xref: /titanic_44/usr/src/uts/common/fs/nfs/nfs_vfsops.c (revision 28167c24ba5be8b7c1d05e02d053f4a55cd21cc9)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  *
26  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
27  *	All rights reserved.
28  */
29 
30 #pragma ident	"%Z%%M%	%I%	%E% SMI"
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/vnode.h>
38 #include <sys/pathname.h>
39 #include <sys/sysmacros.h>
40 #include <sys/kmem.h>
41 #include <sys/mkdev.h>
42 #include <sys/mount.h>
43 #include <sys/mntent.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/policy.h>
54 #include <sys/zone.h>
55 #include <sys/class.h>
56 #include <sys/socket.h>
57 #include <sys/netconfig.h>
58 
59 #include <rpc/types.h>
60 #include <rpc/auth.h>
61 #include <rpc/clnt.h>
62 
63 #include <nfs/nfs.h>
64 #include <nfs/nfs_clnt.h>
65 #include <nfs/rnode.h>
66 #include <nfs/mount.h>
67 #include <nfs/nfs_acl.h>
68 
69 #include <fs/fs_subr.h>
70 
71 /*
72  * From rpcsec module (common/rpcsec).
73  */
74 extern int sec_clnt_loadinfo(struct sec_data *, struct sec_data **, model_t);
75 extern void sec_clnt_freeinfo(struct sec_data *);
76 
77 static int pathconf_get(struct mntinfo *, model_t, struct nfs_args *);
78 static void pathconf_rele(struct mntinfo *);
79 
80 /*
81  * The order and contents of this structure must be kept in sync with that of
82  * rfsreqcnt_v2_tmpl in nfs_stats.c
83  */
84 static char *rfsnames_v2[] = {
85 	"null", "getattr", "setattr", "unused", "lookup", "readlink", "read",
86 	"unused", "write", "create", "remove", "rename", "link", "symlink",
87 	"mkdir", "rmdir", "readdir", "fsstat"
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_v2[] = {
98 	0, 0, 1, 0, 0, 0, 1,
99 	0, 2, 2, 2, 2, 2, 2,
100 	2, 2, 1, 0
101 };
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_v2[] = {
108 	0, 0, 0, 0, 0, 0, 1,
109 	0, 2, 0, 0, 0, 0, 0,
110 	0, 0, 1, 0
111 };
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_v2[] = {
120 	0, 0, 1, 0, 0, 0, 0,
121 	0, 1, 1, 1, 1, 1, 1,
122 	1, 1, 0, 0
123 };
124 
125 /*
126  * nfs vfs operations.
127  */
128 static int	nfs_mount(vfs_t *, vnode_t *, struct mounta *, cred_t *);
129 static int	nfs_unmount(vfs_t *, int, cred_t *);
130 static int	nfs_root(vfs_t *, vnode_t **);
131 static int	nfs_statvfs(vfs_t *, struct statvfs64 *);
132 static int	nfs_sync(vfs_t *, short, cred_t *);
133 static int	nfs_vget(vfs_t *, vnode_t **, fid_t *);
134 static int	nfs_mountroot(vfs_t *, whymountroot_t);
135 static void	nfs_freevfs(vfs_t *);
136 
137 static int	nfsrootvp(vnode_t **, vfs_t *, struct servinfo *,
138 		    int, cred_t *, zone_t *);
139 
140 /*
141  * Initialize the vfs structure
142  */
143 
144 int nfsfstyp;
145 vfsops_t *nfs_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 int rdma_debug = 0;
153 
154 int
155 nfsinit(int fstyp, char *name)
156 {
157 	static const fs_operation_def_t nfs_vfsops_template[] = {
158 		VFSNAME_MOUNT, nfs_mount,
159 		VFSNAME_UNMOUNT, nfs_unmount,
160 		VFSNAME_ROOT, nfs_root,
161 		VFSNAME_STATVFS, nfs_statvfs,
162 		VFSNAME_SYNC, (fs_generic_func_p) nfs_sync,
163 		VFSNAME_VGET, nfs_vget,
164 		VFSNAME_MOUNTROOT, nfs_mountroot,
165 		VFSNAME_FREEVFS, (fs_generic_func_p)nfs_freevfs,
166 		NULL, NULL
167 	};
168 	int error;
169 
170 	error = vfs_setfsops(fstyp, nfs_vfsops_template, &nfs_vfsops);
171 	if (error != 0) {
172 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
173 		    "nfsinit: bad vfs ops template");
174 		return (error);
175 	}
176 
177 	error = vn_make_ops(name, nfs_vnodeops_template, &nfs_vnodeops);
178 	if (error != 0) {
179 		(void) vfs_freevfsops_by_type(fstyp);
180 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
181 		    "nfsinit: bad vnode ops template");
182 		return (error);
183 	}
184 
185 
186 	nfsfstyp = fstyp;
187 
188 	return (0);
189 }
190 
191 void
192 nfsfini(void)
193 {
194 }
195 
196 /*
197  * nfs mount vfsop
198  * Set up mount info record and attach it to vfs struct.
199  */
200 static int
201 nfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
202 {
203 	char *data = uap->dataptr;
204 	int error;
205 	vnode_t *rtvp;			/* the server's root */
206 	mntinfo_t *mi;			/* mount info, pointed at by vfs */
207 	size_t hlen;			/* length of hostname */
208 	size_t nlen;			/* length of netname */
209 	char netname[SYS_NMLN];		/* server's netname */
210 	struct netbuf addr;		/* server's address */
211 	struct netbuf syncaddr;		/* AUTH_DES time sync addr */
212 	struct knetconfig *knconf;	/* transport knetconfig structure */
213 	struct knetconfig *rdma_knconf;	/* rdma transport structure */
214 	rnode_t *rp;
215 	struct servinfo *svp;		/* nfs server info */
216 	struct servinfo *svp_tail = NULL; /* previous nfs server info */
217 	struct servinfo *svp_head;	/* first nfs server info */
218 	struct servinfo *svp_2ndlast;	/* 2nd last in the server info list */
219 	struct sec_data *secdata;	/* security data */
220 	STRUCT_DECL(nfs_args, args);	/* nfs mount arguments */
221 	STRUCT_DECL(knetconfig, knconf_tmp);
222 	STRUCT_DECL(netbuf, addr_tmp);
223 	int flags, addr_type;
224 	char *p, *pf;
225 	zone_t *zone = nfs_zone();
226 
227 	if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
228 		return (error);
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 		 * Remounts need to save the pathconf information.
279 		 * Part of the infamous static kludge.
