xref: /dragonfly/sys/vfs/nfs/nfs_vfsops.c (revision 685c703c)
1 /*
2  * Copyright (c) 1989, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)nfs_vfsops.c	8.12 (Berkeley) 5/20/95
37  * $FreeBSD: src/sys/nfs/nfs_vfsops.c,v 1.91.2.7 2003/01/27 20:04:08 dillon Exp $
38  * $DragonFly: src/sys/vfs/nfs/nfs_vfsops.c,v 1.44 2006/07/18 22:22:15 dillon Exp $
39  */
40 
41 #include "opt_bootp.h"
42 #include "opt_nfsroot.h"
43 
44 #include <sys/param.h>
45 #include <sys/sockio.h>
46 #include <sys/proc.h>
47 #include <sys/vnode.h>
48 #include <sys/fcntl.h>
49 #include <sys/kernel.h>
50 #include <sys/sysctl.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/mbuf.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/systm.h>
57 
58 #include <vm/vm.h>
59 #include <vm/vm_extern.h>
60 #include <vm/vm_zone.h>
61 
62 #include <net/if.h>
63 #include <net/route.h>
64 #include <netinet/in.h>
65 
66 #include <sys/thread2.h>
67 
68 #include "rpcv2.h"
69 #include "nfsproto.h"
70 #include "nfs.h"
71 #include "nfsmount.h"
72 #include "nfsnode.h"
73 #include "xdr_subs.h"
74 #include "nfsm_subs.h"
75 #include "nfsdiskless.h"
76 #include "nfsmountrpc.h"
77 
78 extern int	nfs_mountroot(struct mount *mp);
79 extern void	bootpc_init(void);
80 
81 extern int	nfs_ticks;
82 extern struct vop_ops nfsv2_vnode_vops;
83 extern struct vop_ops nfsv2_fifo_vops;
84 extern struct vop_ops nfsv2_spec_vops;
85 
86 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header");
87 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle");
88 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure");
89 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data");
90 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor");
91 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure");
92 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables");
93 
94 vm_zone_t nfsmount_zone;
95 
96 struct nfsstats	nfsstats;
97 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem");
98 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD,
99 	&nfsstats, nfsstats, "");
100 static int nfs_ip_paranoia = 1;
101 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW,
102 	&nfs_ip_paranoia, 0, "");
103 #ifdef NFS_DEBUG
104 int nfs_debug;
105 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, "");
106 #endif
107 
108 /*
109  * Tunable to determine the Read/Write unit size.  Maximum value
110  * is NFS_MAXDATA.  We also default to NFS_MAXDATA.
111  */
112 static int nfs_io_size = NFS_MAXDATA;
113 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW,
114 	&nfs_io_size, 0, "NFS optimal I/O unit size");
115 
116 static void	nfs_decode_args (struct nfsmount *nmp,
117 			struct nfs_args *argp);
118 static int	mountnfs (struct nfs_args *,struct mount *,
119 			struct sockaddr *,char *,char *,struct vnode **);
120 static int	nfs_mount ( struct mount *mp, char *path, caddr_t data,
121 			struct ucred *cred);
122 static int	nfs_unmount ( struct mount *mp, int mntflags);
123 static int	nfs_root ( struct mount *mp, struct vnode **vpp);
124 static int	nfs_statfs ( struct mount *mp, struct statfs *sbp,
125 			struct ucred *cred);
126 static int	nfs_sync ( struct mount *mp, int waitfor);
127 
128 /*
129  * nfs vfs operations.
130  */
131 static struct vfsops nfs_vfsops = {
132 	.vfs_mount =    	nfs_mount,
133 	.vfs_unmount =  	nfs_unmount,
134 	.vfs_root =     	nfs_root,
135 	.vfs_statfs =    	nfs_statfs,
136 	.vfs_sync =     	nfs_sync,
137 	.vfs_init =     	nfs_init,
138 	.vfs_uninit =    	nfs_uninit
139 };
140 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK);
141 
142 /*
143  * This structure must be filled in by a primary bootstrap or bootstrap
144  * server for a diskless/dataless machine. It is initialized below just
145  * to ensure that it is allocated to initialized data (.data not .bss).
146  */
147 struct nfs_diskless nfs_diskless = { { { 0 } } };
148 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } };
149 int nfs_diskless_valid = 0;
150 
151 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD,
152 	&nfs_diskless_valid, 0, "");
153 
154 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD,
155 	nfsv3_diskless.root_hostnam, 0, "");
156 
157 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD,
158 	&nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr,
159 	"%Ssockaddr_in", "");
160 
161 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD,
162 	nfsv3_diskless.swap_hostnam, 0, "");
163 
164 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD,
165 	&nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr,
166 	"%Ssockaddr_in","");
167 
168 
169 void nfsargs_ntoh (struct nfs_args *);
170 static int nfs_mountdiskless (char *, char *, int,
171 				  struct sockaddr_in *, struct nfs_args *,
172 				  struct thread *, struct vnode **,
173 				  struct mount **);
174 static void nfs_convert_diskless (void);
175 static void nfs_convert_oargs (struct nfs_args *args,
176 				   struct onfs_args *oargs);
177 
178 /*
179  * Calculate the buffer I/O block size to use.  The maximum V2 block size
180  * is typically 8K, the maximum datagram size is typically 16K, and the
181  * maximum V3 block size is typically 32K.  The buffer cache tends to work
182  * best with 16K blocks but we allow 32K for TCP connections.
