xref: /titanic_41/usr/src/uts/common/fs/nfs/nfs3_vfsops.c (revision 1e49577a7fcde812700ded04431b49d67cc57d6d)
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 2010 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 #include <sys/param.h>
32 #include <sys/types.h>
33 #include <sys/systm.h>
34 #include <sys/cred.h>
35 #include <sys/vfs.h>
36 #include <sys/vfs_opreg.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 #include <sys/tsol/tnet.h>
59 
60 #include <rpc/types.h>
61 #include <rpc/auth.h>
62 #include <rpc/clnt.h>
63 
64 #include <nfs/nfs.h>
65 #include <nfs/nfs_clnt.h>
66 #include <nfs/rnode.h>
67 #include <nfs/mount.h>
68 #include <nfs/nfs_acl.h>
69 
70 #include <fs/fs_subr.h>
71 
72 /*
73  * From rpcsec module (common/rpcsec).
74  */
75 extern int sec_clnt_loadinfo(struct sec_data *, struct sec_data **, model_t);
76 extern void sec_clnt_freeinfo(struct sec_data *);
77 
78 /*
79  * The order and contents of this structure must be kept in sync with that of
80  * rfsreqcnt_v3_tmpl in nfs_stats.c
81  */
82 static char *rfsnames_v3[] = {
83 	"null", "getattr", "setattr", "lookup", "access", "readlink", "read",
84 	"write", "create", "mkdir", "symlink", "mknod", "remove", "rmdir",
85 	"rename", "link", "readdir", "readdirplus", "fsstat", "fsinfo",
86 	"pathconf", "commit"
87 };
88 
89 /*
90  * This table maps from NFS protocol number into call type.
91  * Zero means a "Lookup" type call
92  * One  means a "Read" type call
93  * Two  means a "Write" type call
94  * This is used to select a default time-out.
95  */
96 static uchar_t call_type_v3[] = {
97 	0, 0, 1, 0, 0, 0, 1,
98 	2, 2, 2, 2, 2, 2, 2,
99 	2, 2, 1, 2, 0, 0, 0,
100 	2 };
101 
102 /*
103  * Similar table, but to determine which timer to use
104  * (only real reads and writes!)
105  */
106 static uchar_t timer_type_v3[] = {
107 	0, 0, 0, 0, 0, 0, 1,
108 	2, 0, 0, 0, 0, 0, 0,
109 	0, 0, 1, 1, 0, 0, 0,
110 	0 };
111 
112 /*
113  * This table maps from NFS protocol number into a call type
114  * for the semisoft mount option.
115  * Zero means do not repeat operation.
116  * One  means repeat.
117  */
118 static uchar_t ss_call_type_v3[] = {
119 	0, 0, 1, 0, 0, 0, 0,
120 	1, 1, 1, 1, 1, 1, 1,
121 	1, 1, 0, 0, 0, 0, 0,
122 	1 };
123 
124 /*
125  * nfs3 vfs operations.
126  */
127 static int	nfs3_mount(vfs_t *, vnode_t *, struct mounta *, cred_t *);
128 static int	nfs3_unmount(vfs_t *, int, cred_t *);
129 static int	nfs3_root(vfs_t *, vnode_t **);
130 static int	nfs3_statvfs(vfs_t *, struct statvfs64 *);
131 static int	nfs3_sync(vfs_t *, short, cred_t *);
132 static int	nfs3_vget(vfs_t *, vnode_t **, fid_t *);
133 static int	nfs3_mountroot(vfs_t *, whymountroot_t);
134 static void	nfs3_freevfs(vfs_t *);
135 
136 static int	nfs3rootvp(vnode_t **, vfs_t *, struct servinfo *,
137 		    int, cred_t *, zone_t *);
138 
139 /*
140  * Initialize the vfs structure
141  */
142 
143 static int nfs3fstyp;
144 vfsops_t *nfs3_vfsops;
145 
146 /*
147  * Debug variable to check for rdma based
148  * transport startup and cleanup. Controlled
149  * through /etc/system. Off by default.
150  */
151 extern int rdma_debug;
152 
153 int
154 nfs3init(int fstyp, char *name)
155 {
156 	static const fs_operation_def_t nfs3_vfsops_template[] = {
157 		VFSNAME_MOUNT,		{ .vfs_mount = nfs3_mount },
158 		VFSNAME_UNMOUNT,	{ .vfs_unmount = nfs3_unmount },
159 		VFSNAME_ROOT,		{ .vfs_root = nfs3_root },
160 		VFSNAME_STATVFS,	{ .vfs_statvfs = nfs3_statvfs },
161 		VFSNAME_SYNC,		{ .vfs_sync = nfs3_sync },
162 		VFSNAME_VGET,		{ .vfs_vget = nfs3_vget },
163 		VFSNAME_MOUNTROOT,	{ .vfs_mountroot = nfs3_mountroot },
164 		VFSNAME_FREEVFS,	{ .vfs_freevfs = nfs3_freevfs },
165 		NULL,			NULL
166 	};
167 	int error;
168 
169 	error = vfs_setfsops(fstyp, nfs3_vfsops_template, &nfs3_vfsops);
170 	if (error != 0) {
171 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
172 		    "nfs3init: bad vfs ops template");
173 		return (error);
174 	}
175 
176 	error = vn_make_ops(name, nfs3_vnodeops_template, &nfs3_vnodeops);
177 	if (error != 0) {
178 		(void) vfs_freevfsops_by_type(fstyp);
179 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
180 		    "nfs3init: bad vnode ops template");
181 		return (error);
182 	}
183 
184 	nfs3fstyp = fstyp;
185 
186 	return (0);
187 }
188 
189 void
190 nfs3fini(void)
191 {
192 }
193 
194 static void
195 nfs3_free_args(struct nfs_args *nargs, nfs_fhandle *fh)
196 {
197 
198 	if (fh)
199 		kmem_free(fh, sizeof (*fh));
200 
201 	if (nargs->knconf) {
202 		if (nargs->knconf->knc_protofmly)
203 			kmem_free(nargs->knconf->knc_protofmly, KNC_STRSIZE);
204 		if (nargs->knconf->knc_proto)
205 			kmem_free(nargs->knconf->knc_proto, KNC_STRSIZE);
206 		kmem_free(nargs->knconf, sizeof (*nargs->knconf));
207 		nargs->knconf = NULL;
208 	}
209 
210 	if (nargs->fh) {
211 		kmem_free(nargs->fh, strlen(nargs->fh) + 1);
212 		nargs->fh = NULL;
213 	}
214 
215 	if (nargs->hostname) {
216 		kmem_free(nargs->hostname, strlen(nargs->hostname) + 1);
217 		nargs->hostname = NULL;
218 	}
219 
220 	if (nargs->addr) {
221 		if (nargs->addr->buf) {
222 			ASSERT(nargs->addr->len);
223 			kmem_free(nargs->addr->buf, nargs->addr->len);
224 		}
225 		kmem_free(nargs->addr, sizeof (struct netbuf));
226 		nargs->addr = NULL;
227 	}
228 
229 	if (nargs->syncaddr) {
230 		ASSERT(nargs->syncaddr->len);
231 		if (nargs->syncaddr->buf) {
232 			ASSERT(nargs->syncaddr->len);
233 			kmem_free(nargs->syncaddr->buf, nargs->syncaddr->len);
234 		}
235 		kmem_free(nargs->syncaddr, sizeof (struct netbuf));
236 		nargs->syncaddr = NULL;
237 	}
238 
239 	if (nargs->netname) {
240 		kmem_free(nargs->netname, strlen(nargs->netname) + 1);
241 		nargs->netname = NULL;
242 	}
243 
244 	if (nargs->nfs_ext_u.nfs_extA.secdata) {
245 		sec_clnt_freeinfo(nargs->nfs_ext_u.nfs_extA.secdata);
246 		nargs->nfs_ext_u.nfs_extA.secdata = NULL;
247 	}
248 }
249 
250 static int
251 nfs3_copyin(char *data, int datalen, struct nfs_args *nargs, nfs_fhandle *fh)
252 {
253 
254 	int error;
255 	size_t nlen;			/* length of netname */
256 	size_t hlen;			/* length of hostname */
257 	char netname[MAXNETNAMELEN+1];  /* server's netname */
258 	struct netbuf addr;		/* server's address */
259 	struct netbuf syncaddr;		/* AUTH_DES time sync addr */
260 	struct knetconfig *knconf;	/* transport knetconfig structure */
261 	struct sec_data *secdata = NULL;	/* security data */
262 	STRUCT_DECL(nfs_args, args);    	/* nfs mount arguments */
263 	STRUCT_DECL(knetconfig, knconf_tmp);
264 	STRUCT_DECL(netbuf, addr_tmp);
265 	int flags;
266 	char *p, *pf;
267 	char *userbufptr;
268 
269 
270 	bzero(nargs, sizeof (*nargs));
271 
272 	STRUCT_INIT(args, get_udatamodel());
273 	bzero(STRUCT_BUF(args), SIZEOF_STRUCT(nfs_args, DATAMODEL_NATIVE));
274 	if (copyin(data, STRUCT_BUF(args), MIN(datalen, STRUCT_SIZE(args))))
275 		return (EFAULT);
276 
277 	nargs->wsize = STRUCT_FGET(args, wsize);
278 	nargs->rsize = STRUCT_FGET(args, rsize);
279 	nargs->timeo = STRUCT_FGET(args, timeo);
280 	nargs->retrans = STRUCT_FGET(args, retrans);
281 	nargs->acregmin = STRUCT_FGET(args, acregmin);
282 	nargs->acregmax = STRUCT_FGET(args, acregmax);
283 	nargs->acdirmin = STRUCT_FGET(args, acdirmin);
284 	nargs->acdirmax = STRUCT_FGET(args, acdirmax);
285 
286 	flags = STRUCT_FGET(args, flags);
287 	nargs->flags = flags;
288 
289 	addr.buf = NULL;
290 	syncaddr.buf = NULL;
291 
292 	/*
293 	 * Allocate space for a knetconfig structure and
294 	 * its strings and copy in from user-land.
