xref: /titanic_44/usr/src/uts/common/fs/nfs/nfs_server.c (revision ed22c7109fc5dd9e1b7a5d0333bdc7ad2718e2ab)
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  */
25 
26 /*
27  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
28  *	All rights reserved.
29  *	Use is subject to license terms.
30  */
31 
32 #include <sys/param.h>
33 #include <sys/types.h>
34 #include <sys/systm.h>
35 #include <sys/cred.h>
36 #include <sys/proc.h>
37 #include <sys/user.h>
38 #include <sys/buf.h>
39 #include <sys/vfs.h>
40 #include <sys/vnode.h>
41 #include <sys/pathname.h>
42 #include <sys/uio.h>
43 #include <sys/file.h>
44 #include <sys/stat.h>
45 #include <sys/errno.h>
46 #include <sys/socket.h>
47 #include <sys/sysmacros.h>
48 #include <sys/siginfo.h>
49 #include <sys/tiuser.h>
50 #include <sys/statvfs.h>
51 #include <sys/stream.h>
52 #include <sys/strsubr.h>
53 #include <sys/stropts.h>
54 #include <sys/timod.h>
55 #include <sys/t_kuser.h>
56 #include <sys/kmem.h>
57 #include <sys/kstat.h>
58 #include <sys/dirent.h>
59 #include <sys/cmn_err.h>
60 #include <sys/debug.h>
61 #include <sys/unistd.h>
62 #include <sys/vtrace.h>
63 #include <sys/mode.h>
64 #include <sys/acl.h>
65 #include <sys/sdt.h>
66 
67 #include <rpc/types.h>
68 #include <rpc/auth.h>
69 #include <rpc/auth_unix.h>
70 #include <rpc/auth_des.h>
71 #include <rpc/svc.h>
72 #include <rpc/xdr.h>
73 #include <rpc/rpc_rdma.h>
74 
75 #include <nfs/nfs.h>
76 #include <nfs/export.h>
77 #include <nfs/nfssys.h>
78 #include <nfs/nfs_clnt.h>
79 #include <nfs/nfs_acl.h>
80 #include <nfs/nfs_log.h>
81 #include <nfs/nfs_cmd.h>
82 #include <nfs/lm.h>
83 #include <nfs/nfs_dispatch.h>
84 #include <nfs/nfs4_drc.h>
85 
86 #include <sys/modctl.h>
87 #include <sys/cladm.h>
88 #include <sys/clconf.h>
89 
90 #include <sys/tsol/label.h>
91 
92 #define	MAXHOST 32
93 const char *kinet_ntop6(uchar_t *, char *, size_t);
94 
95 /*
96  * Module linkage information.
97  */
98 
99 static struct modlmisc modlmisc = {
100 	&mod_miscops, "NFS server module"
101 };
102 
103 static struct modlinkage modlinkage = {
104 	MODREV_1, (void *)&modlmisc, NULL
105 };
106 
107 kmem_cache_t *nfs_xuio_cache;
108 int nfs_loaned_buffers = 0;
109 
110 int
111 _init(void)
112 {
113 	int status;
114 
115 	if ((status = nfs_srvinit()) != 0) {
116 		cmn_err(CE_WARN, "_init: nfs_srvinit failed");
117 		return (status);
118 	}
119 
120 	status = mod_install((struct modlinkage *)&modlinkage);
121 	if (status != 0) {
122 		/*
123 		 * Could not load module, cleanup previous
124 		 * initialization work.
125 		 */
126 		nfs_srvfini();
127 	}
128 
129 	/*
130 	 * Initialise some placeholders for nfssys() calls. These have
131 	 * to be declared by the nfs module, since that handles nfssys()
132 	 * calls - also used by NFS clients - but are provided by this
133 	 * nfssrv module. These also then serve as confirmation to the
134 	 * relevant code in nfs that nfssrv has been loaded, as they're
135 	 * initially NULL.
136 	 */
137 	nfs_srv_quiesce_func = nfs_srv_quiesce_all;
138 	nfs_srv_dss_func = rfs4_dss_setpaths;
139 
140 	/* setup DSS paths here; must be done before initial server startup */
141 	rfs4_dss_paths = rfs4_dss_oldpaths = NULL;
142 
143 	/* initialize the copy reduction caches */
144 
145 	nfs_xuio_cache = kmem_cache_create("nfs_xuio_cache",
146 	    sizeof (nfs_xuio_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
147 
148 	return (status);
149 }
150 
151 int
152 _fini()
153 {
154 	return (EBUSY);
155 }
156 
157 int
158 _info(struct modinfo *modinfop)
159 {
160 	return (mod_info(&modlinkage, modinfop));
161 }
162 
163 /*
164  * PUBLICFH_CHECK() checks if the dispatch routine supports
165  * RPC_PUBLICFH_OK, if the filesystem is exported public, and if the
166  * incoming request is using the public filehandle. The check duplicates
167  * the exportmatch() call done in checkexport(), and we should consider
168  * modifying those routines to avoid the duplication. For now, we optimize
169  * by calling exportmatch() only after checking that the dispatch routine
170  * supports RPC_PUBLICFH_OK, and if the filesystem is explicitly exported
171  * public (i.e., not the placeholder).
172  */
173 #define	PUBLICFH_CHECK(disp, exi, fsid, xfid) \
174 		((disp->dis_flags & RPC_PUBLICFH_OK) && \
175 		((exi->exi_export.ex_flags & EX_PUBLIC) || \
176 		(exi == exi_public && exportmatch(exi_root, \
177 		fsid, xfid))))
178 
179 static void	nfs_srv_shutdown_all(int);
180 static void	rfs4_server_start(int);
181 static void	nullfree(void);
182 static void	rfs_dispatch(struct svc_req *, SVCXPRT *);
183 static void	acl_dispatch(struct svc_req *, SVCXPRT *);
184 static void	common_dispatch(struct svc_req *, SVCXPRT *,
185 		rpcvers_t, rpcvers_t, char *,
186 		struct rpc_disptable *);
187 static void	hanfsv4_failover(void);
188 static	int	checkauth(struct exportinfo *, struct svc_req *, cred_t *, int,
189 			bool_t);
190 static char	*client_name(struct svc_req *req);
191 static char	*client_addr(struct svc_req *req, char *buf);
192 extern	int	sec_svc_getcred(struct svc_req *, cred_t *cr, char **, int *);
193 extern	bool_t	sec_svc_inrootlist(int, caddr_t, int, caddr_t *);
194 
195 #define	NFSLOG_COPY_NETBUF(exi, xprt, nb)	{		\
196 	(nb)->maxlen = (xprt)->xp_rtaddr.maxlen;		\
197 	(nb)->len = (xprt)->xp_rtaddr.len;			\
198 	(nb)->buf = kmem_alloc((nb)->len, KM_SLEEP);		\
199 	bcopy((xprt)->xp_rtaddr.buf, (nb)->buf, (nb)->len);	\
200 	}
201 
202 /*
203  * Public Filehandle common nfs routines
204  */
205 static int	MCLpath(char **);
206 static void	URLparse(char *);
207 
208 /*
209  * NFS callout table.
210  * This table is used by svc_getreq() to dispatch a request with
211  * a given prog/vers pair to an appropriate service provider
212  * dispatch routine.
213  *
214  * NOTE: ordering is relied upon below when resetting the version min/max
215  * for NFS_PROGRAM.  Careful, if this is ever changed.
216  */
217 static SVC_CALLOUT __nfs_sc_clts[] = {
218 	{ NFS_PROGRAM,	   NFS_VERSMIN,	    NFS_VERSMAX,	rfs_dispatch },
219 	{ NFS_ACL_PROGRAM, NFS_ACL_VERSMIN, NFS_ACL_VERSMAX,	acl_dispatch }
220 };
221 
222 static SVC_CALLOUT_TABLE nfs_sct_clts = {
223 	sizeof (__nfs_sc_clts) / sizeof (__nfs_sc_clts[0]), FALSE,
224 	__nfs_sc_clts
225 };
226 
227 static SVC_CALLOUT __nfs_sc_cots[] = {
228 	{ NFS_PROGRAM,	   NFS_VERSMIN,	    NFS_VERSMAX,	rfs_dispatch },
229 	{ NFS_ACL_PROGRAM, NFS_ACL_VERSMIN, NFS_ACL_VERSMAX,	acl_dispatch }
230 };
231 
232 static SVC_CALLOUT_TABLE nfs_sct_cots = {
233 	sizeof (__nfs_sc_cots) / sizeof (__nfs_sc_cots[0]), FALSE, __nfs_sc_cots
234 };
235 
236 static SVC_CALLOUT __nfs_sc_rdma[] = {
237 	{ NFS_PROGRAM,	   NFS_VERSMIN,	    NFS_VERSMAX,	rfs_dispatch },
238 	{ NFS_ACL_PROGRAM, NFS_ACL_VERSMIN, NFS_ACL_VERSMAX,	acl_dispatch }
239 };
240 
241 static SVC_CALLOUT_TABLE nfs_sct_rdma = {
242 	sizeof (__nfs_sc_rdma) / sizeof (__nfs_sc_rdma[0]), FALSE, __nfs_sc_rdma
243 };
244 rpcvers_t nfs_versmin = NFS_VERSMIN_DEFAULT;
245 rpcvers_t nfs_versmax = NFS_VERSMAX_DEFAULT;
246 
247 /*
248  * Used to track the state of the server so that initialization
249  * can be done properly.
250  */
251 typedef enum {
252 	NFS_SERVER_STOPPED,	/* server state destroyed */
253 	NFS_SERVER_STOPPING,	/* server state being destroyed */
254 	NFS_SERVER_RUNNING,
255 	NFS_SERVER_QUIESCED,	/* server state preserved */
256 	NFS_SERVER_OFFLINE	/* server pool offline */
257 } nfs_server_running_t;
258 
259 static nfs_server_running_t nfs_server_upordown;
260 static kmutex_t nfs_server_upordown_lock;
261 static	kcondvar_t nfs_server_upordown_cv;
262 
263 /*
264  * DSS: distributed stable storage
265  * lists of all DSS paths: current, and before last warmstart
266  */
267 nvlist_t *rfs4_dss_paths, *rfs4_dss_oldpaths;
268 
269 int rfs4_dispatch(struct rpcdisp *, struct svc_req *, SVCXPRT *, char *);
270 bool_t rfs4_minorvers_mismatch(struct svc_req *, SVCXPRT *, void *);
271 
272 /*
273  * RDMA wait variables.
274  */
275 static kcondvar_t rdma_wait_cv;
276 static kmutex_t rdma_wait_mutex;
277 
278 /*
279  * Will be called at the point the server pool is being unregistered
280  * from the pool list. From that point onwards, the pool is waiting
281  * to be drained and as such the server state is stale and pertains
282  * to the old instantiation of the NFS server pool.
283  */
284 void
285 nfs_srv_offline(void)
286 {
287 	mutex_enter(&nfs_server_upordown_lock);
288 	if (nfs_server_upordown == NFS_SERVER_RUNNING) {
289 		nfs_server_upordown = NFS_SERVER_OFFLINE;
290 	}
291 	mutex_exit(&nfs_server_upordown_lock);
292 }
293 
294 /*
295  * Will be called at the point the server pool is being destroyed so
296  * all transports have been closed and no service threads are in
297  * existence.
298  *
299  * If we quiesce the server, we're shutting it down without destroying the
300  * server state. This allows it to warm start subsequently.
301  */
302 void
303 nfs_srv_stop_all(void)
304 {
305 	int quiesce = 0;
306 	nfs_srv_shutdown_all(quiesce);
307 }
308 
309 /*
310  * This alternative shutdown routine can be requested via nfssys()
311  */
312 void
313 nfs_srv_quiesce_all(void)
314 {
315 	int quiesce = 1;
316 	nfs_srv_shutdown_all(quiesce);
317 }
318 
319 static void
320 nfs_srv_shutdown_all(int quiesce) {
321 	mutex_enter(&nfs_server_upordown_lock);
322 	if (quiesce) {
323 		if (nfs_server_upordown == NFS_SERVER_RUNNING ||
324 			nfs_server_upordown == NFS_SERVER_OFFLINE) {
325 			nfs_server_upordown = NFS_SERVER_QUIESCED;
326 			cv_signal(&nfs_server_upordown_cv);
327 
328 			/* reset DSS state, for subsequent warm restart */
329 			rfs4_dss_numnewpaths = 0;
330 			rfs4_dss_newpaths = NULL;
331 
332 			cmn_err(CE_NOTE, "nfs_server: server is now quiesced; "
333 			    "NFSv4 state has been preserved");
334 		}
335 	} else {
336 		if (nfs_server_upordown == NFS_SERVER_OFFLINE) {
337 			nfs_server_upordown = NFS_SERVER_STOPPING;
338 			mutex_exit(&nfs_server_upordown_lock);
339 			rfs4_state_fini();
340 			rfs4_fini_drc(nfs4_drc);
341 			mutex_enter(&nfs_server_upordown_lock);
342 			nfs_server_upordown = NFS_SERVER_STOPPED;
343 			cv_signal(&nfs_server_upordown_cv);
344 		}
345 	}
346 	mutex_exit(&nfs_server_upordown_lock);
347 }
348 
349 static int
350 nfs_srv_set_sc_versions(struct file *fp, SVC_CALLOUT_TABLE **sctpp,
351 			rpcvers_t versmin, rpcvers_t versmax)
352 {
353 	struct strioctl strioc;
354 	struct T_info_ack tinfo;
355 	int		error, retval;
356 
357 	/*
358 	 * Find out what type of transport this is.
359 	 */
360 	strioc.ic_cmd = TI_GETINFO;
361 	strioc.ic_timout = -1;
362 	strioc.ic_len = sizeof (tinfo);
363 	strioc.ic_dp = (char *)&tinfo;
364 	tinfo.PRIM_type = T_INFO_REQ;
365 
366 	error = strioctl(fp->f_vnode, I_STR, (intptr_t)&strioc, 0, K_TO_K,
367 	    CRED(), &retval);
368 	if (error || retval)
369 		return (error);
370 
371 	/*
372 	 * Based on our query of the transport type...
373 	 *
374 	 * Reset the min/max versions based on the caller's request
375 	 * NOTE: This assumes that NFS_PROGRAM is first in the array!!
376 	 * And the second entry is the NFS_ACL_PROGRAM.
377 	 */
378 	switch (tinfo.SERV_type) {
379 	case T_CLTS:
380 		if (versmax == NFS_V4)
381 			return (EINVAL);
382 		__nfs_sc_clts[0].sc_versmin = versmin;
383 		__nfs_sc_clts[0].sc_versmax = versmax;
384 		__nfs_sc_clts[1].sc_versmin = versmin;
385 		__nfs_sc_clts[1].sc_versmax = versmax;
386 		*sctpp = &nfs_sct_clts;
387 		break;
388 	case T_COTS:
389 	case T_COTS_ORD:
390 		__nfs_sc_cots[0].sc_versmin = versmin;
391 		__nfs_sc_cots[0].sc_versmax = versmax;
392 		/* For the NFS_ACL program, check the max version */
393 		if (versmax > NFS_ACL_VERSMAX)
394 			versmax = NFS_ACL_VERSMAX;
395 		__nfs_sc_cots[1].sc_versmin = versmin;
396 		__nfs_sc_cots[1].sc_versmax = versmax;
397 		*sctpp = &nfs_sct_cots;
398 		break;
399 	default:
400 		error = EINVAL;
401 	}
402 
403 	return (error);
404 }
405 
406 /*
407  * NFS Server system call.
408  * Does all of the work of running a NFS server.
409  * uap->fd is the fd of an open transport provider
410  */
411 int
412 nfs_svc(struct nfs_svc_args *arg, model_t model)
413 {
414 	file_t *fp;
415 	SVCMASTERXPRT *xprt;
416 	int error;
417 	int readsize;
418 	char buf[KNC_STRSIZE];
419 	size_t len;
420 	STRUCT_HANDLE(nfs_svc_args, uap);
421 	struct netbuf addrmask;
422 	SVC_CALLOUT_TABLE *sctp = NULL;
423 
424 #ifdef lint
425 	model = model;		/* STRUCT macros don't always refer to it */
426 #endif
427 
428 	STRUCT_SET_HANDLE(uap, model, arg);
429 
430 	/* Check privileges in nfssys() */
431 
432 	if ((fp = getf(STRUCT_FGET(uap, fd))) == NULL)
433 		return (EBADF);
434 
435 	/*
436 	 * Set read buffer size to rsize
437 	 * and add room for RPC headers.
