xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision dc306354b0b29af51801a7632f1e95265a68cd81)
1 /*	$NetBSD: nfs_vfsops.c,v 1.79 1998/11/12 22:31:02 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 #if defined(_KERNEL) && !defined(_LKM)
42 #include "opt_compat_netbsd.h"
43 #endif
44 
45 #include <sys/param.h>
46 #include <sys/conf.h>
47 #include <sys/ioctl.h>
48 #include <sys/signal.h>
49 #include <sys/proc.h>
50 #include <sys/namei.h>
51 #include <sys/device.h>
52 #include <sys/vnode.h>
53 #include <sys/kernel.h>
54 #include <sys/mount.h>
55 #include <sys/buf.h>
56 #include <sys/mbuf.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/systm.h>
60 
61 #include <net/if.h>
62 #include <net/route.h>
63 #include <netinet/in.h>
64 
65 #include <nfs/rpcv2.h>
66 #include <nfs/nfsproto.h>
67 #include <nfs/nfsnode.h>
68 #include <nfs/nfs.h>
69 #include <nfs/nfsmount.h>
70 #include <nfs/xdr_subs.h>
71 #include <nfs/nfsm_subs.h>
72 #include <nfs/nfsdiskless.h>
73 #include <nfs/nqnfs.h>
74 #include <nfs/nfs_var.h>
75 
76 extern struct nfsstats nfsstats;
77 extern int nfs_ticks;
78 
79 int nfs_sysctl __P((int *, u_int, void *, size_t *, void *, size_t,
80 		      struct proc *));
81 
82 /*
83  * nfs vfs operations.
84  */
85 
86 extern struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
87 extern struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
88 extern struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
89 
90 struct vnodeopv_desc *nfs_vnodeopv_descs[] = {
91 	&nfsv2_vnodeop_opv_desc,
92 	&spec_nfsv2nodeop_opv_desc,
93 	&fifo_nfsv2nodeop_opv_desc,
94 	NULL,
95 };
96 
97 struct vfsops nfs_vfsops = {
98 	MOUNT_NFS,
99 	nfs_mount,
100 	nfs_start,
101 	nfs_unmount,
102 	nfs_root,
103 	nfs_quotactl,
104 	nfs_statfs,
105 	nfs_sync,
106 	nfs_vget,
107 	nfs_fhtovp,
108 	nfs_vptofh,
109 	nfs_vfs_init,
110 	nfs_sysctl,
111 	nfs_mountroot,
112 	nfs_vnodeopv_descs,
113 };
114 
115 extern u_int32_t nfs_procids[NFS_NPROCS];
116 extern u_int32_t nfs_prog, nfs_vers;
117 
118 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
119     struct mount **, struct vnode **, struct proc *));
120 
121 #define TRUE	1
122 #define	FALSE	0
123 
124 /*
125  * nfs statfs call
126  */
127 int
128 nfs_statfs(mp, sbp, p)
129 	struct mount *mp;
130 	register struct statfs *sbp;
131 	struct proc *p;
132 {
133 	register struct vnode *vp;
134 	register struct nfs_statfs *sfp;
135 	register caddr_t cp;
136 	register u_int32_t *tl;
137 	register int32_t t1, t2;
138 	caddr_t bpos, dpos, cp2;
139 	struct nfsmount *nmp = VFSTONFS(mp);
140 	int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
141 	struct mbuf *mreq, *mrep = NULL, *md, *mb, *mb2;
142 	struct ucred *cred;
143 	struct nfsnode *np;
144 	u_quad_t tquad;
145 
146 #ifndef nolint
147 	sfp = (struct nfs_statfs *)0;
148 #endif
149 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
150 	if (error)
151 		return (error);
152 	vp = NFSTOV(np);
153 	cred = crget();
154 	cred->cr_ngroups = 0;
155 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
156 		(void)nfs_fsinfo(nmp, vp, cred, p);
157 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
158 	nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
159 	nfsm_fhtom(vp, v3);
160 	nfsm_request(vp, NFSPROC_FSSTAT, p, cred);
161 	if (v3)
162 		nfsm_postop_attr(vp, retattr);
163 	if (error) {
164 		if (mrep != NULL)
165 			m_free(mrep);
166 		goto nfsmout;
167 	}
168 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
169 #ifdef COMPAT_09
170 	sbp->f_type = 2;
171 #else
172 	sbp->f_type = 0;
173 #endif
174 	sbp->f_flags = nmp->nm_flag;
175 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
176 	if (v3) {
177 		sbp->f_bsize = NFS_FABLKSIZE;
178 		fxdr_hyper(&sfp->sf_tbytes, &tquad);
179 		sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
180 		fxdr_hyper(&sfp->sf_fbytes, &tquad);
181 		sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
182 		fxdr_hyper(&sfp->sf_abytes, &tquad);
183 		sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
184 		sbp->f_files = (fxdr_unsigned(int32_t,
185 		    sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
186 		sbp->f_ffree = (fxdr_unsigned(int32_t,
187 		    sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
188 	} else {
189 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
190 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
191 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
192 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
193 		sbp->f_files = 0;
194 		sbp->f_ffree = 0;
195 	}
196 	if (sbp != &mp->mnt_stat) {
197 		memcpy(sbp->f_mntonname, mp->mnt_stat.f_mntonname, MNAMELEN);