280 		 */
281 
282 		if ((mi = VFTOMI(vfsp)) != NULL) {
283 			uint_t new_mi_llock;
284 			uint_t old_mi_llock;
285 
286 			new_mi_llock = (flags & NFSMNT_LLOCK) ? 1 : 0;
287 			old_mi_llock = (mi->mi_flags & MI_LLOCK) ? 1 : 0;
288 			if (old_mi_llock != new_mi_llock)
289 				return (EBUSY);
290 		}
291 		return (pathconf_get((struct mntinfo *)vfsp->vfs_data,
292 		    get_udatamodel(), STRUCT_BUF(args)));
293 	}
294 
295 	mutex_enter(&mvp->v_lock);
296 	if (!(uap->flags & MS_OVERLAY) &&
297 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
298 		mutex_exit(&mvp->v_lock);
299 		return (EBUSY);
300 	}
301 	mutex_exit(&mvp->v_lock);
302 
303 	/* make sure things are zeroed for errout: */
304 	rtvp = NULL;
305 	mi = NULL;
306 	addr.buf = NULL;
307 	syncaddr.buf = NULL;
308 	secdata = NULL;
309 
310 	/*
311 	 * A valid knetconfig structure is required.
312 	 */
313 	if (!(flags & NFSMNT_KNCONF))
314 		return (EINVAL);
315 
316 	/*
317 	 * Allocate a servinfo struct.
318 	 */
319 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
320 	mutex_init(&svp->sv_lock, NULL, MUTEX_DEFAULT, NULL);
321 	if (svp_tail) {
322 		svp_2ndlast = svp_tail;
323 		svp_tail->sv_next = svp;
324 	} else {
325 		svp_head = svp;
326 		svp_2ndlast = svp;
327 	}
328 
329 	svp_tail = svp;
330 
331 	/*
332 	 * Allocate space for a knetconfig structure and
333 	 * its strings and copy in from user-land.
334 	 */
335 	knconf = kmem_zalloc(sizeof (*knconf), KM_SLEEP);
336 	svp->sv_knconf = knconf;
337 	STRUCT_INIT(knconf_tmp, get_udatamodel());
338 	if (copyin(STRUCT_FGETP(args, knconf), STRUCT_BUF(knconf_tmp),
339 	    STRUCT_SIZE(knconf_tmp))) {
340 		sv_free(svp_head);
341 		return (EFAULT);
342 	}
343 
344 	knconf->knc_semantics = STRUCT_FGET(knconf_tmp, knc_semantics);
345 	knconf->knc_protofmly = STRUCT_FGETP(knconf_tmp, knc_protofmly);
346 	knconf->knc_proto = STRUCT_FGETP(knconf_tmp, knc_proto);
347 	if (get_udatamodel() != DATAMODEL_LP64) {
348 		knconf->knc_rdev = expldev(STRUCT_FGET(knconf_tmp, knc_rdev));
349 	} else {
350 		knconf->knc_rdev = STRUCT_FGET(knconf_tmp, knc_rdev);
351 	}
352 
353 	pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
354 	p = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
355 	error = copyinstr(knconf->knc_protofmly, pf, KNC_STRSIZE, NULL);
356 	if (error) {
357 		kmem_free(pf, KNC_STRSIZE);
358 		kmem_free(p, KNC_STRSIZE);
359 		sv_free(svp_head);
360 		return (error);
361 	}
362 	error = copyinstr(knconf->knc_proto, p, KNC_STRSIZE, NULL);
363 	if (error) {
364 		kmem_free(pf, KNC_STRSIZE);
365 		kmem_free(p, KNC_STRSIZE);
366 		sv_free(svp_head);
367 		return (error);
368 	}
369 	knconf->knc_protofmly = pf;
370 	knconf->knc_proto = p;
371 
372 	/*
373 	 * Get server address
374 	 */
375 	STRUCT_INIT(addr_tmp, get_udatamodel());
376 	if (copyin(STRUCT_FGETP(args, addr), STRUCT_BUF(addr_tmp),
377 	    STRUCT_SIZE(addr_tmp))) {
378 		addr.buf = NULL;
379 		error = EFAULT;
380 	} else {
381 		char *userbufptr;
382 
383 		userbufptr = addr.buf = STRUCT_FGETP(addr_tmp, buf);
384 		addr.len = STRUCT_FGET(addr_tmp, len);
385 		addr.buf = kmem_alloc(addr.len, KM_SLEEP);
386 		addr.maxlen = addr.len;
387 		if (copyin(userbufptr, addr.buf, addr.len))
388 			error = EFAULT;
389 	}
390 	svp->sv_addr = addr;
391 	if (error)
392 		goto errout;
393 
394 	/*
395 	 * Get the root fhandle
396 	 */
397 	if (copyin(STRUCT_FGETP(args, fh), &(svp->sv_fhandle.fh_buf),
398 	    NFS_FHSIZE)) {
399 		error = EFAULT;
400 		goto errout;
401 	}
402 	svp->sv_fhandle.fh_len = NFS_FHSIZE;
403 
404 	/*
405 	 * Get server's hostname
406 	 */
407 	if (flags & NFSMNT_HOSTNAME) {
408 		error = copyinstr(STRUCT_FGETP(args, hostname),
409 		    netname, sizeof (netname), &hlen);
410 		if (error)
411 			goto errout;
412 	} else {
413 		char *p = "unknown-host";
414 		hlen = strlen(p) + 1;
415 		(void) strcpy(netname, p);
416 	}
417 	svp->sv_hostnamelen = hlen;
418 	svp->sv_hostname = kmem_alloc(svp->sv_hostnamelen, KM_SLEEP);
419 	(void) strcpy(svp->sv_hostname, netname);
420 
421 	/*
422 	 * RDMA MOUNT SUPPORT FOR NFS v2:
423 	 * Establish, is it possible to use RDMA, if so overload the
424 	 * knconf with rdma specific knconf and free the orignal.
425 	 */
426 	if ((flags & NFSMNT_TRYRDMA) || (flags & NFSMNT_DORDMA)) {
427 		/*
428 		 * Determine the addr type for RDMA, IPv4 or v6.
429 		 */
430 		if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET) == 0)
431 			addr_type = AF_INET;
432 		else if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET6) == 0)
433 			addr_type = AF_INET6;
434 
435 		if (rdma_reachable(addr_type, &svp->sv_addr,
436 			&rdma_knconf) == 0) {
437 			/*
438 			 * If successful, hijack, the orignal knconf and
439 			 * replace with a new one, depending on the flags.
440 			 */
441 			svp->sv_origknconf = svp->sv_knconf;
442 			svp->sv_knconf = rdma_knconf;
443 			knconf = rdma_knconf;
444 		} else {
445 			if (flags & NFSMNT_TRYRDMA) {
446 #ifdef	DEBUG
447 				if (rdma_debug)
448 					zcmn_err(getzoneid(), CE_WARN,
449 					    "no RDMA onboard, revert\n");
450 #endif
451 			}
452 
453 			if (flags & NFSMNT_DORDMA) {
454 				/*
455 				 * If proto=rdma is specified and no RDMA
456 				 * path to this server is avialable then
457 				 * ditch this server.
458 				 * This is not included in the mountable
459 				 * server list or the replica list.