295 	 */
296 	knconf = kmem_zalloc(sizeof (*knconf), KM_SLEEP);
297 	STRUCT_INIT(knconf_tmp, get_udatamodel());
298 	if (copyin(STRUCT_FGETP(args, knconf), STRUCT_BUF(knconf_tmp),
299 	    STRUCT_SIZE(knconf_tmp))) {
300 		kmem_free(knconf, sizeof (*knconf));
301 		return (EFAULT);
302 	}
303 
304 	knconf->knc_semantics = STRUCT_FGET(knconf_tmp, knc_semantics);
305 	knconf->knc_protofmly = STRUCT_FGETP(knconf_tmp, knc_protofmly);
306 	knconf->knc_proto = STRUCT_FGETP(knconf_tmp, knc_proto);
307 	if (get_udatamodel() != DATAMODEL_LP64) {
308 		knconf->knc_rdev = expldev(STRUCT_FGET(knconf_tmp, knc_rdev));
309 	} else {
310 		knconf->knc_rdev = STRUCT_FGET(knconf_tmp, knc_rdev);
311 	}
312 
313 	pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
314 	p = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
315 	error = copyinstr(knconf->knc_protofmly, pf, KNC_STRSIZE, NULL);
316 	if (error) {
317 		kmem_free(pf, KNC_STRSIZE);
318 		kmem_free(p, KNC_STRSIZE);
319 		kmem_free(knconf, sizeof (*knconf));
320 		return (error);
321 	}
322 
323 	error = copyinstr(knconf->knc_proto, p, KNC_STRSIZE, NULL);
324 	if (error) {
325 		kmem_free(pf, KNC_STRSIZE);
326 		kmem_free(p, KNC_STRSIZE);
327 		kmem_free(knconf, sizeof (*knconf));
328 		return (error);
329 	}
330 
331 
332 	knconf->knc_protofmly = pf;
333 	knconf->knc_proto = p;
334 
335 	nargs->knconf = knconf;
336 	/*
337 	 * Get server address
338 	 */
339 	STRUCT_INIT(addr_tmp, get_udatamodel());
340 	if (copyin(STRUCT_FGETP(args, addr), STRUCT_BUF(addr_tmp),
341 	    STRUCT_SIZE(addr_tmp))) {
342 		error = EFAULT;
343 		goto errout;
344 	}
345 
346 	nargs->addr = kmem_alloc(sizeof (struct netbuf), KM_SLEEP);
347 	userbufptr = STRUCT_FGETP(addr_tmp, buf);
348 	addr.len = STRUCT_FGET(addr_tmp, len);
349 	addr.buf = kmem_alloc(addr.len, KM_SLEEP);
350 	addr.maxlen = addr.len;
351 	if (copyin(userbufptr, addr.buf, addr.len)) {
352 		kmem_free(addr.buf, addr.len);
353 		error = EFAULT;
354 		goto errout;
355 	}
356 	bcopy(&addr, nargs->addr, sizeof (struct netbuf));
357 
358 	/*
359 	 * Get the root fhandle
360 	 */
361 
362 	if (copyin(STRUCT_FGETP(args, fh), fh, sizeof (nfs_fhandle))) {
363 		error = EFAULT;
364 		goto errout;
365 	}
366 
367 
368 	/*
369 	 * Get server's hostname
370 	 */
371 	if (flags & NFSMNT_HOSTNAME) {
372 		error = copyinstr(STRUCT_FGETP(args, hostname), netname,
373 		    sizeof (netname), &hlen);
374 	if (error)
375 		goto errout;
376 	nargs->hostname = kmem_zalloc(hlen, KM_SLEEP);
377 	(void) strcpy(nargs->hostname, netname);
378 	} else {
379 	nargs->hostname = NULL;
380 	}
381 
382 
383 	/*
384 	 * If there are syncaddr and netname data, load them in. This is
385 	 * to support data needed for NFSV4 when AUTH_DH is the negotiated
386 	 * flavor via SECINFO. (instead of using MOUNT protocol in V3).
387 	 */
388 	netname[0] = '\0';
389 	if (flags & NFSMNT_SECURE) {
390 		if (STRUCT_FGETP(args, syncaddr) == NULL) {
391 			error = EINVAL;
392 			goto errout;
393 		}
394 		/* get syncaddr */
395 		STRUCT_INIT(addr_tmp, get_udatamodel());
396 		if (copyin(STRUCT_FGETP(args, syncaddr), STRUCT_BUF(addr_tmp),
397 		    STRUCT_SIZE(addr_tmp))) {
398 			error = EINVAL;
399 			goto errout;
400 		}
401 		userbufptr = STRUCT_FGETP(addr_tmp, buf);
402 		syncaddr.len = STRUCT_FGET(addr_tmp, len);
403 		syncaddr.buf = kmem_alloc(syncaddr.len, KM_SLEEP);
404 		syncaddr.maxlen = syncaddr.len;
405 		if (copyin(userbufptr, syncaddr.buf, syncaddr.len)) {
406 			kmem_free(syncaddr.buf, syncaddr.len);
407 			error = EFAULT;
408 			goto errout;
409 		}
410 
411 		nargs->syncaddr = kmem_alloc(sizeof (struct netbuf), KM_SLEEP);
412 		bcopy(&syncaddr, nargs->syncaddr, sizeof (struct netbuf));
413 
414 		ASSERT(STRUCT_FGETP(args, netname));
415 
416 		if (copyinstr(STRUCT_FGETP(args, netname), netname,
417 		    sizeof (netname), &nlen)) {
418 			error = EFAULT;
419 			goto errout;
420 		}
421 
422 		netname[nlen] = '\0';
423 		nargs->netname = kmem_zalloc(nlen, KM_SLEEP);
424 		(void) strcpy(nargs->netname, netname);
425 	}
426 
427 	/*
428 	 * Get the extention data which has the security data structure.
429 	 * This includes data for AUTH_SYS as well.
430 	 */
431 	if (flags & NFSMNT_NEWARGS) {
432 		nargs->nfs_args_ext = STRUCT_FGET(args, nfs_args_ext);
433 		if (nargs->nfs_args_ext == NFS_ARGS_EXTA ||
434 		    nargs->nfs_args_ext == NFS_ARGS_EXTB) {
435 			/*
436 			 * Indicating the application is using the new
437 			 * sec_data structure to pass in the security
438 			 * data.
439 			 */
440 			if (STRUCT_FGETP(args,
441 			    nfs_ext_u.nfs_extA.secdata) != NULL) {
442 				error = sec_clnt_loadinfo(
443 				    (struct sec_data *)STRUCT_FGETP(args,
444 				    nfs_ext_u.nfs_extA.secdata), &secdata,
445 				    get_udatamodel());
446 			}
447 			nargs->nfs_ext_u.nfs_extA.secdata = secdata;
448 		}
449 	}
450 
451 	if (error)
452 		goto errout;
453 
454 	/*
455 	 * Failover support:
456 	 *
457 	 * We may have a linked list of nfs_args structures,
458 	 * which means the user is looking for failover.  If
459 	 * the mount is either not "read-only" or "soft",
460 	 * we want to bail out with EINVAL.
461 	 */
462 	if (nargs->nfs_args_ext == NFS_ARGS_EXTB)
463 		nargs->nfs_ext_u.nfs_extB.next =
464 		    STRUCT_FGETP(args, nfs_ext_u.nfs_extB.next);
465 
466 errout:
467 	if (error)
468 		nfs3_free_args(nargs, fh);
469 
470 	return (error);
471 }
472 
473 
474 /*
475  * nfs mount vfsop
476  * Set up mount info record and attach it to vfs struct.