198 		memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, MNAMELEN);
199 	}
200 	strncpy(&sbp->f_fstypename[0], mp->mnt_op->vfs_name, MFSNAMELEN);
201 	nfsm_reqdone;
202 	vrele(vp);
203 	crfree(cred);
204 	return (error);
205 }
206 
207 /*
208  * nfs version 3 fsinfo rpc call
209  */
210 int
211 nfs_fsinfo(nmp, vp, cred, p)
212 	register struct nfsmount *nmp;
213 	register struct vnode *vp;
214 	struct ucred *cred;
215 	struct proc *p;
216 {
217 	register struct nfsv3_fsinfo *fsp;
218 	register caddr_t cp;
219 	register int32_t t1, t2;
220 	register u_int32_t *tl, pref, max;
221 	caddr_t bpos, dpos, cp2;
222 	int error = 0, retattr;
223 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
224 	u_int64_t maxfsize;
225 
226 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
227 	nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
228 	nfsm_fhtom(vp, 1);
229 	nfsm_request(vp, NFSPROC_FSINFO, p, cred);
230 	nfsm_postop_attr(vp, retattr);
231 	if (!error) {
232 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
233 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
234 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
235 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
236 				~(NFS_FABLKSIZE - 1);
237 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
238 		if (max < nmp->nm_wsize && max > 0) {
239 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
240 			if (nmp->nm_wsize == 0)
241 				nmp->nm_wsize = max;
242 		}
243 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
244 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
245 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
246 				~(NFS_FABLKSIZE - 1);
247 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
248 		if (max < nmp->nm_rsize && max > 0) {
249 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
250 			if (nmp->nm_rsize == 0)
251 				nmp->nm_rsize = max;
252 		}
253 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
254 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
255 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
256 				~(NFS_DIRFRAGSIZ - 1);
257 		if (max < nmp->nm_readdirsize && max > 0) {
258 			nmp->nm_readdirsize = max & ~(NFS_DIRFRAGSIZ - 1);
259 			if (nmp->nm_readdirsize == 0)
260 				nmp->nm_readdirsize = max;
261 		}
262 		/* XXX */
263 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
264 		fxdr_hyper(&fsp->fs_maxfilesize, &maxfsize);
265 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
266 			nmp->nm_maxfilesize = maxfsize;
267 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
268 	}
269 	nfsm_reqdone;
270 	return (error);
271 }
272 
273 /*
274  * Mount a remote root fs via. NFS.  It goes like this:
275  * - Call nfs_boot_init() to fill in the nfs_diskless struct
276  * - build the rootfs mount point and call mountnfs() to do the rest.
277  */
278 int
279 nfs_mountroot()
280 {
281 	struct nfs_diskless *nd;
282 	struct vattr attr;
283 	struct mount *mp;
284 	struct vnode *vp;
285 	struct proc *procp;
286 	long n;
287 	int error;
288 
289 	procp = curproc; /* XXX */
290 
291 	if (root_device->dv_class != DV_IFNET)
292 		return (ENODEV);
293 
294 	/*
295 	 * XXX time must be non-zero when we init the interface or else
296 	 * the arp code will wedge.  [Fixed now in if_ether.c]
297 	 * However, the NFS attribute cache gives false "hits" when
298 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
299 	 */
300 	if (time.tv_sec < NFS_MAXATTRTIMO)
301 		time.tv_sec = NFS_MAXATTRTIMO;
302 
303 	/*
304 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
305 	 * Side effect:  Finds and configures a network interface.
306 	 */
307 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
308 	memset((caddr_t)nd, 0, sizeof(*nd));
309 	error = nfs_boot_init(nd, procp);
310 	if (error)
311 		goto out;
312 
313 	/*
314 	 * Create the root mount point.
315 	 */
316 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp);
317 	if (error)
318 		goto out;
319 	printf("root on %s\n", nd->nd_root.ndm_host);
320 
321 	/*
322 	 * Link it into the mount list.
323 	 */
324 	simple_lock(&mountlist_slock);
325 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
326 	simple_unlock(&mountlist_slock);
327 	rootvp = vp;
328 	mp->mnt_vnodecovered = NULLVP;
329 	vfs_unbusy(mp);
330 
331 	/* Get root attributes (for the time). */
332 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
333 	if (error)
334 		panic("nfs_mountroot: getattr for root");
335 	n = attr.va_atime.tv_sec;
336 #ifdef	DEBUG
337 	printf("root time: 0x%lx\n", n);
338 #endif
339 	inittodr(n);
340 
341 #if 0
342 	/*
343 	 * XXX splnet, so networks will receive...
344 	 * XXX Which port needed this hack?