183  *
184  * We force the block size to be at least a page for buffer cache efficiency.
185  */
186 static
187 int
188 nfs_iosize(int v3, int sotype)
189 {
190 	int iosize;
191 	int iomax;
192 
193 	if (v3) {
194 		if (sotype == SOCK_STREAM)
195 			iomax = NFS_MAXDATA;
196 		else
197 			iomax = NFS_MAXDGRAMDATA;
198 	} else {
199 		iomax = NFS_V2MAXDATA;
200 	}
201 	if ((iosize = nfs_io_size) > iomax)
202 		iosize = iomax;
203 	if (iosize < PAGE_SIZE)
204 		iosize = PAGE_SIZE;
205 
206 	/*
207 	 * This is an aweful hack but until the buffer cache is rewritten
208 	 * we need it.  The problem is that when you combine write() with
209 	 * mmap() the vm_page->valid bits can become weird looking
210 	 * (e.g. 0xfc).  This occurs because NFS uses piecemeal buffers
211 	 * at the file EOF.  To solve the problem the BIO system needs to
212 	 * be guarenteed that the NFS iosize for regular files will be a
213 	 * multiple of PAGE_SIZE so it can invalidate the whole page
214 	 * rather then just the piece of it owned by the buffer when
215 	 * NFS does vinvalbuf() calls.
216 	 */
217 	if (iosize & PAGE_MASK)
218 		iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE;
219 	return iosize;
220 }
221 
222 static void
223 nfs_convert_oargs(args, oargs)
224 	struct nfs_args *args;
225 	struct onfs_args *oargs;
226 {
227 	args->version = NFS_ARGSVERSION;
228 	args->addr = oargs->addr;
229 	args->addrlen = oargs->addrlen;
230 	args->sotype = oargs->sotype;
231 	args->proto = oargs->proto;
232 	args->fh = oargs->fh;
233 	args->fhsize = oargs->fhsize;
234 	args->flags = oargs->flags;
235 	args->wsize = oargs->wsize;
236 	args->rsize = oargs->rsize;
237 	args->readdirsize = oargs->readdirsize;
238 	args->timeo = oargs->timeo;
239 	args->retrans = oargs->retrans;
240 	args->maxgrouplist = oargs->maxgrouplist;
241 	args->readahead = oargs->readahead;
242 	args->deadthresh = oargs->deadthresh;
243 	args->hostname = oargs->hostname;
244 }
245 
246 static void
247 nfs_convert_diskless()
248 {
249 	int i;
250 
251 	bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif,
252 		sizeof(struct ifaliasreq));
253 	bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway,
254 		sizeof(struct sockaddr_in));
255 	nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args);
256 
257 	bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH);
258 	nfsv3_diskless.swap_fhsize = NFSX_V2FH;
259 	for (i = NFSX_V2FH - 1; i >= 0; --i) {
260 		if (nfs_diskless.swap_fh[i])
261 			break;
262 	}
263 	if (i < 0)
264 		nfsv3_diskless.swap_fhsize = 0;
265 
266 	bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr,
267 		sizeof(struct sockaddr_in));
268 	bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN);
269 	nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks;
270 	bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred,
271 		sizeof(struct ucred));
272 	nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args);
273 
274 	bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH);
275 	nfsv3_diskless.root_fhsize = NFSX_V2FH;
276 	for (i = NFSX_V2FH - 1; i >= 0; --i) {
277 		if (nfs_diskless.root_fh[i])
278 			break;
279 	}
280 	if (i < 0)
281 		nfsv3_diskless.root_fhsize = 0;
282 
283 	bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr,
284 		sizeof(struct sockaddr_in));
285 	bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN);
286 	nfsv3_diskless.root_time = nfs_diskless.root_time;
287 	bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam,
288 		MAXHOSTNAMELEN);
289 	nfs_diskless_valid = 3;
290 }
291 
292 /*
293  * nfs statfs call
294  */
295 int
296 nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
297 {
298 	struct vnode *vp;
299 	struct nfs_statfs *sfp;
300 	caddr_t cp;
301 	u_int32_t *tl;
302 	int32_t t1, t2;
303 	caddr_t bpos, dpos, cp2;
304 	struct nfsmount *nmp = VFSTONFS(mp);
305 	thread_t td = curthread;
306 	int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
307 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
308 	struct nfsnode *np;
309 	u_quad_t tquad;
310 
311 #ifndef nolint
312 	sfp = (struct nfs_statfs *)0;
313 #endif
314 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
315 	if (error)
316 		return (error);
317 	vp = NFSTOV(np);
318 	/* ignore the passed cred */
319 	cred = crget();
320 	cred->cr_ngroups = 1;
321 	if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0)
322 		(void)nfs_fsinfo(nmp, vp, td);
323 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
324 	nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
325 	nfsm_fhtom(vp, v3);
326 	nfsm_request(vp, NFSPROC_FSSTAT, td, cred);
327 	if (v3)
328 		nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
329 	if (error) {
330 		if (mrep != NULL)
331 			m_freem(mrep);
