xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs4_vfsops.c (revision 1f0c5e612d8dfd7d1a8f4487cef753357b9909f5)
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 /*
23  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
25  */
26 
27 /*
28  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
29  *	All Rights Reserved
30  */
31 
32 #include <sys/param.h>
33 #include <sys/types.h>
34 #include <sys/systm.h>
35 #include <sys/cred.h>
36 #include <sys/vfs.h>
37 #include <sys/vfs_opreg.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 #include <sys/fs/autofs.h>
79 
80 #include <sys/sdt.h>
81 
82 
83 /*
84  * Arguments passed to thread to free data structures from forced unmount.
85  */
86 
87 typedef struct {
88 	vfs_t	*fm_vfsp;
89 	int	fm_flag;
90 	cred_t	*fm_cr;
91 } freemountargs_t;
92 
93 static void	async_free_mount(vfs_t *, int, cred_t *);
94 static void	nfs4_free_mount(vfs_t *, int, cred_t *);
95 static void	nfs4_free_mount_thread(freemountargs_t *);
96 static int nfs4_chkdup_servinfo4(servinfo4_t *, servinfo4_t *);
97 
98 /*
99  * From rpcsec module (common/rpcsec).
100  */
101 extern int sec_clnt_loadinfo(struct sec_data *, struct sec_data **, model_t);
102 extern void sec_clnt_freeinfo(struct sec_data *);
103 
104 /*
105  * The order and contents of this structure must be kept in sync with that of
106  * rfsreqcnt_v4_tmpl in nfs_stats.c
107  */
108 static char *rfsnames_v4[] = {
109 	"null", "compound", "reserved",	"access", "close", "commit", "create",
110 	"delegpurge", "delegreturn", "getattr",	"getfh", "link", "lock",
111 	"lockt", "locku", "lookup", "lookupp", "nverify", "open", "openattr",
112 	"open_confirm",	"open_downgrade", "putfh", "putpubfh", "putrootfh",
113 	"read", "readdir", "readlink", "remove", "rename", "renew",
114 	"restorefh", "savefh", "secinfo", "setattr", "setclientid",
115 	"setclientid_confirm", "verify", "write"
116 };
117 
118 /*
119  * nfs4_max_mount_retry is the number of times the client will redrive
120  * a mount compound before giving up and returning failure.  The intent
121  * is to redrive mount compounds which fail NFS4ERR_STALE so that
122  * if a component of the server path being mounted goes stale, it can
123  * "recover" by redriving the mount compund (LOOKUP ops).  This recovery
124  * code is needed outside of the recovery framework because mount is a
125  * special case.  The client doesn't create vnodes/rnodes for components
126  * of the server path being mounted.  The recovery code recovers real
127  * client objects, not STALE FHs which map to components of the server
128  * path being mounted.
129  *
130  * We could just fail the mount on the first time, but that would
131  * instantly trigger failover (from nfs4_mount), and the client should
132  * try to re-lookup the STALE FH before doing failover.  The easiest
133  * way to "re-lookup" is to simply redrive the mount compound.
134  */
135 static int nfs4_max_mount_retry = 2;
136 
137 /*
138  * nfs4 vfs operations.
139  */
140 int		nfs4_mount(vfs_t *, vnode_t *, struct mounta *, cred_t *);
141 static int	nfs4_unmount(vfs_t *, int, cred_t *);
142 static int	nfs4_root(vfs_t *, vnode_t **);
143 static int	nfs4_statvfs(vfs_t *, struct statvfs64 *);
144 static int	nfs4_sync(vfs_t *, short, cred_t *);
145 static int	nfs4_vget(vfs_t *, vnode_t **, fid_t *);
146 static int	nfs4_mountroot(vfs_t *, whymountroot_t);
147 static void	nfs4_freevfs(vfs_t *);
148 
149 static int	nfs4rootvp(vnode_t **, vfs_t *, struct servinfo4 *,
150 		    int, cred_t *, zone_t *);
151 
152 vfsops_t	*nfs4_vfsops;
153 
154 int nfs4_vfsinit(void);
155 void nfs4_vfsfini(void);
156 static void nfs4setclientid_init(void);
157 static void nfs4setclientid_fini(void);
158 static void nfs4setclientid_otw(mntinfo4_t *, servinfo4_t *,  cred_t *,
159 		struct nfs4_server *, nfs4_error_t *, int *);
160 static void	destroy_nfs4_server(nfs4_server_t *);
161 static void	remove_mi(nfs4_server_t *, mntinfo4_t *);
162 
163 extern void nfs4_ephemeral_init(void);
164 extern void nfs4_ephemeral_fini(void);
165 
166 /* referral related routines */
167 static servinfo4_t *copy_svp(servinfo4_t *);
168 static void free_knconf_contents(struct knetconfig *k);
169 static char *extract_referral_point(const char *, int);
170 static void setup_newsvpath(servinfo4_t *, int);
171 static void update_servinfo4(servinfo4_t *, fs_location4 *,
172 		struct nfs_fsl_info *, char *, int);
173 
174 /*
175  * Initialize the vfs structure
176  */
177 
178 static int nfs4fstyp;
179 
180 
181 /*
182  * Debug variable to check for rdma based
183  * transport startup and cleanup. Controlled
184  * through /etc/system. Off by default.
185  */
186 extern int rdma_debug;
187 
188 int
189 nfs4init(int fstyp, char *name)
190 {
191 	static const fs_operation_def_t nfs4_vfsops_template[] = {
192 		VFSNAME_MOUNT,		{ .vfs_mount = nfs4_mount },
193 		VFSNAME_UNMOUNT,	{ .vfs_unmount = nfs4_unmount },
194 		VFSNAME_ROOT,		{ .vfs_root = nfs4_root },
195 		VFSNAME_STATVFS,	{ .vfs_statvfs = nfs4_statvfs },
196 		VFSNAME_SYNC,		{ .vfs_sync = nfs4_sync },
197 		VFSNAME_VGET,		{ .vfs_vget = nfs4_vget },
198 		VFSNAME_MOUNTROOT,	{ .vfs_mountroot = nfs4_mountroot },
199 		VFSNAME_FREEVFS,	{ .vfs_freevfs = nfs4_freevfs },
200 		NULL,			NULL
201 	};
202 	int error;
203 
204 	nfs4_vfsops = NULL;
205 	nfs4_vnodeops = NULL;
206 	nfs4_trigger_vnodeops = NULL;
207 
208 	error = vfs_setfsops(fstyp, nfs4_vfsops_template, &nfs4_vfsops);
209 	if (error != 0) {
210 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
211 		    "nfs4init: bad vfs ops template");
212 		goto out;
213 	}
214 
215 	error = vn_make_ops(name, nfs4_vnodeops_template, &nfs4_vnodeops);
216 	if (error != 0) {
217 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
218 		    "nfs4init: bad vnode ops template");
219 		goto out;
220 	}
221 
222 	error = vn_make_ops("nfs4_trigger", nfs4_trigger_vnodeops_template,
223 	    &nfs4_trigger_vnodeops);
224 	if (error != 0) {
225 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
226 		    "nfs4init: bad trigger vnode ops template");
227 		goto out;
228 	}
229 
230 	nfs4fstyp = fstyp;
231 	(void) nfs4_vfsinit();
232 	(void) nfs4_init_dot_entries();
233 
234 out:
235 	if (error) {
236 		if (nfs4_trigger_vnodeops != NULL)
237 			vn_freevnodeops(nfs4_trigger_vnodeops);
238 
239 		if (nfs4_vnodeops != NULL)
240 			vn_freevnodeops(nfs4_vnodeops);
241 
242 		(void) vfs_freevfsops_by_type(fstyp);
243 	}
244 
245 	return (error);
246 }
247 
248 void
249 nfs4fini(void)
250 {
251 	(void) nfs4_destroy_dot_entries();
252 	nfs4_vfsfini();
253 }
254 
255 /*
256  * Create a new sec_data structure to store AUTH_DH related data:
257  * netname, syncaddr, knetconfig. There is no AUTH_F_RPCTIMESYNC
258  * flag set for NFS V4 since we are avoiding to contact the rpcbind
259  * daemon and is using the IP time service (IPPORT_TIMESERVER).
260  *
261  * sec_data can be freed by sec_clnt_freeinfo().
262  */
263 static struct sec_data *
264 create_authdh_data(char *netname, int nlen, struct netbuf *syncaddr,
265     struct knetconfig *knconf)
266 {
267 	struct sec_data *secdata;
268 	dh_k4_clntdata_t *data;
269 	char *pf, *p;
270 
271 	if (syncaddr == NULL || syncaddr->buf == NULL || nlen == 0)
272 		return (NULL);
273 
274 	secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
275 	secdata->flags = 0;
276 
277 	data = kmem_alloc(sizeof (*data), KM_SLEEP);
278 
279 	data->syncaddr.maxlen = syncaddr->maxlen;
280 	data->syncaddr.len = syncaddr->len;
281 	data->syncaddr.buf = (char *)kmem_alloc(syncaddr->len, KM_SLEEP);
282 	bcopy(syncaddr->buf, data->syncaddr.buf, syncaddr->len);
283 
284 	/*
285 	 * duplicate the knconf information for the
286 	 * new opaque data.
287 	 */
288 	data->knconf = kmem_alloc(sizeof (*knconf), KM_SLEEP);
289 	*data->knconf = *knconf;
290 	pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
291 	p = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
292 	bcopy(knconf->knc_protofmly, pf, KNC_STRSIZE);
293 	bcopy(knconf->knc_proto, p, KNC_STRSIZE);
294 	data->knconf->knc_protofmly = pf;
295 	data->knconf->knc_proto = p;
296 
297 	/* move server netname to the sec_data structure */
298 	data->netname = kmem_alloc(nlen, KM_SLEEP);
299 	bcopy(netname, data->netname, nlen);
300 	data->netnamelen = (int)nlen;
301 
302 	secdata->secmod = AUTH_DH;
303 	secdata->rpcflavor = AUTH_DH;
304 	secdata->data = (caddr_t)data;
305 
306 	return (secdata);
307 }
308 
309 /*
310  * Returns (deep) copy of sec_data_t. Allocates all memory required; caller
311  * is responsible for freeing.
312  */
313 sec_data_t *
314 copy_sec_data(sec_data_t *fsecdata)
315 {
316 	sec_data_t *tsecdata;
317 
318 	if (fsecdata == NULL)
319 		return (NULL);
320 
321 	if (fsecdata->rpcflavor == AUTH_DH) {
322 		dh_k4_clntdata_t *fdata = (dh_k4_clntdata_t *)fsecdata->data;
323 
324 		if (fdata == NULL)
325 			return (NULL);
326 
327 		tsecdata = (sec_data_t *)create_authdh_data(fdata->netname,
328 		    fdata->netnamelen, &fdata->syncaddr, fdata->knconf);
329 
330 		return (tsecdata);
331 	}
332 
333 	tsecdata = kmem_zalloc(sizeof (sec_data_t), KM_SLEEP);
334 
335 	tsecdata->secmod = fsecdata->secmod;
336 	tsecdata->rpcflavor = fsecdata->rpcflavor;
337 	tsecdata->flags = fsecdata->flags;
338 	tsecdata->uid = fsecdata->uid;
339 
340 	if (fsecdata->rpcflavor == RPCSEC_GSS) {
341 		gss_clntdata_t *gcd = (gss_clntdata_t *)fsecdata->data;
342 
343 		tsecdata->data = (caddr_t)copy_sec_data_gss(gcd);
344 	} else {
345 		tsecdata->data = NULL;
346 	}
347 
348 	return (tsecdata);
349 }
350 
351 gss_clntdata_t *
352 copy_sec_data_gss(gss_clntdata_t *fdata)
353 {
354 	gss_clntdata_t *tdata;
355 
356 	if (fdata == NULL)
357 		return (NULL);
358 
359 	tdata = kmem_zalloc(sizeof (gss_clntdata_t), KM_SLEEP);
360 
361 	tdata->mechanism.length = fdata->mechanism.length;
362 	tdata->mechanism.elements = kmem_zalloc(fdata->mechanism.length,
363 	    KM_SLEEP);
364 	bcopy(fdata->mechanism.elements, tdata->mechanism.elements,
365 	    fdata->mechanism.length);
366 
367 	tdata->service = fdata->service;
368 
369 	(void) strcpy(tdata->uname, fdata->uname);
370 	(void) strcpy(tdata->inst, fdata->inst);
371 	(void) strcpy(tdata->realm, fdata->realm);
372 
373 	tdata->qop = fdata->qop;
374 
375 	return (tdata);
376 }
377 
378 static int
379 nfs4_chkdup_servinfo4(servinfo4_t *svp_head, servinfo4_t *svp)
380 {
381 	servinfo4_t *si;
382 
383 	/*
384 	 * Iterate over the servinfo4 list to make sure
385 	 * we do not have a duplicate. Skip any servinfo4
386 	 * that has been marked "NOT IN USE"
387 	 */
388 	for (si = svp_head; si; si = si->sv_next) {
389 		(void) nfs_rw_enter_sig(&si->sv_lock, RW_READER, 0);
390 		if (si->sv_flags & SV4_NOTINUSE) {
391 			nfs_rw_exit(&si->sv_lock);
392 			continue;
393 		}
394 		nfs_rw_exit(&si->sv_lock);
395 		if (si == svp)
396 			continue;
397 		if (si->sv_addr.len == svp->sv_addr.len &&
398 		    strcmp(si->sv_knconf->knc_protofmly,
399 		    svp->sv_knconf->knc_protofmly) == 0 &&
400 		    bcmp(si->sv_addr.buf, svp->sv_addr.buf,
401 		    si->sv_addr.len) == 0) {
402 			/* it's a duplicate */
403 			return (1);
404 		}
405 	}
406 	/* it's not a duplicate */
407 	return (0);
408 }
409 
410 void
411 nfs4_free_args(struct nfs_args *nargs)
412 {
413 	if (nargs->knconf) {
414 		if (nargs->knconf->knc_protofmly)
415 			kmem_free(nargs->knconf->knc_protofmly,
416 			    KNC_STRSIZE);
417 		if (nargs->knconf->knc_proto)
418 			kmem_free(nargs->knconf->knc_proto, KNC_STRSIZE);
419 		kmem_free(nargs->knconf, sizeof (*nargs->knconf));
420 		nargs->knconf = NULL;
421 	}
422 
423 	if (nargs->fh) {
424 		kmem_free(nargs->fh, strlen(nargs->fh) + 1);
425 		nargs->fh = NULL;
426 	}
427 
428 	if (nargs->hostname) {
429 		kmem_free(nargs->hostname, strlen(nargs->hostname) + 1);
430 		nargs->hostname = NULL;
431 	}
432 
433 	if (nargs->addr) {
434 		if (nargs->addr->buf) {
435 			ASSERT(nargs->addr->len);
436 			kmem_free(nargs->addr->buf, nargs->addr->len);
437 		}
438 		kmem_free(nargs->addr, sizeof (struct netbuf));
439 		nargs->addr = NULL;
440 	}
441 
442 	if (nargs->syncaddr) {
443 		ASSERT(nargs->syncaddr->len);
444 		if (nargs->syncaddr->buf) {
445 			ASSERT(nargs->syncaddr->len);
446 			kmem_free(nargs->syncaddr->buf, nargs->syncaddr->len);
447 		}
448 		kmem_free(nargs->syncaddr, sizeof (struct netbuf));
449 		nargs->syncaddr = NULL;
450 	}
451 
452 	if (nargs->netname) {
453 		kmem_free(nargs->netname, strlen(nargs->netname) + 1);
454 		nargs->netname = NULL;
455 	}
456 
457 	if (nargs->nfs_ext_u.nfs_extA.secdata) {
458 		sec_clnt_freeinfo(
459 		    nargs->nfs_ext_u.nfs_extA.secdata);
460 		nargs->nfs_ext_u.nfs_extA.secdata = NULL;
461 	}
462 }
463 
464 
465 int
466 nfs4_copyin(char *data, int datalen, struct nfs_args *nargs)
467 {
468 
469 	int error;
470 	size_t hlen;			/* length of hostname */
471 	size_t nlen;			/* length of netname */
472 	char netname[MAXNETNAMELEN+1];	/* server's netname */
473 	struct netbuf addr;		/* server's address */
474 	struct netbuf syncaddr;		/* AUTH_DES time sync addr */
475 	struct knetconfig *knconf;		/* transport structure */
476 	struct sec_data *secdata = NULL;	/* security data */
477 	STRUCT_DECL(nfs_args, args);		/* nfs mount arguments */
478 	STRUCT_DECL(knetconfig, knconf_tmp);
479 	STRUCT_DECL(netbuf, addr_tmp);
480 	int flags;
481 	char *p, *pf;
482 	struct pathname pn;
483 	char *userbufptr;
484 
485 
486 	bzero(nargs, sizeof (*nargs));
487 
488 	STRUCT_INIT(args, get_udatamodel());
489 	bzero(STRUCT_BUF(args), SIZEOF_STRUCT(nfs_args, DATAMODEL_NATIVE));
490 	if (copyin(data, STRUCT_BUF(args), MIN(datalen,
491 	    STRUCT_SIZE(args))))
492 		return (EFAULT);
493 
494 	nargs->wsize = STRUCT_FGET(args, wsize);
495 	nargs->rsize = STRUCT_FGET(args, rsize);
496 	nargs->timeo = STRUCT_FGET(args, timeo);
497 	nargs->retrans = STRUCT_FGET(args, retrans);
498 	nargs->acregmin = STRUCT_FGET(args, acregmin);
499 	nargs->acregmax = STRUCT_FGET(args, acregmax);
500 	nargs->acdirmin = STRUCT_FGET(args, acdirmin);
501 	nargs->acdirmax = STRUCT_FGET(args, acdirmax);
502 
503 	flags = STRUCT_FGET(args, flags);
504 	nargs->flags = flags;
505 
506 	addr.buf = NULL;
507 	syncaddr.buf = NULL;
508 
509 
510 	/*
511 	 * Allocate space for a knetconfig structure and
512 	 * its strings and copy in from user-land.
513 	 */
514 	knconf = kmem_zalloc(sizeof (*knconf), KM_SLEEP);
515 	STRUCT_INIT(knconf_tmp, get_udatamodel());
516 	if (copyin(STRUCT_FGETP(args, knconf), STRUCT_BUF(knconf_tmp),
517 	    STRUCT_SIZE(knconf_tmp))) {
518 		kmem_free(knconf, sizeof (*knconf));
519 		return (EFAULT);
520 	}
521 
522 	knconf->knc_semantics = STRUCT_FGET(knconf_tmp, knc_semantics);
523 	knconf->knc_protofmly = STRUCT_FGETP(knconf_tmp, knc_protofmly);
524 	knconf->knc_proto = STRUCT_FGETP(knconf_tmp, knc_proto);
525 	if (get_udatamodel() != DATAMODEL_LP64) {
526 		knconf->knc_rdev = expldev(STRUCT_FGET(knconf_tmp, knc_rdev));
527 	} else {
528 		knconf->knc_rdev = STRUCT_FGET(knconf_tmp, knc_rdev);
529 	}
530 
531 	pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
532 	p = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
533 	error = copyinstr(knconf->knc_protofmly, pf, KNC_STRSIZE, NULL);
534 	if (error) {
535 		kmem_free(pf, KNC_STRSIZE);
536 		kmem_free(p, KNC_STRSIZE);
537 		kmem_free(knconf, sizeof (*knconf));
538 		return (error);
539 	}
540 
541 	error = copyinstr(knconf->knc_proto, p, KNC_STRSIZE, NULL);
542 	if (error) {
543 		kmem_free(pf, KNC_STRSIZE);
544 		kmem_free(p, KNC_STRSIZE);
545 		kmem_free(knconf, sizeof (*knconf));
546 		return (error);
547 	}
548 
549 
550 	knconf->knc_protofmly = pf;
551 	knconf->knc_proto = p;
552 
553 	nargs->knconf = knconf;
554 
555 	/*
556 	 * Get server address
557 	 */
558 	STRUCT_INIT(addr_tmp, get_udatamodel());
559 	if (copyin(STRUCT_FGETP(args, addr), STRUCT_BUF(addr_tmp),
560 	    STRUCT_SIZE(addr_tmp))) {
561 		error = EFAULT;
562 		goto errout;
563 	}
564 
565 	nargs->addr = kmem_zalloc(sizeof (struct netbuf), KM_SLEEP);
566 	userbufptr = STRUCT_FGETP(addr_tmp, buf);
567 	addr.len = STRUCT_FGET(addr_tmp, len);
568 	addr.buf = kmem_alloc(addr.len, KM_SLEEP);
569 	addr.maxlen = addr.len;
570 	if (copyin(userbufptr, addr.buf, addr.len)) {
571 		kmem_free(addr.buf, addr.len);
572 		error = EFAULT;
573 		goto errout;
574 	}
575 	bcopy(&addr, nargs->addr, sizeof (struct netbuf));
576 
577 	/*
578 	 * Get the root fhandle
579 	 */
580 	error = pn_get(STRUCT_FGETP(args, fh), UIO_USERSPACE, &pn);
581 	if (error)
582 		goto errout;
583 
584 	/* Volatile fh: keep server paths, so use actual-size strings */
585 	nargs->fh = kmem_alloc(pn.pn_pathlen + 1, KM_SLEEP);
586 	bcopy(pn.pn_path, nargs->fh, pn.pn_pathlen);
587 	nargs->fh[pn.pn_pathlen] = '\0';
588 	pn_free(&pn);
589 
590 
591 	/*
592 	 * Get server's hostname
593 	 */
594 	if (flags & NFSMNT_HOSTNAME) {
595 		error = copyinstr(STRUCT_FGETP(args, hostname),
596 		    netname, sizeof (netname), &hlen);
597 		if (error)
598 			goto errout;
599 		nargs->hostname = kmem_zalloc(hlen, KM_SLEEP);
600 		(void) strcpy(nargs->hostname, netname);
601 
602 	} else {
603 		nargs->hostname = NULL;
604 	}
605 
606 
607 	/*
608 	 * If there are syncaddr and netname data, load them in. This is
609 	 * to support data needed for NFSV4 when AUTH_DH is the negotiated
610 	 * flavor via SECINFO. (instead of using MOUNT protocol in V3).
