xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs4_vfsops.c (revision 5e3986cb9bc07bde1dfa1a7028d5ad3e501abae4)
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 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
28  *	All Rights Reserved
29  */
30 
31 #pragma ident	"%Z%%M%	%I%	%E% SMI"
32 
33 #include <sys/param.h>
34 #include <sys/types.h>
35 #include <sys/systm.h>
36 #include <sys/cred.h>
37 #include <sys/vfs.h>
38 #include <sys/vnode.h>
39 #include <sys/pathname.h>
40 #include <sys/sysmacros.h>
41 #include <sys/kmem.h>
42 #include <sys/mkdev.h>
43 #include <sys/mount.h>
44 #include <sys/statvfs.h>
45 #include <sys/errno.h>
46 #include <sys/debug.h>
47 #include <sys/cmn_err.h>
48 #include <sys/utsname.h>
49 #include <sys/bootconf.h>
50 #include <sys/modctl.h>
51 #include <sys/acl.h>
52 #include <sys/flock.h>
53 #include <sys/time.h>
54 #include <sys/disp.h>
55 #include <sys/policy.h>
56 #include <sys/socket.h>
57 #include <sys/netconfig.h>
58 #include <sys/dnlc.h>
59 #include <sys/list.h>
60 #include <sys/mntent.h>
61 #include <sys/tsol/label.h>
62 
63 #include <rpc/types.h>
64 #include <rpc/auth.h>
65 #include <rpc/rpcsec_gss.h>
66 #include <rpc/clnt.h>
67 
68 #include <nfs/nfs.h>
69 #include <nfs/nfs_clnt.h>
70 #include <nfs/mount.h>
71 #include <nfs/nfs_acl.h>
72 
73 #include <fs/fs_subr.h>
74 
75 #include <nfs/nfs4.h>
76 #include <nfs/rnode4.h>
77 #include <nfs/nfs4_clnt.h>
78 
79 /*
80  * Arguments passed to thread to free data structures from forced unmount.
81  */
82 
83 typedef struct {
84 	vfs_t *fm_vfsp;
85 	cred_t *fm_cr;
86 } freemountargs_t;
87 
88 static void	async_free_mount(vfs_t *, cred_t *);
89 static void	nfs4_free_mount(vfs_t *, cred_t *);
90 static void	nfs4_free_mount_thread(freemountargs_t *);
91 static int nfs4_chkdup_servinfo4(servinfo4_t *, servinfo4_t *);
92 
93 /*
94  * From rpcsec module (common/rpcsec).
95  */
96 extern int sec_clnt_loadinfo(struct sec_data *, struct sec_data **, model_t);
97 extern void sec_clnt_freeinfo(struct sec_data *);
98 
99 /*
100  * The order and contents of this structure must be kept in sync with that of
101  * rfsreqcnt_v4_tmpl in nfs_stats.c
102  */
103 static char *rfsnames_v4[] = {
104 	"null", "compound", "reserved",	"access", "close", "commit", "create",
105 	"delegpurge", "delegreturn", "getattr",	"getfh", "link", "lock",
106 	"lockt", "locku", "lookup", "lookupp", "nverify", "open", "openattr",
107 	"open_confirm",	"open_downgrade", "putfh", "putpubfh", "putrootfh",
108 	"read", "readdir", "readlink", "remove", "rename", "renew",
109 	"restorefh", "savefh", "secinfo", "setattr", "setclientid",
110 	"setclientid_confirm", "verify", "write"
111 };
112 
113 /*
114  * nfs4_max_mount_retry is the number of times the client will redrive
115  * a mount compound before giving up and returning failure.  The intent
116  * is to redrive mount compounds which fail NFS4ERR_STALE so that
117  * if a component of the server path being mounted goes stale, it can
118  * "recover" by redriving the mount compund (LOOKUP ops).  This recovery
119  * code is needed outside of the recovery framework because mount is a
120  * special case.  The client doesn't create vnodes/rnodes for components
121  * of the server path being mounted.  The recovery code recovers real
122  * client objects, not STALE FHs which map to components of the server
123  * path being mounted.
124  *
125  * We could just fail the mount on the first time, but that would
126  * instantly trigger failover (from nfs4_mount), and the client should
127  * try to re-lookup the STALE FH before doing failover.  The easiest
128  * way to "re-lookup" is to simply redrive the mount compound.
129  */
130 static int nfs4_max_mount_retry = 2;
131 
132 /*
133  * nfs4 vfs operations.
134  */
135 static int	nfs4_mount(vfs_t *, vnode_t *, struct mounta *, cred_t *);
136 static int	nfs4_unmount(vfs_t *, int, cred_t *);
137 static int	nfs4_root(vfs_t *, vnode_t **);
138 static int	nfs4_statvfs(vfs_t *, struct statvfs64 *);
139 static int	nfs4_sync(vfs_t *, short, cred_t *);
140 static int	nfs4_vget(vfs_t *, vnode_t **, fid_t *);
141 static int	nfs4_mountroot(vfs_t *, whymountroot_t);
142 static void	nfs4_freevfs(vfs_t *);
143 
144 static int	nfs4rootvp(vnode_t **, vfs_t *, struct servinfo4 *,
145 		    int, cred_t *, zone_t *);
146 
147 vfsops_t	*nfs4_vfsops;
148 
149 int nfs4_vfsinit(void);
150 void nfs4_vfsfini(void);
151 static void nfs4setclientid_init(void);
152 static void nfs4setclientid_fini(void);
153 static void nfs4setclientid_otw(mntinfo4_t *, servinfo4_t *,  cred_t *,
154 		struct nfs4_server *, nfs4_error_t *, int *);
155 static void	destroy_nfs4_server(nfs4_server_t *);
156 static void	remove_mi(nfs4_server_t *, mntinfo4_t *);
157 
158 /*
159  * Initialize the vfs structure
160  */
161 
162 static int nfs4fstyp;
163 
164 
165 /*
166  * Debug variable to check for rdma based
167  * transport startup and cleanup. Controlled
168  * through /etc/system. Off by default.
169  */
170 extern int rdma_debug;
171 
172 int
173 nfs4init(int fstyp, char *name)
174 {
175 	static const fs_operation_def_t nfs4_vfsops_template[] = {
176 		VFSNAME_MOUNT, nfs4_mount,
177 		VFSNAME_UNMOUNT, nfs4_unmount,
178 		VFSNAME_ROOT, nfs4_root,
179 		VFSNAME_STATVFS, nfs4_statvfs,
180 		VFSNAME_SYNC, (fs_generic_func_p) nfs4_sync,
181 		VFSNAME_VGET, nfs4_vget,
182 		VFSNAME_MOUNTROOT, nfs4_mountroot,
183 		VFSNAME_FREEVFS, (fs_generic_func_p)nfs4_freevfs,
184 		NULL, NULL
185 	};
186 	int error;
187 
188 	error = vfs_setfsops(fstyp, nfs4_vfsops_template, &nfs4_vfsops);
189 	if (error != 0) {
190 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
191 		    "nfs4init: bad vfs ops template");
192 		return (error);
193 	}
194 
195 	error = vn_make_ops(name, nfs4_vnodeops_template, &nfs4_vnodeops);
196 	if (error != 0) {
197 		(void) vfs_freevfsops_by_type(fstyp);
198 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
199 		    "nfs4init: bad vnode ops template");
200 		return (error);
201 	}
202 
203 	nfs4fstyp = fstyp;
204 
205 	(void) nfs4_vfsinit();
206 
207 	(void) nfs4_init_dot_entries();
208 
209 	return (0);
210 }
211 
212 void
213 nfs4fini(void)
214 {
215 	(void) nfs4_destroy_dot_entries();
216 	nfs4_vfsfini();
217 }
218 
219 /*
220  * Create a new sec_data structure to store AUTH_DH related data:
221  * netname, syncaddr, knetconfig. There is no AUTH_F_RPCTIMESYNC
222  * flag set for NFS V4 since we are avoiding to contact the rpcbind
223  * daemon and is using the IP time service (IPPORT_TIMESERVER).
224  *
225  * sec_data can be freed by sec_clnt_freeinfo().
226  */
227 struct sec_data *
228 create_authdh_data(char *netname, int nlen, struct netbuf *syncaddr,
229 		struct knetconfig *knconf) {
230 	struct sec_data *secdata;
231 	dh_k4_clntdata_t *data;
232 	char *pf, *p;
233 
234 	if (syncaddr == NULL || syncaddr->buf == NULL || nlen == 0)
235 		return (NULL);
236 
237 	secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
238 	secdata->flags = 0;
239 
240 	data = kmem_alloc(sizeof (*data), KM_SLEEP);
241 
242 	data->syncaddr.maxlen = syncaddr->maxlen;
243 	data->syncaddr.len = syncaddr->len;
244 	data->syncaddr.buf = (char *)kmem_alloc(syncaddr->len, KM_SLEEP);
245 	bcopy(syncaddr->buf, data->syncaddr.buf, syncaddr->len);
246 
247 	/*
248 	 * duplicate the knconf information for the
249 	 * new opaque data.
250 	 */
251 	data->knconf = kmem_alloc(sizeof (*knconf), KM_SLEEP);
252 	*data->knconf = *knconf;
253 	pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
254 	p = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
255 	bcopy(knconf->knc_protofmly, pf, KNC_STRSIZE);
256 	bcopy(knconf->knc_proto, p, KNC_STRSIZE);
257 	data->knconf->knc_protofmly = pf;
258 	data->knconf->knc_proto = p;
259 
260 	/* move server netname to the sec_data structure */
261 	data->netname = kmem_alloc(nlen, KM_SLEEP);
262 	bcopy(netname, data->netname, nlen);
263 	data->netnamelen = (int)nlen;
264 
265 	secdata->secmod = AUTH_DH;
266 	secdata->rpcflavor = AUTH_DH;
267 	secdata->data = (caddr_t)data;
268 
269 	return (secdata);
270 }
271 
272 static int
273 nfs4_chkdup_servinfo4(servinfo4_t *svp_head, servinfo4_t *svp)
274 {
275 	servinfo4_t *si;
276 
277 	/*
278 	 * Iterate over the servinfo4 list to make sure
279 	 * we do not have a duplicate. Skip any servinfo4
280 	 * that has been marked "NOT IN USE"
281 	 */
282 	for (si = svp_head; si; si = si->sv_next) {
283 		(void) nfs_rw_enter_sig(&si->sv_lock, RW_READER, 0);
284 		if (si->sv_flags & SV4_NOTINUSE) {
285 			nfs_rw_exit(&si->sv_lock);
286 			continue;
287 		}
288 		nfs_rw_exit(&si->sv_lock);
289 		if (si == svp)
290 			continue;
291 		if (si->sv_addr.len == svp->sv_addr.len &&
292 		    strcmp(si->sv_knconf->knc_protofmly,
293 			svp->sv_knconf->knc_protofmly) == 0 &&
294 		    bcmp(si->sv_addr.buf, svp->sv_addr.buf,
295 			si->sv_addr.len) == 0) {
296 			/* it's a duplicate */
297 			return (1);
298 		}
299 	}
300 	/* it's not a duplicate */
301 	return (0);
302 }
303 
304 /*
305  * nfs mount vfsop
306  * Set up mount info record and attach it to vfs struct.
307  */
308 static int
309 nfs4_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
310 {
311 	char *data = uap->dataptr;
312 	int error;
313 	vnode_t *rtvp;			/* the server's root */
314 	mntinfo4_t *mi;			/* mount info, pointed at by vfs */
315 	size_t hlen;			/* length of hostname */
316 	size_t nlen;			/* length of netname */
317 	char netname[MAXNETNAMELEN+1];	/* server's netname */
318 	struct netbuf addr;		/* server's address */
319 	struct netbuf syncaddr;		/* AUTH_DES time sync addr */
320 	struct knetconfig *knconf;	/* transport knetconfig structure */
321 	struct knetconfig *rdma_knconf;	/* rdma transport structure */
322 	rnode4_t *rp;
323 	struct servinfo4 *svp;		/* nfs server info */
324 	struct servinfo4 *svp_tail = NULL; /* previous nfs server info */
325 	struct servinfo4 *svp_head;	/* first nfs server info */
326 	struct servinfo4 *svp_2ndlast;	/* 2nd last in server info list */
327 	struct sec_data *secdata;	/* security data */
328 	STRUCT_DECL(nfs_args, args);	/* nfs mount arguments */
329 	STRUCT_DECL(knetconfig, knconf_tmp);
330 	STRUCT_DECL(netbuf, addr_tmp);
331 	int flags, addr_type, removed;
332 	char *p, *pf;
333 	struct pathname pn;
334 	char *userbufptr;
335 	zone_t *zone = nfs_zone();
336 	nfs4_error_t n4e;
337 	zone_t *mntzone = NULL;
338 
339 	if (secpolicy_fs_mount(cr, mvp, vfsp) != 0)
340 		return (EPERM);
341 	if (mvp->v_type != VDIR)
342 		return (ENOTDIR);
343 	/*
344 	 * get arguments
345 	 *
346 	 * nfs_args is now versioned and is extensible, so
347 	 * uap->datalen might be different from sizeof (args)
348 	 * in a compatible situation.
349 	 */
350 more:
351 	STRUCT_INIT(args, get_udatamodel());
352 	bzero(STRUCT_BUF(args), SIZEOF_STRUCT(nfs_args, DATAMODEL_NATIVE));
353 	if (copyin(data, STRUCT_BUF(args), MIN(uap->datalen,
354 	    STRUCT_SIZE(args))))
355 		return (EFAULT);
356 
357 	flags = STRUCT_FGET(args, flags);
358 
359 	/*
360 	 * If the request changes the locking type, disallow the remount,
361 	 * because it's questionable whether we can transfer the
362 	 * locking state correctly.
363 	 */
364 	if (uap->flags & MS_REMOUNT) {
365 		if ((mi = VFTOMI4(vfsp)) != NULL) {
366 			uint_t new_mi_llock;
367 			uint_t old_mi_llock;
368 
369 			new_mi_llock = (flags & NFSMNT_LLOCK) ? 1 : 0;
370 			old_mi_llock = (mi->mi_flags & MI4_LLOCK) ? 1 : 0;
371 			if (old_mi_llock != new_mi_llock)
372 				return (EBUSY);
373 		}
374 		return (0);
375 	}
376 
377 	mutex_enter(&mvp->v_lock);
378 	if (!(uap->flags & MS_OVERLAY) &&
379 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
380 		mutex_exit(&mvp->v_lock);
381 		return (EBUSY);
382 	}
383 	mutex_exit(&mvp->v_lock);
384 
385 	/* make sure things are zeroed for errout: */
386 	rtvp = NULL;
387 	mi = NULL;
388 	addr.buf = NULL;
389 	syncaddr.buf = NULL;
390 	secdata = NULL;
391 
392 	/*
393 	 * A valid knetconfig structure is required.
394 	 */
395 	if (!(flags & NFSMNT_KNCONF))
396 		return (EINVAL);
397 
398 	/*
399 	 * Allocate a servinfo4 struct.
400 	 */
401 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
402 	nfs_rw_init(&svp->sv_lock, NULL, RW_DEFAULT, NULL);
403 	if (svp_tail) {
404 		svp_2ndlast = svp_tail;
405 		svp_tail->sv_next = svp;
406 	} else {
407 		svp_head = svp;
408 		svp_2ndlast = svp;
409 	}
410 
411 	svp_tail = svp;
412 
413 	/*
414 	 * Allocate space for a knetconfig structure and
415 	 * its strings and copy in from user-land.
416 	 */
417 	knconf = kmem_zalloc(sizeof (*knconf), KM_SLEEP);
418 	svp->sv_knconf = knconf;
419 	STRUCT_INIT(knconf_tmp, get_udatamodel());
420 	if (copyin(STRUCT_FGETP(args, knconf), STRUCT_BUF(knconf_tmp),
421 	    STRUCT_SIZE(knconf_tmp))) {
422 		sv4_free(svp_head);
423 		return (EFAULT);
424 	}
425 
426 	knconf->knc_semantics = STRUCT_FGET(knconf_tmp, knc_semantics);
427 	knconf->knc_protofmly = STRUCT_FGETP(knconf_tmp, knc_protofmly);
428 	knconf->knc_proto = STRUCT_FGETP(knconf_tmp, knc_proto);
429 	if (get_udatamodel() != DATAMODEL_LP64) {
430 		knconf->knc_rdev = expldev(STRUCT_FGET(knconf_tmp, knc_rdev));
431 	} else {
432 		knconf->knc_rdev = STRUCT_FGET(knconf_tmp, knc_rdev);
433 	}
434 
435 	pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
436 	p = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
437 	error = copyinstr(knconf->knc_protofmly, pf, KNC_STRSIZE, NULL);
438 	if (error) {
439 		kmem_free(pf, KNC_STRSIZE);
440 		kmem_free(p, KNC_STRSIZE);
441 		sv4_free(svp_head);
442 		return (error);
443 	}
444 	error = copyinstr(knconf->knc_proto, p, KNC_STRSIZE, NULL);
445 	if (error) {
446 		kmem_free(pf, KNC_STRSIZE);
447 		kmem_free(p, KNC_STRSIZE);
448 		sv4_free(svp_head);
449 		return (error);
450 	}
451 	if (strcmp(p, NC_UDP) == 0) {
452 		kmem_free(pf, KNC_STRSIZE);
453 		kmem_free(p, KNC_STRSIZE);
454 		sv4_free(svp_head);
455 		return (ENOTSUP);
456 	}
457 	knconf->knc_protofmly = pf;
458 	knconf->knc_proto = p;
459 
460 	/*
461 	 * Get server address
462 	 */
463 	STRUCT_INIT(addr_tmp, get_udatamodel());
464 	if (copyin(STRUCT_FGETP(args, addr), STRUCT_BUF(addr_tmp),
465 	    STRUCT_SIZE(addr_tmp))) {
466 		error = EFAULT;
467 		goto errout;
468 	}
469 
470 	userbufptr = addr.buf = STRUCT_FGETP(addr_tmp, buf);
471 	addr.len = STRUCT_FGET(addr_tmp, len);
472 	addr.buf = kmem_alloc(addr.len, KM_SLEEP);
473 	addr.maxlen = addr.len;
474 	if (copyin(userbufptr, addr.buf, addr.len)) {
475 		kmem_free(addr.buf, addr.len);
476 		error = EFAULT;
477 		goto errout;
478 	}
479 
480 	svp->sv_addr = addr;
481 
482 	/*
483 	 * Get the root fhandle
484 	 */
485 	error = pn_get(STRUCT_FGETP(args, fh), UIO_USERSPACE, &pn);
486 
487 	if (error)
488 		goto errout;
489 
490 	/* Volatile fh: keep server paths, so use actual-size strings */
491 	svp->sv_path = kmem_alloc(pn.pn_pathlen + 1, KM_SLEEP);
492 	bcopy(pn.pn_path, svp->sv_path, pn.pn_pathlen);
493 	svp->sv_path[pn.pn_pathlen] = '\0';
494 	svp->sv_pathlen = pn.pn_pathlen + 1;
495 	pn_free(&pn);
496 
497 	/*
498 	 * Get server's hostname
499 	 */
500 	if (flags & NFSMNT_HOSTNAME) {
501 		error = copyinstr(STRUCT_FGETP(args, hostname),
502 				netname, sizeof (netname), &hlen);
503 		if (error)
504 			goto errout;
505 	} else {
506 		char *p = "unknown-host";
507 		hlen = strlen(p) + 1;
508 		(void) strcpy(netname, p);
509 	}
510 	svp->sv_hostnamelen = hlen;
511 	svp->sv_hostname = kmem_alloc(svp->sv_hostnamelen, KM_SLEEP);
512 	(void) strcpy(svp->sv_hostname, netname);
513 
514 	/*
515 	 * RDMA MOUNT SUPPORT FOR NFS v4.