332 		goto nfsmout;
333 	}
334 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
335 	sbp->f_flags = nmp->nm_flag;
336 	sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype);
337 
338 	if (v3) {
339 		sbp->f_bsize = NFS_FABLKSIZE;
340 		tquad = fxdr_hyper(&sfp->sf_tbytes);
341 		sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
342 		tquad = fxdr_hyper(&sfp->sf_fbytes);
343 		sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
344 		tquad = fxdr_hyper(&sfp->sf_abytes);
345 		sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
346 		sbp->f_files = (fxdr_unsigned(int32_t,
347 		    sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
348 		sbp->f_ffree = (fxdr_unsigned(int32_t,
349 		    sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
350 	} else {
351 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
352 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
353 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
354 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
355 		sbp->f_files = 0;
356 		sbp->f_ffree = 0;
357 	}
358 	if (sbp != &mp->mnt_stat) {
359 		sbp->f_type = mp->mnt_vfc->vfc_typenum;
360 		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
361 	}
362 	m_freem(mrep);
363 nfsmout:
364 	vput(vp);
365 	crfree(cred);
366 	return (error);
367 }
368 
369 /*
370  * nfs version 3 fsinfo rpc call
371  */
372 int
373 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td)
374 {
375 	struct nfsv3_fsinfo *fsp;
376 	caddr_t cp;
377 	int32_t t1, t2;
378 	u_int32_t *tl, pref, max;
379 	caddr_t bpos, dpos, cp2;
380 	int error = 0, retattr;
381 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
382 	u_int64_t maxfsize;
383 
384 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
385 	nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
386 	nfsm_fhtom(vp, 1);
387 	nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ));
388 	nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK);
389 	if (!error) {
390 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
391 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
392 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
393 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
394 				~(NFS_FABLKSIZE - 1);
395 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
396 		if (max < nmp->nm_wsize && max > 0) {
397 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
398 			if (nmp->nm_wsize == 0)
399 				nmp->nm_wsize = max;
400 		}
401 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
402 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
403 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
404 				~(NFS_FABLKSIZE - 1);
405 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
406 		if (max < nmp->nm_rsize && max > 0) {
407 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
408 			if (nmp->nm_rsize == 0)
409 				nmp->nm_rsize = max;
410 		}
411 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
412 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
413 			nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
414 				~(NFS_DIRBLKSIZ - 1);
415 		if (max < nmp->nm_readdirsize && max > 0) {
416 			nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
417 			if (nmp->nm_readdirsize == 0)
418 				nmp->nm_readdirsize = max;
419 		}
420 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
421 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
422 			nmp->nm_maxfilesize = maxfsize;
423 		nmp->nm_state |= NFSSTA_GOTFSINFO;
424 	}
425 	m_freem(mrep);
426 nfsmout:
427 	return (error);
428 }
429 
430 /*
431  * Mount a remote root fs via. nfs. This depends on the info in the
432  * nfs_diskless structure that has been filled in properly by some primary
433  * bootstrap.
434  * It goes something like this:
435  * - do enough of "ifconfig" by calling ifioctl() so that the system
436  *   can talk to the server
437  * - If nfs_diskless.mygateway is filled in, use that address as
438  *   a default gateway.
439  * - build the rootfs mount point and call mountnfs() to do the rest.
440  */
441 int
442 nfs_mountroot(struct mount *mp)
443 {
444 	struct mount  *swap_mp;
445 	struct nfsv3_diskless *nd = &nfsv3_diskless;
446 	struct socket *so;
447 	struct vnode *vp;
448 	struct thread *td = curthread;		/* XXX */
449 	int error, i;
450 	u_long l;
451 	char buf[128];
452 
453 #if defined(BOOTP_NFSROOT) && defined(BOOTP)
454 	bootpc_init();		/* use bootp to get nfs_diskless filled in */
455 #endif
456 
457 	/*
458 	 * XXX time must be non-zero when we init the interface or else
459 	 * the arp code will wedge...
460 	 */
461 	while (mycpu->gd_time_seconds == 0)
462 		tsleep(mycpu, 0, "arpkludge", 10);
463 
464 	/*
465 	 * The boot code may have passed us a diskless structure.