611 	 */
612 	netname[0] = '\0';
613 	if (flags & NFSMNT_SECURE) {
614 
615 		/* get syncaddr */
616 		STRUCT_INIT(addr_tmp, get_udatamodel());
617 		if (copyin(STRUCT_FGETP(args, syncaddr), STRUCT_BUF(addr_tmp),
618 		    STRUCT_SIZE(addr_tmp))) {
619 			error = EINVAL;
620 			goto errout;
621 		}
622 		userbufptr = STRUCT_FGETP(addr_tmp, buf);
623 		syncaddr.len = STRUCT_FGET(addr_tmp, len);
624 		syncaddr.buf = kmem_alloc(syncaddr.len, KM_SLEEP);
625 		syncaddr.maxlen = syncaddr.len;
626 		if (copyin(userbufptr, syncaddr.buf, syncaddr.len)) {
627 			kmem_free(syncaddr.buf, syncaddr.len);
628 			error = EFAULT;
629 			goto errout;
630 		}
631 
632 		nargs->syncaddr = kmem_alloc(sizeof (struct netbuf), KM_SLEEP);
633 		bcopy(&syncaddr, nargs->syncaddr, sizeof (struct netbuf));
634 
635 		/* get server's netname */
636 		if (copyinstr(STRUCT_FGETP(args, netname), netname,
637 		    sizeof (netname), &nlen)) {
638 			error = EFAULT;
639 			goto errout;
640 		}
641 
642 		netname[nlen] = '\0';
643 		nargs->netname = kmem_zalloc(nlen, KM_SLEEP);
644 		(void) strcpy(nargs->netname, netname);
645 	}
646 
647 	/*
648 	 * Get the extention data which has the security data structure.
649 	 * This includes data for AUTH_SYS as well.
650 	 */
651 	if (flags & NFSMNT_NEWARGS) {
652 		nargs->nfs_args_ext = STRUCT_FGET(args, nfs_args_ext);
653 		if (nargs->nfs_args_ext == NFS_ARGS_EXTA ||
654 		    nargs->nfs_args_ext == NFS_ARGS_EXTB) {
655 			/*
656 			 * Indicating the application is using the new
657 			 * sec_data structure to pass in the security
658 			 * data.
659 			 */
660 			if (STRUCT_FGETP(args,
661 			    nfs_ext_u.nfs_extA.secdata) != NULL) {
662 				error = sec_clnt_loadinfo(
663 				    (struct sec_data *)STRUCT_FGETP(args,
664 				    nfs_ext_u.nfs_extA.secdata),
665 				    &secdata, get_udatamodel());
666 			}
667 			nargs->nfs_ext_u.nfs_extA.secdata = secdata;
668 		}
669 	}
670 
671 	if (error)
672 		goto errout;
673 
674 	/*
675 	 * Failover support:
676 	 *
677 	 * We may have a linked list of nfs_args structures,
678 	 * which means the user is looking for failover.  If
679 	 * the mount is either not "read-only" or "soft",
680 	 * we want to bail out with EINVAL.
681 	 */
682 	if (nargs->nfs_args_ext == NFS_ARGS_EXTB)
683 		nargs->nfs_ext_u.nfs_extB.next =
684 		    STRUCT_FGETP(args, nfs_ext_u.nfs_extB.next);
685 
686 errout:
687 	if (error)
688 		nfs4_free_args(nargs);
689 
690 	return (error);
691 }
692 
693 
694 /*
695  * nfs mount vfsop
696  * Set up mount info record and attach it to vfs struct.
697  */
698 int
699 nfs4_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
700 {
701 	char *data = uap->dataptr;
702 	int error;
703 	vnode_t *rtvp;			/* the server's root */
704 	mntinfo4_t *mi;			/* mount info, pointed at by vfs */
705 	struct knetconfig *rdma_knconf;	/* rdma transport structure */
706 	rnode4_t *rp;
707 	struct servinfo4 *svp;		/* nfs server info */
708 	struct servinfo4 *svp_tail = NULL; /* previous nfs server info */
709 	struct servinfo4 *svp_head;	/* first nfs server info */
710 	struct servinfo4 *svp_2ndlast;	/* 2nd last in server info list */
711 	struct sec_data *secdata;	/* security data */
712 	struct nfs_args *args = NULL;
713 	int flags, addr_type, removed;
714 	zone_t *zone = nfs_zone();
715 	nfs4_error_t n4e;
716 	zone_t *mntzone = NULL;
717 
718 	if (secpolicy_fs_mount(cr, mvp, vfsp) != 0)
719 		return (EPERM);
720 	if (mvp->v_type != VDIR)
721 		return (ENOTDIR);
722 
723 	/*
724 	 * get arguments
725 	 *
726 	 * nfs_args is now versioned and is extensible, so
727 	 * uap->datalen might be different from sizeof (args)
728 	 * in a compatible situation.
729 	 */
730 more:
731 	if (!(uap->flags & MS_SYSSPACE)) {
732 		if (args == NULL)
733 			args = kmem_zalloc(sizeof (struct nfs_args), KM_SLEEP);
734 		else
735 			nfs4_free_args(args);
736 		error = nfs4_copyin(data, uap->datalen, args);
737 		if (error) {
738 			if (args) {
739 				kmem_free(args, sizeof (*args));
740 			}
741 			return (error);
742 		}
743 	} else {
744 		args = (struct nfs_args *)data;
745 	}
746 
747 	flags = args->flags;
748 
749 	/*
750 	 * If the request changes the locking type, disallow the remount,
751 	 * because it's questionable whether we can transfer the
752 	 * locking state correctly.
753 	 */
754 	if (uap->flags & MS_REMOUNT) {
755 		if (!(uap->flags & MS_SYSSPACE)) {
756 			nfs4_free_args(args);
757 			kmem_free(args, sizeof (*args));
758 		}
759 		if ((mi = VFTOMI4(vfsp)) != NULL) {
760 			uint_t new_mi_llock;
761 			uint_t old_mi_llock;
762 			new_mi_llock = (flags & NFSMNT_LLOCK) ? 1 : 0;
763 			old_mi_llock = (mi->mi_flags & MI4_LLOCK) ? 1 : 0;
764 			if (old_mi_llock != new_mi_llock)
765 				return (EBUSY);
766 		}
767 		return (0);
768 	}
769 
770 	/*
771 	 * For ephemeral mount trigger stub vnodes, we have two problems
772 	 * to solve: racing threads will likely fail the v_count check, and
773 	 * we want only one to proceed with the mount.
774 	 *
775 	 * For stubs, if the mount has already occurred (via a racing thread),
776 	 * just return success. If not, skip the v_count check and proceed.
777 	 * Note that we are already serialised at this point.
778 	 */
779 	mutex_enter(&mvp->v_lock);
780 	if (vn_matchops(mvp, nfs4_trigger_vnodeops)) {
781 		/* mntpt is a v4 stub vnode */
782 		ASSERT(RP_ISSTUB(VTOR4(mvp)));
783 		ASSERT(!(uap->flags & MS_OVERLAY));
784 		ASSERT(!(mvp->v_flag & VROOT));
785 		if (vn_mountedvfs(mvp) != NULL) {
786 			/* ephemeral mount has already occurred */
787 			ASSERT(uap->flags & MS_SYSSPACE);
788 			mutex_exit(&mvp->v_lock);
789 			return (0);
790 		}
791 	} else {
792 		/* mntpt is a non-v4 or v4 non-stub vnode */
793 		if (!(uap->flags & MS_OVERLAY) &&
794 		    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
795 			mutex_exit(&mvp->v_lock);
796 			if (!(uap->flags & MS_SYSSPACE)) {
797 				nfs4_free_args(args);
798 				kmem_free(args, sizeof (*args));
799 			}
800 			return (EBUSY);
801 		}
802 	}
803 	mutex_exit(&mvp->v_lock);
804 
805 	/* make sure things are zeroed for errout: */
806 	rtvp = NULL;
807 	mi = NULL;
808 	secdata = NULL;
809 
810 	/*
811 	 * A valid knetconfig structure is required.
812 	 */
813 	if (!(flags & NFSMNT_KNCONF) ||
814 	    args->knconf == NULL || args->knconf->knc_protofmly == NULL ||
815 	    args->knconf->knc_proto == NULL ||
816 	    (strcmp(args->knconf->knc_proto, NC_UDP) == 0)) {
817 		if (!(uap->flags & MS_SYSSPACE)) {
818 			nfs4_free_args(args);
819 			kmem_free(args, sizeof (*args));
820 		}
821 		return (EINVAL);
822 	}
823 
824 	if ((strlen(args->knconf->knc_protofmly) >= KNC_STRSIZE) ||
825 	    (strlen(args->knconf->knc_proto) >= KNC_STRSIZE)) {
826 		if (!(uap->flags & MS_SYSSPACE)) {
827 			nfs4_free_args(args);
828 			kmem_free(args, sizeof (*args));
829 		}
830 		return (EINVAL);
831 	}
832 
833 	/*
834 	 * Allocate a servinfo4 struct.
835 	 */
836 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
837 	nfs_rw_init(&svp->sv_lock, NULL, RW_DEFAULT, NULL);
838 	if (svp_tail) {
839 		svp_2ndlast = svp_tail;
840 		svp_tail->sv_next = svp;
841 	} else {
842 		svp_head = svp;
843 		svp_2ndlast = svp;
844 	}
845 
846 	svp_tail = svp;
847 	svp->sv_knconf = args->knconf;
848 	args->knconf = NULL;
849 
850 	/*
851 	 * Get server address
852 	 */
853 	if (args->addr == NULL || args->addr->buf == NULL) {
854 		error = EINVAL;
855 		goto errout;
856 	}
857 
858 	svp->sv_addr.maxlen = args->addr->maxlen;
859 	svp->sv_addr.len = args->addr->len;
860 	svp->sv_addr.buf = args->addr->buf;
861 	args->addr->buf = NULL;
862 
863 	/*
864 	 * Get the root fhandle
865 	 */
866 	if (args->fh == NULL || (strlen(args->fh) >= MAXPATHLEN)) {
867 		error = EINVAL;
868 		goto errout;
869 	}
870 
871 	svp->sv_path = args->fh;
872 	svp->sv_pathlen = strlen(args->fh) + 1;
873 	args->fh = NULL;
874 
875 	/*
876 	 * Get server's hostname
877 	 */
878 	if (flags & NFSMNT_HOSTNAME) {
879 		if (args->hostname == NULL || (strlen(args->hostname) >
880 		    MAXNETNAMELEN)) {
881 			error = EINVAL;
882 			goto errout;
883 		}
884 		svp->sv_hostnamelen = strlen(args->hostname) + 1;
885 		svp->sv_hostname = args->hostname;
886 		args->hostname = NULL;
887 	} else {
888 		char *p = "unknown-host";
889 		svp->sv_hostnamelen = strlen(p) + 1;
890 		svp->sv_hostname = kmem_zalloc(svp->sv_hostnamelen, KM_SLEEP);
891 		(void) strcpy(svp->sv_hostname, p);
892 	}
893 
894 	/*
895 	 * RDMA MOUNT SUPPORT FOR NFS v4.
896 	 * Establish, is it possible to use RDMA, if so overload the
897 	 * knconf with rdma specific knconf and free the orignal knconf.
898 	 */
899 	if ((flags & NFSMNT_TRYRDMA) || (flags & NFSMNT_DORDMA)) {
900 		/*
901 		 * Determine the addr type for RDMA, IPv4 or v6.
902 		 */
903 		if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET) == 0)
904 			addr_type = AF_INET;
905 		else if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET6) == 0)
906 			addr_type = AF_INET6;
907 
908 		if (rdma_reachable(addr_type, &svp->sv_addr,
909 		    &rdma_knconf) == 0) {
910 			/*
911 			 * If successful, hijack the orignal knconf and
912 			 * replace with the new one, depending on the flags.
913 			 */
914 			svp->sv_origknconf = svp->sv_knconf;
915 			svp->sv_knconf = rdma_knconf;
916 		} else {
917 			if (flags & NFSMNT_TRYRDMA) {
918 #ifdef	DEBUG
919 				if (rdma_debug)
920 					zcmn_err(getzoneid(), CE_WARN,
921 					    "no RDMA onboard, revert\n");
922 #endif
923 			}
924 
925 			if (flags & NFSMNT_DORDMA) {
926 				/*
927 				 * If proto=rdma is specified and no RDMA
928 				 * path to this server is avialable then
929 				 * ditch this server.
930 				 * This is not included in the mountable
931 				 * server list or the replica list.
932 				 * Check if more servers are specified;
933 				 * Failover case, otherwise bail out of mount.
934 				 */
935 				if (args->nfs_args_ext == NFS_ARGS_EXTB &&
936 				    args->nfs_ext_u.nfs_extB.next != NULL) {
937 					data = (char *)
938 					    args->nfs_ext_u.nfs_extB.next;
939 					if (uap->flags & MS_RDONLY &&
940 					    !(flags & NFSMNT_SOFT)) {
941 						if (svp_head->sv_next == NULL) {
942 							svp_tail = NULL;
943 							svp_2ndlast = NULL;
944 							sv4_free(svp_head);
945 							goto more;
946 						} else {
947 							svp_tail = svp_2ndlast;
948 							svp_2ndlast->sv_next =
949 							    NULL;
950 							sv4_free(svp);
951 							goto more;
952 						}
953 					}
954 				} else {
955 					/*
956 					 * This is the last server specified
957 					 * in the nfs_args list passed down
958 					 * and its not rdma capable.
959 					 */
960 					if (svp_head->sv_next == NULL) {
961 						/*
962 						 * Is this the only one
963 						 */
964 						error = EINVAL;
965 #ifdef	DEBUG
966 						if (rdma_debug)
967 							zcmn_err(getzoneid(),
968 							    CE_WARN,
969 							    "No RDMA srv");
970 #endif
971 						goto errout;
972 					} else {
973 						/*
974 						 * There is list, since some
975 						 * servers specified before
976 						 * this passed all requirements
977 						 */
978 						svp_tail = svp_2ndlast;
979 						svp_2ndlast->sv_next = NULL;
980 						sv4_free(svp);
981 						goto proceed;
982 					}
983 				}
984 			}
985 		}
986 	}
987 
988 	/*
989 	 * If there are syncaddr and netname data, load them in. This is
990 	 * to support data needed for NFSV4 when AUTH_DH is the negotiated
991 	 * flavor via SECINFO. (instead of using MOUNT protocol in V3).
992 	 */
993 	if (args->flags & NFSMNT_SECURE) {
994 		svp->sv_dhsec = create_authdh_data(args->netname,
995 		    strlen(args->netname),
996 		    args->syncaddr, svp->sv_knconf);
997 	}
998 
999 	/*
1000 	 * Get the extention data which has the security data structure.
1001 	 * This includes data for AUTH_SYS as well.
1002 	 */
1003 	if (flags & NFSMNT_NEWARGS) {
1004 		switch (args->nfs_args_ext) {
1005 		case NFS_ARGS_EXTA:
1006 		case NFS_ARGS_EXTB:
1007 			/*
1008 			 * Indicating the application is using the new
1009 			 * sec_data structure to pass in the security
1010 			 * data.
1011 			 */
1012 			secdata = args->nfs_ext_u.nfs_extA.secdata;
1013 			if (secdata == NULL) {
1014 				error = EINVAL;
1015 			} else if (uap->flags & MS_SYSSPACE) {
1016 				/*
1017 				 * Need to validate the flavor here if
1018 				 * sysspace, userspace was already
1019 				 * validate from the nfs_copyin function.
1020 				 */
1021 				switch (secdata->rpcflavor) {
1022 				case AUTH_NONE:
1023 				case AUTH_UNIX:
1024 				case AUTH_LOOPBACK:
1025 				case AUTH_DES:
1026 				case RPCSEC_GSS:
1027 					break;
1028 				default:
1029 					error = EINVAL;
1030 					goto errout;
1031 				}
1032 			}
1033 			args->nfs_ext_u.nfs_extA.secdata = NULL;
1034 			break;
1035 
1036 		default:
1037 			error = EINVAL;
1038 			break;
1039 		}
1040 
1041 	} else if (flags & NFSMNT_SECURE) {
1042 		/*
1043 		 * NFSMNT_SECURE is deprecated but we keep it
1044 		 * to support the rogue user-generated application
1045 		 * that may use this undocumented interface to do
1046 		 * AUTH_DH security, e.g. our own rexd.
1047 		 *
1048 		 * Also note that NFSMNT_SECURE is used for passing
1049 		 * AUTH_DH info to be used in negotiation.
1050 		 */
1051 		secdata = create_authdh_data(args->netname,
1052 		    strlen(args->netname), args->syncaddr, svp->sv_knconf);
1053 
1054 	} else {
1055 		secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
1056 		secdata->secmod = secdata->rpcflavor = AUTH_SYS;
1057 		secdata->data = NULL;
1058 	}
1059 
1060 	svp->sv_secdata = secdata;
1061 
1062 	/*
1063 	 * User does not explictly specify a flavor, and a user
1064 	 * defined default flavor is passed down.
1065 	 */
1066 	if (flags & NFSMNT_SECDEFAULT) {
1067 		(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
1068 		svp->sv_flags |= SV4_TRYSECDEFAULT;
1069 		nfs_rw_exit(&svp->sv_lock);
1070 	}
1071 
1072 	/*
1073 	 * Failover support:
1074 	 *
1075 	 * We may have a linked list of nfs_args structures,
1076 	 * which means the user is looking for failover.  If
1077 	 * the mount is either not "read-only" or "soft",
1078 	 * we want to bail out with EINVAL.
1079 	 */
1080 	if (args->nfs_args_ext == NFS_ARGS_EXTB &&
1081 	    args->nfs_ext_u.nfs_extB.next != NULL) {
1082 		if (uap->flags & MS_RDONLY && !(flags & NFSMNT_SOFT)) {
1083 			data = (char *)args->nfs_ext_u.nfs_extB.next;
1084 			goto more;
1085 		}
1086 		error = EINVAL;
1087 		goto errout;
1088 	}
1089 
1090 	/*
1091 	 * Determine the zone we're being mounted into.
1092 	 */
1093 	zone_hold(mntzone = zone);		/* start with this assumption */
1094 	if (getzoneid() == GLOBAL_ZONEID) {
1095 		zone_rele(mntzone);
1096 		mntzone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
1097 		ASSERT(mntzone != NULL);
1098 		if (mntzone != zone) {
1099 			error = EBUSY;
1100 			goto errout;
1101 		}
1102 	}
1103 
1104 	if (is_system_labeled()) {
1105 		error = nfs_mount_label_policy(vfsp, &svp->sv_addr,
1106 		    svp->sv_knconf, cr);
1107 
1108 		if (error > 0)
1109 			goto errout;
1110 
1111 		if (error == -1) {
1112 			/* change mount to read-only to prevent write-down */
1113 			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1114 		}
1115 	}
1116 
1117 	/*
1118 	 * Stop the mount from going any further if the zone is going away.
1119 	 */
1120 	if (zone_status_get(mntzone) >= ZONE_IS_SHUTTING_DOWN) {
1121 		error = EBUSY;
1122 		goto errout;
1123 	}
1124 
1125 	/*
1126 	 * Get root vnode.
1127 	 */
1128 proceed:
1129 	error = nfs4rootvp(&rtvp, vfsp, svp_head, flags, cr, mntzone);
1130 	if (error) {
1131 		/* if nfs4rootvp failed, it will free svp_head */
1132 		svp_head = NULL;
1133 		goto errout;
1134 	}
1135 
1136 	mi = VTOMI4(rtvp);
1137 
1138 	/*
1139 	 * Send client id to the server, if necessary
1140 	 */
1141 	nfs4_error_zinit(&n4e);
1142 	nfs4setclientid(mi, cr, FALSE, &n4e);
1143 
1144 	error = n4e.error;
1145 
1146 	if (error)
1147 		goto errout;
1148 
1149 	/*
1150 	 * Set option fields in the mount info record
1151 	 */
1152 
1153 	if (svp_head->sv_next) {
1154 		mutex_enter(&mi->mi_lock);
1155 		mi->mi_flags |= MI4_LLOCK;
1156 		mutex_exit(&mi->mi_lock);
1157 	}
1158 	error = nfs4_setopts(rtvp, DATAMODEL_NATIVE, args);
1159 	if (error)
1160 		goto errout;
1161 
1162 	/*
1163 	 * Time to tie in the mirror mount info at last!
1164 	 */
1165 	if (flags & NFSMNT_EPHEMERAL)
1166 		error = nfs4_record_ephemeral_mount(mi, mvp);
1167 
1168 errout:
1169 	if (error) {
1170 		if (rtvp != NULL) {
1171 			rp = VTOR4(rtvp);
1172 			if (rp->r_flags & R4HASHED)
1173 				rp4_rmhash(rp);
1174 		}
1175 		if (mi != NULL) {
1176 			nfs4_async_stop(vfsp);
1177 			nfs4_async_manager_stop(vfsp);
1178 			nfs4_remove_mi_from_server(mi, NULL);
1179 			if (rtvp != NULL)
1180 				VN_RELE(rtvp);
1181 			if (mntzone != NULL)
1182 				zone_rele(mntzone);
1183 			/* need to remove it from the zone */
1184 			removed = nfs4_mi_zonelist_remove(mi);
1185 			if (removed)
1186 				zone_rele_ref(&mi->mi_zone_ref,
1187 				    ZONE_REF_NFSV4);
1188 			MI4_RELE(mi);
1189 			if (!(uap->flags & MS_SYSSPACE) && args) {
1190 				nfs4_free_args(args);
1191 				kmem_free(args, sizeof (*args));
1192 			}
1193 			return (error);
1194 		}
1195 		if (svp_head)
1196 			sv4_free(svp_head);
1197 	}
1198 
1199 	if (!(uap->flags & MS_SYSSPACE) && args) {
1200 		nfs4_free_args(args);
1201 		kmem_free(args, sizeof (*args));
1202 	}
1203 	if (rtvp != NULL)
1204 		VN_RELE(rtvp);
1205 
1206 	if (mntzone != NULL)
1207 		zone_rele(mntzone);
1208 
1209 	return (error);
1210 }
1211 
1212 #ifdef  DEBUG
1213 #define	VERS_MSG	"NFS4 server "
1214 #else
1215 #define	VERS_MSG	"NFS server "
1216 #endif
1217 
1218 #define	READ_MSG        \
1219 	VERS_MSG "%s returned 0 for read transfer size"
1220 #define	WRITE_MSG       \
1221 	VERS_MSG "%s returned 0 for write transfer size"
1222 #define	SIZE_MSG        \
1223 	VERS_MSG "%s returned 0 for maximum file size"
1224 
1225 /*
1226  * Get the symbolic link text from the server for a given filehandle
1227  * of that symlink.
