xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs3_vfsops.c (revision 734b6a94890be549309b21156f8ed6d4561cac51)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
28  *	All rights reserved.
29  */
30 
31 #pragma ident	"%Z%%M%	%I%	%E% SMI"
32 
33 #include <sys/param.h>
34 #include <sys/types.h>
35 #include <sys/systm.h>
36 #include <sys/cred.h>
37 #include <sys/vfs.h>
38 #include <sys/vnode.h>
39 #include <sys/pathname.h>
40 #include <sys/sysmacros.h>
41 #include <sys/kmem.h>
42 #include <sys/mkdev.h>
43 #include <sys/mount.h>
44 #include <sys/mntent.h>
45 #include <sys/statvfs.h>
46 #include <sys/errno.h>
47 #include <sys/debug.h>
48 #include <sys/cmn_err.h>
49 #include <sys/utsname.h>
50 #include <sys/bootconf.h>
51 #include <sys/modctl.h>
52 #include <sys/acl.h>
53 #include <sys/flock.h>
54 #include <sys/policy.h>
55 #include <sys/zone.h>
56 #include <sys/class.h>
57 #include <sys/socket.h>
58 #include <sys/netconfig.h>
59 #include <sys/tsol/tnet.h>
60 
61 #include <rpc/types.h>
62 #include <rpc/auth.h>
63 #include <rpc/clnt.h>
64 
65 #include <nfs/nfs.h>
66 #include <nfs/nfs_clnt.h>
67 #include <nfs/rnode.h>
68 #include <nfs/mount.h>
69 #include <nfs/nfs_acl.h>
70 
71 #include <fs/fs_subr.h>
72 
73 /*
74  * From rpcsec module (common/rpcsec).
75  */
76 extern int sec_clnt_loadinfo(struct sec_data *, struct sec_data **, model_t);
77 extern void sec_clnt_freeinfo(struct sec_data *);
78 
79 /*
80  * The order and contents of this structure must be kept in sync with that of
81  * rfsreqcnt_v3_tmpl in nfs_stats.c
82  */
83 static char *rfsnames_v3[] = {
84 	"null", "getattr", "setattr", "lookup", "access", "readlink", "read",
85 	"write", "create", "mkdir", "symlink", "mknod", "remove", "rmdir",
86 	"rename", "link", "readdir", "readdirplus", "fsstat", "fsinfo",
87 	"pathconf", "commit"
88 };
89 
90 /*
91  * This table maps from NFS protocol number into call type.
92  * Zero means a "Lookup" type call
93  * One  means a "Read" type call
94  * Two  means a "Write" type call
95  * This is used to select a default time-out.
96  */
97 static uchar_t call_type_v3[] = {
98 	0, 0, 1, 0, 0, 0, 1,
99 	2, 2, 2, 2, 2, 2, 2,
100 	2, 2, 1, 2, 0, 0, 0,
101 	2 };
102 
103 /*
104  * Similar table, but to determine which timer to use
105  * (only real reads and writes!)
106  */
107 static uchar_t timer_type_v3[] = {
108 	0, 0, 0, 0, 0, 0, 1,
109 	2, 0, 0, 0, 0, 0, 0,
110 	0, 0, 1, 1, 0, 0, 0,
111 	0 };
112 
113 /*
114  * This table maps from NFS protocol number into a call type
115  * for the semisoft mount option.
116  * Zero means do not repeat operation.
117  * One  means repeat.
118  */
119 static uchar_t ss_call_type_v3[] = {
120 	0, 0, 1, 0, 0, 0, 0,
121 	1, 1, 1, 1, 1, 1, 1,
122 	1, 1, 0, 0, 0, 0, 0,
123 	1 };
124 
125 /*
126  * nfs3 vfs operations.
127  */
128 static int	nfs3_mount(vfs_t *, vnode_t *, struct mounta *, cred_t *);
129 static int	nfs3_unmount(vfs_t *, int, cred_t *);
130 static int	nfs3_root(vfs_t *, vnode_t **);
131 static int	nfs3_statvfs(vfs_t *, struct statvfs64 *);
132 static int	nfs3_sync(vfs_t *, short, cred_t *);
133 static int	nfs3_vget(vfs_t *, vnode_t **, fid_t *);
134 static int	nfs3_mountroot(vfs_t *, whymountroot_t);
135 static void	nfs3_freevfs(vfs_t *);
136 
137 static int	nfs3rootvp(vnode_t **, vfs_t *, struct servinfo *,
138 		    int, cred_t *, zone_t *);
139 
140 /*
141  * Initialize the vfs structure
142  */
143 
144 static int nfs3fstyp;
145 vfsops_t *nfs3_vfsops;
146 
147 /*
148  * Debug variable to check for rdma based
149  * transport startup and cleanup. Controlled
150  * through /etc/system. Off by default.
151  */
152 extern int rdma_debug;
153 
154 int
155 nfs3init(int fstyp, char *name)
156 {
157 	static const fs_operation_def_t nfs3_vfsops_template[] = {
158 		VFSNAME_MOUNT, nfs3_mount,
159 		VFSNAME_UNMOUNT, nfs3_unmount,
160 		VFSNAME_ROOT, nfs3_root,
161 		VFSNAME_STATVFS, nfs3_statvfs,
162 		VFSNAME_SYNC, (fs_generic_func_p) nfs3_sync,
163 		VFSNAME_VGET, nfs3_vget,
164 		VFSNAME_MOUNTROOT, nfs3_mountroot,
165 		VFSNAME_FREEVFS, (fs_generic_func_p)nfs3_freevfs,
166 		NULL, NULL
167 	};
168 	int error;
169 
170 	error = vfs_setfsops(fstyp, nfs3_vfsops_template, &nfs3_vfsops);
171 	if (error != 0) {
172 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
173 		    "nfs3init: bad vfs ops template");
174 		return (error);
175 	}
176 
177 	error = vn_make_ops(name, nfs3_vnodeops_template, &nfs3_vnodeops);
178 	if (error != 0) {
179 		(void) vfs_freevfsops_by_type(fstyp);
180 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
181 		    "nfs3init: bad vnode ops template");
182 		return (error);
183 	}
184 
185 	nfs3fstyp = fstyp;
186 
187 	return (0);
188 }
189 
190 void
191 nfs3fini(void)
192 {
193 }
194 
195 /*
196  * nfs mount vfsop
197  * Set up mount info record and attach it to vfs struct.
198  */
199 static int
200 nfs3_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
201 {
202 	char *data = uap->dataptr;
203 	int error;
204 	vnode_t *rtvp;			/* the server's root */
205 	mntinfo_t *mi;			/* mount info, pointed at by vfs */
206 	size_t hlen;			/* length of hostname */
207 	size_t nlen;			/* length of netname */
208 	char netname[SYS_NMLN];		/* server's netname */
209 	struct netbuf addr;		/* server's address */
210 	struct netbuf syncaddr;		/* AUTH_DES time sync addr */
211 	struct knetconfig *knconf;	/* transport knetconfig structure */
212 	struct knetconfig *rdma_knconf;	/* rdma transport structure */
213 	rnode_t *rp;
214 	struct servinfo *svp;		/* nfs server info */
215 	struct servinfo *svp_tail = NULL; /* previous nfs server info */
216 	struct servinfo *svp_head;	/* first nfs server info */
217 	struct servinfo *svp_2ndlast;	/* 2nd last in server info list */
218 	struct sec_data *secdata;	/* security data */
219 	STRUCT_DECL(nfs_args, args);	/* nfs mount arguments */
220 	STRUCT_DECL(knetconfig, knconf_tmp);
221 	STRUCT_DECL(netbuf, addr_tmp);
222 	int flags, addr_type;
223 	char *p, *pf;
224 	zone_t *zone = nfs_zone();
225 	zone_t *mntzone = NULL;
226 
227 	if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
228 		return (EPERM);
229 
230 	if (mvp->v_type != VDIR)
231 		return (ENOTDIR);
232 
233 	/*
234 	 * get arguments
235 	 *
236 	 * nfs_args is now versioned and is extensible, so
237 	 * uap->datalen might be different from sizeof (args)
238 	 * in a compatible situation.
