xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision d0fed6c87ddc40a8bffa6f99e7433ddfc864dd83)
1 /*	$NetBSD: nfs_vfsops.c,v 1.58 1997/02/22 02:48:26 fvdl Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_vfsops.c	8.12 (Berkeley) 5/20/95
39  */
40 
41 #include <sys/param.h>
42 #include <sys/conf.h>
43 #include <sys/ioctl.h>
44 #include <sys/signal.h>
45 #include <sys/proc.h>
46 #include <sys/namei.h>
47 #include <sys/device.h>
48 #include <sys/vnode.h>
49 #include <sys/kernel.h>
50 #include <sys/mount.h>
51 #include <sys/buf.h>
52 #include <sys/mbuf.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/systm.h>
56 
57 #include <net/if.h>
58 #include <net/route.h>
59 #include <netinet/in.h>
60 
61 #include <nfs/rpcv2.h>
62 #include <nfs/nfsproto.h>
63 #include <nfs/nfsnode.h>
64 #include <nfs/nfs.h>
65 #include <nfs/nfsmount.h>
66 #include <nfs/xdr_subs.h>
67 #include <nfs/nfsm_subs.h>
68 #include <nfs/nfsdiskless.h>
69 #include <nfs/nqnfs.h>
70 #include <nfs/nfs_var.h>
71 
72 extern struct nfsstats nfsstats;
73 extern int nfs_ticks;
74 
75 #ifdef notyet
76 static int nfs_sysctl(int *, u_int, void *, size_t *, void *, size_t,
77 		      struct proc *);
78 #endif
79 
80 /*
81  * nfs vfs operations.
82  */
83 struct vfsops nfs_vfsops = {
84 	MOUNT_NFS,
85 	nfs_mount,
86 	nfs_start,
87 	nfs_unmount,
88 	nfs_root,
89 	nfs_quotactl,
90 	nfs_statfs,
91 	nfs_sync,
92 	nfs_vget,
93 	nfs_fhtovp,
94 	nfs_vptofh,
95 	nfs_vfs_init,
96 	nfs_mountroot,
97 #ifdef notyet
98 	nfs_sysctl
99 #endif
100 };
101 
102 extern u_int32_t nfs_procids[NFS_NPROCS];
103 extern u_int32_t nfs_prog, nfs_vers;
104 
105 static struct mount *
106 nfs_mount_diskless __P((struct nfs_dlmount *, const char *, int,
107     struct vnode **));
108 
109 #define TRUE	1
110 #define	FALSE	0
111 
112 /*
113  * nfs statfs call
114  */
115 int
116 nfs_statfs(mp, sbp, p)
117 	struct mount *mp;
118 	register struct statfs *sbp;
119 	struct proc *p;
120 {
121 	register struct vnode *vp;
122 	register struct nfs_statfs *sfp;
123 	register caddr_t cp;
124 	register u_int32_t *tl;
125 	register int32_t t1, t2;
126 	caddr_t bpos, dpos, cp2;
127 	struct nfsmount *nmp = VFSTONFS(mp);
128 	int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
129 	struct mbuf *mreq, *mrep = NULL, *md, *mb, *mb2;
130 	struct ucred *cred;
131 	struct nfsnode *np;
132 	u_quad_t tquad;
133 
134 #ifndef nolint
135 	sfp = (struct nfs_statfs *)0;
136 #endif
137 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
138 	if (error)
139 		return (error);
140 	vp = NFSTOV(np);
141 	cred = crget();
142 	cred->cr_ngroups = 0;
143 	if (v3 && (nmp->nm_flag & NFSMNT_GOTFSINFO) == 0)
144 		(void)nfs_fsinfo(nmp, vp, cred, p);
145 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
146 	nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
147 	nfsm_fhtom(vp, v3);
148 	nfsm_request(vp, NFSPROC_FSSTAT, p, cred);
149 	if (v3)
150 		nfsm_postop_attr(vp, retattr);
151 	if (error) {
152 		if (mrep != NULL)
153 			m_free(mrep);
154 		goto nfsmout;
155 	}
156 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
157 #ifdef COMPAT_09
158 	sbp->f_type = 2;
159 #else
160 	sbp->f_type = 0;
161 #endif
162 	sbp->f_flags = nmp->nm_flag;
163 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
164 	if (v3) {
165 		sbp->f_bsize = NFS_FABLKSIZE;
166 		fxdr_hyper(&sfp->sf_tbytes, &tquad);
167 		sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
168 		fxdr_hyper(&sfp->sf_fbytes, &tquad);
169 		sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
170 		fxdr_hyper(&sfp->sf_abytes, &tquad);
171 		sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
172 		sbp->f_files = (fxdr_unsigned(int32_t,
173 		    sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
174 		sbp->f_ffree = (fxdr_unsigned(int32_t,
175 		    sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