438 	 */
439 	readsize = nfs3tsize() + (RPC_MAXDATASIZE - NFS_MAXDATA);
440 	if (readsize < RPC_MAXDATASIZE)
441 		readsize = RPC_MAXDATASIZE;
442 
443 	error = copyinstr((const char *)STRUCT_FGETP(uap, netid), buf,
444 	    KNC_STRSIZE, &len);
445 	if (error) {
446 		releasef(STRUCT_FGET(uap, fd));
447 		return (error);
448 	}
449 
450 	addrmask.len = STRUCT_FGET(uap, addrmask.len);
451 	addrmask.maxlen = STRUCT_FGET(uap, addrmask.maxlen);
452 	addrmask.buf = kmem_alloc(addrmask.maxlen, KM_SLEEP);
453 	error = copyin(STRUCT_FGETP(uap, addrmask.buf), addrmask.buf,
454 	    addrmask.len);
455 	if (error) {
456 		releasef(STRUCT_FGET(uap, fd));
457 		kmem_free(addrmask.buf, addrmask.maxlen);
458 		return (error);
459 	}
460 
461 	nfs_versmin = STRUCT_FGET(uap, versmin);
462 	nfs_versmax = STRUCT_FGET(uap, versmax);
463 
464 	/* Double check the vers min/max ranges */
465 	if ((nfs_versmin > nfs_versmax) ||
466 	    (nfs_versmin < NFS_VERSMIN) ||
467 	    (nfs_versmax > NFS_VERSMAX)) {
468 		nfs_versmin = NFS_VERSMIN_DEFAULT;
469 		nfs_versmax = NFS_VERSMAX_DEFAULT;
470 	}
471 
472 	if (error =
473 	    nfs_srv_set_sc_versions(fp, &sctp, nfs_versmin, nfs_versmax)) {
474 		releasef(STRUCT_FGET(uap, fd));
475 		kmem_free(addrmask.buf, addrmask.maxlen);
476 		return (error);
477 	}
478 
479 	/* Initialize nfsv4 server */
480 	if (nfs_versmax == (rpcvers_t)NFS_V4)
481 		rfs4_server_start(STRUCT_FGET(uap, delegation));
482 
483 	/* Create a transport handle. */
484 	error = svc_tli_kcreate(fp, readsize, buf, &addrmask, &xprt,
485 	    sctp, NULL, NFS_SVCPOOL_ID, TRUE);
486 
487 	if (error)
488 		kmem_free(addrmask.buf, addrmask.maxlen);
489 
490 	releasef(STRUCT_FGET(uap, fd));
491 
492 	/* HA-NFSv4: save the cluster nodeid */
493 	if (cluster_bootflags & CLUSTER_BOOTED)
494 		lm_global_nlmid = clconf_get_nodeid();
495 
496 	return (error);
497 }
498 
499 static void
500 rfs4_server_start(int nfs4_srv_delegation)
501 {
502 	/*
503 	 * Determine if the server has previously been "started" and
504 	 * if not, do the per instance initialization
505 	 */
506 	mutex_enter(&nfs_server_upordown_lock);
507 
508 	if (nfs_server_upordown != NFS_SERVER_RUNNING) {
509 		/* Do we need to stop and wait on the previous server? */
510 		while (nfs_server_upordown == NFS_SERVER_STOPPING ||
511 		    nfs_server_upordown == NFS_SERVER_OFFLINE)
512 			cv_wait(&nfs_server_upordown_cv,
513 			    &nfs_server_upordown_lock);
514 
515 		if (nfs_server_upordown != NFS_SERVER_RUNNING) {
516 			(void) svc_pool_control(NFS_SVCPOOL_ID,
517 			    SVCPSET_UNREGISTER_PROC, (void *)&nfs_srv_offline);
518 			(void) svc_pool_control(NFS_SVCPOOL_ID,
519 			    SVCPSET_SHUTDOWN_PROC, (void *)&nfs_srv_stop_all);
520 
521 			/* is this an nfsd warm start? */
522 			if (nfs_server_upordown == NFS_SERVER_QUIESCED) {
523 				cmn_err(CE_NOTE, "nfs_server: "
524 				    "server was previously quiesced; "
525 				    "existing NFSv4 state will be re-used");
526 
527 				/*
528 				 * HA-NFSv4: this is also the signal
529 				 * that a Resource Group failover has
530 				 * occurred.
531 				 */
532 				if (cluster_bootflags & CLUSTER_BOOTED)
533 					hanfsv4_failover();
534 			} else {
535 				/* cold start */
536 				rfs4_state_init();
537 				nfs4_drc = rfs4_init_drc(nfs4_drc_max,
538 				    nfs4_drc_hash);
539 			}
540 
541 			/*
542 			 * Check to see if delegation is to be
543 			 * enabled at the server
544 			 */
545 			if (nfs4_srv_delegation != FALSE)
546 				rfs4_set_deleg_policy(SRV_NORMAL_DELEGATE);
547 
548 			nfs_server_upordown = NFS_SERVER_RUNNING;
549 		}
550 		cv_signal(&nfs_server_upordown_cv);
551 	}
552 	mutex_exit(&nfs_server_upordown_lock);
553 }
554 
555 /*
556  * If RDMA device available,
557  * start RDMA listener.
558  */
559 int
560 rdma_start(struct rdma_svc_args *rsa)
561 {
562 	int error;
563 	rdma_xprt_group_t started_rdma_xprts;
564 	rdma_stat stat;
565 	int svc_state = 0;
566 
567 	/* Double check the vers min/max ranges */
568 	if ((rsa->nfs_versmin > rsa->nfs_versmax) ||
569 	    (rsa->nfs_versmin < NFS_VERSMIN) ||
570 	    (rsa->nfs_versmax > NFS_VERSMAX)) {
571 		rsa->nfs_versmin = NFS_VERSMIN_DEFAULT;
572 		rsa->nfs_versmax = NFS_VERSMAX_DEFAULT;
573 	}
574 	nfs_versmin = rsa->nfs_versmin;
575 	nfs_versmax = rsa->nfs_versmax;
576 
577 	/* Set the versions in the callout table */
578 	__nfs_sc_rdma[0].sc_versmin = rsa->nfs_versmin;
579 	__nfs_sc_rdma[0].sc_versmax = rsa->nfs_versmax;
580 	/* For the NFS_ACL program, check the max version */
581 	__nfs_sc_rdma[1].sc_versmin = rsa->nfs_versmin;
582 	if (rsa->nfs_versmax > NFS_ACL_VERSMAX)
583 		__nfs_sc_rdma[1].sc_versmax = NFS_ACL_VERSMAX;
584 	else
585 		__nfs_sc_rdma[1].sc_versmax = rsa->nfs_versmax;
586 
587 	/* Initialize nfsv4 server */
588 	if (rsa->nfs_versmax == (rpcvers_t)NFS_V4)
589 		rfs4_server_start(rsa->delegation);
590 
591 	started_rdma_xprts.rtg_count = 0;
592 	started_rdma_xprts.rtg_listhead = NULL;
593 	started_rdma_xprts.rtg_poolid = rsa->poolid;
594 
595 restart:
596 	error = svc_rdma_kcreate(rsa->netid, &nfs_sct_rdma, rsa->poolid,
597 	    &started_rdma_xprts);
598 
599 	svc_state = !error;
600 
601 	while (!error) {
602 
603 		/*
604 		 * wait till either interrupted by a signal on
605 		 * nfs service stop/restart or signalled by a
606 		 * rdma plugin attach/detatch.
607 		 */
608 
609 		stat = rdma_kwait();
610 
611 		/*
612 		 * stop services if running -- either on a HCA detach event
613 		 * or if the nfs service is stopped/restarted.
614 		 */
615 
616 		if ((stat == RDMA_HCA_DETACH || stat == RDMA_INTR) &&
617 		    svc_state) {
618 			rdma_stop(&started_rdma_xprts);
619 			svc_state = 0;
620 		}
621 
622 		/*
623 		 * nfs service stop/restart, break out of the
624 		 * wait loop and return;
625 		 */
626 		if (stat == RDMA_INTR)
627 			return (0);
628 
629 		/*
630 		 * restart stopped services on a HCA attach event
631 		 * (if not already running)
632 		 */
633 
634 		if ((stat == RDMA_HCA_ATTACH) && (svc_state == 0))
635 			goto restart;
636 
637 		/*
638 		 * loop until a nfs service stop/restart
639 		 */
640 	}
641 
642 	return (error);
643 }
644 
645 /* ARGSUSED */
646 void
647 rpc_null(caddr_t *argp, caddr_t *resp)
648 {
649 }
650 
651 /* ARGSUSED */
652 void
653 rpc_null_v3(caddr_t *argp, caddr_t *resp, struct exportinfo *exi,
654     struct svc_req *req, cred_t *cr)
655 {
656 	DTRACE_NFSV3_3(op__null__start, struct svc_req *, req,
657 	    cred_t *, cr, vnode_t *, NULL);
658 	DTRACE_NFSV3_3(op__null__done, struct svc_req *, req,
659 	    cred_t *, cr, vnode_t *, NULL);
660 }
661 
662 /* ARGSUSED */
663 static void
664 rfs_error(caddr_t *argp, caddr_t *resp)
665 {
666 	/* return (EOPNOTSUPP); */
667 }
668 
669 static void
670 nullfree(void)
671 {
672 }
673 
674 static char *rfscallnames_v2[] = {
675 	"RFS2_NULL",
676 	"RFS2_GETATTR",
677 	"RFS2_SETATTR",
678 	"RFS2_ROOT",
679 	"RFS2_LOOKUP",
680 	"RFS2_READLINK",
681 	"RFS2_READ",
682 	"RFS2_WRITECACHE",
683 	"RFS2_WRITE",
684 	"RFS2_CREATE",
685 	"RFS2_REMOVE",
686 	"RFS2_RENAME",
687 	"RFS2_LINK",
688 	"RFS2_SYMLINK",
689 	"RFS2_MKDIR",
690 	"RFS2_RMDIR",
691 	"RFS2_READDIR",
692 	"RFS2_STATFS"
693 };
694 
695 static struct rpcdisp rfsdisptab_v2[] = {
696 	/*
697 	 * NFS VERSION 2
698 	 */
699 
700 	/* RFS_NULL = 0 */
701 	{rpc_null,
702 	    xdr_void, NULL_xdrproc_t, 0,
703 	    xdr_void, NULL_xdrproc_t, 0,
704 	    nullfree, RPC_IDEMPOTENT,
705 	    0},
706 
707 	/* RFS_GETATTR = 1 */
708 	{rfs_getattr,
709 	    xdr_fhandle, xdr_fastfhandle, sizeof (fhandle_t),
710 	    xdr_attrstat, xdr_fastattrstat, sizeof (struct nfsattrstat),
711 	    nullfree, RPC_IDEMPOTENT|RPC_ALLOWANON|RPC_MAPRESP,
712 	    rfs_getattr_getfh},
713 
714 	/* RFS_SETATTR = 2 */
715 	{rfs_setattr,
716 	    xdr_saargs, NULL_xdrproc_t, sizeof (struct nfssaargs),
717 	    xdr_attrstat, xdr_fastattrstat, sizeof (struct nfsattrstat),
718 	    nullfree, RPC_MAPRESP,
719 	    rfs_setattr_getfh},
720 
721 	/* RFS_ROOT = 3 *** NO LONGER SUPPORTED *** */
722 	{rfs_error,
723 	    xdr_void, NULL_xdrproc_t, 0,
724 	    xdr_void, NULL_xdrproc_t, 0,
725 	    nullfree, RPC_IDEMPOTENT,
726 	    0},
727 
728 	/* RFS_LOOKUP = 4 */
729 	{rfs_lookup,
730 	    xdr_diropargs, NULL_xdrproc_t, sizeof (struct nfsdiropargs),
731 	    xdr_diropres, xdr_fastdiropres, sizeof (struct nfsdiropres),
732 	    nullfree, RPC_IDEMPOTENT|RPC_MAPRESP|RPC_PUBLICFH_OK,
733 	    rfs_lookup_getfh},
734 
735 	/* RFS_READLINK = 5 */
736 	{rfs_readlink,
737 	    xdr_fhandle, xdr_fastfhandle, sizeof (fhandle_t),
738 	    xdr_rdlnres, NULL_xdrproc_t, sizeof (struct nfsrdlnres),
739 	    rfs_rlfree, RPC_IDEMPOTENT,
740 	    rfs_readlink_getfh},
741 
742 	/* RFS_READ = 6 */
743 	{rfs_read,
744 	    xdr_readargs, NULL_xdrproc_t, sizeof (struct nfsreadargs),
745 	    xdr_rdresult, NULL_xdrproc_t, sizeof (struct nfsrdresult),
746 	    rfs_rdfree, RPC_IDEMPOTENT,
747 	    rfs_read_getfh},
748 
749 	/* RFS_WRITECACHE = 7 *** NO LONGER SUPPORTED *** */
750 	{rfs_error,
751 	    xdr_void, NULL_xdrproc_t, 0,
752 	    xdr_void, NULL_xdrproc_t, 0,
753 	    nullfree, RPC_IDEMPOTENT,
754 	    0},
755 
756 	/* RFS_WRITE = 8 */
757 	{rfs_write,
758 	    xdr_writeargs, NULL_xdrproc_t, sizeof (struct nfswriteargs),
759 	    xdr_attrstat, xdr_fastattrstat, sizeof (struct nfsattrstat),
760 	    nullfree, RPC_MAPRESP,
761 	    rfs_write_getfh},
762 
763 	/* RFS_CREATE = 9 */
764 	{rfs_create,
765 	    xdr_creatargs, NULL_xdrproc_t, sizeof (struct nfscreatargs),
766 	    xdr_diropres, xdr_fastdiropres, sizeof (struct nfsdiropres),
767 	    nullfree, RPC_MAPRESP,
768 	    rfs_create_getfh},
769 
770 	/* RFS_REMOVE = 10 */
771 	{rfs_remove,
772 	    xdr_diropargs, NULL_xdrproc_t, sizeof (struct nfsdiropargs),
773 #ifdef _LITTLE_ENDIAN
774 	    xdr_enum, xdr_fastenum, sizeof (enum nfsstat),
775 #else
776 	    xdr_enum, NULL_xdrproc_t, sizeof (enum nfsstat),
777 #endif
778 	    nullfree, RPC_MAPRESP,
779 	    rfs_remove_getfh},
780 
781 	/* RFS_RENAME = 11 */
782 	{rfs_rename,
783 	    xdr_rnmargs, NULL_xdrproc_t, sizeof (struct nfsrnmargs),
784 #ifdef _LITTLE_ENDIAN
785 	    xdr_enum, xdr_fastenum, sizeof (enum nfsstat),
786 #else
787 	    xdr_enum, NULL_xdrproc_t, sizeof (enum nfsstat),
788 #endif
789 	    nullfree, RPC_MAPRESP,
790 	    rfs_rename_getfh},
791 
792 	/* RFS_LINK = 12 */
793 	{rfs_link,
794 	    xdr_linkargs, NULL_xdrproc_t, sizeof (struct nfslinkargs),
795 #ifdef _LITTLE_ENDIAN
796 	    xdr_enum, xdr_fastenum, sizeof (enum nfsstat),
797 #else
798 	    xdr_enum, NULL_xdrproc_t, sizeof (enum nfsstat),
799 #endif
800 	    nullfree, RPC_MAPRESP,
801 	    rfs_link_getfh},
802 
803 	/* RFS_SYMLINK = 13 */
804 	{rfs_symlink,
805 	    xdr_slargs, NULL_xdrproc_t, sizeof (struct nfsslargs),
806 #ifdef _LITTLE_ENDIAN
807 	    xdr_enum, xdr_fastenum, sizeof (enum nfsstat),
808 #else
809 	    xdr_enum, NULL_xdrproc_t, sizeof (enum nfsstat),
810 #endif
811 	    nullfree, RPC_MAPRESP,