466 	 */
467 	if (nfs_diskless_valid == 1)
468 		nfs_convert_diskless();
469 
470 #define SINP(sockaddr)	((struct sockaddr_in *)(sockaddr))
471 	printf("nfs_mountroot: interface %s ip %s",
472 		nd->myif.ifra_name,
473 		inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr));
474 	printf(" bcast %s",
475 		inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr));
476 	printf(" mask %s\n",
477 		inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr));
478 #undef SINP
479 
480 	/*
481 	 * XXX splnet, so networks will receive...
482 	 */
483 	crit_enter();
484 
485 	/*
486 	 * BOOTP does not necessarily have to be compiled into the kernel
487 	 * for an NFS root to work.  If we inherited the network
488 	 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured
489 	 * out our interface for us and all we need to do is ifconfig the
490 	 * interface.  We only do this if the interface has not already been
491 	 * ifconfig'd by e.g. BOOTP.
492 	 */
493 	error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td);
494 	if (error) {
495 		panic("nfs_mountroot: socreate(%04x): %d",
496 			nd->myif.ifra_addr.sa_family, error);
497 	}
498 
499 	error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred);
500 	if (error)
501 		panic("nfs_mountroot: SIOCAIFADDR: %d", error);
502 
503 	soclose(so, FNONBLOCK);
504 
505 	/*
506 	 * If the gateway field is filled in, set it as the default route.
507 	 */
508 	if (nd->mygateway.sin_len != 0) {
509 		struct sockaddr_in mask, sin;
510 
511 		bzero((caddr_t)&mask, sizeof(mask));
512 		sin = mask;
513 		sin.sin_family = AF_INET;
514 		sin.sin_len = sizeof(sin);
515 		printf("nfs_mountroot: gateway %s\n",
516 			inet_ntoa(nd->mygateway.sin_addr));
517 		error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin,
518 					(struct sockaddr *)&nd->mygateway,
519 					(struct sockaddr *)&mask,
520 					RTF_UP | RTF_GATEWAY);
521 		if (error)
522 			printf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error);
523 	}
524 
525 	/*
526 	 * Create the rootfs mount point.
527 	 */
528 	nd->root_args.fh = nd->root_fh;
529 	nd->root_args.fhsize = nd->root_fhsize;
530 	l = ntohl(nd->root_saddr.sin_addr.s_addr);
531 	snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
532 		(l >> 24) & 0xff, (l >> 16) & 0xff,
533 		(l >>  8) & 0xff, (l >>  0) & 0xff,nd->root_hostnam);
534 	printf("NFS ROOT: %s\n",buf);
535 	if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY,
536 	    &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) {
537 		if (swap_mp) {
538 			mp->mnt_vfc->vfc_refcount--;
539 			free(swap_mp, M_MOUNT);
540 		}
541 		crit_exit();
542 		return (error);
543 	}
544 
545 	swap_mp = NULL;
546 	if (nd->swap_nblks) {
547 
548 		/* Convert to DEV_BSIZE instead of Kilobyte */
549 		nd->swap_nblks *= 2;
550 
551 		/*
552 		 * Create a fake mount point just for the swap vnode so that the
553 		 * swap file can be on a different server from the rootfs.
554 		 */
555 		nd->swap_args.fh = nd->swap_fh;
556 		nd->swap_args.fhsize = nd->swap_fhsize;
557 		l = ntohl(nd->swap_saddr.sin_addr.s_addr);
558 		snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s",
559 			(l >> 24) & 0xff, (l >> 16) & 0xff,
560 			(l >>  8) & 0xff, (l >>  0) & 0xff,nd->swap_hostnam);
561 		printf("NFS SWAP: %s\n",buf);
562 		if ((error = nfs_mountdiskless(buf, "/swap", 0,
563 		    &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) {
564 			crit_exit();
565 			return (error);
566 		}
567 		vfs_unbusy(swap_mp);
568 
569 		VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size =
570 				nd->swap_nblks * DEV_BSIZE ;
571 
572 		/*
573 		 * Since the swap file is not the root dir of a file system,
574 		 * hack it to a regular file.
575 		 */
576 		vp->v_flag = 0;
577 		vref(vp);
578 		nfs_setvtype(vp, VREG);
579 		swaponvp(td, vp, nd->swap_nblks);
580 	}
581 
582 	mp->mnt_flag |= MNT_ROOTFS;
583 	mp->mnt_vnodecovered = NULLVP;
584 	vfs_unbusy(mp);
585 
586 	/*
587 	 * This is not really an nfs issue, but it is much easier to
588 	 * set hostname here and then let the "/etc/rc.xxx" files
589 	 * mount the right /var based upon its preset value.
590 	 */
591 	bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN);
592 	hostname[MAXHOSTNAMELEN - 1] = '\0';
593 	for (i = 0; i < MAXHOSTNAMELEN; i++)
594 		if (hostname[i] == '\0')
595 			break;
596 	inittodr(ntohl(nd->root_time));
597 	crit_exit();
598 	return (0);
599 }
600 
601 /*
602  * Internal version of mount system call for diskless setup.