1228  *
1229  *      (get symlink text) PUTFH READLINK
1230  */
1231 static int
1232 getlinktext_otw(mntinfo4_t *mi, nfs_fh4 *fh, char **linktextp, cred_t *cr,
1233     int flags)
1234 {
1235 	COMPOUND4args_clnt args;
1236 	COMPOUND4res_clnt res;
1237 	int doqueue;
1238 	nfs_argop4 argop[2];
1239 	nfs_resop4 *resop;
1240 	READLINK4res *lr_res;
1241 	uint_t len;
1242 	bool_t needrecov = FALSE;
1243 	nfs4_recov_state_t recov_state;
1244 	nfs4_sharedfh_t *sfh;
1245 	nfs4_error_t e;
1246 	int num_retry = nfs4_max_mount_retry;
1247 	int recovery = !(flags & NFS4_GETFH_NEEDSOP);
1248 
1249 	sfh = sfh4_get(fh, mi);
1250 	recov_state.rs_flags = 0;
1251 	recov_state.rs_num_retry_despite_err = 0;
1252 
1253 recov_retry:
1254 	nfs4_error_zinit(&e);
1255 
1256 	args.array_len = 2;
1257 	args.array = argop;
1258 	args.ctag = TAG_GET_SYMLINK;
1259 
1260 	if (! recovery) {
1261 		e.error = nfs4_start_op(mi, NULL, NULL, &recov_state);
1262 		if (e.error) {
1263 			sfh4_rele(&sfh);
1264 			return (e.error);
1265 		}
1266 	}
1267 
1268 	/* 0. putfh symlink fh */
1269 	argop[0].argop = OP_CPUTFH;
1270 	argop[0].nfs_argop4_u.opcputfh.sfh = sfh;
1271 
1272 	/* 1. readlink */
1273 	argop[1].argop = OP_READLINK;
1274 
1275 	doqueue = 1;
1276 
1277 	rfs4call(mi, &args, &res, cr, &doqueue, 0, &e);
1278 
1279 	needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp);
1280 
1281 	if (needrecov && !recovery && num_retry-- > 0) {
1282 
1283 		NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
1284 		    "getlinktext_otw: initiating recovery\n"));
1285 
1286 		if (nfs4_start_recovery(&e, mi, NULL, NULL, NULL, NULL,
1287 		    OP_READLINK, NULL, NULL, NULL) == FALSE) {
1288 			nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
1289 			if (!e.error)
1290 				xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1291 			goto recov_retry;
1292 		}
1293 	}
1294 
1295 	/*
1296 	 * If non-NFS4 pcol error and/or we weren't able to recover.
1297 	 */
1298 	if (e.error != 0) {
1299 		if (! recovery)
1300 			nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
1301 		sfh4_rele(&sfh);
1302 		return (e.error);
1303 	}
1304 
1305 	if (res.status) {
1306 		e.error = geterrno4(res.status);
1307 		xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1308 		if (! recovery)
1309 			nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
1310 		sfh4_rele(&sfh);
1311 		return (e.error);
1312 	}
1313 
1314 	/* res.status == NFS4_OK */
1315 	ASSERT(res.status == NFS4_OK);
1316 
1317 	resop = &res.array[1];  /* readlink res */
1318 	lr_res = &resop->nfs_resop4_u.opreadlink;
1319 
1320 	/* treat symlink name as data */
1321 	*linktextp = utf8_to_str((utf8string *)&lr_res->link, &len, NULL);
1322 
1323 	if (! recovery)
1324 		nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
1325 	sfh4_rele(&sfh);
1326 	xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1327 	return (0);
1328 }
1329 
1330 /*
1331  * Skip over consecutive slashes and "/./" in a pathname.
1332  */
1333 void
1334 pathname_skipslashdot(struct pathname *pnp)
1335 {
1336 	char *c1, *c2;
1337 
1338 	while (pnp->pn_pathlen > 0 && *pnp->pn_path == '/') {
1339 
1340 		c1 = pnp->pn_path + 1;
1341 		c2 = pnp->pn_path + 2;
1342 
1343 		if (*c1 == '.' && (*c2 == '/' || *c2 == '\0')) {
1344 			pnp->pn_path = pnp->pn_path + 2; /* skip "/." */
1345 			pnp->pn_pathlen = pnp->pn_pathlen - 2;
1346 		} else {
1347 			pnp->pn_path++;
1348 			pnp->pn_pathlen--;
1349 		}
1350 	}
1351 }
1352 
1353 /*
1354  * Resolve a symbolic link path. The symlink is in the nth component of
1355  * svp->sv_path and has an nfs4 file handle "fh".
1356  * Upon return, the sv_path will point to the new path that has the nth
1357  * component resolved to its symlink text.
1358  */
1359 int
1360 resolve_sympath(mntinfo4_t *mi, servinfo4_t *svp, int nth, nfs_fh4 *fh,
1361     cred_t *cr, int flags)
1362 {
1363 	char *oldpath;
1364 	char *symlink, *newpath;
1365 	struct pathname oldpn, newpn;
1366 	char component[MAXNAMELEN];
1367 	int i, addlen, error = 0;
1368 	int oldpathlen;
1369 
1370 	/* Get the symbolic link text over the wire. */
1371 	error = getlinktext_otw(mi, fh, &symlink, cr, flags);
1372 
1373 	if (error || symlink == NULL || strlen(symlink) == 0)
1374 		return (error);
1375 
1376 	/*
1377 	 * Compose the new pathname.
1378 	 * Note:
1379 	 *    - only the nth component is resolved for the pathname.
1380 	 *    - pathname.pn_pathlen does not count the ending null byte.
1381 	 */
1382 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1383 	oldpath = svp->sv_path;
1384 	oldpathlen = svp->sv_pathlen;
1385 	if (error = pn_get(oldpath, UIO_SYSSPACE, &oldpn)) {
1386 		nfs_rw_exit(&svp->sv_lock);
1387 		kmem_free(symlink, strlen(symlink) + 1);
1388 		return (error);
1389 	}
1390 	nfs_rw_exit(&svp->sv_lock);
1391 	pn_alloc(&newpn);
1392 
1393 	/*
1394 	 * Skip over previous components from the oldpath so that the
1395 	 * oldpn.pn_path will point to the symlink component. Skip
1396 	 * leading slashes and "/./" (no OP_LOOKUP on ".") so that
1397 	 * pn_getcompnent can get the component.
1398 	 */
1399 	for (i = 1; i < nth; i++) {
1400 		pathname_skipslashdot(&oldpn);
1401 		error = pn_getcomponent(&oldpn, component);
1402 		if (error)
1403 			goto out;
1404 	}
1405 
1406 	/*
1407 	 * Copy the old path upto the component right before the symlink
1408 	 * if the symlink is not an absolute path.
1409 	 */
1410 	if (symlink[0] != '/') {
1411 		addlen = oldpn.pn_path - oldpn.pn_buf;
1412 		bcopy(oldpn.pn_buf, newpn.pn_path, addlen);
1413 		newpn.pn_pathlen += addlen;
1414 		newpn.pn_path += addlen;
1415 		newpn.pn_buf[newpn.pn_pathlen] = '/';
1416 		newpn.pn_pathlen++;
1417 		newpn.pn_path++;
1418 	}
1419 
1420 	/* copy the resolved symbolic link text */
1421 	addlen = strlen(symlink);
1422 	if (newpn.pn_pathlen + addlen >= newpn.pn_bufsize) {
1423 		error = ENAMETOOLONG;
1424 		goto out;
1425 	}
1426 	bcopy(symlink, newpn.pn_path, addlen);
1427 	newpn.pn_pathlen += addlen;
1428 	newpn.pn_path += addlen;
1429 
1430 	/*
1431 	 * Check if there is any remaining path after the symlink component.
1432 	 * First, skip the symlink component.
1433 	 */
1434 	pathname_skipslashdot(&oldpn);
1435 	if (error = pn_getcomponent(&oldpn, component))
1436 		goto out;
1437 
1438 	addlen = pn_pathleft(&oldpn); /* includes counting the slash */
1439 
1440 	/*
1441 	 * Copy the remaining path to the new pathname if there is any.
1442 	 */
1443 	if (addlen > 0) {
1444 		if (newpn.pn_pathlen + addlen >= newpn.pn_bufsize) {
1445 			error = ENAMETOOLONG;
1446 			goto out;
1447 		}
1448 		bcopy(oldpn.pn_path, newpn.pn_path, addlen);
1449 		newpn.pn_pathlen += addlen;
1450 	}
1451 	newpn.pn_buf[newpn.pn_pathlen] = '\0';
1452 
1453 	/* get the newpath and store it in the servinfo4_t */
1454 	newpath = kmem_alloc(newpn.pn_pathlen + 1, KM_SLEEP);
1455 	bcopy(newpn.pn_buf, newpath, newpn.pn_pathlen);
1456 	newpath[newpn.pn_pathlen] = '\0';
1457 
1458 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
1459 	svp->sv_path = newpath;
1460 	svp->sv_pathlen = strlen(newpath) + 1;
1461 	nfs_rw_exit(&svp->sv_lock);
1462 
1463 	kmem_free(oldpath, oldpathlen);
1464 out:
1465 	kmem_free(symlink, strlen(symlink) + 1);
1466 	pn_free(&newpn);
1467 	pn_free(&oldpn);
1468 
1469 	return (error);
1470 }
1471 
1472 /*
1473  * This routine updates servinfo4 structure with the new referred server
1474  * info.
1475  * nfsfsloc has the location related information
1476  * fsp has the hostname and pathname info.
1477  * new path = pathname from referral + part of orig pathname(based on nth).
1478  */
1479 static void
1480 update_servinfo4(servinfo4_t *svp, fs_location4 *fsp,
1481     struct nfs_fsl_info *nfsfsloc, char *orig_path, int nth)
1482 {
1483 	struct knetconfig *knconf, *svknconf;
1484 	struct netbuf *saddr;
1485 	sec_data_t	*secdata;
1486 	utf8string *host;
1487 	int i = 0, num_slashes = 0;
1488 	char *p, *spath, *op, *new_path;
1489 
1490 	/* Update knconf */
1491 	knconf = svp->sv_knconf;
1492 	free_knconf_contents(knconf);
1493 	bzero(knconf, sizeof (struct knetconfig));
1494 	svknconf = nfsfsloc->knconf;
1495 	knconf->knc_semantics = svknconf->knc_semantics;
1496 	knconf->knc_protofmly = kmem_zalloc(KNC_STRSIZE, KM_SLEEP);
1497 	knconf->knc_proto = kmem_zalloc(KNC_STRSIZE, KM_SLEEP);
1498 	knconf->knc_rdev = svknconf->knc_rdev;
1499 	bcopy(svknconf->knc_protofmly, knconf->knc_protofmly, KNC_STRSIZE);
1500 	bcopy(svknconf->knc_proto, knconf->knc_proto, KNC_STRSIZE);
1501 
1502 	/* Update server address */
1503 	saddr = &svp->sv_addr;
1504 	if (saddr->buf != NULL)
1505 		kmem_free(saddr->buf, saddr->maxlen);
1506 	saddr->buf  = kmem_alloc(nfsfsloc->addr->maxlen, KM_SLEEP);
1507 	saddr->len = nfsfsloc->addr->len;
1508 	saddr->maxlen = nfsfsloc->addr->maxlen;
1509 	bcopy(nfsfsloc->addr->buf, saddr->buf, nfsfsloc->addr->len);
1510 
1511 	/* Update server name */
1512 	host = fsp->server_val;
1513 	kmem_free(svp->sv_hostname, svp->sv_hostnamelen);
1514 	svp->sv_hostname = kmem_zalloc(host->utf8string_len + 1, KM_SLEEP);
1515 	bcopy(host->utf8string_val, svp->sv_hostname, host->utf8string_len);
1516 	svp->sv_hostname[host->utf8string_len] = '\0';
1517 	svp->sv_hostnamelen = host->utf8string_len + 1;
1518 
1519 	/*
1520 	 * Update server path.
1521 	 * We need to setup proper path here.
1522 	 * For ex., If we got a path name serv1:/rp/aaa/bbb
1523 	 * where aaa is a referral and points to serv2:/rpool/aa
1524 	 * we need to set the path to serv2:/rpool/aa/bbb
1525 	 * The first part of this below code generates /rpool/aa
1526 	 * and the second part appends /bbb to the server path.
1527 	 */
1528 	spath = p = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1529 	*p++ = '/';
1530 	for (i = 0; i < fsp->rootpath.pathname4_len; i++) {
1531 		component4 *comp;
1532 
1533 		comp = &fsp->rootpath.pathname4_val[i];
1534 		/* If no space, null the string and bail */
1535 		if ((p - spath) + comp->utf8string_len + 1 > MAXPATHLEN) {
1536 			p = spath + MAXPATHLEN - 1;
1537 			spath[0] = '\0';
1538 			break;
1539 		}
1540 		bcopy(comp->utf8string_val, p, comp->utf8string_len);
1541 		p += comp->utf8string_len;
1542 		*p++ = '/';
1543 	}
1544 	if (fsp->rootpath.pathname4_len != 0)
1545 		*(p - 1) = '\0';
1546 	else
1547 		*p = '\0';
1548 	p = spath;
1549 
1550 	new_path = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1551 	(void) strlcpy(new_path, p, MAXPATHLEN);
1552 	kmem_free(p, MAXPATHLEN);
1553 	i = strlen(new_path);
1554 
1555 	for (op = orig_path; *op; op++) {
1556 		if (*op == '/')
1557 			num_slashes++;
1558 		if (num_slashes == nth + 2) {
1559 			while (*op != '\0') {
1560 				new_path[i] = *op;
1561 				i++;
1562 				op++;
1563 			}
1564 			break;
1565 		}
1566 	}
1567 	new_path[i] = '\0';
1568 
1569 	kmem_free(svp->sv_path, svp->sv_pathlen);
1570 	svp->sv_pathlen = strlen(new_path) + 1;
1571 	svp->sv_path = kmem_alloc(svp->sv_pathlen, KM_SLEEP);
1572 	bcopy(new_path, svp->sv_path, svp->sv_pathlen);
1573 	kmem_free(new_path, MAXPATHLEN);
1574 
1575 	/*
1576 	 * All the security data is specific to old server.
1577 	 * Clean it up except secdata which deals with mount options.
1578 	 * We need to inherit that data. Copy secdata into our new servinfo4.
1579 	 */
1580 	if (svp->sv_dhsec) {
1581 		sec_clnt_freeinfo(svp->sv_dhsec);
1582 		svp->sv_dhsec = NULL;
1583 	}
1584 	if (svp->sv_save_secinfo &&
1585 	    svp->sv_save_secinfo != svp->sv_secinfo) {
1586 		secinfo_free(svp->sv_save_secinfo);
1587 		svp->sv_save_secinfo = NULL;
1588 	}
1589 	if (svp->sv_secinfo) {
1590 		secinfo_free(svp->sv_secinfo);
1591 		svp->sv_secinfo = NULL;
1592 	}
1593 	svp->sv_currsec = NULL;
1594 
1595 	secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
1596 	*secdata = *svp->sv_secdata;
1597 	secdata->data = NULL;
1598 	if (svp->sv_secdata) {
1599 		sec_clnt_freeinfo(svp->sv_secdata);
1600 		svp->sv_secdata = NULL;
1601 	}
1602 	svp->sv_secdata = secdata;
1603 }
1604 
1605 /*
1606  * Resolve a referral. The referral is in the n+1th component of
1607  * svp->sv_path and has a parent nfs4 file handle "fh".
1608  * Upon return, the sv_path will point to the new path that has referral
1609  * component resolved to its referred path and part of original path.
1610  * Hostname and other address information is also updated.
1611  */
1612 int
1613 resolve_referral(mntinfo4_t *mi, servinfo4_t *svp, cred_t *cr, int nth,
1614     nfs_fh4 *fh)
1615 {
1616 	nfs4_sharedfh_t	*sfh;
1617 	struct nfs_fsl_info nfsfsloc;
1618 	nfs4_ga_res_t garp;
1619 	COMPOUND4res_clnt callres;
1620 	fs_location4	*fsp;
1621 	char *nm, *orig_path;
1622 	int orig_pathlen = 0, ret = -1, index;
1623 
1624 	if (svp->sv_pathlen <= 0)
1625 		return (ret);
1626 
1627 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
1628 	orig_pathlen = svp->sv_pathlen;
1629 	orig_path = kmem_alloc(orig_pathlen, KM_SLEEP);
1630 	bcopy(svp->sv_path, orig_path, orig_pathlen);
1631 	nm = extract_referral_point(svp->sv_path, nth);
1632 	setup_newsvpath(svp, nth);
1633 	nfs_rw_exit(&svp->sv_lock);
1634 
1635 	sfh = sfh4_get(fh, mi);
1636 	index = nfs4_process_referral(mi, sfh, nm, cr,
1637 	    &garp, &callres, &nfsfsloc);
1638 	sfh4_rele(&sfh);
1639 	kmem_free(nm, MAXPATHLEN);
1640 	if (index < 0) {
1641 		kmem_free(orig_path, orig_pathlen);
1642 		return (index);
1643 	}
1644 
1645 	fsp =  &garp.n4g_ext_res->n4g_fslocations.locations_val[index];
1646 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
1647 	update_servinfo4(svp, fsp, &nfsfsloc, orig_path, nth);
1648 	nfs_rw_exit(&svp->sv_lock);
1649 
1650 	mutex_enter(&mi->mi_lock);
1651 	mi->mi_vfs_referral_loop_cnt++;
1652 	mutex_exit(&mi->mi_lock);
1653 
1654 	ret = 0;
1655 bad:
1656 	/* Free up XDR memory allocated in nfs4_process_referral() */
1657 	xdr_free(xdr_nfs_fsl_info, (char *)&nfsfsloc);
1658 	xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&callres);
1659 	kmem_free(orig_path, orig_pathlen);
1660 
1661 	return (ret);
1662 }
1663 
1664 /*
1665  * Get the root filehandle for the given filesystem and server, and update
1666  * svp.
1667  *
1668  * If NFS4_GETFH_NEEDSOP is set, then use nfs4_start_fop and nfs4_end_fop
1669  * to coordinate with recovery.  Otherwise, the caller is assumed to be
1670  * the recovery thread or have already done a start_fop.
1671  *
1672  * Errors are returned by the nfs4_error_t parameter.
1673  */
1674 static void
1675 nfs4getfh_otw(struct mntinfo4 *mi, servinfo4_t *svp, vtype_t *vtp,
1676     int flags, cred_t *cr, nfs4_error_t *ep)
1677 {
1678 	COMPOUND4args_clnt args;
1679 	COMPOUND4res_clnt res;
1680 	int doqueue = 1;
1681 	nfs_argop4 *argop;
1682 	nfs_resop4 *resop;
1683 	nfs4_ga_res_t *garp;
1684 	int num_argops;
1685 	lookup4_param_t lookuparg;
1686 	nfs_fh4 *tmpfhp;
1687 	nfs_fh4 *resfhp;
1688 	bool_t needrecov = FALSE;
1689 	nfs4_recov_state_t recov_state;
1690 	int llndx;
1691 	int nthcomp;
1692 	int recovery = !(flags & NFS4_GETFH_NEEDSOP);
1693 
1694 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1695 	ASSERT(svp->sv_path != NULL);
1696 	if (svp->sv_path[0] == '\0') {
1697 		nfs_rw_exit(&svp->sv_lock);
1698 		nfs4_error_init(ep, EINVAL);
1699 		return;
1700 	}
1701 	nfs_rw_exit(&svp->sv_lock);
1702 
1703 	recov_state.rs_flags = 0;
1704 	recov_state.rs_num_retry_despite_err = 0;
1705 
1706 recov_retry:
1707 	if (mi->mi_vfs_referral_loop_cnt >= NFS4_REFERRAL_LOOP_MAX) {
1708 		DTRACE_PROBE3(nfs4clnt__debug__referral__loop, mntinfo4 *,
1709 		    mi, servinfo4_t *, svp, char *, "nfs4getfh_otw");
1710 		nfs4_error_init(ep, EINVAL);
1711 		return;
1712 	}
1713 	nfs4_error_zinit(ep);
1714 
1715 	if (!recovery) {
1716 		ep->error = nfs4_start_fop(mi, NULL, NULL, OH_MOUNT,
1717 		    &recov_state, NULL);
1718 
1719 		/*
1720 		 * If recovery has been started and this request as
1721 		 * initiated by a mount, then we must wait for recovery
1722 		 * to finish before proceeding, otherwise, the error
1723 		 * cleanup would remove data structures needed by the
1724 		 * recovery thread.
1725 		 */
1726 		if (ep->error) {
1727 			mutex_enter(&mi->mi_lock);
1728 			if (mi->mi_flags & MI4_MOUNTING) {
1729 				mi->mi_flags |= MI4_RECOV_FAIL;
1730 				mi->mi_error = EIO;
1731 
1732 				NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
1733 				    "nfs4getfh_otw: waiting 4 recovery\n"));
1734 
1735 				while (mi->mi_flags & MI4_RECOV_ACTIV)
1736 					cv_wait(&mi->mi_failover_cv,
1737 					    &mi->mi_lock);
1738 			}
1739 			mutex_exit(&mi->mi_lock);
1740 			return;
1741 		}
1742 
1743 		/*
1744 		 * If the client does not specify a specific flavor to use
1745 		 * and has not gotten a secinfo list from the server yet,
1746 		 * retrieve the secinfo list from the server and use a
1747 		 * flavor from the list to mount.
1748 		 *
1749 		 * If fail to get the secinfo list from the server, then
1750 		 * try the default flavor.