176 	} else {
177 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
178 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
179 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
180 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
181 		sbp->f_files = 0;
182 		sbp->f_ffree = 0;
183 	}
184 	if (sbp != &mp->mnt_stat) {
185 		bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
186 		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
187 	}
188 	strncpy(&sbp->f_fstypename[0], mp->mnt_op->vfs_name, MFSNAMELEN);
189 	nfsm_reqdone;
190 	vrele(vp);
191 	crfree(cred);
192 	return (error);
193 }
194 
195 /*
196  * nfs version 3 fsinfo rpc call
197  */
198 int
199 nfs_fsinfo(nmp, vp, cred, p)
200 	register struct nfsmount *nmp;
201 	register struct vnode *vp;
202 	struct ucred *cred;
203 	struct proc *p;
204 {
205 	register struct nfsv3_fsinfo *fsp;
206 	register caddr_t cp;
207 	register int32_t t1, t2;
208 	register u_int32_t *tl, pref, max;
209 	caddr_t bpos, dpos, cp2;
210 	int error = 0, retattr;
211 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
212 
213 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
214 	nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
215 	nfsm_fhtom(vp, 1);
216 	nfsm_request(vp, NFSPROC_FSINFO, p, cred);
217 	nfsm_postop_attr(vp, retattr);
218 	if (!error) {
219 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
220 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
221 		if (pref < nmp->nm_wsize)
222 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
223 				~(NFS_FABLKSIZE - 1);
224 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
225 		if (max < nmp->nm_wsize) {
226 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
227 			if (nmp->nm_wsize == 0)
228 				nmp->nm_wsize = max;
229 		}
230 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
231 		if (pref < nmp->nm_rsize)
232 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
233 				~(NFS_FABLKSIZE - 1);
234 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
235 		if (max < nmp->nm_rsize) {
236 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
237 			if (nmp->nm_rsize == 0)
238 				nmp->nm_rsize = max;
239 		}
240 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
241 		if (pref < nmp->nm_readdirsize)
242 			nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
243 				~(NFS_DIRBLKSIZ - 1);
244 		if (max < nmp->nm_readdirsize) {
245 			nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
246 			if (nmp->nm_readdirsize == 0)
247 				nmp->nm_readdirsize = max;
248 		}
249 		nmp->nm_flag |= NFSMNT_GOTFSINFO;
250 	}
251 	nfsm_reqdone;
252 	return (error);
253 }
254 
255 /*
256  * Mount a remote root fs via. NFS.  It goes like this:
257  * - Call nfs_boot_init() to fill in the nfs_diskless struct
258  *   (using RARP, bootparam RPC, mountd RPC)
259  * - hand craft the swap nfs vnode hanging off a fake mount point
260  *	if swdevt[0].sw_dev == NODEV
261  * - build the rootfs mount point and call mountnfs() to do the rest.
262  */
263 int
264 nfs_mountroot()
265 {
266 	struct nfs_diskless nd;
267 	struct vattr attr;
268 	struct mount *mp;
269 	struct vnode *vp;
270 	struct proc *procp;
271 	long n;
272 	int error;
273 
274 	procp = curproc; /* XXX */
275 
276 	if (root_device->dv_class != DV_IFNET)
277 		return (ENODEV);
278 
279 	/*
280 	 * XXX time must be non-zero when we init the interface or else
281 	 * the arp code will wedge.  [Fixed now in if_ether.c]
282 	 * However, the NFS attribute cache gives false "hits" when
283 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
284 	 */
285 	if (time.tv_sec < NFS_MAXATTRTIMO)
286 		time.tv_sec = NFS_MAXATTRTIMO;
287 
288 	/*
289 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
290 	 * Side effect:  Finds and configures a network interface.