516 	 * Establish, is it possible to use RDMA, if so overload the
517 	 * knconf with rdma specific knconf and free the orignal knconf.
518 	 */
519 	if ((flags & NFSMNT_TRYRDMA) || (flags & NFSMNT_DORDMA)) {
520 		/*
521 		 * Determine the addr type for RDMA, IPv4 or v6.
522 		 */
523 		if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET) == 0)
524 			addr_type = AF_INET;
525 		else if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET6) == 0)
526 			addr_type = AF_INET6;
527 
528 		if (rdma_reachable(addr_type, &svp->sv_addr,
529 			&rdma_knconf) == 0) {
530 			/*
531 			 * If successful, hijack the orignal knconf and
532 			 * replace with the new one, depending on the flags.
533 			 */
534 			svp->sv_origknconf = svp->sv_knconf;
535 			svp->sv_knconf = rdma_knconf;
536 			knconf = rdma_knconf;
537 		} else {
538 			if (flags & NFSMNT_TRYRDMA) {
539 #ifdef	DEBUG
540 				if (rdma_debug)
541 					zcmn_err(getzoneid(), CE_WARN,
542 					    "no RDMA onboard, revert\n");
543 #endif
544 			}
545 
546 			if (flags & NFSMNT_DORDMA) {
547 				/*
548 				 * If proto=rdma is specified and no RDMA
549 				 * path to this server is avialable then
550 				 * ditch this server.
551 				 * This is not included in the mountable
552 				 * server list or the replica list.
553 				 * Check if more servers are specified;
554 				 * Failover case, otherwise bail out of mount.
555 				 */
556 				if (STRUCT_FGET(args, nfs_args_ext) ==
557 				    NFS_ARGS_EXTB && STRUCT_FGETP(args,
558 					nfs_ext_u.nfs_extB.next) != NULL) {
559 					if (uap->flags & MS_RDONLY &&
560 					    !(flags & NFSMNT_SOFT)) {
561 						data = (char *)
562 						    STRUCT_FGETP(args,
563 						nfs_ext_u.nfs_extB.next);
564 						if (svp_head->sv_next == NULL) {
565 							svp_tail = NULL;
566 							svp_2ndlast = NULL;
567 							sv4_free(svp_head);
568 							goto more;
569 						} else {
570 							svp_tail = svp_2ndlast;
571 							svp_2ndlast->sv_next =
572 							    NULL;
573 							sv4_free(svp);
574 							goto more;
575 						}
576 					}
577 				} else {
578 					/*
579 					 * This is the last server specified
580 					 * in the nfs_args list passed down
581 					 * and its not rdma capable.
582 					 */
583 					if (svp_head->sv_next == NULL) {
584 						/*
585 						 * Is this the only one
586 						 */
587 						error = EINVAL;
588 #ifdef	DEBUG
589 						if (rdma_debug)
590 							zcmn_err(getzoneid(),
591 							    CE_WARN,
592 							    "No RDMA srv");
593 #endif
594 						goto errout;
595 					} else {
596 						/*
597 						 * There is list, since some
598 						 * servers specified before
599 						 * this passed all requirements
600 						 */
601 						svp_tail = svp_2ndlast;
602 						svp_2ndlast->sv_next = NULL;
603 						sv4_free(svp);
604 						goto proceed;
605 					}
606 				}
607 			}
608 		}
609 	}
610 
611 	/*
612 	 * If there are syncaddr and netname data, load them in. This is
613 	 * to support data needed for NFSV4 when AUTH_DH is the negotiated
614 	 * flavor via SECINFO. (instead of using MOUNT protocol in V3).
615 	 */
616 	netname[0] = '\0';
617 	if (flags & NFSMNT_SECURE) {
618 
619 		/* get syncaddr */
620 		STRUCT_INIT(addr_tmp, get_udatamodel());
621 		if (copyin(STRUCT_FGETP(args, syncaddr), STRUCT_BUF(addr_tmp),
622 			STRUCT_SIZE(addr_tmp))) {
623 			error = EINVAL;
624 			goto errout;
625 		}
626 		userbufptr = STRUCT_FGETP(addr_tmp, buf);
627 		syncaddr.len = STRUCT_FGET(addr_tmp, len);
628 		syncaddr.buf = kmem_alloc(syncaddr.len, KM_SLEEP);
629 		syncaddr.maxlen = syncaddr.len;
630 		if (copyin(userbufptr, syncaddr.buf, syncaddr.len)) {
631 			kmem_free(syncaddr.buf, syncaddr.len);
632 			error = EFAULT;
633 			goto errout;
634 		}
635 
636 		/* get server's netname */
637 		if (copyinstr(STRUCT_FGETP(args, netname), netname,
638 				sizeof (netname), &nlen)) {
639 			kmem_free(syncaddr.buf, syncaddr.len);
640 			error = EFAULT;
641 			goto errout;
642 		}
643 		netname[nlen] = '\0';
644 
645 		svp->sv_dhsec = create_authdh_data(netname, nlen, &syncaddr,
646 						knconf);
647 	}
648 
649 	/*
650 	 * Get the extention data which has the security data structure.
651 	 * This includes data for AUTH_SYS as well.
652 	 */
653 	if (flags & NFSMNT_NEWARGS) {
654 		switch (STRUCT_FGET(args, nfs_args_ext)) {
655 		case NFS_ARGS_EXTA:
656 		case NFS_ARGS_EXTB:
657 			/*
658 			 * Indicating the application is using the new
659 			 * sec_data structure to pass in the security
660 			 * data.
661 			 */
662 			if (STRUCT_FGETP(args,
663 			    nfs_ext_u.nfs_extA.secdata) == NULL) {
664 				error = EINVAL;
665 			} else {
666 				error = sec_clnt_loadinfo(
667 				    (struct sec_data *)STRUCT_FGETP(args,
668 					nfs_ext_u.nfs_extA.secdata),
669 				    &secdata, get_udatamodel());
670 			}
671 			break;
672 
673 		default:
674 			error = EINVAL;
675 			break;
676 		}
677 
678 	} else if (flags & NFSMNT_SECURE) {
679 		/*
680 		 * NFSMNT_SECURE is deprecated but we keep it
681 		 * to support the rouge user generated application
682 		 * that may use this undocumented interface to do
683 		 * AUTH_DH security.
684 		 */
685 		secdata = create_authdh_data(netname, nlen, &syncaddr, knconf);
686 
687 	} else {
688 		secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
689 		secdata->secmod = secdata->rpcflavor = AUTH_SYS;
690 		secdata->data = NULL;
691 	}
692 
693 	svp->sv_secdata = secdata;
694 
695 	/* syncaddr is no longer needed. */
696 	if (syncaddr.buf != NULL)
697 		kmem_free(syncaddr.buf, syncaddr.len);
698 
699 	/*
700 	 * User does not explictly specify a flavor, and a user
701 	 * defined default flavor is passed down.
702 	 */
703 	if (flags & NFSMNT_SECDEFAULT) {
704 		(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
705 		svp->sv_flags |= SV4_TRYSECDEFAULT;
706 		nfs_rw_exit(&svp->sv_lock);
707 	}
708 
709 	/*
710 	 * Failover support:
711 	 *
712 	 * We may have a linked list of nfs_args structures,
713 	 * which means the user is looking for failover.  If
714 	 * the mount is either not "read-only" or "soft",
715 	 * we want to bail out with EINVAL.
716 	 */
717 	if (STRUCT_FGET(args, nfs_args_ext) == NFS_ARGS_EXTB &&
718 	    STRUCT_FGETP(args, nfs_ext_u.nfs_extB.next) != NULL) {
719 		if (uap->flags & MS_RDONLY && !(flags & NFSMNT_SOFT)) {
720 			data = (char *)STRUCT_FGETP(args,
721 			    nfs_ext_u.nfs_extB.next);
722 			goto more;
723 		}
724 		error = EINVAL;
725 		goto errout;
726 	}
727 
728 	/*
729 	 * Determine the zone we're being mounted into.
730 	 */
731 	zone_hold(mntzone = zone);		/* start with this assumption */
732 	if (getzoneid() == GLOBAL_ZONEID) {
733 		zone_rele(mntzone);
734 		mntzone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
735 		ASSERT(mntzone != NULL);
736 		if (mntzone != zone) {
737 			error = EBUSY;
738 			goto errout;
739 		}
740 	}
741 
742 	if (is_system_labeled()) {
743 		error = nfs_mount_label_policy(vfsp, &svp->sv_addr,
744 		    svp->sv_knconf, cr);
745 
746 		if (error > 0)
747 			goto errout;
748 
749 		if (error == -1) {
750 			/* change mount to read-only to prevent write-down */
751 			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
752 		}
753 	}
754 
755 	/*
756 	 * Stop the mount from going any further if the zone is going away.
757 	 */
758 	if (zone_status_get(mntzone) >= ZONE_IS_SHUTTING_DOWN) {
759 		error = EBUSY;
760 		goto errout;
761 	}
762 
763 	/*
764 	 * Get root vnode.
765 	 */
766 proceed:
767 	error = nfs4rootvp(&rtvp, vfsp, svp_head, flags, cr, mntzone);
768 
769 	if (error) {
770 		/* if nfs4rootvp failed, it will free svp_head */
771 		svp_head = NULL;
772 		goto errout;
773 	}
774 
775 	mi = VTOMI4(rtvp);
776 
777 	/*
778 	 * Send client id to the server, if necessary
779 	 */
780 	nfs4_error_zinit(&n4e);
781 	nfs4setclientid(mi, cr, FALSE, &n4e);
782 	error = n4e.error;
783 
784 	if (error)
785 		goto errout;
786 
787 	/*
788 	 * Set option fields in the mount info record
789 	 */
790 
791 	if (svp_head->sv_next) {
792 		mutex_enter(&mi->mi_lock);
793 		mi->mi_flags |= MI4_LLOCK;
794 		mutex_exit(&mi->mi_lock);
795 	}
796 
797 	error = nfs4_setopts(rtvp, get_udatamodel(), STRUCT_BUF(args));
798 
799 errout:
800 	if (error) {
801 		if (rtvp != NULL) {
802 			rp = VTOR4(rtvp);
803 			if (rp->r_flags & R4HASHED)
804 				rp4_rmhash(rp);
805 		}
806 		if (mi != NULL) {
807 			nfs4_async_stop(vfsp);
808 			nfs4_async_manager_stop(vfsp);
809 			nfs4_remove_mi_from_server(mi, NULL);
810 			if (rtvp != NULL)
811 				VN_RELE(rtvp);
812 			if (mntzone != NULL)
813 				zone_rele(mntzone);
814 			/* need to remove it from the zone */
815 			removed = nfs4_mi_zonelist_remove(mi);
816 			if (removed)
817 				zone_rele(mi->mi_zone);
818 			MI4_RELE(mi);
819 			return (error);
820 		}
821 		if (svp_head)
822 			sv4_free(svp_head);
823 	}
824 
825 	if (rtvp != NULL)
826 		VN_RELE(rtvp);
827 
828 	if (mntzone != NULL)
829 		zone_rele(mntzone);
830 
831 	return (error);
832 }
833 
834 #ifdef	DEBUG
835 #define	VERS_MSG	"NFS4 server "
836 #else
837 #define	VERS_MSG	"NFS server "
838 #endif
839 
840 #define	READ_MSG	\
841 	VERS_MSG "%s returned 0 for read transfer size"
842 #define	WRITE_MSG	\
843 	VERS_MSG "%s returned 0 for write transfer size"
844 #define	SIZE_MSG	\
845 	VERS_MSG "%s returned 0 for maximum file size"
846 
847 /*
848  * Get the symbolic link text from the server for a given filehandle
849  * of that symlink.
850  *
851  *	(get symlink text) PUTFH READLINK
852  */
853 static int
854 getlinktext_otw(mntinfo4_t *mi, nfs_fh4 *fh, char **linktextp, cred_t *cr,
855 		int flags)
856 {
857 	COMPOUND4args_clnt args;
858 	COMPOUND4res_clnt res;
859 	int doqueue;
860 	nfs_argop4 argop[2];
861 	nfs_resop4 *resop;
862 	READLINK4res *lr_res;
863 	uint_t len;
864 	bool_t needrecov = FALSE;
865 	nfs4_recov_state_t recov_state;
866 	nfs4_sharedfh_t *sfh;
867 	nfs4_error_t e;
868 	int num_retry = nfs4_max_mount_retry;
869 	int recovery = !(flags & NFS4_GETFH_NEEDSOP);
870 
871 	sfh = sfh4_get(fh, mi);
872 	recov_state.rs_flags = 0;
873 	recov_state.rs_num_retry_despite_err = 0;
874 
875 recov_retry:
876 	nfs4_error_zinit(&e);
877 
878 	args.array_len = 2;
879 	args.array = argop;
880 	args.ctag = TAG_GET_SYMLINK;
881 
882 	if (! recovery) {
883 		e.error = nfs4_start_op(mi, NULL, NULL, &recov_state);
884 		if (e.error) {
885 			sfh4_rele(&sfh);
886 			return (e.error);
887 		}
888 	}
889 
890 	/* 0. putfh symlink fh */
891 	argop[0].argop = OP_CPUTFH;
892 	argop[0].nfs_argop4_u.opcputfh.sfh = sfh;
893 
894 	/* 1. readlink */
895 	argop[1].argop = OP_READLINK;
896 
897 	doqueue = 1;
898 
899 	rfs4call(mi, &args, &res, cr, &doqueue, 0, &e);
900 
901 	needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp);
902 
903 	if (needrecov && !recovery && num_retry-- > 0) {
904 
905 		NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
906 		    "getlinktext_otw: initiating recovery\n"));
907 
908 		if (nfs4_start_recovery(&e, mi, NULL, NULL, NULL, NULL,
909 		    OP_READLINK, NULL) == FALSE) {
910 			nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
911 			if (!e.error)
912 				(void) xdr_free(xdr_COMPOUND4res_clnt,
913 						(caddr_t)&res);
914 			goto recov_retry;
915 		}
916 	}
917 
918 	/*
919 	 * If non-NFS4 pcol error and/or we weren't able to recover.
920 	 */
921 	if (e.error != 0) {
922 		if (! recovery)
923 			nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
924 		sfh4_rele(&sfh);
925 		return (e.error);
926 	}
927 
928 	if (res.status) {
929 		e.error = geterrno4(res.status);
930 		(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
931 		if (! recovery)
932 			nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
933 		sfh4_rele(&sfh);
934 		return (e.error);
935 	}
936 
937 	/* res.status == NFS4_OK */
938 	ASSERT(res.status == NFS4_OK);
939 
940 	resop = &res.array[1];	/* readlink res */
941 	lr_res = &resop->nfs_resop4_u.opreadlink;
942 
943 	/* treat symlink name as data */
944 	*linktextp = utf8_to_str(&lr_res->link, &len, NULL);
945 
946 	if (! recovery)
947 		nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
948 	sfh4_rele(&sfh);
949 	(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
950 
951 	return (0);
952 }
953 
954 /*
955  * Skip over consecutive slashes and "/./" in a pathname.
956  */
957 void
958 pathname_skipslashdot(struct pathname *pnp)
959 {
960 	char *c1, *c2;
961 
962 	while (pnp->pn_pathlen > 0 && *pnp->pn_path == '/') {
963 
964 		c1 = pnp->pn_path + 1;
965 		c2 = pnp->pn_path + 2;
966 
967 		if (*c1 == '.' && (*c2 == '/' || *c2 == '\0')) {
968 			pnp->pn_path = pnp->pn_path + 2; /* skip "/." */
969 			pnp->pn_pathlen = pnp->pn_pathlen - 2;
970 		} else {
971 			pnp->pn_path++;
972 			pnp->pn_pathlen--;
973 		}
974 	}
975 }
976 
977 /*
978  * Resolve a symbolic link path. The symlink is in the nth component of
979  * svp->sv_path and has an nfs4 file handle "fh".
980  * Upon return, the sv_path will point to the new path that has the nth
981  * component resolved to its symlink text.
982  */
983 int
984 resolve_sympath(mntinfo4_t *mi, servinfo4_t *svp, int nth, nfs_fh4 *fh,
985 		cred_t *cr, int flags)
986 {
987 	char *oldpath;
988 	char *symlink, *newpath;
989 	struct pathname oldpn, newpn;
990 	char component[MAXNAMELEN];
991 	int i, addlen, error = 0;
992 	int oldpathlen;
993 
994 	/* Get the symbolic link text over the wire. */
995 	error = getlinktext_otw(mi, fh, &symlink, cr, flags);
996 
997 	if (error || symlink == NULL || strlen(symlink) == 0)
998 		return (error);
999 
1000 	/*
1001 	 * Compose the new pathname.
1002 	 * Note:
1003 	 *    - only the nth component is resolved for the pathname.
1004 	 *    - pathname.pn_pathlen does not count the ending null byte.
