" has been removed.
3093 			 */
3094 			cmpret = strcmp(path, newpath);
3095 			if (cmpret < 0)
3096 				break;
3097 			if (cmpret == 0) {
3098 				found = 1;
3099 				break;
3100 			}
3101 		}
3102 
3103 		if (found == 0) {
3104 			unsigned index = dss_path->index;
3105 			rfs4_servinst_t *sip = dss_path->sip;
3106 			rfs4_dss_path_t *path_next = dss_path->next;
3107 
3108 			/*
3109 			 * This path has been removed.
3110 			 * We must clear out the servinst reference to
3111 			 * it, since it's now owned by another
3112 			 * node: we should not attempt to touch it.
3113 			 */
3114 			ASSERT(dss_path == sip->dss_paths[index]);
3115 			sip->dss_paths[index] = NULL;
3116 
3117 			/* remove from "currently-serving" list, and destroy */
3118 			remque(dss_path);
3119 			/* allow for NUL */
3120 			kmem_free(dss_path->path, strlen(dss_path->path) + 1);
3121 			kmem_free(dss_path, sizeof (rfs4_dss_path_t));
3122 
3123 			dss_path = path_next;
3124 		} else {
3125 			/* path was found; not removed */
3126 			dss_path = dss_path->next;
3127 		}
3128 	} while (dss_path != rfs4_dss_pathlist);
3129 
3130 	/*
3131 	 * Now, look for added paths: RGs that have been failed-over
3132 	 * to this node.
3133 	 * Walk the "passed-in" rfs4_dss_newpaths array from nfsd and,
3134 	 * for each path, check if it is on the "currently-serving"
3135 	 * rfs4_dss_pathlist. If not, that RG path has been added.
3136 	 *
3137 	 * Note: we don't do duplicate detection here; nfsd does that for us.
3138 	 *
3139 	 * Note: numadded_paths <= rfs4_dss_numnewpaths, which gives us
3140 	 * an upper bound for the size needed for added_paths[numadded_paths].
3141 	 */
3142 
3143 	/* probably more space than we need, but guaranteed to be enough */
3144 	if (rfs4_dss_numnewpaths > 0) {
3145 		size_t sz = rfs4_dss_numnewpaths * sizeof (char *);
3146 		added_paths = kmem_zalloc(sz, KM_SLEEP);
3147 	}
3148 
3149 	/* walk the "passed-in" rfs4_dss_newpaths array from nfsd */
3150 	for (i = 0; i < rfs4_dss_numnewpaths; i++) {
3151 		int found = 0;
3152 		char *newpath = rfs4_dss_newpaths[i];
3153 
3154 		dss_path = rfs4_dss_pathlist;
3155 		do {
3156 			char *path = dss_path->path;
3157 
3158 			/* used only for non-HA */
3159 			if (strcmp(path, NFS4_DSS_VAR_DIR) == 0) {
3160 				dss_path = dss_path->next;
3161 				continue;
3162 			}
3163 
3164 			if (strncmp(path, newpath, strlen(path)) == 0) {
3165 				found = 1;
3166 				break;
3167 			}
3168 
3169 			dss_path = dss_path->next;
3170 		} while (dss_path != rfs4_dss_pathlist);
3171 
3172 		if (found == 0) {
3173 			added_paths[numadded_paths] = newpath;
3174 			numadded_paths++;
3175 		}
3176 	}
3177 
3178 	/* did we find any added paths? */
3179 	if (numadded_paths > 0) {
3180 		/* create a new server instance, and start its grace period */
3181 		start_grace = 1;
3182 		rfs4_servinst_create(start_grace, numadded_paths, added_paths);
3183 
3184 		/* read in the stable storage state from these paths */
3185 		rfs4_dss_readstate(numadded_paths, added_paths);
3186 
3187 		/*
3188 		 * Multiple failovers during a grace period will cause
3189 		 * clients of the same resource group to be partitioned
3190 		 * into different server instances, with different
3191 		 * grace periods.  Since clients of the same resource
3192 		 * group must be subject to the same grace period,
3193 		 * we need to reset all currently active grace periods.