291 	 */
292 	bzero((caddr_t) &nd, sizeof(nd));
293 	error = nfs_boot_init(&nd, procp);
294 	if (error)
295 		return (error);
296 
297 	/*
298 	 * Create the root mount point.
299 	 */
300 	nfs_boot_getfh(&nd.nd_boot, "root", &nd.nd_root);
301 	mp = nfs_mount_diskless(&nd.nd_root, "/", 0, &vp);
302 	printf("root on %s\n", nd.nd_root.ndm_host);
303 
304 	/*
305 	 * Link it into the mount list.
306 	 */
307 #ifdef Lite2_integrated
308 	simple_lock(&mountlist_slock);
309 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
310 	simple_unlock(&mountlist_slock);
311 	rootvp = vp;
312 	vfs_unbusy(mp, procp);
313 #else
314 	if (vfs_lock(mp))
315 		panic("nfs_mountroot: vfs_lock");
316 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
317 	mp->mnt_vnodecovered = NULLVP;
318 	vfs_unlock(mp);
319 	rootvp = vp;
320 #endif
321 
322 	/* Get root attributes (for the time). */
323 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
324 	if (error) panic("nfs_mountroot: getattr for root");
325 	n = attr.va_mtime.tv_sec;
326 #ifdef	DEBUG
327 	printf("root time: 0x%lx\n", n);
328 #endif
329 	inittodr(n);
330 
331 	/*
332 	 * XXX splnet, so networks will receive...
333 	 */
334 	splnet();
335 
336 #ifdef notyet
337 	/* Set up swap credentials. */
338 	proc0.p_ucred->cr_uid = ntohl(nd.swap_ucred.cr_uid);
339 	proc0.p_ucred->cr_gid = ntohl(nd.swap_ucred.cr_gid);
340 	if ((proc0.p_ucred->cr_ngroups = ntohs(nd.swap_ucred.cr_ngroups)) >
341 		NGROUPS)
342 		proc0.p_ucred->cr_ngroups = NGROUPS;
343 	for (i = 0; i < proc0.p_ucred->cr_ngroups; i++)
344 	    proc0.p_ucred->cr_groups[i] = ntohl(nd.swap_ucred.cr_groups[i]);
345 #endif
346 
347 	/*
348 	 * "Mount" the swap device.
349 	 *
350 	 * On a "dataless" configuration (swap on disk) we will have:
351 	 *	(swdevt[0].sw_dev != NODEV) identifying the swap device.
352 	 */
353 	if (bdevvp(swapdev, &swapdev_vp))
354 		panic("nfs_mountroot: can't setup swap vp");
355 	if (swdevt[0].sw_dev != NODEV) {
356 		printf("swap on device 0x%x\n", swdevt[0].sw_dev);
357 		return (0);
358 	}
359 
360 	/*
361 	 * If swapping to an nfs node:  (swdevt[0].sw_dev == NODEV)
362 	 * Create a fake mount point just for the swap vnode so that the
363 	 * swap file can be on a different server from the rootfs.
364 	 */
365 	nfs_boot_getfh(&nd.nd_boot, "swap", &nd.nd_swap);
366 	mp = nfs_mount_diskless(&nd.nd_swap, "/swap", 0, &vp);
367 #ifdef Lite2_integrated
368 	vfs_unbusy(mp, procp);
369 #endif
370 	printf("swap on %s\n", nd.nd_swap.ndm_host);
371 
372 	/*
373 	 * Since the swap file is not the root dir of a file system,
374 	 * hack it to a regular file.
375 	 */
376 	vp->v_type = VREG;
377 	vp->v_flag = 0;
378 	swdevt[0].sw_vp = vp;
379 
380 	/*
381 	 * Find out how large the swap file is.