1005 	 */
1006 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1007 	oldpath = svp->sv_path;
1008 	oldpathlen = svp->sv_pathlen;
1009 	if (error = pn_get(oldpath, UIO_SYSSPACE, &oldpn)) {
1010 		nfs_rw_exit(&svp->sv_lock);
1011 		kmem_free(symlink, strlen(symlink) + 1);
1012 		return (error);
1013 	}
1014 	nfs_rw_exit(&svp->sv_lock);
1015 	pn_alloc(&newpn);
1016 
1017 	/*
1018 	 * Skip over previous components from the oldpath so that the
1019 	 * oldpn.pn_path will point to the symlink component. Skip
1020 	 * leading slashes and "/./" (no OP_LOOKUP on ".") so that
1021 	 * pn_getcompnent can get the component.
1022 	 */
1023 	for (i = 1; i < nth; i++) {
1024 		pathname_skipslashdot(&oldpn);
1025 		error = pn_getcomponent(&oldpn, component);
1026 		if (error)
1027 			goto out;
1028 	}
1029 
1030 	/*
1031 	 * Copy the old path upto the component right before the symlink
1032 	 * if the symlink is not an absolute path.
1033 	 */
1034 	if (symlink[0] != '/') {
1035 		addlen = oldpn.pn_path - oldpn.pn_buf;
1036 		bcopy(oldpn.pn_buf, newpn.pn_path, addlen);
1037 		newpn.pn_pathlen += addlen;
1038 		newpn.pn_path += addlen;
1039 		newpn.pn_buf[newpn.pn_pathlen] = '/';
1040 		newpn.pn_pathlen++;
1041 		newpn.pn_path++;
1042 	}
1043 
1044 	/* copy the resolved symbolic link text */
1045 	addlen = strlen(symlink);
1046 	if (newpn.pn_pathlen + addlen >= newpn.pn_bufsize) {
1047 		error = ENAMETOOLONG;
1048 		goto out;
1049 	}
1050 	bcopy(symlink, newpn.pn_path, addlen);
1051 	newpn.pn_pathlen += addlen;
1052 	newpn.pn_path += addlen;
1053 
1054 	/*
1055 	 * Check if there is any remaining path after the symlink component.
1056 	 * First, skip the symlink component.
1057 	 */
1058 	pathname_skipslashdot(&oldpn);
1059 	if (error = pn_getcomponent(&oldpn, component))
1060 		goto out;
1061 
1062 	addlen = pn_pathleft(&oldpn); /* includes counting the slash */
1063 
1064 	/*
1065 	 * Copy the remaining path to the new pathname if there is any.
1066 	 */
1067 	if (addlen > 0) {
1068 		if (newpn.pn_pathlen + addlen >= newpn.pn_bufsize) {
1069 			error = ENAMETOOLONG;
1070 			goto out;
1071 		}
1072 		bcopy(oldpn.pn_path, newpn.pn_path, addlen);
1073 		newpn.pn_pathlen += addlen;
1074 	}
1075 	newpn.pn_buf[newpn.pn_pathlen] = '\0';
1076 
1077 	/* get the newpath and store it in the servinfo4_t */
1078 	newpath = kmem_alloc(newpn.pn_pathlen + 1, KM_SLEEP);
1079 	bcopy(newpn.pn_buf, newpath, newpn.pn_pathlen);
1080 	newpath[newpn.pn_pathlen] = '\0';
1081 
1082 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
1083 	svp->sv_path = newpath;
1084 	svp->sv_pathlen = strlen(newpath) + 1;
1085 	nfs_rw_exit(&svp->sv_lock);
1086 
1087 	kmem_free(oldpath, oldpathlen);
1088 out:
1089 	kmem_free(symlink, strlen(symlink) + 1);
1090 	pn_free(&newpn);
1091 	pn_free(&oldpn);
1092 
1093 	return (error);
1094 }
1095 
1096 /*
1097  * Get the root filehandle for the given filesystem and server, and update
1098  * svp.
1099  *
1100  * If NFS4_GETFH_NEEDSOP is set, then use nfs4_start_fop and nfs4_end_fop
1101  * to coordinate with recovery.  Otherwise, the caller is assumed to be
1102  * the recovery thread or have already done a start_fop.
1103  *
1104  * Errors are returned by the nfs4_error_t parameter.
1105  */
1106 
1107 static void
1108 nfs4getfh_otw(struct mntinfo4 *mi, servinfo4_t *svp, vtype_t *vtp,
1109 		int flags, cred_t *cr, nfs4_error_t *ep)
1110 {
1111 	COMPOUND4args_clnt args;
1112 	COMPOUND4res_clnt res;
1113 	int doqueue = 1;
1114 	nfs_argop4 *argop;
1115 	nfs_resop4 *resop;
1116 	nfs4_ga_res_t *garp;
1117 	int num_argops;
1118 	lookup4_param_t lookuparg;
1119 	nfs_fh4 *tmpfhp;
1120 	nfs_fh4 *resfhp;
1121 	bool_t needrecov = FALSE;
1122 	nfs4_recov_state_t recov_state;
1123 	int llndx;
1124 	int nthcomp;
1125 	int recovery = !(flags & NFS4_GETFH_NEEDSOP);
1126 
1127 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1128 	ASSERT(svp->sv_path != NULL);
1129 	if (svp->sv_path[0] == '\0') {
1130 		nfs_rw_exit(&svp->sv_lock);
1131 		nfs4_error_init(ep, EINVAL);
1132 		return;
1133 	}
1134 	nfs_rw_exit(&svp->sv_lock);
1135 
1136 	recov_state.rs_flags = 0;
1137 	recov_state.rs_num_retry_despite_err = 0;
1138 recov_retry:
1139 	nfs4_error_zinit(ep);
1140 
1141 	if (!recovery) {
1142 		ep->error = nfs4_start_fop(mi, NULL, NULL, OH_MOUNT,
1143 				&recov_state, NULL);
1144 
1145 		/*
1146 		 * If recovery has been started and this request as
1147 		 * initiated by a mount, then we must wait for recovery
1148 		 * to finish before proceeding, otherwise, the error
1149 		 * cleanup would remove data structures needed by the
1150 		 * recovery thread.
1151 		 */
1152 		if (ep->error) {
1153 			mutex_enter(&mi->mi_lock);
1154 			if (mi->mi_flags & MI4_MOUNTING) {
1155 				mi->mi_flags |= MI4_RECOV_FAIL;
1156 				mi->mi_error = EIO;
1157 
1158 				NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
1159 				    "nfs4getfh_otw: waiting 4 recovery\n"));
1160 
1161 				while (mi->mi_flags & MI4_RECOV_ACTIV)
1162 					cv_wait(&mi->mi_failover_cv,
1163 					    &mi->mi_lock);
1164 			}
1165 			mutex_exit(&mi->mi_lock);
1166 			return;
1167 		}
1168 
1169 		/*
1170 		 * If the client does not specify a specific flavor to use
1171 		 * and has not gotten a secinfo list from the server yet,
1172 		 * retrieve the secinfo list from the server and use a
1173 		 * flavor from the list to mount.
1174 		 *
1175 		 * If fail to get the secinfo list from the server, then
1176 		 * try the default flavor.
1177 		 */
1178 		if ((svp->sv_flags & SV4_TRYSECDEFAULT) &&
1179 		    svp->sv_secinfo == NULL) {
1180 			(void) nfs4_secinfo_path(mi, cr, FALSE);
1181 		}
1182 	}
1183 
1184 	if (recovery)
1185 		args.ctag = TAG_REMAP_MOUNT;
1186 	else
1187 		args.ctag = TAG_MOUNT;
1188 
1189 	lookuparg.l4_getattrs = LKP4_ALL_ATTRIBUTES;
1190 	lookuparg.argsp = &args;
1191 	lookuparg.resp = &res;
1192 	lookuparg.header_len = 2;	/* Putrootfh, getfh */
1193 	lookuparg.trailer_len = 0;
1194 	lookuparg.ga_bits = FATTR4_FSINFO_MASK;
1195 	lookuparg.mi = mi;
1196 
1197 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1198 	ASSERT(svp->sv_path != NULL);
1199 	llndx = nfs4lookup_setup(svp->sv_path, &lookuparg, 0);
1200 	nfs_rw_exit(&svp->sv_lock);
1201 
1202 	argop = args.array;
1203 	num_argops = args.array_len;
1204 
1205 	/* choose public or root filehandle */
1206 	if (flags & NFS4_GETFH_PUBLIC)
1207 		argop[0].argop = OP_PUTPUBFH;
1208 	else
1209 		argop[0].argop = OP_PUTROOTFH;
1210 
1211 	/* get fh */
1212 	argop[1].argop = OP_GETFH;
1213 
1214 	NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE,
1215 	    "nfs4getfh_otw: %s call, mi 0x%p",
1216 	    needrecov ? "recov" : "first", (void *)mi));
1217 
1218 	rfs4call(mi, &args, &res, cr, &doqueue, RFSCALL_SOFT, ep);
1219 
1220 	needrecov = nfs4_needs_recovery(ep, FALSE, mi->mi_vfsp);
1221 
1222 	if (needrecov) {
1223 		bool_t abort;
1224 
1225 		if (recovery) {
1226 			nfs4args_lookup_free(argop, num_argops);
1227 			kmem_free(argop,
1228 					lookuparg.arglen * sizeof (nfs_argop4));
1229 			if (!ep->error)
1230 				(void) xdr_free(xdr_COMPOUND4res_clnt,
1231 								(caddr_t)&res);
1232 			return;
1233 		}
1234 
1235 		NFS4_DEBUG(nfs4_client_recov_debug,
1236 		    (CE_NOTE, "nfs4getfh_otw: initiating recovery\n"));
1237 
1238 		abort = nfs4_start_recovery(ep, mi, NULL,
1239 			    NULL, NULL, NULL, OP_GETFH, NULL);
1240 		if (!ep->error) {
1241 			ep->error = geterrno4(res.status);
1242 			(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1243 		}
1244 		nfs4args_lookup_free(argop, num_argops);
1245 		kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4));
1246 		nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state, needrecov);
1247 		/* have another go? */
1248 		if (abort == FALSE)
1249 			goto recov_retry;
1250 		return;
1251 	}
1252 
1253 	/*
1254 	 * No recovery, but check if error is set.
1255 	 */
1256 	if (ep->error)  {
1257 		nfs4args_lookup_free(argop, num_argops);
1258 		kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4));
1259 		if (!recovery)
1260 			nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state,
1261 				needrecov);
1262 		return;
1263 	}
1264 
1265 is_link_err:
1266 
1267 	/* for non-recovery errors */
1268 	if (res.status && res.status != NFS4ERR_SYMLINK) {
1269 		if (!recovery) {
1270 			nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state,
1271 				needrecov);
1272 		}
1273 		nfs4args_lookup_free(argop, num_argops);
1274 		kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4));
1275 		(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1276 		return;
1277 	}
1278 
1279 	/*
1280 	 * If any intermediate component in the path is a symbolic link,
1281 	 * resolve the symlink, then try mount again using the new path.
1282 	 */
1283 	if (res.status == NFS4ERR_SYMLINK) {
1284 		int where;
1285 
1286 		/*
1287 		 * This must be from OP_LOOKUP failure. The (cfh) for this
1288 		 * OP_LOOKUP is a symlink node. Found out where the
1289 		 * OP_GETFH is for the (cfh) that is a symlink node.
1290 		 *
1291 		 * Example:
1292 		 * (mount) PUTROOTFH, GETFH, LOOKUP comp1, GETFH, GETATTR,
1293 		 * LOOKUP comp2, GETFH, GETATTR, LOOKUP comp3, GETFH, GETATTR
1294 		 *
1295 		 * LOOKUP comp3 fails with SYMLINK because comp2 is a symlink.
1296 		 * In this case, where = 7, nthcomp = 2.
1297 		 */
1298 		where = res.array_len - 2;
1299 		ASSERT(where > 0);
1300 
1301 		resop = &res.array[where - 1];
1302 		ASSERT(resop->resop == OP_GETFH);
1303 		tmpfhp = &resop->nfs_resop4_u.opgetfh.object;
1304 		nthcomp = res.array_len/3 - 1;
1305 
1306 		/*
1307 		 * Need to call nfs4_end_op before resolve_sympath to avoid
1308 		 * potential nfs4_start_op deadlock.
1309 		 */
1310 		if (!recovery)
1311 			nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state,
1312 				needrecov);
1313 
1314 		ep->error = resolve_sympath(mi, svp, nthcomp, tmpfhp, cr,
1315 					    flags);
1316 
1317 		nfs4args_lookup_free(argop, num_argops);
1318 		kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4));
1319 		(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1320 
1321 		if (ep->error)
1322 			return;
1323 
1324 		goto recov_retry;
1325 	}
1326 
1327 	/* getfh */
1328 	resop = &res.array[res.array_len - 2];
1329 	ASSERT(resop->resop == OP_GETFH);
1330 	resfhp = &resop->nfs_resop4_u.opgetfh.object;
1331 
1332 	/* getattr fsinfo res */
1333 	resop++;
1334 	garp = &resop->nfs_resop4_u.opgetattr.ga_res;
1335 
1336 	*vtp = garp->n4g_va.va_type;
1337 
1338 	mi->mi_fh_expire_type = garp->n4g_ext_res->n4g_fet;
1339 
1340 	mutex_enter(&mi->mi_lock);
1341 	if (garp->n4g_ext_res->n4g_pc4.pc4_link_support)
1342 		mi->mi_flags |= MI4_LINK;
1343 	if (garp->n4g_ext_res->n4g_pc4.pc4_symlink_support)
1344 		mi->mi_flags |= MI4_SYMLINK;
1345 	if (garp->n4g_ext_res->n4g_suppattrs & FATTR4_ACL_MASK)
1346 		mi->mi_flags |= MI4_ACL;
1347 	mutex_exit(&mi->mi_lock);
1348 
1349 	if (garp->n4g_ext_res->n4g_maxread == 0)
1350 		mi->mi_tsize =
1351 			MIN(MAXBSIZE, mi->mi_tsize);
1352 	else
1353 		mi->mi_tsize =
1354 			MIN(garp->n4g_ext_res->n4g_maxread,
1355 			    mi->mi_tsize);
1356 
1357 	if (garp->n4g_ext_res->n4g_maxwrite == 0)
1358 		mi->mi_stsize =
1359 			MIN(MAXBSIZE, mi->mi_stsize);
1360 	else
1361 		mi->mi_stsize =
1362 			MIN(garp->n4g_ext_res->n4g_maxwrite,
1363 			    mi->mi_stsize);
1364 
1365 	if (garp->n4g_ext_res->n4g_maxfilesize != 0)
1366 		mi->mi_maxfilesize =
1367 			MIN(garp->n4g_ext_res->n4g_maxfilesize,
1368 			    mi->mi_maxfilesize);
1369 
1370 	/*
1371 	 * If the final component is a a symbolic link, resolve the symlink,
1372 	 * then try mount again using the new path.
1373 	 *
1374 	 * Assume no symbolic link for root filesysm "/".
1375 	 */
1376 	if (*vtp == VLNK) {
1377 		/*
1378 		 * nthcomp is the total result length minus
1379 		 * the 1st 2 OPs (PUTROOTFH, GETFH),
1380 		 * then divided by 3 (LOOKUP,GETFH,GETATTR)
1381 		 *
1382 		 * e.g. PUTROOTFH GETFH LOOKUP 1st-comp GETFH GETATTR
1383 		 *	LOOKUP 2nd-comp GETFH GETATTR
1384 		 *
1385 		 *	(8 - 2)/3 = 2
1386 		 */
1387 		nthcomp = (res.array_len - 2)/3;
1388 
1389 		/*
1390 		 * Need to call nfs4_end_op before resolve_sympath to avoid
1391 		 * potential nfs4_start_op deadlock. See RFE 4777612.
1392 		 */
1393 		if (!recovery)
1394 			nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state,
1395 				needrecov);
1396 
1397 		ep->error = resolve_sympath(mi, svp, nthcomp, resfhp, cr,
1398 					flags);
1399 
1400 		nfs4args_lookup_free(argop, num_argops);
1401 		kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4));
1402 		(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1403 
1404 		if (ep->error)
1405 			return;
1406 
1407 		goto recov_retry;
1408 	}
1409 
1410 	/*
1411 	 * We need to figure out where in the compound the getfh
1412 	 * for the parent directory is. If the object to be mounted is
1413 	 * the root, then there is no lookup at all:
1414 	 * PUTROOTFH, GETFH.
1415 	 * If the object to be mounted is in the root, then the compound is:
1416 	 * PUTROOTFH, GETFH, LOOKUP, GETFH, GETATTR.
1417 	 * In either of these cases, the index of the GETFH is 1.
1418 	 * If it is not at the root, then it's something like:
1419 	 * PUTROOTFH, GETFH, LOOKUP, GETFH, GETATTR,
1420 	 * LOOKUP, GETFH, GETATTR
1421 	 * In this case, the index is llndx (last lookup index) - 2.
1422 	 */
1423 	if (llndx == -1 || llndx == 2)
1424 		resop = &res.array[1];
1425 	else {
1426 		ASSERT(llndx > 2);
1427 		resop = &res.array[llndx-2];
1428 	}
1429 
1430 	ASSERT(resop->resop == OP_GETFH);
1431 	tmpfhp = &resop->nfs_resop4_u.opgetfh.object;
1432 
1433 	/* save the filehandles for the replica */
1434 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
1435 	ASSERT(tmpfhp->nfs_fh4_len <= NFS4_FHSIZE);
1436 	svp->sv_pfhandle.fh_len = tmpfhp->nfs_fh4_len;
1437 	bcopy(tmpfhp->nfs_fh4_val, svp->sv_pfhandle.fh_buf,
1438 	    tmpfhp->nfs_fh4_len);
1439 	ASSERT(resfhp->nfs_fh4_len <= NFS4_FHSIZE);
1440 	svp->sv_fhandle.fh_len = resfhp->nfs_fh4_len;
1441 	bcopy(resfhp->nfs_fh4_val, svp->sv_fhandle.fh_buf, resfhp->nfs_fh4_len);
1442 
1443 	/* initialize fsid and supp_attrs for server fs */
1444 	svp->sv_fsid = garp->n4g_fsid;
1445 	svp->sv_supp_attrs =
1446 		garp->n4g_ext_res->n4g_suppattrs | FATTR4_MANDATTR_MASK;
1447 
1448 	nfs_rw_exit(&svp->sv_lock);
1449 
1450 	nfs4args_lookup_free(argop, num_argops);
1451 	kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4));
1452 	(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1453 	if (!recovery)
1454 		nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state, needrecov);
1455 }
1456 
1457 static ushort_t nfs4_max_threads = 8;	/* max number of active async threads */
1458 static uint_t nfs4_bsize = 32 * 1024;	/* client `block' size */
1459 static uint_t nfs4_async_clusters = 1;	/* # of reqs from each async queue */
1460 static uint_t nfs4_cots_timeo = NFS_COTS_TIMEO;
1461 
1462 /*
1463  * Remap the root filehandle for the given filesystem.