1751 		 */
1752 		if ((svp->sv_flags & SV4_TRYSECDEFAULT) &&
1753 		    svp->sv_secinfo == NULL) {
1754 			(void) nfs4_secinfo_path(mi, cr, FALSE);
1755 		}
1756 	}
1757 
1758 	if (recovery)
1759 		args.ctag = TAG_REMAP_MOUNT;
1760 	else
1761 		args.ctag = TAG_MOUNT;
1762 
1763 	lookuparg.l4_getattrs = LKP4_ALL_ATTRIBUTES;
1764 	lookuparg.argsp = &args;
1765 	lookuparg.resp = &res;
1766 	lookuparg.header_len = 2;	/* Putrootfh, getfh */
1767 	lookuparg.trailer_len = 0;
1768 	lookuparg.ga_bits = FATTR4_FSINFO_MASK;
1769 	lookuparg.mi = mi;
1770 
1771 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1772 	ASSERT(svp->sv_path != NULL);
1773 	llndx = nfs4lookup_setup(svp->sv_path, &lookuparg, 0);
1774 	nfs_rw_exit(&svp->sv_lock);
1775 
1776 	argop = args.array;
1777 	num_argops = args.array_len;
1778 
1779 	/* choose public or root filehandle */
1780 	if (flags & NFS4_GETFH_PUBLIC)
1781 		argop[0].argop = OP_PUTPUBFH;
1782 	else
1783 		argop[0].argop = OP_PUTROOTFH;
1784 
1785 	/* get fh */
1786 	argop[1].argop = OP_GETFH;
1787 
1788 	NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE,
1789 	    "nfs4getfh_otw: %s call, mi 0x%p",
1790 	    needrecov ? "recov" : "first", (void *)mi));
1791 
1792 	rfs4call(mi, &args, &res, cr, &doqueue, RFSCALL_SOFT, ep);
1793 
1794 	needrecov = nfs4_needs_recovery(ep, FALSE, mi->mi_vfsp);
1795 
1796 	if (needrecov) {
1797 		bool_t abort;
1798 
1799 		if (recovery) {
1800 			nfs4args_lookup_free(argop, num_argops);
1801 			kmem_free(argop,
1802 			    lookuparg.arglen * sizeof (nfs_argop4));
1803 			if (!ep->error)
1804 				xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1805 			return;
1806 		}
1807 
1808 		NFS4_DEBUG(nfs4_client_recov_debug,
1809 		    (CE_NOTE, "nfs4getfh_otw: initiating recovery\n"));
1810 
1811 		abort = nfs4_start_recovery(ep, mi, NULL,
1812 		    NULL, NULL, NULL, OP_GETFH, NULL, NULL, NULL);
1813 		if (!ep->error) {
1814 			ep->error = geterrno4(res.status);
1815 			xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1816 		}
1817 		nfs4args_lookup_free(argop, num_argops);
1818 		kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4));
1819 		nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state, needrecov);
1820 		/* have another go? */
1821 		if (abort == FALSE)
1822 			goto recov_retry;
1823 		return;
1824 	}
1825 
1826 	/*
1827 	 * No recovery, but check if error is set.
1828 	 */
1829 	if (ep->error)  {
1830 		nfs4args_lookup_free(argop, num_argops);
1831 		kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4));
1832 		if (!recovery)
1833 			nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state,
1834 			    needrecov);
1835 		return;
1836 	}
1837 
1838 is_link_err:
1839 
1840 	/* for non-recovery errors */
1841 	if (res.status && res.status != NFS4ERR_SYMLINK &&
1842 	    res.status != NFS4ERR_MOVED) {
1843 		if (!recovery) {
1844 			nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state,
1845 			    needrecov);
1846 		}
1847 		nfs4args_lookup_free(argop, num_argops);
1848 		kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4));
1849 		xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1850 		return;
1851 	}
1852 
1853 	/*
1854 	 * If any intermediate component in the path is a symbolic link,
1855 	 * resolve the symlink, then try mount again using the new path.
1856 	 */
1857 	if (res.status == NFS4ERR_SYMLINK || res.status == NFS4ERR_MOVED) {
1858 		int where;
1859 
1860 		/*
1861 		 * Need to call nfs4_end_op before resolve_sympath to avoid
1862 		 * potential nfs4_start_op deadlock.
1863 		 */
1864 		if (!recovery)
1865 			nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state,
1866 			    needrecov);
1867 
1868 		/*
1869 		 * This must be from OP_LOOKUP failure. The (cfh) for this
1870 		 * OP_LOOKUP is a symlink node. Found out where the
1871 		 * OP_GETFH is for the (cfh) that is a symlink node.
1872 		 *
1873 		 * Example:
1874 		 * (mount) PUTROOTFH, GETFH, LOOKUP comp1, GETFH, GETATTR,
1875 		 * LOOKUP comp2, GETFH, GETATTR, LOOKUP comp3, GETFH, GETATTR
1876 		 *
1877 		 * LOOKUP comp3 fails with SYMLINK because comp2 is a symlink.
1878 		 * In this case, where = 7, nthcomp = 2.
1879 		 */
1880 		where = res.array_len - 2;
1881 		ASSERT(where > 0);
1882 
1883 		if (res.status == NFS4ERR_SYMLINK) {
1884 
1885 			resop = &res.array[where - 1];
1886 			ASSERT(resop->resop == OP_GETFH);
1887 			tmpfhp = &resop->nfs_resop4_u.opgetfh.object;
1888 			nthcomp = res.array_len/3 - 1;
1889 			ep->error = resolve_sympath(mi, svp, nthcomp,
1890 			    tmpfhp, cr, flags);
1891 
1892 		} else if (res.status == NFS4ERR_MOVED) {
1893 
1894 			resop = &res.array[where - 2];
1895 			ASSERT(resop->resop == OP_GETFH);
1896 			tmpfhp = &resop->nfs_resop4_u.opgetfh.object;
1897 			nthcomp = res.array_len/3 - 1;
1898 			ep->error = resolve_referral(mi, svp, cr, nthcomp,
1899 			    tmpfhp);
1900 		}
1901 
1902 		nfs4args_lookup_free(argop, num_argops);
1903 		kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4));
1904 		xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1905 
1906 		if (ep->error)
1907 			return;
1908 
1909 		goto recov_retry;
1910 	}
1911 
1912 	/* getfh */
1913 	resop = &res.array[res.array_len - 2];
1914 	ASSERT(resop->resop == OP_GETFH);
1915 	resfhp = &resop->nfs_resop4_u.opgetfh.object;
1916 
1917 	/* getattr fsinfo res */
1918 	resop++;
1919 	garp = &resop->nfs_resop4_u.opgetattr.ga_res;
1920 
1921 	*vtp = garp->n4g_va.va_type;
1922 
1923 	mi->mi_fh_expire_type = garp->n4g_ext_res->n4g_fet;
1924 
1925 	mutex_enter(&mi->mi_lock);
1926 	if (garp->n4g_ext_res->n4g_pc4.pc4_link_support)
1927 		mi->mi_flags |= MI4_LINK;
1928 	if (garp->n4g_ext_res->n4g_pc4.pc4_symlink_support)
1929 		mi->mi_flags |= MI4_SYMLINK;
1930 	if (garp->n4g_ext_res->n4g_suppattrs & FATTR4_ACL_MASK)
1931 		mi->mi_flags |= MI4_ACL;
1932 	mutex_exit(&mi->mi_lock);
1933 
1934 	if (garp->n4g_ext_res->n4g_maxread == 0)
1935 		mi->mi_tsize =
1936 		    MIN(MAXBSIZE, mi->mi_tsize);
1937 	else
1938 		mi->mi_tsize =
1939 		    MIN(garp->n4g_ext_res->n4g_maxread,
1940 		    mi->mi_tsize);
1941 
1942 	if (garp->n4g_ext_res->n4g_maxwrite == 0)
1943 		mi->mi_stsize =
1944 		    MIN(MAXBSIZE, mi->mi_stsize);
1945 	else
1946 		mi->mi_stsize =
1947 		    MIN(garp->n4g_ext_res->n4g_maxwrite,
1948 		    mi->mi_stsize);
1949 
1950 	if (garp->n4g_ext_res->n4g_maxfilesize != 0)
1951 		mi->mi_maxfilesize =
1952 		    MIN(garp->n4g_ext_res->n4g_maxfilesize,
1953 		    mi->mi_maxfilesize);
1954 
1955 	/*
1956 	 * If the final component is a a symbolic link, resolve the symlink,
1957 	 * then try mount again using the new path.
1958 	 *
1959 	 * Assume no symbolic link for root filesysm "/".
1960 	 */
1961 	if (*vtp == VLNK) {
1962 		/*
1963 		 * nthcomp is the total result length minus
1964 		 * the 1st 2 OPs (PUTROOTFH, GETFH),
1965 		 * then divided by 3 (LOOKUP,GETFH,GETATTR)
1966 		 *
1967 		 * e.g. PUTROOTFH GETFH LOOKUP 1st-comp GETFH GETATTR
1968 		 *	LOOKUP 2nd-comp GETFH GETATTR
1969 		 *
1970 		 *	(8 - 2)/3 = 2
1971 		 */
1972 		nthcomp = (res.array_len - 2)/3;
1973 
1974 		/*
1975 		 * Need to call nfs4_end_op before resolve_sympath to avoid
1976 		 * potential nfs4_start_op deadlock. See RFE 4777612.
1977 		 */
1978 		if (!recovery)
1979 			nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state,
1980 			    needrecov);
1981 
1982 		ep->error = resolve_sympath(mi, svp, nthcomp, resfhp, cr,
1983 		    flags);
1984 
1985 		nfs4args_lookup_free(argop, num_argops);
1986 		kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4));
1987 		xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1988 
1989 		if (ep->error)
1990 			return;
1991 
1992 		goto recov_retry;
1993 	}
1994 
1995 	/*
1996 	 * We need to figure out where in the compound the getfh
1997 	 * for the parent directory is. If the object to be mounted is
1998 	 * the root, then there is no lookup at all:
1999 	 * PUTROOTFH, GETFH.
2000 	 * If the object to be mounted is in the root, then the compound is:
2001 	 * PUTROOTFH, GETFH, LOOKUP, GETFH, GETATTR.
2002 	 * In either of these cases, the index of the GETFH is 1.
2003 	 * If it is not at the root, then it's something like:
2004 	 * PUTROOTFH, GETFH, LOOKUP, GETFH, GETATTR,
2005 	 * LOOKUP, GETFH, GETATTR
2006 	 * In this case, the index is llndx (last lookup index) - 2.
2007 	 */
2008 	if (llndx == -1 || llndx == 2)
2009 		resop = &res.array[1];
2010 	else {
2011 		ASSERT(llndx > 2);
2012 		resop = &res.array[llndx-2];
2013 	}
2014 
2015 	ASSERT(resop->resop == OP_GETFH);
2016 	tmpfhp = &resop->nfs_resop4_u.opgetfh.object;
2017 
2018 	/* save the filehandles for the replica */
2019 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
2020 	ASSERT(tmpfhp->nfs_fh4_len <= NFS4_FHSIZE);
2021 	svp->sv_pfhandle.fh_len = tmpfhp->nfs_fh4_len;
2022 	bcopy(tmpfhp->nfs_fh4_val, svp->sv_pfhandle.fh_buf,
2023 	    tmpfhp->nfs_fh4_len);
2024 	ASSERT(resfhp->nfs_fh4_len <= NFS4_FHSIZE);
2025 	svp->sv_fhandle.fh_len = resfhp->nfs_fh4_len;
2026 	bcopy(resfhp->nfs_fh4_val, svp->sv_fhandle.fh_buf, resfhp->nfs_fh4_len);
2027 
2028 	/* initialize fsid and supp_attrs for server fs */
2029 	svp->sv_fsid = garp->n4g_fsid;
2030 	svp->sv_supp_attrs =
2031 	    garp->n4g_ext_res->n4g_suppattrs | FATTR4_MANDATTR_MASK;
2032 
2033 	nfs_rw_exit(&svp->sv_lock);
2034 	nfs4args_lookup_free(argop, num_argops);
2035 	kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4));
2036 	xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2037 	if (!recovery)
2038 		nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state, needrecov);
2039 }
2040 
2041 /*
2042  * Save a copy of Servinfo4_t structure.
2043  * We might need when there is a failure in getting file handle
2044  * in case of a referral to replace servinfo4 struct and try again.
2045  */
2046 static struct servinfo4 *
2047 copy_svp(servinfo4_t *nsvp)
2048 {
2049 	servinfo4_t *svp = NULL;
2050 	struct knetconfig *sknconf, *tknconf;
2051 	struct netbuf *saddr, *taddr;
2052 
2053 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
2054 	nfs_rw_init(&svp->sv_lock, NULL, RW_DEFAULT, NULL);
2055 	svp->sv_flags = nsvp->sv_flags;
2056 	svp->sv_fsid = nsvp->sv_fsid;
2057 	svp->sv_hostnamelen = nsvp->sv_hostnamelen;
2058 	svp->sv_pathlen = nsvp->sv_pathlen;
2059 	svp->sv_supp_attrs = nsvp->sv_supp_attrs;
2060 
2061 	svp->sv_path = kmem_alloc(svp->sv_pathlen, KM_SLEEP);
2062 	svp->sv_hostname = kmem_alloc(svp->sv_hostnamelen, KM_SLEEP);
2063 	bcopy(nsvp->sv_hostname, svp->sv_hostname, svp->sv_hostnamelen);
2064 	bcopy(nsvp->sv_path, svp->sv_path, svp->sv_pathlen);
2065 
2066 	saddr = &nsvp->sv_addr;
2067 	taddr = &svp->sv_addr;
2068 	taddr->maxlen = saddr->maxlen;
2069 	taddr->len = saddr->len;
2070 	if (saddr->len > 0) {
2071 		taddr->buf = kmem_zalloc(saddr->maxlen, KM_SLEEP);
2072 		bcopy(saddr->buf, taddr->buf, saddr->len);
2073 	}
2074 
2075 	svp->sv_knconf = kmem_zalloc(sizeof (struct knetconfig), KM_SLEEP);
2076 	sknconf = nsvp->sv_knconf;
2077 	tknconf = svp->sv_knconf;
2078 	tknconf->knc_semantics = sknconf->knc_semantics;
2079 	tknconf->knc_rdev = sknconf->knc_rdev;
2080 	if (sknconf->knc_proto != NULL) {
2081 		tknconf->knc_proto = kmem_zalloc(KNC_STRSIZE, KM_SLEEP);
2082 		bcopy(sknconf->knc_proto, (char *)tknconf->knc_proto,
2083 		    KNC_STRSIZE);
2084 	}
2085 	if (sknconf->knc_protofmly != NULL) {
2086 		tknconf->knc_protofmly = kmem_zalloc(KNC_STRSIZE, KM_SLEEP);
2087 		bcopy(sknconf->knc_protofmly, (char *)tknconf->knc_protofmly,
2088 		    KNC_STRSIZE);
2089 	}
2090 
2091 	if (nsvp->sv_origknconf != NULL) {
2092 		svp->sv_origknconf = kmem_zalloc(sizeof (struct knetconfig),
2093 		    KM_SLEEP);
2094 		sknconf = nsvp->sv_origknconf;
2095 		tknconf = svp->sv_origknconf;
2096 		tknconf->knc_semantics = sknconf->knc_semantics;
2097 		tknconf->knc_rdev = sknconf->knc_rdev;
2098 		if (sknconf->knc_proto != NULL) {
2099 			tknconf->knc_proto = kmem_zalloc(KNC_STRSIZE, KM_SLEEP);
2100 			bcopy(sknconf->knc_proto, (char *)tknconf->knc_proto,
2101 			    KNC_STRSIZE);
2102 		}
2103 		if (sknconf->knc_protofmly != NULL) {
2104 			tknconf->knc_protofmly = kmem_zalloc(KNC_STRSIZE,
2105 			    KM_SLEEP);
2106 			bcopy(sknconf->knc_protofmly,
2107 			    (char *)tknconf->knc_protofmly, KNC_STRSIZE);
2108 		}
2109 	}
2110 
2111 	svp->sv_secdata = copy_sec_data(nsvp->sv_secdata);
2112 	svp->sv_dhsec = copy_sec_data(svp->sv_dhsec);
2113 	/*
2114 	 * Rest of the security information is not copied as they are built
2115 	 * with the information available from secdata and dhsec.
2116 	 */
2117 	svp->sv_next = NULL;
2118 
2119 	return (svp);
2120 }
2121 
2122 servinfo4_t *
2123 restore_svp(mntinfo4_t *mi, servinfo4_t *svp, servinfo4_t *origsvp)
2124 {
2125 	servinfo4_t *srvnext, *tmpsrv;
2126 
2127 	if (strcmp(svp->sv_hostname, origsvp->sv_hostname) != 0) {
2128 		/*
2129 		 * Since the hostname changed, we must be dealing
2130 		 * with a referral, and the lookup failed.  We will
2131 		 * restore the whole servinfo4_t to what it was before.
2132 		 */
2133 		srvnext = svp->sv_next;
2134 		svp->sv_next = NULL;
2135 		tmpsrv = copy_svp(origsvp);
2136 		sv4_free(svp);
2137 		svp = tmpsrv;
2138 		svp->sv_next = srvnext;
2139 		mutex_enter(&mi->mi_lock);
2140 		mi->mi_servers = svp;
2141 		mi->mi_curr_serv = svp;
2142 		mutex_exit(&mi->mi_lock);
2143 
2144 	} else if (origsvp->sv_pathlen != svp->sv_pathlen) {
2145 
2146 		/*
2147 		 * For symlink case: restore original path because
2148 		 * it might have contained symlinks that were
2149 		 * expanded by nfsgetfh_otw before the failure occurred.
2150 		 */
2151 		(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
2152 		kmem_free(svp->sv_path, svp->sv_pathlen);
2153 		svp->sv_path =
2154 		    kmem_alloc(origsvp->sv_pathlen, KM_SLEEP);
2155 		svp->sv_pathlen = origsvp->sv_pathlen;
2156 		bcopy(origsvp->sv_path, svp->sv_path,
2157 		    origsvp->sv_pathlen);
2158 		nfs_rw_exit(&svp->sv_lock);
2159 	}
2160 	return (svp);
2161 }
2162 
2163 static ushort_t nfs4_max_threads = 8;	/* max number of active async threads */
2164 uint_t nfs4_bsize = 32 * 1024;	/* client `block' size */
2165 static uint_t nfs4_async_clusters = 1;	/* # of reqs from each async queue */
2166 static uint_t nfs4_cots_timeo = NFS_COTS_TIMEO;
2167 
2168 /*
2169  * Remap the root filehandle for the given filesystem.
2170  *
2171  * results returned via the nfs4_error_t parameter.
2172  */
2173 void
2174 nfs4_remap_root(mntinfo4_t *mi, nfs4_error_t *ep, int flags)
2175 {
2176 	struct servinfo4 *svp, *origsvp;
2177 	vtype_t vtype;
2178 	nfs_fh4 rootfh;
2179 	int getfh_flags;
2180 	int num_retry;
2181 
2182 	mutex_enter(&mi->mi_lock);
2183 
2184 remap_retry:
2185 	svp = mi->mi_curr_serv;
2186 	getfh_flags =
2187 	    (flags & NFS4_REMAP_NEEDSOP) ? NFS4_GETFH_NEEDSOP : 0;
2188 	getfh_flags |=
2189 	    (mi->mi_flags & MI4_PUBLIC) ? NFS4_GETFH_PUBLIC : 0;
2190 	mutex_exit(&mi->mi_lock);
2191 
2192 	/*
2193 	 * Just in case server path being mounted contains
2194 	 * symlinks and fails w/STALE, save the initial sv_path
2195 	 * so we can redrive the initial mount compound with the
2196 	 * initial sv_path -- not a symlink-expanded version.
2197 	 *
2198 	 * This could only happen if a symlink was expanded
2199 	 * and the expanded mount compound failed stale.  Because
2200 	 * it could be the case that the symlink was removed at
2201 	 * the server (and replaced with another symlink/dir,
2202 	 * we need to use the initial sv_path when attempting
2203 	 * to re-lookup everything and recover.
2204 	 */
2205 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
2206 	origsvp = copy_svp(svp);
2207 	nfs_rw_exit(&svp->sv_lock);
2208 
2209 	num_retry = nfs4_max_mount_retry;
2210 
2211 	do {
2212 		/*
2213 		 * Get the root fh from the server.  Retry nfs4_max_mount_retry
2214 		 * (2) times if it fails with STALE since the recovery
2215 		 * infrastructure doesn't do STALE recovery for components
2216 		 * of the server path to the object being mounted.
2217 		 */
2218 		nfs4getfh_otw(mi, svp, &vtype, getfh_flags, CRED(), ep);
2219 
2220 		if (ep->error == 0 && ep->stat == NFS4_OK)
2221 			break;
2222 
2223 		/*
2224 		 * For some reason, the mount compound failed.  Before
2225 		 * retrying, we need to restore original conditions.
2226 		 */
2227 		svp = restore_svp(mi, svp, origsvp);
2228 
2229 	} while (num_retry-- > 0);
2230 
2231 	sv4_free(origsvp);
2232 
2233 	if (ep->error != 0 || ep->stat != 0) {
2234 		return;
2235 	}
2236 
2237 	if (vtype != VNON && vtype != mi->mi_type) {
2238 		/* shouldn't happen */
2239 		zcmn_err(mi->mi_zone->zone_id, CE_WARN,
2240 		    "nfs4_remap_root: server root vnode type (%d) doesn't "
2241 		    "match mount info (%d)", vtype, mi->mi_type);
2242 	}
2243 
2244 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
2245 	rootfh.nfs_fh4_val = svp->sv_fhandle.fh_buf;
2246 	rootfh.nfs_fh4_len = svp->sv_fhandle.fh_len;
2247 	nfs_rw_exit(&svp->sv_lock);
2248 	sfh4_update(mi->mi_rootfh, &rootfh);
2249 
2250 	/*
2251 	 * It's possible that recovery took place on the filesystem
2252 	 * and the server has been updated between the time we did
2253 	 * the nfs4getfh_otw and now. Re-drive the otw operation
2254 	 * to make sure we have a good fh.
2255 	 */
2256 	mutex_enter(&mi->mi_lock);
2257 	if (mi->mi_curr_serv != svp)
2258 		goto remap_retry;
2259 
2260 	mutex_exit(&mi->mi_lock);
2261 }
2262 
2263 static int
2264 nfs4rootvp(vnode_t **rtvpp, vfs_t *vfsp, struct servinfo4 *svp_head,
2265     int flags, cred_t *cr, zone_t *zone)
2266 {
2267 	vnode_t *rtvp = NULL;
2268 	mntinfo4_t *mi;
2269 	dev_t nfs_dev;
2270 	int error = 0;
2271 	rnode4_t *rp;
2272 	int i, len;
2273 	struct vattr va;
2274 	vtype_t vtype = VNON;
2275 	vtype_t tmp_vtype = VNON;
2276 	struct servinfo4 *firstsvp = NULL, *svp = svp_head;
2277 	nfs4_oo_hash_bucket_t *bucketp;
2278 	nfs_fh4 fh;
2279 	char *droptext = "";
2280 	struct nfs_stats *nfsstatsp;
2281 	nfs4_fname_t *mfname;
2282 	nfs4_error_t e;
2283 	int num_retry, removed;
2284 	cred_t *lcr = NULL, *tcr = cr;
2285 	struct servinfo4 *origsvp;
2286 	char *resource;
2287 
2288 	nfsstatsp = zone_getspecific(nfsstat_zone_key, nfs_zone());
2289 	ASSERT(nfsstatsp != NULL);
2290 
2291 	ASSERT(nfs_zone() == zone);
2292 	ASSERT(crgetref(cr));
2293 
2294 	/*
2295 	 * Create a mount record and link it to the vfs struct.