382 	 */
383 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
384 	if (error)
385 		panic("nfs_mountroot: getattr for swap");
386 	n = (long) (attr.va_size >> DEV_BSHIFT);
387 #ifdef	DEBUG
388 	printf("swap size: 0x%lx (blocks)\n", n);
389 #endif
390 	swdevt[0].sw_nblks = n;
391 
392 	return (0);
393 }
394 
395 /*
396  * Internal version of mount system call for diskless setup.
397  */
398 static struct mount *
399 nfs_mount_diskless(ndmntp, mntname, mntflag, vpp)
400 	struct nfs_dlmount *ndmntp;
401 	const char *mntname;
402 	int mntflag;
403 	struct vnode **vpp;
404 {
405 	struct mount *mp;
406 	struct mbuf *m;
407 	int error;
408 
409 #ifdef Lite2_integrated
410 	vfs_rootmountalloc("nfs", mntname, &mp);
411 #else
412 	/* Create the mount point. */
413 	mp = (struct mount *)malloc((u_long)sizeof(struct mount),
414 	    M_MOUNT, M_WAITOK);
415 	bzero((char *)mp, (u_long)sizeof(struct mount));
416 #endif
417 	mp->mnt_op = &nfs_vfsops;
418 	mp->mnt_flag = mntflag;
419 
420 	/* Get mbuf for server sockaddr. */
421 	m = m_get(M_WAIT, MT_SONAME);
422 	if (m == NULL)
423 		panic("nfs_mountroot: mget soname for %s", mntname);
424 	bcopy((caddr_t)ndmntp->ndm_args.addr, mtod(m, caddr_t),
425 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
426 
427 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
428 			 ndmntp->ndm_args.hostname, vpp);
429 	if (error)
430 		panic("nfs_mountroot: mount %s failed: %d", mntname, error);
431 
432 	return (mp);
433 }
434 
435 void
436 nfs_decode_args(nmp, argp)
437 	struct nfsmount *nmp;
438 	struct nfs_args *argp;
439 {
440 	int s;
441 	int adjsock;
442 	int maxio;
443 
444 	s = splsoftnet();
445 
446 	/*
447 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
448 	 * no sense in that context.
449 	 */
450 	if (argp->sotype == SOCK_STREAM)
451 		argp->flags &= ~NFSMNT_NOCONN;
452 
453 	/* Re-bind if rsrvd port requested and wasn't on one */
454 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
455 		  && (argp->flags & NFSMNT_RESVPORT);
456 	/* Also re-bind if we're switching to/from a connected UDP socket */
457 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
458 		    (argp->flags & NFSMNT_NOCONN));
459 
460 	/* Update flags atomically.  Don't change the lock bits. */
461 	nmp->nm_flag =
462 	    (argp->flags & ~NFSMNT_INTERNAL) | (nmp->nm_flag & NFSMNT_INTERNAL);
463 	splx(s);
464 
465 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
466 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
467 		if (nmp->nm_timeo < NFS_MINTIMEO)
468 			nmp->nm_timeo = NFS_MINTIMEO;
469 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
470 			nmp->nm_timeo = NFS_MAXTIMEO;
471 	}
472 
473 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
474 		nmp->nm_retry = argp->retrans;
475 		if (nmp->nm_retry > NFS_MAXREXMIT)
476 			nmp->nm_retry = NFS_MAXREXMIT;
477 	}
478 
479 	if (argp->flags & NFSMNT_NFSV3) {
480 		if (argp->sotype == SOCK_DGRAM)
481 			maxio = NFS_MAXDGRAMDATA;
482 		else
483 			maxio = NFS_MAXDATA;
484 	} else
485 		maxio = NFS_V2MAXDATA;
486 
487 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
488 		int osize = nmp->nm_wsize;
489 		nmp->nm_wsize = argp->wsize;
490 		/* Round down to multiple of blocksize */
491 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
492 		if (nmp->nm_wsize <= 0)
493 			nmp->nm_wsize = NFS_FABLKSIZE;