239 	 */
240 more:
241 	STRUCT_INIT(args, get_udatamodel());
242 	bzero(STRUCT_BUF(args), SIZEOF_STRUCT(nfs_args, DATAMODEL_NATIVE));
243 	if (copyin(data, STRUCT_BUF(args), MIN(uap->datalen,
244 	    STRUCT_SIZE(args))))
245 		return (EFAULT);
246 
247 	flags = STRUCT_FGET(args, flags);
248 
249 	if (uap->flags & MS_REMOUNT) {
250 		size_t n;
251 		char name[FSTYPSZ];
252 
253 		if (uap->flags & MS_SYSSPACE)
254 			error = copystr(uap->fstype, name, FSTYPSZ, &n);
255 		else
256 			error = copyinstr(uap->fstype, name, FSTYPSZ, &n);
257 
258 		if (error) {
259 			if (error == ENAMETOOLONG)
260 				return (EINVAL);
261 			return (error);
262 		}
263 
264 		/*
265 		 * This check is to ensure that the request is a
266 		 * genuine nfs remount request.
267 		 */
268 
269 		if (strncmp(name, "nfs", 3) != 0)
270 			return (EINVAL);
271 
272 		/*
273 		 * If the request changes the locking type, disallow the
274 		 * remount,
275 		 * because it's questionable whether we can transfer the
276 		 * locking state correctly.
277 		 */
278 
279 		if ((mi = VFTOMI(vfsp)) != NULL) {
280 			uint_t new_mi_llock;
281 			uint_t old_mi_llock;
282 
283 			new_mi_llock = (flags & NFSMNT_LLOCK) ? 1 : 0;
284 			old_mi_llock = (mi->mi_flags & MI_LLOCK) ? 1 : 0;
285 			if (old_mi_llock != new_mi_llock)
286 				return (EBUSY);
287 		}
288 		return (0);
289 	}
290 
291 	mutex_enter(&mvp->v_lock);
292 	if (!(uap->flags & MS_OVERLAY) &&
293 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
294 		mutex_exit(&mvp->v_lock);
295 		return (EBUSY);
296 	}
297 	mutex_exit(&mvp->v_lock);
298 
299 	/* make sure things are zeroed for errout: */
300 	rtvp = NULL;
301 	mi = NULL;
302 	addr.buf = NULL;
303 	syncaddr.buf = NULL;
304 	secdata = NULL;
305 
306 	/*
307 	 * A valid knetconfig structure is required.
308 	 */
309 	if (!(flags & NFSMNT_KNCONF))
310 		return (EINVAL);
311 
312 	/*
313 	 * Allocate a servinfo struct.
314 	 */
315 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
316 	mutex_init(&svp->sv_lock, NULL, MUTEX_DEFAULT, NULL);
317 	if (svp_tail) {
318 		svp_2ndlast = svp_tail;
319 		svp_tail->sv_next = svp;
320 	} else {
321 		svp_head = svp;
322 		svp_2ndlast = svp;
323 	}
324 
325 	svp_tail = svp;
326 
327 	/*
328 	 * Allocate space for a knetconfig structure and
329 	 * its strings and copy in from user-land.
330 	 */
331 	knconf = kmem_zalloc(sizeof (*knconf), KM_SLEEP);
332 	svp->sv_knconf = knconf;
333 	STRUCT_INIT(knconf_tmp, get_udatamodel());
334 	if (copyin(STRUCT_FGETP(args, knconf), STRUCT_BUF(knconf_tmp),
335 	    STRUCT_SIZE(knconf_tmp))) {
336 		sv_free(svp_head);
337 		return (EFAULT);
338 	}
339 
340 	knconf->knc_semantics = STRUCT_FGET(knconf_tmp, knc_semantics);
341 	knconf->knc_protofmly = STRUCT_FGETP(knconf_tmp, knc_protofmly);
342 	knconf->knc_proto = STRUCT_FGETP(knconf_tmp, knc_proto);
343 	if (get_udatamodel() != DATAMODEL_LP64) {
344 		knconf->knc_rdev = expldev(STRUCT_FGET(knconf_tmp, knc_rdev));
345 	} else {
346 		knconf->knc_rdev = STRUCT_FGET(knconf_tmp, knc_rdev);
347 	}
348 
349 	pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
350 	p = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
351 	error = copyinstr(knconf->knc_protofmly, pf, KNC_STRSIZE, NULL);
352 	if (error) {
353 		kmem_free(pf, KNC_STRSIZE);
354 		kmem_free(p, KNC_STRSIZE);
355 		sv_free(svp_head);
356 		return (error);
357 	}
358 	error = copyinstr(knconf->knc_proto, p, KNC_STRSIZE, NULL);
359 	if (error) {
360 		kmem_free(pf, KNC_STRSIZE);
361 		kmem_free(p, KNC_STRSIZE);
362 		sv_free(svp_head);
363 		return (error);
364 	}
365 	knconf->knc_protofmly = pf;
366 	knconf->knc_proto = p;
367 
368 	/*
369 	 * Get server address
370 	 */
371 	STRUCT_INIT(addr_tmp, get_udatamodel());
372 	if (copyin(STRUCT_FGETP(args, addr), STRUCT_BUF(addr_tmp),
373 	    STRUCT_SIZE(addr_tmp))) {
374 		addr.buf = NULL;
375 		error = EFAULT;
376 	} else {
377 		char *userbufptr;
378 
379 		userbufptr = addr.buf = STRUCT_FGETP(addr_tmp, buf);
380 		addr.len = STRUCT_FGET(addr_tmp, len);
381 		addr.buf = kmem_alloc(addr.len, KM_SLEEP);
382 		addr.maxlen = addr.len;
383 		if (copyin(userbufptr, addr.buf, addr.len))
384 			error = EFAULT;
385 	}
386 	svp->sv_addr = addr;
387 	if (error)
388 		goto errout;
389 
390 	/*
391 	 * Get the root fhandle
392 	 */
393 	if (copyin(STRUCT_FGETP(args, fh), &svp->sv_fhandle,
394 	    sizeof (svp->sv_fhandle))) {
395 		error = EFAULT;
396 		goto errout;
397 	}
398 
399 	/*
400 	 * Check the root fhandle length
401 	 */
402 	if (svp->sv_fhandle.fh_len > NFS3_FHSIZE ||
403 		svp->sv_fhandle.fh_len <= 0) {
404 		error = EINVAL;
405 #ifdef DEBUG
406 		zcmn_err(getzoneid(), CE_WARN,
407 		    "nfs3_mount: got an invalid fhandle. fh_len = %d",
408 		    svp->sv_fhandle.fh_len);
409 		svp->sv_fhandle.fh_len = NFS_FHANDLE_LEN;
410 		nfs_printfhandle(&svp->sv_fhandle);
411 #endif
412 		goto errout;
413 	}
414 
415 	/*
416 	 * Get server's hostname
417 	 */
418 	if (flags & NFSMNT_HOSTNAME) {
419 		error = copyinstr(STRUCT_FGETP(args, hostname),
420 		    netname, sizeof (netname), &hlen);
421 		if (error)
422 			goto errout;
423 	} else {
424 		char *p = "unknown-host";
425 		hlen = strlen(p) + 1;
426 		(void) strcpy(netname, p);
427 	}
428 	svp->sv_hostnamelen = hlen;
429 	svp->sv_hostname = kmem_alloc(svp->sv_hostnamelen, KM_SLEEP);
430 	(void) strcpy(svp->sv_hostname, netname);
431 
432 	/*
433 	 * RDMA MOUNT SUPPORT FOR NFS v3:
434 	 * Establish, is it possible to use RDMA, if so overload the
435 	 * knconf with rdma specific knconf and free the orignal.