460 				 * Check if more servers are specified;
461 				 * Failover case, otherwise bail out of mount.
462 				 */
463 				if (STRUCT_FGET(args, nfs_args_ext) ==
464 				    NFS_ARGS_EXTB && STRUCT_FGETP(args,
465 					nfs_ext_u.nfs_extB.next) != NULL) {
466 					if (uap->flags & MS_RDONLY &&
467 					    !(flags & NFSMNT_SOFT)) {
468 						data = (char *)
469 						    STRUCT_FGETP(args,
470 						nfs_ext_u.nfs_extB.next);
471 						if (svp_head->sv_next == NULL) {
472 							svp_tail = NULL;
473 							svp_2ndlast = NULL;
474 							sv_free(svp_head);
475 							goto more;
476 						} else {
477 							svp_tail = svp_2ndlast;
478 							svp_2ndlast->sv_next =
479 							    NULL;
480 							sv_free(svp);
481 							goto more;
482 						}
483 					}
484 				} else {
485 					/*
486 					 * This is the last server specified
487 					 * in the nfs_args list passed down
488 					 * and its not rdma capable.
489 					 */
490 					if (svp_head->sv_next == NULL) {
491 						/*
492 						 * Is this the only one
493 						 */
494 						error = EINVAL;
495 #ifdef	DEBUG
496 						if (rdma_debug)
497 							zcmn_err(getzoneid(),
498 							    CE_WARN,
499 							    "No RDMA srv");
500 #endif
501 						goto errout;
502 					} else {
503 						/*
504 						 * There is list, since some
505 						 * servers specified before
506 						 * this passed all requirements
507 						 */
508 						svp_tail = svp_2ndlast;
509 						svp_2ndlast->sv_next = NULL;
510 						sv_free(svp);
511 						goto proceed;
512 					}
513 				}
514 			}
515 		}
516 	}
517 
518 	/*
519 	 * Get the extention data which has the new security data structure.
520 	 */
521 	if (flags & NFSMNT_NEWARGS) {
522 		switch (STRUCT_FGET(args, nfs_args_ext)) {
523 		case NFS_ARGS_EXTA:
524 		case NFS_ARGS_EXTB:
525 			/*
526 			 * Indicating the application is using the new
527 			 * sec_data structure to pass in the security
528 			 * data.
529 			 */
530 			if (STRUCT_FGETP(args,
531 			    nfs_ext_u.nfs_extA.secdata) == NULL) {
532 				error = EINVAL;
533 			} else {
534 				error = sec_clnt_loadinfo(
535 				    (struct sec_data *)STRUCT_FGETP(args,
536 					nfs_ext_u.nfs_extA.secdata),
537 				    &secdata, get_udatamodel());
538 			}
539 			break;
540 
541 		default:
542 			error = EINVAL;
543 			break;
544 		}
545 	} else if (flags & NFSMNT_SECURE) {
546 		/*
547 		 * Keep this for backward compatibility to support
548 		 * NFSMNT_SECURE/NFSMNT_RPCTIMESYNC flags.
549 		 */
550 		if (STRUCT_FGETP(args, syncaddr) == NULL) {
551 			error = EINVAL;
552 		} else {
553 			/*
554 			 * get time sync address.
555 			 */
556 			if (copyin(STRUCT_FGETP(args, syncaddr), &addr_tmp,
557 			    STRUCT_SIZE(addr_tmp))) {
558 				syncaddr.buf = NULL;
559 				error = EFAULT;
560 			} else {
561 				char *userbufptr;
562 
563 				userbufptr = syncaddr.buf =
564 				    STRUCT_FGETP(addr_tmp, buf);
565 				syncaddr.len =
566 				    STRUCT_FGET(addr_tmp, len);
567 				syncaddr.buf = kmem_alloc(syncaddr.len,
568 				    KM_SLEEP);
569 				syncaddr.maxlen = syncaddr.len;
570 
571 				if (copyin(userbufptr, syncaddr.buf,
572 				    syncaddr.len))
573 					error = EFAULT;
574 			}
575 
576 			/*
577 			 * get server's netname
578 			 */
579 			if (!error) {
580 				error = copyinstr(STRUCT_FGETP(args, netname),
581 				    netname, sizeof (netname), &nlen);
582 				netname[nlen] = '\0';
583 			}
584 
585 			if (error && syncaddr.buf != NULL) {
586 				kmem_free(syncaddr.buf, syncaddr.len);
587 				syncaddr.buf = NULL;
588 			}
589 		}
590 
591 		/*
592 		 * Move security related data to the sec_data structure.
593 		 */
594 		if (!error) {
595 			dh_k4_clntdata_t *data;
596 			char *pf, *p;
597 
598 			secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
599 			if (flags & NFSMNT_RPCTIMESYNC)
600 				secdata->flags |= AUTH_F_RPCTIMESYNC;
601 			data = kmem_alloc(sizeof (*data), KM_SLEEP);
602 			data->syncaddr = syncaddr;
603 
604 			/*
605 			 * duplicate the knconf information for the
606 			 * new opaque data.
607 			 */
608 			data->knconf = kmem_alloc(sizeof (*knconf), KM_SLEEP);
609 			*data->knconf = *knconf;
610 			pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
611 			p = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
612 			bcopy(knconf->knc_protofmly, pf, KNC_STRSIZE);
613 			bcopy(knconf->knc_proto, pf, KNC_STRSIZE);
614 			data->knconf->knc_protofmly = pf;
615 			data->knconf->knc_proto = p;
616 
617 			/* move server netname to the sec_data structure */
618 			if (nlen != 0) {
619 				data->netname = kmem_alloc(nlen, KM_SLEEP);
620 				bcopy(netname, data->netname, nlen);
621 				data->netnamelen = (int)nlen;
622 			}
623 			secdata->secmod = secdata->rpcflavor = AUTH_DES;
624 			secdata->data = (caddr_t)data;
625 		}
626 	} else {
627 		secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
628 		secdata->secmod = secdata->rpcflavor = AUTH_UNIX;
629 		secdata->data = NULL;
630 	}
631 	svp->sv_secdata = secdata;
632 	if (error)
633 		goto errout;
634 
635 	/*
636 	 * See bug 1180236.
637 	 * If mount secure failed, we will fall back to AUTH_NONE
638 	 * and try again.  nfs3rootvp() will turn this back off.
639 	 *
640 	 * The NFS Version 2 mount uses GETATTR and STATFS procedures.
641 	 * The server does not care if these procedures have the proper
642 	 * authentication flavor, so if mount retries using AUTH_NONE
643 	 * that does not require a credential setup for root then the
644 	 * automounter would work without requiring root to be
645 	 * keylogged into AUTH_DES.