1464  *
1465  * results returned via the nfs4_error_t parameter.
1466  */
1467 void
1468 nfs4_remap_root(mntinfo4_t *mi, nfs4_error_t *ep, int flags)
1469 {
1470 	struct servinfo4 *svp;
1471 	vtype_t vtype;
1472 	nfs_fh4 rootfh;
1473 	int getfh_flags;
1474 	char *orig_sv_path;
1475 	int orig_sv_pathlen, num_retry;
1476 
1477 	mutex_enter(&mi->mi_lock);
1478 
1479 remap_retry:
1480 	svp = mi->mi_curr_serv;
1481 	getfh_flags =
1482 		(flags & NFS4_REMAP_NEEDSOP) ? NFS4_GETFH_NEEDSOP : 0;
1483 	getfh_flags |=
1484 		(mi->mi_flags & MI4_PUBLIC) ? NFS4_GETFH_PUBLIC : 0;
1485 	mutex_exit(&mi->mi_lock);
1486 
1487 	/*
1488 	 * Just in case server path being mounted contains
1489 	 * symlinks and fails w/STALE, save the initial sv_path
1490 	 * so we can redrive the initial mount compound with the
1491 	 * initial sv_path -- not a symlink-expanded version.
1492 	 *
1493 	 * This could only happen if a symlink was expanded
1494 	 * and the expanded mount compound failed stale.  Because
1495 	 * it could be the case that the symlink was removed at
1496 	 * the server (and replaced with another symlink/dir,
1497 	 * we need to use the initial sv_path when attempting
1498 	 * to re-lookup everything and recover.
1499 	 */
1500 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1501 	orig_sv_pathlen = svp->sv_pathlen;
1502 	orig_sv_path = kmem_alloc(orig_sv_pathlen, KM_SLEEP);
1503 	bcopy(svp->sv_path, orig_sv_path, orig_sv_pathlen);
1504 	nfs_rw_exit(&svp->sv_lock);
1505 
1506 	num_retry = nfs4_max_mount_retry;
1507 
1508 	do {
1509 		/*
1510 		 * Get the root fh from the server.  Retry nfs4_max_mount_retry
1511 		 * (2) times if it fails with STALE since the recovery
1512 		 * infrastructure doesn't do STALE recovery for components
1513 		 * of the server path to the object being mounted.
1514 		 */
1515 		nfs4getfh_otw(mi, svp, &vtype, getfh_flags, CRED(), ep);
1516 
1517 		if (ep->error == 0 && ep->stat == NFS4_OK)
1518 			break;
1519 
1520 		/*
1521 		 * For some reason, the mount compound failed.  Before
1522 		 * retrying, we need to restore the original sv_path
1523 		 * because it might have contained symlinks that were
1524 		 * expanded by nfsgetfh_otw before the failure occurred.
1525 		 * replace current sv_path with orig sv_path -- just in case
1526 		 * it changed due to embedded symlinks.
1527 		 */
1528 		(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1529 		if (orig_sv_pathlen != svp->sv_pathlen) {
1530 			kmem_free(svp->sv_path, svp->sv_pathlen);
1531 			svp->sv_path = kmem_alloc(orig_sv_pathlen, KM_SLEEP);
1532 			svp->sv_pathlen = orig_sv_pathlen;
1533 		}
1534 		bcopy(orig_sv_path, svp->sv_path, orig_sv_pathlen);
1535 		nfs_rw_exit(&svp->sv_lock);
1536 
1537 	} while (num_retry-- > 0);
1538 
1539 	kmem_free(orig_sv_path, orig_sv_pathlen);
1540 
1541 	if (ep->error != 0 || ep->stat != 0) {
1542 		return;
1543 	}
1544 
1545 	if (vtype != VNON && vtype != mi->mi_type) {
1546 		/* shouldn't happen */
1547 		zcmn_err(mi->mi_zone->zone_id, CE_WARN,
1548 			"nfs4_remap_root: server root vnode type (%d) doesn't "
1549 			"match mount info (%d)", vtype, mi->mi_type);
1550 	}
1551 
1552 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1553 	rootfh.nfs_fh4_val = svp->sv_fhandle.fh_buf;
1554 	rootfh.nfs_fh4_len = svp->sv_fhandle.fh_len;
1555 	nfs_rw_exit(&svp->sv_lock);
1556 	sfh4_update(mi->mi_rootfh, &rootfh);
1557 
1558 	/*
1559 	 * It's possible that recovery took place on the filesystem
1560 	 * and the server has been updated between the time we did
1561 	 * the nfs4getfh_otw and now. Re-drive the otw operation
1562 	 * to make sure we have a good fh.
1563 	 */
1564 	mutex_enter(&mi->mi_lock);
1565 	if (mi->mi_curr_serv != svp)
1566 		goto remap_retry;
1567 
1568 	mutex_exit(&mi->mi_lock);
1569 }
1570 
1571 static int
1572 nfs4rootvp(vnode_t **rtvpp, vfs_t *vfsp, struct servinfo4 *svp_head,
1573 	int flags, cred_t *cr, zone_t *zone)
1574 {
1575 	vnode_t *rtvp = NULL;
1576 	mntinfo4_t *mi;
1577 	dev_t nfs_dev;
1578 	int error = 0;
1579 	rnode4_t *rp;
1580 	int i;
1581 	struct vattr va;
1582 	vtype_t vtype = VNON;
1583 	vtype_t tmp_vtype = VNON;
1584 	struct servinfo4 *firstsvp = NULL, *svp = svp_head;
1585 	nfs4_oo_hash_bucket_t *bucketp;
1586 	nfs_fh4 fh;
1587 	char *droptext = "";
1588 	struct nfs_stats *nfsstatsp;
1589 	nfs4_fname_t *mfname;
1590 	nfs4_error_t e;
1591 	char *orig_sv_path;
1592 	int orig_sv_pathlen, num_retry, removed;
1593 	cred_t *lcr = NULL, *tcr = cr;
1594 
1595 	nfsstatsp = zone_getspecific(nfsstat_zone_key, nfs_zone());
1596 	ASSERT(nfsstatsp != NULL);
1597 
1598 	ASSERT(nfs_zone() == zone);
1599 	ASSERT(crgetref(cr));
1600 
1601 	/*
1602 	 * Create a mount record and link it to the vfs struct.
1603 	 */
1604 	mi = kmem_zalloc(sizeof (*mi), KM_SLEEP);
1605 	mutex_init(&mi->mi_lock, NULL, MUTEX_DEFAULT, NULL);
1606 	nfs_rw_init(&mi->mi_recovlock, NULL, RW_DEFAULT, NULL);
1607 	nfs_rw_init(&mi->mi_rename_lock, NULL, RW_DEFAULT, NULL);
1608 	nfs_rw_init(&mi->mi_fh_lock, NULL, RW_DEFAULT, NULL);
1609 
1610 	if (!(flags & NFSMNT_SOFT))
1611 		mi->mi_flags |= MI4_HARD;
1612 	if ((flags & NFSMNT_NOPRINT))
1613 		mi->mi_flags |= MI4_NOPRINT;
1614 	if (flags & NFSMNT_INT)
1615 		mi->mi_flags |= MI4_INT;
1616 	if (flags & NFSMNT_PUBLIC)
1617 		mi->mi_flags |= MI4_PUBLIC;
1618 	mi->mi_retrans = NFS_RETRIES;
1619 	if (svp->sv_knconf->knc_semantics == NC_TPI_COTS_ORD ||
1620 	    svp->sv_knconf->knc_semantics == NC_TPI_COTS)
1621 		mi->mi_timeo = nfs4_cots_timeo;
1622 	else
1623 		mi->mi_timeo = NFS_TIMEO;
1624 	mi->mi_prog = NFS_PROGRAM;
1625 	mi->mi_vers = NFS_V4;
1626 	mi->mi_rfsnames = rfsnames_v4;
1627 	mi->mi_reqs = nfsstatsp->nfs_stats_v4.rfsreqcnt_ptr;
1628 	cv_init(&mi->mi_failover_cv, NULL, CV_DEFAULT, NULL);
1629 	mi->mi_servers = svp;
1630 	mi->mi_curr_serv = svp;
1631 	mi->mi_acregmin = SEC2HR(ACREGMIN);
1632 	mi->mi_acregmax = SEC2HR(ACREGMAX);
1633 	mi->mi_acdirmin = SEC2HR(ACDIRMIN);
1634 	mi->mi_acdirmax = SEC2HR(ACDIRMAX);
1635 	mi->mi_fh_expire_type = FH4_PERSISTENT;
1636 	mi->mi_clientid_next = NULL;
1637 	mi->mi_clientid_prev = NULL;
1638 	mi->mi_grace_wait = 0;
1639 	mi->mi_error = 0;
1640 	mi->mi_srvsettime = 0;
1641 
1642 	mi->mi_count = 1;
1643 
1644 	mi->mi_tsize = nfs4_tsize(svp->sv_knconf);
1645 	mi->mi_stsize = mi->mi_tsize;
1646 
1647 	if (flags & NFSMNT_DIRECTIO)
1648 		mi->mi_flags |= MI4_DIRECTIO;
1649 
1650 	mi->mi_flags |= MI4_MOUNTING;
1651 
1652 	/*
1653 	 * Make a vfs struct for nfs.  We do this here instead of below
1654 	 * because rtvp needs a vfs before we can do a getattr on it.
1655 	 *
1656 	 * Assign a unique device id to the mount
1657 	 */
1658 	mutex_enter(&nfs_minor_lock);
1659 	do {
1660 		nfs_minor = (nfs_minor + 1) & MAXMIN32;
1661 		nfs_dev = makedevice(nfs_major, nfs_minor);
1662 	} while (vfs_devismounted(nfs_dev));
1663 	mutex_exit(&nfs_minor_lock);
1664 
1665 	vfsp->vfs_dev = nfs_dev;
1666 	vfs_make_fsid(&vfsp->vfs_fsid, nfs_dev, nfs4fstyp);
1667 	vfsp->vfs_data = (caddr_t)mi;
1668 	vfsp->vfs_fstype = nfsfstyp;
1669 	vfsp->vfs_bsize = nfs4_bsize;
1670 
1671 	/*
1672 	 * Initialize fields used to support async putpage operations.
1673 	 */
1674 	for (i = 0; i < NFS4_ASYNC_TYPES; i++)
1675 		mi->mi_async_clusters[i] = nfs4_async_clusters;
1676 	mi->mi_async_init_clusters = nfs4_async_clusters;
1677 	mi->mi_async_curr = &mi->mi_async_reqs[0];
1678 	mi->mi_max_threads = nfs4_max_threads;
1679 	mutex_init(&mi->mi_async_lock, NULL, MUTEX_DEFAULT, NULL);
1680 	cv_init(&mi->mi_async_reqs_cv, NULL, CV_DEFAULT, NULL);
1681 	cv_init(&mi->mi_async_work_cv, NULL, CV_DEFAULT, NULL);
1682 	cv_init(&mi->mi_async_cv, NULL, CV_DEFAULT, NULL);
1683 	cv_init(&mi->mi_inact_req_cv, NULL, CV_DEFAULT, NULL);
1684 
1685 	mi->mi_vfsp = vfsp;
1686 	zone_hold(mi->mi_zone = zone);
1687 	nfs4_mi_zonelist_add(mi);
1688 
1689 	/*
1690 	 * Initialize the <open owner/cred> hash table.
1691 	 */
1692 	for (i = 0; i < NFS4_NUM_OO_BUCKETS; i++) {
1693 		bucketp = &(mi->mi_oo_list[i]);
1694 		mutex_init(&bucketp->b_lock, NULL, MUTEX_DEFAULT, NULL);
1695 		list_create(&bucketp->b_oo_hash_list,
1696 		    sizeof (nfs4_open_owner_t),
1697 		    offsetof(nfs4_open_owner_t, oo_hash_node));
1698 	}
1699 
1700 	/*
1701 	 * Initialize the freed open owner list.
1702 	 */
1703 	mi->mi_foo_num = 0;
1704 	mi->mi_foo_max = NFS4_NUM_FREED_OPEN_OWNERS;
1705 	list_create(&mi->mi_foo_list, sizeof (nfs4_open_owner_t),
1706 	    offsetof(nfs4_open_owner_t, oo_foo_node));
1707 
1708 	list_create(&mi->mi_lost_state, sizeof (nfs4_lost_rqst_t),
1709 	    offsetof(nfs4_lost_rqst_t, lr_node));
1710 
1711 	list_create(&mi->mi_bseqid_list, sizeof (nfs4_bseqid_entry_t),
1712 	    offsetof(nfs4_bseqid_entry_t, bs_node));
1713 
1714 	/*
1715 	 * Initialize the msg buffer.
1716 	 */
1717 	list_create(&mi->mi_msg_list, sizeof (nfs4_debug_msg_t),
1718 	    offsetof(nfs4_debug_msg_t, msg_node));
1719 	mi->mi_msg_count = 0;
1720 	mutex_init(&mi->mi_msg_list_lock, NULL, MUTEX_DEFAULT, NULL);
1721 
1722 	/*
1723 	 * Initialize kstats
1724 	 */
1725 	nfs4_mnt_kstat_init(vfsp);
1726 
1727 	/*
1728 	 * Initialize the shared filehandle pool, and get the fname for
1729 	 * the filesystem root.
1730 	 */
1731 	sfh4_createtab(&mi->mi_filehandles);
1732 	mi->mi_fname = fn_get(NULL, ".");
1733 
1734 	/*
1735 	 * Save server path we're attempting to mount.
1736 	 */
1737 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
1738 	orig_sv_pathlen = svp_head->sv_pathlen;
1739 	orig_sv_path = kmem_alloc(svp_head->sv_pathlen, KM_SLEEP);
1740 	bcopy(svp_head->sv_path, orig_sv_path, svp_head->sv_pathlen);
1741 	nfs_rw_exit(&svp->sv_lock);
1742 
1743 	/*
1744 	 * Make the GETFH call to get root fh for each replica.
1745 	 */
1746 	if (svp_head->sv_next)
1747 		droptext = ", dropping replica";
1748 
1749 	/*
1750 	 * If the uid is set then set the creds for secure mounts
1751 	 * by proxy processes such as automountd.
1752 	 */
1753 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1754 	if (svp->sv_secdata->uid != 0) {
1755 		lcr = crdup(cr);
1756 		(void) crsetugid(lcr, svp->sv_secdata->uid, crgetgid(cr));
1757 		tcr = lcr;
1758 	}
1759 	nfs_rw_exit(&svp->sv_lock);
1760 	for (svp = svp_head; svp; svp = svp->sv_next) {
1761 		if (nfs4_chkdup_servinfo4(svp_head, svp)) {
1762 			nfs_cmn_err(error, CE_WARN,
1763 				VERS_MSG "Host %s is a duplicate%s",
1764 				svp->sv_hostname, droptext);
1765 			(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
1766 			svp->sv_flags |= SV4_NOTINUSE;
1767 			nfs_rw_exit(&svp->sv_lock);
1768 			continue;
1769 		}
1770 		mi->mi_curr_serv = svp;
1771 
1772 		/*
1773 		 * Just in case server path being mounted contains
1774 		 * symlinks and fails w/STALE, save the initial sv_path
1775 		 * so we can redrive the initial mount compound with the
1776 		 * initial sv_path -- not a symlink-expanded version.
1777 		 *
1778 		 * This could only happen if a symlink was expanded
1779 		 * and the expanded mount compound failed stale.  Because
1780 		 * it could be the case that the symlink was removed at
1781 		 * the server (and replaced with another symlink/dir,
1782 		 * we need to use the initial sv_path when attempting
1783 		 * to re-lookup everything and recover.
1784 		 *
1785 		 * Other mount errors should evenutally be handled here also
1786 		 * (NFS4ERR_DELAY, NFS4ERR_RESOURCE).  For now, all mount
1787 		 * failures will result in mount being redriven a few times.
1788 		 */
1789 		num_retry = nfs4_max_mount_retry;
1790 		do {
1791 			nfs4getfh_otw(mi, svp, &tmp_vtype,
1792 			    ((flags & NFSMNT_PUBLIC) ? NFS4_GETFH_PUBLIC : 0) |
1793 			    NFS4_GETFH_NEEDSOP, tcr, &e);
1794 
1795 			if (e.error == 0 && e.stat == NFS4_OK)
1796 				break;
1797 
1798 			/*
1799 			 * replace current sv_path with orig sv_path -- just in
1800 			 * case it changed due to embedded symlinks.
1801 			 */
1802 			(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1803 			if (orig_sv_pathlen != svp->sv_pathlen) {
1804 				kmem_free(svp->sv_path, svp->sv_pathlen);
1805 				svp->sv_path = kmem_alloc(orig_sv_pathlen,
1806 							KM_SLEEP);
1807 				svp->sv_pathlen = orig_sv_pathlen;
1808 			}
1809 			bcopy(orig_sv_path, svp->sv_path, orig_sv_pathlen);
1810 			nfs_rw_exit(&svp->sv_lock);
1811 
1812 		} while (num_retry-- > 0);
1813 
1814 		error = e.error ? e.error : geterrno4(e.stat);
1815 		if (error) {
1816 			nfs_cmn_err(error, CE_WARN,
1817 				VERS_MSG "initial call to %s failed%s: %m",
1818 				svp->sv_hostname, droptext);
1819 			(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
1820 			svp->sv_flags |= SV4_NOTINUSE;
1821 			nfs_rw_exit(&svp->sv_lock);
1822 			mi->mi_flags &= ~MI4_RECOV_FAIL;
1823 			mi->mi_error = 0;
1824 			continue;
1825 		}
1826 
1827 		if (tmp_vtype == VBAD) {
1828 			zcmn_err(mi->mi_zone->zone_id, CE_WARN,
1829 				VERS_MSG "%s returned a bad file type for "
1830 				"root%s", svp->sv_hostname, droptext);
1831 			(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
1832 			svp->sv_flags |= SV4_NOTINUSE;
1833 			nfs_rw_exit(&svp->sv_lock);
1834 			continue;
1835 		}
1836 
1837 		if (vtype == VNON) {
1838 			vtype = tmp_vtype;
1839 		} else if (vtype != tmp_vtype) {
1840 			zcmn_err(mi->mi_zone->zone_id, CE_WARN,
1841 				VERS_MSG "%s returned a different file type "
1842 				"for root%s", svp->sv_hostname, droptext);
1843 			(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
1844 			svp->sv_flags |= SV4_NOTINUSE;
1845 			nfs_rw_exit(&svp->sv_lock);
1846 			continue;
1847 		}
1848 		if (firstsvp == NULL)
1849 			firstsvp = svp;
1850 	}
1851 
1852 	kmem_free(orig_sv_path, orig_sv_pathlen);
1853 
1854 	if (firstsvp == NULL) {
1855 		if (error == 0)
1856 			error = ENOENT;
1857 		goto bad;
1858 	}
1859 
1860 	mi->mi_curr_serv = svp = firstsvp;
1861 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1862 	ASSERT((mi->mi_curr_serv->sv_flags & SV4_NOTINUSE) == 0);
1863 	fh.nfs_fh4_len = svp->sv_fhandle.fh_len;
1864 	fh.nfs_fh4_val = svp->sv_fhandle.fh_buf;
1865 	mi->mi_rootfh = sfh4_get(&fh, mi);
1866 	fh.nfs_fh4_len = svp->sv_pfhandle.fh_len;
1867 	fh.nfs_fh4_val = svp->sv_pfhandle.fh_buf;
1868 	mi->mi_srvparentfh = sfh4_get(&fh, mi);
1869 	nfs_rw_exit(&svp->sv_lock);
1870 
1871 	/*
1872 	 * Make the root vnode without attributes.