812 	    rfs_symlink_getfh},
813 
814 	/* RFS_MKDIR = 14 */
815 	{rfs_mkdir,
816 	    xdr_creatargs, NULL_xdrproc_t, sizeof (struct nfscreatargs),
817 	    xdr_diropres, xdr_fastdiropres, sizeof (struct nfsdiropres),
818 	    nullfree, RPC_MAPRESP,
819 	    rfs_mkdir_getfh},
820 
821 	/* RFS_RMDIR = 15 */
822 	{rfs_rmdir,
823 	    xdr_diropargs, NULL_xdrproc_t, sizeof (struct nfsdiropargs),
824 #ifdef _LITTLE_ENDIAN
825 	    xdr_enum, xdr_fastenum, sizeof (enum nfsstat),
826 #else
827 	    xdr_enum, NULL_xdrproc_t, sizeof (enum nfsstat),
828 #endif
829 	    nullfree, RPC_MAPRESP,
830 	    rfs_rmdir_getfh},
831 
832 	/* RFS_READDIR = 16 */
833 	{rfs_readdir,
834 	    xdr_rddirargs, NULL_xdrproc_t, sizeof (struct nfsrddirargs),
835 	    xdr_putrddirres, NULL_xdrproc_t, sizeof (struct nfsrddirres),
836 	    rfs_rddirfree, RPC_IDEMPOTENT,
837 	    rfs_readdir_getfh},
838 
839 	/* RFS_STATFS = 17 */
840 	{rfs_statfs,
841 	    xdr_fhandle, xdr_fastfhandle, sizeof (fhandle_t),
842 	    xdr_statfs, xdr_faststatfs, sizeof (struct nfsstatfs),
843 	    nullfree, RPC_IDEMPOTENT|RPC_ALLOWANON|RPC_MAPRESP,
844 	    rfs_statfs_getfh},
845 };
846 
847 static char *rfscallnames_v3[] = {
848 	"RFS3_NULL",
849 	"RFS3_GETATTR",
850 	"RFS3_SETATTR",
851 	"RFS3_LOOKUP",
852 	"RFS3_ACCESS",
853 	"RFS3_READLINK",
854 	"RFS3_READ",
855 	"RFS3_WRITE",
856 	"RFS3_CREATE",
857 	"RFS3_MKDIR",
858 	"RFS3_SYMLINK",
859 	"RFS3_MKNOD",
860 	"RFS3_REMOVE",
861 	"RFS3_RMDIR",
862 	"RFS3_RENAME",
863 	"RFS3_LINK",
864 	"RFS3_READDIR",
865 	"RFS3_READDIRPLUS",
866 	"RFS3_FSSTAT",
867 	"RFS3_FSINFO",
868 	"RFS3_PATHCONF",
869 	"RFS3_COMMIT"
870 };
871 
872 static struct rpcdisp rfsdisptab_v3[] = {
873 	/*
874 	 * NFS VERSION 3
875 	 */
876 
877 	/* RFS_NULL = 0 */
878 	{rpc_null_v3,
879 	    xdr_void, NULL_xdrproc_t, 0,
880 	    xdr_void, NULL_xdrproc_t, 0,
881 	    nullfree, RPC_IDEMPOTENT,
882 	    0},
883 
884 	/* RFS3_GETATTR = 1 */
885 	{rfs3_getattr,
886 	    xdr_nfs_fh3_server, NULL_xdrproc_t, sizeof (GETATTR3args),
887 	    xdr_GETATTR3res, NULL_xdrproc_t, sizeof (GETATTR3res),
888 	    nullfree, (RPC_IDEMPOTENT | RPC_ALLOWANON),
889 	    rfs3_getattr_getfh},
890 
891 	/* RFS3_SETATTR = 2 */
892 	{rfs3_setattr,
893 	    xdr_SETATTR3args, NULL_xdrproc_t, sizeof (SETATTR3args),
894 	    xdr_SETATTR3res, NULL_xdrproc_t, sizeof (SETATTR3res),
895 	    nullfree, 0,
896 	    rfs3_setattr_getfh},
897 
898 	/* RFS3_LOOKUP = 3 */
899 	{rfs3_lookup,
900 	    xdr_diropargs3, NULL_xdrproc_t, sizeof (LOOKUP3args),
901 	    xdr_LOOKUP3res, NULL_xdrproc_t, sizeof (LOOKUP3res),
902 	    nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK),
903 	    rfs3_lookup_getfh},
904 
905 	/* RFS3_ACCESS = 4 */
906 	{rfs3_access,
907 	    xdr_ACCESS3args, NULL_xdrproc_t, sizeof (ACCESS3args),
908 	    xdr_ACCESS3res, NULL_xdrproc_t, sizeof (ACCESS3res),
909 	    nullfree, RPC_IDEMPOTENT,
910 	    rfs3_access_getfh},
911 
912 	/* RFS3_READLINK = 5 */
913 	{rfs3_readlink,
914 	    xdr_nfs_fh3_server, NULL_xdrproc_t, sizeof (READLINK3args),
915 	    xdr_READLINK3res, NULL_xdrproc_t, sizeof (READLINK3res),
916 	    rfs3_readlink_free, RPC_IDEMPOTENT,
917 	    rfs3_readlink_getfh},
918 
919 	/* RFS3_READ = 6 */
920 	{rfs3_read,
921 	    xdr_READ3args, NULL_xdrproc_t, sizeof (READ3args),
922 	    xdr_READ3res, NULL_xdrproc_t, sizeof (READ3res),
923 	    rfs3_read_free, RPC_IDEMPOTENT,
924 	    rfs3_read_getfh},
925 
926 	/* RFS3_WRITE = 7 */
927 	{rfs3_write,
928 	    xdr_WRITE3args, NULL_xdrproc_t, sizeof (WRITE3args),
929 	    xdr_WRITE3res, NULL_xdrproc_t, sizeof (WRITE3res),
930 	    nullfree, 0,
931 	    rfs3_write_getfh},
932 
933 	/* RFS3_CREATE = 8 */
934 	{rfs3_create,
935 	    xdr_CREATE3args, NULL_xdrproc_t, sizeof (CREATE3args),
936 	    xdr_CREATE3res, NULL_xdrproc_t, sizeof (CREATE3res),
937 	    nullfree, 0,
938 	    rfs3_create_getfh},
939 
940 	/* RFS3_MKDIR = 9 */
941 	{rfs3_mkdir,
942 	    xdr_MKDIR3args, NULL_xdrproc_t, sizeof (MKDIR3args),
943 	    xdr_MKDIR3res, NULL_xdrproc_t, sizeof (MKDIR3res),
944 	    nullfree, 0,
945 	    rfs3_mkdir_getfh},
946 
947 	/* RFS3_SYMLINK = 10 */
948 	{rfs3_symlink,
949 	    xdr_SYMLINK3args, NULL_xdrproc_t, sizeof (SYMLINK3args),
950 	    xdr_SYMLINK3res, NULL_xdrproc_t, sizeof (SYMLINK3res),
951 	    nullfree, 0,
952 	    rfs3_symlink_getfh},
953 
954 	/* RFS3_MKNOD = 11 */
955 	{rfs3_mknod,
956 	    xdr_MKNOD3args, NULL_xdrproc_t, sizeof (MKNOD3args),
957 	    xdr_MKNOD3res, NULL_xdrproc_t, sizeof (MKNOD3res),
958 	    nullfree, 0,
959 	    rfs3_mknod_getfh},
960 
961 	/* RFS3_REMOVE = 12 */
962 	{rfs3_remove,
963 	    xdr_diropargs3, NULL_xdrproc_t, sizeof (REMOVE3args),
964 	    xdr_REMOVE3res, NULL_xdrproc_t, sizeof (REMOVE3res),
965 	    nullfree, 0,
966 	    rfs3_remove_getfh},
967 
968 	/* RFS3_RMDIR = 13 */
969 	{rfs3_rmdir,
970 	    xdr_diropargs3, NULL_xdrproc_t, sizeof (RMDIR3args),
971 	    xdr_RMDIR3res, NULL_xdrproc_t, sizeof (RMDIR3res),
972 	    nullfree, 0,
973 	    rfs3_rmdir_getfh},
974 
975 	/* RFS3_RENAME = 14 */
976 	{rfs3_rename,
977 	    xdr_RENAME3args, NULL_xdrproc_t, sizeof (RENAME3args),
978 	    xdr_RENAME3res, NULL_xdrproc_t, sizeof (RENAME3res),
979 	    nullfree, 0,
980 	    rfs3_rename_getfh},
981 
982 	/* RFS3_LINK = 15 */
983 	{rfs3_link,
984 	    xdr_LINK3args, NULL_xdrproc_t, sizeof (LINK3args),
985 	    xdr_LINK3res, NULL_xdrproc_t, sizeof (LINK3res),
986 	    nullfree, 0,
987 	    rfs3_link_getfh},
988 
989 	/* RFS3_READDIR = 16 */
990 	{rfs3_readdir,
991 	    xdr_READDIR3args, NULL_xdrproc_t, sizeof (READDIR3args),
992 	    xdr_READDIR3res, NULL_xdrproc_t, sizeof (READDIR3res),
993 	    rfs3_readdir_free, RPC_IDEMPOTENT,
994 	    rfs3_readdir_getfh},
995 
996 	/* RFS3_READDIRPLUS = 17 */
997 	{rfs3_readdirplus,
998 	    xdr_READDIRPLUS3args, NULL_xdrproc_t, sizeof (READDIRPLUS3args),
999 	    xdr_READDIRPLUS3res, NULL_xdrproc_t, sizeof (READDIRPLUS3res),
1000 	    rfs3_readdirplus_free, RPC_AVOIDWORK,
1001 	    rfs3_readdirplus_getfh},
1002 
1003 	/* RFS3_FSSTAT = 18 */
1004 	{rfs3_fsstat,
1005 	    xdr_nfs_fh3_server, NULL_xdrproc_t, sizeof (FSSTAT3args),
1006 	    xdr_FSSTAT3res, NULL_xdrproc_t, sizeof (FSSTAT3res),
1007 	    nullfree, RPC_IDEMPOTENT,
1008 	    rfs3_fsstat_getfh},
1009 
1010 	/* RFS3_FSINFO = 19 */
1011 	{rfs3_fsinfo,
1012 	    xdr_nfs_fh3_server, NULL_xdrproc_t, sizeof (FSINFO3args),
1013 	    xdr_FSINFO3res, NULL_xdrproc_t, sizeof (FSINFO3res),
1014 	    nullfree, RPC_IDEMPOTENT|RPC_ALLOWANON,
1015 	    rfs3_fsinfo_getfh},
1016 
1017 	/* RFS3_PATHCONF = 20 */
1018 	{rfs3_pathconf,
1019 	    xdr_nfs_fh3_server, NULL_xdrproc_t, sizeof (PATHCONF3args),
1020 	    xdr_PATHCONF3res, NULL_xdrproc_t, sizeof (PATHCONF3res),
1021 	    nullfree, RPC_IDEMPOTENT,
1022 	    rfs3_pathconf_getfh},
1023 
1024 	/* RFS3_COMMIT = 21 */
1025 	{rfs3_commit,
1026 	    xdr_COMMIT3args, NULL_xdrproc_t, sizeof (COMMIT3args),
1027 	    xdr_COMMIT3res, NULL_xdrproc_t, sizeof (COMMIT3res),
1028 	    nullfree, RPC_IDEMPOTENT,
1029 	    rfs3_commit_getfh},
1030 };
1031 
1032 static char *rfscallnames_v4[] = {
1033 	"RFS4_NULL",
1034 	"RFS4_COMPOUND",
1035 	"RFS4_NULL",
1036 	"RFS4_NULL",
1037 	"RFS4_NULL",
1038 	"RFS4_NULL",
1039 	"RFS4_NULL",
1040 	"RFS4_NULL",
1041 	"RFS4_CREATE"
1042 };
1043 
1044 static struct rpcdisp rfsdisptab_v4[] = {
1045 	/*
1046 	 * NFS VERSION 4
1047 	 */
1048 
1049 	/* RFS_NULL = 0 */
1050 	{rpc_null,
1051 	    xdr_void, NULL_xdrproc_t, 0,
1052 	    xdr_void, NULL_xdrproc_t, 0,
1053 	    nullfree, RPC_IDEMPOTENT, 0},
1054 
1055 	/* RFS4_compound = 1 */
1056 	{rfs4_compound,
1057 	    xdr_COMPOUND4args_srv, NULL_xdrproc_t, sizeof (COMPOUND4args),
1058 	    xdr_COMPOUND4res_srv, NULL_xdrproc_t, sizeof (COMPOUND4res),
1059 	    rfs4_compound_free, 0, 0},
1060 };
1061 
1062 union rfs_args {
1063 	/*
1064 	 * NFS VERSION 2
1065 	 */
1066 
1067 	/* RFS_NULL = 0 */
1068 
1069 	/* RFS_GETATTR = 1 */
1070 	fhandle_t nfs2_getattr_args;
1071 
1072 	/* RFS_SETATTR = 2 */
1073 	struct nfssaargs nfs2_setattr_args;
1074 
1075 	/* RFS_ROOT = 3 *** NO LONGER SUPPORTED *** */
1076 
1077 	/* RFS_LOOKUP = 4 */
1078 	struct nfsdiropargs nfs2_lookup_args;
1079 
1080 	/* RFS_READLINK = 5 */
1081 	fhandle_t nfs2_readlink_args;
1082 
1083 	/* RFS_READ = 6 */
1084 	struct nfsreadargs nfs2_read_args;
1085 
1086 	/* RFS_WRITECACHE = 7 *** NO LONGER SUPPORTED *** */
1087 
1088 	/* RFS_WRITE = 8 */
1089 	struct nfswriteargs nfs2_write_args;
1090 
1091 	/* RFS_CREATE = 9 */
1092 	struct nfscreatargs nfs2_create_args;
1093 
1094 	/* RFS_REMOVE = 10 */
1095 	struct nfsdiropargs nfs2_remove_args;
1096 
1097 	/* RFS_RENAME = 11 */
1098 	struct nfsrnmargs nfs2_rename_args;
1099 
1100 	/* RFS_LINK = 12 */
1101 	struct nfslinkargs nfs2_link_args;
1102 
1103 	/* RFS_SYMLINK = 13 */
1104 	struct nfsslargs nfs2_symlink_args;
1105 
1106 	/* RFS_MKDIR = 14 */
1107 	struct nfscreatargs nfs2_mkdir_args;
1108 
1109 	/* RFS_RMDIR = 15 */
1110 	struct nfsdiropargs nfs2_rmdir_args;
1111 
1112 	/* RFS_READDIR = 16 */
1113 	struct nfsrddirargs nfs2_readdir_args;
1114 
1115 	/* RFS_STATFS = 17 */
1116 	fhandle_t nfs2_statfs_args;
1117 
1118 	/*
1119 	 * NFS VERSION 3
1120 	 */
1121 
1122 	/* RFS_NULL = 0 */
1123 
1124 	/* RFS3_GETATTR = 1 */
1125 	GETATTR3args nfs3_getattr_args;
1126 
1127 	/* RFS3_SETATTR = 2 */
1128 	SETATTR3args nfs3_setattr_args;
1129 
1130 	/* RFS3_LOOKUP = 3 */
1131 	LOOKUP3args nfs3_lookup_args;
1132 
1133 	/* RFS3_ACCESS = 4 */
1134 	ACCESS3args nfs3_access_args;
1135 
1136 	/* RFS3_READLINK = 5 */
1137 	READLINK3args nfs3_readlink_args;
1138 
1139 	/* RFS3_READ = 6 */
1140 	READ3args nfs3_read_args;
1141 
1142 	/* RFS3_WRITE = 7 */
1143 	WRITE3args nfs3_write_args;
1144 
1145 	/* RFS3_CREATE = 8 */
1146 	CREATE3args nfs3_create_args;
1147 
1148 	/* RFS3_MKDIR = 9 */
1149 	MKDIR3args nfs3_mkdir_args;
1150 
1151 	/* RFS3_SYMLINK = 10 */
1152 	SYMLINK3args nfs3_symlink_args;
1153 
1154 	/* RFS3_MKNOD = 11 */
1155 	MKNOD3args nfs3_mknod_args;
1156 
1157 	/* RFS3_REMOVE = 12 */
1158 	REMOVE3args nfs3_remove_args;
1159 
1160 	/* RFS3_RMDIR = 13 */
1161 	RMDIR3args nfs3_rmdir_args;
1162 
1163 	/* RFS3_RENAME = 14 */
1164 	RENAME3args nfs3_rename_args;
1165 
1166 	/* RFS3_LINK = 15 */
1167 	LINK3args nfs3_link_args;
1168 
1169 	/* RFS3_READDIR = 16 */
1170 	READDIR3args nfs3_readdir_args;
1171 
1172 	/* RFS3_READDIRPLUS = 17 */
1173 	READDIRPLUS3args nfs3_readdirplus_args;
1174 
1175 	/* RFS3_FSSTAT = 18 */
1176 	FSSTAT3args nfs3_fsstat_args;
1177 
1178 	/* RFS3_FSINFO = 19 */
1179 	FSINFO3args nfs3_fsinfo_args;
1180 
1181 	/* RFS3_PATHCONF = 20 */
1182 	PATHCONF3args nfs3_pathconf_args;
1183 
1184 	/* RFS3_COMMIT = 21 */
1185 	COMMIT3args nfs3_commit_args;
1186 
1187 	/*
1188 	 * NFS VERSION 4
1189 	 */
1190 
1191 	/* RFS_NULL = 0 */
1192 
1193 	/* COMPUND = 1 */
1194 	COMPOUND4args nfs4_compound_args;
1195 };
1196 
1197 union rfs_res {
1198 	/*
1199 	 * NFS VERSION 2
1200 	 */
1201 
1202 	/* RFS_NULL = 0 */
1203 