494 		adjsock |= (nmp->nm_wsize != osize);
495 	}
496 	if (nmp->nm_wsize > maxio)
497 		nmp->nm_wsize = maxio;
498 	if (nmp->nm_wsize > MAXBSIZE)
499 		nmp->nm_wsize = MAXBSIZE;
500 
501 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
502 		int osize = nmp->nm_rsize;
503 		nmp->nm_rsize = argp->rsize;
504 		/* Round down to multiple of blocksize */
505 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
506 		if (nmp->nm_rsize <= 0)
507 			nmp->nm_rsize = NFS_FABLKSIZE;
508 		adjsock |= (nmp->nm_rsize != osize);
509 	}
510 	if (nmp->nm_rsize > maxio)
511 		nmp->nm_rsize = maxio;
512 	if (nmp->nm_rsize > MAXBSIZE)
513 		nmp->nm_rsize = MAXBSIZE;
514 
515 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
516 		nmp->nm_readdirsize = argp->readdirsize;
517 		/* Round down to multiple of blocksize */
518 		nmp->nm_readdirsize &= ~(NFS_DIRBLKSIZ - 1);
519 		if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
520 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
521 	} else if (argp->flags & NFSMNT_RSIZE)
522 		nmp->nm_readdirsize = nmp->nm_rsize;
523 
524 	if (nmp->nm_readdirsize > maxio)
525 		nmp->nm_readdirsize = maxio;
526 
527 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
528 		argp->maxgrouplist <= NFS_MAXGRPS)
529 		nmp->nm_numgrps = argp->maxgrouplist;
530 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
531 		argp->readahead <= NFS_MAXRAHEAD)
532 		nmp->nm_readahead = argp->readahead;
533 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
534 		argp->leaseterm <= NQ_MAXLEASE)
535 		nmp->nm_leaseterm = argp->leaseterm;
536 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
537 		argp->deadthresh <= NQ_NEVERDEAD)
538 		nmp->nm_deadthresh = argp->deadthresh;
539 
540 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
541 		    (nmp->nm_soproto != argp->proto));
542 	nmp->nm_sotype = argp->sotype;
543 	nmp->nm_soproto = argp->proto;
544 
545 	if (nmp->nm_so && adjsock) {
546 		nfs_disconnect(nmp);
547 		if (nmp->nm_sotype == SOCK_DGRAM)
548 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
549 				printf("nfs_args: retrying connect\n");
550 				(void) tsleep((caddr_t)&lbolt,
551 					      PSOCK, "nfscon", 0);
552 			}
553 	}
554 }
555 
556 /*
557  * VFS Operations.
558  *
559  * mount system call
560  * It seems a bit dumb to copyinstr() the host and path here and then
561  * bcopy() them in mountnfs(), but I wanted to detect errors before
562  * doing the sockargs() call because sockargs() allocates an mbuf and
563  * an error after that means that I have to release the mbuf.
564  */
565 /* ARGSUSED */
566 int
567 nfs_mount(mp, path, data, ndp, p)
568 	struct mount *mp;
569 	const char *path;
570 	void *data;
571 	struct nameidata *ndp;
572 	struct proc *p;
573 {
574 	int error;
575 	struct nfs_args args;
576 	struct mbuf *nam;
577 	struct vnode *vp;
578 	char pth[MNAMELEN], hst[MNAMELEN];
579 	size_t len;
580 	u_char nfh[NFSX_V3FHMAX];
581 
582 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
583 	if (error)
584 		return (error);
585 	if (args.version != NFS_ARGSVERSION)
586 		return (EPROGMISMATCH);
587 	if (mp->mnt_flag & MNT_UPDATE) {
588 		register struct nfsmount *nmp = VFSTONFS(mp);
589 
590 		if (nmp == NULL)
591 			return (EIO);
592 		/*
593 		 * When doing an update, we can't change from or to
594 		 * v3 and/or nqnfs.