1873 	 */
1874 	mfname = mi->mi_fname;
1875 	fn_hold(mfname);
1876 	rtvp = makenfs4node_by_fh(mi->mi_rootfh, NULL,
1877 	    &mfname, NULL, mi, cr, gethrtime());
1878 	rtvp->v_type = vtype;
1879 
1880 	mi->mi_curread = mi->mi_tsize;
1881 	mi->mi_curwrite = mi->mi_stsize;
1882 
1883 	/*
1884 	 * Start the manager thread responsible for handling async worker
1885 	 * threads.
1886 	 */
1887 	MI4_HOLD(mi);
1888 	VFS_HOLD(vfsp);	/* add reference for thread */
1889 	mi->mi_manager_thread = zthread_create(NULL, 0, nfs4_async_manager,
1890 					vfsp, 0, minclsyspri);
1891 	ASSERT(mi->mi_manager_thread != NULL);
1892 
1893 	/*
1894 	 * Create the thread that handles over-the-wire calls for
1895 	 * VOP_INACTIVE.
1896 	 * This needs to happen after the manager thread is created.
1897 	 */
1898 	MI4_HOLD(mi);
1899 	mi->mi_inactive_thread = zthread_create(NULL, 0, nfs4_inactive_thread,
1900 					mi, 0, minclsyspri);
1901 	ASSERT(mi->mi_inactive_thread != NULL);
1902 
1903 	/* If we didn't get a type, get one now */
1904 	if (rtvp->v_type == VNON) {
1905 		va.va_mask = AT_TYPE;
1906 		error = nfs4getattr(rtvp, &va, tcr);
1907 		if (error)
1908 			goto bad;
1909 		rtvp->v_type = va.va_type;
1910 	}
1911 
1912 	mi->mi_type = rtvp->v_type;
1913 
1914 	mutex_enter(&mi->mi_lock);
1915 	mi->mi_flags &= ~MI4_MOUNTING;
1916 	mutex_exit(&mi->mi_lock);
1917 
1918 	*rtvpp = rtvp;
1919 	if (lcr != NULL)
1920 		crfree(lcr);
1921 
1922 	return (0);
1923 bad:
1924 	/*
1925 	 * An error occurred somewhere, need to clean up...
1926 	 */
1927 	if (lcr != NULL)
1928 		crfree(lcr);
1929 	if (rtvp != NULL) {
1930 		/*
1931 		 * We need to release our reference to the root vnode and
1932 		 * destroy the mntinfo4 struct that we just created.
1933 		 */
1934 		rp = VTOR4(rtvp);
1935 		if (rp->r_flags & R4HASHED)
1936 			rp4_rmhash(rp);
1937 		VN_RELE(rtvp);
1938 	}
1939 	nfs4_async_stop(vfsp);
1940 	nfs4_async_manager_stop(vfsp);
1941 	removed = nfs4_mi_zonelist_remove(mi);
1942 	if (removed)
1943 		zone_rele(mi->mi_zone);
1944 
1945 	/*
1946 	 * This releases the initial "hold" of the mi since it will never
1947 	 * be referenced by the vfsp.  Also, when mount returns to vfs.c
1948 	 * with an error, the vfsp will be destroyed, not rele'd.
1949 	 */
1950 	MI4_RELE(mi);
1951 
1952 	*rtvpp = NULL;
1953 	return (error);
1954 }
1955 
1956 /*
1957  * vfs operations
1958  */
1959 static int
1960 nfs4_unmount(vfs_t *vfsp, int flag, cred_t *cr)
1961 {
1962 	mntinfo4_t *mi;
1963 	ushort_t omax;
1964 	int removed;
1965 
1966 	if (secpolicy_fs_unmount(cr, vfsp) != 0)
1967 		return (EPERM);
1968 
1969 	mi = VFTOMI4(vfsp);
1970 
1971 	if (flag & MS_FORCE) {
1972 		vfsp->vfs_flag |= VFS_UNMOUNTED;
1973 		if (nfs_zone() != mi->mi_zone) {
1974 			/*
1975 			 * If the request is coming from the wrong zone,
1976 			 * we don't want to create any new threads, and
1977 			 * performance is not a concern.  Do everything
1978 			 * inline.
1979 			 */
1980 			NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE,
1981 			    "nfs4_unmount x-zone forced unmount of vfs %p\n",
1982 			    (void *)vfsp));
1983 			nfs4_free_mount(vfsp, cr);
1984 		} else {
1985 			/*
1986 			 * Free data structures asynchronously, to avoid
1987 			 * blocking the current thread (for performance
1988 			 * reasons only).
1989 			 */
1990 			async_free_mount(vfsp, cr);
1991 		}
1992 		return (0);
1993 	}
1994 	/*
1995 	 * Wait until all asynchronous putpage operations on
1996 	 * this file system are complete before flushing rnodes
1997 	 * from the cache.
1998 	 */
1999 	omax = mi->mi_max_threads;
2000 	if (nfs4_async_stop_sig(vfsp)) {
2001 		return (EINTR);
2002 	}
2003 	r4flush(vfsp, cr);
2004 	/*
2005 	 * If there are any active vnodes on this file system,
2006 	 * then the file system is busy and can't be umounted.
2007 	 */
2008 	if (check_rtable4(vfsp)) {
2009 		mutex_enter(&mi->mi_async_lock);
2010 		mi->mi_max_threads = omax;
2011 		mutex_exit(&mi->mi_async_lock);
2012 		return (EBUSY);
2013 	}
2014 	/*
2015 	 * The unmount can't fail from now on, and there are no active
2016 	 * files that could require over-the-wire calls to the server,
2017 	 * so stop the async manager and the inactive thread.
2018 	 */
2019 	nfs4_async_manager_stop(vfsp);
2020 	/*
2021 	 * Destroy all rnodes belonging to this file system from the
2022 	 * rnode hash queues and purge any resources allocated to
2023 	 * them.
2024 	 */
2025 	destroy_rtable4(vfsp, cr);
2026 	vfsp->vfs_flag |= VFS_UNMOUNTED;
2027 
2028 	nfs4_remove_mi_from_server(mi, NULL);
2029 	removed = nfs4_mi_zonelist_remove(mi);
2030 	if (removed)
2031 		zone_rele(mi->mi_zone);
2032 
2033 	return (0);
2034 }
2035 
2036 /*
2037  * find root of nfs
2038  */
2039 static int
2040 nfs4_root(vfs_t *vfsp, vnode_t **vpp)
2041 {
2042 	mntinfo4_t *mi;
2043 	vnode_t *vp;
2044 	nfs4_fname_t *mfname;
2045 	servinfo4_t *svp;
2046 
2047 	mi = VFTOMI4(vfsp);
2048 
2049 	if (nfs_zone() != mi->mi_zone)
2050 		return (EPERM);
2051 
2052 	svp = mi->mi_curr_serv;
2053 	if (svp) {
2054 		(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
2055 		if (svp->sv_flags & SV4_ROOT_STALE) {
2056 			nfs_rw_exit(&svp->sv_lock);
2057 
2058 			(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
2059 			if (svp->sv_flags & SV4_ROOT_STALE) {
2060 				svp->sv_flags &= ~SV4_ROOT_STALE;
2061 				nfs_rw_exit(&svp->sv_lock);
2062 				return (ENOENT);
2063 			}
2064 			nfs_rw_exit(&svp->sv_lock);
2065 		} else
2066 			nfs_rw_exit(&svp->sv_lock);
2067 	}
2068 
2069 	mfname = mi->mi_fname;
2070 	fn_hold(mfname);
2071 	vp = makenfs4node_by_fh(mi->mi_rootfh, NULL, &mfname, NULL,
2072 	    VFTOMI4(vfsp), CRED(), gethrtime());
2073 
2074 	if (VTOR4(vp)->r_flags & R4STALE) {
2075 		VN_RELE(vp);
2076 		return (ENOENT);
2077 	}
2078 
2079 	ASSERT(vp->v_type == VNON || vp->v_type == mi->mi_type);
2080 
2081 	vp->v_type = mi->mi_type;
2082 
2083 	*vpp = vp;
2084 
2085 	return (0);
2086 }
2087 
2088 static int
2089 nfs4_statfs_otw(vnode_t *vp, struct statvfs64 *sbp, cred_t *cr)
2090 {
2091 	int error;
2092 	nfs4_ga_res_t gar;
2093 	nfs4_ga_ext_res_t ger;
2094 
2095 	gar.n4g_ext_res = &ger;
2096 
2097 	if (error = nfs4_attr_otw(vp, TAG_FSINFO, &gar,
2098 	    NFS4_STATFS_ATTR_MASK, cr))
2099 		return (error);
2100 
2101 	*sbp = gar.n4g_ext_res->n4g_sb;
2102 
2103 	return (0);
2104 }
2105 
2106 /*
2107  * Get file system statistics.
2108  */
2109 static int
2110 nfs4_statvfs(vfs_t *vfsp, struct statvfs64 *sbp)
2111 {
2112 	int error;
2113 	vnode_t *vp;
2114 	cred_t *cr;
2115 
2116 	error = nfs4_root(vfsp, &vp);
2117 	if (error)
2118 		return (error);
2119 
2120 	cr = CRED();
2121 
2122 	error = nfs4_statfs_otw(vp, sbp, cr);
2123 	if (!error) {
2124 		(void) strncpy(sbp->f_basetype,
2125 			vfssw[vfsp->vfs_fstype].vsw_name, FSTYPSZ);
2126 		sbp->f_flag = vf_to_stf(vfsp->vfs_flag);
2127 	} else {
2128 		nfs4_purge_stale_fh(error, vp, cr);
2129 	}
2130 
2131 	VN_RELE(vp);
2132 
2133 	return (error);
2134 }
2135 
2136 static kmutex_t nfs4_syncbusy;
2137 
2138 /*
2139  * Flush dirty nfs files for file system vfsp.
2140  * If vfsp == NULL, all nfs files are flushed.
2141  *
2142  * SYNC_CLOSE in flag is passed to us to
2143  * indicate that we are shutting down and or
2144  * rebooting.
2145  */
2146 static int
2147 nfs4_sync(vfs_t *vfsp, short flag, cred_t *cr)
2148 {
2149 	/*
2150 	 * Cross-zone calls are OK here, since this translates to a
2151 	 * VOP_PUTPAGE(B_ASYNC), which gets picked up by the right zone.
2152 	 */
2153 	if (!(flag & SYNC_ATTR) && mutex_tryenter(&nfs4_syncbusy) != 0) {
2154 		r4flush(vfsp, cr);
2155 		mutex_exit(&nfs4_syncbusy);
2156 	}
2157 
2158 	/*
2159 	 * if SYNC_CLOSE is set then we know that
2160 	 * the system is rebooting, mark the mntinfo
2161 	 * for later examination.
2162 	 */
2163 	if (vfsp && (flag & SYNC_CLOSE)) {
2164 		mntinfo4_t *mi;
2165 
2166 		mi = VFTOMI4(vfsp);
2167 		if (!(mi->mi_flags & MI4_SHUTDOWN)) {
2168 			mutex_enter(&mi->mi_lock);
2169 			mi->mi_flags |= MI4_SHUTDOWN;
2170 			mutex_exit(&mi->mi_lock);
2171 		}
2172 	}
2173 	return (0);
2174 }
2175 
2176 /*
2177  * vget is difficult, if not impossible, to support in v4 because we don't
2178  * know the parent directory or name, which makes it impossible to create a
2179  * useful shadow vnode.  And we need the shadow vnode for things like
2180  * OPEN.
2181  */
2182 
2183 /* ARGSUSED */
2184 /*
2185  * XXX Check nfs4_vget_pseudo() for dependency.
2186  */
2187 static int
2188 nfs4_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
2189 {
2190 	return (EREMOTE);
2191 }
2192 
2193 /*
2194  * nfs4_mountroot get called in the case where we are diskless booting.  All
2195  * we need from here is the ability to get the server info and from there we
2196  * can simply call nfs4_rootvp.
2197  */
2198 /* ARGSUSED */
2199 static int
2200 nfs4_mountroot(vfs_t *vfsp, whymountroot_t why)
2201 {
2202 	vnode_t *rtvp;
2203 	char root_hostname[SYS_NMLN+1];
2204 	struct servinfo4 *svp;
2205 	int error;
2206 	int vfsflags;
2207 	size_t size;
2208 	char *root_path;
2209 	struct pathname pn;
2210 	char *name;
2211 	cred_t *cr;
2212 	mntinfo4_t *mi;
2213 	struct nfs_args args;		/* nfs mount arguments */
2214 	static char token[10];
2215 	nfs4_error_t n4e;
2216 
2217 	bzero(&args, sizeof (args));
2218 
2219 	/* do this BEFORE getfile which causes xid stamps to be initialized */
2220 	clkset(-1L);		/* hack for now - until we get time svc? */
2221 
2222 	if (why == ROOT_REMOUNT) {
2223 		/*
2224 		 * Shouldn't happen.
2225 		 */
2226 		panic("nfs4_mountroot: why == ROOT_REMOUNT");
2227 	}
2228 
2229 	if (why == ROOT_UNMOUNT) {
2230 		/*
2231 		 * Nothing to do for NFS.
2232 		 */
2233 		return (0);
2234 	}
2235 
2236 	/*
2237 	 * why == ROOT_INIT
2238 	 */
2239 
2240 	name = token;
2241 	*name = 0;
2242 	(void) getfsname("root", name, sizeof (token));
2243 
2244 	pn_alloc(&pn);
2245 	root_path = pn.pn_path;
2246 
2247 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
2248 	nfs_rw_init(&svp->sv_lock, NULL, RW_DEFAULT, NULL);
2249 	svp->sv_knconf = kmem_zalloc(sizeof (*svp->sv_knconf), KM_SLEEP);
2250 	svp->sv_knconf->knc_protofmly = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
2251 	svp->sv_knconf->knc_proto = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
2252 
2253 	/*
2254 	 * Get server address
2255 	 * Get the root path
2256 	 * Get server's transport
2257 	 * Get server's hostname
2258 	 * Get options
2259 	 */
2260 	args.addr = &svp->sv_addr;
2261 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
2262 	args.fh = (char *)&svp->sv_fhandle;
2263 	args.knconf = svp->sv_knconf;
2264 	args.hostname = root_hostname;
2265 	vfsflags = 0;
2266 	if (error = mount_root(*name ? name : "root", root_path, NFS_V4,
2267 	    &args, &vfsflags)) {
2268 		if (error == EPROTONOSUPPORT)
2269 			nfs_cmn_err(error, CE_WARN, "nfs4_mountroot: "
2270 			    "mount_root failed: server doesn't support NFS V4");
2271 		else
2272 			nfs_cmn_err(error, CE_WARN,
2273 			    "nfs4_mountroot: mount_root failed: %m");
2274 		nfs_rw_exit(&svp->sv_lock);
2275 		sv4_free(svp);
2276 		pn_free(&pn);
2277 		return (error);
2278 	}
2279 	nfs_rw_exit(&svp->sv_lock);
2280 	svp->sv_hostnamelen = (int)(strlen(root_hostname) + 1);
2281 	svp->sv_hostname = kmem_alloc(svp->sv_hostnamelen, KM_SLEEP);
2282 	(void) strcpy(svp->sv_hostname, root_hostname);
2283 
2284 	svp->sv_pathlen = (int)(strlen(root_path) + 1);
2285 	svp->sv_path = kmem_alloc(svp->sv_pathlen, KM_SLEEP);
2286 	(void) strcpy(svp->sv_path, root_path);
2287 
2288 	/*
2289 	 * Force root partition to always be mounted with AUTH_UNIX for now
2290 	 */
2291 	svp->sv_secdata = kmem_alloc(sizeof (*svp->sv_secdata), KM_SLEEP);
2292 	svp->sv_secdata->secmod = AUTH_UNIX;
2293 	svp->sv_secdata->rpcflavor = AUTH_UNIX;
2294 	svp->sv_secdata->data = NULL;
2295 
2296 	cr = crgetcred();
2297 	rtvp = NULL;
2298 
2299 	error = nfs4rootvp(&rtvp, vfsp, svp, args.flags, cr, global_zone);
2300 
2301 	if (error) {
2302 		crfree(cr);
2303 		pn_free(&pn);
2304 		goto errout;
2305 	}
2306 
2307 	mi = VTOMI4(rtvp);
2308 
2309 	/*
2310 	 * Send client id to the server, if necessary
2311 	 */
2312 	nfs4_error_zinit(&n4e);
2313 	nfs4setclientid(mi, cr, FALSE, &n4e);
2314 	error = n4e.error;
2315 
2316 	crfree(cr);
2317 
2318 	if (error) {
2319 		pn_free(&pn);
2320 		goto errout;
2321 	}
2322 
2323 	error = nfs4_setopts(rtvp, DATAMODEL_NATIVE, &args);
2324 	if (error) {
2325 		nfs_cmn_err(error, CE_WARN,
2326 		    "nfs4_mountroot: invalid root mount options");
2327 		pn_free(&pn);
2328 		goto errout;
2329 	}
2330 
2331 	(void) vfs_lock_wait(vfsp);
2332 	vfs_add(NULL, vfsp, vfsflags);
2333 	vfs_unlock(vfsp);
2334 
2335 	size = strlen(svp->sv_hostname);
2336 	(void) strcpy(rootfs.bo_name, svp->sv_hostname);
2337 	rootfs.bo_name[size] = ':';
2338 	(void) strcpy(&rootfs.bo_name[size + 1], root_path);
2339 
2340 	pn_free(&pn);
2341 
2342 errout:
2343 	if (error) {
2344 		sv4_free(svp);
2345 		nfs4_async_stop(vfsp);
2346 		nfs4_async_manager_stop(vfsp);
2347 	}
2348 
2349 	if (rtvp != NULL)
2350 		VN_RELE(rtvp);
2351 
2352 	return (error);
2353 }
2354 
2355 /*
2356  * Initialization routine for VFS routines.  Should only be called once
2357  */
2358 int
2359 nfs4_vfsinit(void)
2360 {
2361 	mutex_init(&nfs4_syncbusy, NULL, MUTEX_DEFAULT, NULL);
2362 	nfs4setclientid_init();
2363 	return (0);
2364 }
2365 
2366 void
2367 nfs4_vfsfini(void)
2368 {
2369 	nfs4setclientid_fini();
2370 	mutex_destroy(&nfs4_syncbusy);
2371 }
2372 
2373 void
2374 nfs4_freevfs(vfs_t *vfsp)
2375 {
2376 	mntinfo4_t *mi;
2377 
2378 	/* need to release the initial hold */
2379 	mi = VFTOMI4(vfsp);
2380 	MI4_RELE(mi);
2381 }
2382 
2383 /*
2384  * Client side SETCLIENTID and SETCLIENTID_CONFIRM
2385  */
2386 struct nfs4_server nfs4_server_lst =
2387 	{ &nfs4_server_lst, &nfs4_server_lst };
2388 
2389 kmutex_t nfs4_server_lst_lock;
2390 
2391 static void
2392 nfs4setclientid_init(void)
2393 {
2394 	mutex_init(&nfs4_server_lst_lock, NULL, MUTEX_DEFAULT, NULL);
2395 }
2396 
2397 static void
2398 nfs4setclientid_fini(void)
2399 {
2400 	mutex_destroy(&nfs4_server_lst_lock);
2401 }
2402 
2403 int nfs4_retry_sclid_delay = NFS4_RETRY_SCLID_DELAY;
2404 int nfs4_num_sclid_retries = NFS4_NUM_SCLID_RETRIES;
2405 
2406 /*
2407  * Set the clientid for the server for "mi".  No-op if the clientid is
2408  * already set.