477  */
478 static int
479 nfs3_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
480 {
481 	struct nfs_args	*args = NULL;
482 	nfs_fhandle	*fhandle = NULL;
483 	char *data = uap->dataptr;
484 	int error;
485 	vnode_t *rtvp;			/* the server's root */
486 	mntinfo_t *mi;			/* mount info, pointed at by vfs */
487 	size_t nlen;			/* length of netname */
488 	struct knetconfig *knconf;	/* transport knetconfig structure */
489 	struct knetconfig *rdma_knconf;	/* rdma transport structure */
490 	rnode_t *rp;
491 	struct servinfo *svp;		/* nfs server info */
492 	struct servinfo *svp_tail = NULL; /* previous nfs server info */
493 	struct servinfo *svp_head;	/* first nfs server info */
494 	struct servinfo *svp_2ndlast;	/* 2nd last in server info list */
495 	struct sec_data *secdata;	/* security data */
496 	int flags, addr_type;
497 	zone_t *zone = nfs_zone();
498 	zone_t *mntzone = NULL;
499 
500 
501 	if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
502 		return (EPERM);
503 
504 	if (mvp->v_type != VDIR)
505 		return (ENOTDIR);
506 
507 	/*
508 	 * get arguments
509 	 *
510 	 * nfs_args is now versioned and is extensible, so
511 	 * uap->datalen might be different from sizeof (args)
512 	 * in a compatible situation.
513 	 */
514 
515 more:
516 
517 	if (!(uap->flags & MS_SYSSPACE)) {
518 		if (args == NULL)
519 			args = kmem_alloc(sizeof (struct nfs_args), KM_SLEEP);
520 		else {
521 			nfs3_free_args(args, fhandle);
522 			fhandle = NULL;
523 		}
524 		if (fhandle == NULL)
525 			fhandle = kmem_alloc(sizeof (nfs_fhandle), KM_SLEEP);
526 		error = nfs3_copyin(data, uap->datalen, args, fhandle);
527 		if (error) {
528 			if (args)
529 				kmem_free(args, sizeof (*args));
530 			return (error);
531 		}
532 	} else {
533 		args = (struct nfs_args *)data;
534 		fhandle = (nfs_fhandle *)args->fh;
535 	}
536 
537 
538 	flags = args->flags;
539 
540 	if (uap->flags & MS_REMOUNT) {
541 		size_t	n;
542 		char	name[FSTYPSZ];
543 
544 		if (uap->flags & MS_SYSSPACE) {
545 			error = copystr(uap->fstype, name, FSTYPSZ, &n);
546 		} else {
547 			nfs3_free_args(args, fhandle);
548 			kmem_free(args, sizeof (*args));
549 			error = copyinstr(uap->fstype, name, FSTYPSZ, &n);
550 		}
551 		if (error) {
552 			if (error == ENAMETOOLONG)
553 				return (EINVAL);
554 			return (error);
555 		}
556 
557 		/*
558 		 * This check is to ensure that the request is a
559 		 * genuine nfs remount request.
560 		 */
561 
562 		if (strncmp(name, "nfs", 3) != 0)
563 			return (EINVAL);
564 
565 		/*
566 		 * If the request changes the locking type, disallow the
567 		 * remount,
568 		 * because it's questionable whether we can transfer the
569 		 * locking state correctly.
570 		 */
571 
572 		if ((mi = VFTOMI(vfsp)) != NULL) {
573 			uint_t new_mi_llock;
574 			uint_t old_mi_llock;
575 
576 			new_mi_llock = (flags & NFSMNT_LLOCK) ? 1 : 0;
577 			old_mi_llock = (mi->mi_flags & MI_LLOCK) ? 1 : 0;
578 			if (old_mi_llock != new_mi_llock)
579 				return (EBUSY);
580 		}
581 		return (0);
582 	}
583 
584 	mutex_enter(&mvp->v_lock);
585 	if (!(uap->flags & MS_OVERLAY) &&
586 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
587 		mutex_exit(&mvp->v_lock);
588 		if (!(uap->flags & MS_SYSSPACE)) {
589 			nfs3_free_args(args, fhandle);
590 			kmem_free(args, sizeof (*args));
591 		}
592 		return (EBUSY);
593 	}
594 	mutex_exit(&mvp->v_lock);
595 
596 	/* make sure things are zeroed for errout: */
597 	rtvp = NULL;
598 	mi = NULL;
599 	secdata = NULL;
600 
601 	/*
602 	 * A valid knetconfig structure is required.
603 	 */
604 	if (!(flags & NFSMNT_KNCONF)) {
605 		if (!(uap->flags & MS_SYSSPACE)) {
606 			nfs3_free_args(args, fhandle);
607 			kmem_free(args, sizeof (*args));
608 		}
609 		return (EINVAL);
610 	}
611 
612 	if ((strlen(args->knconf->knc_protofmly) >= KNC_STRSIZE) ||
613 	    (strlen(args->knconf->knc_proto) >= KNC_STRSIZE)) {
614 		if (!(uap->flags & MS_SYSSPACE)) {
615 			nfs3_free_args(args, fhandle);
616 			kmem_free(args, sizeof (*args));
617 		}
618 		return (EINVAL);
619 	}
620 
621 	/*
622 	 * Allocate a servinfo struct.
623 	 */
624 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
625 	mutex_init(&svp->sv_lock, NULL, MUTEX_DEFAULT, NULL);
626 	if (svp_tail) {
627 		svp_2ndlast = svp_tail;
628 		svp_tail->sv_next = svp;
629 	} else {
630 		svp_head = svp;
631 		svp_2ndlast = svp;
632 	}
633 
634 	svp_tail = svp;
635 
636 	svp->sv_knconf = args->knconf;
637 	args->knconf = NULL;
638 
639 	if (args->addr == NULL || args->addr->buf == NULL) {
640 		error = EINVAL;
641 		goto errout;
642 	}
643 
644 	svp->sv_addr.maxlen = args->addr->maxlen;
645 	svp->sv_addr.len = args->addr->len;
646 	svp->sv_addr.buf = args->addr->buf;
647 	args->addr->buf = NULL;
648 
649 	/*
650 	 * Check the root fhandle length
651 	 */
652 	ASSERT(fhandle);
653 	if (fhandle->fh_len > NFS3_FHSIZE || fhandle->fh_len == 0) {
654 		error = EINVAL;
655 #ifdef DEBUG
656 		zcmn_err(getzoneid(), CE_WARN,
657 		    "nfs3_mount: got an invalid fhandle. fh_len = %d",
658 		    fhandle->fh_len);
659 		fhandle->fh_len = NFS_FHANDLE_LEN;
660 		nfs_printfhandle(fhandle);
661 #endif
662 		goto errout;
663 	}
664 
665 	bcopy(&fhandle->fh_buf, &svp->sv_fhandle.fh_buf, fhandle->fh_len);
666 	svp->sv_fhandle.fh_len = fhandle->fh_len;
667 
668 	/*
669 	 * Get server's hostname
670 	 */
671 	if (flags & NFSMNT_HOSTNAME) {
672 		if (args->hostname == NULL) {
673 			error = EINVAL;
674 			goto errout;
675 		}
676 		svp->sv_hostnamelen = strlen(args->hostname) + 1;
677 		svp->sv_hostname = args->hostname;
678 		args->hostname = NULL;
679 	} else {
680 		char *p = "unknown-host";
681 		svp->sv_hostnamelen = strlen(p) + 1;
682 		svp->sv_hostname = kmem_zalloc(svp->sv_hostnamelen, KM_SLEEP);
683 		(void) strcpy(svp->sv_hostname, p);
684 	}
685 
686 
687 	/*
688 	 * RDMA MOUNT SUPPORT FOR NFS v3:
689 	 * Establish, is it possible to use RDMA, if so overload the
690 	 * knconf with rdma specific knconf and free the orignal.
691 	 */
692 	if ((flags & NFSMNT_TRYRDMA) || (flags & NFSMNT_DORDMA)) {
693 		/*
694 		 * Determine the addr type for RDMA, IPv4 or v6.
695 		 */
696 		if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET) == 0)
697 			addr_type = AF_INET;
698 		else if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET6) == 0)
699 			addr_type = AF_INET6;
700 
701 		if (rdma_reachable(addr_type, &svp->sv_addr,
702 		    &rdma_knconf) == 0) {
703 			/*
704 			 * If successful, hijack the orignal knconf and
705 			 * replace with a new one, depending on the flags.
706 			 */
707 			svp->sv_origknconf = svp->sv_knconf;
708 			svp->sv_knconf = rdma_knconf;
709 			knconf = rdma_knconf;
710 		} else {
711 			if (flags & NFSMNT_TRYRDMA) {
712 #ifdef	DEBUG
713 				if (rdma_debug)
714 					zcmn_err(getzoneid(), CE_WARN,
715 					    "no RDMA onboard, revert\n");
716 #endif
717 			}
718 
719 			if (flags & NFSMNT_DORDMA) {
720 				/*
721 				 * If proto=rdma is specified and no RDMA
722 				 * path to this server is avialable then
723 				 * ditch this server.