3194 		 */
3195 		rfs4_grace_reset_all();
3196 	}
3197 
3198 	if (rfs4_dss_numnewpaths > 0)
3199 		kmem_free(added_paths, rfs4_dss_numnewpaths * sizeof (char *));
3200 }
3201 
3202 /*
3203  * Used by NFSv3 and NFSv4 server to query label of
3204  * a pathname component during lookup/access ops.
3205  */
3206 ts_label_t *
3207 nfs_getflabel(vnode_t *vp, struct exportinfo *exi)
3208 {
3209 	zone_t *zone;
3210 	ts_label_t *zone_label;
3211 	char *path;
3212 
3213 	mutex_enter(&vp->v_lock);
3214 	if (vp->v_path != vn_vpath_empty) {
3215 		zone = zone_find_by_any_path(vp->v_path, B_FALSE);
3216 		mutex_exit(&vp->v_lock);
3217 	} else {
3218 		/*
3219 		 * v_path not cached. Fall back on pathname of exported
3220 		 * file system as we rely on pathname from which we can
3221 		 * derive a label. The exported file system portion of
3222 		 * path is sufficient to obtain a label.
3223 		 */
3224 		path = exi->exi_export.ex_path;
3225 		if (path == NULL) {
3226 			mutex_exit(&vp->v_lock);
3227 			return (NULL);
3228 		}
3229 		zone = zone_find_by_any_path(path, B_FALSE);
3230 		mutex_exit(&vp->v_lock);
3231 	}
3232 	/*
3233 	 * Caller has verified that the file is either
3234 	 * exported or visible. So if the path falls in
3235 	 * global zone, admin_low is returned; otherwise
3236 	 * the zone's label is returned.
3237 	 */
3238 	zone_label = zone->zone_slabel;
3239 	label_hold(zone_label);
3240 	zone_rele(zone);
3241 	return (zone_label);
3242 }
3243 
3244 /*
3245  * TX NFS routine used by NFSv3 and NFSv4 to do label check
3246  * on client label and server's file object lable.
3247  */
3248 boolean_t
3249 do_rfs_label_check(bslabel_t *clabel, vnode_t *vp, int flag,
3250     struct exportinfo *exi)
3251 {
3252 	bslabel_t *slabel;
3253 	ts_label_t *tslabel;
3254 	boolean_t result;
3255 
3256 	if ((tslabel = nfs_getflabel(vp, exi)) == NULL) {
3257 		return (B_FALSE);
3258 	}
3259 	slabel = label2bslabel(tslabel);
3260 	DTRACE_PROBE4(tx__rfs__log__info__labelcheck, char *,
3261 	    "comparing server's file label(1) with client label(2) (vp(3))",
3262 	    bslabel_t *, slabel, bslabel_t *, clabel, vnode_t *, vp);
3263 
3264 	if (flag == EQUALITY_CHECK)
3265 		result = blequal(clabel, slabel);
3266 	else
3267 		result = bldominates(clabel, slabel);
3268 	label_rele(tslabel);
3269 	return (result);
3270 }
3271 
3272 /*
3273  * Callback function to return the loaned buffers.
3274  * Calls VOP_RETZCBUF() only after all uio_iov[]
3275  * buffers are returned. nu_ref maintains the count.
3276  */
3277 void
3278 rfs_free_xuio(void *free_arg)
3279 {
3280 	uint_t ref;
3281 	nfs_xuio_t *nfsuiop = (nfs_xuio_t *)free_arg;
3282 
3283 	ref = atomic_dec_uint_nv(&nfsuiop->nu_ref);
3284 
3285 	/*
3286 	 * Call VOP_RETZCBUF() only when all the iov buffers
3287 	 * are sent OTW.