1204 	/* RFS_GETATTR = 1 */
1205 	struct nfsattrstat nfs2_getattr_res;
1206 
1207 	/* RFS_SETATTR = 2 */
1208 	struct nfsattrstat nfs2_setattr_res;
1209 
1210 	/* RFS_ROOT = 3 *** NO LONGER SUPPORTED *** */
1211 
1212 	/* RFS_LOOKUP = 4 */
1213 	struct nfsdiropres nfs2_lookup_res;
1214 
1215 	/* RFS_READLINK = 5 */
1216 	struct nfsrdlnres nfs2_readlink_res;
1217 
1218 	/* RFS_READ = 6 */
1219 	struct nfsrdresult nfs2_read_res;
1220 
1221 	/* RFS_WRITECACHE = 7 *** NO LONGER SUPPORTED *** */
1222 
1223 	/* RFS_WRITE = 8 */
1224 	struct nfsattrstat nfs2_write_res;
1225 
1226 	/* RFS_CREATE = 9 */
1227 	struct nfsdiropres nfs2_create_res;
1228 
1229 	/* RFS_REMOVE = 10 */
1230 	enum nfsstat nfs2_remove_res;
1231 
1232 	/* RFS_RENAME = 11 */
1233 	enum nfsstat nfs2_rename_res;
1234 
1235 	/* RFS_LINK = 12 */
1236 	enum nfsstat nfs2_link_res;
1237 
1238 	/* RFS_SYMLINK = 13 */
1239 	enum nfsstat nfs2_symlink_res;
1240 
1241 	/* RFS_MKDIR = 14 */
1242 	struct nfsdiropres nfs2_mkdir_res;
1243 
1244 	/* RFS_RMDIR = 15 */
1245 	enum nfsstat nfs2_rmdir_res;
1246 
1247 	/* RFS_READDIR = 16 */
1248 	struct nfsrddirres nfs2_readdir_res;
1249 
1250 	/* RFS_STATFS = 17 */
1251 	struct nfsstatfs nfs2_statfs_res;
1252 
1253 	/*
1254 	 * NFS VERSION 3
1255 	 */
1256 
1257 	/* RFS_NULL = 0 */
1258 
1259 	/* RFS3_GETATTR = 1 */
1260 	GETATTR3res nfs3_getattr_res;
1261 
1262 	/* RFS3_SETATTR = 2 */
1263 	SETATTR3res nfs3_setattr_res;
1264 
1265 	/* RFS3_LOOKUP = 3 */
1266 	LOOKUP3res nfs3_lookup_res;
1267 
1268 	/* RFS3_ACCESS = 4 */
1269 	ACCESS3res nfs3_access_res;
1270 
1271 	/* RFS3_READLINK = 5 */
1272 	READLINK3res nfs3_readlink_res;
1273 
1274 	/* RFS3_READ = 6 */
1275 	READ3res nfs3_read_res;
1276 
1277 	/* RFS3_WRITE = 7 */
1278 	WRITE3res nfs3_write_res;
1279 
1280 	/* RFS3_CREATE = 8 */
1281 	CREATE3res nfs3_create_res;
1282 
1283 	/* RFS3_MKDIR = 9 */
1284 	MKDIR3res nfs3_mkdir_res;
1285 
1286 	/* RFS3_SYMLINK = 10 */
1287 	SYMLINK3res nfs3_symlink_res;
1288 
1289 	/* RFS3_MKNOD = 11 */
1290 	MKNOD3res nfs3_mknod_res;
1291 
1292 	/* RFS3_REMOVE = 12 */
1293 	REMOVE3res nfs3_remove_res;
1294 
1295 	/* RFS3_RMDIR = 13 */
1296 	RMDIR3res nfs3_rmdir_res;
1297 
1298 	/* RFS3_RENAME = 14 */
1299 	RENAME3res nfs3_rename_res;
1300 
1301 	/* RFS3_LINK = 15 */
1302 	LINK3res nfs3_link_res;
1303 
1304 	/* RFS3_READDIR = 16 */
1305 	READDIR3res nfs3_readdir_res;
1306 
1307 	/* RFS3_READDIRPLUS = 17 */
1308 	READDIRPLUS3res nfs3_readdirplus_res;
1309 
1310 	/* RFS3_FSSTAT = 18 */
1311 	FSSTAT3res nfs3_fsstat_res;
1312 
1313 	/* RFS3_FSINFO = 19 */
1314 	FSINFO3res nfs3_fsinfo_res;
1315 
1316 	/* RFS3_PATHCONF = 20 */
1317 	PATHCONF3res nfs3_pathconf_res;
1318 
1319 	/* RFS3_COMMIT = 21 */
1320 	COMMIT3res nfs3_commit_res;
1321 
1322 	/*
1323 	 * NFS VERSION 4
1324 	 */
1325 
1326 	/* RFS_NULL = 0 */
1327 
1328 	/* RFS4_COMPOUND = 1 */
1329 	COMPOUND4res nfs4_compound_res;
1330 
1331 };
1332 
1333 static struct rpc_disptable rfs_disptable[] = {
1334 	{sizeof (rfsdisptab_v2) / sizeof (rfsdisptab_v2[0]),
1335 	    rfscallnames_v2,
1336 	    &rfsproccnt_v2_ptr, rfsdisptab_v2},
1337 	{sizeof (rfsdisptab_v3) / sizeof (rfsdisptab_v3[0]),
1338 	    rfscallnames_v3,
1339 	    &rfsproccnt_v3_ptr, rfsdisptab_v3},
1340 	{sizeof (rfsdisptab_v4) / sizeof (rfsdisptab_v4[0]),
1341 	    rfscallnames_v4,
1342 	    &rfsproccnt_v4_ptr, rfsdisptab_v4},
1343 };
1344 
1345 /*
1346  * If nfs_portmon is set, then clients are required to use privileged
1347  * ports (ports < IPPORT_RESERVED) in order to get NFS services.
1348  *
1349  * N.B.: this attempt to carry forward the already ill-conceived notion
1350  * of privileged ports for TCP/UDP is really quite ineffectual.  Not only
1351  * is it transport-dependent, it's laughably easy to spoof.  If you're
1352  * really interested in security, you must start with secure RPC instead.
1353  */
1354 static int nfs_portmon = 0;
1355 
1356 #ifdef DEBUG
1357 static int cred_hits = 0;
1358 static int cred_misses = 0;
1359 #endif
1360 
1361 
1362 #ifdef DEBUG
1363 /*
1364  * Debug code to allow disabling of rfs_dispatch() use of
1365  * fastxdrargs() and fastxdrres() calls for testing purposes.
1366  */
1367 static int rfs_no_fast_xdrargs = 0;
1368 static int rfs_no_fast_xdrres = 0;
1369 #endif
1370 
1371 union acl_args {
1372 	/*
1373 	 * ACL VERSION 2
1374 	 */
1375 
1376 	/* ACL2_NULL = 0 */
1377 
1378 	/* ACL2_GETACL = 1 */
1379 	GETACL2args acl2_getacl_args;
1380 
1381 	/* ACL2_SETACL = 2 */
1382 	SETACL2args acl2_setacl_args;
1383 
1384 	/* ACL2_GETATTR = 3 */
1385 	GETATTR2args acl2_getattr_args;
1386 
1387 	/* ACL2_ACCESS = 4 */
1388 	ACCESS2args acl2_access_args;
1389 
1390 	/* ACL2_GETXATTRDIR = 5 */
1391 	GETXATTRDIR2args acl2_getxattrdir_args;
1392 
1393 	/*
1394 	 * ACL VERSION 3
1395 	 */
1396 
1397 	/* ACL3_NULL = 0 */
1398 
1399 	/* ACL3_GETACL = 1 */
1400 	GETACL3args acl3_getacl_args;
1401 
1402 	/* ACL3_SETACL = 2 */
1403 	SETACL3args acl3_setacl;
1404 
1405 	/* ACL3_GETXATTRDIR = 3 */
1406 	GETXATTRDIR3args acl3_getxattrdir_args;
1407 
1408 };
1409 
1410 union acl_res {
1411 	/*
1412 	 * ACL VERSION 2
1413 	 */
1414 
1415 	/* ACL2_NULL = 0 */
1416 
1417 	/* ACL2_GETACL = 1 */
1418 	GETACL2res acl2_getacl_res;
1419 
1420 	/* ACL2_SETACL = 2 */
1421 	SETACL2res acl2_setacl_res;
1422 
1423 	/* ACL2_GETATTR = 3 */
1424 	GETATTR2res acl2_getattr_res;
1425 
1426 	/* ACL2_ACCESS = 4 */
1427 	ACCESS2res acl2_access_res;
1428 
1429 	/* ACL2_GETXATTRDIR = 5 */
1430 	GETXATTRDIR2args acl2_getxattrdir_res;
1431 
1432 	/*
1433 	 * ACL VERSION 3
1434 	 */
1435 
1436 	/* ACL3_NULL = 0 */
1437 
1438 	/* ACL3_GETACL = 1 */
1439 	GETACL3res acl3_getacl_res;
1440 
1441 	/* ACL3_SETACL = 2 */
1442 	SETACL3res acl3_setacl_res;
1443 
1444 	/* ACL3_GETXATTRDIR = 3 */
1445 	GETXATTRDIR3res acl3_getxattrdir_res;
1446 
1447 };
1448 
1449 static bool_t
1450 auth_tooweak(struct svc_req *req, char *res)
1451 {
1452 
1453 	if (req->rq_vers == NFS_VERSION && req->rq_proc == RFS_LOOKUP) {
1454 		struct nfsdiropres *dr = (struct nfsdiropres *)res;
1455 		if (dr->dr_status == WNFSERR_CLNT_FLAVOR)
1456 			return (TRUE);
1457 	} else if (req->rq_vers == NFS_V3 && req->rq_proc == NFSPROC3_LOOKUP) {
1458 		LOOKUP3res *resp = (LOOKUP3res *)res;
1459 		if (resp->status == WNFSERR_CLNT_FLAVOR)
1460 			return (TRUE);
1461 	}
1462 	return (FALSE);
1463 }
1464 
1465 
1466 static void
1467 common_dispatch(struct svc_req *req, SVCXPRT *xprt, rpcvers_t min_vers,
1468 		rpcvers_t max_vers, char *pgmname,
1469 		struct rpc_disptable *disptable)
1470 {
1471 	int which;
1472 	rpcvers_t vers;
1473 	char *args;
1474 	union {
1475 			union rfs_args ra;
1476 			union acl_args aa;
1477 		} args_buf;
1478 	char *res;
1479 	union {
1480 			union rfs_res rr;
1481 			union acl_res ar;
1482 		} res_buf;
1483 	struct rpcdisp *disp = NULL;
1484 	int dis_flags = 0;
1485 	cred_t *cr;
1486 	int error = 0;
1487 	int anon_ok;
1488 	struct exportinfo *exi = NULL;
1489 	unsigned int nfslog_rec_id;
1490 	int dupstat;
1491 	struct dupreq *dr;
1492 	int authres;
1493 	bool_t publicfh_ok = FALSE;
1494 	enum_t auth_flavor;
1495 	bool_t dupcached = FALSE;
1496 	struct netbuf	nb;
1497 	bool_t logging_enabled = FALSE;
1498 	struct exportinfo *nfslog_exi = NULL;
1499 	char **procnames;
1500 	char cbuf[INET6_ADDRSTRLEN];	/* to hold both IPv4 and IPv6 addr */
1501 
1502 	vers = req->rq_vers;
1503 
1504 	if (vers < min_vers || vers > max_vers) {
1505 		svcerr_progvers(req->rq_xprt, min_vers, max_vers);
1506 		error++;
1507 		cmn_err(CE_NOTE, "%s: bad version number %u", pgmname, vers);
1508 		goto done;
1509 	}
1510 	vers -= min_vers;
1511 
1512 	which = req->rq_proc;
1513 	if (which < 0 || which >= disptable[(int)vers].dis_nprocs) {
1514 		svcerr_noproc(req->rq_xprt);
1515 		error++;
1516 		goto done;
1517 	}
1518 
1519 	(*(disptable[(int)vers].dis_proccntp))[which].value.ui64++;
1520 
1521 	disp = &disptable[(int)vers].dis_table[which];
1522 	procnames = disptable[(int)vers].dis_procnames;
1523 
1524 	auth_flavor = req->rq_cred.oa_flavor;
1525 
1526 	/*
1527 	 * Deserialize into the args struct.
1528 	 */
1529 	args = (char *)&args_buf;
1530 
1531 #ifdef DEBUG
1532 	if (rfs_no_fast_xdrargs || (auth_flavor == RPCSEC_GSS) ||
1533 	    disp->dis_fastxdrargs == NULL_xdrproc_t ||
1534 	    !SVC_GETARGS(xprt, disp->dis_fastxdrargs, (char *)&args))
1535 #else
1536 	if ((auth_flavor == RPCSEC_GSS) ||
1537 	    disp->dis_fastxdrargs == NULL_xdrproc_t ||
1538 	    !SVC_GETARGS(xprt, disp->dis_fastxdrargs, (char *)&args))
1539 #endif
1540 	{
1541 		bzero(args, disp->dis_argsz);
1542 		if (!SVC_GETARGS(xprt, disp->dis_xdrargs, args)) {
1543 			error++;
1544 			/*
1545 			 * Check if we are outside our capabilities.
1546 			 */
1547 			if (rfs4_minorvers_mismatch(req, xprt, (void *)args))
1548 				goto done;
1549 
1550 			svcerr_decode(xprt);
1551 			cmn_err(CE_NOTE,
1552 			    "Failed to decode arguments for %s version %u "
1553 			    "procedure %s client %s%s",
1554 			    pgmname, vers + min_vers, procnames[which],
1555 			    client_name(req), client_addr(req, cbuf));
1556 			goto done;
1557 		}
1558 	}
1559 
1560 	/*
1561 	 * If Version 4 use that specific dispatch function.
1562 	 */
1563 	if (req->rq_vers == 4) {
1564 		error += rfs4_dispatch(disp, req, xprt, args);
1565 		goto done;
1566 	}
1567 
1568 	dis_flags = disp->dis_flags;
1569 
1570 	/*
1571 	 * Find export information and check authentication,
1572 	 * setting the credential if everything is ok.
1573 	 */
1574 	if (disp->dis_getfh != NULL) {
1575 		void *fh;
1576 		fsid_t *fsid;
1577 		fid_t *fid, *xfid;
1578 		fhandle_t *fh2;
1579 		nfs_fh3 *fh3;
1580 
1581 		fh = (*disp->dis_getfh)(args);
1582 		switch (req->rq_vers) {
1583 		case NFS_VERSION:
1584 			fh2 = (fhandle_t *)fh;
1585 			fsid = &fh2->fh_fsid;
1586 			fid = (fid_t *)&fh2->fh_len;
1587 			xfid = (fid_t *)&fh2->fh_xlen;
1588 			break;
1589 		case NFS_V3:
1590 			fh3 = (nfs_fh3 *)fh;
1591 			fsid = &fh3->fh3_fsid;
1592 			fid = FH3TOFIDP(fh3);
1593 			xfid = FH3TOXFIDP(fh3);
1594 			break;
1595 		}
1596 
1597 		/*
1598 		 * Fix for bug 1038302 - corbin
1599 		 * There is a problem here if anonymous access is
1600 		 * disallowed.  If the current request is part of the
1601 		 * client's mount process for the requested filesystem,
1602 		 * then it will carry root (uid 0) credentials on it, and
1603 		 * will be denied by checkauth if that client does not
1604 		 * have explicit root=0 permission.  This will cause the
1605 		 * client's mount operation to fail.  As a work-around,
1606 		 * we check here to see if the request is a getattr or
1607 		 * statfs operation on the exported vnode itself, and
1608 		 * pass a flag to checkauth with the result of this test.
1609 		 *
1610 		 * The filehandle refers to the mountpoint itself if
1611 		 * the fh_data and fh_xdata portions of the filehandle
1612 		 * are equal.
1613 		 *
1614 		 * Added anon_ok argument to checkauth().