724 				 * This is not included in the mountable
725 				 * server list or the replica list.
726 				 * Check if more servers are specified;
727 				 * Failover case, otherwise bail out of mount.
728 				 */
729 				if (args->nfs_args_ext == NFS_ARGS_EXTB &&
730 				    args->nfs_ext_u.nfs_extB.next != NULL) {
731 					data = (char *)
732 					    args->nfs_ext_u.nfs_extB.next;
733 					if (uap->flags & MS_RDONLY &&
734 					    !(flags & NFSMNT_SOFT)) {
735 						if (svp_head->sv_next == NULL) {
736 							svp_tail = NULL;
737 							svp_2ndlast = NULL;
738 							sv_free(svp_head);
739 							goto more;
740 						} else {
741 							svp_tail = svp_2ndlast;
742 							svp_2ndlast->sv_next =
743 							    NULL;
744 							sv_free(svp);
745 							goto more;
746 						}
747 					}
748 				} else {
749 					/*
750 					 * This is the last server specified
751 					 * in the nfs_args list passed down
752 					 * and its not rdma capable.
753 					 */
754 					if (svp_head->sv_next == NULL) {
755 						/*
756 						 * Is this the only one
757 						 */
758 						error = EINVAL;
759 #ifdef	DEBUG
760 						if (rdma_debug)
761 							zcmn_err(getzoneid(),
762 							    CE_WARN,
763 							    "No RDMA srv");
764 #endif
765 						goto errout;
766 					} else {
767 						/*
768 						 * There is list, since some
769 						 * servers specified before
770 						 * this passed all requirements
771 						 */
772 						svp_tail = svp_2ndlast;
773 						svp_2ndlast->sv_next = NULL;
774 						sv_free(svp);
775 						goto proceed;
776 					}
777 				}
778 			}
779 		}
780 	}
781 
782 	/*
783 	 * Get the extention data which has the new security data structure.
784 	 */
785 	if (flags & NFSMNT_NEWARGS) {
786 		switch (args->nfs_args_ext) {
787 		case NFS_ARGS_EXTA:
788 		case NFS_ARGS_EXTB:
789 			/*
790 			 * Indicating the application is using the new
791 			 * sec_data structure to pass in the security
792 			 * data.
793 			 */
794 			secdata = args->nfs_ext_u.nfs_extA.secdata;
795 			if (args->nfs_ext_u.nfs_extA.secdata == NULL) {
796 				error = EINVAL;
797 			} else {
798 				/*
799 				 * Need to validate the flavor here if
800 				 * sysspace, userspace was already
801 				 * validate from the nfs_copyin function.
802 				 */
803 				switch (secdata->rpcflavor) {
804 				case AUTH_NONE:
805 				case AUTH_UNIX:
806 				case AUTH_LOOPBACK:
807 				case AUTH_DES:
808 				case RPCSEC_GSS:
809 					args->nfs_ext_u.nfs_extA.secdata = NULL;
810 					break;
811 				default:
812 					error = EINVAL;
813 					goto errout;
814 				}
815 			}
816 			break;
817 
818 		default:
819 			error = EINVAL;
820 			break;
821 		}
822 	} else if (flags & NFSMNT_SECURE) {
823 		/*
824 		 * Keep this for backward compatibility to support
825 		 * NFSMNT_SECURE/NFSMNT_RPCTIMESYNC flags.
826 		 */
827 		if (args->syncaddr == NULL || args->syncaddr->buf == NULL) {
828 			error = EINVAL;
829 			goto errout;
830 		}
831 		/*
832 		 * Move security related data to the sec_data structure.
833 		 */
834 		{
835 			dh_k4_clntdata_t *data;
836 			char *pf, *p;
837 			secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
838 			if (flags & NFSMNT_RPCTIMESYNC)
839 				secdata->flags |= AUTH_F_RPCTIMESYNC;
840 			data = kmem_alloc(sizeof (*data), KM_SLEEP);
841 			bcopy(args->syncaddr, &data->syncaddr,
842 			    sizeof (*args->syncaddr));
843 
844 			/*
845 			 * duplicate the knconf information for the
846 			 * new opaque data.
847 			 */
848 			data->knconf = kmem_alloc(sizeof (*knconf), KM_SLEEP);
849 			*data->knconf = *knconf;
850 			pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
851 			p = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
852 			bcopy(knconf->knc_protofmly, pf, KNC_STRSIZE);
853 			bcopy(knconf->knc_proto, pf, KNC_STRSIZE);
854 			data->knconf->knc_protofmly = pf;
855 			data->knconf->knc_proto = p;
856 
857 			nlen = strlen(args->hostname) + 1;
858 			/* move server netname to the sec_data structure */
859 			if (nlen != 0) {
860 				data->netname = kmem_alloc(nlen, KM_SLEEP);
861 				bcopy(args->hostname, data->netname, nlen);
862 				data->netnamelen = nlen;
863 			}
864 			secdata->secmod = secdata->rpcflavor = AUTH_DES;
865 			secdata->data = (caddr_t)data;
866 		}
867 	} else 	{
868 		secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
869 		secdata->secmod = secdata->rpcflavor = AUTH_UNIX;
870 		secdata->data = NULL;
871 	}
872 
873 	svp->sv_secdata = secdata;
874 	if (error)
875 		goto errout;
876 
877 	/*
878 	 * See bug 1180236.
879 	 * If mount secure failed, we will fall back to AUTH_NONE
880 	 * and try again.  nfs3rootvp() will turn this back off.
881 	 *
882 	 * The NFS Version 3 mount uses the FSINFO and GETATTR
883 	 * procedures.  The server should not care if these procedures
884 	 * have the proper security flavor, so if mount retries using
885 	 * AUTH_NONE that does not require a credential setup for root
886 	 * then the automounter would work without requiring root to be
887 	 * keylogged into AUTH_DES.
888 	 */
889 	if (secdata->rpcflavor != AUTH_UNIX &&
890 	    secdata->rpcflavor != AUTH_LOOPBACK)
891 		secdata->flags |= AUTH_F_TRYNONE;
892 
893 	/*
894 	 * Failover support:
895 	 *
896 	 * We may have a linked list of nfs_args structures,
897 	 * which means the user is looking for failover.  If
898 	 * the mount is either not "read-only" or "soft",
899 	 * we want to bail out with EINVAL.
900 	 */
901 	if (args->nfs_args_ext == NFS_ARGS_EXTB &&
902 	    args->nfs_ext_u.nfs_extB.next != NULL) {
903 		if (uap->flags & MS_RDONLY && !(flags & NFSMNT_SOFT)) {
904 			data = (char *)args->nfs_ext_u.nfs_extB.next;
905 			goto more;
906 		}
907 		error = EINVAL;
908 		goto errout;
909 	}
910 
911 	/*
912 	 * Determine the zone we're being mounted into.
913 	 */
914 	zone_hold(mntzone = zone);		/* start with this assumption */
915 	if (getzoneid() == GLOBAL_ZONEID) {
916 		zone_rele(mntzone);
917 		mntzone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
918 		ASSERT(mntzone != NULL);
919 		if (mntzone != zone) {
920 			error = EBUSY;
921 			goto errout;
922 		}
923 	}
924 
925 	if (is_system_labeled()) {
926 		error = nfs_mount_label_policy(vfsp, &svp->sv_addr,
927 		    svp->sv_knconf, cr);
928 
929 		if (error > 0)
930 			goto errout;
931 
932 		if (error == -1) {
933 			/* change mount to read-only to prevent write-down */
934 			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
935 		}
936 	}
937 
938 	/*
939 	 * Stop the mount from going any further if the zone is going away.
940 	 */
941 	if (zone_status_get(mntzone) >= ZONE_IS_SHUTTING_DOWN) {
942 		error = EBUSY;
943 		goto errout;
944 	}
945 
946 	/*
947 	 * Get root vnode.