595 		 */
596 		args.flags = (args.flags & ~(NFSMNT_NFSV3|NFSMNT_NQNFS)) |
597 		    (nmp->nm_flag & (NFSMNT_NFSV3|NFSMNT_NQNFS));
598 		nfs_decode_args(nmp, &args);
599 		return (0);
600 	}
601 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
602 	if (error)
603 		return (error);
604 	error = copyinstr(path, pth, MNAMELEN-1, &len);
605 	if (error)
606 		return (error);
607 	bzero(&pth[len], MNAMELEN - len);
608 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
609 	if (error)
610 		return (error);
611 	bzero(&hst[len], MNAMELEN - len);
612 	/* sockargs() call must be after above copyin() calls */
613 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
614 	if (error)
615 		return (error);
616 	args.fh = nfh;
617 	error = mountnfs(&args, mp, nam, pth, hst, &vp);
618 	return (error);
619 }
620 
621 /*
622  * Common code for mount and mountroot
623  */
624 int
625 mountnfs(argp, mp, nam, pth, hst, vpp)
626 	register struct nfs_args *argp;
627 	register struct mount *mp;
628 	struct mbuf *nam;
629 	const char *pth, *hst;
630 	struct vnode **vpp;
631 {
632 	register struct nfsmount *nmp;
633 	struct nfsnode *np;
634 	int error;
635 	struct vattr attrs;
636 
637 	if (mp->mnt_flag & MNT_UPDATE) {
638 		nmp = VFSTONFS(mp);
639 		/* update paths, file handles, etc, here	XXX */
640 		m_freem(nam);
641 		return (0);
642 	} else {
643 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
644 		    M_NFSMNT, M_WAITOK);
645 		bzero((caddr_t)nmp, sizeof (struct nfsmount));
646 		mp->mnt_data = (qaddr_t)nmp;
647 		TAILQ_INIT(&nmp->nm_uidlruhead);
648 		TAILQ_INIT(&nmp->nm_bufq);
649 	}
650 #ifdef Lite2_integrated
651 	vfs_getnewfsid(mp, makefstype(MOUNT_NFS));
652 #else
653 	getnewfsid(mp, makefstype(MOUNT_NFS));
654 #endif
655 	nmp->nm_mountp = mp;
656 	if (argp->flags & NFSMNT_NQNFS)
657 		/*
658 		 * We have to set mnt_maxsymlink to a non-zero value so
659 		 * that COMPAT_43 routines will know that we are setting
660 		 * the d_type field in directories (and can zero it for
661 		 * unsuspecting binaries).
662 		 */
663 		mp->mnt_maxsymlinklen = 1;
664 	nmp->nm_timeo = NFS_TIMEO;
665 	nmp->nm_retry = NFS_RETRANS;
666 	nmp->nm_wsize = NFS_WSIZE;
667 	nmp->nm_rsize = NFS_RSIZE;
668 	nmp->nm_readdirsize = NFS_READDIRSIZE;
669 	nmp->nm_numgrps = NFS_MAXGRPS;
670 	nmp->nm_readahead = NFS_DEFRAHEAD;
671 	nmp->nm_leaseterm = NQ_DEFLEASE;
672 	nmp->nm_deadthresh = NQ_DEADTHRESH;
673 	CIRCLEQ_INIT(&nmp->nm_timerhead);
674 	nmp->nm_inprog = NULLVP;
675 	nmp->nm_fhsize = argp->fhsize;
676 	bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
677 #ifdef COMPAT_09
678 	mp->mnt_stat.f_type = 2;
679 #else
680 	mp->mnt_stat.f_type = 0;
681 #endif
682 	strncpy(&mp->mnt_stat.f_fstypename[0], mp->mnt_op->vfs_name, MFSNAMELEN);
683 	bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
684 	bcopy(pth, mp->mnt_stat.f_mntonname, MNAMELEN);
685 	nmp->nm_nam = nam;
686 
687 	/* Set up the sockets and per-host congestion */
688 	nmp->nm_sotype = argp->sotype;
689 	nmp->nm_soproto = argp->proto;
690 
691 	nfs_decode_args(nmp, argp);
692 
693 	/*
694 	 * For Connection based sockets (TCP,...) defer the connect until
695 	 * the first request, in case the server is not responding.
696 	 */
697 	if (nmp->nm_sotype == SOCK_DGRAM &&
698 		(error = nfs_connect(nmp, (struct nfsreq *)0)))
699 		goto bad;
700 
701 	/*
702 	 * This is silly, but it has to be set so that vinifod() works.