603  */
604 static int
605 nfs_mountdiskless(char *path, char *which, int mountflag,
606 	struct sockaddr_in *sin, struct nfs_args *args, struct thread *td,
607 	struct vnode **vpp, struct mount **mpp)
608 {
609 	struct mount *mp;
610 	struct sockaddr *nam;
611 	int didalloc = 0;
612 	int error;
613 
614 	mp = *mpp;
615 
616 	if (mp == NULL) {
617 		if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) {
618 			printf("nfs_mountroot: NFS not configured");
619 			return (error);
620 		}
621 		didalloc = 1;
622 	}
623 	mp->mnt_kern_flag = 0;
624 	mp->mnt_flag = mountflag;
625 	nam = dup_sockaddr((struct sockaddr *)sin);
626 
627 #if defined(BOOTP) || defined(NFS_ROOT)
628 	if (args->fhsize == 0) {
629 		printf("NFS_ROOT: No FH passed from loader, attempting mount rpc...");
630 		args->fhsize = 0;
631 		error = md_mount(sin, which, args->fh, &args->fhsize, args, td);
632 		if (error) {
633 			printf("failed.\n");
634 			goto haderror;
635 		}
636 		printf("success!\n");
637 	}
638 #endif
639 
640 	if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) {
641 #if defined(BOOTP) || defined(NFS_ROOT)
642 haderror:
643 #endif
644 		printf("nfs_mountroot: mount %s on %s: %d", path, which, error);
645 		mp->mnt_vfc->vfc_refcount--;
646 		vfs_unbusy(mp);
647 		if (didalloc)
648 			free(mp, M_MOUNT);
649 		FREE(nam, M_SONAME);
650 		return (error);
651 	}
652 	*mpp = mp;
653 	return (0);
654 }
655 
656 static void
657 nfs_decode_args(nmp, argp)
658 	struct nfsmount *nmp;
659 	struct nfs_args *argp;
660 {
661 	int adjsock;
662 	int maxio;
663 
664 	crit_enter();
665 	/*
666 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
667 	 * no sense in that context.
668 	 */
669 	if (argp->sotype == SOCK_STREAM)
670 		nmp->nm_flag &= ~NFSMNT_NOCONN;
671 
672 	/* Also clear RDIRPLUS if not NFSv3, it crashes some servers */
673 	if ((argp->flags & NFSMNT_NFSV3) == 0)
674 		nmp->nm_flag &= ~NFSMNT_RDIRPLUS;
675 
676 	/* Re-bind if rsrvd port requested and wasn't on one */
677 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
678 		  && (argp->flags & NFSMNT_RESVPORT);
679 	/* Also re-bind if we're switching to/from a connected UDP socket */
680 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
681 		    (argp->flags & NFSMNT_NOCONN));
682 
683 	/* Update flags atomically.  Don't change the lock bits. */
684 	nmp->nm_flag = argp->flags | nmp->nm_flag;
685 	crit_exit();
686 
687 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
688 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
689 		if (nmp->nm_timeo < NFS_MINTIMEO)
690 			nmp->nm_timeo = NFS_MINTIMEO;
691 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
692 			nmp->nm_timeo = NFS_MAXTIMEO;
693 	}
694 
695 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
696 		nmp->nm_retry = argp->retrans;
697 		if (nmp->nm_retry > NFS_MAXREXMIT)
698 			nmp->nm_retry = NFS_MAXREXMIT;
699 	}
700 
701 	maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
702 
703 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
704 		nmp->nm_wsize = argp->wsize;
705 		/* Round down to multiple of blocksize */
706 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
707 		if (nmp->nm_wsize <= 0)
708 			nmp->nm_wsize = NFS_FABLKSIZE;
709 	}
710 	if (nmp->nm_wsize > maxio)
711 		nmp->nm_wsize = maxio;
712 	if (nmp->nm_wsize > MAXBSIZE)
713 		nmp->nm_wsize = MAXBSIZE;
714 
715 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
716 		nmp->nm_rsize = argp->rsize;
717 		/* Round down to multiple of blocksize */
718 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
719 		if (nmp->nm_rsize <= 0)
720 			nmp->nm_rsize = NFS_FABLKSIZE;
721 	}
722 	if (nmp->nm_rsize > maxio)
723 		nmp->nm_rsize = maxio;
724 	if (nmp->nm_rsize > MAXBSIZE)
725 		nmp->nm_rsize = MAXBSIZE;
726 
727 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
728 		nmp->nm_readdirsize = argp->readdirsize;
729 	}
730 	if (nmp->nm_readdirsize > maxio)
731 		nmp->nm_readdirsize = maxio;
732 	if (nmp->nm_readdirsize > nmp->nm_rsize)
733 		nmp->nm_readdirsize = nmp->nm_rsize;
734 
735 	if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0)
736 		nmp->nm_acregmin = argp->acregmin;