646 	 */
647 	if (secdata->rpcflavor != AUTH_UNIX &&
648 	    secdata->rpcflavor != AUTH_LOOPBACK)
649 		secdata->flags |= AUTH_F_TRYNONE;
650 
651 	/*
652 	 * Failover support:
653 	 *
654 	 * We may have a linked list of nfs_args structures,
655 	 * which means the user is looking for failover.  If
656 	 * the mount is either not "read-only" or "soft",
657 	 * we want to bail out with EINVAL.
658 	 */
659 	if (STRUCT_FGET(args, nfs_args_ext) == NFS_ARGS_EXTB &&
660 	    STRUCT_FGETP(args, nfs_ext_u.nfs_extB.next) != NULL) {
661 		if (uap->flags & MS_RDONLY && !(flags & NFSMNT_SOFT)) {
662 			data = (char *)STRUCT_FGETP(args,
663 			    nfs_ext_u.nfs_extB.next);
664 			goto more;
665 		}
666 		error = EINVAL;
667 		goto errout;
668 	}
669 
670 	/*
671 	 * Determine the zone we're being mounted into.
672 	 */
673 	if (getzoneid() == GLOBAL_ZONEID) {
674 		zone_t *mntzone;
675 
676 		mntzone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
677 		ASSERT(mntzone != NULL);
678 		zone_rele(mntzone);
679 		if (mntzone != zone) {
680 			error = EBUSY;
681 			goto errout;
682 		}
683 	}
684 
685 	/*
686 	 * Stop the mount from going any further if the zone is going away.
687 	 */
688 	if (zone_status_get(curproc->p_zone) >= ZONE_IS_SHUTTING_DOWN) {
689 		error = EBUSY;
690 		goto errout;
691 	}
692 
693 	/*
694 	 * Get root vnode.
695 	 */
696 proceed:
697 	error = nfsrootvp(&rtvp, vfsp, svp_head, flags, cr, zone);
698 
699 	if (error)
700 		goto errout;
701 
702 	/*
703 	 * Set option fields in the mount info record
704 	 */
705 	mi = VTOMI(rtvp);
706 
707 	if (svp_head->sv_next)
708 		mi->mi_flags |= MI_LLOCK;
709 
710 	error = nfs_setopts(rtvp, get_udatamodel(), STRUCT_BUF(args));
711 	if (!error) {
712 		/* static pathconf kludge */
713 		error = pathconf_get(mi, get_udatamodel(), STRUCT_BUF(args));
714 	}
715 
716 errout:
717 	if (error) {
718 		if (rtvp != NULL) {
719 			rp = VTOR(rtvp);
720 			if (rp->r_flags & RHASHED)
721 				rp_rmhash(rp);
722 		}
723 		sv_free(svp_head);
724 		if (mi != NULL) {
725 			nfs_async_stop(vfsp);
726 			nfs_async_manager_stop(vfsp);
727 			if (mi->mi_io_kstats) {
728 				kstat_delete(mi->mi_io_kstats);
729 				mi->mi_io_kstats = NULL;
730 			}
731 			if (mi->mi_ro_kstats) {
732 				kstat_delete(mi->mi_ro_kstats);
733 				mi->mi_ro_kstats = NULL;
734 			}
735 			nfs_free_mi(mi);
736 		}
737 	}
738 
739 	if (rtvp != NULL)
740 		VN_RELE(rtvp);
741 
742 	return (error);
743 }
744 
745 /*
746  * The pathconf information is kept on a linked list of kmem_alloc'ed
747  * structs. We search the list & add a new struct iff there is no other
748  * struct with the same information.
749  * See sys/pathconf.h for ``the rest of the story.''
750  */
751 static struct pathcnf *allpc = NULL;
752 
753 static int
754 pathconf_get(struct mntinfo *mi, model_t model, struct nfs_args *args)
755 {
756 	struct pathcnf *p;
757 	struct pathcnf pc;
758 	STRUCT_DECL(pathcnf, pc_tmp);
759 	STRUCT_HANDLE(nfs_args, ap);
760 	int i;
761 
762 #ifdef lint
763 	model = model;
764 #endif
765 
766 	STRUCT_INIT(pc_tmp, model);
767 	STRUCT_SET_HANDLE(ap, model, args);
768 
769 	if (mi->mi_pathconf != NULL) {
770 		pathconf_rele(mi);
771 		mi->mi_pathconf = NULL;
772 	}
773 	if ((STRUCT_FGET(ap, flags) & NFSMNT_POSIX) &&
774 	    STRUCT_FGETP(ap, pathconf) != NULL) {
775 		if (copyin(STRUCT_FGETP(ap, pathconf), STRUCT_BUF(pc_tmp),
776 		    STRUCT_SIZE(pc_tmp)))
777 			return (EFAULT);
778 		if (_PC_ISSET(_PC_ERROR, STRUCT_FGET(pc_tmp, pc_mask)))
779 			return (EINVAL);
780 
781 		pc.pc_link_max = STRUCT_FGET(pc_tmp, pc_link_max);
782 		pc.pc_max_canon = STRUCT_FGET(pc_tmp, pc_max_canon);
783 		pc.pc_max_input = STRUCT_FGET(pc_tmp, pc_max_input);
784 		pc.pc_name_max = STRUCT_FGET(pc_tmp, pc_name_max);
785 		pc.pc_path_max = STRUCT_FGET(pc_tmp, pc_path_max);
786 		pc.pc_pipe_buf = STRUCT_FGET(pc_tmp, pc_pipe_buf);
787 		pc.pc_vdisable = STRUCT_FGET(pc_tmp, pc_vdisable);
788 		pc.pc_xxx = STRUCT_FGET(pc_tmp, pc_xxx);
789 		for (i = 0; i < _PC_N; i++)
790 			pc.pc_mask[i] = STRUCT_FGET(pc_tmp, pc_mask[i]);
791 
792 		for (p = allpc; p != NULL; p = p->pc_next) {
793 			if (PCCMP(p, &pc) == 0)
794 				break;
795 		}
796 		if (p != NULL) {
797 			mi->mi_pathconf = p;
798 			p->pc_refcnt++;
799 		} else {
800 			p = kmem_alloc(sizeof (*p), KM_SLEEP);
801 			*p = pc;
802 			p->pc_next = allpc;
803 			p->pc_refcnt = 1;
804 			allpc = mi->mi_pathconf = p;
805 		}
806 	}
807 	return (0);
808 }
809 
810 /*
811  * release the static pathconf information
812  */
813 static void
814 pathconf_rele(struct mntinfo *mi)
815 {
816 	if (mi->mi_pathconf != NULL) {
817 		if (--mi->mi_pathconf->pc_refcnt == 0) {
818 			struct pathcnf *p;
819 			struct pathcnf *p2;
820 
821 			p2 = p = allpc;
822 			while (p != NULL && p != mi->mi_pathconf) {
823 				p2 = p;
824 				p = p->pc_next;
825 			}
826 			if (p == NULL) {
827 				panic("mi->pathconf");
828 				/*NOTREACHED*/
829 			}
830 			if (p == allpc)
831 				allpc = p->pc_next;
832 			else
833 				p2->pc_next = p->pc_next;
834 			kmem_free(p, sizeof (*p));
835 			mi->mi_pathconf = NULL;
836 		}
837 	}
838 }
839 
840 static int nfs_dynamic = 1;	/* global variable to enable dynamic retrans. */
841 static ushort_t nfs_max_threads = 8;	/* max number of active async threads */
842 static uint_t nfs_async_clusters = 1;	/* # of reqs from each async queue */
843 static uint_t nfs_cots_timeo = NFS_COTS_TIMEO;
844 
845 static int
846 nfsrootvp(vnode_t **rtvpp, vfs_t *vfsp, struct servinfo *svp,
847 	int flags, cred_t *cr, zone_t *zone)
848 {
849 	vnode_t *rtvp;
850 	mntinfo_t *mi;
851 	dev_t nfs_dev;
852 	struct vattr va;
853 	int error;
854 	rnode_t *rp;
855 	int i;
856 	struct nfs_stats *nfsstatsp;
857 	cred_t *lcr = NULL, *tcr = cr;
858 
859 	nfsstatsp = zone_getspecific(nfsstat_zone_key, nfs_zone());
860 	ASSERT(nfsstatsp != NULL);
861 
862 	/*
863 	 * Create a mount record and link it to the vfs struct.