948 	 */
949 proceed:
950 	error = nfs3rootvp(&rtvp, vfsp, svp_head, flags, cr, mntzone);
951 
952 	if (error)
953 		goto errout;
954 
955 	/*
956 	 * Set option fields in the mount info record
957 	 */
958 	mi = VTOMI(rtvp);
959 
960 	if (svp_head->sv_next)
961 		mi->mi_flags |= MI_LLOCK;
962 
963 	error = nfs_setopts(rtvp, DATAMODEL_NATIVE, args);
964 
965 errout:
966 	if (rtvp != NULL) {
967 		if (error) {
968 			rp = VTOR(rtvp);
969 			if (rp->r_flags & RHASHED)
970 				rp_rmhash(rp);
971 		}
972 		VN_RELE(rtvp);
973 	}
974 
975 	if (error) {
976 		sv_free(svp_head);
977 		if (mi != NULL) {
978 			nfs_async_stop(vfsp);
979 			nfs_async_manager_stop(vfsp);
980 			if (mi->mi_io_kstats) {
981 				kstat_delete(mi->mi_io_kstats);
982 				mi->mi_io_kstats = NULL;
983 			}
984 			if (mi->mi_ro_kstats) {
985 				kstat_delete(mi->mi_ro_kstats);
986 				mi->mi_ro_kstats = NULL;
987 			}
988 			nfs_free_mi(mi);
989 		}
990 	}
991 
992 
993 	if (!(uap->flags & MS_SYSSPACE)) {
994 		nfs3_free_args(args, fhandle);
995 		kmem_free(args, sizeof (*args));
996 	}
997 
998 	if (mntzone != NULL)
999 		zone_rele(mntzone);
1000 
1001 	return (error);
1002 }
1003 
1004 static int nfs3_dynamic = 0;	/* global variable to enable dynamic retrans. */
1005 static ushort_t nfs3_max_threads = 8;	/* max number of active async threads */
1006 uint_t nfs3_bsize = 32 * 1024;	/* client `block' size */
1007 static uint_t nfs3_async_clusters = 1;	/* # of reqs from each async queue */
1008 static uint_t nfs3_cots_timeo = NFS_COTS_TIMEO;
1009 
1010 static int
1011 nfs3rootvp(vnode_t **rtvpp, vfs_t *vfsp, struct servinfo *svp,
1012 	int flags, cred_t *cr, zone_t *zone)
1013 {
1014 	vnode_t *rtvp;
1015 	mntinfo_t *mi;
1016 	dev_t nfs_dev;
1017 	struct vattr va;
1018 	struct FSINFO3args args;
1019 	struct FSINFO3res res;
1020 	int error;
1021 	int douprintf;
1022 	rnode_t *rp;
1023 	int i;
1024 	uint_t max_transfer_size;
1025 	struct nfs_stats *nfsstatsp;
1026 	cred_t *lcr = NULL, *tcr = cr;
1027 
1028 	nfsstatsp = zone_getspecific(nfsstat_zone_key, nfs_zone());
1029 	ASSERT(nfsstatsp != NULL);
1030 
1031 	ASSERT(nfs_zone() == zone);
1032 	/*
1033 	 * Create a mount record and link it to the vfs struct.
1034 	 */
1035 	mi = kmem_zalloc(sizeof (*mi), KM_SLEEP);
1036 	mutex_init(&mi->mi_lock, NULL, MUTEX_DEFAULT, NULL);
1037 	mutex_init(&mi->mi_remap_lock, NULL, MUTEX_DEFAULT, NULL);
1038 	mi->mi_flags = MI_ACL | MI_EXTATTR;
1039 	if (!(flags & NFSMNT_SOFT))
1040 		mi->mi_flags |= MI_HARD;
1041 	if ((flags & NFSMNT_SEMISOFT))
1042 		mi->mi_flags |= MI_SEMISOFT;
1043 	if ((flags & NFSMNT_NOPRINT))
1044 		mi->mi_flags |= MI_NOPRINT;
1045 	if (flags & NFSMNT_INT)
1046 		mi->mi_flags |= MI_INT;
1047 	mi->mi_retrans = NFS_RETRIES;
1048 	if (svp->sv_knconf->knc_semantics == NC_TPI_COTS_ORD ||
1049 	    svp->sv_knconf->knc_semantics == NC_TPI_COTS)
1050 		mi->mi_timeo = nfs3_cots_timeo;
1051 	else
1052 		mi->mi_timeo = NFS_TIMEO;
1053 	mi->mi_prog = NFS_PROGRAM;
1054 	mi->mi_vers = NFS_V3;
1055 	mi->mi_rfsnames = rfsnames_v3;
1056 	mi->mi_reqs = nfsstatsp->nfs_stats_v3.rfsreqcnt_ptr;
1057 	mi->mi_call_type = call_type_v3;
1058 	mi->mi_ss_call_type = ss_call_type_v3;
1059 	mi->mi_timer_type = timer_type_v3;
1060 	mi->mi_aclnames = aclnames_v3;
1061 	mi->mi_aclreqs = nfsstatsp->nfs_stats_v3.aclreqcnt_ptr;
1062 	mi->mi_acl_call_type = acl_call_type_v3;
1063 	mi->mi_acl_ss_call_type = acl_ss_call_type_v3;
1064 	mi->mi_acl_timer_type = acl_timer_type_v3;
1065 	cv_init(&mi->mi_failover_cv, NULL, CV_DEFAULT, NULL);
1066 	mi->mi_servers = svp;
1067 	mi->mi_curr_serv = svp;
1068 	mi->mi_acregmin = SEC2HR(ACREGMIN);
1069 	mi->mi_acregmax = SEC2HR(ACREGMAX);
1070 	mi->mi_acdirmin = SEC2HR(ACDIRMIN);
1071 	mi->mi_acdirmax = SEC2HR(ACDIRMAX);
1072 
1073 	if (nfs3_dynamic)
1074 		mi->mi_flags |= MI_DYNAMIC;
1075 
1076 	if (flags & NFSMNT_DIRECTIO)
1077 		mi->mi_flags |= MI_DIRECTIO;
1078 
1079 	/*
1080 	 * Make a vfs struct for nfs.  We do this here instead of below
1081 	 * because rtvp needs a vfs before we can do a getattr on it.
1082 	 *
1083 	 * Assign a unique device id to the mount
1084 	 */
1085 	mutex_enter(&nfs_minor_lock);
1086 	do {
1087 		nfs_minor = (nfs_minor + 1) & MAXMIN32;
1088 		nfs_dev = makedevice(nfs_major, nfs_minor);
1089 	} while (vfs_devismounted(nfs_dev));
1090 	mutex_exit(&nfs_minor_lock);
1091 
1092 	vfsp->vfs_dev = nfs_dev;
1093 	vfs_make_fsid(&vfsp->vfs_fsid, nfs_dev, nfs3fstyp);
1094 	vfsp->vfs_data = (caddr_t)mi;
1095 	vfsp->vfs_fstype = nfsfstyp;
1096 
1097 	/*
1098 	 * Verify that nfs3_bsize tuneable is set to an
1099 	 * acceptable value.  It be a multiple of PAGESIZE or
1100 	 * file corruption can occur.
1101 	 */
1102 	if (nfs3_bsize & PAGEOFFSET)
1103 		nfs3_bsize &= PAGEMASK;
1104 	if (nfs3_bsize < PAGESIZE)
1105 		nfs3_bsize = PAGESIZE;
1106 	vfsp->vfs_bsize = nfs3_bsize;
1107 
1108 	/*
1109 	 * Initialize fields used to support async putpage operations.
1110 	 */
1111 	for (i = 0; i < NFS_ASYNC_TYPES; i++)
1112 		mi->mi_async_clusters[i] = nfs3_async_clusters;
1113 	mi->mi_async_init_clusters = nfs3_async_clusters;
1114 	mi->mi_async_curr[NFS_ASYNC_QUEUE] =
1115 	    mi->mi_async_curr[NFS_ASYNC_PGOPS_QUEUE] = &mi->mi_async_reqs[0];
1116 	mi->mi_max_threads = nfs3_max_threads;
1117 	mutex_init(&mi->mi_async_lock, NULL, MUTEX_DEFAULT, NULL);
1118 	cv_init(&mi->mi_async_reqs_cv, NULL, CV_DEFAULT, NULL);
1119 	cv_init(&mi->mi_async_work_cv[NFS_ASYNC_QUEUE], NULL, CV_DEFAULT, NULL);
1120 	cv_init(&mi->mi_async_work_cv[NFS_ASYNC_PGOPS_QUEUE], NULL,
1121 	    CV_DEFAULT, NULL);
1122 	cv_init(&mi->mi_async_cv, NULL, CV_DEFAULT, NULL);
1123 
1124 	mi->mi_vfsp = vfsp;
1125 	zone_hold(mi->mi_zone = zone);
1126 	nfs_mi_zonelist_add(mi);
1127 
1128 	/*
1129 	 * Make the root vnode, use it to get attributes,
1130 	 * then remake it with the attributes.
1131 	 */
1132 	rtvp = makenfs3node((nfs_fh3 *)&svp->sv_fhandle,
1133 	    NULL, vfsp, gethrtime(), cr, NULL, NULL);
1134 
1135 	/*
1136 	 * Make the FSINFO calls, primarily at this point to
1137 	 * determine the transfer size.  For client failover,
1138 	 * we'll want this to be the minimum bid from any
1139 	 * server, so that we don't overrun stated limits.
1140 	 *
1141 	 * While we're looping, we'll turn off AUTH_F_TRYNONE,
1142 	 * which is only for the mount operation.