436 	 */
437 	if ((flags & NFSMNT_TRYRDMA) || (flags & NFSMNT_DORDMA)) {
438 		/*
439 		 * Determine the addr type for RDMA, IPv4 or v6.
440 		 */
441 		if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET) == 0)
442 			addr_type = AF_INET;
443 		else if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET6) == 0)
444 			addr_type = AF_INET6;
445 
446 		if (rdma_reachable(addr_type, &svp->sv_addr,
447 			&rdma_knconf) == 0) {
448 			/*
449 			 * If successful, hijack the orignal knconf and
450 			 * replace with a new one, depending on the flags.
451 			 */
452 			svp->sv_origknconf = svp->sv_knconf;
453 			svp->sv_knconf = rdma_knconf;
454 			knconf = rdma_knconf;
455 		} else {
456 			if (flags & NFSMNT_TRYRDMA) {
457 #ifdef	DEBUG
458 				if (rdma_debug)
459 					zcmn_err(getzoneid(), CE_WARN,
460 					    "no RDMA onboard, revert\n");
461 #endif
462 			}
463 
464 			if (flags & NFSMNT_DORDMA) {
465 				/*
466 				 * If proto=rdma is specified and no RDMA
467 				 * path to this server is avialable then
468 				 * ditch this server.
469 				 * This is not included in the mountable
470 				 * server list or the replica list.
471 				 * Check if more servers are specified;
472 				 * Failover case, otherwise bail out of mount.
473 				 */
474 				if (STRUCT_FGET(args, nfs_args_ext) ==
475 				    NFS_ARGS_EXTB && STRUCT_FGETP(args,
476 					nfs_ext_u.nfs_extB.next) != NULL) {
477 					if (uap->flags & MS_RDONLY &&
478 					    !(flags & NFSMNT_SOFT)) {
479 						data = (char *)
480 						    STRUCT_FGETP(args,
481 						nfs_ext_u.nfs_extB.next);
482 						if (svp_head->sv_next == NULL) {
483 							svp_tail = NULL;
484 							svp_2ndlast = NULL;
485 							sv_free(svp_head);
486 							goto more;
487 						} else {
488 							svp_tail = svp_2ndlast;
489 							svp_2ndlast->sv_next =
490 							    NULL;
491 							sv_free(svp);
492 							goto more;
493 						}
494 					}
495 				} else {
496 					/*
497 					 * This is the last server specified
498 					 * in the nfs_args list passed down
499 					 * and its not rdma capable.
500 					 */
501 					if (svp_head->sv_next == NULL) {
502 						/*
503 						 * Is this the only one
504 						 */
505 						error = EINVAL;
506 #ifdef	DEBUG
507 						if (rdma_debug)
508 							zcmn_err(getzoneid(),
509 							    CE_WARN,
510 							    "No RDMA srv");
511 #endif
512 						goto errout;
513 					} else {
514 						/*
515 						 * There is list, since some
516 						 * servers specified before
517 						 * this passed all requirements
518 						 */
519 						svp_tail = svp_2ndlast;
520 						svp_2ndlast->sv_next = NULL;
521 						sv_free(svp);
522 						goto proceed;
523 					}
524 				}
525 			}
526 		}
527 	}
528 
529 	/*
530 	 * Get the extention data which has the new security data structure.
531 	 */
532 	if (flags & NFSMNT_NEWARGS) {
533 		switch (STRUCT_FGET(args, nfs_args_ext)) {
534 		case NFS_ARGS_EXTA:
535 		case NFS_ARGS_EXTB:
536 			/*
537 			 * Indicating the application is using the new
538 			 * sec_data structure to pass in the security
539 			 * data.
540 			 */
541 			if (STRUCT_FGETP(args,
542 			    nfs_ext_u.nfs_extA.secdata) == NULL) {
543 				error = EINVAL;
544 			} else {
545 				error = sec_clnt_loadinfo(
546 				    (struct sec_data *)STRUCT_FGETP(args,
547 					nfs_ext_u.nfs_extA.secdata),
548 				    &secdata, get_udatamodel());
549 			}
550 			break;
551 
552 		default:
553 			error = EINVAL;
554 			break;
555 		}
556 	} else if (flags & NFSMNT_SECURE) {
557 		/*
558 		 * Keep this for backward compatibility to support
559 		 * NFSMNT_SECURE/NFSMNT_RPCTIMESYNC flags.
560 		 */
561 		if (STRUCT_FGETP(args, syncaddr) == NULL) {
562 			error = EINVAL;
563 		} else {
564 			/*
565 			 * get time sync address.
566 			 */
567 			if (copyin(STRUCT_FGETP(args, syncaddr), &addr_tmp,
568 			    STRUCT_SIZE(addr_tmp))) {
569 				syncaddr.buf = NULL;
570 				error = EFAULT;
571 			} else {
572 				char *userbufptr;
573 
574 				userbufptr = syncaddr.buf =
575 				    STRUCT_FGETP(addr_tmp, buf);
576 				syncaddr.len =
577 				    STRUCT_FGET(addr_tmp, len);
578 				syncaddr.buf = kmem_alloc(syncaddr.len,
579 				    KM_SLEEP);
580 				syncaddr.maxlen = syncaddr.len;
581 
582 				if (copyin(userbufptr, syncaddr.buf,
583 				    syncaddr.len))
584 					error = EFAULT;
585 			}
586 
587 			/*
588 			 * get server's netname
589 			 */
590 			if (!error) {
591 				error = copyinstr(STRUCT_FGETP(args, netname),
592 				    netname, sizeof (netname), &nlen);
593 				netname[nlen] = '\0';
594 			}
595 
596 			if (error && syncaddr.buf != NULL) {
597 				kmem_free(syncaddr.buf, syncaddr.len);
598 				syncaddr.buf = NULL;
599 			}
600 		}
601 
602 		/*
603 		 * Move security related data to the sec_data structure.
604 		 */
605 		if (!error) {
606 			dh_k4_clntdata_t *data;
607 			char *pf, *p;
608 
609 			secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
610 			if (flags & NFSMNT_RPCTIMESYNC)
611 				secdata->flags |= AUTH_F_RPCTIMESYNC;
612 			data = kmem_alloc(sizeof (*data), KM_SLEEP);
613 			data->syncaddr = syncaddr;
614 
615 			/*
616 			 * duplicate the knconf information for the
617 			 * new opaque data.