864 	 */
865 	mi = kmem_zalloc(sizeof (*mi), KM_SLEEP);
866 	mutex_init(&mi->mi_lock, NULL, MUTEX_DEFAULT, NULL);
867 	mutex_init(&mi->mi_remap_lock, NULL, MUTEX_DEFAULT, NULL);
868 	mi->mi_flags = MI_ACL | MI_EXTATTR;
869 	if (!(flags & NFSMNT_SOFT))
870 		mi->mi_flags |= MI_HARD;
871 	if ((flags & NFSMNT_SEMISOFT))
872 		mi->mi_flags |= MI_SEMISOFT;
873 	if ((flags & NFSMNT_NOPRINT))
874 		mi->mi_flags |= MI_NOPRINT;
875 	if (flags & NFSMNT_INT)
876 		mi->mi_flags |= MI_INT;
877 	mi->mi_retrans = NFS_RETRIES;
878 	if (svp->sv_knconf->knc_semantics == NC_TPI_COTS_ORD ||
879 	    svp->sv_knconf->knc_semantics == NC_TPI_COTS)
880 		mi->mi_timeo = nfs_cots_timeo;
881 	else
882 		mi->mi_timeo = NFS_TIMEO;
883 	mi->mi_prog = NFS_PROGRAM;
884 	mi->mi_vers = NFS_VERSION;
885 	mi->mi_rfsnames = rfsnames_v2;
886 	mi->mi_reqs = nfsstatsp->nfs_stats_v2.rfsreqcnt_ptr;
887 	mi->mi_call_type = call_type_v2;
888 	mi->mi_ss_call_type = ss_call_type_v2;
889 	mi->mi_timer_type = timer_type_v2;
890 	mi->mi_aclnames = aclnames_v2;
891 	mi->mi_aclreqs = nfsstatsp->nfs_stats_v2.aclreqcnt_ptr;
892 	mi->mi_acl_call_type = acl_call_type_v2;
893 	mi->mi_acl_ss_call_type = acl_ss_call_type_v2;
894 	mi->mi_acl_timer_type = acl_timer_type_v2;
895 	cv_init(&mi->mi_failover_cv, NULL, CV_DEFAULT, NULL);
896 	mi->mi_servers = svp;
897 	mi->mi_curr_serv = svp;
898 	mi->mi_acregmin = SEC2HR(ACREGMIN);
899 	mi->mi_acregmax = SEC2HR(ACREGMAX);
900 	mi->mi_acdirmin = SEC2HR(ACDIRMIN);
901 	mi->mi_acdirmax = SEC2HR(ACDIRMAX);
902 
903 	if (nfs_dynamic)
904 		mi->mi_flags |= MI_DYNAMIC;
905 
906 	if (flags & NFSMNT_DIRECTIO)
907 		mi->mi_flags |= MI_DIRECTIO;
908 
909 	/*
910 	 * Make a vfs struct for nfs.  We do this here instead of below
911 	 * because rtvp needs a vfs before we can do a getattr on it.
912 	 *
913 	 * Assign a unique device id to the mount
914 	 */
915 	mutex_enter(&nfs_minor_lock);
916 	do {
917 		nfs_minor = (nfs_minor + 1) & MAXMIN32;
918 		nfs_dev = makedevice(nfs_major, nfs_minor);
919 	} while (vfs_devismounted(nfs_dev));
920 	mutex_exit(&nfs_minor_lock);
921 
922 	vfsp->vfs_dev = nfs_dev;
923 	vfs_make_fsid(&vfsp->vfs_fsid, nfs_dev, nfsfstyp);
924 	vfsp->vfs_data = (caddr_t)mi;
925 	vfsp->vfs_fstype = nfsfstyp;
926 	vfsp->vfs_bsize = NFS_MAXDATA;
927 
928 	/*
929 	 * Initialize fields used to support async putpage operations.
930 	 */
931 	for (i = 0; i < NFS_ASYNC_TYPES; i++)
932 		mi->mi_async_clusters[i] = nfs_async_clusters;
933 	mi->mi_async_init_clusters = nfs_async_clusters;
934 	mi->mi_async_curr = &mi->mi_async_reqs[0];
935 	mi->mi_max_threads = nfs_max_threads;
936 	mutex_init(&mi->mi_async_lock, NULL, MUTEX_DEFAULT, NULL);
937 	cv_init(&mi->mi_async_reqs_cv, NULL, CV_DEFAULT, NULL);
938 	cv_init(&mi->mi_async_work_cv, NULL, CV_DEFAULT, NULL);
939 	cv_init(&mi->mi_async_cv, NULL, CV_DEFAULT, NULL);
940 
941 	mi->mi_vfsp = vfsp;
942 	zone_hold(mi->mi_zone = zone);
943 	nfs_mi_zonelist_add(mi);
944 
945 	/*
946 	 * Make the root vnode, use it to get attributes,
947 	 * then remake it with the attributes.
948 	 */
949 	rtvp = makenfsnode((fhandle_t *)svp->sv_fhandle.fh_buf,
950 	    NULL, vfsp, gethrtime(), cr, NULL, NULL);
951 
952 	va.va_mask = AT_ALL;
953 
954 	/*
955 	 * If the uid is set then set the creds for secure mounts
956 	 * by proxy processes such as automountd.