1615 		 */
1616 
1617 		if ((dis_flags & RPC_ALLOWANON) && EQFID(fid, xfid))
1618 			anon_ok = 1;
1619 		else
1620 			anon_ok = 0;
1621 
1622 		cr = xprt->xp_cred;
1623 		ASSERT(cr != NULL);
1624 #ifdef DEBUG
1625 		if (crgetref(cr) != 1) {
1626 			crfree(cr);
1627 			cr = crget();
1628 			xprt->xp_cred = cr;
1629 			cred_misses++;
1630 		} else
1631 			cred_hits++;
1632 #else
1633 		if (crgetref(cr) != 1) {
1634 			crfree(cr);
1635 			cr = crget();
1636 			xprt->xp_cred = cr;
1637 		}
1638 #endif
1639 
1640 		exi = checkexport(fsid, xfid);
1641 
1642 		if (exi != NULL) {
1643 			publicfh_ok = PUBLICFH_CHECK(disp, exi, fsid, xfid);
1644 
1645 			/*
1646 			 * Don't allow non-V4 clients access
1647 			 * to pseudo exports
1648 			 */
1649 			if (PSEUDO(exi)) {
1650 				svcerr_weakauth(xprt);
1651 				error++;
1652 				goto done;
1653 			}
1654 
1655 			authres = checkauth(exi, req, cr, anon_ok, publicfh_ok);
1656 			/*
1657 			 * authres >  0: authentication OK - proceed
1658 			 * authres == 0: authentication weak - return error
1659 			 * authres <  0: authentication timeout - drop
1660 			 */
1661 			if (authres <= 0) {
1662 				if (authres == 0) {
1663 					svcerr_weakauth(xprt);
1664 					error++;
1665 				}
1666 				goto done;
1667 			}
1668 		}
1669 	} else
1670 		cr = NULL;
1671 
1672 	if ((dis_flags & RPC_MAPRESP) && (auth_flavor != RPCSEC_GSS)) {
1673 		res = (char *)SVC_GETRES(xprt, disp->dis_ressz);
1674 		if (res == NULL)
1675 			res = (char *)&res_buf;
1676 	} else
1677 		res = (char *)&res_buf;
1678 
1679 	if (!(dis_flags & RPC_IDEMPOTENT)) {
1680 		dupstat = SVC_DUP_EXT(xprt, req, res, disp->dis_ressz, &dr,
1681 		    &dupcached);
1682 
1683 		switch (dupstat) {
1684 		case DUP_ERROR:
1685 			svcerr_systemerr(xprt);
1686 			error++;
1687 			goto done;
1688 			/* NOTREACHED */
1689 		case DUP_INPROGRESS:
1690 			if (res != (char *)&res_buf)
1691 				SVC_FREERES(xprt);
1692 			error++;
1693 			goto done;
1694 			/* NOTREACHED */
1695 		case DUP_NEW:
1696 		case DUP_DROP:
1697 			curthread->t_flag |= T_DONTPEND;
1698 
1699 			(*disp->dis_proc)(args, res, exi, req, cr);
1700 
1701 			curthread->t_flag &= ~T_DONTPEND;
1702 			if (curthread->t_flag & T_WOULDBLOCK) {
1703 				curthread->t_flag &= ~T_WOULDBLOCK;
1704 				SVC_DUPDONE_EXT(xprt, dr, res, NULL,
1705 				    disp->dis_ressz, DUP_DROP);
1706 				if (res != (char *)&res_buf)
1707 					SVC_FREERES(xprt);
1708 				error++;
1709 				goto done;
1710 			}
1711 			if (dis_flags & RPC_AVOIDWORK) {
1712 				SVC_DUPDONE_EXT(xprt, dr, res, NULL,
1713 				    disp->dis_ressz, DUP_DROP);
1714 			} else {
1715 				SVC_DUPDONE_EXT(xprt, dr, res,
1716 				    disp->dis_resfree == nullfree ? NULL :
1717 				    disp->dis_resfree,
1718 				    disp->dis_ressz, DUP_DONE);
1719 				dupcached = TRUE;
1720 			}
1721 			break;
1722 		case DUP_DONE:
1723 			break;
1724 		}
1725 
1726 	} else {
1727 		curthread->t_flag |= T_DONTPEND;
1728 
1729 		(*disp->dis_proc)(args, res, exi, req, cr);
1730 
1731 		curthread->t_flag &= ~T_DONTPEND;
1732 		if (curthread->t_flag & T_WOULDBLOCK) {
1733 			curthread->t_flag &= ~T_WOULDBLOCK;
1734 			if (res != (char *)&res_buf)
1735 				SVC_FREERES(xprt);
1736 			error++;
1737 			goto done;
1738 		}
1739 	}
1740 
1741 	if (auth_tooweak(req, res)) {
1742 		svcerr_weakauth(xprt);
1743 		error++;
1744 		goto done;
1745 	}
1746 
1747 	/*
1748 	 * Check to see if logging has been enabled on the server.
1749 	 * If so, then obtain the export info struct to be used for
1750 	 * the later writing of the log record.  This is done for
1751 	 * the case that a lookup is done across a non-logged public
1752 	 * file system.
1753 	 */
1754 	if (nfslog_buffer_list != NULL) {
1755 		nfslog_exi = nfslog_get_exi(exi, req, res, &nfslog_rec_id);
1756 		/*
1757 		 * Is logging enabled?
1758 		 */
1759 		logging_enabled = (nfslog_exi != NULL);
1760 
1761 		/*
1762 		 * Copy the netbuf for logging purposes, before it is
1763 		 * freed by svc_sendreply().
1764 		 */
1765 		if (logging_enabled) {
1766 			NFSLOG_COPY_NETBUF(nfslog_exi, xprt, &nb);
1767 			/*
1768 			 * If RPC_MAPRESP flag set (i.e. in V2 ops) the
1769 			 * res gets copied directly into the mbuf and
1770 			 * may be freed soon after the sendreply. So we
1771 			 * must copy it here to a safe place...
1772 			 */
1773 			if (res != (char *)&res_buf) {
1774 				bcopy(res, (char *)&res_buf, disp->dis_ressz);
1775 			}
1776 		}
1777 	}
1778 
1779 	/*
1780 	 * Serialize and send results struct
1781 	 */
1782 #ifdef DEBUG
1783 	if (rfs_no_fast_xdrres == 0 && res != (char *)&res_buf)
1784 #else
1785 	if (res != (char *)&res_buf)
1786 #endif
1787 	{
1788 		if (!svc_sendreply(xprt, disp->dis_fastxdrres, res)) {
1789 			cmn_err(CE_NOTE, "%s: bad sendreply", pgmname);
1790 			svcerr_systemerr(xprt);
1791 			error++;
1792 		}
1793 	} else {
1794 		if (!svc_sendreply(xprt, disp->dis_xdrres, res)) {
1795 			cmn_err(CE_NOTE, "%s: bad sendreply", pgmname);
1796 			svcerr_systemerr(xprt);
1797 			error++;
1798 		}
1799 	}
1800 
1801 	/*
1802 	 * Log if needed
1803 	 */
1804 	if (logging_enabled) {
1805 		nfslog_write_record(nfslog_exi, req, args, (char *)&res_buf,
1806 		    cr, &nb, nfslog_rec_id, NFSLOG_ONE_BUFFER);
1807 		exi_rele(nfslog_exi);
1808 		kmem_free((&nb)->buf, (&nb)->len);
1809 	}
1810 
1811 	/*
1812 	 * Free results struct. With the addition of NFS V4 we can
1813 	 * have non-idempotent procedures with functions.
1814 	 */
1815 	if (disp->dis_resfree != nullfree && dupcached == FALSE) {
1816 		(*disp->dis_resfree)(res);
1817 	}
1818 
1819 done:
1820 	/*
1821 	 * Free arguments struct
1822 	 */
1823 	if (disp) {
1824 		if (!SVC_FREEARGS(xprt, disp->dis_xdrargs, args)) {
1825 			cmn_err(CE_NOTE, "%s: bad freeargs", pgmname);
1826 			error++;
1827 		}
1828 	} else {
1829 		if (!SVC_FREEARGS(xprt, (xdrproc_t)0, (caddr_t)0)) {
1830 			cmn_err(CE_NOTE, "%s: bad freeargs", pgmname);
1831 			error++;
1832 		}
1833 	}
1834 
1835 	if (exi != NULL)
1836 		exi_rele(exi);
1837 
1838 	global_svstat_ptr[req->rq_vers][NFS_BADCALLS].value.ui64 += error;
1839 
1840 	global_svstat_ptr[req->rq_vers][NFS_CALLS].value.ui64++;
1841 }
1842 
1843 static void
1844 rfs_dispatch(struct svc_req *req, SVCXPRT *xprt)
1845 {
1846 	common_dispatch(req, xprt, NFS_VERSMIN, NFS_VERSMAX,
1847 	    "NFS", rfs_disptable);
1848 }
1849 
1850 static char *aclcallnames_v2[] = {
1851 	"ACL2_NULL",
1852 	"ACL2_GETACL",
1853 	"ACL2_SETACL",
1854 	"ACL2_GETATTR",
1855 	"ACL2_ACCESS",
1856 	"ACL2_GETXATTRDIR"
1857 };
1858 
1859 static struct rpcdisp acldisptab_v2[] = {
1860 	/*
1861 	 * ACL VERSION 2
1862 	 */
1863 
1864 	/* ACL2_NULL = 0 */
1865 	{rpc_null,
1866 	    xdr_void, NULL_xdrproc_t, 0,
1867 	    xdr_void, NULL_xdrproc_t, 0,
1868 	    nullfree, RPC_IDEMPOTENT,
1869 	    0},
1870 
1871 	/* ACL2_GETACL = 1 */
1872 	{acl2_getacl,
1873 	    xdr_GETACL2args, xdr_fastGETACL2args, sizeof (GETACL2args),
1874 	    xdr_GETACL2res, NULL_xdrproc_t, sizeof (GETACL2res),
1875 	    acl2_getacl_free, RPC_IDEMPOTENT,
1876 	    acl2_getacl_getfh},
1877 
1878 	/* ACL2_SETACL = 2 */
1879 	{acl2_setacl,
1880 	    xdr_SETACL2args, NULL_xdrproc_t, sizeof (SETACL2args),
1881 #ifdef _LITTLE_ENDIAN
1882 	    xdr_SETACL2res, xdr_fastSETACL2res, sizeof (SETACL2res),
1883 #else
1884 	    xdr_SETACL2res, NULL_xdrproc_t, sizeof (SETACL2res),
1885 #endif
1886 	    nullfree, RPC_MAPRESP,
1887 	    acl2_setacl_getfh},
1888 
1889 	/* ACL2_GETATTR = 3 */
1890 	{acl2_getattr,
1891 	    xdr_GETATTR2args, xdr_fastGETATTR2args, sizeof (GETATTR2args),
1892 #ifdef _LITTLE_ENDIAN
1893 	    xdr_GETATTR2res, xdr_fastGETATTR2res, sizeof (GETATTR2res),
1894 #else
1895 	    xdr_GETATTR2res, NULL_xdrproc_t, sizeof (GETATTR2res),
1896 #endif
1897 	    nullfree, RPC_IDEMPOTENT|RPC_ALLOWANON|RPC_MAPRESP,
1898 	    acl2_getattr_getfh},
1899 
1900 	/* ACL2_ACCESS = 4 */
1901 	{acl2_access,
1902 	    xdr_ACCESS2args, xdr_fastACCESS2args, sizeof (ACCESS2args),
1903 #ifdef _LITTLE_ENDIAN
1904 	    xdr_ACCESS2res, xdr_fastACCESS2res, sizeof (ACCESS2res),
1905 #else
1906 	    xdr_ACCESS2res, NULL_xdrproc_t, sizeof (ACCESS2res),
1907 #endif
1908 	    nullfree, RPC_IDEMPOTENT|RPC_MAPRESP,
1909 	    acl2_access_getfh},
1910 
1911 	/* ACL2_GETXATTRDIR = 5 */
1912 	{acl2_getxattrdir,
1913 	    xdr_GETXATTRDIR2args, NULL_xdrproc_t, sizeof (GETXATTRDIR2args),
1914 	    xdr_GETXATTRDIR2res, NULL_xdrproc_t, sizeof (GETXATTRDIR2res),
1915 	    nullfree, RPC_IDEMPOTENT,
1916 	    acl2_getxattrdir_getfh},
1917 };
1918 
1919 static char *aclcallnames_v3[] = {
1920 	"ACL3_NULL",
1921 	"ACL3_GETACL",
1922 	"ACL3_SETACL",
1923 	"ACL3_GETXATTRDIR"
1924 };
1925 
1926 static struct rpcdisp acldisptab_v3[] = {
1927 	/*
1928 	 * ACL VERSION 3
1929 	 */
1930 
1931 	/* ACL3_NULL = 0 */
1932 	{rpc_null,
1933 	    xdr_void, NULL_xdrproc_t, 0,
1934 	    xdr_void, NULL_xdrproc_t, 0,
1935 	    nullfree, RPC_IDEMPOTENT,
1936 	    0},
1937 
1938 	/* ACL3_GETACL = 1 */
1939 	{acl3_getacl,
1940 	    xdr_GETACL3args, NULL_xdrproc_t, sizeof (GETACL3args),
1941 	    xdr_GETACL3res, NULL_xdrproc_t, sizeof (GETACL3res),
1942 	    acl3_getacl_free, RPC_IDEMPOTENT,
1943 	    acl3_getacl_getfh},
1944 
1945 	/* ACL3_SETACL = 2 */
1946 	{acl3_setacl,
1947 	    xdr_SETACL3args, NULL_xdrproc_t, sizeof (SETACL3args),
1948 	    xdr_SETACL3res, NULL_xdrproc_t, sizeof (SETACL3res),
1949 	    nullfree, 0,
1950 	    acl3_setacl_getfh},
1951 
1952 	/* ACL3_GETXATTRDIR = 3 */
1953 	{acl3_getxattrdir,
1954 	    xdr_GETXATTRDIR3args, NULL_xdrproc_t, sizeof (GETXATTRDIR3args),
1955 	    xdr_GETXATTRDIR3res, NULL_xdrproc_t, sizeof (GETXATTRDIR3res),
1956 	    nullfree, RPC_IDEMPOTENT,
1957 	    acl3_getxattrdir_getfh},
1958 };
1959 
1960 static struct rpc_disptable acl_disptable[] = {
1961 	{sizeof (acldisptab_v2) / sizeof (acldisptab_v2[0]),
1962 		aclcallnames_v2,
1963 		&aclproccnt_v2_ptr, acldisptab_v2},
1964 	{sizeof (acldisptab_v3) / sizeof (acldisptab_v3[0]),
1965 		aclcallnames_v3,
1966 		&aclproccnt_v3_ptr, acldisptab_v3},
1967 };
1968 
1969 static void
1970 acl_dispatch(struct svc_req *req, SVCXPRT *xprt)
1971 {
1972 	common_dispatch(req, xprt, NFS_ACL_VERSMIN, NFS_ACL_VERSMAX,
1973 	    "ACL", acl_disptable);
1974 }
1975 
1976 int
1977 checkwin(int flavor, int window, struct svc_req *req)
1978 {
1979 	struct authdes_cred *adc;
1980 
1981 	switch (flavor) {
1982 	case AUTH_DES:
1983 		adc = (struct authdes_cred *)req->rq_clntcred;
1984 		if (adc->adc_fullname.window > window)
1985 			return (0);
1986 		break;
1987 
1988 	default:
1989 		break;
1990 	}
1991 	return (1);
1992 }
1993 
1994 
1995 /*
1996  * checkauth() will check the access permission against the export
1997  * information.  Then map root uid/gid to appropriate uid/gid.
1998  *
1999  * This routine is used by NFS V3 and V2 code.
2000  */
2001 static int
2002 checkauth(struct exportinfo *exi, struct svc_req *req, cred_t *cr, int anon_ok,
2003     bool_t publicfh_ok)
2004 {
2005 	int i, nfsflavor, rpcflavor, stat, access;
2006 	struct secinfo *secp;
2007 	caddr_t principal;
2008 	char buf[INET6_ADDRSTRLEN]; /* to hold both IPv4 and IPv6 addr */
2009 	int anon_res = 0;
2010 
2011 	/*
2012 	 * Check for privileged port number
2013 	 * N.B.:  this assumes that we know the format of a netbuf.
2014 	 */
2015 	if (nfs_portmon) {
2016 		struct sockaddr *ca;
2017 		ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2018 
2019 		if (ca == NULL)
2020 			return (0);
2021 
2022 		if ((ca->sa_family == AF_INET &&
2023 		    ntohs(((struct sockaddr_in *)ca)->sin_port) >=
2024 		    IPPORT_RESERVED) ||
2025 		    (ca->sa_family == AF_INET6 &&
2026 		    ntohs(((struct sockaddr_in6 *)ca)->sin6_port) >=
2027 		    IPPORT_RESERVED)) {
2028 			cmn_err(CE_NOTE,
2029 			    "nfs_server: client %s%ssent NFS request from "
2030 			    "unprivileged port",
2031 			    client_name(req), client_addr(req, buf));
2032 			return (0);
2033 		}
2034 	}
2035 
2036 	/*
2037 	 *  return 1 on success or 0 on failure
2038 	 */
2039 	stat = sec_svc_getcred(req, cr, &principal, &nfsflavor);
2040 
2041 	/*
2042 	 * A failed AUTH_UNIX svc_get_cred() implies we couldn't set
2043 	 * the credentials; below we map that to anonymous.