618 			 */
619 			data->knconf = kmem_alloc(sizeof (*knconf), KM_SLEEP);
620 			*data->knconf = *knconf;
621 			pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
622 			p = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
623 			bcopy(knconf->knc_protofmly, pf, KNC_STRSIZE);
624 			bcopy(knconf->knc_proto, pf, KNC_STRSIZE);
625 			data->knconf->knc_protofmly = pf;
626 			data->knconf->knc_proto = p;
627 
628 			/* move server netname to the sec_data structure */
629 			if (nlen != 0) {
630 				data->netname = kmem_alloc(nlen, KM_SLEEP);
631 				bcopy(netname, data->netname, nlen);
632 				data->netnamelen = (int)nlen;
633 			}
634 			secdata->secmod = secdata->rpcflavor = AUTH_DES;
635 			secdata->data = (caddr_t)data;
636 		}
637 	} else {
638 		secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
639 		secdata->secmod = secdata->rpcflavor = AUTH_UNIX;
640 		secdata->data = NULL;
641 	}
642 	svp->sv_secdata = secdata;
643 	if (error)
644 		goto errout;
645 
646 	/*
647 	 * See bug 1180236.
648 	 * If mount secure failed, we will fall back to AUTH_NONE
649 	 * and try again.  nfs3rootvp() will turn this back off.
650 	 *
651 	 * The NFS Version 3 mount uses the FSINFO and GETATTR
652 	 * procedures.  The server should not care if these procedures
653 	 * have the proper security flavor, so if mount retries using
654 	 * AUTH_NONE that does not require a credential setup for root
655 	 * then the automounter would work without requiring root to be
656 	 * keylogged into AUTH_DES.
657 	 */
658 	if (secdata->rpcflavor != AUTH_UNIX &&
659 	    secdata->rpcflavor != AUTH_LOOPBACK)
660 		secdata->flags |= AUTH_F_TRYNONE;
661 
662 	/*
663 	 * Failover support:
664 	 *
665 	 * We may have a linked list of nfs_args structures,
666 	 * which means the user is looking for failover.  If
667 	 * the mount is either not "read-only" or "soft",
668 	 * we want to bail out with EINVAL.
669 	 */
670 	if (STRUCT_FGET(args, nfs_args_ext) == NFS_ARGS_EXTB &&
671 	    STRUCT_FGETP(args, nfs_ext_u.nfs_extB.next) != NULL) {
672 		if (uap->flags & MS_RDONLY && !(flags & NFSMNT_SOFT)) {
673 			data = (char *)STRUCT_FGETP(args,
674 			    nfs_ext_u.nfs_extB.next);
675 			goto more;
676 		}
677 		error = EINVAL;
678 		goto errout;
679 	}
680 
681 	/*
682 	 * Determine the zone we're being mounted into.
683 	 */
684 	zone_hold(mntzone = zone);		/* start with this assumption */
685 	if (getzoneid() == GLOBAL_ZONEID) {
686 		zone_rele(mntzone);
687 		mntzone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
688 		ASSERT(mntzone != NULL);
689 		if (mntzone != zone) {
690 			error = EBUSY;
691 			goto errout;
692 		}
693 	}
694 
695 	if (is_system_labeled()) {
696 		error = nfs_mount_label_policy(vfsp, &svp->sv_addr,
697 		    svp->sv_knconf, cr);
698 
699 		if (error > 0)
700 			goto errout;
701 
702 		if (error == -1) {
703 			/* change mount to read-only to prevent write-down */
704 			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
705 		}
706 	}
707 
708 	/*
709 	 * Stop the mount from going any further if the zone is going away.
710 	 */
711 	if (zone_status_get(mntzone) >= ZONE_IS_SHUTTING_DOWN) {
712 		error = EBUSY;
713 		goto errout;
714 	}
715 
716 	/*
717 	 * Get root vnode.
718 	 */
719 proceed:
720 	error = nfs3rootvp(&rtvp, vfsp, svp_head, flags, cr, mntzone);
721 
722 	if (error)
723 		goto errout;
724 
725 	/*
726 	 * Set option fields in the mount info record
727 	 */
728 	mi = VTOMI(rtvp);
729 
730 	if (svp_head->sv_next)
731 		mi->mi_flags |= MI_LLOCK;
732 
733 	error = nfs_setopts(rtvp, get_udatamodel(), STRUCT_BUF(args));
734 
735 errout:
736 	if (error) {
737 		if (rtvp != NULL) {
738 			rp = VTOR(rtvp);
739 			if (rp->r_flags & RHASHED)
740 				rp_rmhash(rp);
741 		}
742 		sv_free(svp_head);
743 		if (mi != NULL) {
744 			nfs_async_stop(vfsp);
745 			nfs_async_manager_stop(vfsp);
746 			if (mi->mi_io_kstats) {
747 				kstat_delete(mi->mi_io_kstats);
748 				mi->mi_io_kstats = NULL;
749 			}
750 			if (mi->mi_ro_kstats) {
751 				kstat_delete(mi->mi_ro_kstats);
752 				mi->mi_ro_kstats = NULL;
753 			}
754 			nfs_free_mi(mi);
755 		}
756 	}
757 
758 	if (rtvp != NULL)
759 		VN_RELE(rtvp);
760 
761 	if (mntzone != NULL)
762 		zone_rele(mntzone);
763 
764 	return (error);
765 }
766 
767 static int nfs3_dynamic = 0;	/* global variable to enable dynamic retrans. */
768 static ushort_t nfs3_max_threads = 8;	/* max number of active async threads */
769 static uint_t nfs3_bsize = 32 * 1024;	/* client `block' size */
770 static uint_t nfs3_async_clusters = 1;	/* # of reqs from each async queue */
771 static uint_t nfs3_cots_timeo = NFS_COTS_TIMEO;
772 
773 static int
774 nfs3rootvp(vnode_t **rtvpp, vfs_t *vfsp, struct servinfo *svp,
775 	int flags, cred_t *cr, zone_t *zone)
776 {
777 	vnode_t *rtvp;
778 	mntinfo_t *mi;
779 	dev_t nfs_dev;
780 	struct vattr va;
781 	struct FSINFO3args args;
782 	struct FSINFO3res res;
783 	int error;
784 	int douprintf;
785 	rnode_t *rp;
786 	int i;
787 	uint_t max_transfer_size;
788 	struct nfs_stats *nfsstatsp;
789 	cred_t *lcr = NULL, *tcr = cr;
790 
791 	nfsstatsp = zone_getspecific(nfsstat_zone_key, nfs_zone());
792 	ASSERT(nfsstatsp != NULL);
793 
794 	ASSERT(nfs_zone() == zone);
795 	/*
796 	 * Create a mount record and link it to the vfs struct.