2409  *
2410  * The recovery boolean should be set to TRUE if this function was called
2411  * by the recovery code, and FALSE otherwise.  This is used to determine
2412  * if we need to call nfs4_start/end_op as well as grab the mi_recovlock
2413  * for adding a mntinfo4_t to a nfs4_server_t.
2414  *
2415  * Error is returned via 'n4ep'.  If there was a 'n4ep->stat' error, then
2416  * 'n4ep->error' is set to geterrno4(n4ep->stat).
2417  */
2418 void
2419 nfs4setclientid(mntinfo4_t *mi, cred_t *cr, bool_t recovery, nfs4_error_t *n4ep)
2420 {
2421 	struct nfs4_server *np;
2422 	struct servinfo4 *svp = mi->mi_curr_serv;
2423 	nfs4_recov_state_t recov_state;
2424 	int num_retries = 0;
2425 	bool_t retry;
2426 	cred_t *lcr = NULL;
2427 	int retry_inuse = 1; /* only retry once on NFS4ERR_CLID_INUSE */
2428 	time_t lease_time = 0;
2429 
2430 	recov_state.rs_flags = 0;
2431 	recov_state.rs_num_retry_despite_err = 0;
2432 	ASSERT(n4ep != NULL);
2433 
2434 recov_retry:
2435 	retry = FALSE;
2436 	nfs4_error_zinit(n4ep);
2437 	if (!recovery)
2438 		(void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_READER, 0);
2439 
2440 	mutex_enter(&nfs4_server_lst_lock);
2441 	np = servinfo4_to_nfs4_server(svp); /* This locks np if it is found */
2442 	mutex_exit(&nfs4_server_lst_lock);
2443 	if (!np) {
2444 		struct nfs4_server *tnp;
2445 		np = new_nfs4_server(svp, cr);
2446 		mutex_enter(&np->s_lock);
2447 
2448 		mutex_enter(&nfs4_server_lst_lock);
2449 		tnp = servinfo4_to_nfs4_server(svp);
2450 		if (tnp) {
2451 			/*
2452 			 * another thread snuck in and put server on list.
2453 			 * since we aren't adding it to the nfs4_server_list
2454 			 * we need to set the ref count to 0 and destroy it.
2455 			 */
2456 			np->s_refcnt = 0;
2457 			destroy_nfs4_server(np);
2458 			np = tnp;
2459 		} else {
2460 			/*
2461 			 * do not give list a reference until everything
2462 			 * succeeds
2463 			 */
2464 			insque(np, &nfs4_server_lst);
2465 		}
2466 		mutex_exit(&nfs4_server_lst_lock);
2467 	}
2468 	ASSERT(MUTEX_HELD(&np->s_lock));
2469 	/*
2470 	 * If we find the server already has N4S_CLIENTID_SET, then
2471 	 * just return, we've already done SETCLIENTID to that server
2472 	 */
2473 	if (np->s_flags & N4S_CLIENTID_SET) {
2474 		/* add mi to np's mntinfo4_list */
2475 		nfs4_add_mi_to_server(np, mi);
2476 		if (!recovery)
2477 			nfs_rw_exit(&mi->mi_recovlock);
2478 		mutex_exit(&np->s_lock);
2479 		nfs4_server_rele(np);
2480 		return;
2481 	}
2482 	mutex_exit(&np->s_lock);
2483 
2484 
2485 	/*
2486 	 * Drop the mi_recovlock since nfs4_start_op will
2487 	 * acquire it again for us.
2488 	 */
2489 	if (!recovery) {
2490 		nfs_rw_exit(&mi->mi_recovlock);
2491 
2492 		n4ep->error = nfs4_start_op(mi, NULL, NULL, &recov_state);
2493 		if (n4ep->error) {
2494 			nfs4_server_rele(np);
2495 			return;
2496 		}
2497 	}
2498 
2499 	mutex_enter(&np->s_lock);
2500 	while (np->s_flags & N4S_CLIENTID_PEND) {
2501 		if (!cv_wait_sig(&np->s_clientid_pend, &np->s_lock)) {
2502 			mutex_exit(&np->s_lock);
2503 			nfs4_server_rele(np);
2504 			if (!recovery)
2505 				nfs4_end_op(mi, NULL, NULL, &recov_state,
2506 				    recovery);
2507 			n4ep->error = EINTR;
2508 			return;
2509 		}
2510 	}
2511 
2512 	if (np->s_flags & N4S_CLIENTID_SET) {
2513 		/* XXX copied/pasted from above */
2514 		/* add mi to np's mntinfo4_list */
2515 		nfs4_add_mi_to_server(np, mi);
2516 		mutex_exit(&np->s_lock);
2517 		nfs4_server_rele(np);
2518 		if (!recovery)
2519 			nfs4_end_op(mi, NULL, NULL, &recov_state, recovery);
2520 		return;
2521 	}
2522 
2523 	/*
2524 	 * Reset the N4S_CB_PINGED flag. This is used to
2525 	 * indicate if we have received a CB_NULL from the
2526 	 * server. Also we reset the waiter flag.
2527 	 */
2528 	np->s_flags &= ~(N4S_CB_PINGED | N4S_CB_WAITER);
2529 	/* any failure must now clear this flag */
2530 	np->s_flags |= N4S_CLIENTID_PEND;
2531 	mutex_exit(&np->s_lock);
2532 	nfs4setclientid_otw(mi, svp, cr, np, n4ep, &retry_inuse);
2533 
2534 	if (n4ep->error == EACCES) {
2535 		/*
2536 		 * If the uid is set then set the creds for secure mounts
2537 		 * by proxy processes such as automountd.
2538 		 */
2539 		(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
2540 		if (svp->sv_secdata->uid != 0) {
2541 			lcr = crdup(cr);
2542 			(void) crsetugid(lcr, svp->sv_secdata->uid,
2543 			    crgetgid(cr));
2544 		}
2545 		nfs_rw_exit(&svp->sv_lock);
2546 
2547 		if (lcr != NULL) {
2548 			mutex_enter(&np->s_lock);
2549 			crfree(np->s_cred);
2550 			np->s_cred = lcr;
2551 			mutex_exit(&np->s_lock);
2552 			nfs4setclientid_otw(mi, svp, lcr, np, n4ep,
2553 				&retry_inuse);
2554 		}
2555 	}
2556 	mutex_enter(&np->s_lock);
2557 	lease_time = np->s_lease_time;
2558 	np->s_flags &= ~N4S_CLIENTID_PEND;
2559 	mutex_exit(&np->s_lock);
2560 
2561 	if (n4ep->error != 0 || n4ep->stat != NFS4_OK) {
2562 		/*
2563 		 * Start recovery if failover is a possibility.  If
2564 		 * invoked by the recovery thread itself, then just
2565 		 * return and let it handle the failover first.  NB:
2566 		 * recovery is not allowed if the mount is in progress
2567 		 * since the infrastructure is not sufficiently setup
2568 		 * to allow it.  Just return the error (after suitable
2569 		 * retries).
2570 		 */
2571 		if (FAILOVER_MOUNT4(mi) && nfs4_try_failover(n4ep)) {
2572 			(void) nfs4_start_recovery(n4ep, mi, NULL,
2573 				    NULL, NULL, NULL, OP_SETCLIENTID, NULL);
2574 			/*
2575 			 * Don't retry here, just return and let
2576 			 * recovery take over.
2577 			 */
2578 			if (recovery)
2579 				retry = FALSE;
2580 		} else if (nfs4_rpc_retry_error(n4ep->error) ||
2581 			    n4ep->stat == NFS4ERR_RESOURCE ||
2582 			    n4ep->stat == NFS4ERR_STALE_CLIENTID) {
2583 
2584 				retry = TRUE;
2585 				/*
2586 				 * Always retry if in recovery or once had
2587 				 * contact with the server (but now it's
2588 				 * overloaded).
2589 				 */
2590 				if (recovery == TRUE ||
2591 				    n4ep->error == ETIMEDOUT ||
2592 				    n4ep->error == ECONNRESET)
2593 					num_retries = 0;
2594 		} else if (retry_inuse && n4ep->error == 0 &&
2595 			    n4ep->stat == NFS4ERR_CLID_INUSE) {
2596 				retry = TRUE;
2597 				num_retries = 0;
2598 		}
2599 	} else {
2600 		/*
2601 		 * Since everything succeeded give the list a reference count if
2602 		 * it hasn't been given one by add_new_nfs4_server() or if this
2603 		 * is not a recovery situation in which case it is already on
2604 		 * the list.
2605 		 */
2606 		mutex_enter(&np->s_lock);
2607 		if ((np->s_flags & N4S_INSERTED) == 0) {
2608 			np->s_refcnt++;
2609 			np->s_flags |= N4S_INSERTED;
2610 		}
2611 		mutex_exit(&np->s_lock);
2612 	}
2613 
2614 	if (!recovery)
2615 		nfs4_end_op(mi, NULL, NULL, &recov_state, recovery);
2616 
2617 
2618 	if (retry && num_retries++ < nfs4_num_sclid_retries) {
2619 		if (retry_inuse) {
2620 			delay(SEC_TO_TICK(lease_time + nfs4_retry_sclid_delay));
2621 			retry_inuse = 0;
2622 		} else
2623 			delay(SEC_TO_TICK(nfs4_retry_sclid_delay));
2624 
2625 		nfs4_server_rele(np);
2626 		goto recov_retry;
2627 	}
2628 
2629 
2630 	if (n4ep->error == 0)
2631 		n4ep->error = geterrno4(n4ep->stat);
2632 
2633 	/* broadcast before release in case no other threads are waiting */
2634 	cv_broadcast(&np->s_clientid_pend);
2635 	nfs4_server_rele(np);
2636 }
2637 
2638 int nfs4setclientid_otw_debug = 0;
2639 
2640 /*
2641  * This function handles the recovery of STALE_CLIENTID for SETCLIENTID_CONFRIM,
2642  * but nothing else; the calling function must be designed to handle those
2643  * other errors.
2644  */
2645 static void
2646 nfs4setclientid_otw(mntinfo4_t *mi, struct servinfo4 *svp,  cred_t *cr,
2647 	struct nfs4_server *np, nfs4_error_t *ep, int *retry_inusep)
2648 {
2649 	COMPOUND4args_clnt args;
2650 	COMPOUND4res_clnt res;
2651 	nfs_argop4 argop[3];
2652 	SETCLIENTID4args *s_args;
2653 	SETCLIENTID4resok *s_resok;
2654 	int doqueue = 1;
2655 	nfs4_ga_res_t *garp = NULL;
2656 	timespec_t prop_time, after_time;
2657 	verifier4 verf;
2658 	clientid4 tmp_clientid;
2659 
2660 	ASSERT(!MUTEX_HELD(&np->s_lock));
2661 
2662 	args.ctag = TAG_SETCLIENTID;
2663 
2664 	args.array = argop;
2665 	args.array_len = 3;
2666 
2667 	/* PUTROOTFH */
2668 	argop[0].argop = OP_PUTROOTFH;
2669 
2670 	/* GETATTR */
2671 	argop[1].argop = OP_GETATTR;
2672 	argop[1].nfs_argop4_u.opgetattr.attr_request = FATTR4_LEASE_TIME_MASK;
2673 	argop[1].nfs_argop4_u.opgetattr.mi = mi;
2674 
2675 	/* SETCLIENTID */
2676 	argop[2].argop = OP_SETCLIENTID;
2677 
2678 	s_args = &argop[2].nfs_argop4_u.opsetclientid;
2679 
2680 	mutex_enter(&np->s_lock);
2681 
2682 	s_args->client.verifier = np->clidtosend.verifier;
2683 	s_args->client.id_len = np->clidtosend.id_len;
2684 	ASSERT(s_args->client.id_len <= NFS4_OPAQUE_LIMIT);
2685 	s_args->client.id_val = np->clidtosend.id_val;
2686 
2687 	/*
2688 	 * Callback needs to happen on non-RDMA transport
2689 	 * Check if we have saved the original knetconfig
2690 	 * if so, use that instead.
2691 	 */
2692 	if (svp->sv_origknconf != NULL)
2693 		nfs4_cb_args(np, svp->sv_origknconf, s_args);
2694 	else
2695 		nfs4_cb_args(np, svp->sv_knconf, s_args);
2696 
2697 	mutex_exit(&np->s_lock);
2698 
2699 	rfs4call(mi, &args, &res, cr, &doqueue, 0, ep);
2700 
2701 	if (ep->error)
2702 		return;
2703 
2704 	/* getattr lease_time res */
2705 	if (res.array_len >= 2) {
2706 		garp = &res.array[1].nfs_resop4_u.opgetattr.ga_res;
2707 
2708 #ifndef _LP64
2709 		/*
2710 		 * The 32 bit client cannot handle a lease time greater than
2711 		 * (INT32_MAX/1000000).  This is due to the use of the
2712 		 * lease_time in calls to drv_usectohz() in
2713 		 * nfs4_renew_lease_thread().  The problem is that
2714 		 * drv_usectohz() takes a time_t (which is just a long = 4
2715 		 * bytes) as its parameter.  The lease_time is multiplied by
2716 		 * 1000000 to convert seconds to usecs for the parameter.  If
2717 		 * a number bigger than (INT32_MAX/1000000) is used then we
2718 		 * overflow on the 32bit client.
2719 		 */
2720 		if (garp->n4g_ext_res->n4g_leasetime > (INT32_MAX/1000000)) {
2721 			garp->n4g_ext_res->n4g_leasetime = INT32_MAX/1000000;
2722 		}
2723 #endif
2724 
2725 		mutex_enter(&np->s_lock);
2726 		np->s_lease_time = garp->n4g_ext_res->n4g_leasetime;
2727 
2728 		/*
2729 		 * Keep track of the lease period for the mi's
2730 		 * mi_msg_list.  We need an appropiate time
2731 		 * bound to associate past facts with a current
2732 		 * event.  The lease period is perfect for this.
2733 		 */
2734 		mutex_enter(&mi->mi_msg_list_lock);
2735 		mi->mi_lease_period = np->s_lease_time;
2736 		mutex_exit(&mi->mi_msg_list_lock);
2737 		mutex_exit(&np->s_lock);
2738 	}
2739 
2740 
2741 	if (res.status == NFS4ERR_CLID_INUSE) {
2742 		clientaddr4 *clid_inuse;
2743 
2744 		if (!(*retry_inusep)) {
2745 			clid_inuse = &res.array->nfs_resop4_u.
2746 				opsetclientid.SETCLIENTID4res_u.client_using;
2747 
2748 			zcmn_err(mi->mi_zone->zone_id, CE_NOTE,
2749 			    "NFS4 mount (SETCLIENTID failed)."
2750 			    "  nfs4_client_id.id is in"
2751 			    "use already by: r_netid<%s> r_addr<%s>",
2752 			    clid_inuse->r_netid, clid_inuse->r_addr);
2753 		}
2754 
2755 		/*
2756 		 * XXX - The client should be more robust in its
2757 		 * handling of clientid in use errors (regen another
2758 		 * clientid and try again?)
2759 		 */
2760 		(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2761 		return;
2762 	}
2763 
2764 	if (res.status) {
2765 		(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2766 		return;
2767 	}
2768 
2769 	s_resok = &res.array[2].nfs_resop4_u.