957 	 */
958 	if (svp->sv_secdata->uid != 0 &&
959 	    svp->sv_secdata->rpcflavor == RPCSEC_GSS) {
960 		lcr = crdup(cr);
961 		(void) crsetugid(lcr, svp->sv_secdata->uid, crgetgid(cr));
962 		tcr = lcr;
963 	}
964 
965 	error = nfsgetattr(rtvp, &va, tcr);
966 	if (error)
967 		goto bad;
968 	rtvp->v_type = va.va_type;
969 
970 	/*
971 	 * Poll every server to get the filesystem stats; we're
972 	 * only interested in the server's transfer size, and we
973 	 * want the minimum.
974 	 *
975 	 * While we're looping, we'll turn off AUTH_F_TRYNONE,
976 	 * which is only for the mount operation.
977 	 */
978 
979 	mi->mi_tsize = MIN(NFS_MAXDATA, nfstsize());
980 	mi->mi_stsize = MIN(NFS_MAXDATA, nfstsize());
981 
982 	for (svp = mi->mi_servers; svp != NULL; svp = svp->sv_next) {
983 		struct nfsstatfs fs;
984 		int douprintf;
985 
986 		douprintf = 1;
987 		mi->mi_curr_serv = svp;
988 
989 		error = rfs2call(mi, RFS_STATFS,
990 			xdr_fhandle, (caddr_t)svp->sv_fhandle.fh_buf,
991 			xdr_statfs, (caddr_t)&fs, tcr, &douprintf,
992 			&fs.fs_status, 0, NULL);
993 		if (error)
994 			goto bad;
995 		mi->mi_stsize = MIN(mi->mi_stsize, fs.fs_tsize);
996 		svp->sv_secdata->flags &= ~AUTH_F_TRYNONE;
997 	}
998 	mi->mi_curr_serv = mi->mi_servers;
999 	mi->mi_curread = mi->mi_tsize;
1000 	mi->mi_curwrite = mi->mi_stsize;
1001 
1002 	/*
1003 	 * Start the manager thread responsible for handling async worker
1004 	 * threads.
1005 	 */
1006 	VFS_HOLD(vfsp);	/* add reference for thread */
1007 	mi->mi_manager_thread = zthread_create(NULL, 0, nfs_async_manager,
1008 					vfsp, 0, minclsyspri);
1009 	ASSERT(mi->mi_manager_thread != NULL);
1010 
1011 	/*
1012 	 * Initialize kstats
1013 	 */
1014 	nfs_mnt_kstat_init(vfsp);
1015 
1016 	mi->mi_type = rtvp->v_type;
1017 
1018 	*rtvpp = rtvp;
1019 	if (lcr != NULL)
1020 		crfree(lcr);
1021 
1022 	return (0);
1023 bad:
1024 	/*
1025 	 * An error occurred somewhere, need to clean up...
1026 	 * We need to release our reference to the root vnode and
1027 	 * destroy the mntinfo struct that we just created.
1028 	 */
1029 	if (lcr != NULL)
1030 		crfree(lcr);
1031 	rp = VTOR(rtvp);
1032 	if (rp->r_flags & RHASHED)
1033 		rp_rmhash(rp);
1034 	VN_RELE(rtvp);
1035 	nfs_async_stop(vfsp);
1036 	nfs_async_manager_stop(vfsp);
1037 	if (mi->mi_io_kstats) {
1038 		kstat_delete(mi->mi_io_kstats);
1039 		mi->mi_io_kstats = NULL;
1040 	}
1041 	if (mi->mi_ro_kstats) {
1042 		kstat_delete(mi->mi_ro_kstats);
1043 		mi->mi_ro_kstats = NULL;
1044 	}
1045 	nfs_free_mi(mi);
1046 	*rtvpp = NULL;
1047 	return (error);
1048 }
1049 
1050 /*
1051  * vfs operations
1052  */
1053 static int
1054 nfs_unmount(vfs_t *vfsp, int flag, cred_t *cr)
1055 {
1056 	mntinfo_t *mi;
1057 	ushort_t omax;
1058 
1059 	if (secpolicy_fs_unmount(cr, vfsp) != 0)
1060 		return (EPERM);
1061 
1062 	mi = VFTOMI(vfsp);
1063 	if (flag & MS_FORCE) {
1064 		vfsp->vfs_flag |= VFS_UNMOUNTED;
1065 		/*
1066 		 * We need to stop the manager thread explicitly; the worker
1067 		 * threads can time out and exit on their own.
1068 		 */
1069 		nfs_async_manager_stop(vfsp);
1070 		destroy_rtable(vfsp, cr);
1071 		if (mi->mi_io_kstats) {
1072 			kstat_delete(mi->mi_io_kstats);
1073 			mi->mi_io_kstats = NULL;
1074 		}
1075 		if (mi->mi_ro_kstats) {
1076 			kstat_delete(mi->mi_ro_kstats);
1077 			mi->mi_ro_kstats = NULL;
1078 		}
1079 		return (0);
1080 	}
1081 	/*
1082 	 * Wait until all asynchronous putpage operations on
1083 	 * this file system are complete before flushing rnodes
1084 	 * from the cache.
1085 	 */
1086 	omax = mi->mi_max_threads;
1087 	if (nfs_async_stop_sig(vfsp)) {
1088 		return (EINTR);
1089 	}
1090 	rflush(vfsp, cr);
1091 	/*
1092 	 * If there are any active vnodes on this file system,
1093 	 * then the file system is busy and can't be umounted.
1094 	 */
1095 	if (check_rtable(vfsp)) {
1096 		mutex_enter(&mi->mi_async_lock);
1097 		mi->mi_max_threads = omax;
1098 		mutex_exit(&mi->mi_async_lock);
1099 		return (EBUSY);
1100 	}
1101 	/*
1102 	 * The unmount can't fail from now on; stop the manager thread.
1103 	 */
1104 	nfs_async_manager_stop(vfsp);
1105 	/*
1106 	 * Destroy all rnodes belonging to this file system from the
1107 	 * rnode hash queues and purge any resources allocated to
1108 	 * them.
1109 	 */
1110 	destroy_rtable(vfsp, cr);
1111 	if (mi->mi_io_kstats) {
1112 		kstat_delete(mi->mi_io_kstats);
1113 		mi->mi_io_kstats = NULL;
1114 	}
1115 	if (mi->mi_ro_kstats) {
1116 		kstat_delete(mi->mi_ro_kstats);
1117 		mi->mi_ro_kstats = NULL;
1118 	}
1119 	return (0);
1120 }
1121 
1122 /*
1123  * find root of nfs
1124  */
1125 static int
1126 nfs_root(vfs_t *vfsp, vnode_t **vpp)
1127 {
1128 	mntinfo_t *mi;
1129 	vnode_t *vp;
1130 	servinfo_t *svp;
1131 
1132 	mi = VFTOMI(vfsp);
1133 
1134 	if (nfs_zone() != mi->mi_zone)
1135 		return (EPERM);
1136 
1137 	svp = mi->mi_curr_serv;
1138 	if (svp && (svp->sv_flags & SV_ROOT_STALE)) {
1139 		mutex_enter(&svp->sv_lock);
1140 		svp->sv_flags &= ~SV_ROOT_STALE;
1141 		mutex_exit(&svp->sv_lock);
1142 		return (ENOENT);
1143 	}
1144 
1145 	vp = makenfsnode((fhandle_t *)mi->mi_curr_serv->sv_fhandle.fh_buf,
1146 	    NULL, vfsp, gethrtime(), CRED(), NULL, NULL);
1147 
1148 	if (VTOR(vp)->r_flags & RSTALE) {
1149 		VN_RELE(vp);
1150 		return (ENOENT);
1151 	}
1152 
1153 	ASSERT(vp->v_type == VNON || vp->v_type == mi->mi_type);
1154 
1155 	vp->v_type = mi->mi_type;
1156 
1157 	*vpp = vp;
1158 
1159 	return (0);
1160 }
1161 
1162 /*
1163  * Get file system statistics.