737 	else
738 		nmp->nm_acregmin = NFS_MINATTRTIMO;
739 	if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0)
740 		nmp->nm_acregmax = argp->acregmax;
741 	else
742 		nmp->nm_acregmax = NFS_MAXATTRTIMO;
743 	if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0)
744 		nmp->nm_acdirmin = argp->acdirmin;
745 	else
746 		nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
747 	if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0)
748 		nmp->nm_acdirmax = argp->acdirmax;
749 	else
750 		nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
751 	if (nmp->nm_acdirmin > nmp->nm_acdirmax)
752 		nmp->nm_acdirmin = nmp->nm_acdirmax;
753 	if (nmp->nm_acregmin > nmp->nm_acregmax)
754 		nmp->nm_acregmin = nmp->nm_acregmax;
755 
756 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) {
757 		if (argp->maxgrouplist <= NFS_MAXGRPS)
758 			nmp->nm_numgrps = argp->maxgrouplist;
759 		else
760 			nmp->nm_numgrps = NFS_MAXGRPS;
761 	}
762 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) {
763 		if (argp->readahead <= NFS_MAXRAHEAD)
764 			nmp->nm_readahead = argp->readahead;
765 		else
766 			nmp->nm_readahead = NFS_MAXRAHEAD;
767 	}
768 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) {
769 		if (argp->deadthresh <= NFS_NEVERDEAD)
770 			nmp->nm_deadthresh = argp->deadthresh;
771 		else
772 			nmp->nm_deadthresh = NFS_NEVERDEAD;
773 	}
774 
775 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
776 		    (nmp->nm_soproto != argp->proto));
777 	nmp->nm_sotype = argp->sotype;
778 	nmp->nm_soproto = argp->proto;
779 
780 	if (nmp->nm_so && adjsock) {
781 		nfs_safedisconnect(nmp);
782 		if (nmp->nm_sotype == SOCK_DGRAM)
783 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
784 				printf("nfs_args: retrying connect\n");
785 				(void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0);
786 			}
787 	}
788 }
789 
790 /*
791  * VFS Operations.
792  *
793  * mount system call
794  * It seems a bit dumb to copyinstr() the host and path here and then
795  * bcopy() them in mountnfs(), but I wanted to detect errors before
796  * doing the sockargs() call because sockargs() allocates an mbuf and
797  * an error after that means that I have to release the mbuf.
798  */
799 /* ARGSUSED */
800 static int
801 nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
802 {
803 	int error;
804 	struct nfs_args args;
805 	struct sockaddr *nam;
806 	struct vnode *vp;
807 	char pth[MNAMELEN], hst[MNAMELEN];
808 	size_t len;
809 	u_char nfh[NFSX_V3FHMAX];
810 
811 	if (path == NULL) {
812 		nfs_mountroot(mp);
813 		return (0);
814 	}
815 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
816 	if (error)
817 		return (error);
818 	if (args.version != NFS_ARGSVERSION) {
819 #ifdef COMPAT_PRELITE2
820 		/*
821 		 * If the argument version is unknown, then assume the
822 		 * caller is a pre-lite2 4.4BSD client and convert its
823 		 * arguments.
824 		 */
825 		struct onfs_args oargs;
826 		error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args));
827 		if (error)
828 			return (error);
829 		nfs_convert_oargs(&args,&oargs);
830 #else /* !COMPAT_PRELITE2 */
831 		return (EPROGMISMATCH);
832 #endif /* COMPAT_PRELITE2 */
833 	}
834 	if (mp->mnt_flag & MNT_UPDATE) {
835 		struct nfsmount *nmp = VFSTONFS(mp);
836 
837 		if (nmp == NULL)
838 			return (EIO);
839 		/*
840 		 * When doing an update, we can't change from or to
841 		 * v3, or change cookie translation
842 		 */
843 		args.flags = (args.flags &
844 		    ~(NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/)) |
845 		    (nmp->nm_flag &
846 			(NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/));
847 		nfs_decode_args(nmp, &args);
848 		return (0);
849 	}
850 
851 	/*
852 	 * Make the nfs_ip_paranoia sysctl serve as the default connection
853 	 * or no-connection mode for those protocols that support
854 	 * no-connection mode (the flag will be cleared later for protocols
855 	 * that do not support no-connection mode).  This will allow a client
856 	 * to receive replies from a different IP then the request was
857 	 * sent to.  Note: default value for nfs_ip_paranoia is 1 (paranoid),
858 	 * not 0.