703 	 * We do not want to do an nfs_statfs() here since we can get
704 	 * stuck on a dead server and we are holding a lock on the mount
705 	 * point.
706 	 */
707 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
708 	/*
709 	 * A reference count is needed on the nfsnode representing the
710 	 * remote root.  If this object is not persistent, then backward
711 	 * traversals of the mount point (i.e. "..") will not work if
712 	 * the nfsnode gets flushed out of the cache. Ufs does not have
713 	 * this problem, because one can identify root inodes by their
714 	 * number == ROOTINO (2).
715 	 */
716 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
717 	if (error)
718 		goto bad;
719 	*vpp = NFSTOV(np);
720 	VOP_GETATTR(*vpp, &attrs, curproc->p_ucred, curproc); /* XXX */
721 
722 	return (0);
723 bad:
724 	nfs_disconnect(nmp);
725 	free((caddr_t)nmp, M_NFSMNT);
726 	m_freem(nam);
727 	return (error);
728 }
729 
730 /*
731  * unmount system call
732  */
733 int
734 nfs_unmount(mp, mntflags, p)
735 	struct mount *mp;
736 	int mntflags;
737 	struct proc *p;
738 {
739 	register struct nfsmount *nmp;
740 	struct nfsnode *np;
741 	struct vnode *vp;
742 	int error, flags = 0;
743 
744 	if (mntflags & MNT_FORCE)
745 		flags |= FORCECLOSE;
746 	nmp = VFSTONFS(mp);
747 	/*
748 	 * Goes something like this..
749 	 * - Check for activity on the root vnode (other than ourselves).
750 	 * - Call vflush() to clear out vnodes for this file system,
751 	 *   except for the root vnode.
752 	 * - Decrement reference on the vnode representing remote root.
753 	 * - Close the socket
754 	 * - Free up the data structures
755 	 */
756 	/*
757 	 * We need to decrement the ref. count on the nfsnode representing
758 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
759 	 * has done vput on this vnode, otherwise we would get deadlock!
760 	 */
761 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
762 	if (error)
763 		return(error);
764 	vp = NFSTOV(np);
765 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
766 		vput(vp);
767 		return (EBUSY);
768 	}
769 
770 	/*
771 	 * Must handshake with nqnfs_clientd() if it is active.
772 	 */
773 	nmp->nm_flag |= NFSMNT_DISMINPROG;
774 	while (nmp->nm_inprog != NULLVP)
775 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
776 	error = vflush(mp, vp, flags);
777 	if (error) {
778 		vput(vp);
779 		nmp->nm_flag &= ~NFSMNT_DISMINPROG;
780 		return (error);
781 	}
782 
783 	/*
784 	 * We are now committed to the unmount.
785 	 * For NQNFS, let the server daemon free the nfsmount structure.
786 	 */
787 	if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB))
788 		nmp->nm_flag |= NFSMNT_DISMNT;
789 
790 	/*
791 	 * There are two reference counts to get rid of here.
792 	 */
793 	vrele(vp);
794 	vrele(vp);
795 	vgone(vp);
796 	nfs_disconnect(nmp);
797 	m_freem(nmp->nm_nam);
798 
799 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
800 		free((caddr_t)nmp, M_NFSMNT);
801 	return (0);
802 }
803 
804 /*
805  * Return root of a filesystem
806  */
807 int
808 nfs_root(mp, vpp)
809 	struct mount *mp;
810 	struct vnode **vpp;
811 {
812 	register struct vnode *vp;
813 	struct nfsmount *nmp;
814 	struct nfsnode *np;
815 	int error;
816 
817 	nmp = VFSTONFS(mp);
818 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
819 	if (error)
820 		return (error);
821 	vp = NFSTOV(np);
822 	if (vp->v_type == VNON)
823 		vp->v_type = VDIR;
824 	vp->v_flag = VROOT;
825 	*vpp = vp;
826 	return (0);
827 }
828 
829 extern int syncprt;
830 
831 /*
832  * Flush out the buffer cache
833  */
834 /* ARGSUSED */
835 int
836 nfs_sync(mp, waitfor, cred, p)
837 	struct mount *mp;
838 	int waitfor;
839 	struct ucred *cred;
840 	struct proc *p;
841 {
842 	register struct vnode *vp;
843 	int error, allerror = 0;
844 
845 	/*
846 	 * Force stale buffer cache information to be flushed.