3288 	 */
3289 	if (ref != 0)
3290 		return;
3291 
3292 	if (((uio_t *)nfsuiop)->uio_extflg & UIO_XUIO) {
3293 		(void) VOP_RETZCBUF(nfsuiop->nu_vp, (xuio_t *)free_arg, NULL,
3294 		    NULL);
3295 		VN_RELE(nfsuiop->nu_vp);
3296 	}
3297 
3298 	kmem_cache_free(nfs_xuio_cache, free_arg);
3299 }
3300 
3301 xuio_t *
3302 rfs_setup_xuio(vnode_t *vp)
3303 {
3304 	nfs_xuio_t *nfsuiop;
3305 
3306 	nfsuiop = kmem_cache_alloc(nfs_xuio_cache, KM_SLEEP);
3307 
3308 	bzero(nfsuiop, sizeof (nfs_xuio_t));
3309 	nfsuiop->nu_vp = vp;
3310 
3311 	/*
3312 	 * ref count set to 1. more may be added
3313 	 * if multiple mblks refer to multiple iov's.
3314 	 * This is done in uio_to_mblk().
3315 	 */
3316 
3317 	nfsuiop->nu_ref = 1;
3318 
3319 	nfsuiop->nu_frtn.free_func = rfs_free_xuio;
3320 	nfsuiop->nu_frtn.free_arg = (char *)nfsuiop;
3321 
3322 	nfsuiop->nu_uio.xu_type = UIOTYPE_ZEROCOPY;
3323 
3324 	return (&nfsuiop->nu_uio);
3325 }
3326 
3327 mblk_t *
3328 uio_to_mblk(uio_t *uiop)
3329 {
3330 	struct iovec *iovp;
3331 	int i;
3332 	mblk_t *mp, *mp1;
3333 	nfs_xuio_t *nfsuiop = (nfs_xuio_t *)uiop;
3334 
3335 	if (uiop->uio_iovcnt == 0)
3336 		return (NULL);
3337 
3338 	iovp = uiop->uio_iov;
3339 	mp = mp1 = esballoca((uchar_t *)iovp->iov_base, iovp->iov_len,
3340 	    BPRI_MED, &nfsuiop->nu_frtn);
3341 	ASSERT(mp != NULL);
3342 
3343 	mp->b_wptr += iovp->iov_len;
3344 	mp->b_datap->db_type = M_DATA;
3345 
3346 	for (i = 1; i < uiop->uio_iovcnt; i++) {
3347 		iovp = (uiop->uio_iov + i);
3348 
3349 		mp1->b_cont = esballoca(
3350 		    (uchar_t *)iovp->iov_base, iovp->iov_len, BPRI_MED,
3351 		    &nfsuiop->nu_frtn);
3352 
3353 		mp1 = mp1->b_cont;
3354 		ASSERT(mp1 != NULL);
3355 		mp1->b_wptr += iovp->iov_len;
3356 		mp1->b_datap->db_type = M_DATA;
3357 	}
3358 
3359 	nfsuiop->nu_ref = uiop->uio_iovcnt;
3360 
3361 	return (mp);
3362 }
3363 
3364 /*
3365  * Allocate memory to hold data for a read request of len bytes.
3366  *
3367  * We don't allocate buffers greater than kmem_max_cached in size to avoid
3368  * allocating memory from the kmem_oversized arena.  If we allocate oversized
3369  * buffers, we incur heavy cross-call activity when freeing these large buffers
3370  * in the TCP receive path. Note that we can't set b_wptr here since the
3371  * length of the data returned may differ from the length requested when
3372  * reading the end of a file; we set b_wptr in rfs_rndup_mblks() once the
3373  * length of the read is known.