797 	 */
798 	mi = kmem_zalloc(sizeof (*mi), KM_SLEEP);
799 	mutex_init(&mi->mi_lock, NULL, MUTEX_DEFAULT, NULL);
800 	mutex_init(&mi->mi_remap_lock, NULL, MUTEX_DEFAULT, NULL);
801 	mi->mi_flags = MI_ACL | MI_EXTATTR;
802 	if (!(flags & NFSMNT_SOFT))
803 		mi->mi_flags |= MI_HARD;
804 	if ((flags & NFSMNT_SEMISOFT))
805 		mi->mi_flags |= MI_SEMISOFT;
806 	if ((flags & NFSMNT_NOPRINT))
807 		mi->mi_flags |= MI_NOPRINT;
808 	if (flags & NFSMNT_INT)
809 		mi->mi_flags |= MI_INT;
810 	mi->mi_retrans = NFS_RETRIES;
811 	if (svp->sv_knconf->knc_semantics == NC_TPI_COTS_ORD ||
812 	    svp->sv_knconf->knc_semantics == NC_TPI_COTS)
813 		mi->mi_timeo = nfs3_cots_timeo;
814 	else
815 		mi->mi_timeo = NFS_TIMEO;
816 	mi->mi_prog = NFS_PROGRAM;
817 	mi->mi_vers = NFS_V3;
818 	mi->mi_rfsnames = rfsnames_v3;
819 	mi->mi_reqs = nfsstatsp->nfs_stats_v3.rfsreqcnt_ptr;
820 	mi->mi_call_type = call_type_v3;
821 	mi->mi_ss_call_type = ss_call_type_v3;
822 	mi->mi_timer_type = timer_type_v3;
823 	mi->mi_aclnames = aclnames_v3;
824 	mi->mi_aclreqs = nfsstatsp->nfs_stats_v3.aclreqcnt_ptr;
825 	mi->mi_acl_call_type = acl_call_type_v3;
826 	mi->mi_acl_ss_call_type = acl_ss_call_type_v3;
827 	mi->mi_acl_timer_type = acl_timer_type_v3;
828 	cv_init(&mi->mi_failover_cv, NULL, CV_DEFAULT, NULL);
829 	mi->mi_servers = svp;
830 	mi->mi_curr_serv = svp;
831 	mi->mi_acregmin = SEC2HR(ACREGMIN);
832 	mi->mi_acregmax = SEC2HR(ACREGMAX);
833 	mi->mi_acdirmin = SEC2HR(ACDIRMIN);
834 	mi->mi_acdirmax = SEC2HR(ACDIRMAX);
835 
836 	if (nfs3_dynamic)
837 		mi->mi_flags |= MI_DYNAMIC;
838 
839 	if (flags & NFSMNT_DIRECTIO)
840 		mi->mi_flags |= MI_DIRECTIO;
841 
842 	/*
843 	 * Make a vfs struct for nfs.  We do this here instead of below
844 	 * because rtvp needs a vfs before we can do a getattr on it.
845 	 *
846 	 * Assign a unique device id to the mount
847 	 */
848 	mutex_enter(&nfs_minor_lock);
849 	do {
850 		nfs_minor = (nfs_minor + 1) & MAXMIN32;
851 		nfs_dev = makedevice(nfs_major, nfs_minor);
852 	} while (vfs_devismounted(nfs_dev));
853 	mutex_exit(&nfs_minor_lock);
854 
855 	vfsp->vfs_dev = nfs_dev;
856 	vfs_make_fsid(&vfsp->vfs_fsid, nfs_dev, nfs3fstyp);
857 	vfsp->vfs_data = (caddr_t)mi;
858 	vfsp->vfs_fstype = nfsfstyp;
859 
860 	/*
861 	 * Verify that nfs3_bsize tuneable is set to an
862 	 * acceptable value.  It be a multiple of PAGESIZE or
863 	 * file corruption can occur.
864 	 */
865 	if (nfs3_bsize & PAGEOFFSET)
866 		nfs3_bsize &= PAGEMASK;
867 	if (nfs3_bsize < PAGESIZE)
868 		nfs3_bsize = PAGESIZE;
869 	vfsp->vfs_bsize = nfs3_bsize;
870 
871 	/*
872 	 * Initialize fields used to support async putpage operations.
873 	 */
874 	for (i = 0; i < NFS_ASYNC_TYPES; i++)
875 		mi->mi_async_clusters[i] = nfs3_async_clusters;
876 	mi->mi_async_init_clusters = nfs3_async_clusters;
877 	mi->mi_async_curr = &mi->mi_async_reqs[0];
878 	mi->mi_max_threads = nfs3_max_threads;
879 	mutex_init(&mi->mi_async_lock, NULL, MUTEX_DEFAULT, NULL);
880 	cv_init(&mi->mi_async_reqs_cv, NULL, CV_DEFAULT, NULL);
881 	cv_init(&mi->mi_async_work_cv, NULL, CV_DEFAULT, NULL);
882 	cv_init(&mi->mi_async_cv, NULL, CV_DEFAULT, NULL);
883 
884 	mi->mi_vfsp = vfsp;
885 	zone_hold(mi->mi_zone = zone);
886 	nfs_mi_zonelist_add(mi);
887 
888 	/*
889 	 * Make the root vnode, use it to get attributes,
890 	 * then remake it with the attributes.
891 	 */
892 	rtvp = makenfs3node((nfs_fh3 *)&svp->sv_fhandle,
893 	    NULL, vfsp, gethrtime(), cr, NULL, NULL);
894 
895 	/*
896 	 * Make the FSINFO calls, primarily at this point to
897 	 * determine the transfer size.  For client failover,
898 	 * we'll want this to be the minimum bid from any
899 	 * server, so that we don't overrun stated limits.
900 	 *
901 	 * While we're looping, we'll turn off AUTH_F_TRYNONE,
902 	 * which is only for the mount operation.
903 	 */
904 
905 	mi->mi_tsize = nfs3_tsize(svp->sv_knconf);
906 	mi->mi_stsize = mi->mi_tsize;
907 
908 	mi->mi_curread = nfs3_bsize;
909 	mi->mi_curwrite = mi->mi_curread;
910 
911 	/*
912 	 * If the uid is set then set the creds for secure mounts
913 	 * by proxy processes such as automountd.
914 	 */
915 	if (svp->sv_secdata->uid != 0 &&
916 	    svp->sv_secdata->rpcflavor == RPCSEC_GSS) {
917 		lcr = crdup(cr);
918 		(void) crsetugid(lcr, svp->sv_secdata->uid, crgetgid(cr));
919 		tcr = lcr;
920 	}
921 
922 	for (svp = mi->mi_servers; svp != NULL; svp = svp->sv_next) {
923 		douprintf = 1;
924 		mi->mi_curr_serv = svp;
925 		max_transfer_size = nfs3_tsize(svp->sv_knconf);
926 		mi->mi_tsize = MIN(max_transfer_size, mi->mi_tsize);
927 		mi->mi_stsize = MIN(max_transfer_size, mi->mi_stsize);
928 		mi->mi_curread = MIN(max_transfer_size, mi->mi_curread);
929 		mi->mi_curwrite = MIN(max_transfer_size, mi->mi_curwrite);
930 		args.fsroot = *(nfs_fh3 *)&svp->sv_fhandle;
931 
932 		error = rfs3call(mi, NFSPROC3_FSINFO,
933 		    xdr_nfs_fh3, (caddr_t)&args,
934 		    xdr_FSINFO3res, (caddr_t)&res, tcr,
935 		    &douprintf, &res.status, 0, NULL);
936 		if (error)
937 			goto bad;
938 		error = geterrno3(res.status);
939 		if (error)
940 			goto bad;
941 
942 		/* get type of root node */
943 		if (res.resok.obj_attributes.attributes) {
944 			if (res.resok.obj_attributes.attr.type < NF3REG ||
945 			    res.resok.obj_attributes.attr.type > NF3FIFO) {
946 #ifdef DEBUG
947 				zcmn_err(getzoneid(), CE_WARN,
948 			    "NFS3 server %s returned a bad file type for root",
949 				    svp->sv_hostname);
950 #else
951 				zcmn_err(getzoneid(), CE_WARN,
952 			    "NFS server %s returned a bad file type for root",
953 				    svp->sv_hostname);
954 #endif
955 				error = EINVAL;
956 				goto bad;
957 			} else {
958 				if (rtvp->v_type != VNON &&
959 		rtvp->v_type != nf3_to_vt[res.resok.obj_attributes.attr.type]) {
960 #ifdef DEBUG