847 	 */
848 loop:
849 	for (vp = mp->mnt_vnodelist.lh_first;
850 	     vp != NULL;
851 	     vp = vp->v_mntvnodes.le_next) {
852 		/*
853 		 * If the vnode that we are about to sync is no longer
854 		 * associated with this mount point, start over.
855 		 */
856 		if (vp->v_mount != mp)
857 			goto loop;
858 		if (VOP_ISLOCKED(vp) || vp->v_dirtyblkhd.lh_first == NULL)
859 			continue;
860 #ifdef Lite2_integrated
861 		if (vget(vp, LK_EXCLUSIVE, p))
862 #else
863 		if (vget(vp, 1))
864 #endif
865 			goto loop;
866 		error = VOP_FSYNC(vp, cred, waitfor, p);
867 		if (error)
868 			allerror = error;
869 		vput(vp);
870 	}
871 	return (allerror);
872 }
873 
874 /*
875  * NFS flat namespace lookup.
876  * Currently unsupported.
877  */
878 /* ARGSUSED */
879 int
880 nfs_vget(mp, ino, vpp)
881 	struct mount *mp;
882 	ino_t ino;
883 	struct vnode **vpp;
884 {
885 
886 	return (EOPNOTSUPP);
887 }
888 
889 #ifdef notyet
890 /*
891  * Do that sysctl thang...
892  */
893 static int
894 nfs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
895 	   size_t newlen, struct proc *p)
896 {
897 	int rv;
898 
899 	/*
900 	 * All names at this level are terminal.
901 	 */
902 	if(namelen > 1)
903 		return ENOTDIR;	/* overloaded */
904 
905 	switch(name[0]) {
906 	case NFS_NFSSTATS:
907 		if(!oldp) {
908 			*oldlenp = sizeof nfsstats;
909 			return 0;
910 		}
911 
912 		if(*oldlenp < sizeof nfsstats) {
913 			*oldlenp = sizeof nfsstats;
914 			return ENOMEM;
915 		}
916 
917 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
918 		if(rv) return rv;
919 
920 		if(newp && newlen != sizeof nfsstats)
921 			return EINVAL;
922 
923 		if(newp) {
924 			return copyin(newp, &nfsstats, sizeof nfsstats);
925 		}
926 		return 0;
927 
928 	default:
929 		return EOPNOTSUPP;
930 	}
931 }
932 #endif
933 
934 
935 /*
936  * At this point, this should never happen
937  */
938 /* ARGSUSED */
939 int
940 nfs_fhtovp(mp, fhp, nam, vpp, exflagsp, credanonp)
941 	register struct mount *mp;
942 	struct fid *fhp;
943 	struct mbuf *nam;
944 	struct vnode **vpp;
945 	int *exflagsp;
946 	struct ucred **credanonp;
947 {
948 
949 	return (EINVAL);
950 }
951 
952 /*
953  * Vnode pointer to File handle, should never happen either
954  */
955 /* ARGSUSED */
956 int
957 nfs_vptofh(vp, fhp)
958 	struct vnode *vp;
959 	struct fid *fhp;
960 {
961 
962 	return (EINVAL);
963 }
964 
965 /*
966  * Vfs start routine, a no-op.
967  */
968 /* ARGSUSED */
969 int
970 nfs_start(mp, flags, p)
971 	struct mount *mp;
972 	int flags;
973 	struct proc *p;
974 {
975 
976 	return (0);
977 }
978 
979 /*
980  * Do operations associated with quotas, not supported
981  */
982 /* ARGSUSED */
983 int
984 nfs_quotactl(mp, cmd, uid, arg, p)
985 	struct mount *mp;
986 	int cmd;
987 	uid_t uid;
988 	caddr_t arg;
989 	struct proc *p;
990 {
991 
992 	return (EOPNOTSUPP);
993 }
994