2044 	 */
2045 	if (!stat && nfsflavor != AUTH_UNIX) {
2046 		cmn_err(CE_NOTE,
2047 		    "nfs_server: couldn't get unix cred for %s",
2048 		    client_name(req));
2049 		return (0);
2050 	}
2051 
2052 	/*
2053 	 * Short circuit checkauth() on operations that support the
2054 	 * public filehandle, and if the request for that operation
2055 	 * is using the public filehandle. Note that we must call
2056 	 * sec_svc_getcred() first so that xp_cookie is set to the
2057 	 * right value. Normally xp_cookie is just the RPC flavor
2058 	 * of the the request, but in the case of RPCSEC_GSS it
2059 	 * could be a pseudo flavor.
2060 	 */
2061 	if (publicfh_ok)
2062 		return (1);
2063 
2064 	rpcflavor = req->rq_cred.oa_flavor;
2065 	/*
2066 	 * Check if the auth flavor is valid for this export
2067 	 */
2068 	access = nfsauth_access(exi, req);
2069 	if (access & NFSAUTH_DROP)
2070 		return (-1);	/* drop the request */
2071 
2072 	if (access & NFSAUTH_DENIED) {
2073 		/*
2074 		 * If anon_ok == 1 and we got NFSAUTH_DENIED, it was
2075 		 * probably due to the flavor not matching during the
2076 		 * the mount attempt. So map the flavor to AUTH_NONE
2077 		 * so that the credentials get mapped to the anonymous
2078 		 * user.
2079 		 */
2080 		if (anon_ok == 1)
2081 			rpcflavor = AUTH_NONE;
2082 		else
2083 			return (0);	/* deny access */
2084 
2085 	} else if (access & NFSAUTH_MAPNONE) {
2086 		/*
2087 		 * Access was granted even though the flavor mismatched
2088 		 * because AUTH_NONE was one of the exported flavors.
2089 		 */
2090 		rpcflavor = AUTH_NONE;
2091 
2092 	} else if (access & NFSAUTH_WRONGSEC) {
2093 		/*
2094 		 * NFSAUTH_WRONGSEC is used for NFSv4. If we get here,
2095 		 * it means a client ignored the list of allowed flavors
2096 		 * returned via the MOUNT protocol. So we just disallow it!
2097 		 */
2098 		return (0);
2099 	}
2100 
2101 	switch (rpcflavor) {
2102 	case AUTH_NONE:
2103 		anon_res = crsetugid(cr, exi->exi_export.ex_anon,
2104 		    exi->exi_export.ex_anon);
2105 		(void) crsetgroups(cr, 0, NULL);
2106 		break;
2107 
2108 	case AUTH_UNIX:
2109 		if (!stat || crgetuid(cr) == 0 && !(access & NFSAUTH_ROOT)) {
2110 			anon_res = crsetugid(cr, exi->exi_export.ex_anon,
2111 			    exi->exi_export.ex_anon);
2112 			(void) crsetgroups(cr, 0, NULL);
2113 		} else if (!stat || crgetuid(cr) == 0 &&
2114 		    access & NFSAUTH_ROOT) {
2115 			/*
2116 			 * It is root, so apply rootid to get real UID
2117 			 * Find the secinfo structure.  We should be able
2118 			 * to find it by the time we reach here.
2119 			 * nfsauth_access() has done the checking.
2120 			 */
2121 			secp = NULL;
2122 			for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2123 				struct secinfo *sptr;
2124 				sptr = &exi->exi_export.ex_secinfo[i];
2125 				if (sptr->s_secinfo.sc_nfsnum == nfsflavor) {
2126 					secp = sptr;
2127 					break;
2128 				}
2129 			}
2130 			if (secp != NULL) {
2131 				(void) crsetugid(cr, secp->s_rootid,
2132 				    secp->s_rootid);
2133 				(void) crsetgroups(cr, 0, NULL);
2134 			}
2135 		}
2136 		break;
2137 
2138 	case AUTH_DES:
2139 	case RPCSEC_GSS:
2140 		/*
2141 		 *  Find the secinfo structure.  We should be able
2142 		 *  to find it by the time we reach here.
2143 		 *  nfsauth_access() has done the checking.
2144 		 */
2145 		secp = NULL;
2146 		for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2147 			if (exi->exi_export.ex_secinfo[i].s_secinfo.sc_nfsnum ==
2148 			    nfsflavor) {
2149 				secp = &exi->exi_export.ex_secinfo[i];
2150 				break;
2151 			}
2152 		}
2153 
2154 		if (!secp) {
2155 			cmn_err(CE_NOTE, "nfs_server: client %s%shad "
2156 			    "no secinfo data for flavor %d",
2157 			    client_name(req), client_addr(req, buf),
2158 			    nfsflavor);
2159 			return (0);
2160 		}
2161 
2162 		if (!checkwin(rpcflavor, secp->s_window, req)) {
2163 			cmn_err(CE_NOTE,
2164 			    "nfs_server: client %s%sused invalid "
2165 			    "auth window value",
2166 			    client_name(req), client_addr(req, buf));
2167 			return (0);
2168 		}
2169 
2170 		/*
2171 		 * Map root principals listed in the share's root= list to root,
2172 		 * and map any others principals that were mapped to root by RPC
2173 		 * to anon.
2174 		 */
2175 		if (principal && sec_svc_inrootlist(rpcflavor, principal,
2176 		    secp->s_rootcnt, secp->s_rootnames)) {
2177 			if (crgetuid(cr) == 0 && secp->s_rootid == 0)
2178 				return (1);
2179 
2180 
2181 			(void) crsetugid(cr, secp->s_rootid, secp->s_rootid);
2182 
2183 			/*
2184 			 * NOTE: If and when kernel-land privilege tracing is
2185 			 * added this may have to be replaced with code that
2186 			 * retrieves root's supplementary groups (e.g., using
2187 			 * kgss_get_group_info().  In the meantime principals
2188 			 * mapped to uid 0 get all privileges, so setting cr's
2189 			 * supplementary groups for them does nothing.
2190 			 */
2191 			(void) crsetgroups(cr, 0, NULL);
2192 
2193 			return (1);
2194 		}
2195 
2196 		/*
2197 		 * Not a root princ, or not in root list, map UID 0/nobody to
2198 		 * the anon ID for the share.  (RPC sets cr's UIDs and GIDs to
2199 		 * UID_NOBODY and GID_NOBODY, respectively.)
2200 		 */
2201 		if (crgetuid(cr) != 0 &&
2202 		    (crgetuid(cr) != UID_NOBODY || crgetgid(cr) != GID_NOBODY))
2203 			return (1);
2204 
2205 		anon_res = crsetugid(cr, exi->exi_export.ex_anon,
2206 		    exi->exi_export.ex_anon);
2207 		(void) crsetgroups(cr, 0, NULL);
2208 		break;
2209 	default:
2210 		return (0);
2211 	} /* switch on rpcflavor */
2212 
2213 	/*
2214 	 * Even if anon access is disallowed via ex_anon == -1, we allow
2215 	 * this access if anon_ok is set.  So set creds to the default
2216 	 * "nobody" id.
2217 	 */
2218 	if (anon_res != 0) {
2219 		if (anon_ok == 0) {
2220 			cmn_err(CE_NOTE,
2221 			    "nfs_server: client %s%ssent wrong "
2222 			    "authentication for %s",
2223 			    client_name(req), client_addr(req, buf),
2224 			    exi->exi_export.ex_path ?
2225 			    exi->exi_export.ex_path : "?");
2226 			return (0);
2227 		}
2228 
2229 		if (crsetugid(cr, UID_NOBODY, GID_NOBODY) != 0)
2230 			return (0);
2231 	}
2232 
2233 	return (1);
2234 }
2235 
2236 /*
2237  * returns 0 on failure, -1 on a drop, -2 on wrong security flavor,
2238  * and 1 on success
2239  */
2240 int
2241 checkauth4(struct compound_state *cs, struct svc_req *req)
2242 {
2243 	int i, rpcflavor, access;
2244 	struct secinfo *secp;
2245 	char buf[MAXHOST + 1];
2246 	int anon_res = 0, nfsflavor;
2247 	struct exportinfo *exi;
2248 	cred_t	*cr;
2249 	caddr_t	principal;
2250 
2251 	exi = cs->exi;
2252 	cr = cs->cr;
2253 	principal = cs->principal;
2254 	nfsflavor = cs->nfsflavor;
2255 
2256 	ASSERT(cr != NULL);
2257 
2258 	rpcflavor = req->rq_cred.oa_flavor;
2259 	cs->access &= ~CS_ACCESS_LIMITED;
2260 
2261 	/*
2262 	 * Check for privileged port number
2263 	 * N.B.:  this assumes that we know the format of a netbuf.
2264 	 */
2265 	if (nfs_portmon) {
2266 		struct sockaddr *ca;
2267 		ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2268 
2269 		if (ca == NULL)
2270 			return (0);
2271 
2272 		if ((ca->sa_family == AF_INET &&
2273 		    ntohs(((struct sockaddr_in *)ca)->sin_port) >=
2274 		    IPPORT_RESERVED) ||
2275 		    (ca->sa_family == AF_INET6 &&
2276 		    ntohs(((struct sockaddr_in6 *)ca)->sin6_port) >=
2277 		    IPPORT_RESERVED)) {
2278 			cmn_err(CE_NOTE,
2279 			    "nfs_server: client %s%ssent NFSv4 request from "
2280 			    "unprivileged port",
2281 			    client_name(req), client_addr(req, buf));
2282 			return (0);
2283 		}
2284 	}
2285 
2286 	/*
2287 	 * Check the access right per auth flavor on the vnode of
2288 	 * this export for the given request.
2289 	 */
2290 	access = nfsauth4_access(cs->exi, cs->vp, req);
2291 
2292 	if (access & NFSAUTH_WRONGSEC)
2293 		return (-2);	/* no access for this security flavor */
2294 
2295 	if (access & NFSAUTH_DROP)
2296 		return (-1);	/* drop the request */
2297 
2298 	if (access & NFSAUTH_DENIED) {
2299 
2300 		if (exi->exi_export.ex_seccnt > 0)
2301 			return (0);	/* deny access */
2302 
2303 	} else if (access & NFSAUTH_LIMITED) {
2304 
2305 		cs->access |= CS_ACCESS_LIMITED;
2306 
2307 	} else if (access & NFSAUTH_MAPNONE) {
2308 		/*
2309 		 * Access was granted even though the flavor mismatched
2310 		 * because AUTH_NONE was one of the exported flavors.
2311 		 */
2312 		rpcflavor = AUTH_NONE;
2313 	}
2314 
2315 	/*
2316 	 * XXX probably need to redo some of it for nfsv4?
2317 	 * return 1 on success or 0 on failure
2318 	 */
2319 
2320 	switch (rpcflavor) {
2321 	case AUTH_NONE:
2322 		anon_res = crsetugid(cr, exi->exi_export.ex_anon,
2323 		    exi->exi_export.ex_anon);
2324 		(void) crsetgroups(cr, 0, NULL);
2325 		break;
2326 
2327 	case AUTH_UNIX:
2328 		if (crgetuid(cr) == 0 && !(access & NFSAUTH_ROOT)) {
2329 			anon_res = crsetugid(cr, exi->exi_export.ex_anon,
2330 			    exi->exi_export.ex_anon);
2331 			(void) crsetgroups(cr, 0, NULL);
2332 		} else if (crgetuid(cr) == 0 && access & NFSAUTH_ROOT) {
2333 			/*
2334 			 * It is root, so apply rootid to get real UID
2335 			 * Find the secinfo structure.  We should be able
2336 			 * to find it by the time we reach here.
2337 			 * nfsauth_access() has done the checking.
2338 			 */
2339 			secp = NULL;
2340 			for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2341 				struct secinfo *sptr;
2342 				sptr = &exi->exi_export.ex_secinfo[i];
2343 				if (sptr->s_secinfo.sc_nfsnum == nfsflavor) {
2344 					secp = &exi->exi_export.ex_secinfo[i];
2345 					break;
2346 				}
2347 			}
2348 			if (secp != NULL) {
2349 				(void) crsetugid(cr, secp->s_rootid,
2350 				    secp->s_rootid);
2351 				(void) crsetgroups(cr, 0, NULL);
2352 			}
2353 		}
2354 		break;
2355 
2356 	default:
2357 		/*
2358 		 *  Find the secinfo structure.  We should be able
2359 		 *  to find it by the time we reach here.
2360 		 *  nfsauth_access() has done the checking.
2361 		 */
2362 		secp = NULL;
2363 		for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2364 			if (exi->exi_export.ex_secinfo[i].s_secinfo.sc_nfsnum ==
2365 			    nfsflavor) {
2366 				secp = &exi->exi_export.ex_secinfo[i];
2367 				break;
2368 			}
2369 		}
2370 
2371 		if (!secp) {
2372 			cmn_err(CE_NOTE, "nfs_server: client %s%shad "
2373 			    "no secinfo data for flavor %d",
2374 			    client_name(req), client_addr(req, buf),
2375 			    nfsflavor);
2376 			return (0);
2377 		}
2378 
2379 		if (!checkwin(rpcflavor, secp->s_window, req)) {
2380 			cmn_err(CE_NOTE,
2381 			    "nfs_server: client %s%sused invalid "
2382 			    "auth window value",
2383 			    client_name(req), client_addr(req, buf));
2384 			return (0);
2385 		}
2386 
2387 		/*
2388 		 * Map root principals listed in the share's root= list to root,
2389 		 * and map any others principals that were mapped to root by RPC
2390 		 * to anon. If not going to anon, set to rootid (root_mapping).
2391 		 */
2392 		if (principal && sec_svc_inrootlist(rpcflavor, principal,
2393 		    secp->s_rootcnt, secp->s_rootnames)) {
2394 			if (crgetuid(cr) == 0 && secp->s_rootid == 0)
2395 				return (1);
2396 
2397 			(void) crsetugid(cr, secp->s_rootid, secp->s_rootid);
2398 
2399 			/*
2400 			 * NOTE: If and when kernel-land privilege tracing is
2401 			 * added this may have to be replaced with code that
2402 			 * retrieves root's supplementary groups (e.g., using
2403 			 * kgss_get_group_info().  In the meantime principals
2404 			 * mapped to uid 0 get all privileges, so setting cr's
2405 			 * supplementary groups for them does nothing.
2406 			 */
2407 			(void) crsetgroups(cr, 0, NULL);
2408 
2409 			return (1);
2410 		}
2411 
2412 		/*
2413 		 * Not a root princ, or not in root list, map UID 0/nobody to
2414 		 * the anon ID for the share.  (RPC sets cr's UIDs and GIDs to
2415 		 * UID_NOBODY and GID_NOBODY, respectively.)
2416 		 */
2417 		if (crgetuid(cr) != 0 &&
2418 		    (crgetuid(cr) != UID_NOBODY || crgetgid(cr) != GID_NOBODY))
2419 			return (1);
2420 
2421 		anon_res = crsetugid(cr, exi->exi_export.ex_anon,
2422 		    exi->exi_export.ex_anon);
2423 		(void) crsetgroups(cr, 0, NULL);
2424 		break;
2425 	} /* switch on rpcflavor */
2426 
2427 	/*
2428 	 * Even if anon access is disallowed via ex_anon == -1, we allow
2429 	 * this access if anon_ok is set.  So set creds to the default
2430 	 * "nobody" id.
2431 	 */
2432 
2433 	if (anon_res != 0) {
2434 		cmn_err(CE_NOTE,
2435 		    "nfs_server: client %s%ssent wrong "
2436 		    "authentication for %s",
2437 		    client_name(req), client_addr(req, buf),
2438 		    exi->exi_export.ex_path ?
2439 		    exi->exi_export.ex_path : "?");
2440 		return (0);
2441 	}
2442 
2443 	return (1);
2444 }
2445 
2446 
2447 static char *
2448 client_name(struct svc_req *req)
2449 {
2450 	char *hostname = NULL;
2451 
2452 	/*
2453 	 * If it's a Unix cred then use the
2454 	 * hostname from the credential.
2455 	 */
2456 	if (req->rq_cred.oa_flavor == AUTH_UNIX) {
2457 		hostname = ((struct authunix_parms *)
2458 		    req->rq_clntcred)->aup_machname;
2459 	}
2460 	if (hostname == NULL)
2461 		hostname = "";
2462 
2463 	return (hostname);
2464 }
2465 
2466 static char *
2467 client_addr(struct svc_req *req, char *buf)
2468 {
2469 	struct sockaddr *ca;
2470 	uchar_t *b;
2471 	char *frontspace = "";
2472 
2473 	/*
2474 	 * We assume we are called in tandem with client_name and the
2475 	 * format string looks like "...client %s%sblah blah..."