1143 	 */
1144 
1145 	mi->mi_tsize = nfs3_tsize(svp->sv_knconf);
1146 	mi->mi_stsize = mi->mi_tsize;
1147 
1148 	mi->mi_curread = nfs3_bsize;
1149 	mi->mi_curwrite = mi->mi_curread;
1150 
1151 	/*
1152 	 * If the uid is set then set the creds for secure mounts
1153 	 * by proxy processes such as automountd.
1154 	 */
1155 	if (svp->sv_secdata->uid != 0 &&
1156 	    svp->sv_secdata->rpcflavor == RPCSEC_GSS) {
1157 		lcr = crdup(cr);
1158 		(void) crsetugid(lcr, svp->sv_secdata->uid, crgetgid(cr));
1159 		tcr = lcr;
1160 	}
1161 
1162 	for (svp = mi->mi_servers; svp != NULL; svp = svp->sv_next) {
1163 		douprintf = 1;
1164 		mi->mi_curr_serv = svp;
1165 		max_transfer_size = nfs3_tsize(svp->sv_knconf);
1166 		mi->mi_tsize = MIN(max_transfer_size, mi->mi_tsize);
1167 		mi->mi_stsize = MIN(max_transfer_size, mi->mi_stsize);
1168 		mi->mi_curread = MIN(max_transfer_size, mi->mi_curread);
1169 		mi->mi_curwrite = MIN(max_transfer_size, mi->mi_curwrite);
1170 		args.fsroot = *(nfs_fh3 *)&svp->sv_fhandle;
1171 
1172 		error = rfs3call(mi, NFSPROC3_FSINFO,
1173 		    xdr_nfs_fh3, (caddr_t)&args,
1174 		    xdr_FSINFO3res, (caddr_t)&res, tcr,
1175 		    &douprintf, &res.status, 0, NULL);
1176 		if (error)
1177 			goto bad;
1178 		error = geterrno3(res.status);
1179 		if (error)
1180 			goto bad;
1181 
1182 		/* get type of root node */
1183 		if (res.resok.obj_attributes.attributes) {
1184 			if (res.resok.obj_attributes.attr.type < NF3REG ||
1185 			    res.resok.obj_attributes.attr.type > NF3FIFO) {
1186 #ifdef DEBUG
1187 				zcmn_err(getzoneid(), CE_WARN,
1188 			    "NFS3 server %s returned a bad file type for root",
1189 				    svp->sv_hostname);
1190 #else
1191 				zcmn_err(getzoneid(), CE_WARN,
1192 			    "NFS server %s returned a bad file type for root",
1193 				    svp->sv_hostname);
1194 #endif
1195 				error = EINVAL;
1196 				goto bad;
1197 			} else {
1198 				if (rtvp->v_type != VNON && rtvp->v_type !=
1199 				    nf3_to_vt[res.resok.obj_attributes.attr.
1200 				    type]) {
1201 #ifdef DEBUG
1202 					zcmn_err(getzoneid(), CE_WARN,
1203 		"NFS3 server %s returned a different file type for root",
1204 					    svp->sv_hostname);
1205 #else
1206 					zcmn_err(getzoneid(), CE_WARN,
1207 		"NFS server %s returned a different file type for root",
1208 					    svp->sv_hostname);
1209 #endif
1210 					error = EINVAL;
1211 					goto bad;
1212 				}
1213 				rtvp->v_type =
1214 				    nf3_to_vt[res.resok.obj_attributes.attr.
1215 				    type];
1216 			}
1217 		}
1218 
1219 		if (res.resok.rtmax != 0) {
1220 			mi->mi_tsize = MIN(res.resok.rtmax, mi->mi_tsize);
1221 			if (res.resok.rtpref != 0) {
1222 				mi->mi_curread = MIN(res.resok.rtpref,
1223 				    mi->mi_curread);
1224 			} else {
1225 				mi->mi_curread = MIN(res.resok.rtmax,
1226 				    mi->mi_curread);
1227 			}
1228 		} else if (res.resok.rtpref != 0) {
1229 			mi->mi_tsize = MIN(res.resok.rtpref, mi->mi_tsize);
1230 			mi->mi_curread = MIN(res.resok.rtpref, mi->mi_curread);
1231 		} else {
1232 #ifdef DEBUG
1233 			zcmn_err(getzoneid(), CE_WARN,
1234 			    "NFS3 server %s returned 0 for read transfer sizes",
1235 			    svp->sv_hostname);
1236 #else
1237 			zcmn_err(getzoneid(), CE_WARN,
1238 			    "NFS server %s returned 0 for read transfer sizes",
1239 			    svp->sv_hostname);
1240 #endif
1241 			error = EIO;
1242 			goto bad;
1243 		}
1244 		if (res.resok.wtmax != 0) {
1245 			mi->mi_stsize = MIN(res.resok.wtmax, mi->mi_stsize);
1246 			if (res.resok.wtpref != 0) {
1247 				mi->mi_curwrite = MIN(res.resok.wtpref,
1248 				    mi->mi_curwrite);
1249 			} else {
1250 				mi->mi_curwrite = MIN(res.resok.wtmax,
1251 				    mi->mi_curwrite);
1252 			}
1253 		} else if (res.resok.wtpref != 0) {
1254 			mi->mi_stsize = MIN(res.resok.wtpref, mi->mi_stsize);
1255 			mi->mi_curwrite = MIN(res.resok.wtpref,
1256 			    mi->mi_curwrite);
1257 		} else {
1258 #ifdef DEBUG
1259 			zcmn_err(getzoneid(), CE_WARN,
1260 			"NFS3 server %s returned 0 for write transfer sizes",
1261 			    svp->sv_hostname);
1262 #else
1263 			zcmn_err(getzoneid(), CE_WARN,
1264 			"NFS server %s returned 0 for write transfer sizes",
1265 			    svp->sv_hostname);
1266 #endif
1267 			error = EIO;
1268 			goto bad;
1269 		}
1270 
1271 		/*
1272 		 * These signal the ability of the server to create
1273 		 * hard links and symbolic links, so they really
1274 		 * aren't relevant if there is more than one server.
1275 		 * We'll set them here, though it probably looks odd.
1276 		 */
1277 		if (res.resok.properties & FSF3_LINK)
1278 			mi->mi_flags |= MI_LINK;
1279 		if (res.resok.properties & FSF3_SYMLINK)
1280 			mi->mi_flags |= MI_SYMLINK;
1281 
1282 		/* Pick up smallest non-zero maxfilesize value */
1283 		if (res.resok.maxfilesize) {
1284 			if (mi->mi_maxfilesize) {
1285 				mi->mi_maxfilesize = MIN(mi->mi_maxfilesize,
1286 				    res.resok.maxfilesize);
1287 			} else
1288 				mi->mi_maxfilesize = res.resok.maxfilesize;
1289 		}
1290 
1291 		/*
1292 		 * AUTH_F_TRYNONE is only for the mount operation,
1293 		 * so turn it back off.
1294 		 */
1295 		svp->sv_secdata->flags &= ~AUTH_F_TRYNONE;
1296 	}
1297 	mi->mi_curr_serv = mi->mi_servers;
1298 
1299 	/*
1300 	 * Start the thread responsible for handling async worker threads.
1301 	 */
1302 	VFS_HOLD(vfsp);	/* add reference for thread */
1303 	mi->mi_manager_thread = zthread_create(NULL, 0, nfs_async_manager,
1304 	    vfsp, 0, minclsyspri);
1305 	ASSERT(mi->mi_manager_thread != NULL);
1306 
1307 	/*
1308 	 * Initialize kstats
1309 	 */
1310 	nfs_mnt_kstat_init(vfsp);
1311 
1312 	/* If we didn't get a type, get one now */
1313 	if (rtvp->v_type == VNON) {
1314 		va.va_mask = AT_ALL;
1315 
1316 		error = nfs3getattr(rtvp, &va, tcr);
1317 		if (error)
1318 			goto bad;
1319 		rtvp->v_type = va.va_type;
1320 	}
1321 
1322 	mi->mi_type = rtvp->v_type;
1323 
1324 	*rtvpp = rtvp;
1325 	if (lcr != NULL)
1326 		crfree(lcr);
1327 
1328 	return (0);
1329 bad:
1330 	/*
1331 	 * An error occurred somewhere, need to clean up...
1332 	 * We need to release our reference to the root vnode and
1333 	 * destroy the mntinfo struct that we just created.