2770 		opsetclientid.SETCLIENTID4res_u.resok4;
2771 
2772 	tmp_clientid = s_resok->clientid;
2773 
2774 	verf = s_resok->setclientid_confirm;
2775 
2776 #ifdef	DEBUG
2777 	if (nfs4setclientid_otw_debug) {
2778 		union {
2779 			clientid4	clientid;
2780 			int		foo[2];
2781 		} cid;
2782 
2783 		cid.clientid = s_resok->clientid;
2784 
2785 		zcmn_err(mi->mi_zone->zone_id, CE_NOTE,
2786 		"nfs4setclientid_otw: OK, clientid = %x,%x, "
2787 		"verifier = %" PRIx64 "\n", cid.foo[0], cid.foo[1], verf);
2788 	}
2789 #endif
2790 
2791 	(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2792 
2793 	/* Confirm the client id and get the lease_time attribute */
2794 
2795 	args.ctag = TAG_SETCLIENTID_CF;
2796 
2797 	args.array = argop;
2798 	args.array_len = 1;
2799 
2800 	argop[0].argop = OP_SETCLIENTID_CONFIRM;
2801 
2802 	argop[0].nfs_argop4_u.opsetclientid_confirm.clientid = tmp_clientid;
2803 	argop[0].nfs_argop4_u.opsetclientid_confirm.setclientid_confirm = verf;
2804 
2805 	/* used to figure out RTT for np */
2806 	gethrestime(&prop_time);
2807 
2808 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4setlientid_otw: "
2809 		"start time: %ld sec %ld nsec", prop_time.tv_sec,
2810 		prop_time.tv_nsec));
2811 
2812 	rfs4call(mi, &args, &res, cr, &doqueue, 0, ep);
2813 
2814 	gethrestime(&after_time);
2815 	mutex_enter(&np->s_lock);
2816 	np->propagation_delay.tv_sec =
2817 		MAX(1, after_time.tv_sec - prop_time.tv_sec);
2818 	mutex_exit(&np->s_lock);
2819 
2820 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4setlcientid_otw: "
2821 		"finish time: %ld sec ", after_time.tv_sec));
2822 
2823 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4setclientid_otw: "
2824 		"propagation delay set to %ld sec",
2825 		np->propagation_delay.tv_sec));
2826 
2827 	if (ep->error)
2828 		return;
2829 
2830 	if (res.status == NFS4ERR_CLID_INUSE) {
2831 		clientaddr4 *clid_inuse;
2832 
2833 		if (!(*retry_inusep)) {
2834 			clid_inuse = &res.array->nfs_resop4_u.
2835 				opsetclientid.SETCLIENTID4res_u.client_using;
2836 
2837 			zcmn_err(mi->mi_zone->zone_id, CE_NOTE,
2838 			    "SETCLIENTID_CONFIRM failed.  "
2839 			    "nfs4_client_id.id is in use already by: "
2840 			    "r_netid<%s> r_addr<%s>",
2841 			    clid_inuse->r_netid, clid_inuse->r_addr);
2842 		}
2843 
2844 		(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2845 		return;
2846 	}
2847 
2848 	if (res.status) {
2849 		(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2850 		return;
2851 	}
2852 
2853 	mutex_enter(&np->s_lock);
2854 	np->clientid = tmp_clientid;
2855 	np->s_flags |= N4S_CLIENTID_SET;
2856 
2857 	/* Add mi to np's mntinfo4 list */
2858 	nfs4_add_mi_to_server(np, mi);
2859 
2860 	if (np->lease_valid == NFS4_LEASE_NOT_STARTED) {
2861 		/*
2862 		 * Start lease management thread.
2863 		 * Keep trying until we succeed.
2864 		 */
2865 
2866 		np->s_refcnt++;		/* pass reference to thread */
2867 		(void) zthread_create(NULL, 0, nfs4_renew_lease_thread, np, 0,
2868 				    minclsyspri);
2869 	}
2870 	mutex_exit(&np->s_lock);
2871 
2872 	(void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2873 }
2874 
2875 /*
2876  * Add mi to sp's mntinfo4_list if it isn't already in the list.  Makes
2877  * mi's clientid the same as sp's.
2878  * Assumes sp is locked down.
2879  */
2880 void
2881 nfs4_add_mi_to_server(nfs4_server_t *sp, mntinfo4_t *mi)
2882 {
2883 	mntinfo4_t *tmi;
2884 	int in_list = 0;
2885 
2886 	ASSERT(nfs_rw_lock_held(&mi->mi_recovlock, RW_READER) ||
2887 	    nfs_rw_lock_held(&mi->mi_recovlock, RW_WRITER));
2888 	ASSERT(sp != &nfs4_server_lst);
2889 	ASSERT(MUTEX_HELD(&sp->s_lock));
2890 
2891 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
2892 		"nfs4_add_mi_to_server: add mi %p to sp %p",
2893 		    (void*)mi, (void*)sp));
2894 
2895 	for (tmi = sp->mntinfo4_list;
2896 	    tmi != NULL;
2897 	    tmi = tmi->mi_clientid_next) {
2898 		if (tmi == mi) {
2899 			NFS4_DEBUG(nfs4_client_lease_debug,
2900 				(CE_NOTE,
2901 				"nfs4_add_mi_to_server: mi in list"));
2902 			in_list = 1;
2903 		}
2904 	}
2905 
2906 	/*
2907 	 * First put a hold on the mntinfo4's vfsp so that references via
2908 	 * mntinfo4_list will be valid.
2909 	 */
2910 	if (!in_list)
2911 		VFS_HOLD(mi->mi_vfsp);
2912 
2913 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4_add_mi_to_server: "
2914 		"hold vfs %p for mi: %p", (void*)mi->mi_vfsp, (void*)mi));
2915 
2916 	if (!in_list) {
2917 		if (sp->mntinfo4_list)
2918 			sp->mntinfo4_list->mi_clientid_prev = mi;
2919 		mi->mi_clientid_next = sp->mntinfo4_list;
2920 		sp->mntinfo4_list = mi;
2921 		mi->mi_srvsettime = gethrestime_sec();
2922 	}
2923 
2924 	/* set mi's clientid to that of sp's for later matching */
2925 	mi->mi_clientid = sp->clientid;
2926 
2927 	/*
2928 	 * Update the clientid for any other mi's belonging to sp.  This
2929 	 * must be done here while we hold sp->s_lock, so that
2930 	 * find_nfs4_server() continues to work.
2931 	 */
2932 
2933 	for (tmi = sp->mntinfo4_list;
2934 	    tmi != NULL;
2935 	    tmi = tmi->mi_clientid_next) {
2936 		if (tmi != mi) {
2937 			tmi->mi_clientid = sp->clientid;
2938 		}
2939 	}
2940 }
2941 
2942 /*
2943  * Remove the mi from sp's mntinfo4_list and release its reference.
2944  * Exception: if mi still has open files, flag it for later removal (when
2945  * all the files are closed).
2946  *
2947  * If this is the last mntinfo4 in sp's list then tell the lease renewal
2948  * thread to exit.
2949  */
2950 static void
2951 nfs4_remove_mi_from_server_nolock(mntinfo4_t *mi, nfs4_server_t *sp)
2952 {
2953 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
2954 		"nfs4_remove_mi_from_server_nolock: remove mi %p from sp %p",
2955 		(void*)mi, (void*)sp));
2956 
2957 	ASSERT(sp != NULL);
2958 	ASSERT(MUTEX_HELD(&sp->s_lock));
2959 	ASSERT(mi->mi_open_files >= 0);
2960 
2961 	/*
2962 	 * First make sure this mntinfo4 can be taken off of the list,
2963 	 * ie: it doesn't have any open files remaining.
2964 	 */
2965 	if (mi->mi_open_files > 0) {
2966 		NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
2967 			"nfs4_remove_mi_from_server_nolock: don't "
2968 			"remove mi since it still has files open"));
2969 
2970 		mutex_enter(&mi->mi_lock);
2971 		mi->mi_flags |= MI4_REMOVE_ON_LAST_CLOSE;
2972 		mutex_exit(&mi->mi_lock);
2973 		return;
2974 	}
2975 
2976 	VFS_HOLD(mi->mi_vfsp);
2977 	remove_mi(sp, mi);
2978 	VFS_RELE(mi->mi_vfsp);
2979 
2980 	if (sp->mntinfo4_list == NULL) {
2981 		/* last fs unmounted, kill the thread */
2982 		NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
2983 			"remove_mi_from_nfs4_server_nolock: kill the thread"));
2984 		nfs4_mark_srv_dead(sp);
2985 	}
2986 }
2987 
2988 /*
2989  * Remove mi from sp's mntinfo4_list and release the vfs reference.
2990  */
2991 static void
2992 remove_mi(nfs4_server_t *sp, mntinfo4_t *mi)
2993 {
2994 	ASSERT(MUTEX_HELD(&sp->s_lock));
2995 
2996 	/*
2997 	 * We release a reference, and the caller must still have a
2998 	 * reference.
2999 	 */
3000 	ASSERT(mi->mi_vfsp->vfs_count >= 2);
3001 
3002 	if (mi->mi_clientid_prev) {
3003 		mi->mi_clientid_prev->mi_clientid_next = mi->mi_clientid_next;
3004 	} else {
3005 		/* This is the first mi in sp's mntinfo4_list */
3006 		/*
3007 		 * Make sure the first mntinfo4 in the list is the actual
3008 		 * mntinfo4 passed in.
3009 		 */
3010 		ASSERT(sp->mntinfo4_list == mi);
3011 
3012 		sp->mntinfo4_list = mi->mi_clientid_next;
3013 	}
3014 	if (mi->mi_clientid_next)
3015 		mi->mi_clientid_next->mi_clientid_prev = mi->mi_clientid_prev;
3016 
3017 	/* Now mark the mntinfo4's links as being removed */
3018 	mi->mi_clientid_prev = mi->mi_clientid_next = NULL;
3019 
3020 	VFS_RELE(mi->mi_vfsp);
3021 }
3022 
3023 /*
3024  * Free all the entries in sp's mntinfo4_list.
3025  */
3026 static void
3027 remove_all_mi(nfs4_server_t *sp)
3028 {
3029 	mntinfo4_t *mi;
3030 
3031 	ASSERT(MUTEX_HELD(&sp->s_lock));
3032 
3033 	while (sp->mntinfo4_list != NULL) {
3034 		mi = sp->mntinfo4_list;
3035 		/*
3036 		 * Grab a reference in case there is only one left (which
3037 		 * remove_mi() frees).
3038 		 */
3039 		VFS_HOLD(mi->mi_vfsp);
3040 		remove_mi(sp, mi);
3041 		VFS_RELE(mi->mi_vfsp);
3042 	}
3043 }
3044 
3045 /*
3046  * Remove the mi from sp's mntinfo4_list as above, and rele the vfs.
3047  *
3048  * This version can be called with a null nfs4_server_t arg,
3049  * and will either find the right one and handle locking, or
3050  * do nothing because the mi wasn't added to an sp's mntinfo4_list.
3051  */
3052 void
3053 nfs4_remove_mi_from_server(mntinfo4_t *mi, nfs4_server_t *esp)
3054 {
3055 	nfs4_server_t	*sp;
3056 
3057 	if (esp == NULL) {
3058 		(void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_READER, 0);
3059 		sp = find_nfs4_server_all(mi, 1);
3060 	} else
3061 		sp = esp;
3062 
3063 	if (sp != NULL)
3064 		nfs4_remove_mi_from_server_nolock(mi, sp);
3065 
3066 	/*
3067 	 * If we had a valid esp as input, the calling function will be
3068 	 * responsible for unlocking the esp nfs4_server.
3069 	 */
3070 	if (esp == NULL) {
3071 		if (sp != NULL)
3072 			mutex_exit(&sp->s_lock);
3073 		nfs_rw_exit(&mi->mi_recovlock);
3074 		if (sp != NULL)
3075 			nfs4_server_rele(sp);
3076 	}
3077 }
3078 
3079 /*
3080  * Return TRUE if the given server has any non-unmounted filesystems.
3081  */
3082 
3083 bool_t
3084 nfs4_fs_active(nfs4_server_t *sp)
3085 {
3086 	mntinfo4_t *mi;
3087 
3088 	ASSERT(MUTEX_HELD(&sp->s_lock));
3089 
3090 	for (mi = sp->mntinfo4_list; mi != NULL; mi = mi->mi_clientid_next) {
3091 		if (!(mi->mi_vfsp->vfs_flag & VFS_UNMOUNTED))
3092 			return (TRUE);
3093 	}
3094 
3095 	return (FALSE);
3096 }
3097 
3098 /*
3099  * Mark sp as finished and notify any waiters.
3100  */
3101 
3102 void
3103 nfs4_mark_srv_dead(nfs4_server_t *sp)
3104 {
3105 	ASSERT(MUTEX_HELD(&sp->s_lock));
3106 
3107 	sp->s_thread_exit = NFS4_THREAD_EXIT;
3108 	cv_broadcast(&sp->cv_thread_exit);
3109 }
3110 
3111 /*
3112  * Create a new nfs4_server_t structure.
3113  * Returns new node unlocked and not in list, but with a reference count of
3114  * 1.
3115  */
3116 struct nfs4_server *
3117 new_nfs4_server(struct servinfo4 *svp, cred_t *cr)
3118 {
3119 	struct nfs4_server *np;
3120 	timespec_t tt;
3121 	union {
3122 		struct {
3123 			uint32_t sec;
3124 			uint32_t subsec;
3125 		} un_curtime;
3126 		verifier4	un_verifier;
3127 	} nfs4clientid_verifier;
3128 	char id_val[] = "Solaris: %s, NFSv4 kernel client";
3129 	int len;
3130 
3131 	np = kmem_zalloc(sizeof (struct nfs4_server), KM_SLEEP);
3132 	np->saddr.len = svp->sv_addr.len;
3133 	np->saddr.maxlen = svp->sv_addr.maxlen;
3134 	np->saddr.buf = kmem_alloc(svp->sv_addr.maxlen, KM_SLEEP);
3135 	bcopy(svp->sv_addr.buf, np->saddr.buf, svp->sv_addr.len);
3136 	np->s_refcnt = 1;
3137 
3138 	/*
3139 	 * Build the nfs_client_id4 for this server mount.  Ensure
3140 	 * the verifier is useful and that the identification is
3141 	 * somehow based on the server's address for the case of
3142 	 * multi-homed servers.
3143 	 */
3144 	nfs4clientid_verifier.un_verifier = 0;
3145 	gethrestime(&tt);
3146 	nfs4clientid_verifier.un_curtime.sec = (uint32_t)tt.tv_sec;
3147 	nfs4clientid_verifier.un_curtime.subsec = (uint32_t)tt.tv_nsec;
3148 	np->clidtosend.verifier = nfs4clientid_verifier.un_verifier;
3149 
3150 	/*
3151 	 * calculate the length of the opaque identifier.  Subtract 2
3152 	 * for the "%s" and add the traditional +1 for null
3153 	 * termination.
3154 	 */
3155 	len = strlen(id_val) - 2 + strlen(uts_nodename()) + 1;
3156 	np->clidtosend.id_len = len + np->saddr.maxlen;
3157 
3158 	np->clidtosend.id_val = kmem_alloc(np->clidtosend.id_len, KM_SLEEP);
3159 	(void) sprintf(np->clidtosend.id_val, id_val, uts_nodename());
3160 	bcopy(np->saddr.buf, &np->clidtosend.id_val[len], np->saddr.len);
3161 
3162 	np->s_flags = 0;
3163 	np->mntinfo4_list = NULL;
3164 	/* save cred for issuing rfs4calls inside the renew thread */
3165 	crhold(cr);
3166 	np->s_cred = cr;
3167 	cv_init(&np->cv_thread_exit, NULL, CV_DEFAULT, NULL);
3168 	mutex_init(&np->s_lock, NULL, MUTEX_DEFAULT, NULL);
3169 	nfs_rw_init(&np->s_recovlock, NULL, RW_DEFAULT, NULL);
3170 	list_create(&np->s_deleg_list, sizeof (rnode4_t),
3171 	    offsetof(rnode4_t, r_deleg_link));
3172 	np->s_thread_exit = 0;
3173 	np->state_ref_count = 0;
3174 	np->lease_valid = NFS4_LEASE_NOT_STARTED;
3175 	cv_init(&np->s_cv_otw_count, NULL, CV_DEFAULT, NULL);
3176 	cv_init(&np->s_clientid_pend, NULL, CV_DEFAULT, NULL);
3177 	np->s_otw_call_count = 0;
3178 	cv_init(&np->wait_cb_null, NULL, CV_DEFAULT, NULL);
3179 	np->zoneid = getzoneid();
3180 	np->zone_globals = nfs4_get_callback_globals();
3181 	ASSERT(np->zone_globals != NULL);
3182 	return (np);
3183 }
3184 
3185 /*
3186  * Create a new nfs4_server_t structure and add it to the list.
3187  * Returns new node locked; reference must eventually be freed.
3188  */
3189 static struct nfs4_server *
3190 add_new_nfs4_server(struct servinfo4 *svp, cred_t *cr)
3191 {
3192 	nfs4_server_t *sp;
3193 
3194 	ASSERT(MUTEX_HELD(&nfs4_server_lst_lock));
3195 	sp = new_nfs4_server(svp, cr);
3196 	mutex_enter(&sp->s_lock);
3197 	insque(sp, &nfs4_server_lst);
3198 	sp->s_refcnt++;			/* list gets a reference */
3199 	sp->s_flags |= N4S_INSERTED;
3200 	sp->clientid = 0;
3201 	return (sp);
3202 }
3203 
3204 int nfs4_server_t_debug = 0;
3205 
3206 #ifdef lint
3207 extern void
3208 dumpnfs4slist(char *, mntinfo4_t *, clientid4, servinfo4_t *);
3209 #endif
3210 
3211 #ifndef lint
3212 #ifdef DEBUG
3213 void
3214 dumpnfs4slist(char *txt, mntinfo4_t *mi, clientid4 clientid, servinfo4_t *srv_p)
3215 {
3216 	int hash16(void *p, int len);
3217 	nfs4_server_t *np;
3218 
3219 	NFS4_DEBUG(nfs4_server_t_debug, (CE_NOTE,
3220 	    "dumping nfs4_server_t list in %s", txt));
3221 	NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT,
3222 	    "mi 0x%p, want clientid %llx, addr %d/%04X",
3223 	    mi, (longlong_t)clientid, srv_p->sv_addr.len,
3224 	    hash16((void *)srv_p->sv_addr.buf, srv_p->sv_addr.len)));
3225 	for (np = nfs4_server_lst.forw; np != &nfs4_server_lst;
3226 	    np = np->forw) {
3227 		NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT,
3228 		    "node 0x%p,    clientid %llx, addr %d/%04X, cnt %d",
3229 		    np, (longlong_t)np->clientid, np->saddr.len,
3230 		    hash16((void *)np->saddr.buf, np->saddr.len),
3231 		    np->state_ref_count));
3232 		if (np->saddr.len == srv_p->sv_addr.len &&
3233 		    bcmp(np->saddr.buf, srv_p->sv_addr.buf,
3234 		    np->saddr.len) == 0)
3235 			NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT,
3236 			    " - address matches"));
3237 		if (np->clientid == clientid || np->clientid == 0)
3238 			NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT,
3239 			    " - clientid matches"));
3240 		if (np->s_thread_exit != NFS4_THREAD_EXIT)
3241 			NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT,
3242 			    " - thread not exiting"));
3243 	}
3244 	delay(hz);
3245 }
3246 #endif
3247 #endif
3248 
3249 
3250 /*
3251  * Move a mntinfo4_t from one server list to another.