2476 	 *
2477 	 * If it's a Unix cred then client_name returned
2478 	 * a host name, so we need insert a space between host name
2479 	 * and IP address.
2480 	 */
2481 	if (req->rq_cred.oa_flavor == AUTH_UNIX)
2482 		frontspace = " ";
2483 
2484 	/*
2485 	 * Convert the caller's IP address to a dotted string
2486 	 */
2487 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2488 
2489 	if (ca->sa_family == AF_INET) {
2490 		b = (uchar_t *)&((struct sockaddr_in *)ca)->sin_addr;
2491 		(void) sprintf(buf, "%s(%d.%d.%d.%d) ", frontspace,
2492 		    b[0] & 0xFF, b[1] & 0xFF, b[2] & 0xFF, b[3] & 0xFF);
2493 	} else if (ca->sa_family == AF_INET6) {
2494 		struct sockaddr_in6 *sin6;
2495 		sin6 = (struct sockaddr_in6 *)ca;
2496 		(void) kinet_ntop6((uchar_t *)&sin6->sin6_addr,
2497 		    buf, INET6_ADDRSTRLEN);
2498 
2499 	} else {
2500 
2501 		/*
2502 		 * No IP address to print. If there was a host name
2503 		 * printed, then we print a space.
2504 		 */
2505 		(void) sprintf(buf, frontspace);
2506 	}
2507 
2508 	return (buf);
2509 }
2510 
2511 /*
2512  * NFS Server initialization routine.  This routine should only be called
2513  * once.  It performs the following tasks:
2514  *	- Call sub-initialization routines (localize access to variables)
2515  *	- Initialize all locks
2516  *	- initialize the version 3 write verifier
2517  */
2518 int
2519 nfs_srvinit(void)
2520 {
2521 	int error;
2522 
2523 	error = nfs_exportinit();
2524 	if (error != 0)
2525 		return (error);
2526 	error = rfs4_srvrinit();
2527 	if (error != 0) {
2528 		nfs_exportfini();
2529 		return (error);
2530 	}
2531 	rfs_srvrinit();
2532 	rfs3_srvrinit();
2533 	nfsauth_init();
2534 
2535 	/* Init the stuff to control start/stop */
2536 	nfs_server_upordown = NFS_SERVER_STOPPED;
2537 	mutex_init(&nfs_server_upordown_lock, NULL, MUTEX_DEFAULT, NULL);
2538 	cv_init(&nfs_server_upordown_cv, NULL, CV_DEFAULT, NULL);
2539 	mutex_init(&rdma_wait_mutex, NULL, MUTEX_DEFAULT, NULL);
2540 	cv_init(&rdma_wait_cv, NULL, CV_DEFAULT, NULL);
2541 
2542 	return (0);
2543 }
2544 
2545 /*
2546  * NFS Server finalization routine. This routine is called to cleanup the
2547  * initialization work previously performed if the NFS server module could
2548  * not be loaded correctly.
2549  */
2550 void
2551 nfs_srvfini(void)
2552 {
2553 	nfsauth_fini();
2554 	rfs3_srvrfini();
2555 	rfs_srvrfini();
2556 	nfs_exportfini();
2557 
2558 	mutex_destroy(&nfs_server_upordown_lock);
2559 	cv_destroy(&nfs_server_upordown_cv);
2560 	mutex_destroy(&rdma_wait_mutex);
2561 	cv_destroy(&rdma_wait_cv);
2562 }
2563 
2564 /*
2565  * Set up an iovec array of up to cnt pointers.
2566  */
2567 
2568 void
2569 mblk_to_iov(mblk_t *m, int cnt, struct iovec *iovp)
2570 {
2571 	while (m != NULL && cnt-- > 0) {
2572 		iovp->iov_base = (caddr_t)m->b_rptr;
2573 		iovp->iov_len = (m->b_wptr - m->b_rptr);
2574 		iovp++;
2575 		m = m->b_cont;
2576 	}
2577 }
2578 
2579 /*
2580  * Common code between NFS Version 2 and NFS Version 3 for the public
2581  * filehandle multicomponent lookups.
2582  */
2583 
2584 /*
2585  * Public filehandle evaluation of a multi-component lookup, following
2586  * symbolic links, if necessary. This may result in a vnode in another
2587  * filesystem, which is OK as long as the other filesystem is exported.
2588  *
2589  * Note that the exi will be set either to NULL or a new reference to the
2590  * exportinfo struct that corresponds to the vnode of the multi-component path.
2591  * It is the callers responsibility to release this reference.
2592  */
2593 int
2594 rfs_publicfh_mclookup(char *p, vnode_t *dvp, cred_t *cr, vnode_t **vpp,
2595     struct exportinfo **exi, struct sec_ol *sec)
2596 {
2597 	int pathflag;
2598 	vnode_t *mc_dvp = NULL;
2599 	vnode_t *realvp;
2600 	int error;
2601 
2602 	*exi = NULL;
2603 
2604 	/*
2605 	 * check if the given path is a url or native path. Since p is
2606 	 * modified by MCLpath(), it may be empty after returning from
2607 	 * there, and should be checked.
2608 	 */
2609 	if ((pathflag = MCLpath(&p)) == -1)
2610 		return (EIO);
2611 
2612 	/*
2613 	 * If pathflag is SECURITY_QUERY, turn the SEC_QUERY bit
2614 	 * on in sec->sec_flags. This bit will later serve as an
2615 	 * indication in makefh_ol() or makefh3_ol() to overload the
2616 	 * filehandle to contain the sec modes used by the server for
2617 	 * the path.
2618 	 */
2619 	if (pathflag == SECURITY_QUERY) {
2620 		if ((sec->sec_index = (uint_t)(*p)) > 0) {
2621 			sec->sec_flags |= SEC_QUERY;
2622 			p++;
2623 			if ((pathflag = MCLpath(&p)) == -1)
2624 				return (EIO);
2625 		} else {
2626 			cmn_err(CE_NOTE,
2627 			    "nfs_server: invalid security index %d, "
2628 			    "violating WebNFS SNEGO protocol.", sec->sec_index);
2629 			return (EIO);
2630 		}
2631 	}
2632 
2633 	if (p[0] == '\0') {
2634 		error = ENOENT;
2635 		goto publicfh_done;
2636 	}
2637 
2638 	error = rfs_pathname(p, &mc_dvp, vpp, dvp, cr, pathflag);
2639 
2640 	/*
2641 	 * If name resolves to "/" we get EINVAL since we asked for
2642 	 * the vnode of the directory that the file is in. Try again
2643 	 * with NULL directory vnode.
2644 	 */
2645 	if (error == EINVAL) {
2646 		error = rfs_pathname(p, NULL, vpp, dvp, cr, pathflag);
2647 		if (!error) {
2648 			ASSERT(*vpp != NULL);
2649 			if ((*vpp)->v_type == VDIR) {
2650 				VN_HOLD(*vpp);
2651 				mc_dvp = *vpp;
2652 			} else {
2653 				/*
2654 				 * This should not happen, the filesystem is
2655 				 * in an inconsistent state. Fail the lookup
2656 				 * at this point.
2657 				 */
2658 				VN_RELE(*vpp);
2659 				error = EINVAL;
2660 			}
2661 		}
2662 	}
2663 
2664 	if (error)
2665 		goto publicfh_done;
2666 
2667 	if (*vpp == NULL) {
2668 		error = ENOENT;
2669 		goto publicfh_done;
2670 	}
2671 
2672 	ASSERT(mc_dvp != NULL);
2673 	ASSERT(*vpp != NULL);
2674 
2675 	if ((*vpp)->v_type == VDIR) {
2676 		do {
2677 			/*
2678 			 * *vpp may be an AutoFS node, so we perform
2679 			 * a VOP_ACCESS() to trigger the mount of the intended
2680 			 * filesystem, so we can perform the lookup in the
2681 			 * intended filesystem.
2682 			 */
2683 			(void) VOP_ACCESS(*vpp, 0, 0, cr, NULL);
2684 
2685 			/*
2686 			 * If vnode is covered, get the
2687 			 * the topmost vnode.
2688 			 */
2689 			if (vn_mountedvfs(*vpp) != NULL) {
2690 				error = traverse(vpp);
2691 				if (error) {
2692 					VN_RELE(*vpp);
2693 					goto publicfh_done;
2694 				}
2695 			}
2696 
2697 			if (VOP_REALVP(*vpp, &realvp, NULL) == 0 &&
2698 			    realvp != *vpp) {
2699 				/*
2700 				 * If realvp is different from *vpp
2701 				 * then release our reference on *vpp, so that
2702 				 * the export access check be performed on the
2703 				 * real filesystem instead.
2704 				 */
2705 				VN_HOLD(realvp);
2706 				VN_RELE(*vpp);
2707 				*vpp = realvp;
2708 			} else {
2709 				break;
2710 			}
2711 		/* LINTED */
2712 		} while (TRUE);
2713 
2714 		/*
2715 		 * Let nfs_vptexi() figure what the real parent is.
2716 		 */
2717 		VN_RELE(mc_dvp);
2718 		mc_dvp = NULL;
2719 
2720 	} else {
2721 		/*
2722 		 * If vnode is covered, get the
2723 		 * the topmost vnode.
2724 		 */
2725 		if (vn_mountedvfs(mc_dvp) != NULL) {
2726 			error = traverse(&mc_dvp);
2727 			if (error) {
2728 				VN_RELE(*vpp);
2729 				goto publicfh_done;
2730 			}
2731 		}
2732 
2733 		if (VOP_REALVP(mc_dvp, &realvp, NULL) == 0 &&
2734 		    realvp != mc_dvp) {
2735 			/*
2736 			 * *vpp is a file, obtain realvp of the parent
2737 			 * directory vnode.
2738 			 */
2739 			VN_HOLD(realvp);
2740 			VN_RELE(mc_dvp);
2741 			mc_dvp = realvp;
2742 		}
2743 	}
2744 
2745 	/*
2746 	 * The pathname may take us from the public filesystem to another.
2747 	 * If that's the case then just set the exportinfo to the new export
2748 	 * and build filehandle for it. Thanks to per-access checking there's
2749 	 * no security issues with doing this. If the client is not allowed
2750 	 * access to this new export then it will get an access error when it
2751 	 * tries to use the filehandle
2752 	 */
2753 	if (error = nfs_check_vpexi(mc_dvp, *vpp, kcred, exi)) {
2754 		VN_RELE(*vpp);
2755 		goto publicfh_done;
2756 	}
2757 
2758 	/*
2759 	 * Not allowed access to pseudo exports.
2760 	 */
2761 	if (PSEUDO(*exi)) {
2762 		error = ENOENT;
2763 		VN_RELE(*vpp);
2764 		goto publicfh_done;
2765 	}
2766 
2767 	/*
2768 	 * Do a lookup for the index file. We know the index option doesn't
2769 	 * allow paths through handling in the share command, so mc_dvp will
2770 	 * be the parent for the index file vnode, if its present. Use
2771 	 * temporary pointers to preserve and reuse the vnode pointers of the
2772 	 * original directory in case there's no index file. Note that the
2773 	 * index file is a native path, and should not be interpreted by
2774 	 * the URL parser in rfs_pathname()
2775 	 */
2776 	if (((*exi)->exi_export.ex_flags & EX_INDEX) &&
2777 	    ((*vpp)->v_type == VDIR) && (pathflag == URLPATH)) {
2778 		vnode_t *tvp, *tmc_dvp;	/* temporary vnode pointers */
2779 
2780 		tmc_dvp = mc_dvp;
2781 		mc_dvp = tvp = *vpp;
2782 
2783 		error = rfs_pathname((*exi)->exi_export.ex_index, NULL, vpp,
2784 		    mc_dvp, cr, NATIVEPATH);
2785 
2786 		if (error == ENOENT) {
2787 			*vpp = tvp;
2788 			mc_dvp = tmc_dvp;
2789 			error = 0;
2790 		} else {	/* ok or error other than ENOENT */
2791 			if (tmc_dvp)
2792 				VN_RELE(tmc_dvp);
2793 			if (error)
2794 				goto publicfh_done;
2795 
2796 			/*
2797 			 * Found a valid vp for index "filename". Sanity check
2798 			 * for odd case where a directory is provided as index
2799 			 * option argument and leads us to another filesystem
2800 			 */
2801 
2802 			/* Release the reference on the old exi value */
2803 			ASSERT(*exi != NULL);
2804 			exi_rele(*exi);
2805 
2806 			if (error = nfs_check_vpexi(mc_dvp, *vpp, kcred, exi)) {
2807 				VN_RELE(*vpp);
2808 				goto publicfh_done;
2809 			}
2810 		}
2811 	}
2812 
2813 publicfh_done:
2814 	if (mc_dvp)
2815 		VN_RELE(mc_dvp);
2816 
2817 	return (error);
2818 }
2819 
2820 /*
2821  * Evaluate a multi-component path
2822  */
2823 int
2824 rfs_pathname(
2825 	char *path,			/* pathname to evaluate */
2826 	vnode_t **dirvpp,		/* ret for ptr to parent dir vnode */
2827 	vnode_t **compvpp,		/* ret for ptr to component vnode */
2828 	vnode_t *startdvp,		/* starting vnode */
2829 	cred_t *cr,			/* user's credential */
2830 	int pathflag)			/* flag to identify path, e.g. URL */
2831 {
2832 	char namebuf[TYPICALMAXPATHLEN];
2833 	struct pathname pn;
2834 	int error;
2835 
2836 	/*
2837 	 * If pathname starts with '/', then set startdvp to root.
2838 	 */
2839 	if (*path == '/') {
2840 		while (*path == '/')
2841 			path++;
2842 
2843 		startdvp = rootdir;
2844 	}
2845 
2846 	error = pn_get_buf(path, UIO_SYSSPACE, &pn, namebuf, sizeof (namebuf));
2847 	if (error == 0) {
2848 		/*
2849 		 * Call the URL parser for URL paths to modify the original
2850 		 * string to handle any '%' encoded characters that exist.
2851 		 * Done here to avoid an extra bcopy in the lookup.
2852 		 * We need to be careful about pathlen's. We know that
2853 		 * rfs_pathname() is called with a non-empty path. However,
2854 		 * it could be emptied due to the path simply being all /'s,
2855 		 * which is valid to proceed with the lookup, or due to the
2856 		 * URL parser finding an encoded null character at the
2857 		 * beginning of path which should not proceed with the lookup.
2858 		 */
2859 		if (pn.pn_pathlen != 0 && pathflag == URLPATH) {
2860 			URLparse(pn.pn_path);
2861 			if ((pn.pn_pathlen = strlen(pn.pn_path)) == 0)
2862 				return (ENOENT);
2863 		}
2864 		VN_HOLD(startdvp);
2865 		error = lookuppnvp(&pn, NULL, NO_FOLLOW, dirvpp, compvpp,
2866 		    rootdir, startdvp, cr);
2867 	}
2868 	if (error == ENAMETOOLONG) {
2869 		/*
2870 		 * This thread used a pathname > TYPICALMAXPATHLEN bytes long.
2871 		 */
2872 		if (error = pn_get(path, UIO_SYSSPACE, &pn))
2873 			return (error);
2874 		if (pn.pn_pathlen != 0 && pathflag == URLPATH) {
2875 			URLparse(pn.pn_path);
2876 			if ((pn.pn_pathlen = strlen(pn.pn_path)) == 0) {
2877 				pn_free(&pn);
2878 				return (ENOENT);
2879 			}
2880 		}
2881 		VN_HOLD(startdvp);
2882 		error = lookuppnvp(&pn, NULL, NO_FOLLOW, dirvpp, compvpp,
2883 		    rootdir, startdvp, cr);
2884 		pn_free(&pn);
2885 	}
2886 
2887 	return (error);
2888 }
2889 
2890 /*
2891  * Adapt the multicomponent lookup path depending on the pathtype
2892  */
2893 static int
2894 MCLpath(char **path)
2895 {
2896 	unsigned char c = (unsigned char)**path;
2897 
2898 	/*
2899 	 * If the MCL path is between 0x20 and 0x7E (graphic printable
2900 	 * character of the US-ASCII coded character set), its a URL path,
2901 	 * per RFC 1738.
2902 	 */
2903 	if (c >= 0x20 && c <= 0x7E)
2904 		return (URLPATH);
2905 
2906 	/*
2907 	 * If the first octet of the MCL path is not an ASCII character
2908 	 * then it must be interpreted as a tag value that describes the
2909 	 * format of the remaining octets of the MCL path.
2910 	 *
2911 	 * If the first octet of the MCL path is 0x81 it is a query
2912 	 * for the security info.