859 	 */
860 	if (nfs_ip_paranoia == 0)
861 		args.flags |= NFSMNT_NOCONN;
862 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
863 		return (EINVAL);
864 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
865 	if (error)
866 		return (error);
867 	error = copyinstr(path, pth, MNAMELEN-1, &len);
868 	if (error)
869 		return (error);
870 	bzero(&pth[len], MNAMELEN - len);
871 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
872 	if (error)
873 		return (error);
874 	bzero(&hst[len], MNAMELEN - len);
875 	/* sockargs() call must be after above copyin() calls */
876 	error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen);
877 	if (error)
878 		return (error);
879 	args.fh = nfh;
880 	error = mountnfs(&args, mp, nam, pth, hst, &vp);
881 	return (error);
882 }
883 
884 /*
885  * Common code for mount and mountroot
886  */
887 static int
888 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam,
889 	char *pth, char *hst, struct vnode **vpp)
890 {
891 	struct nfsmount *nmp;
892 	struct nfsnode *np;
893 	int error;
894 
895 	if (mp->mnt_flag & MNT_UPDATE) {
896 		nmp = VFSTONFS(mp);
897 		/* update paths, file handles, etc, here	XXX */
898 		FREE(nam, M_SONAME);
899 		return (0);
900 	} else {
901 		nmp = zalloc(nfsmount_zone);
902 		bzero((caddr_t)nmp, sizeof (struct nfsmount));
903 		TAILQ_INIT(&nmp->nm_uidlruhead);
904 		TAILQ_INIT(&nmp->nm_bioq);
905 		mp->mnt_data = (qaddr_t)nmp;
906 	}
907 	vfs_getnewfsid(mp);
908 	nmp->nm_mountp = mp;
909 
910 	/*
911 	 * V2 can only handle 32 bit filesizes.  A 4GB-1 limit may be too
912 	 * high, depending on whether we end up with negative offsets in
913 	 * the client or server somewhere.  2GB-1 may be safer.
914 	 *
915 	 * For V3, nfs_fsinfo will adjust this as necessary.  Assume maximum
916 	 * that we can handle until we find out otherwise.
917 	 * XXX Our "safe" limit on the client is what we can store in our
918 	 * buffer cache using signed(!) block numbers.
919 	 */
920 	if ((argp->flags & NFSMNT_NFSV3) == 0)
921 		nmp->nm_maxfilesize = 0xffffffffLL;
922 	else
923 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
924 
925 	nmp->nm_timeo = NFS_TIMEO;
926 	nmp->nm_retry = NFS_RETRANS;
927 	nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype);
928 	nmp->nm_rsize = nmp->nm_wsize;
929 	nmp->nm_readdirsize = NFS_READDIRSIZE;
930 	nmp->nm_numgrps = NFS_MAXGRPS;
931 	nmp->nm_readahead = NFS_DEFRAHEAD;
932 	nmp->nm_deadthresh = NFS_DEADTHRESH;
933 	nmp->nm_fhsize = argp->fhsize;
934 	bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
935 	bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
936 	nmp->nm_nam = nam;
937 	/* Set up the sockets and per-host congestion */
938 	nmp->nm_sotype = argp->sotype;
939 	nmp->nm_soproto = argp->proto;
940 	nmp->nm_cred = crhold(proc0.p_ucred);
941 
942 	nfs_decode_args(nmp, argp);
943 
944 	/*
945 	 * For Connection based sockets (TCP,...) defer the connect until
946 	 * the first request, in case the server is not responding.
947 	 */
948 	if (nmp->nm_sotype == SOCK_DGRAM &&
949 		(error = nfs_connect(nmp, (struct nfsreq *)0)))
950 		goto bad;
951 
952 	/*
953 	 * This is silly, but it has to be set so that vinifod() works.
954 	 * We do not want to do an nfs_statfs() here since we can get
955 	 * stuck on a dead server and we are holding a lock on the mount
956 	 * point.
957 	 */
958 	mp->mnt_stat.f_iosize =
959 		nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype);
960 
961 	/*
962 	 * Install vop_ops for our vnops
963 	 */
964 	vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops);
965 	vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops);
966 	vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops);
967 
968 	/*
969 	 * A reference count is needed on the nfsnode representing the
970 	 * remote root.  If this object is not persistent, then backward
971 	 * traversals of the mount point (i.e. "..") will not work if
972 	 * the nfsnode gets flushed out of the cache. Ufs does not have
973 	 * this problem, because one can identify root inodes by their
974 	 * number == ROOTINO (2).
975 	 */
976 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
977 	if (error)
978 		goto bad;
979 	*vpp = NFSTOV(np);
980 
981 	/*
982 	 * Retrieval of mountpoint attributes is delayed until nfs_rot
983 	 * or nfs_statfs are first called.  This will happen either when
984 	 * we first traverse the mount point or if somebody does a df(1).
985 	 *
986 	 * NFSSTA_GOTFSINFO is used to flag if we have successfully
987 	 * retrieved mountpoint attributes.  In the case of NFSv3 we
988 	 * also flag static fsinfo.
989 	 */
990 	if (*vpp != NULL)
991 		(*vpp)->v_type = VNON;
992 
993 	/*
994 	 * Lose the lock but keep the ref.
995 	 */
996 	VOP_UNLOCK(*vpp, 0);
997 
998 	return (0);
999 bad:
1000 	nfs_disconnect(nmp);
1001 	nfs_free_mount(nmp);
1002 	FREE(nam, M_SONAME);
1003 	return (error);
1004 }
1005 
1006 /*
1007  * unmount system call
1008  */
1009 static int
1010 nfs_unmount(struct mount *mp, int mntflags)
1011 {
1012 	struct nfsmount *nmp;
1013 	int error, flags = 0;
1014 
1015 	if (mntflags & MNT_FORCE)
1016 		flags |= FORCECLOSE;
1017 	nmp = VFSTONFS(mp);
1018 	/*
1019 	 * Goes something like this..