961 					zcmn_err(getzoneid(), CE_WARN,
962 		"NFS3 server %s returned a different file type for root",
963 					    svp->sv_hostname);
964 #else
965 					zcmn_err(getzoneid(), CE_WARN,
966 		"NFS server %s returned a different file type for root",
967 					    svp->sv_hostname);
968 #endif
969 					error = EINVAL;
970 					goto bad;
971 				}
972 				rtvp->v_type =
973 				nf3_to_vt[res.resok.obj_attributes.attr.type];
974 			}
975 		}
976 
977 		if (res.resok.rtmax != 0) {
978 			mi->mi_tsize = MIN(res.resok.rtmax, mi->mi_tsize);
979 			if (res.resok.rtpref != 0) {
980 				mi->mi_curread = MIN(res.resok.rtpref,
981 						    mi->mi_curread);
982 			} else {
983 				mi->mi_curread = MIN(res.resok.rtmax,
984 						    mi->mi_curread);
985 			}
986 		} else if (res.resok.rtpref != 0) {
987 			mi->mi_tsize = MIN(res.resok.rtpref, mi->mi_tsize);
988 			mi->mi_curread = MIN(res.resok.rtpref, mi->mi_curread);
989 		} else {
990 #ifdef DEBUG
991 			zcmn_err(getzoneid(), CE_WARN,
992 			    "NFS3 server %s returned 0 for read transfer sizes",
993 			    svp->sv_hostname);
994 #else
995 			zcmn_err(getzoneid(), CE_WARN,
996 			    "NFS server %s returned 0 for read transfer sizes",
997 			    svp->sv_hostname);
998 #endif
999 			error = EIO;
1000 			goto bad;
1001 		}
1002 		if (res.resok.wtmax != 0) {
1003 			mi->mi_stsize = MIN(res.resok.wtmax, mi->mi_stsize);
1004 			if (res.resok.wtpref != 0) {
1005 				mi->mi_curwrite = MIN(res.resok.wtpref,
1006 						    mi->mi_curwrite);
1007 			} else {
1008 				mi->mi_curwrite = MIN(res.resok.wtmax,
1009 						    mi->mi_curwrite);
1010 			}
1011 		} else if (res.resok.wtpref != 0) {
1012 			mi->mi_stsize = MIN(res.resok.wtpref, mi->mi_stsize);
1013 			mi->mi_curwrite = MIN(res.resok.wtpref,
1014 					    mi->mi_curwrite);
1015 		} else {
1016 #ifdef DEBUG
1017 			zcmn_err(getzoneid(), CE_WARN,
1018 			"NFS3 server %s returned 0 for write transfer sizes",
1019 			    svp->sv_hostname);
1020 #else
1021 			zcmn_err(getzoneid(), CE_WARN,
1022 			"NFS server %s returned 0 for write transfer sizes",
1023 			    svp->sv_hostname);
1024 #endif
1025 			error = EIO;
1026 			goto bad;
1027 		}
1028 
1029 		/*
1030 		 * These signal the ability of the server to create
1031 		 * hard links and symbolic links, so they really
1032 		 * aren't relevant if there is more than one server.
1033 		 * We'll set them here, though it probably looks odd.
1034 		 */
1035 		if (res.resok.properties & FSF3_LINK)
1036 			mi->mi_flags |= MI_LINK;
1037 		if (res.resok.properties & FSF3_SYMLINK)
1038 			mi->mi_flags |= MI_SYMLINK;
1039 
1040 		/* Pick up smallest non-zero maxfilesize value */
1041 		if (res.resok.maxfilesize) {
1042 			if (mi->mi_maxfilesize) {
1043 				mi->mi_maxfilesize = MIN(mi->mi_maxfilesize,
1044 							res.resok.maxfilesize);
1045 			} else
1046 				mi->mi_maxfilesize = res.resok.maxfilesize;
1047 		}
1048 
1049 		/*
1050 		 * AUTH_F_TRYNONE is only for the mount operation,
1051 		 * so turn it back off.
1052 		 */
1053 		svp->sv_secdata->flags &= ~AUTH_F_TRYNONE;
1054 	}
1055 	mi->mi_curr_serv = mi->mi_servers;
1056 
1057 	/*
1058 	 * Start the thread responsible for handling async worker threads.
1059 	 */
1060 	VFS_HOLD(vfsp);	/* add reference for thread */
1061 	mi->mi_manager_thread = zthread_create(NULL, 0, nfs_async_manager,
1062 					vfsp, 0, minclsyspri);
1063 	ASSERT(mi->mi_manager_thread != NULL);
1064 
1065 	/*
1066 	 * Initialize kstats
1067 	 */
1068 	nfs_mnt_kstat_init(vfsp);
1069 
1070 	/* If we didn't get a type, get one now */
1071 	if (rtvp->v_type == VNON) {
1072 		va.va_mask = AT_ALL;
1073 
1074 		error = nfs3getattr(rtvp, &va, tcr);
1075 		if (error)
1076 			goto bad;
1077 		rtvp->v_type = va.va_type;
1078 	}
1079 
1080 	mi->mi_type = rtvp->v_type;
1081 
1082 	*rtvpp = rtvp;
1083 	if (lcr != NULL)
1084 		crfree(lcr);
1085 
1086 	return (0);
1087 bad:
1088 	/*
1089 	 * An error occurred somewhere, need to clean up...
1090 	 * We need to release our reference to the root vnode and
1091 	 * destroy the mntinfo struct that we just created.
1092 	 */
1093 	if (lcr != NULL)
1094 		crfree(lcr);
1095 	rp = VTOR(rtvp);
1096 	if (rp->r_flags & RHASHED)
1097 		rp_rmhash(rp);
1098 	VN_RELE(rtvp);
1099 	nfs_async_stop(vfsp);
1100 	nfs_async_manager_stop(vfsp);
1101 	if (mi->mi_io_kstats) {
1102 		kstat_delete(mi->mi_io_kstats);
1103 		mi->mi_io_kstats = NULL;
1104 	}
1105 	if (mi->mi_ro_kstats) {
1106 		kstat_delete(mi->mi_ro_kstats);
1107 		mi->mi_ro_kstats = NULL;
1108 	}
1109 	nfs_free_mi(mi);
1110 	*rtvpp = NULL;
1111 	return (error);
1112 }
1113 
1114 /*
1115  * vfs operations
1116  */
1117 static int
1118 nfs3_unmount(vfs_t *vfsp, int flag, cred_t *cr)
1119 {
1120 	mntinfo_t *mi;
1121 	ushort_t omax;
1122 
1123 	if (secpolicy_fs_unmount(cr, vfsp) != 0)
1124 		return (EPERM);
1125 
1126 	mi = VFTOMI(vfsp);
1127 	if (flag & MS_FORCE) {
1128 		vfsp->vfs_flag |= VFS_UNMOUNTED;
1129 		/*
1130 		 * We need to stop the manager thread explicitly; the worker
1131 		 * threads can time out and exit on their own.
1132 		 */
1133 		nfs_async_manager_stop(vfsp);
1134 		destroy_rtable(vfsp, cr);
1135 		if (mi->mi_io_kstats) {
1136 			kstat_delete(mi->mi_io_kstats);
1137 			mi->mi_io_kstats = NULL;
1138 		}
1139 		if (mi->mi_ro_kstats) {
1140 			kstat_delete(mi->mi_ro_kstats);
1141 			mi->mi_ro_kstats = NULL;
1142 		}
1143 		return (0);
1144 	}
1145 	/*
1146 	 * Wait until all asynchronous putpage operations on
1147 	 * this file system are complete before flushing rnodes
1148 	 * from the cache.
1149 	 */
1150 	omax = mi->mi_max_threads;
1151 	if (nfs_async_stop_sig(vfsp)) {
1152 		return (EINTR);
1153 	}
1154 	rflush(vfsp, cr);
1155 	/*
1156 	 * If there are any active vnodes on this file system,
1157 	 * then the file system is busy and can't be umounted.