3252  * Locking of the two nfs4_server_t nodes will be done in list order.
3253  *
3254  * Returns NULL if the current nfs4_server_t for the filesystem could not
3255  * be found (e.g., due to forced unmount).  Otherwise returns a reference
3256  * to the new nfs4_server_t, which must eventually be freed.
3257  */
3258 nfs4_server_t *
3259 nfs4_move_mi(mntinfo4_t *mi, servinfo4_t *old, servinfo4_t *new)
3260 {
3261 	nfs4_server_t *p, *op = NULL, *np = NULL;
3262 	int num_open;
3263 	zoneid_t zoneid = nfs_zoneid();
3264 
3265 	ASSERT(nfs_zone() == mi->mi_zone);
3266 
3267 	mutex_enter(&nfs4_server_lst_lock);
3268 #ifdef DEBUG
3269 	if (nfs4_server_t_debug)
3270 		dumpnfs4slist("nfs4_move_mi", mi, (clientid4)0, new);
3271 #endif
3272 	for (p = nfs4_server_lst.forw; p != &nfs4_server_lst; p = p->forw) {
3273 		if (p->zoneid != zoneid)
3274 			continue;
3275 		if (p->saddr.len == old->sv_addr.len &&
3276 		    bcmp(p->saddr.buf, old->sv_addr.buf, p->saddr.len) == 0 &&
3277 		    p->s_thread_exit != NFS4_THREAD_EXIT) {
3278 			op = p;
3279 			mutex_enter(&op->s_lock);
3280 			op->s_refcnt++;
3281 		}
3282 		if (p->saddr.len == new->sv_addr.len &&
3283 		    bcmp(p->saddr.buf, new->sv_addr.buf, p->saddr.len) == 0 &&
3284 		    p->s_thread_exit != NFS4_THREAD_EXIT) {
3285 			np = p;
3286 			mutex_enter(&np->s_lock);
3287 		}
3288 		if (op != NULL && np != NULL)
3289 			break;
3290 	}
3291 	if (op == NULL) {
3292 		/*
3293 		 * Filesystem has been forcibly unmounted.  Bail out.
3294 		 */
3295 		if (np != NULL)
3296 			mutex_exit(&np->s_lock);
3297 		mutex_exit(&nfs4_server_lst_lock);
3298 		return (NULL);
3299 	}
3300 	if (np != NULL) {
3301 		np->s_refcnt++;
3302 	} else {
3303 #ifdef DEBUG
3304 		NFS4_DEBUG(nfs4_client_failover_debug, (CE_NOTE,
3305 		    "nfs4_move_mi: no target nfs4_server, will create."));
3306 #endif
3307 		np = add_new_nfs4_server(new, kcred);
3308 	}
3309 	mutex_exit(&nfs4_server_lst_lock);
3310 
3311 	NFS4_DEBUG(nfs4_client_failover_debug, (CE_NOTE,
3312 	    "nfs4_move_mi: for mi 0x%p, "
3313 	    "old servinfo4 0x%p, new servinfo4 0x%p, "
3314 	    "old nfs4_server 0x%p, new nfs4_server 0x%p, ",
3315 	    (void*)mi, (void*)old, (void*)new,
3316 	    (void*)op, (void*)np));
3317 	ASSERT(op != NULL && np != NULL);
3318 
3319 	/* discard any delegations */
3320 	nfs4_deleg_discard(mi, op);
3321 
3322 	num_open = mi->mi_open_files;
3323 	mi->mi_open_files = 0;
3324 	op->state_ref_count -= num_open;
3325 	ASSERT(op->state_ref_count >= 0);
3326 	np->state_ref_count += num_open;
3327 	nfs4_remove_mi_from_server_nolock(mi, op);
3328 	mi->mi_open_files = num_open;
3329 	NFS4_DEBUG(nfs4_client_failover_debug, (CE_NOTE,
3330 	    "nfs4_move_mi: mi_open_files %d, op->cnt %d, np->cnt %d",
3331 	    mi->mi_open_files, op->state_ref_count, np->state_ref_count));
3332 
3333 	nfs4_add_mi_to_server(np, mi);
3334 
3335 	mutex_exit(&op->s_lock);
3336 	nfs4_server_rele(op);
3337 	mutex_exit(&np->s_lock);
3338 
3339 	return (np);
3340 }
3341 
3342 /*
3343  * Need to have the nfs4_server_lst_lock.
3344  * Search the nfs4_server list to find a match on this servinfo4
3345  * based on its address.
3346  *
3347  * Returns NULL if no match is found.  Otherwise returns a reference (which
3348  * must eventually be freed) to a locked nfs4_server.
3349  */
3350 nfs4_server_t *
3351 servinfo4_to_nfs4_server(servinfo4_t *srv_p)
3352 {
3353 	nfs4_server_t *np;
3354 	zoneid_t zoneid = nfs_zoneid();
3355 
3356 	ASSERT(MUTEX_HELD(&nfs4_server_lst_lock));
3357 	for (np = nfs4_server_lst.forw; np != &nfs4_server_lst; np = np->forw) {
3358 		if (np->zoneid == zoneid &&
3359 		    np->saddr.len == srv_p->sv_addr.len &&
3360 		    bcmp(np->saddr.buf, srv_p->sv_addr.buf,
3361 			    np->saddr.len) == 0 &&
3362 		    np->s_thread_exit != NFS4_THREAD_EXIT) {
3363 			mutex_enter(&np->s_lock);
3364 			np->s_refcnt++;
3365 			return (np);
3366 		}
3367 	}
3368 	return (NULL);
3369 }
3370 
3371 /*
3372  * Search the nfs4_server_lst to find a match based on clientid and
3373  * addr.
3374  * Locks the nfs4_server down if it is found and returns a reference that
3375  * must eventually be freed.
3376  *
3377  * Returns NULL it no match is found.  This means one of two things: either
3378  * mi is in the process of being mounted, or mi has been unmounted.
3379  *
3380  * The caller should be holding mi->mi_recovlock, and it should continue to
3381  * hold the lock until done with the returned nfs4_server_t.  Once
3382  * mi->mi_recovlock is released, there is no guarantee that the returned
3383  * mi->nfs4_server_t will continue to correspond to mi.
3384  */
3385 nfs4_server_t *
3386 find_nfs4_server(mntinfo4_t *mi)
3387 {
3388 	return (find_nfs4_server_all(mi, 0));
3389 }
3390 
3391 /*
3392  * Same as above, but takes an "all" parameter which can be
3393  * set to 1 if the caller wishes to find nfs4_server_t's which
3394  * have been marked for termination by the exit of the renew
3395  * thread.  This should only be used by operations which are
3396  * cleaning up and will not cause an OTW op.
3397  */
3398 nfs4_server_t *
3399 find_nfs4_server_all(mntinfo4_t *mi, int all)
3400 {
3401 	nfs4_server_t *np;
3402 	servinfo4_t *svp;
3403 	zoneid_t zoneid = mi->mi_zone->zone_id;
3404 
3405 	ASSERT(nfs_rw_lock_held(&mi->mi_recovlock, RW_READER) ||
3406 	    nfs_rw_lock_held(&mi->mi_recovlock, RW_WRITER));
3407 	/*
3408 	 * This can be called from nfs4_unmount() which can be called from the
3409 	 * global zone, hence it's legal for the global zone to muck with
3410 	 * another zone's server list, as long as it doesn't try to contact
3411 	 * them.
3412 	 */
3413 	ASSERT(zoneid == getzoneid() || getzoneid() == GLOBAL_ZONEID ||
3414 	    nfs_global_client_only != 0);
3415 
3416 	/*
3417 	 * The nfs4_server_lst_lock global lock is held when we get a new
3418 	 * clientid (via SETCLIENTID OTW).  Holding this global lock and
3419 	 * mi_recovlock (READER is fine) ensures that the nfs4_server
3420 	 * and this mntinfo4 can't get out of sync, so the following search is
3421 	 * always valid.
3422 	 */
3423 	mutex_enter(&nfs4_server_lst_lock);
3424 #ifdef DEBUG
3425 	if (nfs4_server_t_debug) {
3426 		/* mi->mi_clientid is unprotected, ok for debug output */
3427 		dumpnfs4slist("find_nfs4_server", mi, mi->mi_clientid,
3428 			mi->mi_curr_serv);
3429 	}
3430 #endif
3431 	for (np = nfs4_server_lst.forw; np != &nfs4_server_lst; np = np->forw) {
3432 		mutex_enter(&np->s_lock);
3433 		svp = mi->mi_curr_serv;
3434 
3435 		if (np->zoneid == zoneid &&
3436 		    np->clientid == mi->mi_clientid &&
3437 		    np->saddr.len == svp->sv_addr.len &&
3438 		    bcmp(np->saddr.buf, svp->sv_addr.buf, np->saddr.len) == 0 &&
3439 		    (np->s_thread_exit != NFS4_THREAD_EXIT || all != 0)) {
3440 			mutex_exit(&nfs4_server_lst_lock);
3441 			np->s_refcnt++;
3442 			return (np);
3443 		}
3444 		mutex_exit(&np->s_lock);
3445 	}
3446 	mutex_exit(&nfs4_server_lst_lock);
3447 
3448 	return (NULL);
3449 }
3450 
3451 /*
3452  * Release the reference to sp and destroy it if that's the last one.
3453  */
3454 
3455 void
3456 nfs4_server_rele(nfs4_server_t *sp)
3457 {
3458 	mutex_enter(&sp->s_lock);
3459 	ASSERT(sp->s_refcnt > 0);
3460 	sp->s_refcnt--;
3461 	if (sp->s_refcnt > 0) {
3462 		mutex_exit(&sp->s_lock);
3463 		return;
3464 	}
3465 	mutex_exit(&sp->s_lock);
3466 
3467 	mutex_enter(&nfs4_server_lst_lock);
3468 	mutex_enter(&sp->s_lock);
3469 	if (sp->s_refcnt > 0) {
3470 		mutex_exit(&sp->s_lock);
3471 		mutex_exit(&nfs4_server_lst_lock);
3472 		return;
3473 	}
3474 	remque(sp);
3475 	sp->forw = sp->back = NULL;
3476 	mutex_exit(&nfs4_server_lst_lock);
3477 	destroy_nfs4_server(sp);
3478 }
3479 
3480 static void
3481 destroy_nfs4_server(nfs4_server_t *sp)
3482 {
3483 	ASSERT(MUTEX_HELD(&sp->s_lock));
3484 	ASSERT(sp->s_refcnt == 0);
3485 	ASSERT(sp->s_otw_call_count == 0);
3486 
3487 	remove_all_mi(sp);
3488 
3489 	crfree(sp->s_cred);
3490 	kmem_free(sp->saddr.buf, sp->saddr.maxlen);
3491 	kmem_free(sp->clidtosend.id_val, sp->clidtosend.id_len);
3492 	mutex_exit(&sp->s_lock);
3493 
3494 	/* destroy the nfs4_server */
3495 	nfs4callback_destroy(sp);
3496 	list_destroy(&sp->s_deleg_list);
3497 	mutex_destroy(&sp->s_lock);
3498 	cv_destroy(&sp->cv_thread_exit);
3499 	cv_destroy(&sp->s_cv_otw_count);
3500 	cv_destroy(&sp->s_clientid_pend);
3501 	cv_destroy(&sp->wait_cb_null);
3502 	nfs_rw_destroy(&sp->s_recovlock);
3503 	kmem_free(sp, sizeof (*sp));
3504 }
3505 
3506 /*
3507  * Lock sp, but only if it's still active (in the list and hasn't been
3508  * flagged as exiting) or 'all' is non-zero.
3509  * Returns TRUE if sp got locked and adds a reference to sp.
3510  */
3511 bool_t
3512 nfs4_server_vlock(nfs4_server_t *sp, int all)
3513 {
3514 	nfs4_server_t *np;
3515 
3516 	mutex_enter(&nfs4_server_lst_lock);
3517 	for (np = nfs4_server_lst.forw; np != &nfs4_server_lst; np = np->forw) {
3518 		if (sp == np && (np->s_thread_exit != NFS4_THREAD_EXIT ||
3519 		    all != 0)) {
3520 			mutex_enter(&np->s_lock);
3521 			np->s_refcnt++;
3522 			mutex_exit(&nfs4_server_lst_lock);
3523 			return (TRUE);
3524 		}
3525 	}
3526 	mutex_exit(&nfs4_server_lst_lock);
3527 	return (FALSE);
3528 }
3529 
3530 /*
3531  * Fork off a thread to free the data structures for a mount.
3532  */
3533 
3534 static void
3535 async_free_mount(vfs_t *vfsp, cred_t *cr)
3536 {
3537 	freemountargs_t *args;
3538 	args = kmem_alloc(sizeof (freemountargs_t), KM_SLEEP);
3539 	args->fm_vfsp = vfsp;
3540 	VFS_HOLD(vfsp);
3541 	MI4_HOLD(VFTOMI4(vfsp));
3542 	args->fm_cr = cr;
3543 	crhold(cr);
3544 	(void) zthread_create(NULL, 0, nfs4_free_mount_thread, args, 0,
3545 	    minclsyspri);
3546 }
3547 
3548 static void
3549 nfs4_free_mount_thread(freemountargs_t *args)
3550 {
3551 	mntinfo4_t *mi;
3552 	nfs4_free_mount(args->fm_vfsp, args->fm_cr);
3553 	mi = VFTOMI4(args->fm_vfsp);
3554 	crfree(args->fm_cr);
3555 	VFS_RELE(args->fm_vfsp);
3556 	MI4_RELE(mi);
3557 	kmem_free(args, sizeof (freemountargs_t));
3558 	zthread_exit();
3559 	/* NOTREACHED */
3560 }
3561 
3562 /*
3563  * Thread to free the data structures for a given filesystem.
3564  */
3565 static void
3566 nfs4_free_mount(vfs_t *vfsp, cred_t *cr)
3567 {
3568 	mntinfo4_t	*mi = VFTOMI4(vfsp);
3569 	nfs4_server_t	*sp;
3570 	callb_cpr_t	cpr_info;
3571 	kmutex_t	cpr_lock;
3572 	boolean_t	async_thread;
3573 	int		removed;
3574 
3575 	/*
3576 	 * We need to participate in the CPR framework if this is a kernel
3577 	 * thread.
3578 	 */
3579 	async_thread = (curproc == nfs_zone()->zone_zsched);
3580 	if (async_thread) {
3581 		mutex_init(&cpr_lock, NULL, MUTEX_DEFAULT, NULL);
3582 		CALLB_CPR_INIT(&cpr_info, &cpr_lock, callb_generic_cpr,
3583 		    "nfsv4AsyncUnmount");
3584 	}
3585 
3586 	/*
3587 	 * We need to wait for all outstanding OTW calls
3588 	 * and recovery to finish before we remove the mi
3589 	 * from the nfs4_server_t, as current pending
3590 	 * calls might still need this linkage (in order
3591 	 * to find a nfs4_server_t from a mntinfo4_t).
3592 	 */
3593 	(void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_READER, FALSE);
3594 	sp = find_nfs4_server(mi);
3595 	nfs_rw_exit(&mi->mi_recovlock);
3596 
3597 	if (sp) {
3598 		while (sp->s_otw_call_count != 0) {
3599 			if (async_thread) {
3600 				mutex_enter(&cpr_lock);
3601 				CALLB_CPR_SAFE_BEGIN(&cpr_info);
3602 				mutex_exit(&cpr_lock);
3603 			}
3604 			cv_wait(&sp->s_cv_otw_count, &sp->s_lock);
3605 			if (async_thread) {
3606 				mutex_enter(&cpr_lock);
3607 				CALLB_CPR_SAFE_END(&cpr_info, &cpr_lock);
3608 				mutex_exit(&cpr_lock);
3609 			}
3610 		}
3611 		mutex_exit(&sp->s_lock);
3612 		nfs4_server_rele(sp);
3613 		sp = NULL;
3614 	}
3615 
3616 
3617 	mutex_enter(&mi->mi_lock);
3618 	while (mi->mi_in_recovery != 0) {
3619 		if (async_thread) {
3620 			mutex_enter(&cpr_lock);
3621 			CALLB_CPR_SAFE_BEGIN(&cpr_info);
3622 			mutex_exit(&cpr_lock);
3623 		}
3624 		cv_wait(&mi->mi_cv_in_recov, &mi->mi_lock);
3625 		if (async_thread) {
3626 			mutex_enter(&cpr_lock);
3627 			CALLB_CPR_SAFE_END(&cpr_info, &cpr_lock);
3628 			mutex_exit(&cpr_lock);
3629 		}
3630 	}
3631 	mutex_exit(&mi->mi_lock);
3632 
3633 	/*
3634 	 * The original purge of the dnlc via 'dounmount'
3635 	 * doesn't guarantee that another dnlc entry was not
3636 	 * added while we waitied for all outstanding OTW
3637 	 * and recovery calls to finish.  So re-purge the
3638 	 * dnlc now.
3639 	 */
3640 	(void) dnlc_purge_vfsp(vfsp, 0);
3641 
3642 	/*
3643 	 * We need to explicitly stop the manager thread; the asyc worker
3644 	 * threads can timeout and exit on their own.
3645 	 */
3646 	mutex_enter(&mi->mi_async_lock);
3647 	mi->mi_max_threads = 0;
3648 	cv_broadcast(&mi->mi_async_work_cv);
3649 	mutex_exit(&mi->mi_async_lock);
3650 	if (mi->mi_manager_thread)
3651 		nfs4_async_manager_stop(vfsp);
3652 
3653 	destroy_rtable4(vfsp, cr);
3654 
3655 	nfs4_remove_mi_from_server(mi, NULL);
3656 
3657 	if (async_thread) {
3658 		mutex_enter(&cpr_lock);
3659 		CALLB_CPR_EXIT(&cpr_info);	/* drops cpr_lock */
3660 		mutex_destroy(&cpr_lock);
3661 	}
3662 
3663 	removed = nfs4_mi_zonelist_remove(mi);
3664 	if (removed)
3665 		zone_rele(mi->mi_zone);
3666 }
3667