2296 	 */
2297 	mi = kmem_zalloc(sizeof (*mi), KM_SLEEP);
2298 	mutex_init(&mi->mi_lock, NULL, MUTEX_DEFAULT, NULL);
2299 	nfs_rw_init(&mi->mi_recovlock, NULL, RW_DEFAULT, NULL);
2300 	nfs_rw_init(&mi->mi_rename_lock, NULL, RW_DEFAULT, NULL);
2301 	nfs_rw_init(&mi->mi_fh_lock, NULL, RW_DEFAULT, NULL);
2302 
2303 	if (!(flags & NFSMNT_SOFT))
2304 		mi->mi_flags |= MI4_HARD;
2305 	if ((flags & NFSMNT_NOPRINT))
2306 		mi->mi_flags |= MI4_NOPRINT;
2307 	if (flags & NFSMNT_INT)
2308 		mi->mi_flags |= MI4_INT;
2309 	if (flags & NFSMNT_PUBLIC)
2310 		mi->mi_flags |= MI4_PUBLIC;
2311 	if (flags & NFSMNT_MIRRORMOUNT)
2312 		mi->mi_flags |= MI4_MIRRORMOUNT;
2313 	if (flags & NFSMNT_REFERRAL)
2314 		mi->mi_flags |= MI4_REFERRAL;
2315 	mi->mi_retrans = NFS_RETRIES;
2316 	if (svp->sv_knconf->knc_semantics == NC_TPI_COTS_ORD ||
2317 	    svp->sv_knconf->knc_semantics == NC_TPI_COTS)
2318 		mi->mi_timeo = nfs4_cots_timeo;
2319 	else
2320 		mi->mi_timeo = NFS_TIMEO;
2321 	mi->mi_prog = NFS_PROGRAM;
2322 	mi->mi_vers = NFS_V4;
2323 	mi->mi_rfsnames = rfsnames_v4;
2324 	mi->mi_reqs = nfsstatsp->nfs_stats_v4.rfsreqcnt_ptr;
2325 	cv_init(&mi->mi_failover_cv, NULL, CV_DEFAULT, NULL);
2326 	mi->mi_servers = svp;
2327 	mi->mi_curr_serv = svp;
2328 	mi->mi_acregmin = SEC2HR(ACREGMIN);
2329 	mi->mi_acregmax = SEC2HR(ACREGMAX);
2330 	mi->mi_acdirmin = SEC2HR(ACDIRMIN);
2331 	mi->mi_acdirmax = SEC2HR(ACDIRMAX);
2332 	mi->mi_fh_expire_type = FH4_PERSISTENT;
2333 	mi->mi_clientid_next = NULL;
2334 	mi->mi_clientid_prev = NULL;
2335 	mi->mi_srv = NULL;
2336 	mi->mi_grace_wait = 0;
2337 	mi->mi_error = 0;
2338 	mi->mi_srvsettime = 0;
2339 	mi->mi_srvset_cnt = 0;
2340 
2341 	mi->mi_count = 1;
2342 
2343 	mi->mi_tsize = nfs4_tsize(svp->sv_knconf);
2344 	mi->mi_stsize = mi->mi_tsize;
2345 
2346 	if (flags & NFSMNT_DIRECTIO)
2347 		mi->mi_flags |= MI4_DIRECTIO;
2348 
2349 	mi->mi_flags |= MI4_MOUNTING;
2350 
2351 	/*
2352 	 * Make a vfs struct for nfs.  We do this here instead of below
2353 	 * because rtvp needs a vfs before we can do a getattr on it.
2354 	 *
2355 	 * Assign a unique device id to the mount
2356 	 */
2357 	mutex_enter(&nfs_minor_lock);
2358 	do {
2359 		nfs_minor = (nfs_minor + 1) & MAXMIN32;
2360 		nfs_dev = makedevice(nfs_major, nfs_minor);
2361 	} while (vfs_devismounted(nfs_dev));
2362 	mutex_exit(&nfs_minor_lock);
2363 
2364 	vfsp->vfs_dev = nfs_dev;
2365 	vfs_make_fsid(&vfsp->vfs_fsid, nfs_dev, nfs4fstyp);
2366 	vfsp->vfs_data = (caddr_t)mi;
2367 	vfsp->vfs_fstype = nfsfstyp;
2368 	vfsp->vfs_bsize = nfs4_bsize;
2369 
2370 	/*
2371 	 * Initialize fields used to support async putpage operations.
2372 	 */
2373 	for (i = 0; i < NFS4_ASYNC_TYPES; i++)
2374 		mi->mi_async_clusters[i] = nfs4_async_clusters;
2375 	mi->mi_async_init_clusters = nfs4_async_clusters;
2376 	mi->mi_async_curr[NFS4_ASYNC_QUEUE] =
2377 	    mi->mi_async_curr[NFS4_ASYNC_PGOPS_QUEUE] = &mi->mi_async_reqs[0];
2378 	mi->mi_max_threads = nfs4_max_threads;
2379 	mutex_init(&mi->mi_async_lock, NULL, MUTEX_DEFAULT, NULL);
2380 	cv_init(&mi->mi_async_reqs_cv, NULL, CV_DEFAULT, NULL);
2381 	cv_init(&mi->mi_async_work_cv[NFS4_ASYNC_QUEUE], NULL, CV_DEFAULT,
2382 	    NULL);
2383 	cv_init(&mi->mi_async_work_cv[NFS4_ASYNC_PGOPS_QUEUE], NULL,
2384 	    CV_DEFAULT, NULL);
2385 	cv_init(&mi->mi_async_cv, NULL, CV_DEFAULT, NULL);
2386 	cv_init(&mi->mi_inact_req_cv, NULL, CV_DEFAULT, NULL);
2387 
2388 	mi->mi_vfsp = vfsp;
2389 	mi->mi_zone = zone;
2390 	zone_init_ref(&mi->mi_zone_ref);
2391 	zone_hold_ref(zone, &mi->mi_zone_ref, ZONE_REF_NFSV4);
2392 	nfs4_mi_zonelist_add(mi);
2393 
2394 	/*
2395 	 * Initialize the <open owner/cred> hash table.
2396 	 */
2397 	for (i = 0; i < NFS4_NUM_OO_BUCKETS; i++) {
2398 		bucketp = &(mi->mi_oo_list[i]);
2399 		mutex_init(&bucketp->b_lock, NULL, MUTEX_DEFAULT, NULL);
2400 		list_create(&bucketp->b_oo_hash_list,
2401 		    sizeof (nfs4_open_owner_t),
2402 		    offsetof(nfs4_open_owner_t, oo_hash_node));
2403 	}
2404 
2405 	/*
2406 	 * Initialize the freed open owner list.
2407 	 */
2408 	mi->mi_foo_num = 0;
2409 	mi->mi_foo_max = NFS4_NUM_FREED_OPEN_OWNERS;
2410 	list_create(&mi->mi_foo_list, sizeof (nfs4_open_owner_t),
2411 	    offsetof(nfs4_open_owner_t, oo_foo_node));
2412 
2413 	list_create(&mi->mi_lost_state, sizeof (nfs4_lost_rqst_t),
2414 	    offsetof(nfs4_lost_rqst_t, lr_node));
2415 
2416 	list_create(&mi->mi_bseqid_list, sizeof (nfs4_bseqid_entry_t),
2417 	    offsetof(nfs4_bseqid_entry_t, bs_node));
2418 
2419 	/*
2420 	 * Initialize the msg buffer.
2421 	 */
2422 	list_create(&mi->mi_msg_list, sizeof (nfs4_debug_msg_t),
2423 	    offsetof(nfs4_debug_msg_t, msg_node));
2424 	mi->mi_msg_count = 0;
2425 	mutex_init(&mi->mi_msg_list_lock, NULL, MUTEX_DEFAULT, NULL);
2426 
2427 	/*
2428 	 * Initialize kstats
2429 	 */
2430 	nfs4_mnt_kstat_init(vfsp);
2431 
2432 	/*
2433 	 * Initialize the shared filehandle pool.
2434 	 */
2435 	sfh4_createtab(&mi->mi_filehandles);
2436 
2437 	/*
2438 	 * Save server path we're attempting to mount.
2439 	 */
2440 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
2441 	origsvp = copy_svp(svp);
2442 	nfs_rw_exit(&svp->sv_lock);
2443 
2444 	/*
2445 	 * Make the GETFH call to get root fh for each replica.
2446 	 */
2447 	if (svp_head->sv_next)
2448 		droptext = ", dropping replica";
2449 
2450 	/*
2451 	 * If the uid is set then set the creds for secure mounts
2452 	 * by proxy processes such as automountd.
2453 	 */
2454 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
2455 	if (svp->sv_secdata->uid != 0 &&
2456 	    svp->sv_secdata->rpcflavor == RPCSEC_GSS) {
2457 		lcr = crdup(cr);
2458 		(void) crsetugid(lcr, svp->sv_secdata->uid, crgetgid(cr));
2459 		tcr = lcr;
2460 	}
2461 	nfs_rw_exit(&svp->sv_lock);
2462 	for (svp = svp_head; svp; svp = svp->sv_next) {
2463 		if (nfs4_chkdup_servinfo4(svp_head, svp)) {
2464 			nfs_cmn_err(error, CE_WARN,
2465 			    VERS_MSG "Host %s is a duplicate%s",
2466 			    svp->sv_hostname, droptext);
2467 			(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
2468 			svp->sv_flags |= SV4_NOTINUSE;
2469 			nfs_rw_exit(&svp->sv_lock);
2470 			continue;
2471 		}
2472 		mi->mi_curr_serv = svp;
2473 
2474 		/*
2475 		 * Just in case server path being mounted contains
2476 		 * symlinks and fails w/STALE, save the initial sv_path
2477 		 * so we can redrive the initial mount compound with the
2478 		 * initial sv_path -- not a symlink-expanded version.
2479 		 *
2480 		 * This could only happen if a symlink was expanded
2481 		 * and the expanded mount compound failed stale.  Because
2482 		 * it could be the case that the symlink was removed at
2483 		 * the server (and replaced with another symlink/dir,
2484 		 * we need to use the initial sv_path when attempting
2485 		 * to re-lookup everything and recover.
2486 		 *
2487 		 * Other mount errors should evenutally be handled here also
2488 		 * (NFS4ERR_DELAY, NFS4ERR_RESOURCE).  For now, all mount
2489 		 * failures will result in mount being redriven a few times.
2490 		 */
2491 		num_retry = nfs4_max_mount_retry;
2492 		do {
2493 			nfs4getfh_otw(mi, svp, &tmp_vtype,
2494 			    ((flags & NFSMNT_PUBLIC) ? NFS4_GETFH_PUBLIC : 0) |
2495 			    NFS4_GETFH_NEEDSOP, tcr, &e);
2496 
2497 			if (e.error == 0 && e.stat == NFS4_OK)
2498 				break;
2499 
2500 			/*
2501 			 * For some reason, the mount compound failed.  Before
2502 			 * retrying, we need to restore original conditions.
2503 			 */
2504 			svp = restore_svp(mi, svp, origsvp);
2505 			svp_head = svp;
2506 
2507 		} while (num_retry-- > 0);
2508 		error = e.error ? e.error : geterrno4(e.stat);
2509 		if (error) {
2510 			nfs_cmn_err(error, CE_WARN,
2511 			    VERS_MSG "initial call to %s failed%s: %m",
2512 			    svp->sv_hostname, droptext);
2513 			(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
2514 			svp->sv_flags |= SV4_NOTINUSE;
2515 			nfs_rw_exit(&svp->sv_lock);
2516 			mi->mi_flags &= ~MI4_RECOV_FAIL;
2517 			mi->mi_error = 0;
2518 			continue;
2519 		}
2520 
2521 		if (tmp_vtype == VBAD) {
2522 			zcmn_err(mi->mi_zone->zone_id, CE_WARN,
2523 			    VERS_MSG "%s returned a bad file type for "
2524 			    "root%s", svp->sv_hostname, droptext);
2525 			(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
2526 			svp->sv_flags |= SV4_NOTINUSE;
2527 			nfs_rw_exit(&svp->sv_lock);
2528 			continue;
2529 		}
2530 
2531 		if (vtype == VNON) {
2532 			vtype = tmp_vtype;
2533 		} else if (vtype != tmp_vtype) {
2534 			zcmn_err(mi->mi_zone->zone_id, CE_WARN,
2535 			    VERS_MSG "%s returned a different file type "
2536 			    "for root%s", svp->sv_hostname, droptext);
2537 			(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
2538 			svp->sv_flags |= SV4_NOTINUSE;
2539 			nfs_rw_exit(&svp->sv_lock);
2540 			continue;
2541 		}
2542 		if (firstsvp == NULL)
2543 			firstsvp = svp;
2544 	}
2545 
2546 	if (firstsvp == NULL) {
2547 		if (error == 0)
2548 			error = ENOENT;
2549 		goto bad;
2550 	}
2551 
2552 	mi->mi_curr_serv = svp = firstsvp;
2553 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
2554 	ASSERT((mi->mi_curr_serv->sv_flags & SV4_NOTINUSE) == 0);
2555 	fh.nfs_fh4_len = svp->sv_fhandle.fh_len;
2556 	fh.nfs_fh4_val = svp->sv_fhandle.fh_buf;
2557 	mi->mi_rootfh = sfh4_get(&fh, mi);
2558 	fh.nfs_fh4_len = svp->sv_pfhandle.fh_len;
2559 	fh.nfs_fh4_val = svp->sv_pfhandle.fh_buf;
2560 	mi->mi_srvparentfh = sfh4_get(&fh, mi);
2561 	nfs_rw_exit(&svp->sv_lock);
2562 
2563 	/*
2564 	 * Get the fname for filesystem root.
2565 	 */
2566 	mi->mi_fname = fn_get(NULL, ".", mi->mi_rootfh);
2567 	mfname = mi->mi_fname;
2568 	fn_hold(mfname);
2569 
2570 	/*
2571 	 * Make the root vnode without attributes.
2572 	 */
2573 	rtvp = makenfs4node_by_fh(mi->mi_rootfh, NULL,
2574 	    &mfname, NULL, mi, cr, gethrtime());
2575 	rtvp->v_type = vtype;
2576 
2577 	mi->mi_curread = mi->mi_tsize;
2578 	mi->mi_curwrite = mi->mi_stsize;
2579 
2580 	/*
2581 	 * Start the manager thread responsible for handling async worker
2582 	 * threads.
2583 	 */
2584 	MI4_HOLD(mi);
2585 	VFS_HOLD(vfsp);	/* add reference for thread */
2586 	mi->mi_manager_thread = zthread_create(NULL, 0, nfs4_async_manager,
2587 	    vfsp, 0, minclsyspri);
2588 	ASSERT(mi->mi_manager_thread != NULL);
2589 
2590 	/*
2591 	 * Create the thread that handles over-the-wire calls for
2592 	 * VOP_INACTIVE.
2593 	 * This needs to happen after the manager thread is created.
2594 	 */
2595 	MI4_HOLD(mi);
2596 	mi->mi_inactive_thread = zthread_create(NULL, 0, nfs4_inactive_thread,
2597 	    mi, 0, minclsyspri);
2598 	ASSERT(mi->mi_inactive_thread != NULL);
2599 
2600 	/* If we didn't get a type, get one now */
2601 	if (rtvp->v_type == VNON) {
2602 		va.va_mask = AT_TYPE;
2603 		error = nfs4getattr(rtvp, &va, tcr);
2604 		if (error)
2605 			goto bad;
2606 		rtvp->v_type = va.va_type;
2607 	}
2608 
2609 	mi->mi_type = rtvp->v_type;
2610 
2611 	mutex_enter(&mi->mi_lock);
2612 	mi->mi_flags &= ~MI4_MOUNTING;
2613 	mutex_exit(&mi->mi_lock);
2614 
2615 	/* Update VFS with new server and path info */
2616 	if ((strcmp(svp->sv_hostname, origsvp->sv_hostname) != 0) ||
2617 	    (strcmp(svp->sv_path, origsvp->sv_path) != 0)) {
2618 		len = svp->sv_hostnamelen + svp->sv_pathlen;
2619 		resource = kmem_zalloc(len, KM_SLEEP);
2620 		(void) strcat(resource, svp->sv_hostname);
2621 		(void) strcat(resource, ":");
2622 		(void) strcat(resource, svp->sv_path);
2623 		vfs_setresource(vfsp, resource, 0);
2624 		kmem_free(resource, len);
2625 	}
2626 
2627 	sv4_free(origsvp);
2628 	*rtvpp = rtvp;
2629 	if (lcr != NULL)
2630 		crfree(lcr);
2631 
2632 	return (0);
2633 bad:
2634 	/*
2635 	 * An error occurred somewhere, need to clean up...
2636 	 */
2637 	if (lcr != NULL)
2638 		crfree(lcr);
2639 
2640 	if (rtvp != NULL) {
2641 		/*
2642 		 * We need to release our reference to the root vnode and
2643 		 * destroy the mntinfo4 struct that we just created.
2644 		 */
2645 		rp = VTOR4(rtvp);
2646 		if (rp->r_flags & R4HASHED)
2647 			rp4_rmhash(rp);
2648 		VN_RELE(rtvp);
2649 	}
2650 	nfs4_async_stop(vfsp);
2651 	nfs4_async_manager_stop(vfsp);
2652 	removed = nfs4_mi_zonelist_remove(mi);
2653 	if (removed)
2654 		zone_rele_ref(&mi->mi_zone_ref, ZONE_REF_NFSV4);
2655 
2656 	/*
2657 	 * This releases the initial "hold" of the mi since it will never
2658 	 * be referenced by the vfsp.  Also, when mount returns to vfs.c
2659 	 * with an error, the vfsp will be destroyed, not rele'd.
2660 	 */
2661 	MI4_RELE(mi);
2662 
2663 	if (origsvp != NULL)
2664 		sv4_free(origsvp);
2665 
2666 	*rtvpp = NULL;
2667 	return (error);
2668 }
2669 
2670 /*
2671  * vfs operations
2672  */
2673 static int
2674 nfs4_unmount(vfs_t *vfsp, int flag, cred_t *cr)
2675 {
2676 	mntinfo4_t		*mi;
2677 	ushort_t		omax;
2678 	int			removed;
2679 
2680 	bool_t			must_unlock;
2681 
2682 	nfs4_ephemeral_tree_t	*eph_tree;
2683 
2684 	if (secpolicy_fs_unmount(cr, vfsp) != 0)
2685 		return (EPERM);
2686 
2687 	mi = VFTOMI4(vfsp);
2688 
2689 	if (flag & MS_FORCE) {
2690 		vfsp->vfs_flag |= VFS_UNMOUNTED;
2691 		if (nfs_zone() != mi->mi_zone) {
2692 			/*
2693 			 * If the request is coming from the wrong zone,
2694 			 * we don't want to create any new threads, and
2695 			 * performance is not a concern.  Do everything
2696 			 * inline.
2697 			 */
2698 			NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE,
2699 			    "nfs4_unmount x-zone forced unmount of vfs %p\n",
2700 			    (void *)vfsp));
2701 			nfs4_free_mount(vfsp, flag, cr);
2702 		} else {
2703 			/*
2704 			 * Free data structures asynchronously, to avoid
2705 			 * blocking the current thread (for performance
2706 			 * reasons only).
2707 			 */
2708 			async_free_mount(vfsp, flag, cr);
2709 		}
2710 
2711 		return (0);
2712 	}
2713 
2714 	/*
2715 	 * Wait until all asynchronous putpage operations on
2716 	 * this file system are complete before flushing rnodes
2717 	 * from the cache.
2718 	 */
2719 	omax = mi->mi_max_threads;
2720 	if (nfs4_async_stop_sig(vfsp))
2721 		return (EINTR);
2722 
2723 	r4flush(vfsp, cr);
2724 
2725 	/*
2726 	 * About the only reason that this would fail would be
2727 	 * that the harvester is already busy tearing down this
2728 	 * node. So we fail back to the caller and let them try
2729 	 * again when needed.
2730 	 */
2731 	if (nfs4_ephemeral_umount(mi, flag, cr,
2732 	    &must_unlock, &eph_tree)) {
2733 		ASSERT(must_unlock == FALSE);
2734 		mutex_enter(&mi->mi_async_lock);
2735 		mi->mi_max_threads = omax;
2736 		mutex_exit(&mi->mi_async_lock);
2737 
2738 		return (EBUSY);
2739 	}
2740 
2741 	/*
2742 	 * If there are any active vnodes on this file system,
2743 	 * then the file system is busy and can't be unmounted.
2744 	 */
2745 	if (check_rtable4(vfsp)) {
2746 		nfs4_ephemeral_umount_unlock(&must_unlock, &eph_tree);
2747 
2748 		mutex_enter(&mi->mi_async_lock);
2749 		mi->mi_max_threads = omax;
2750 		mutex_exit(&mi->mi_async_lock);
2751 
2752 		return (EBUSY);
2753 	}
2754 
2755 	/*
2756 	 * The unmount can't fail from now on, so record any
2757 	 * ephemeral changes.
2758 	 */
2759 	nfs4_ephemeral_umount_activate(mi, &must_unlock, &eph_tree);
2760 
2761 	/*
2762 	 * There are no active files that could require over-the-wire
2763 	 * calls to the server, so stop the async manager and the
2764 	 * inactive thread.
2765 	 */
2766 	nfs4_async_manager_stop(vfsp);
2767 
2768 	/*
2769 	 * Destroy all rnodes belonging to this file system from the
2770 	 * rnode hash queues and purge any resources allocated to
2771 	 * them.