1158 	 */
1159 	if (check_rtable(vfsp)) {
1160 		mutex_enter(&mi->mi_async_lock);
1161 		mi->mi_max_threads = omax;
1162 		mutex_exit(&mi->mi_async_lock);
1163 		return (EBUSY);
1164 	}
1165 	/*
1166 	 * The unmount can't fail from now on; stop the worker thread manager.
1167 	 */
1168 	nfs_async_manager_stop(vfsp);
1169 	/*
1170 	 * Destroy all rnodes belonging to this file system from the
1171 	 * rnode hash queues and purge any resources allocated to
1172 	 * them.
1173 	 */
1174 	destroy_rtable(vfsp, cr);
1175 	if (mi->mi_io_kstats) {
1176 		kstat_delete(mi->mi_io_kstats);
1177 		mi->mi_io_kstats = NULL;
1178 	}
1179 	if (mi->mi_ro_kstats) {
1180 		kstat_delete(mi->mi_ro_kstats);
1181 		mi->mi_ro_kstats = NULL;
1182 	}
1183 	return (0);
1184 }
1185 
1186 /*
1187  * find root of nfs
1188  */
1189 static int
1190 nfs3_root(vfs_t *vfsp, vnode_t **vpp)
1191 {
1192 	mntinfo_t *mi;
1193 	vnode_t *vp;
1194 	servinfo_t *svp;
1195 
1196 	mi = VFTOMI(vfsp);
1197 
1198 	if (nfs_zone() != mi->mi_zone)
1199 		return (EPERM);
1200 
1201 	svp = mi->mi_curr_serv;
1202 	if (svp && (svp->sv_flags & SV_ROOT_STALE)) {
1203 		mutex_enter(&svp->sv_lock);
1204 		svp->sv_flags &= ~SV_ROOT_STALE;
1205 		mutex_exit(&svp->sv_lock);
1206 		return (ENOENT);
1207 	}
1208 
1209 	vp = makenfs3node((nfs_fh3 *)&mi->mi_curr_serv->sv_fhandle,
1210 	    NULL, vfsp, gethrtime(), CRED(), NULL, NULL);
1211 
1212 	if (VTOR(vp)->r_flags & RSTALE) {
1213 		VN_RELE(vp);
1214 		return (ENOENT);
1215 	}
1216 
1217 	ASSERT(vp->v_type == VNON || vp->v_type == mi->mi_type);
1218 
1219 	vp->v_type = mi->mi_type;
1220 
1221 	*vpp = vp;
1222 
1223 	return (0);
1224 }
1225 
1226 /*
1227  * Get file system statistics.
1228  */
1229 static int
1230 nfs3_statvfs(vfs_t *vfsp, struct statvfs64 *sbp)
1231 {
1232 	int error;
1233 	struct mntinfo *mi;
1234 	struct FSSTAT3args args;
1235 	struct FSSTAT3res res;
1236 	int douprintf;
1237 	failinfo_t fi;
1238 	vnode_t *vp;
1239 	cred_t *cr;
1240 	hrtime_t t;
1241 
1242 	mi = VFTOMI(vfsp);
1243 	if (nfs_zone() != mi->mi_zone)
1244 		return (EPERM);
1245 	error = nfs3_root(vfsp, &vp);
1246 	if (error)
1247 		return (error);
1248 
1249 	cr = CRED();
1250 
1251 	args.fsroot = *VTOFH3(vp);
1252 	fi.vp = vp;
1253 	fi.fhp = (caddr_t)&args.fsroot;
1254 	fi.copyproc = nfs3copyfh;
1255 	fi.lookupproc = nfs3lookup;
1256 	fi.xattrdirproc = acl_getxattrdir3;
1257 
1258 	douprintf = 1;
1259 
1260 	t = gethrtime();
1261 
1262 	error = rfs3call(mi, NFSPROC3_FSSTAT,
1263 	    xdr_nfs_fh3, (caddr_t)&args,
1264 	    xdr_FSSTAT3res, (caddr_t)&res, cr,
1265 	    &douprintf, &res.status, 0, &fi);
1266 
1267 	if (error) {
1268 		VN_RELE(vp);
1269 		return (error);
1270 	}
1271 
1272 	error = geterrno3(res.status);
1273 	if (!error) {
1274 		nfs3_cache_post_op_attr(vp, &res.resok.obj_attributes, t, cr);
1275 		sbp->f_bsize = MAXBSIZE;
1276 		sbp->f_frsize = DEV_BSIZE;
1277 		/*
1278 		 * Allow -1 fields to pass through unconverted.  These
1279 		 * indicate "don't know" fields.
1280 		 */
1281 		if (res.resok.tbytes == (size3)-1)
1282 			sbp->f_blocks = (fsblkcnt64_t)res.resok.tbytes;
1283 		else {
1284 			sbp->f_blocks = (fsblkcnt64_t)
1285 			    (res.resok.tbytes / DEV_BSIZE);
1286 		}
1287 		if (res.resok.fbytes == (size3)-1)
1288 			sbp->f_bfree = (fsblkcnt64_t)res.resok.fbytes;
1289 		else {
1290 			sbp->f_bfree = (fsblkcnt64_t)
1291 			    (res.resok.fbytes / DEV_BSIZE);
1292 		}
1293 		if (res.resok.abytes == (size3)-1)
1294 			sbp->f_bavail = (fsblkcnt64_t)res.resok.abytes;
1295 		else {
1296 			sbp->f_bavail = (fsblkcnt64_t)
1297 			    (res.resok.abytes / DEV_BSIZE);
1298 		}
1299 		sbp->f_files = (fsfilcnt64_t)res.resok.tfiles;
1300 		sbp->f_ffree = (fsfilcnt64_t)res.resok.ffiles;
1301 		sbp->f_favail = (fsfilcnt64_t)res.resok.afiles;
1302 		sbp->f_fsid = (unsigned long)vfsp->vfs_fsid.val[0];
1303 		(void) strncpy(sbp->f_basetype,
1304 		    vfssw[vfsp->vfs_fstype].vsw_name, FSTYPSZ);
1305 		sbp->f_flag = vf_to_stf(vfsp->vfs_flag);
1306 		sbp->f_namemax = (ulong_t)-1;
1307 	} else {
1308 		nfs3_cache_post_op_attr(vp, &res.resfail.obj_attributes, t, cr);
1309 		PURGE_STALE_FH(error, vp, cr);
1310 	}
1311 
1312 	VN_RELE(vp);
1313 
1314 	return (error);
1315 }
1316 
1317 static kmutex_t nfs3_syncbusy;
1318 
1319 /*
1320  * Flush dirty nfs files for file system vfsp.
1321  * If vfsp == NULL, all nfs files are flushed.
1322  */
1323 /* ARGSUSED */
1324 static int
1325 nfs3_sync(vfs_t *vfsp, short flag, cred_t *cr)
1326 {
1327 	/*
1328 	 * Cross-zone calls are OK here, since this translates to a
1329 	 * VOP_PUTPAGE(B_ASYNC), which gets picked up by the right zone.