1164  */
1165 static int
1166 nfs_statvfs(vfs_t *vfsp, struct statvfs64 *sbp)
1167 {
1168 	int error;
1169 	mntinfo_t *mi;
1170 	struct nfsstatfs fs;
1171 	int douprintf;
1172 	failinfo_t fi;
1173 	vnode_t *vp;
1174 
1175 	error = nfs_root(vfsp, &vp);
1176 	if (error)
1177 		return (error);
1178 
1179 	mi = VFTOMI(vfsp);
1180 	douprintf = 1;
1181 	fi.vp = vp;
1182 	fi.fhp = NULL;		/* no need to update, filehandle not copied */
1183 	fi.copyproc = nfscopyfh;
1184 	fi.lookupproc = nfslookup;
1185 	fi.xattrdirproc = acl_getxattrdir2;
1186 
1187 	error = rfs2call(mi, RFS_STATFS,
1188 			xdr_fhandle, (caddr_t)VTOFH(vp),
1189 			xdr_statfs, (caddr_t)&fs, CRED(), &douprintf,
1190 			&fs.fs_status, 0, &fi);
1191 
1192 	if (!error) {
1193 		error = geterrno(fs.fs_status);
1194 		if (!error) {
1195 			mutex_enter(&mi->mi_lock);
1196 			if (mi->mi_stsize) {
1197 				mi->mi_stsize = MIN(mi->mi_stsize, fs.fs_tsize);
1198 			} else {
1199 				mi->mi_stsize = fs.fs_tsize;
1200 				mi->mi_curwrite = mi->mi_stsize;
1201 			}
1202 			mutex_exit(&mi->mi_lock);
1203 			sbp->f_bsize = fs.fs_bsize;
1204 			sbp->f_frsize = fs.fs_bsize;
1205 			sbp->f_blocks = (fsblkcnt64_t)fs.fs_blocks;
1206 			sbp->f_bfree = (fsblkcnt64_t)fs.fs_bfree;
1207 			/*
1208 			 * Some servers may return negative available
1209 			 * block counts.  They may do this because they
1210 			 * calculate the number of available blocks by
1211 			 * subtracting the number of used blocks from
1212 			 * the total number of blocks modified by the
1213 			 * minimum free value.  For example, if the
1214 			 * minumum free percentage is 10 and the file
1215 			 * system is greater than 90 percent full, then
1216 			 * 90 percent of the total blocks minus the
1217 			 * actual number of used blocks may be a
1218 			 * negative number.
1219 			 *
1220 			 * In this case, we need to sign extend the
1221 			 * negative number through the assignment from
1222 			 * the 32 bit bavail count to the 64 bit bavail
1223 			 * count.
1224 			 *
1225 			 * We need to be able to discern between there
1226 			 * just being a lot of available blocks on the
1227 			 * file system and the case described above.
1228 			 * We are making the assumption that it does
1229 			 * not make sense to have more available blocks
1230 			 * than there are free blocks.  So, if there
1231 			 * are, then we treat the number as if it were
1232 			 * a negative number and arrange to have it
1233 			 * sign extended when it is converted from 32
1234 			 * bits to 64 bits.
1235 			 */
1236 			if (fs.fs_bavail <= fs.fs_bfree)
1237 				sbp->f_bavail = (fsblkcnt64_t)fs.fs_bavail;
1238 			else {
1239 				sbp->f_bavail =
1240 					(fsblkcnt64_t)((long)fs.fs_bavail);
1241 			}
1242 			sbp->f_files = (fsfilcnt64_t)-1;
1243 			sbp->f_ffree = (fsfilcnt64_t)-1;
1244 			sbp->f_favail = (fsfilcnt64_t)-1;
1245 			sbp->f_fsid = (unsigned long)vfsp->vfs_fsid.val[0];
1246 			(void) strncpy(sbp->f_basetype,
1247 				vfssw[vfsp->vfs_fstype].vsw_name, FSTYPSZ);
1248 			sbp->f_flag = vf_to_stf(vfsp->vfs_flag);
1249 			sbp->f_namemax = (uint32_t)-1;
1250 		} else {
1251 			PURGE_STALE_FH(error, vp, CRED());
1252 		}
1253 	}
1254 
1255 	VN_RELE(vp);
1256 
1257 	return (error);
1258 }
1259 
1260 static kmutex_t nfs_syncbusy;
1261 
1262 /*
1263  * Flush dirty nfs files for file system vfsp.
1264  * If vfsp == NULL, all nfs files are flushed.
1265  */
1266 /* ARGSUSED */
1267 static int
1268 nfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
1269 {
1270 	/*
1271 	 * Cross-zone calls are OK here, since this translates to a
1272 	 * VOP_PUTPAGE(B_ASYNC), which gets picked up by the right zone.