2772 	 */
2773 	destroy_rtable4(vfsp, cr);
2774 	vfsp->vfs_flag |= VFS_UNMOUNTED;
2775 
2776 	nfs4_remove_mi_from_server(mi, NULL);
2777 	removed = nfs4_mi_zonelist_remove(mi);
2778 	if (removed)
2779 		zone_rele_ref(&mi->mi_zone_ref, ZONE_REF_NFSV4);
2780 
2781 	return (0);
2782 }
2783 
2784 /*
2785  * find root of nfs
2786  */
2787 static int
2788 nfs4_root(vfs_t *vfsp, vnode_t **vpp)
2789 {
2790 	mntinfo4_t *mi;
2791 	vnode_t *vp;
2792 	nfs4_fname_t *mfname;
2793 	servinfo4_t *svp;
2794 
2795 	mi = VFTOMI4(vfsp);
2796 
2797 	if (nfs_zone() != mi->mi_zone)
2798 		return (EPERM);
2799 
2800 	svp = mi->mi_curr_serv;
2801 	if (svp) {
2802 		(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
2803 		if (svp->sv_flags & SV4_ROOT_STALE) {
2804 			nfs_rw_exit(&svp->sv_lock);
2805 
2806 			(void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0);
2807 			if (svp->sv_flags & SV4_ROOT_STALE) {
2808 				svp->sv_flags &= ~SV4_ROOT_STALE;
2809 				nfs_rw_exit(&svp->sv_lock);
2810 				return (ENOENT);
2811 			}
2812 			nfs_rw_exit(&svp->sv_lock);
2813 		} else
2814 			nfs_rw_exit(&svp->sv_lock);
2815 	}
2816 
2817 	mfname = mi->mi_fname;
2818 	fn_hold(mfname);
2819 	vp = makenfs4node_by_fh(mi->mi_rootfh, NULL, &mfname, NULL,
2820 	    VFTOMI4(vfsp), CRED(), gethrtime());
2821 
2822 	if (VTOR4(vp)->r_flags & R4STALE) {
2823 		VN_RELE(vp);
2824 		return (ENOENT);
2825 	}
2826 
2827 	ASSERT(vp->v_type == VNON || vp->v_type == mi->mi_type);
2828 
2829 	vp->v_type = mi->mi_type;
2830 
2831 	*vpp = vp;
2832 
2833 	return (0);
2834 }
2835 
2836 static int
2837 nfs4_statfs_otw(vnode_t *vp, struct statvfs64 *sbp, cred_t *cr)
2838 {
2839 	int error;
2840 	nfs4_ga_res_t gar;
2841 	nfs4_ga_ext_res_t ger;
2842 
2843 	gar.n4g_ext_res = &ger;
2844 
2845 	if (error = nfs4_attr_otw(vp, TAG_FSINFO, &gar,
2846 	    NFS4_STATFS_ATTR_MASK, cr))
2847 		return (error);
2848 
2849 	*sbp = gar.n4g_ext_res->n4g_sb;
2850 
2851 	return (0);
2852 }
2853 
2854 /*
2855  * Get file system statistics.
2856  */
2857 static int
2858 nfs4_statvfs(vfs_t *vfsp, struct statvfs64 *sbp)
2859 {
2860 	int error;
2861 	vnode_t *vp;
2862 	cred_t *cr;
2863 
2864 	error = nfs4_root(vfsp, &vp);
2865 	if (error)
2866 		return (error);
2867 
2868 	cr = CRED();
2869 
2870 	error = nfs4_statfs_otw(vp, sbp, cr);
2871 	if (!error) {
2872 		(void) strncpy(sbp->f_basetype,
2873 		    vfssw[vfsp->vfs_fstype].vsw_name, FSTYPSZ);
2874 		sbp->f_flag = vf_to_stf(vfsp->vfs_flag);
2875 	} else {
2876 		nfs4_purge_stale_fh(error, vp, cr);
2877 	}
2878 
2879 	VN_RELE(vp);
2880 
2881 	return (error);
2882 }
2883 
2884 static kmutex_t nfs4_syncbusy;
2885 
2886 /*
2887  * Flush dirty nfs files for file system vfsp.
2888  * If vfsp == NULL, all nfs files are flushed.
2889  *
2890  * SYNC_CLOSE in flag is passed to us to
2891  * indicate that we are shutting down and or
2892  * rebooting.
2893  */
2894 static int
2895 nfs4_sync(vfs_t *vfsp, short flag, cred_t *cr)
2896 {
2897 	/*
2898 	 * Cross-zone calls are OK here, since this translates to a
2899 	 * VOP_PUTPAGE(B_ASYNC), which gets picked up by the right zone.
2900 	 */
2901 	if (!(flag & SYNC_ATTR) && mutex_tryenter(&nfs4_syncbusy) != 0) {
2902 		r4flush(vfsp, cr);
2903 		mutex_exit(&nfs4_syncbusy);
2904 	}
2905 
2906 	/*
2907 	 * if SYNC_CLOSE is set then we know that
2908 	 * the system is rebooting, mark the mntinfo
2909 	 * for later examination.
2910 	 */
2911 	if (vfsp && (flag & SYNC_CLOSE)) {
2912 		mntinfo4_t *mi;
2913 
2914 		mi = VFTOMI4(vfsp);
2915 		if (!(mi->mi_flags & MI4_SHUTDOWN)) {
2916 			mutex_enter(&mi->mi_lock);
2917 			mi->mi_flags |= MI4_SHUTDOWN;
2918 			mutex_exit(&mi->mi_lock);
2919 		}
2920 	}
2921 	return (0);
2922 }
2923 
2924 /*
2925  * vget is difficult, if not impossible, to support in v4 because we don't
2926  * know the parent directory or name, which makes it impossible to create a
2927  * useful shadow vnode.  And we need the shadow vnode for things like
2928  * OPEN.
2929  */
2930 
2931 /* ARGSUSED */
2932 /*
2933  * XXX Check nfs4_vget_pseudo() for dependency.
2934  */
2935 static int
2936 nfs4_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
2937 {
2938 	return (EREMOTE);
2939 }
2940 
2941 /*
2942  * nfs4_mountroot get called in the case where we are diskless booting.  All
2943  * we need from here is the ability to get the server info and from there we
2944  * can simply call nfs4_rootvp.
2945  */
2946 /* ARGSUSED */
2947 static int
2948 nfs4_mountroot(vfs_t *vfsp, whymountroot_t why)
2949 {
2950 	vnode_t *rtvp;
2951 	char root_hostname[SYS_NMLN+1];
2952 	struct servinfo4 *svp;
2953 	int error;
2954 	int vfsflags;
2955 	size_t size;
2956 	char *root_path;
2957 	struct pathname pn;
2958 	char *name;
2959 	cred_t *cr;
2960 	mntinfo4_t *mi;
2961 	struct nfs_args args;		/* nfs mount arguments */
2962 	static char token[10];
2963 	nfs4_error_t n4e;
2964 
2965 	bzero(&args, sizeof (args));
2966 
2967 	/* do this BEFORE getfile which causes xid stamps to be initialized */
2968 	clkset(-1L);		/* hack for now - until we get time svc? */
2969 
2970 	if (why == ROOT_REMOUNT) {
2971 		/*
2972 		 * Shouldn't happen.
2973 		 */
2974 		panic("nfs4_mountroot: why == ROOT_REMOUNT");
2975 	}
2976 
2977 	if (why == ROOT_UNMOUNT) {
2978 		/*
2979 		 * Nothing to do for NFS.
2980 		 */
2981 		return (0);
2982 	}
2983 
2984 	/*
2985 	 * why == ROOT_INIT
2986 	 */
2987 
2988 	name = token;
2989 	*name = 0;
2990 	(void) getfsname("root", name, sizeof (token));
2991 
2992 	pn_alloc(&pn);
2993 	root_path = pn.pn_path;
2994 
2995 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
2996 	nfs_rw_init(&svp->sv_lock, NULL, RW_DEFAULT, NULL);
2997 	svp->sv_knconf = kmem_zalloc(sizeof (*svp->sv_knconf), KM_SLEEP);
2998 	svp->sv_knconf->knc_protofmly = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
2999 	svp->sv_knconf->knc_proto = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
3000 
3001 	/*
3002 	 * Get server address
3003 	 * Get the root path
3004 	 * Get server's transport
3005 	 * Get server's hostname
3006 	 * Get options
3007 	 */
3008 	args.addr = &svp->sv_addr;
3009 	(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
3010 	args.fh = (char *)&svp->sv_fhandle;
3011 	args.knconf = svp->sv_knconf;
3012 	args.hostname = root_hostname;
3013 	vfsflags = 0;
3014 	if (error = mount_root(*name ? name : "root", root_path, NFS_V4,
3015 	    &args, &vfsflags)) {
3016 		if (error == EPROTONOSUPPORT)
3017 			nfs_cmn_err(error, CE_WARN, "nfs4_mountroot: "
3018 			    "mount_root failed: server doesn't support NFS V4");
3019 		else
3020 			nfs_cmn_err(error, CE_WARN,
3021 			    "nfs4_mountroot: mount_root failed: %m");
3022 		nfs_rw_exit(&svp->sv_lock);
3023 		sv4_free(svp);
3024 		pn_free(&pn);
3025 		return (error);
3026 	}
3027 	nfs_rw_exit(&svp->sv_lock);
3028 	svp->sv_hostnamelen = (int)(strlen(root_hostname) + 1);
3029 	svp->sv_hostname = kmem_alloc(svp->sv_hostnamelen, KM_SLEEP);
3030 	(void) strcpy(svp->sv_hostname, root_hostname);
3031 
3032 	svp->sv_pathlen = (int)(strlen(root_path) + 1);
3033 	svp->sv_path = kmem_alloc(svp->sv_pathlen, KM_SLEEP);
3034 	(void) strcpy(svp->sv_path, root_path);
3035 
3036 	/*
3037 	 * Force root partition to always be mounted with AUTH_UNIX for now
3038 	 */
3039 	svp->sv_secdata = kmem_alloc(sizeof (*svp->sv_secdata), KM_SLEEP);
3040 	svp->sv_secdata->secmod = AUTH_UNIX;
3041 	svp->sv_secdata->rpcflavor = AUTH_UNIX;
3042 	svp->sv_secdata->data = NULL;
3043 
3044 	cr = crgetcred();
3045 	rtvp = NULL;
3046 
3047 	error = nfs4rootvp(&rtvp, vfsp, svp, args.flags, cr, global_zone);
3048 
3049 	if (error) {
3050 		crfree(cr);
3051 		pn_free(&pn);
3052 		sv4_free(svp);
3053 		return (error);
3054 	}
3055 
3056 	mi = VTOMI4(rtvp);
3057 
3058 	/*
3059 	 * Send client id to the server, if necessary
3060 	 */
3061 	nfs4_error_zinit(&n4e);
3062 	nfs4setclientid(mi, cr, FALSE, &n4e);
3063 	error = n4e.error;
3064 
3065 	crfree(cr);
3066 
3067 	if (error) {
3068 		pn_free(&pn);
3069 		goto errout;
3070 	}
3071 
3072 	error = nfs4_setopts(rtvp, DATAMODEL_NATIVE, &args);
3073 	if (error) {
3074 		nfs_cmn_err(error, CE_WARN,
3075 		    "nfs4_mountroot: invalid root mount options");
3076 		pn_free(&pn);
3077 		goto errout;
3078 	}
3079 
3080 	(void) vfs_lock_wait(vfsp);
3081 	vfs_add(NULL, vfsp, vfsflags);
3082 	vfs_unlock(vfsp);
3083 
3084 	size = strlen(svp->sv_hostname);
3085 	(void) strcpy(rootfs.bo_name, svp->sv_hostname);
3086 	rootfs.bo_name[size] = ':';
3087 	(void) strcpy(&rootfs.bo_name[size + 1], root_path);
3088 
3089 	pn_free(&pn);
3090 
3091 errout:
3092 	if (error) {
3093 		sv4_free(svp);
3094 		nfs4_async_stop(vfsp);
3095 		nfs4_async_manager_stop(vfsp);
3096 	}
3097 
3098 	if (rtvp != NULL)
3099 		VN_RELE(rtvp);
3100 
3101 	return (error);
3102 }
3103 
3104 /*
3105  * Initialization routine for VFS routines.  Should only be called once
3106  */
3107 int
3108 nfs4_vfsinit(void)
3109 {
3110 	mutex_init(&nfs4_syncbusy, NULL, MUTEX_DEFAULT, NULL);
3111 	nfs4setclientid_init();
3112 	nfs4_ephemeral_init();
3113 	return (0);
3114 }
3115 
3116 void
3117 nfs4_vfsfini(void)
3118 {
3119 	nfs4_ephemeral_fini();
3120 	nfs4setclientid_fini();
3121 	mutex_destroy(&nfs4_syncbusy);
3122 }
3123 
3124 void
3125 nfs4_freevfs(vfs_t *vfsp)
3126 {
3127 	mntinfo4_t *mi;
3128 
3129 	/* need to release the initial hold */
3130 	mi = VFTOMI4(vfsp);
3131 
3132 	/*
3133 	 * At this point, we can no longer reference the vfs
3134 	 * and need to inform other holders of the reference
3135 	 * to the mntinfo4_t.
3136 	 */
3137 	mi->mi_vfsp = NULL;
3138 
3139 	MI4_RELE(mi);
3140 }
3141 
3142 /*
3143  * Client side SETCLIENTID and SETCLIENTID_CONFIRM
3144  */
3145 struct nfs4_server nfs4_server_lst =
3146 	{ &nfs4_server_lst, &nfs4_server_lst };
3147 
3148 kmutex_t nfs4_server_lst_lock;
3149 
3150 static void
3151 nfs4setclientid_init(void)
3152 {
3153 	mutex_init(&nfs4_server_lst_lock, NULL, MUTEX_DEFAULT, NULL);
3154 }
3155 
3156 static void
3157 nfs4setclientid_fini(void)
3158 {
3159 	mutex_destroy(&nfs4_server_lst_lock);
3160 }
3161 
3162 int nfs4_retry_sclid_delay = NFS4_RETRY_SCLID_DELAY;
3163 int nfs4_num_sclid_retries = NFS4_NUM_SCLID_RETRIES;
3164 
3165 /*
3166  * Set the clientid for the server for "mi".  No-op if the clientid is
3167  * already set.
3168  *
3169  * The recovery boolean should be set to TRUE if this function was called
3170  * by the recovery code, and FALSE otherwise.  This is used to determine
3171  * if we need to call nfs4_start/end_op as well as grab the mi_recovlock
3172  * for adding a mntinfo4_t to a nfs4_server_t.
3173  *
3174  * Error is returned via 'n4ep'.  If there was a 'n4ep->stat' error, then
3175  * 'n4ep->error' is set to geterrno4(n4ep->stat).
3176  */
3177 void
3178 nfs4setclientid(mntinfo4_t *mi, cred_t *cr, bool_t recovery, nfs4_error_t *n4ep)
3179 {
3180 	struct nfs4_server *np;
3181 	struct servinfo4 *svp = mi->mi_curr_serv;
3182 	nfs4_recov_state_t recov_state;
3183 	int num_retries = 0;
3184 	bool_t retry;
3185 	cred_t *lcr = NULL;
3186 	int retry_inuse = 1; /* only retry once on NFS4ERR_CLID_INUSE */
3187 	time_t lease_time = 0;
3188 
3189 	recov_state.rs_flags = 0;
3190 	recov_state.rs_num_retry_despite_err = 0;
3191 	ASSERT(n4ep != NULL);
3192 
3193 recov_retry:
3194 	retry = FALSE;
3195 	nfs4_error_zinit(n4ep);
3196 	if (!recovery)
3197 		(void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_READER, 0);
3198 
3199 	mutex_enter(&nfs4_server_lst_lock);
3200 	np = servinfo4_to_nfs4_server(svp); /* This locks np if it is found */
3201 	mutex_exit(&nfs4_server_lst_lock);
3202 	if (!np) {
3203 		struct nfs4_server *tnp;
3204 		np = new_nfs4_server(svp, cr);
3205 		mutex_enter(&np->s_lock);
3206 
3207 		mutex_enter(&nfs4_server_lst_lock);
3208 		tnp = servinfo4_to_nfs4_server(svp);
3209 		if (tnp) {
3210 			/*
3211 			 * another thread snuck in and put server on list.
3212 			 * since we aren't adding it to the nfs4_server_list
3213 			 * we need to set the ref count to 0 and destroy it.
3214 			 */
3215 			np->s_refcnt = 0;
3216 			destroy_nfs4_server(np);
3217 			np = tnp;
3218 		} else {
3219 			/*
3220 			 * do not give list a reference until everything
3221 			 * succeeds
3222 			 */
3223 			insque(np, &nfs4_server_lst);
3224 		}
3225 		mutex_exit(&nfs4_server_lst_lock);
3226 	}
3227 	ASSERT(MUTEX_HELD(&np->s_lock));
3228 	/*
3229 	 * If we find the server already has N4S_CLIENTID_SET, then
3230 	 * just return, we've already done SETCLIENTID to that server
3231 	 */
3232 	if (np->s_flags & N4S_CLIENTID_SET) {
3233 		/* add mi to np's mntinfo4_list */
3234 		nfs4_add_mi_to_server(np, mi);
3235 		if (!recovery)
3236 			nfs_rw_exit(&mi->mi_recovlock);
3237 		mutex_exit(&np->s_lock);
3238 		nfs4_server_rele(np);
3239 		return;
3240 	}
3241 	mutex_exit(&np->s_lock);
3242 
3243 
3244 	/*
3245 	 * Drop the mi_recovlock since nfs4_start_op will
3246 	 * acquire it again for us.
3247 	 */
3248 	if (!recovery) {
3249 		nfs_rw_exit(&mi->mi_recovlock);
3250 
3251 		n4ep->error = nfs4_start_op(mi, NULL, NULL, &recov_state);
3252 		if (n4ep->error) {
3253 			nfs4_server_rele(np);
3254 			return;
3255 		}
3256 	}
3257 
3258 	mutex_enter(&np->s_lock);
3259 	while (np->s_flags & N4S_CLIENTID_PEND) {
3260 		if (!cv_wait_sig(&np->s_clientid_pend, &np->s_lock)) {
3261 			mutex_exit(&np->s_lock);
3262 			nfs4_server_rele(np);
3263 			if (!recovery)
3264 				nfs4_end_op(mi, NULL, NULL, &recov_state,
3265 				    recovery);
3266 			n4ep->error = EINTR;
3267 			return;
3268 		}
3269 	}
3270 
3271 	if (np->s_flags & N4S_CLIENTID_SET) {
3272 		/* XXX copied/pasted from above */
3273 		/* add mi to np's mntinfo4_list */
3274 		nfs4_add_mi_to_server(np, mi);
3275 		mutex_exit(&np->s_lock);
3276 		nfs4_server_rele(np);
3277 		if (!recovery)
3278 			nfs4_end_op(mi, NULL, NULL, &recov_state, recovery);
3279 		return;
3280 	}
3281 
3282 	/*
3283 	 * Reset the N4S_CB_PINGED flag. This is used to
3284 	 * indicate if we have received a CB_NULL from the
3285 	 * server. Also we reset the waiter flag.
3286 	 */
3287 	np->s_flags &= ~(N4S_CB_PINGED | N4S_CB_WAITER);
3288 	/* any failure must now clear this flag */
3289 	np->s_flags |= N4S_CLIENTID_PEND;
3290 	mutex_exit(&np->s_lock);
3291 	nfs4setclientid_otw(mi, svp, cr, np, n4ep, &retry_inuse);
3292 
3293 	if (n4ep->error == EACCES) {
3294 		/*
3295 		 * If the uid is set then set the creds for secure mounts
3296 		 * by proxy processes such as automountd.
3297 		 */
3298 		(void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
3299 		if (svp->sv_secdata->uid != 0) {
3300 			lcr = crdup(cr);
3301 			(void) crsetugid(lcr, svp->sv_secdata->uid,
3302 			    crgetgid(cr));
3303 		}
3304 		nfs_rw_exit(&svp->sv_lock);
3305 
3306 		if (lcr != NULL) {
3307 			mutex_enter(&np->s_lock);
3308 			crfree(np->s_cred);
3309 			np->s_cred = lcr;
3310 			mutex_exit(&np->s_lock);
3311 			nfs4setclientid_otw(mi, svp, lcr, np, n4ep,
3312 			    &retry_inuse);
3313 		}
3314 	}
3315 	mutex_enter(&np->s_lock);
3316 	lease_time = np->s_lease_time;
3317 	np->s_flags &= ~N4S_CLIENTID_PEND;
3318 	mutex_exit(&np->s_lock);
3319 
3320 	if (n4ep->error != 0 || n4ep->stat != NFS4_OK) {
3321 		/*
3322 		 * Start recovery if failover is a possibility.  If
3323 		 * invoked by the recovery thread itself, then just
3324 		 * return and let it handle the failover first.  NB:
3325 		 * recovery is not allowed if the mount is in progress
3326 		 * since the infrastructure is not sufficiently setup
3327 		 * to allow it.  Just return the error (after suitable
3328 		 * retries).
3329 		 */
3330 		if (FAILOVER_MOUNT4(mi) && nfs4_try_failover(n4ep)) {
3331 			(void) nfs4_start_recovery(n4ep, mi, NULL,
3332 			    NULL, NULL, NULL, OP_SETCLIENTID, NULL, NULL, NULL);
3333 			/*
3334 			 * Don't retry here, just return and let
3335 			 * recovery take over.
3336 			 */
3337 			if (recovery)
3338 				retry = FALSE;
3339 		} else if (nfs4_rpc_retry_error(n4ep->error) ||
3340 		    n4ep->stat == NFS4ERR_RESOURCE ||
3341 		    n4ep->stat == NFS4ERR_STALE_CLIENTID) {
3342 
3343 			retry = TRUE;
3344 			/*
3345 			 * Always retry if in recovery or once had
3346 			 * contact with the server (but now it's
3347 			 * overloaded).
3348 			 */
3349 			if (recovery == TRUE ||
3350 			    n4ep->error == ETIMEDOUT ||
3351 			    n4ep->error == ECONNRESET)
3352 				num_retries = 0;
3353 		} else if (retry_inuse && n4ep->error == 0 &&
3354 		    n4ep->stat == NFS4ERR_CLID_INUSE) {
3355 			retry = TRUE;
3356 			num_retries = 0;
3357 		}
3358 	} else {
3359 		/*
3360 		 * Since everything succeeded give the list a reference count if
3361 		 * it hasn't been given one by add_new_nfs4_server() or if this
3362 		 * is not a recovery situation in which case it is already on
3363 		 * the list.
3364 		 */
3365 		mutex_enter(&np->s_lock);
3366 		if ((np->s_flags & N4S_INSERTED) == 0) {
3367 			np->s_refcnt++;
3368 			np->s_flags |= N4S_INSERTED;
3369 		}
3370 		mutex_exit(&np->s_lock);
3371 	}
3372 
3373 	if (!recovery)
3374 		nfs4_end_op(mi, NULL, NULL, &recov_state, recovery);
3375 
3376 
3377 	if (retry && num_retries++ < nfs4_num_sclid_retries) {
3378 		if (retry_inuse) {
3379 			delay(SEC_TO_TICK(lease_time + nfs4_retry_sclid_delay));
3380 			retry_inuse = 0;
3381 		} else
3382 			delay(SEC_TO_TICK(nfs4_retry_sclid_delay));
3383 
3384 		nfs4_server_rele(np);
3385 		goto recov_retry;
3386 	}
3387 
3388 
3389 	if (n4ep->error == 0)
3390 		n4ep->error = geterrno4(n4ep->stat);
3391 
3392 	/* broadcast before release in case no other threads are waiting */
3393 	cv_broadcast(&np->s_clientid_pend);
3394 	nfs4_server_rele(np);
3395 }
3396 
3397 int nfs4setclientid_otw_debug = 0;
3398 
3399 /*
3400  * This function handles the recovery of STALE_CLIENTID for SETCLIENTID_CONFRIM,
3401  * but nothing else; the calling function must be designed to handle those
3402  * other errors.