1330 	 */
1331 	if (!(flag & SYNC_ATTR) && mutex_tryenter(&nfs3_syncbusy) != 0) {
1332 		rflush(vfsp, cr);
1333 		mutex_exit(&nfs3_syncbusy);
1334 	}
1335 	return (0);
1336 }
1337 
1338 /* ARGSUSED */
1339 static int
1340 nfs3_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1341 {
1342 	int error;
1343 	nfs_fh3 fh;
1344 	vnode_t *vp;
1345 	struct vattr va;
1346 
1347 	if (fidp->fid_len > NFS3_FHSIZE) {
1348 		*vpp = NULL;
1349 		return (ESTALE);
1350 	}
1351 
1352 	if (nfs_zone() != VFTOMI(vfsp)->mi_zone)
1353 		return (EPERM);
1354 	fh.fh3_length = fidp->fid_len;
1355 	bcopy(fidp->fid_data, fh.fh3_u.data, fh.fh3_length);
1356 
1357 	vp = makenfs3node(&fh, NULL, vfsp, gethrtime(), CRED(), NULL, NULL);
1358 
1359 	if (VTOR(vp)->r_flags & RSTALE) {
1360 		VN_RELE(vp);
1361 		*vpp = NULL;
1362 		return (ENOENT);
1363 	}
1364 
1365 	if (vp->v_type == VNON) {
1366 		va.va_mask = AT_ALL;
1367 		error = nfs3getattr(vp, &va, CRED());
1368 		if (error) {
1369 			VN_RELE(vp);
1370 			*vpp = NULL;
1371 			return (error);
1372 		}
1373 		vp->v_type = va.va_type;
1374 	}
1375 
1376 	*vpp = vp;
1377 
1378 	return (0);
1379 }
1380 
1381 /* ARGSUSED */
1382 static int
1383 nfs3_mountroot(vfs_t *vfsp, whymountroot_t why)
1384 {
1385 	vnode_t *rtvp;
1386 	char root_hostname[SYS_NMLN+1];
1387 	struct servinfo *svp;
1388 	int error;
1389 	int vfsflags;
1390 	size_t size;
1391 	char *root_path;
1392 	struct pathname pn;
1393 	char *name;
1394 	cred_t *cr;
1395 	struct nfs_args args;		/* nfs mount arguments */
1396 	static char token[10];
1397 
1398 	bzero(&args, sizeof (args));
1399 
1400 	/* do this BEFORE getfile which causes xid stamps to be initialized */
1401 	clkset(-1L);		/* hack for now - until we get time svc? */
1402 
1403 	if (why == ROOT_REMOUNT) {
1404 		/*
1405 		 * Shouldn't happen.
1406 		 */
1407 		panic("nfs3_mountroot: why == ROOT_REMOUNT");
1408 	}
1409 
1410 	if (why == ROOT_UNMOUNT) {
1411 		/*
1412 		 * Nothing to do for NFS.
1413 		 */
1414 		return (0);
1415 	}
1416 
1417 	/*
1418 	 * why == ROOT_INIT
1419 	 */
1420 
1421 	name = token;
1422 	*name = 0;
1423 	getfsname("root", name, sizeof (token));
1424 
1425 	pn_alloc(&pn);
1426 	root_path = pn.pn_path;
1427 
1428 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
1429 	svp->sv_knconf = kmem_zalloc(sizeof (*svp->sv_knconf), KM_SLEEP);
1430 	svp->sv_knconf->knc_protofmly = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
1431 	svp->sv_knconf->knc_proto = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
1432 
1433 	/*
1434 	 * Get server address
1435 	 * Get the root fhandle
1436 	 * Get server's transport
1437 	 * Get server's hostname
1438 	 * Get options
1439 	 */
1440 	args.addr = &svp->sv_addr;
1441 	args.fh = (char *)&svp->sv_fhandle;
1442 	args.knconf = svp->sv_knconf;
1443 	args.hostname = root_hostname;
1444 	vfsflags = 0;
1445 	if (error = mount_root(*name ? name : "root", root_path, NFS_V3,
1446 	    &args, &vfsflags)) {
1447 		if (error == EPROTONOSUPPORT)
1448 			nfs_cmn_err(error, CE_WARN, "nfs3_mountroot: "
1449 			    "mount_root failed: server doesn't support NFS V3");
1450 		else
1451 			nfs_cmn_err(error, CE_WARN,
1452 			    "nfs3_mountroot: mount_root failed: %m");
1453 		sv_free(svp);
1454 		pn_free(&pn);
1455 		return (error);
1456 	}
1457 	svp->sv_hostnamelen = (int)(strlen(root_hostname) + 1);
1458 	svp->sv_hostname = kmem_alloc(svp->sv_hostnamelen, KM_SLEEP);
1459 	(void) strcpy(svp->sv_hostname, root_hostname);
1460 
1461 	/*
1462 	 * Force root partition to always be mounted with AUTH_UNIX for now
1463 	 */
1464 	svp->sv_secdata = kmem_alloc(sizeof (*svp->sv_secdata), KM_SLEEP);
1465 	svp->sv_secdata->secmod = AUTH_UNIX;
1466 	svp->sv_secdata->rpcflavor = AUTH_UNIX;
1467 	svp->sv_secdata->data = NULL;
1468 
1469 	cr = crgetcred();
1470 	rtvp = NULL;
1471 
1472 	error = nfs3rootvp(&rtvp, vfsp, svp, args.flags, cr, global_zone);
1473 
1474 	crfree(cr);
1475 
1476 	if (error) {
1477 		pn_free(&pn);
1478 		goto errout;
1479 	}
1480 
1481 	error = nfs_setopts(rtvp, DATAMODEL_NATIVE, &args);
1482 	if (error) {
1483 		nfs_cmn_err(error, CE_WARN,
1484 		    "nfs3_mountroot: invalid root mount options");
1485 		pn_free(&pn);
1486 		goto errout;
1487 	}
1488 
1489 	(void) vfs_lock_wait(vfsp);
1490 	vfs_add(NULL, vfsp, vfsflags);
1491 	vfs_unlock(vfsp);
1492 
1493 	size = strlen(svp->sv_hostname);
1494 	(void) strcpy(rootfs.bo_name, svp->sv_hostname);
1495 	rootfs.bo_name[size] = ':';
1496 	(void) strcpy(&rootfs.bo_name[size + 1], root_path);
1497 
1498 	pn_free(&pn);
1499 
1500 errout:
1501 	if (error) {
1502 		sv_free(svp);
1503 		nfs_async_stop(vfsp);
1504 		nfs_async_manager_stop(vfsp);
1505 	}
1506 
1507 	if (rtvp != NULL)
1508 		VN_RELE(rtvp);
1509 
1510 	return (error);
1511 }
1512 
1513 /*
1514  * Initialization routine for VFS routines.  Should only be called once
1515  */
1516 int
1517 nfs3_vfsinit(void)
1518 {
1519 	mutex_init(&nfs3_syncbusy, NULL, MUTEX_DEFAULT, NULL);
1520 	return (0);
1521 }
1522 
1523 void
1524 nfs3_vfsfini(void)
1525 {
1526 	mutex_destroy(&nfs3_syncbusy);
1527 }
1528 
1529 void
1530 nfs3_freevfs(vfs_t *vfsp)
1531 {
1532 	mntinfo_t *mi;
1533 	servinfo_t *svp;
1534 
1535 	/* free up the resources */
1536 	mi = VFTOMI(vfsp);
1537 	svp = mi->mi_servers;
1538 	mi->mi_servers = mi->mi_curr_serv = NULL;
1539 	sv_free(svp);
1540 
1541 	/*
1542 	 * By this time we should have already deleted the
1543 	 * mi kstats in the unmount code. If they are still around
1544 	 * somethings wrong
1545 	 */
1546 	ASSERT(mi->mi_io_kstats == NULL);
1547 	nfs_free_mi(mi);
1548 }
1549