1273 	 */
1274 	if (!(flag & SYNC_ATTR) && mutex_tryenter(&nfs_syncbusy) != 0) {
1275 		rflush(vfsp, cr);
1276 		mutex_exit(&nfs_syncbusy);
1277 	}
1278 	return (0);
1279 }
1280 
1281 /* ARGSUSED */
1282 static int
1283 nfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1284 {
1285 	int error;
1286 	vnode_t *vp;
1287 	struct vattr va;
1288 	struct nfs_fid *nfsfidp = (struct nfs_fid *)fidp;
1289 	zoneid_t zoneid = VFTOMI(vfsp)->mi_zone->zone_id;
1290 
1291 	if (nfs_zone() != VFTOMI(vfsp)->mi_zone)
1292 		return (EPERM);
1293 	if (fidp->fid_len != (sizeof (*nfsfidp) - sizeof (short))) {
1294 #ifdef DEBUG
1295 		zcmn_err(zoneid, CE_WARN,
1296 		    "nfs_vget: bad fid len, %d/%d", fidp->fid_len,
1297 		    (int)(sizeof (*nfsfidp) - sizeof (short)));
1298 #endif
1299 		*vpp = NULL;
1300 		return (ESTALE);
1301 	}
1302 
1303 	vp = makenfsnode((fhandle_t *)(nfsfidp->nf_data), NULL, vfsp,
1304 	    gethrtime(), CRED(), NULL, NULL);
1305 
1306 	if (VTOR(vp)->r_flags & RSTALE) {
1307 		VN_RELE(vp);
1308 		*vpp = NULL;
1309 		return (ENOENT);
1310 	}
1311 
1312 	if (vp->v_type == VNON) {
1313 		va.va_mask = AT_ALL;
1314 		error = nfsgetattr(vp, &va, CRED());
1315 		if (error) {
1316 			VN_RELE(vp);
1317 			*vpp = NULL;
1318 			return (error);
1319 		}
1320 		vp->v_type = va.va_type;
1321 	}
1322 
1323 	*vpp = vp;
1324 
1325 	return (0);
1326 }
1327 
1328 /* ARGSUSED */
1329 static int
1330 nfs_mountroot(vfs_t *vfsp, whymountroot_t why)
1331 {
1332 	vnode_t *rtvp;
1333 	char root_hostname[SYS_NMLN+1];
1334 	struct servinfo *svp;
1335 	int error;
1336 	int vfsflags;
1337 	size_t size;
1338 	char *root_path;
1339 	struct pathname pn;
1340 	char *name;
1341 	cred_t *cr;
1342 	struct nfs_args args;		/* nfs mount arguments */
1343 	static char token[10];
1344 
1345 	bzero(&args, sizeof (args));
1346 
1347 	/* do this BEFORE getfile which causes xid stamps to be initialized */
1348 	clkset(-1L);		/* hack for now - until we get time svc? */
1349 
1350 	if (why == ROOT_REMOUNT) {
1351 		/*
1352 		 * Shouldn't happen.
1353 		 */
1354 		panic("nfs_mountroot: why == ROOT_REMOUNT");
1355 	}
1356 
1357 	if (why == ROOT_UNMOUNT) {
1358 		/*
1359 		 * Nothing to do for NFS.
1360 		 */
1361 		return (0);
1362 	}
1363 
1364 	/*
1365 	 * why == ROOT_INIT
1366 	 */
1367 
1368 	name = token;
1369 	*name = 0;
1370 	getfsname("root", name, sizeof (token));
1371 
1372 	pn_alloc(&pn);
1373 	root_path = pn.pn_path;
1374 
1375 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
1376 	svp->sv_knconf = kmem_zalloc(sizeof (*svp->sv_knconf), KM_SLEEP);
1377 	svp->sv_knconf->knc_protofmly = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
1378 	svp->sv_knconf->knc_proto = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
1379 
1380 	/*
1381 	 * Get server address
1382 	 * Get the root fhandle
1383 	 * Get server's transport
1384 	 * Get server's hostname
1385 	 * Get options
1386 	 */
1387 	args.addr = &svp->sv_addr;
1388 	args.fh = (char *)&svp->sv_fhandle.fh_buf;
1389 	args.knconf = svp->sv_knconf;
1390 	args.hostname = root_hostname;
1391 	vfsflags = 0;
1392 	if (error = mount_root(*name ? name : "root", root_path, NFS_VERSION,
1393 	    &args, &vfsflags)) {
1394 		nfs_cmn_err(error, CE_WARN,
1395 		    "nfs_mountroot: mount_root failed: %m");
1396 		sv_free(svp);
1397 		pn_free(&pn);
1398 		return (error);
1399 	}
1400 	svp->sv_fhandle.fh_len = NFS_FHSIZE;
1401 	svp->sv_hostnamelen = (int)(strlen(root_hostname) + 1);
1402 	svp->sv_hostname = kmem_alloc(svp->sv_hostnamelen, KM_SLEEP);
1403 	(void) strcpy(svp->sv_hostname, root_hostname);
1404 
1405 	/*
1406 	 * Force root partition to always be mounted with AUTH_UNIX for now
1407 	 */
1408 	svp->sv_secdata = kmem_alloc(sizeof (*svp->sv_secdata), KM_SLEEP);
1409 	svp->sv_secdata->secmod = AUTH_UNIX;
1410 	svp->sv_secdata->rpcflavor = AUTH_UNIX;
1411 	svp->sv_secdata->data = NULL;
1412 
1413 	cr = crgetcred();
1414 	rtvp = NULL;
1415 
1416 	error = nfsrootvp(&rtvp, vfsp, svp, args.flags, cr, global_zone);
1417 
1418 	crfree(cr);
1419 
1420 	if (error) {
1421 		pn_free(&pn);
1422 		goto errout;
1423 	}
1424 
1425 	error = nfs_setopts(rtvp, DATAMODEL_NATIVE, &args);
1426 	if (error) {
1427 		nfs_cmn_err(error, CE_WARN,
1428 		    "nfs_mountroot: invalid root mount options");
1429 		pn_free(&pn);
1430 		goto errout;
1431 	}
1432 
1433 	(void) vfs_lock_wait(vfsp);
1434 	vfs_add(NULL, vfsp, vfsflags);
1435 	vfs_unlock(vfsp);
1436 
1437 	size = strlen(svp->sv_hostname);
1438 	(void) strcpy(rootfs.bo_name, svp->sv_hostname);
1439 	rootfs.bo_name[size] = ':';
1440 	(void) strcpy(&rootfs.bo_name[size + 1], root_path);
1441 
1442 	pn_free(&pn);
1443 
1444 errout:
1445 	if (error) {
1446 		sv_free(svp);
1447 		nfs_async_stop(vfsp);
1448 		nfs_async_manager_stop(vfsp);
1449 	}
1450 
1451 	if (rtvp != NULL)
1452 		VN_RELE(rtvp);
1453 
1454 	return (error);
1455 }
1456 
1457 /*
1458  * Initialization routine for VFS routines.  Should only be called once
1459  */
1460 int
1461 nfs_vfsinit(void)
1462 {
1463 	mutex_init(&nfs_syncbusy, NULL, MUTEX_DEFAULT, NULL);
1464 	return (0);
1465 }
1466 
1467 void
1468 nfs_vfsfini(void)
1469 {
1470 	mutex_destroy(&nfs_syncbusy);
1471 }
1472 
1473 void
1474 nfs_freevfs(vfs_t *vfsp)
1475 {
1476 	mntinfo_t *mi;
1477 	servinfo_t *svp;
1478 
1479 	/* free up the resources */
1480 	mi = VFTOMI(vfsp);
1481 	pathconf_rele(mi);
1482 	svp = mi->mi_servers;
1483 	mi->mi_servers = mi->mi_curr_serv = NULL;
1484 	sv_free(svp);
1485 
1486 	/*
1487 	 * By this time we should have already deleted the
1488 	 * mi kstats in the unmount code. If they are still around
1489 	 * somethings wrong
1490 	 */
1491 	ASSERT(mi->mi_io_kstats == NULL);
1492 	nfs_free_mi(mi);
1493 }
1494