1334 	 */
1335 	if (lcr != NULL)
1336 		crfree(lcr);
1337 	rp = VTOR(rtvp);
1338 	if (rp->r_flags & RHASHED)
1339 		rp_rmhash(rp);
1340 	VN_RELE(rtvp);
1341 	nfs_async_stop(vfsp);
1342 	nfs_async_manager_stop(vfsp);
1343 	if (mi->mi_io_kstats) {
1344 		kstat_delete(mi->mi_io_kstats);
1345 		mi->mi_io_kstats = NULL;
1346 	}
1347 	if (mi->mi_ro_kstats) {
1348 		kstat_delete(mi->mi_ro_kstats);
1349 		mi->mi_ro_kstats = NULL;
1350 	}
1351 	nfs_free_mi(mi);
1352 	*rtvpp = NULL;
1353 	return (error);
1354 }
1355 
1356 /*
1357  * vfs operations
1358  */
1359 static int
1360 nfs3_unmount(vfs_t *vfsp, int flag, cred_t *cr)
1361 {
1362 	mntinfo_t *mi;
1363 	ushort_t omax;
1364 
1365 	if (secpolicy_fs_unmount(cr, vfsp) != 0)
1366 		return (EPERM);
1367 
1368 	mi = VFTOMI(vfsp);
1369 	if (flag & MS_FORCE) {
1370 
1371 		vfsp->vfs_flag |= VFS_UNMOUNTED;
1372 
1373 		/*
1374 		 * We are about to stop the async manager.
1375 		 * Let every one know not to schedule any
1376 		 * more async requests
1377 		 */
1378 		mutex_enter(&mi->mi_async_lock);
1379 		mi->mi_max_threads = 0;
1380 		NFS_WAKEALL_ASYNC_WORKERS(mi->mi_async_work_cv);
1381 		mutex_exit(&mi->mi_async_lock);
1382 
1383 		/*
1384 		 * We need to stop the manager thread explicitly; the worker
1385 		 * threads can time out and exit on their own.
1386 		 */
1387 		nfs_async_manager_stop(vfsp);
1388 		destroy_rtable(vfsp, cr);
1389 		if (mi->mi_io_kstats) {
1390 			kstat_delete(mi->mi_io_kstats);
1391 			mi->mi_io_kstats = NULL;
1392 		}
1393 		if (mi->mi_ro_kstats) {
1394 			kstat_delete(mi->mi_ro_kstats);
1395 			mi->mi_ro_kstats = NULL;
1396 		}
1397 		return (0);
1398 	}
1399 	/*
1400 	 * Wait until all asynchronous putpage operations on
1401 	 * this file system are complete before flushing rnodes
1402 	 * from the cache.
1403 	 */
1404 	omax = mi->mi_max_threads;
1405 	if (nfs_async_stop_sig(vfsp)) {
1406 		return (EINTR);
1407 	}
1408 	rflush(vfsp, cr);
1409 	/*
1410 	 * If there are any active vnodes on this file system,
1411 	 * then the file system is busy and can't be umounted.
1412 	 */
1413 	if (check_rtable(vfsp)) {
1414 		mutex_enter(&mi->mi_async_lock);
1415 		mi->mi_max_threads = omax;
1416 		mutex_exit(&mi->mi_async_lock);
1417 		return (EBUSY);
1418 	}
1419 	/*
1420 	 * The unmount can't fail from now on; stop the worker thread manager.
1421 	 */
1422 	nfs_async_manager_stop(vfsp);
1423 	/*
1424 	 * Destroy all rnodes belonging to this file system from the
1425 	 * rnode hash queues and purge any resources allocated to
1426 	 * them.
1427 	 */
1428 	destroy_rtable(vfsp, cr);
1429 	if (mi->mi_io_kstats) {
1430 		kstat_delete(mi->mi_io_kstats);
1431 		mi->mi_io_kstats = NULL;
1432 	}
1433 	if (mi->mi_ro_kstats) {
1434 		kstat_delete(mi->mi_ro_kstats);
1435 		mi->mi_ro_kstats = NULL;
1436 	}
1437 	return (0);
1438 }
1439 
1440 /*
1441  * find root of nfs
1442  */
1443 static int
1444 nfs3_root(vfs_t *vfsp, vnode_t **vpp)
1445 {
1446 	mntinfo_t *mi;
1447 	vnode_t *vp;
1448 	servinfo_t *svp;
1449 	rnode_t *rp;
1450 	int error = 0;
1451 
1452 	mi = VFTOMI(vfsp);
1453 
1454 	if (nfs_zone() != mi->mi_zone)
1455 		return (EPERM);
1456 
1457 	svp = mi->mi_curr_serv;
1458 	if (svp && (svp->sv_flags & SV_ROOT_STALE)) {
1459 		mutex_enter(&svp->sv_lock);
1460 		svp->sv_flags &= ~SV_ROOT_STALE;
1461 		mutex_exit(&svp->sv_lock);
1462 		error = ENOENT;
1463 	}
1464 
1465 	vp = makenfs3node((nfs_fh3 *)&mi->mi_curr_serv->sv_fhandle,
1466 	    NULL, vfsp, gethrtime(), CRED(), NULL, NULL);
1467 
1468 	/*
1469 	 * if the SV_ROOT_STALE flag was reset above, reset the
1470 	 * RSTALE flag if needed and return an error
1471 	 */
1472 	if (error == ENOENT) {
1473 		rp = VTOR(vp);
1474 		if (svp && rp->r_flags & RSTALE) {
1475 			mutex_enter(&rp->r_statelock);
1476 			rp->r_flags &= ~RSTALE;
1477 			mutex_exit(&rp->r_statelock);
1478 		}
1479 		VN_RELE(vp);
1480 		return (error);
1481 	}
1482 
1483 	ASSERT(vp->v_type == VNON || vp->v_type == mi->mi_type);
1484 
1485 	vp->v_type = mi->mi_type;
1486 
1487 	*vpp = vp;
1488 
1489 	return (0);
1490 }
1491 
1492 /*
1493  * Get file system statistics.
1494  */
1495 static int
1496 nfs3_statvfs(vfs_t *vfsp, struct statvfs64 *sbp)
1497 {
1498 	int error;
1499 	struct mntinfo *mi;
1500 	struct FSSTAT3args args;
1501 	struct FSSTAT3res res;
1502 	int douprintf;
1503 	failinfo_t fi;
1504 	vnode_t *vp;
1505 	cred_t *cr;
1506 	hrtime_t t;
1507 
1508 	mi = VFTOMI(vfsp);
1509 	if (nfs_zone() != mi->mi_zone)
1510 		return (EPERM);
1511 	error = nfs3_root(vfsp, &vp);
1512 	if (error)
1513 		return (error);
1514 
1515 	cr = CRED();
1516 
1517 	args.fsroot = *VTOFH3(vp);
1518 	fi.vp = vp;
1519 	fi.fhp = (caddr_t)&args.fsroot;
1520 	fi.copyproc = nfs3copyfh;
1521 	fi.lookupproc = nfs3lookup;
1522 	fi.xattrdirproc = acl_getxattrdir3;
1523 
1524 	douprintf = 1;
1525 
1526 	t = gethrtime();
1527 
1528 	error = rfs3call(mi, NFSPROC3_FSSTAT,
1529 	    xdr_nfs_fh3, (caddr_t)&args,
1530 	    xdr_FSSTAT3res, (caddr_t)&res, cr,
1531 	    &douprintf, &res.status, 0, &fi);
1532 
1533 	if (error) {
1534 		VN_RELE(vp);
1535 		return (error);
1536 	}
1537 
1538 	error = geterrno3(res.status);
1539 	if (!error) {
1540 		nfs3_cache_post_op_attr(vp, &res.resok.obj_attributes, t, cr);
1541 		sbp->f_bsize = MAXBSIZE;
1542 		sbp->f_frsize = DEV_BSIZE;
1543 		/*
1544 		 * Allow -1 fields to pass through unconverted.  These
1545 		 * indicate "don't know" fields.
1546 		 */
1547 		if (res.resok.tbytes == (size3)-1)
1548 			sbp->f_blocks = (fsblkcnt64_t)res.resok.tbytes;
1549 		else {
1550 			sbp->f_blocks = (fsblkcnt64_t)
1551 			    (res.resok.tbytes / DEV_BSIZE);
1552 		}
1553 		if (res.resok.fbytes == (size3)-1)
1554 			sbp->f_bfree = (fsblkcnt64_t)res.resok.fbytes;
1555 		else {
1556 			sbp->f_bfree = (fsblkcnt64_t)
1557 			    (res.resok.fbytes / DEV_BSIZE);
1558 		}
1559 		if (res.resok.abytes == (size3)-1)
1560 			sbp->f_bavail = (fsblkcnt64_t)res.resok.abytes;
1561 		else {
1562 			sbp->f_bavail = (fsblkcnt64_t)
1563 			    (res.resok.abytes / DEV_BSIZE);
1564 		}
1565 		sbp->f_files = (fsfilcnt64_t)res.resok.tfiles;
1566 		sbp->f_ffree = (fsfilcnt64_t)res.resok.ffiles;
1567 		sbp->f_favail = (fsfilcnt64_t)res.resok.afiles;
1568 		sbp->f_fsid = (unsigned long)vfsp->vfs_fsid.val[0];
1569 		(void) strncpy(sbp->f_basetype,
1570 		    vfssw[vfsp->vfs_fstype].vsw_name, FSTYPSZ);
1571 		sbp->f_flag = vf_to_stf(vfsp->vfs_flag);
1572 		sbp->f_namemax = (ulong_t)-1;
1573 	} else {
1574 		nfs3_cache_post_op_attr(vp, &res.resfail.obj_attributes, t, cr);
1575 		PURGE_STALE_FH(error, vp, cr);
1576 	}
1577 
1578 	VN_RELE(vp);
1579 
1580 	return (error);
1581 }
1582 
1583 static kmutex_t nfs3_syncbusy;
1584 
1585 /*
1586  * Flush dirty nfs files for file system vfsp.