3403  */
3404 static void
3405 nfs4setclientid_otw(mntinfo4_t *mi, struct servinfo4 *svp,  cred_t *cr,
3406     struct nfs4_server *np, nfs4_error_t *ep, int *retry_inusep)
3407 {
3408 	COMPOUND4args_clnt args;
3409 	COMPOUND4res_clnt res;
3410 	nfs_argop4 argop[3];
3411 	SETCLIENTID4args *s_args;
3412 	SETCLIENTID4resok *s_resok;
3413 	int doqueue = 1;
3414 	nfs4_ga_res_t *garp = NULL;
3415 	timespec_t prop_time, after_time;
3416 	verifier4 verf;
3417 	clientid4 tmp_clientid;
3418 
3419 	ASSERT(!MUTEX_HELD(&np->s_lock));
3420 
3421 	args.ctag = TAG_SETCLIENTID;
3422 
3423 	args.array = argop;
3424 	args.array_len = 3;
3425 
3426 	/* PUTROOTFH */
3427 	argop[0].argop = OP_PUTROOTFH;
3428 
3429 	/* GETATTR */
3430 	argop[1].argop = OP_GETATTR;
3431 	argop[1].nfs_argop4_u.opgetattr.attr_request = FATTR4_LEASE_TIME_MASK;
3432 	argop[1].nfs_argop4_u.opgetattr.mi = mi;
3433 
3434 	/* SETCLIENTID */
3435 	argop[2].argop = OP_SETCLIENTID;
3436 
3437 	s_args = &argop[2].nfs_argop4_u.opsetclientid;
3438 
3439 	mutex_enter(&np->s_lock);
3440 
3441 	s_args->client.verifier = np->clidtosend.verifier;
3442 	s_args->client.id_len = np->clidtosend.id_len;
3443 	ASSERT(s_args->client.id_len <= NFS4_OPAQUE_LIMIT);
3444 	s_args->client.id_val = np->clidtosend.id_val;
3445 
3446 	/*
3447 	 * Callback needs to happen on non-RDMA transport
3448 	 * Check if we have saved the original knetconfig
3449 	 * if so, use that instead.
3450 	 */
3451 	if (svp->sv_origknconf != NULL)
3452 		nfs4_cb_args(np, svp->sv_origknconf, s_args);
3453 	else
3454 		nfs4_cb_args(np, svp->sv_knconf, s_args);
3455 
3456 	mutex_exit(&np->s_lock);
3457 
3458 	rfs4call(mi, &args, &res, cr, &doqueue, 0, ep);
3459 
3460 	if (ep->error)
3461 		return;
3462 
3463 	/* getattr lease_time res */
3464 	if ((res.array_len >= 2) &&
3465 	    (res.array[1].nfs_resop4_u.opgetattr.status == NFS4_OK)) {
3466 		garp = &res.array[1].nfs_resop4_u.opgetattr.ga_res;
3467 
3468 #ifndef _LP64
3469 		/*
3470 		 * The 32 bit client cannot handle a lease time greater than
3471 		 * (INT32_MAX/1000000).  This is due to the use of the
3472 		 * lease_time in calls to drv_usectohz() in
3473 		 * nfs4_renew_lease_thread().  The problem is that
3474 		 * drv_usectohz() takes a time_t (which is just a long = 4
3475 		 * bytes) as its parameter.  The lease_time is multiplied by
3476 		 * 1000000 to convert seconds to usecs for the parameter.  If
3477 		 * a number bigger than (INT32_MAX/1000000) is used then we
3478 		 * overflow on the 32bit client.
3479 		 */
3480 		if (garp->n4g_ext_res->n4g_leasetime > (INT32_MAX/1000000)) {
3481 			garp->n4g_ext_res->n4g_leasetime = INT32_MAX/1000000;
3482 		}
3483 #endif
3484 
3485 		mutex_enter(&np->s_lock);
3486 		np->s_lease_time = garp->n4g_ext_res->n4g_leasetime;
3487 
3488 		/*
3489 		 * Keep track of the lease period for the mi's
3490 		 * mi_msg_list.  We need an appropiate time
3491 		 * bound to associate past facts with a current
3492 		 * event.  The lease period is perfect for this.
3493 		 */
3494 		mutex_enter(&mi->mi_msg_list_lock);
3495 		mi->mi_lease_period = np->s_lease_time;
3496 		mutex_exit(&mi->mi_msg_list_lock);
3497 		mutex_exit(&np->s_lock);
3498 	}
3499 
3500 
3501 	if (res.status == NFS4ERR_CLID_INUSE) {
3502 		clientaddr4 *clid_inuse;
3503 
3504 		if (!(*retry_inusep)) {
3505 			clid_inuse = &res.array->nfs_resop4_u.
3506 			    opsetclientid.SETCLIENTID4res_u.client_using;
3507 
3508 			zcmn_err(mi->mi_zone->zone_id, CE_NOTE,
3509 			    "NFS4 mount (SETCLIENTID failed)."
3510 			    "  nfs4_client_id.id is in"
3511 			    "use already by: r_netid<%s> r_addr<%s>",
3512 			    clid_inuse->r_netid, clid_inuse->r_addr);
3513 		}
3514 
3515 		/*
3516 		 * XXX - The client should be more robust in its
3517 		 * handling of clientid in use errors (regen another
3518 		 * clientid and try again?)
3519 		 */
3520 		xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3521 		return;
3522 	}
3523 
3524 	if (res.status) {
3525 		xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3526 		return;
3527 	}
3528 
3529 	s_resok = &res.array[2].nfs_resop4_u.
3530 	    opsetclientid.SETCLIENTID4res_u.resok4;
3531 
3532 	tmp_clientid = s_resok->clientid;
3533 
3534 	verf = s_resok->setclientid_confirm;
3535 
3536 #ifdef	DEBUG
3537 	if (nfs4setclientid_otw_debug) {
3538 		union {
3539 			clientid4	clientid;
3540 			int		foo[2];
3541 		} cid;
3542 
3543 		cid.clientid = s_resok->clientid;
3544 
3545 		zcmn_err(mi->mi_zone->zone_id, CE_NOTE,
3546 		"nfs4setclientid_otw: OK, clientid = %x,%x, "
3547 		"verifier = %" PRIx64 "\n", cid.foo[0], cid.foo[1], verf);
3548 	}
3549 #endif
3550 
3551 	xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3552 
3553 	/* Confirm the client id and get the lease_time attribute */
3554 
3555 	args.ctag = TAG_SETCLIENTID_CF;
3556 
3557 	args.array = argop;
3558 	args.array_len = 1;
3559 
3560 	argop[0].argop = OP_SETCLIENTID_CONFIRM;
3561 
3562 	argop[0].nfs_argop4_u.opsetclientid_confirm.clientid = tmp_clientid;
3563 	argop[0].nfs_argop4_u.opsetclientid_confirm.setclientid_confirm = verf;
3564 
3565 	/* used to figure out RTT for np */
3566 	gethrestime(&prop_time);
3567 
3568 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4setlientid_otw: "
3569 	    "start time: %ld sec %ld nsec", prop_time.tv_sec,
3570 	    prop_time.tv_nsec));
3571 
3572 	rfs4call(mi, &args, &res, cr, &doqueue, 0, ep);
3573 
3574 	gethrestime(&after_time);
3575 	mutex_enter(&np->s_lock);
3576 	np->propagation_delay.tv_sec =
3577 	    MAX(1, after_time.tv_sec - prop_time.tv_sec);
3578 	mutex_exit(&np->s_lock);
3579 
3580 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4setlcientid_otw: "
3581 	    "finish time: %ld sec ", after_time.tv_sec));
3582 
3583 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4setclientid_otw: "
3584 	    "propagation delay set to %ld sec",
3585 	    np->propagation_delay.tv_sec));
3586 
3587 	if (ep->error)
3588 		return;
3589 
3590 	if (res.status == NFS4ERR_CLID_INUSE) {
3591 		clientaddr4 *clid_inuse;
3592 
3593 		if (!(*retry_inusep)) {
3594 			clid_inuse = &res.array->nfs_resop4_u.
3595 			    opsetclientid.SETCLIENTID4res_u.client_using;
3596 
3597 			zcmn_err(mi->mi_zone->zone_id, CE_NOTE,
3598 			    "SETCLIENTID_CONFIRM failed.  "
3599 			    "nfs4_client_id.id is in use already by: "
3600 			    "r_netid<%s> r_addr<%s>",
3601 			    clid_inuse->r_netid, clid_inuse->r_addr);
3602 		}
3603 
3604 		xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3605 		return;
3606 	}
3607 
3608 	if (res.status) {
3609 		xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3610 		return;
3611 	}
3612 
3613 	mutex_enter(&np->s_lock);
3614 	np->clientid = tmp_clientid;
3615 	np->s_flags |= N4S_CLIENTID_SET;
3616 
3617 	/* Add mi to np's mntinfo4 list */
3618 	nfs4_add_mi_to_server(np, mi);
3619 
3620 	if (np->lease_valid == NFS4_LEASE_NOT_STARTED) {
3621 		/*
3622 		 * Start lease management thread.
3623 		 * Keep trying until we succeed.
3624 		 */
3625 
3626 		np->s_refcnt++;		/* pass reference to thread */
3627 		(void) zthread_create(NULL, 0, nfs4_renew_lease_thread, np, 0,
3628 		    minclsyspri);
3629 	}
3630 	mutex_exit(&np->s_lock);
3631 
3632 	xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3633 }
3634 
3635 /*
3636  * Add mi to sp's mntinfo4_list if it isn't already in the list.  Makes
3637  * mi's clientid the same as sp's.
3638  * Assumes sp is locked down.
3639  */
3640 void
3641 nfs4_add_mi_to_server(nfs4_server_t *sp, mntinfo4_t *mi)
3642 {
3643 	mntinfo4_t *tmi;
3644 	int in_list = 0;
3645 
3646 	ASSERT(nfs_rw_lock_held(&mi->mi_recovlock, RW_READER) ||
3647 	    nfs_rw_lock_held(&mi->mi_recovlock, RW_WRITER));
3648 	ASSERT(sp != &nfs4_server_lst);
3649 	ASSERT(MUTEX_HELD(&sp->s_lock));
3650 
3651 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3652 	    "nfs4_add_mi_to_server: add mi %p to sp %p",
3653 	    (void*)mi, (void*)sp));
3654 
3655 	for (tmi = sp->mntinfo4_list;
3656 	    tmi != NULL;
3657 	    tmi = tmi->mi_clientid_next) {
3658 		if (tmi == mi) {
3659 			NFS4_DEBUG(nfs4_client_lease_debug,
3660 			    (CE_NOTE,
3661 			    "nfs4_add_mi_to_server: mi in list"));
3662 			in_list = 1;
3663 		}
3664 	}
3665 
3666 	/*
3667 	 * First put a hold on the mntinfo4's vfsp so that references via
3668 	 * mntinfo4_list will be valid.
3669 	 */
3670 	if (!in_list)
3671 		VFS_HOLD(mi->mi_vfsp);
3672 
3673 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4_add_mi_to_server: "
3674 	    "hold vfs %p for mi: %p", (void*)mi->mi_vfsp, (void*)mi));
3675 
3676 	if (!in_list) {
3677 		if (sp->mntinfo4_list)
3678 			sp->mntinfo4_list->mi_clientid_prev = mi;
3679 		mi->mi_clientid_next = sp->mntinfo4_list;
3680 		mi->mi_srv = sp;
3681 		sp->mntinfo4_list = mi;
3682 		mi->mi_srvsettime = gethrestime_sec();
3683 		mi->mi_srvset_cnt++;
3684 	}
3685 
3686 	/* set mi's clientid to that of sp's for later matching */
3687 	mi->mi_clientid = sp->clientid;
3688 
3689 	/*
3690 	 * Update the clientid for any other mi's belonging to sp.  This
3691 	 * must be done here while we hold sp->s_lock, so that
3692 	 * find_nfs4_server() continues to work.
3693 	 */
3694 
3695 	for (tmi = sp->mntinfo4_list;
3696 	    tmi != NULL;
3697 	    tmi = tmi->mi_clientid_next) {
3698 		if (tmi != mi) {
3699 			tmi->mi_clientid = sp->clientid;
3700 		}
3701 	}
3702 }
3703 
3704 /*
3705  * Remove the mi from sp's mntinfo4_list and release its reference.
3706  * Exception: if mi still has open files, flag it for later removal (when
3707  * all the files are closed).
3708  *
3709  * If this is the last mntinfo4 in sp's list then tell the lease renewal
3710  * thread to exit.
3711  */
3712 static void
3713 nfs4_remove_mi_from_server_nolock(mntinfo4_t *mi, nfs4_server_t *sp)
3714 {
3715 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3716 	    "nfs4_remove_mi_from_server_nolock: remove mi %p from sp %p",
3717 	    (void*)mi, (void*)sp));
3718 
3719 	ASSERT(sp != NULL);
3720 	ASSERT(MUTEX_HELD(&sp->s_lock));
3721 	ASSERT(mi->mi_open_files >= 0);
3722 
3723 	/*
3724 	 * First make sure this mntinfo4 can be taken off of the list,
3725 	 * ie: it doesn't have any open files remaining.
3726 	 */
3727 	if (mi->mi_open_files > 0) {
3728 		NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3729 		    "nfs4_remove_mi_from_server_nolock: don't "
3730 		    "remove mi since it still has files open"));
3731 
3732 		mutex_enter(&mi->mi_lock);
3733 		mi->mi_flags |= MI4_REMOVE_ON_LAST_CLOSE;
3734 		mutex_exit(&mi->mi_lock);
3735 		return;
3736 	}
3737 
3738 	VFS_HOLD(mi->mi_vfsp);
3739 	remove_mi(sp, mi);
3740 	VFS_RELE(mi->mi_vfsp);
3741 
3742 	if (sp->mntinfo4_list == NULL) {
3743 		/* last fs unmounted, kill the thread */
3744 		NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3745 		    "remove_mi_from_nfs4_server_nolock: kill the thread"));
3746 		nfs4_mark_srv_dead(sp);
3747 	}
3748 }
3749 
3750 /*
3751  * Remove mi from sp's mntinfo4_list and release the vfs reference.
3752  */
3753 static void
3754 remove_mi(nfs4_server_t *sp, mntinfo4_t *mi)
3755 {
3756 	ASSERT(MUTEX_HELD(&sp->s_lock));
3757 
3758 	/*
3759 	 * We release a reference, and the caller must still have a
3760 	 * reference.
3761 	 */
3762 	ASSERT(mi->mi_vfsp->vfs_count >= 2);
3763 
3764 	if (mi->mi_clientid_prev) {
3765 		mi->mi_clientid_prev->mi_clientid_next = mi->mi_clientid_next;
3766 	} else {
3767 		/* This is the first mi in sp's mntinfo4_list */
3768 		/*
3769 		 * Make sure the first mntinfo4 in the list is the actual
3770 		 * mntinfo4 passed in.
3771 		 */
3772 		ASSERT(sp->mntinfo4_list == mi);
3773 
3774 		sp->mntinfo4_list = mi->mi_clientid_next;
3775 	}
3776 	if (mi->mi_clientid_next)
3777 		mi->mi_clientid_next->mi_clientid_prev = mi->mi_clientid_prev;
3778 
3779 	/* Now mark the mntinfo4's links as being removed */
3780 	mi->mi_clientid_prev = mi->mi_clientid_next = NULL;
3781 	mi->mi_srv = NULL;
3782 	mi->mi_srvset_cnt++;
3783 
3784 	VFS_RELE(mi->mi_vfsp);
3785 }
3786 
3787 /*
3788  * Free all the entries in sp's mntinfo4_list.
3789  */
3790 static void
3791 remove_all_mi(nfs4_server_t *sp)
3792 {
3793 	mntinfo4_t *mi;
3794 
3795 	ASSERT(MUTEX_HELD(&sp->s_lock));
3796 
3797 	while (sp->mntinfo4_list != NULL) {
3798 		mi = sp->mntinfo4_list;
3799 		/*
3800 		 * Grab a reference in case there is only one left (which
3801 		 * remove_mi() frees).
3802 		 */
3803 		VFS_HOLD(mi->mi_vfsp);
3804 		remove_mi(sp, mi);
3805 		VFS_RELE(mi->mi_vfsp);
3806 	}
3807 }
3808 
3809 /*
3810  * Remove the mi from sp's mntinfo4_list as above, and rele the vfs.
3811  *
3812  * This version can be called with a null nfs4_server_t arg,
3813  * and will either find the right one and handle locking, or
3814  * do nothing because the mi wasn't added to an sp's mntinfo4_list.
3815  */
3816 void
3817 nfs4_remove_mi_from_server(mntinfo4_t *mi, nfs4_server_t *esp)
3818 {
3819 	nfs4_server_t	*sp;
3820 
3821 	if (esp) {
3822 		nfs4_remove_mi_from_server_nolock(mi, esp);
3823 		return;
3824 	}
3825 
3826 	(void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_READER, 0);
3827 	if (sp = find_nfs4_server_all(mi, 1)) {
3828 		nfs4_remove_mi_from_server_nolock(mi, sp);
3829 		mutex_exit(&sp->s_lock);
3830 		nfs4_server_rele(sp);
3831 	}
3832 	nfs_rw_exit(&mi->mi_recovlock);
3833 }
3834 
3835 /*
3836  * Return TRUE if the given server has any non-unmounted filesystems.
3837  */
3838 
3839 bool_t
3840 nfs4_fs_active(nfs4_server_t *sp)
3841 {
3842 	mntinfo4_t *mi;
3843 
3844 	ASSERT(MUTEX_HELD(&sp->s_lock));
3845 
3846 	for (mi = sp->mntinfo4_list; mi != NULL; mi = mi->mi_clientid_next) {
3847 		if (!(mi->mi_vfsp->vfs_flag & VFS_UNMOUNTED))
3848 			return (TRUE);
3849 	}
3850 
3851 	return (FALSE);
3852 }
3853 
3854 /*
3855  * Mark sp as finished and notify any waiters.
3856  */
3857 
3858 void
3859 nfs4_mark_srv_dead(nfs4_server_t *sp)
3860 {
3861 	ASSERT(MUTEX_HELD(&sp->s_lock));
3862 
3863 	sp->s_thread_exit = NFS4_THREAD_EXIT;
3864 	cv_broadcast(&sp->cv_thread_exit);
3865 }
3866 
3867 /*
3868  * Create a new nfs4_server_t structure.
3869  * Returns new node unlocked and not in list, but with a reference count of
3870  * 1.
3871  */
3872 struct nfs4_server *
3873 new_nfs4_server(struct servinfo4 *svp, cred_t *cr)
3874 {
3875 	struct nfs4_server *np;
3876 	timespec_t tt;
3877 	union {
3878 		struct {
3879 			uint32_t sec;
3880 			uint32_t subsec;
3881 		} un_curtime;
3882 		verifier4	un_verifier;
3883 	} nfs4clientid_verifier;
3884 	/*
3885 	 * We change this ID string carefully and with the Solaris
3886 	 * NFS server behaviour in mind.  "+referrals" indicates
3887 	 * a client that can handle an NFSv4 referral.
3888 	 */
3889 	char id_val[] = "Solaris: %s, NFSv4 kernel client +referrals";
3890 	int len;
3891 
3892 	np = kmem_zalloc(sizeof (struct nfs4_server), KM_SLEEP);
3893 	np->saddr.len = svp->sv_addr.len;
3894 	np->saddr.maxlen = svp->sv_addr.maxlen;
3895 	np->saddr.buf = kmem_alloc(svp->sv_addr.maxlen, KM_SLEEP);
3896 	bcopy(svp->sv_addr.buf, np->saddr.buf, svp->sv_addr.len);
3897 	np->s_refcnt = 1;
3898 
3899 	/*
3900 	 * Build the nfs_client_id4 for this server mount.  Ensure
3901 	 * the verifier is useful and that the identification is
3902 	 * somehow based on the server's address for the case of
3903 	 * multi-homed servers.
3904 	 */
3905 	nfs4clientid_verifier.un_verifier = 0;
3906 	gethrestime(&tt);
3907 	nfs4clientid_verifier.un_curtime.sec = (uint32_t)tt.tv_sec;
3908 	nfs4clientid_verifier.un_curtime.subsec = (uint32_t)tt.tv_nsec;
3909 	np->clidtosend.verifier = nfs4clientid_verifier.un_verifier;
3910 
3911 	/*
3912 	 * calculate the length of the opaque identifier.  Subtract 2
3913 	 * for the "%s" and add the traditional +1 for null
3914 	 * termination.
3915 	 */
3916 	len = strlen(id_val) - 2 + strlen(uts_nodename()) + 1;
3917 	np->clidtosend.id_len = len + np->saddr.maxlen;
3918 
3919 	np->clidtosend.id_val = kmem_alloc(np->clidtosend.id_len, KM_SLEEP);
3920 	(void) sprintf(np->clidtosend.id_val, id_val, uts_nodename());
3921 	bcopy(np->saddr.buf, &np->clidtosend.id_val[len], np->saddr.len);
3922 
3923 	np->s_flags = 0;
3924 	np->mntinfo4_list = NULL;
3925 	/* save cred for issuing rfs4calls inside the renew thread */
3926 	crhold(cr);
3927 	np->s_cred = cr;
3928 	cv_init(&np->cv_thread_exit, NULL, CV_DEFAULT, NULL);
3929 	mutex_init(&np->s_lock, NULL, MUTEX_DEFAULT, NULL);
3930 	nfs_rw_init(&np->s_recovlock, NULL, RW_DEFAULT, NULL);
3931 	list_create(&np->s_deleg_list, sizeof (rnode4_t),
3932 	    offsetof(rnode4_t, r_deleg_link));
3933 	np->s_thread_exit = 0;
3934 	np->state_ref_count = 0;
3935 	np->lease_valid = NFS4_LEASE_NOT_STARTED;
3936 	cv_init(&np->s_cv_otw_count, NULL, CV_DEFAULT, NULL);
3937 	cv_init(&np->s_clientid_pend, NULL, CV_DEFAULT, NULL);
3938 	np->s_otw_call_count = 0;
3939 	cv_init(&np->wait_cb_null, NULL, CV_DEFAULT, NULL);
3940 	np->zoneid = getzoneid();
3941 	np->zone_globals = nfs4_get_callback_globals();
3942 	ASSERT(np->zone_globals != NULL);
3943 	return (np);
3944 }
3945 
3946 /*
3947  * Create a new nfs4_server_t structure and add it to the list.