1020 	 * - Call vflush() to clear out vnodes for this file system
1021 	 * - Close the socket
1022 	 * - Free up the data structures
1023 	 */
1024 	/* In the forced case, cancel any outstanding requests. */
1025 	if (flags & FORCECLOSE) {
1026 		error = nfs_nmcancelreqs(nmp);
1027 		if (error)
1028 			return (error);
1029 	}
1030 	/*
1031 	 * Must handshake with nfs_clientd() if it is active. XXX
1032 	 */
1033 	nmp->nm_state |= NFSSTA_DISMINPROG;
1034 
1035 	/* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */
1036 	error = vflush(mp, 1, flags);
1037 	if (error) {
1038 		nmp->nm_state &= ~NFSSTA_DISMINPROG;
1039 		return (error);
1040 	}
1041 
1042 	/*
1043 	 * We are now committed to the unmount.
1044 	 * For NQNFS, let the server daemon free the nfsmount structure.
1045 	 */
1046 	if (nmp->nm_flag & NFSMNT_KERB)
1047 		nmp->nm_state |= NFSSTA_DISMNT;
1048 
1049 	nfs_disconnect(nmp);
1050 	FREE(nmp->nm_nam, M_SONAME);
1051 
1052 	if ((nmp->nm_flag & NFSMNT_KERB) == 0)
1053 		nfs_free_mount(nmp);
1054 	return (0);
1055 }
1056 
1057 void
1058 nfs_free_mount(struct nfsmount *nmp)
1059 {
1060 	if (nmp->nm_cred)  {
1061 		crfree(nmp->nm_cred);
1062 		nmp->nm_cred = NULL;
1063 	}
1064 	zfree(nfsmount_zone, nmp);
1065 }
1066 
1067 /*
1068  * Return root of a filesystem
1069  */
1070 static int
1071 nfs_root(mp, vpp)
1072 	struct mount *mp;
1073 	struct vnode **vpp;
1074 {
1075 	struct vnode *vp;
1076 	struct nfsmount *nmp;
1077 	struct vattr attrs;
1078 	struct nfsnode *np;
1079 	int error;
1080 
1081 	nmp = VFSTONFS(mp);
1082 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
1083 	if (error)
1084 		return (error);
1085 	vp = NFSTOV(np);
1086 
1087 	/*
1088 	 * Get transfer parameters and root vnode attributes
1089 	 */
1090 	if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) {
1091 	    if (nmp->nm_flag & NFSMNT_NFSV3) {
1092 		nfs_fsinfo(nmp, vp, curthread);
1093 		mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype);
1094 	    } else {
1095 		if ((error = VOP_GETATTR(vp, &attrs)) == 0)
1096 			nmp->nm_state |= NFSSTA_GOTFSINFO;
1097 
1098 	    }
1099 	}
1100 	if (vp->v_type == VNON)
1101 	    nfs_setvtype(vp, VDIR);
1102 	vp->v_flag = VROOT;
1103 	*vpp = vp;
1104 	return (0);
1105 }
1106 
1107 extern int syncprt;
1108 
1109 struct scaninfo {
1110 	int rescan;
1111 	int waitfor;
1112 	int allerror;
1113 };
1114 
1115 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1116 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1117 
1118 /*
1119  * Flush out the buffer cache
1120  */
1121 /* ARGSUSED */
1122 static int
1123 nfs_sync(struct mount *mp, int waitfor)
1124 {
1125 	struct scaninfo scaninfo;
1126 	int error;
1127 
1128 	scaninfo.rescan = 0;
1129 	scaninfo.waitfor = waitfor;
1130 	scaninfo.allerror = 0;
1131 
1132 	/*
1133 	 * Force stale buffer cache information to be flushed.
1134 	 */
1135 	error = 0;
1136 	while (error == 0 && scaninfo.rescan) {
1137 		scaninfo.rescan = 0;
1138 		error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1,
1139 					nfs_sync_scan2, &scaninfo);
1140 	}
1141 	return(error);
1142 }
1143 
1144 static
1145 int
1146 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1147 {
1148     struct scaninfo *info = data;
1149 
1150     if (VOP_ISLOCKED(vp, NULL) || RB_EMPTY(&vp->v_rbdirty_tree))
1151 	return(-1);
1152     if (info->waitfor == MNT_LAZY)
1153 	return(-1);
1154     return(0);
1155 }
1156 
1157 static
1158 int
1159 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1160 {
1161     struct scaninfo *info = data;
1162     int error;
1163 
1164     error = VOP_FSYNC(vp, info->waitfor);
1165     if (error)
1166 	info->allerror = error;
1167     return(0);
1168 }
1169 
1170