1587  * If vfsp == NULL, all nfs files are flushed.
1588  */
1589 /* ARGSUSED */
1590 static int
1591 nfs3_sync(vfs_t *vfsp, short flag, cred_t *cr)
1592 {
1593 	/*
1594 	 * Cross-zone calls are OK here, since this translates to a
1595 	 * VOP_PUTPAGE(B_ASYNC), which gets picked up by the right zone.
1596 	 */
1597 	if (!(flag & SYNC_ATTR) && mutex_tryenter(&nfs3_syncbusy) != 0) {
1598 		rflush(vfsp, cr);
1599 		mutex_exit(&nfs3_syncbusy);
1600 	}
1601 	return (0);
1602 }
1603 
1604 /* ARGSUSED */
1605 static int
1606 nfs3_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1607 {
1608 	int error;
1609 	nfs_fh3 fh;
1610 	vnode_t *vp;
1611 	struct vattr va;
1612 
1613 	if (fidp->fid_len > NFS3_FHSIZE) {
1614 		*vpp = NULL;
1615 		return (ESTALE);
1616 	}
1617 
1618 	if (nfs_zone() != VFTOMI(vfsp)->mi_zone)
1619 		return (EPERM);
1620 	fh.fh3_length = fidp->fid_len;
1621 	bcopy(fidp->fid_data, fh.fh3_u.data, fh.fh3_length);
1622 
1623 	vp = makenfs3node(&fh, NULL, vfsp, gethrtime(), CRED(), NULL, NULL);
1624 
1625 	if (VTOR(vp)->r_flags & RSTALE) {
1626 		VN_RELE(vp);
1627 		*vpp = NULL;
1628 		return (ENOENT);
1629 	}
1630 
1631 	if (vp->v_type == VNON) {
1632 		va.va_mask = AT_ALL;
1633 		error = nfs3getattr(vp, &va, CRED());
1634 		if (error) {
1635 			VN_RELE(vp);
1636 			*vpp = NULL;
1637 			return (error);
1638 		}
1639 		vp->v_type = va.va_type;
1640 	}
1641 
1642 	*vpp = vp;
1643 
1644 	return (0);
1645 }
1646 
1647 /* ARGSUSED */
1648 static int
1649 nfs3_mountroot(vfs_t *vfsp, whymountroot_t why)
1650 {
1651 	vnode_t *rtvp;
1652 	char root_hostname[SYS_NMLN+1];
1653 	struct servinfo *svp;
1654 	int error;
1655 	int vfsflags;
1656 	size_t size;
1657 	char *root_path;
1658 	struct pathname pn;
1659 	char *name;
1660 	cred_t *cr;
1661 	struct nfs_args args;		/* nfs mount arguments */
1662 	static char token[10];
1663 
1664 	bzero(&args, sizeof (args));
1665 
1666 	/* do this BEFORE getfile which causes xid stamps to be initialized */
1667 	clkset(-1L);		/* hack for now - until we get time svc? */
1668 
1669 	if (why == ROOT_REMOUNT) {
1670 		/*
1671 		 * Shouldn't happen.
1672 		 */
1673 		panic("nfs3_mountroot: why == ROOT_REMOUNT");
1674 	}
1675 
1676 	if (why == ROOT_UNMOUNT) {
1677 		/*
1678 		 * Nothing to do for NFS.
1679 		 */
1680 		return (0);
1681 	}
1682 
1683 	/*
1684 	 * why == ROOT_INIT
1685 	 */
1686 
1687 	name = token;
1688 	*name = 0;
1689 	getfsname("root", name, sizeof (token));
1690 
1691 	pn_alloc(&pn);
1692 	root_path = pn.pn_path;
1693 
1694 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
1695 	svp->sv_knconf = kmem_zalloc(sizeof (*svp->sv_knconf), KM_SLEEP);
1696 	svp->sv_knconf->knc_protofmly = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
1697 	svp->sv_knconf->knc_proto = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
1698 
1699 	/*
1700 	 * Get server address
1701 	 * Get the root fhandle
1702 	 * Get server's transport
1703 	 * Get server's hostname
1704 	 * Get options
1705 	 */
1706 	args.addr = &svp->sv_addr;
1707 	args.fh = (char *)&svp->sv_fhandle;
1708 	args.knconf = svp->sv_knconf;
1709 	args.hostname = root_hostname;
1710 	vfsflags = 0;
1711 	if (error = mount_root(*name ? name : "root", root_path, NFS_V3,
1712 	    &args, &vfsflags)) {
1713 		if (error == EPROTONOSUPPORT)
1714 			nfs_cmn_err(error, CE_WARN, "nfs3_mountroot: "
1715 			    "mount_root failed: server doesn't support NFS V3");
1716 		else
1717 			nfs_cmn_err(error, CE_WARN,
1718 			    "nfs3_mountroot: mount_root failed: %m");
1719 		sv_free(svp);
1720 		pn_free(&pn);
1721 		return (error);
1722 	}
1723 	svp->sv_hostnamelen = (int)(strlen(root_hostname) + 1);
1724 	svp->sv_hostname = kmem_alloc(svp->sv_hostnamelen, KM_SLEEP);
1725 	(void) strcpy(svp->sv_hostname, root_hostname);
1726 
1727 	/*
1728 	 * Force root partition to always be mounted with AUTH_UNIX for now
1729 	 */
1730 	svp->sv_secdata = kmem_alloc(sizeof (*svp->sv_secdata), KM_SLEEP);
1731 	svp->sv_secdata->secmod = AUTH_UNIX;
1732 	svp->sv_secdata->rpcflavor = AUTH_UNIX;
1733 	svp->sv_secdata->data = NULL;
1734 
1735 	cr = crgetcred();
1736 	rtvp = NULL;
1737 
1738 	error = nfs3rootvp(&rtvp, vfsp, svp, args.flags, cr, global_zone);
1739 
1740 	crfree(cr);
1741 
1742 	if (error) {
1743 		pn_free(&pn);
1744 		sv_free(svp);
1745 		return (error);
1746 	}
1747 
1748 	error = nfs_setopts(rtvp, DATAMODEL_NATIVE, &args);
1749 	if (error) {
1750 		nfs_cmn_err(error, CE_WARN,
1751 		    "nfs3_mountroot: invalid root mount options");
1752 		pn_free(&pn);
1753 		goto errout;
1754 	}
1755 
1756 	(void) vfs_lock_wait(vfsp);
1757 	vfs_add(NULL, vfsp, vfsflags);
1758 	vfs_unlock(vfsp);
1759 
1760 	size = strlen(svp->sv_hostname);
1761 	(void) strcpy(rootfs.bo_name, svp->sv_hostname);
1762 	rootfs.bo_name[size] = ':';
1763 	(void) strcpy(&rootfs.bo_name[size + 1], root_path);
1764 
1765 	pn_free(&pn);
1766 
1767 errout:
1768 	if (error) {
1769 		sv_free(svp);
1770 		nfs_async_stop(vfsp);
1771 		nfs_async_manager_stop(vfsp);
1772 	}
1773 
1774 	if (rtvp != NULL)
1775 		VN_RELE(rtvp);
1776 
1777 	return (error);
1778 }
1779 
1780 /*
1781  * Initialization routine for VFS routines.  Should only be called once
1782  */
1783 int
1784 nfs3_vfsinit(void)
1785 {
1786 	mutex_init(&nfs3_syncbusy, NULL, MUTEX_DEFAULT, NULL);
1787 	return (0);
1788 }
1789 
1790 void
1791 nfs3_vfsfini(void)
1792 {
1793 	mutex_destroy(&nfs3_syncbusy);
1794 }
1795 
1796 void
1797 nfs3_freevfs(vfs_t *vfsp)
1798 {
1799 	mntinfo_t *mi;
1800 	servinfo_t *svp;
1801 
1802 	/* free up the resources */
1803 	mi = VFTOMI(vfsp);
1804 	svp = mi->mi_servers;
1805 	mi->mi_servers = mi->mi_curr_serv = NULL;
1806 	sv_free(svp);
1807 
1808 	/*
1809 	 * By this time we should have already deleted the
1810 	 * mi kstats in the unmount code. If they are still around
1811 	 * somethings wrong
1812 	 */
1813 	ASSERT(mi->mi_io_kstats == NULL);
1814 	nfs_free_mi(mi);
1815 }
1816