3374  */
3375 mblk_t *
3376 rfs_read_alloc(uint_t len, struct iovec **iov, int *iovcnt)
3377 {
3378 	struct iovec *iovarr;
3379 	mblk_t *mp, **mpp = &mp;
3380 	size_t mpsize;
3381 	uint_t remain = len;
3382 	int i, err = 0;
3383 
3384 	*iovcnt = howmany(len, kmem_max_cached);
3385 
3386 	iovarr = kmem_alloc(*iovcnt * sizeof (struct iovec), KM_SLEEP);
3387 	*iov = iovarr;
3388 
3389 	for (i = 0; i < *iovcnt; remain -= mpsize, i++) {
3390 		ASSERT(remain <= len);
3391 		/*
3392 		 * We roundup the size we allocate to a multiple of
3393 		 * BYTES_PER_XDR_UNIT (4 bytes) so that the call to
3394 		 * xdrmblk_putmblk() never fails.
3395 		 */
3396 		ASSERT(kmem_max_cached % BYTES_PER_XDR_UNIT == 0);
3397 		mpsize = MIN(kmem_max_cached, remain);
3398 		*mpp = allocb_wait(RNDUP(mpsize), BPRI_MED, STR_NOSIG, &err);
3399 		ASSERT(*mpp != NULL);
3400 		ASSERT(err == 0);
3401 
3402 		iovarr[i].iov_base = (caddr_t)(*mpp)->b_rptr;
3403 		iovarr[i].iov_len = mpsize;
3404 		mpp = &(*mpp)->b_cont;
3405 	}
3406 	return (mp);
3407 }
3408 
3409 void
3410 rfs_rndup_mblks(mblk_t *mp, uint_t len, int buf_loaned)
3411 {
3412 	int i;
3413 	int alloc_err = 0;
3414 	mblk_t *rmp;
3415 	uint_t mpsize, remainder;
3416 
3417 	remainder = P2NPHASE(len, BYTES_PER_XDR_UNIT);
3418 
3419 	/*
3420 	 * Non copy-reduction case.  This function assumes that blocks were
3421 	 * allocated in multiples of BYTES_PER_XDR_UNIT bytes, which makes this
3422 	 * padding safe without bounds checking.
3423 	 */
3424 	if (!buf_loaned) {
3425 		/*
3426 		 * Set the size of each mblk in the chain until we've consumed
3427 		 * the specified length for all but the last one.
3428 		 */
3429 		while ((mpsize = MBLKSIZE(mp)) < len) {
3430 			ASSERT(mpsize % BYTES_PER_XDR_UNIT == 0);
3431 			mp->b_wptr += mpsize;
3432 			len -= mpsize;
3433 			mp = mp->b_cont;
3434 			ASSERT(mp != NULL);
3435 		}
3436 
3437 		ASSERT(len + remainder <= mpsize);
3438 		mp->b_wptr += len;
3439 		for (i = 0; i < remainder; i++)
3440 			*mp->b_wptr++ = '\0';
3441 		return;
3442 	}
3443 
3444 	/*
3445 	 * No remainder mblk required.
3446 	 */
3447 	if (remainder == 0)
3448 		return;
3449 
3450 	/*
3451 	 * Get to the last mblk in the chain.
3452 	 */
3453 	while (mp->b_cont != NULL)
3454 		mp = mp->b_cont;
3455 
3456 	/*
3457 	 * In case of copy-reduction mblks, the size of the mblks are fixed
3458 	 * and are of the size of the loaned buffers.  Allocate a remainder
3459 	 * mblk and chain it to the data buffers. This is sub-optimal, but not
3460 	 * expected to happen commonly.
3461 	 */
3462 	rmp = allocb_wait(remainder, BPRI_MED, STR_NOSIG, &alloc_err);
3463 	ASSERT(rmp != NULL);
3464 	ASSERT(alloc_err == 0);
3465 
3466 	for (i = 0; i < remainder; i++)
3467 		*rmp->b_wptr++ = '\0';
3468 
3469 	rmp->b_datap->db_type = M_DATA;
3470 	mp->b_cont = rmp;
3471 }
3472