2913 	 */
2914 	switch (c) {
2915 	case 0x80:	/* native path, i.e. MCL via mount protocol */
2916 		(*path)++;
2917 		return (NATIVEPATH);
2918 	case 0x81:	/* security query */
2919 		(*path)++;
2920 		return (SECURITY_QUERY);
2921 	default:
2922 		return (-1);
2923 	}
2924 }
2925 
2926 #define	fromhex(c)  ((c >= '0' && c <= '9') ? (c - '0') : \
2927 			((c >= 'A' && c <= 'F') ? (c - 'A' + 10) :\
2928 			((c >= 'a' && c <= 'f') ? (c - 'a' + 10) : 0)))
2929 
2930 /*
2931  * The implementation of URLparse guarantees that the final string will
2932  * fit in the original one. Replaces '%' occurrences followed by 2 characters
2933  * with its corresponding hexadecimal character.
2934  */
2935 static void
2936 URLparse(char *str)
2937 {
2938 	char *p, *q;
2939 
2940 	p = q = str;
2941 	while (*p) {
2942 		*q = *p;
2943 		if (*p++ == '%') {
2944 			if (*p) {
2945 				*q = fromhex(*p) * 16;
2946 				p++;
2947 				if (*p) {
2948 					*q += fromhex(*p);
2949 					p++;
2950 				}
2951 			}
2952 		}
2953 		q++;
2954 	}
2955 	*q = '\0';
2956 }
2957 
2958 
2959 /*
2960  * Get the export information for the lookup vnode, and verify its
2961  * useable.
2962  */
2963 int
2964 nfs_check_vpexi(vnode_t *mc_dvp, vnode_t *vp, cred_t *cr,
2965     struct exportinfo **exi)
2966 {
2967 	int walk;
2968 	int error = 0;
2969 
2970 	*exi = nfs_vptoexi(mc_dvp, vp, cr, &walk, NULL, FALSE);
2971 	if (*exi == NULL)
2972 		error = EACCES;
2973 	else {
2974 		/*
2975 		 * If nosub is set for this export then
2976 		 * a lookup relative to the public fh
2977 		 * must not terminate below the
2978 		 * exported directory.
2979 		 */
2980 		if ((*exi)->exi_export.ex_flags & EX_NOSUB && walk > 0)
2981 			error = EACCES;
2982 	}
2983 
2984 	return (error);
2985 }
2986 
2987 /*
2988  * Do the main work of handling HA-NFSv4 Resource Group failover on
2989  * Sun Cluster.
2990  * We need to detect whether any RG admin paths have been added or removed,
2991  * and adjust resources accordingly.
2992  * Currently we're using a very inefficient algorithm, ~ 2 * O(n**2). In
2993  * order to scale, the list and array of paths need to be held in more
2994  * suitable data structures.
2995  */
2996 static void
2997 hanfsv4_failover(void)
2998 {
2999 	int i, start_grace, numadded_paths = 0;
3000 	char **added_paths = NULL;
3001 	rfs4_dss_path_t *dss_path;
3002 
3003 	/*
3004 	 * Note: currently, rfs4_dss_pathlist cannot be NULL, since
3005 	 * it will always include an entry for NFS4_DSS_VAR_DIR. If we
3006 	 * make the latter dynamically specified too, the following will
3007 	 * need to be adjusted.
3008 	 */
3009 
3010 	/*
3011 	 * First, look for removed paths: RGs that have been failed-over
3012 	 * away from this node.
3013 	 * Walk the "currently-serving" rfs4_dss_pathlist and, for each
3014 	 * path, check if it is on the "passed-in" rfs4_dss_newpaths array
3015 	 * from nfsd. If not, that RG path has been removed.
3016 	 *
3017 	 * Note that nfsd has sorted rfs4_dss_newpaths for us, and removed
3018 	 * any duplicates.
3019 	 */
3020 	dss_path = rfs4_dss_pathlist;
3021 	do {
3022 		int found = 0;
3023 		char *path = dss_path->path;
3024 
3025 		/* used only for non-HA so may not be removed */
3026 		if (strcmp(path, NFS4_DSS_VAR_DIR) == 0) {
3027 			dss_path = dss_path->next;
3028 			continue;
3029 		}
3030 
3031 		for (i = 0; i < rfs4_dss_numnewpaths; i++) {
3032 			int cmpret;
3033 			char *newpath = rfs4_dss_newpaths[i];
3034 
3035 			/*
3036 			 * Since nfsd has sorted rfs4_dss_newpaths for us,
3037 			 * once the return from strcmp is negative we know
3038 			 * we've passed the point where "path" should be,
3039 			 * and can stop searching: "path" has been removed.
3040 			 */
3041 			cmpret = strcmp(path, newpath);
3042 			if (cmpret < 0)
3043 				break;
3044 			if (cmpret == 0) {
3045 				found = 1;
3046 				break;
3047 			}
3048 		}
3049 
3050 		if (found == 0) {
3051 			unsigned index = dss_path->index;
3052 			rfs4_servinst_t *sip = dss_path->sip;
3053 			rfs4_dss_path_t *path_next = dss_path->next;
3054 
3055 			/*
3056 			 * This path has been removed.
3057 			 * We must clear out the servinst reference to
3058 			 * it, since it's now owned by another
3059 			 * node: we should not attempt to touch it.
3060 			 */
3061 			ASSERT(dss_path == sip->dss_paths[index]);
3062 			sip->dss_paths[index] = NULL;
3063 
3064 			/* remove from "currently-serving" list, and destroy */
3065 			remque(dss_path);
3066 			/* allow for NUL */
3067 			kmem_free(dss_path->path, strlen(dss_path->path) + 1);
3068 			kmem_free(dss_path, sizeof (rfs4_dss_path_t));
3069 
3070 			dss_path = path_next;
3071 		} else {
3072 			/* path was found; not removed */
3073 			dss_path = dss_path->next;
3074 		}
3075 	} while (dss_path != rfs4_dss_pathlist);
3076 
3077 	/*
3078 	 * Now, look for added paths: RGs that have been failed-over
3079 	 * to this node.
3080 	 * Walk the "passed-in" rfs4_dss_newpaths array from nfsd and,
3081 	 * for each path, check if it is on the "currently-serving"
3082 	 * rfs4_dss_pathlist. If not, that RG path has been added.
3083 	 *
3084 	 * Note: we don't do duplicate detection here; nfsd does that for us.
3085 	 *
3086 	 * Note: numadded_paths <= rfs4_dss_numnewpaths, which gives us
3087 	 * an upper bound for the size needed for added_paths[numadded_paths].
3088 	 */
3089 
3090 	/* probably more space than we need, but guaranteed to be enough */
3091 	if (rfs4_dss_numnewpaths > 0) {
3092 		size_t sz = rfs4_dss_numnewpaths * sizeof (char *);
3093 		added_paths = kmem_zalloc(sz, KM_SLEEP);
3094 	}
3095 
3096 	/* walk the "passed-in" rfs4_dss_newpaths array from nfsd */
3097 	for (i = 0; i < rfs4_dss_numnewpaths; i++) {
3098 		int found = 0;
3099 		char *newpath = rfs4_dss_newpaths[i];
3100 
3101 		dss_path = rfs4_dss_pathlist;
3102 		do {
3103 			char *path = dss_path->path;
3104 
3105 			/* used only for non-HA */
3106 			if (strcmp(path, NFS4_DSS_VAR_DIR) == 0) {
3107 				dss_path = dss_path->next;
3108 				continue;
3109 			}
3110 
3111 			if (strncmp(path, newpath, strlen(path)) == 0) {
3112 				found = 1;
3113 				break;
3114 			}
3115 
3116 			dss_path = dss_path->next;
3117 		} while (dss_path != rfs4_dss_pathlist);
3118 
3119 		if (found == 0) {
3120 			added_paths[numadded_paths] = newpath;
3121 			numadded_paths++;
3122 		}
3123 	}
3124 
3125 	/* did we find any added paths? */
3126 	if (numadded_paths > 0) {
3127 		/* create a new server instance, and start its grace period */
3128 		start_grace = 1;
3129 		rfs4_servinst_create(start_grace, numadded_paths, added_paths);
3130 
3131 		/* read in the stable storage state from these paths */
3132 		rfs4_dss_readstate(numadded_paths, added_paths);
3133 
3134 		/*
3135 		 * Multiple failovers during a grace period will cause
3136 		 * clients of the same resource group to be partitioned
3137 		 * into different server instances, with different
3138 		 * grace periods.  Since clients of the same resource
3139 		 * group must be subject to the same grace period,
3140 		 * we need to reset all currently active grace periods.
3141 		 */
3142 		rfs4_grace_reset_all();
3143 	}
3144 
3145 	if (rfs4_dss_numnewpaths > 0)
3146 		kmem_free(added_paths, rfs4_dss_numnewpaths * sizeof (char *));
3147 }
3148 
3149 /*
3150  * Used by NFSv3 and NFSv4 server to query label of
3151  * a pathname component during lookup/access ops.
3152  */
3153 ts_label_t *
3154 nfs_getflabel(vnode_t *vp, struct exportinfo *exi)
3155 {
3156 	zone_t *zone;
3157 	ts_label_t *zone_label;
3158 	char *path;
3159 
3160 	mutex_enter(&vp->v_lock);
3161 	if (vp->v_path != NULL) {
3162 		zone = zone_find_by_any_path(vp->v_path, B_FALSE);
3163 		mutex_exit(&vp->v_lock);
3164 	} else {
3165 		/*
3166 		 * v_path not cached. Fall back on pathname of exported
3167 		 * file system as we rely on pathname from which we can
3168 		 * derive a label. The exported file system portion of
3169 		 * path is sufficient to obtain a label.
3170 		 */
3171 		path = exi->exi_export.ex_path;
3172 		if (path == NULL) {
3173 			mutex_exit(&vp->v_lock);
3174 			return (NULL);
3175 		}
3176 		zone = zone_find_by_any_path(path, B_FALSE);
3177 		mutex_exit(&vp->v_lock);
3178 	}
3179 	/*
3180 	 * Caller has verified that the file is either
3181 	 * exported or visible. So if the path falls in
3182 	 * global zone, admin_low is returned; otherwise
3183 	 * the zone's label is returned.
3184 	 */
3185 	zone_label = zone->zone_slabel;
3186 	label_hold(zone_label);
3187 	zone_rele(zone);
3188 	return (zone_label);
3189 }
3190 
3191 /*
3192  * TX NFS routine used by NFSv3 and NFSv4 to do label check
3193  * on client label and server's file object lable.
3194  */
3195 boolean_t
3196 do_rfs_label_check(bslabel_t *clabel, vnode_t *vp, int flag,
3197     struct exportinfo *exi)
3198 {
3199 	bslabel_t *slabel;
3200 	ts_label_t *tslabel;
3201 	boolean_t result;
3202 
3203 	if ((tslabel = nfs_getflabel(vp, exi)) == NULL) {
3204 		return (B_FALSE);
3205 	}
3206 	slabel = label2bslabel(tslabel);
3207 	DTRACE_PROBE4(tx__rfs__log__info__labelcheck, char *,
3208 	    "comparing server's file label(1) with client label(2) (vp(3))",
3209 	    bslabel_t *, slabel, bslabel_t *, clabel, vnode_t *, vp);
3210 
3211 	if (flag == EQUALITY_CHECK)
3212 		result = blequal(clabel, slabel);
3213 	else
3214 		result = bldominates(clabel, slabel);
3215 	label_rele(tslabel);
3216 	return (result);
3217 }
3218 
3219 /*
3220  * Callback function to return the loaned buffers.
3221  * Calls VOP_RETZCBUF() only after all uio_iov[]
3222  * buffers are returned. nu_ref maintains the count.
3223  */
3224 void
3225 rfs_free_xuio(void *free_arg)
3226 {
3227 	uint_t ref;
3228 	nfs_xuio_t *nfsuiop = (nfs_xuio_t *)free_arg;
3229 
3230 	ref = atomic_dec_uint_nv(&nfsuiop->nu_ref);
3231 
3232 	/*
3233 	 * Call VOP_RETZCBUF() only when all the iov buffers
3234 	 * are sent OTW.
3235 	 */
3236 	if (ref != 0)
3237 		return;
3238 
3239 	if (((uio_t *)nfsuiop)->uio_extflg & UIO_XUIO) {
3240 		(void) VOP_RETZCBUF(nfsuiop->nu_vp, (xuio_t *)free_arg, NULL,
3241 		    NULL);
3242 		VN_RELE(nfsuiop->nu_vp);
3243 	}
3244 
3245 	kmem_cache_free(nfs_xuio_cache, free_arg);
3246 }
3247 
3248 xuio_t *
3249 rfs_setup_xuio(vnode_t *vp)
3250 {
3251 	nfs_xuio_t *nfsuiop;
3252 
3253 	nfsuiop = kmem_cache_alloc(nfs_xuio_cache, KM_SLEEP);
3254 
3255 	bzero(nfsuiop, sizeof (nfs_xuio_t));
3256 	nfsuiop->nu_vp = vp;
3257 
3258 	/*
3259 	 * ref count set to 1. more may be added
3260 	 * if multiple mblks refer to multiple iov's.
3261 	 * This is done in uio_to_mblk().
3262 	 */
3263 
3264 	nfsuiop->nu_ref = 1;
3265 
3266 	nfsuiop->nu_frtn.free_func = rfs_free_xuio;
3267 	nfsuiop->nu_frtn.free_arg = (char *)nfsuiop;
3268 
3269 	nfsuiop->nu_uio.xu_type = UIOTYPE_ZEROCOPY;
3270 
3271 	return (&nfsuiop->nu_uio);
3272 }
3273 
3274 mblk_t *
3275 uio_to_mblk(uio_t *uiop)
3276 {
3277 	struct iovec *iovp;
3278 	int i;
3279 	mblk_t *mp, *mp1;
3280 	nfs_xuio_t *nfsuiop = (nfs_xuio_t *)uiop;
3281 
3282 	if (uiop->uio_iovcnt == 0)
3283 		return (NULL);
3284 
3285 	iovp = uiop->uio_iov;
3286 	mp = mp1 = esballoca((uchar_t *)iovp->iov_base, iovp->iov_len,
3287 	    BPRI_MED, &nfsuiop->nu_frtn);
3288 	ASSERT(mp != NULL);
3289 
3290 	mp->b_wptr += iovp->iov_len;
3291 	mp->b_datap->db_type = M_DATA;
3292 
3293 	for (i = 1; i < uiop->uio_iovcnt; i++) {
3294 		iovp = (uiop->uio_iov + i);
3295 
3296 		mp1->b_cont = esballoca(
3297 		    (uchar_t *)iovp->iov_base, iovp->iov_len, BPRI_MED,
3298 		    &nfsuiop->nu_frtn);
3299 
3300 		mp1 = mp1->b_cont;
3301 		ASSERT(mp1 != NULL);
3302 		mp1->b_wptr += iovp->iov_len;
3303 		mp1->b_datap->db_type = M_DATA;
3304 	}
3305 
3306 	nfsuiop->nu_ref = uiop->uio_iovcnt;
3307 
3308 	return (mp);
3309 }
3310 
3311 void
3312 rfs_rndup_mblks(mblk_t *mp, uint_t len, int buf_loaned)
3313 {
3314 	int i, rndup;
3315 	int alloc_err = 0;
3316 	mblk_t *rmp;
3317 
3318 	rndup = BYTES_PER_XDR_UNIT - (len % BYTES_PER_XDR_UNIT);
3319 
3320 	/* single mblk_t non copy-reduction case */
3321 	if (!buf_loaned) {
3322 		mp->b_wptr += len;
3323 		if (rndup != BYTES_PER_XDR_UNIT) {
3324 			for (i = 0; i < rndup; i++)
3325 				*mp->b_wptr++ = '\0';
3326 		}
3327 		return;
3328 	}
3329 
3330 	/* no need for extra rndup */
3331 	if (rndup == BYTES_PER_XDR_UNIT)
3332 		return;
3333 
3334 	while (mp->b_cont)
3335 		mp = mp->b_cont;
3336 
3337 	/*
3338 	 * In case of copy-reduction mblks, the size of the mblks
3339 	 * are fixed and are of the size of the loaned buffers.
3340 	 * Allocate a roundup mblk and chain it to the data
3341 	 * buffers. This is sub-optimal, but not expected to
3342 	 * happen in regular common workloads.
3343 	 */
3344 
3345 	rmp = allocb_wait(rndup, BPRI_MED, STR_NOSIG, &alloc_err);
3346 	ASSERT(rmp != NULL);
3347 	ASSERT(alloc_err == 0);
3348 
3349 	for (i = 0; i < rndup; i++)
3350 		*rmp->b_wptr++ = '\0';
3351 
3352 	rmp->b_datap->db_type = M_DATA;
3353 	mp->b_cont = rmp;
3354 }
3355