3948  * Returns new node locked; reference must eventually be freed.
3949  */
3950 static struct nfs4_server *
3951 add_new_nfs4_server(struct servinfo4 *svp, cred_t *cr)
3952 {
3953 	nfs4_server_t *sp;
3954 
3955 	ASSERT(MUTEX_HELD(&nfs4_server_lst_lock));
3956 	sp = new_nfs4_server(svp, cr);
3957 	mutex_enter(&sp->s_lock);
3958 	insque(sp, &nfs4_server_lst);
3959 	sp->s_refcnt++;			/* list gets a reference */
3960 	sp->s_flags |= N4S_INSERTED;
3961 	sp->clientid = 0;
3962 	return (sp);
3963 }
3964 
3965 int nfs4_server_t_debug = 0;
3966 
3967 #ifdef lint
3968 extern void
3969 dumpnfs4slist(char *, mntinfo4_t *, clientid4, servinfo4_t *);
3970 #endif
3971 
3972 #ifndef lint
3973 #ifdef DEBUG
3974 void
3975 dumpnfs4slist(char *txt, mntinfo4_t *mi, clientid4 clientid, servinfo4_t *srv_p)
3976 {
3977 	int hash16(void *p, int len);
3978 	nfs4_server_t *np;
3979 
3980 	NFS4_DEBUG(nfs4_server_t_debug, (CE_NOTE,
3981 	    "dumping nfs4_server_t list in %s", txt));
3982 	NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT,
3983 	    "mi 0x%p, want clientid %llx, addr %d/%04X",
3984 	    mi, (longlong_t)clientid, srv_p->sv_addr.len,
3985 	    hash16((void *)srv_p->sv_addr.buf, srv_p->sv_addr.len)));
3986 	for (np = nfs4_server_lst.forw; np != &nfs4_server_lst;
3987 	    np = np->forw) {
3988 		NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT,
3989 		    "node 0x%p,    clientid %llx, addr %d/%04X, cnt %d",
3990 		    np, (longlong_t)np->clientid, np->saddr.len,
3991 		    hash16((void *)np->saddr.buf, np->saddr.len),
3992 		    np->state_ref_count));
3993 		if (np->saddr.len == srv_p->sv_addr.len &&
3994 		    bcmp(np->saddr.buf, srv_p->sv_addr.buf,
3995 		    np->saddr.len) == 0)
3996 			NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT,
3997 			    " - address matches"));
3998 		if (np->clientid == clientid || np->clientid == 0)
3999 			NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT,
4000 			    " - clientid matches"));
4001 		if (np->s_thread_exit != NFS4_THREAD_EXIT)
4002 			NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT,
4003 			    " - thread not exiting"));
4004 	}
4005 	delay(hz);
4006 }
4007 #endif
4008 #endif
4009 
4010 
4011 /*
4012  * Move a mntinfo4_t from one server list to another.
4013  * Locking of the two nfs4_server_t nodes will be done in list order.
4014  *
4015  * Returns NULL if the current nfs4_server_t for the filesystem could not
4016  * be found (e.g., due to forced unmount).  Otherwise returns a reference
4017  * to the new nfs4_server_t, which must eventually be freed.
4018  */
4019 nfs4_server_t *
4020 nfs4_move_mi(mntinfo4_t *mi, servinfo4_t *old, servinfo4_t *new)
4021 {
4022 	nfs4_server_t *p, *op = NULL, *np = NULL;
4023 	int num_open;
4024 	zoneid_t zoneid = nfs_zoneid();
4025 
4026 	ASSERT(nfs_zone() == mi->mi_zone);
4027 
4028 	mutex_enter(&nfs4_server_lst_lock);
4029 #ifdef DEBUG
4030 	if (nfs4_server_t_debug)
4031 		dumpnfs4slist("nfs4_move_mi", mi, (clientid4)0, new);
4032 #endif
4033 	for (p = nfs4_server_lst.forw; p != &nfs4_server_lst; p = p->forw) {
4034 		if (p->zoneid != zoneid)
4035 			continue;
4036 		if (p->saddr.len == old->sv_addr.len &&
4037 		    bcmp(p->saddr.buf, old->sv_addr.buf, p->saddr.len) == 0 &&
4038 		    p->s_thread_exit != NFS4_THREAD_EXIT) {
4039 			op = p;
4040 			mutex_enter(&op->s_lock);
4041 			op->s_refcnt++;
4042 		}
4043 		if (p->saddr.len == new->sv_addr.len &&
4044 		    bcmp(p->saddr.buf, new->sv_addr.buf, p->saddr.len) == 0 &&
4045 		    p->s_thread_exit != NFS4_THREAD_EXIT) {
4046 			np = p;
4047 			mutex_enter(&np->s_lock);
4048 		}
4049 		if (op != NULL && np != NULL)
4050 			break;
4051 	}
4052 	if (op == NULL) {
4053 		/*
4054 		 * Filesystem has been forcibly unmounted.  Bail out.
4055 		 */
4056 		if (np != NULL)
4057 			mutex_exit(&np->s_lock);
4058 		mutex_exit(&nfs4_server_lst_lock);
4059 		return (NULL);
4060 	}
4061 	if (np != NULL) {
4062 		np->s_refcnt++;
4063 	} else {
4064 #ifdef DEBUG
4065 		NFS4_DEBUG(nfs4_client_failover_debug, (CE_NOTE,
4066 		    "nfs4_move_mi: no target nfs4_server, will create."));
4067 #endif
4068 		np = add_new_nfs4_server(new, kcred);
4069 	}
4070 	mutex_exit(&nfs4_server_lst_lock);
4071 
4072 	NFS4_DEBUG(nfs4_client_failover_debug, (CE_NOTE,
4073 	    "nfs4_move_mi: for mi 0x%p, "
4074 	    "old servinfo4 0x%p, new servinfo4 0x%p, "
4075 	    "old nfs4_server 0x%p, new nfs4_server 0x%p, ",
4076 	    (void*)mi, (void*)old, (void*)new,
4077 	    (void*)op, (void*)np));
4078 	ASSERT(op != NULL && np != NULL);
4079 
4080 	/* discard any delegations */
4081 	nfs4_deleg_discard(mi, op);
4082 
4083 	num_open = mi->mi_open_files;
4084 	mi->mi_open_files = 0;
4085 	op->state_ref_count -= num_open;
4086 	ASSERT(op->state_ref_count >= 0);
4087 	np->state_ref_count += num_open;
4088 	nfs4_remove_mi_from_server_nolock(mi, op);
4089 	mi->mi_open_files = num_open;
4090 	NFS4_DEBUG(nfs4_client_failover_debug, (CE_NOTE,
4091 	    "nfs4_move_mi: mi_open_files %d, op->cnt %d, np->cnt %d",
4092 	    mi->mi_open_files, op->state_ref_count, np->state_ref_count));
4093 
4094 	nfs4_add_mi_to_server(np, mi);
4095 
4096 	mutex_exit(&op->s_lock);
4097 	mutex_exit(&np->s_lock);
4098 	nfs4_server_rele(op);
4099 
4100 	return (np);
4101 }
4102 
4103 /*
4104  * Need to have the nfs4_server_lst_lock.
4105  * Search the nfs4_server list to find a match on this servinfo4
4106  * based on its address.
4107  *
4108  * Returns NULL if no match is found.  Otherwise returns a reference (which
4109  * must eventually be freed) to a locked nfs4_server.
4110  */
4111 nfs4_server_t *
4112 servinfo4_to_nfs4_server(servinfo4_t *srv_p)
4113 {
4114 	nfs4_server_t *np;
4115 	zoneid_t zoneid = nfs_zoneid();
4116 
4117 	ASSERT(MUTEX_HELD(&nfs4_server_lst_lock));
4118 	for (np = nfs4_server_lst.forw; np != &nfs4_server_lst; np = np->forw) {
4119 		if (np->zoneid == zoneid &&
4120 		    np->saddr.len == srv_p->sv_addr.len &&
4121 		    bcmp(np->saddr.buf, srv_p->sv_addr.buf,
4122 		    np->saddr.len) == 0 &&
4123 		    np->s_thread_exit != NFS4_THREAD_EXIT) {
4124 			mutex_enter(&np->s_lock);
4125 			np->s_refcnt++;
4126 			return (np);
4127 		}
4128 	}
4129 	return (NULL);
4130 }
4131 
4132 /*
4133  * Locks the nfs4_server down if it is found and returns a reference that
4134  * must eventually be freed.
4135  */
4136 static nfs4_server_t *
4137 lookup_nfs4_server(nfs4_server_t *sp, int any_state)
4138 {
4139 	nfs4_server_t *np;
4140 
4141 	mutex_enter(&nfs4_server_lst_lock);
4142 	for (np = nfs4_server_lst.forw; np != &nfs4_server_lst; np = np->forw) {
4143 		mutex_enter(&np->s_lock);
4144 		if (np == sp && np->s_refcnt > 0 &&
4145 		    (np->s_thread_exit != NFS4_THREAD_EXIT || any_state)) {
4146 			mutex_exit(&nfs4_server_lst_lock);
4147 			np->s_refcnt++;
4148 			return (np);
4149 		}
4150 		mutex_exit(&np->s_lock);
4151 	}
4152 	mutex_exit(&nfs4_server_lst_lock);
4153 
4154 	return (NULL);
4155 }
4156 
4157 /*
4158  * The caller should be holding mi->mi_recovlock, and it should continue to
4159  * hold the lock until done with the returned nfs4_server_t.  Once
4160  * mi->mi_recovlock is released, there is no guarantee that the returned
4161  * mi->nfs4_server_t will continue to correspond to mi.
4162  */
4163 nfs4_server_t *
4164 find_nfs4_server(mntinfo4_t *mi)
4165 {
4166 	ASSERT(nfs_rw_lock_held(&mi->mi_recovlock, RW_READER) ||
4167 	    nfs_rw_lock_held(&mi->mi_recovlock, RW_WRITER));
4168 
4169 	return (lookup_nfs4_server(mi->mi_srv, 0));
4170 }
4171 
4172 /*
4173  * Same as above, but takes an "any_state" parameter which can be
4174  * set to 1 if the caller wishes to find nfs4_server_t's which
4175  * have been marked for termination by the exit of the renew
4176  * thread.  This should only be used by operations which are
4177  * cleaning up and will not cause an OTW op.
4178  */
4179 nfs4_server_t *
4180 find_nfs4_server_all(mntinfo4_t *mi, int any_state)
4181 {
4182 	ASSERT(nfs_rw_lock_held(&mi->mi_recovlock, RW_READER) ||
4183 	    nfs_rw_lock_held(&mi->mi_recovlock, RW_WRITER));
4184 
4185 	return (lookup_nfs4_server(mi->mi_srv, any_state));
4186 }
4187 
4188 /*
4189  * Lock sp, but only if it's still active (in the list and hasn't been
4190  * flagged as exiting) or 'any_state' is non-zero.
4191  * Returns TRUE if sp got locked and adds a reference to sp.
4192  */
4193 bool_t
4194 nfs4_server_vlock(nfs4_server_t *sp, int any_state)
4195 {
4196 	return (lookup_nfs4_server(sp, any_state) != NULL);
4197 }
4198 
4199 /*
4200  * Release the reference to sp and destroy it if that's the last one.
4201  */
4202 
4203 void
4204 nfs4_server_rele(nfs4_server_t *sp)
4205 {
4206 	mutex_enter(&sp->s_lock);
4207 	ASSERT(sp->s_refcnt > 0);
4208 	sp->s_refcnt--;
4209 	if (sp->s_refcnt > 0) {
4210 		mutex_exit(&sp->s_lock);
4211 		return;
4212 	}
4213 	mutex_exit(&sp->s_lock);
4214 
4215 	mutex_enter(&nfs4_server_lst_lock);
4216 	mutex_enter(&sp->s_lock);
4217 	if (sp->s_refcnt > 0) {
4218 		mutex_exit(&sp->s_lock);
4219 		mutex_exit(&nfs4_server_lst_lock);
4220 		return;
4221 	}
4222 	remque(sp);
4223 	sp->forw = sp->back = NULL;
4224 	mutex_exit(&nfs4_server_lst_lock);
4225 	destroy_nfs4_server(sp);
4226 }
4227 
4228 static void
4229 destroy_nfs4_server(nfs4_server_t *sp)
4230 {
4231 	ASSERT(MUTEX_HELD(&sp->s_lock));
4232 	ASSERT(sp->s_refcnt == 0);
4233 	ASSERT(sp->s_otw_call_count == 0);
4234 
4235 	remove_all_mi(sp);
4236 
4237 	crfree(sp->s_cred);
4238 	kmem_free(sp->saddr.buf, sp->saddr.maxlen);
4239 	kmem_free(sp->clidtosend.id_val, sp->clidtosend.id_len);
4240 	mutex_exit(&sp->s_lock);
4241 
4242 	/* destroy the nfs4_server */
4243 	nfs4callback_destroy(sp);
4244 	list_destroy(&sp->s_deleg_list);
4245 	mutex_destroy(&sp->s_lock);
4246 	cv_destroy(&sp->cv_thread_exit);
4247 	cv_destroy(&sp->s_cv_otw_count);
4248 	cv_destroy(&sp->s_clientid_pend);
4249 	cv_destroy(&sp->wait_cb_null);
4250 	nfs_rw_destroy(&sp->s_recovlock);
4251 	kmem_free(sp, sizeof (*sp));
4252 }
4253 
4254 /*
4255  * Fork off a thread to free the data structures for a mount.
4256  */
4257 
4258 static void
4259 async_free_mount(vfs_t *vfsp, int flag, cred_t *cr)
4260 {
4261 	freemountargs_t *args;
4262 	args = kmem_alloc(sizeof (freemountargs_t), KM_SLEEP);
4263 	args->fm_vfsp = vfsp;
4264 	VFS_HOLD(vfsp);
4265 	MI4_HOLD(VFTOMI4(vfsp));
4266 	args->fm_flag = flag;
4267 	args->fm_cr = cr;
4268 	crhold(cr);
4269 	(void) zthread_create(NULL, 0, nfs4_free_mount_thread, args, 0,
4270 	    minclsyspri);
4271 }
4272 
4273 static void
4274 nfs4_free_mount_thread(freemountargs_t *args)
4275 {
4276 	mntinfo4_t *mi;
4277 	nfs4_free_mount(args->fm_vfsp, args->fm_flag, args->fm_cr);
4278 	mi = VFTOMI4(args->fm_vfsp);
4279 	crfree(args->fm_cr);
4280 	VFS_RELE(args->fm_vfsp);
4281 	MI4_RELE(mi);
4282 	kmem_free(args, sizeof (freemountargs_t));
4283 	zthread_exit();
4284 	/* NOTREACHED */
4285 }
4286 
4287 /*
4288  * Thread to free the data structures for a given filesystem.
4289  */
4290 static void
4291 nfs4_free_mount(vfs_t *vfsp, int flag, cred_t *cr)
4292 {
4293 	mntinfo4_t		*mi = VFTOMI4(vfsp);
4294 	nfs4_server_t		*sp;
4295 	callb_cpr_t		cpr_info;
4296 	kmutex_t		cpr_lock;
4297 	boolean_t		async_thread;
4298 	int			removed;
4299 
4300 	bool_t			must_unlock;
4301 	nfs4_ephemeral_tree_t	*eph_tree;
4302 
4303 	/*
4304 	 * We need to participate in the CPR framework if this is a kernel
4305 	 * thread.
4306 	 */
4307 	async_thread = (curproc == nfs_zone()->zone_zsched);
4308 	if (async_thread) {
4309 		mutex_init(&cpr_lock, NULL, MUTEX_DEFAULT, NULL);
4310 		CALLB_CPR_INIT(&cpr_info, &cpr_lock, callb_generic_cpr,
4311 		    "nfsv4AsyncUnmount");
4312 	}
4313 
4314 	/*
4315 	 * We need to wait for all outstanding OTW calls
4316 	 * and recovery to finish before we remove the mi
4317 	 * from the nfs4_server_t, as current pending
4318 	 * calls might still need this linkage (in order
4319 	 * to find a nfs4_server_t from a mntinfo4_t).
4320 	 */
4321 	(void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_READER, FALSE);
4322 	sp = find_nfs4_server(mi);
4323 	nfs_rw_exit(&mi->mi_recovlock);
4324 
4325 	if (sp) {
4326 		while (sp->s_otw_call_count != 0) {
4327 			if (async_thread) {
4328 				mutex_enter(&cpr_lock);
4329 				CALLB_CPR_SAFE_BEGIN(&cpr_info);
4330 				mutex_exit(&cpr_lock);
4331 			}
4332 			cv_wait(&sp->s_cv_otw_count, &sp->s_lock);
4333 			if (async_thread) {
4334 				mutex_enter(&cpr_lock);
4335 				CALLB_CPR_SAFE_END(&cpr_info, &cpr_lock);
4336 				mutex_exit(&cpr_lock);
4337 			}
4338 		}
4339 		mutex_exit(&sp->s_lock);
4340 		nfs4_server_rele(sp);
4341 		sp = NULL;
4342 	}
4343 
4344 	mutex_enter(&mi->mi_lock);
4345 	while (mi->mi_in_recovery != 0) {
4346 		if (async_thread) {
4347 			mutex_enter(&cpr_lock);
4348 			CALLB_CPR_SAFE_BEGIN(&cpr_info);
4349 			mutex_exit(&cpr_lock);
4350 		}
4351 		cv_wait(&mi->mi_cv_in_recov, &mi->mi_lock);
4352 		if (async_thread) {
4353 			mutex_enter(&cpr_lock);
4354 			CALLB_CPR_SAFE_END(&cpr_info, &cpr_lock);
4355 			mutex_exit(&cpr_lock);
4356 		}
4357 	}
4358 	mutex_exit(&mi->mi_lock);
4359 
4360 	/*
4361 	 * If we got an error, then do not nuke the
4362 	 * tree. Either the harvester is busy reclaiming
4363 	 * this node or we ran into some busy condition.
4364 	 *
4365 	 * The harvester will eventually come along and cleanup.
4366 	 * The only problem would be the root mount point.
4367 	 *
4368 	 * Since the busy node can occur for a variety
4369 	 * of reasons and can result in an entry staying
4370 	 * in df output but no longer accessible from the
4371 	 * directory tree, we are okay.
4372 	 */
4373 	if (!nfs4_ephemeral_umount(mi, flag, cr,
4374 	    &must_unlock, &eph_tree))
4375 		nfs4_ephemeral_umount_activate(mi, &must_unlock,
4376 		    &eph_tree);
4377 
4378 	/*
4379 	 * The original purge of the dnlc via 'dounmount'
4380 	 * doesn't guarantee that another dnlc entry was not
4381 	 * added while we waitied for all outstanding OTW
4382 	 * and recovery calls to finish.  So re-purge the
4383 	 * dnlc now.
4384 	 */
4385 	(void) dnlc_purge_vfsp(vfsp, 0);
4386 
4387 	/*
4388 	 * We need to explicitly stop the manager thread; the asyc worker
4389 	 * threads can timeout and exit on their own.
4390 	 */
4391 	mutex_enter(&mi->mi_async_lock);
4392 	mi->mi_max_threads = 0;
4393 	NFS4_WAKEALL_ASYNC_WORKERS(mi->mi_async_work_cv);
4394 	mutex_exit(&mi->mi_async_lock);
4395 	if (mi->mi_manager_thread)
4396 		nfs4_async_manager_stop(vfsp);
4397 
4398 	destroy_rtable4(vfsp, cr);
4399 
4400 	nfs4_remove_mi_from_server(mi, NULL);
4401 
4402 	if (async_thread) {
4403 		mutex_enter(&cpr_lock);
4404 		CALLB_CPR_EXIT(&cpr_info);	/* drops cpr_lock */
4405 		mutex_destroy(&cpr_lock);
4406 	}
4407 
4408 	removed = nfs4_mi_zonelist_remove(mi);
4409 	if (removed)
4410 		zone_rele_ref(&mi->mi_zone_ref, ZONE_REF_NFSV4);
4411 }
4412 
4413 /* Referral related sub-routines */
4414 
4415 /* Freeup knetconfig */
4416 static void
4417 free_knconf_contents(struct knetconfig *k)
4418 {
4419 	if (k == NULL)
4420 		return;
4421 	if (k->knc_protofmly)
4422 		kmem_free(k->knc_protofmly, KNC_STRSIZE);
4423 	if (k->knc_proto)
4424 		kmem_free(k->knc_proto, KNC_STRSIZE);
4425 }
4426 
4427 /*
4428  * This updates newpath variable with exact name component from the
4429  * path which gave us a NFS4ERR_MOVED error.
4430  * If the path is /rp/aaa/bbb and nth value is 1, aaa is returned.
4431  */
4432 static char *
4433 extract_referral_point(const char *svp, int nth)
4434 {
4435 	int num_slashes = 0;
4436 	const char *p;
4437 	char *newpath = NULL;
4438 	int i = 0;
4439 
4440 	newpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
4441 	for (p = svp; *p; p++) {
4442 		if (*p == '/')
4443 			num_slashes++;
4444 		if (num_slashes == nth + 1) {
4445 			p++;
4446 			while (*p != '/') {
4447 				if (*p == '\0')
4448 					break;
4449 				newpath[i] = *p;
4450 				i++;
4451 				p++;
4452 			}
4453 			newpath[i++] = '\0';
4454 			break;
4455 		}
4456 	}
4457 	return (newpath);
4458 }
4459 
4460 /*
4461  * This sets up a new path in sv_path to do a lookup of the referral point.
4462  * If the path is /rp/aaa/bbb and the referral point is aaa,
4463  * this updates /rp/aaa. This path will be used to get referral
4464  * location.
4465  */
4466 static void
4467 setup_newsvpath(servinfo4_t *svp, int nth)
4468 {
4469 	int num_slashes = 0, pathlen, i = 0;
4470 	char *newpath, *p;
4471 
4472 	newpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
4473 	for (p = svp->sv_path; *p; p++) {
4474 		newpath[i] =  *p;
4475 		if (*p == '/')
4476 			num_slashes++;
4477 		if (num_slashes == nth + 1) {
4478 			newpath[i] = '\0';
4479 			pathlen = strlen(newpath) + 1;
4480 			kmem_free(svp->sv_path, svp->sv_pathlen);
4481 			svp->sv_path = kmem_alloc(pathlen, KM_SLEEP);
4482 			svp->sv_pathlen = pathlen;
4483 			bcopy(newpath, svp->sv_path, pathlen);
4484 			break;
4485 		}
4486 		i++;
4487 	}
4488 	kmem_free(newpath, MAXPATHLEN);
4489 }
4490