345 	 * XXX Should already be at spl0.
346 	 */
347 	splnet();
348 #endif
349 
350 #if 0	/* swap now comes in from swapctl(2) */
351 #ifdef notyet
352 	/* Set up swap credentials. */
353 	proc0.p_ucred->cr_uid = 0;
354 	proc0.p_ucred->cr_gid = 0;
355 	proc0.p_ucred->cr_ngroups = 0;
356 #endif
357 
358 	/*
359 	 * "Mount" the swap device.
360 	 *
361 	 * On a "dataless" configuration (swap on disk) we will have:
362 	 *	(swdevt[0].sw_dev != NODEV) identifying the swap device.
363 	 */
364 	if (bdevvp(swapdev, &swapdev_vp))
365 		panic("nfs_mountroot: can't setup swap vp");
366 	if (swdevt[0].sw_dev != NODEV) {
367 		printf("swap on device 0x%x\n", swdevt[0].sw_dev);
368 		error = 0;
369 		goto out;
370 	}
371 
372 	/*
373 	 * If swapping to an nfs node:  (swdevt[0].sw_dev == NODEV)
374 	 * Create a fake mount point just for the swap vnode so that the
375 	 * swap file can be on a different server from the rootfs.
376 	 */
377 	error = nfs_mount_diskless(&nd->nd_swap, "/swap", &mp, &vp, procp);
378 	if (error) {
379 		printf("nfs_boot: warning: mount(swap), error=%d\n", error);
380 		error = 0;
381 		goto out;
382 	}
383 	vfs_unbusy(mp);
384 	printf("swap on %s\n", nd->nd_swap.ndm_host);
385 
386 	/*
387 	 * Since the swap file is not the root dir of a file system,
388 	 * hack it to a regular file.
389 	 */
390 	vp->v_type = VREG;
391 	vp->v_flag = 0;
392 	swdevt[0].sw_vp = vp;
393 
394 	/*
395 	 * Find out how large the swap file is.
396 	 */
397 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
398 	if (error)
399 		panic("nfs_mountroot: getattr for swap");
400 	n = (long) (attr.va_size >> DEV_BSHIFT);
401 #ifdef	DEBUG
402 	printf("swap size: 0x%lx (blocks)\n", n);
403 #endif
404 	swdevt[0].sw_nblks = n;
405 #endif
406 
407 out:
408 	free(nd, M_NFSMNT);
409 	return (error);
410 }
411 
412 /*
413  * Internal version of mount system call for diskless setup.
414  * Separate function because we used to call it twice.
415  * (once for root and once for swap)
416  */
417 static int
418 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p)
419 	struct nfs_dlmount *ndmntp;
420 	const char *mntname;	/* mount point name */
421 	struct mount **mpp;
422 	struct vnode **vpp;
423 	struct proc *p;
424 {
425 	struct mount *mp;
426 	struct mbuf *m;
427 	int error;
428 
429 	vfs_rootmountalloc(MOUNT_NFS, (char *)mntname, &mp);
430 
431 	mp->mnt_op = &nfs_vfsops;
432 
433 	/*
434 	 * Historical practice expects NFS root file systems to
435 	 * be initially mounted r/w.
436 	 */
437 	mp->mnt_flag &= ~MNT_RDONLY;
438 
439 	/* Get mbuf for server sockaddr. */
440 	m = m_get(M_WAIT, MT_SONAME);
441 	if (m == NULL)
442 		panic("nfs_mountroot: mget soname for %s", mntname);
443 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
444 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
445 
446 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
447 			 ndmntp->ndm_args.hostname, vpp, p);
448 	if (error) {
449 		mp->mnt_op->vfs_refcount--;
450 		vfs_unbusy(mp);
451 		printf("nfs_mountroot: mount %s failed: %d\n",
452 		       mntname, error);
453 		free(mp, M_MOUNT);
454 	} else
455 		*mpp = mp;
456 
457 	return (error);
458 }
459 
460 void
461 nfs_decode_args(nmp, argp)
462 	struct nfsmount *nmp;
463 	struct nfs_args *argp;
464 {
465 	int s;
466 	int adjsock;
467 	int maxio;
468 
469 	s = splsoftnet();
470 
471 	/*
472 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
473 	 * no sense in that context.
474 	 */
475 	if (argp->sotype == SOCK_STREAM)
476 		argp->flags &= ~NFSMNT_NOCONN;
477 
478 	/*
479 	 * Cookie translation is not needed for v2, silently ignore it.
480 	 */
481 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
482 	    NFSMNT_XLATECOOKIE)
483 		argp->flags &= ~NFSMNT_XLATECOOKIE;
484 
485 	/* Re-bind if rsrvd port requested and wasn't on one */
486 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
487 		  && (argp->flags & NFSMNT_RESVPORT);
488 	/* Also re-bind if we're switching to/from a connected UDP socket */
489 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
490 		    (argp->flags & NFSMNT_NOCONN));
491 
492 	/* Update flags atomically.  Don't change the lock bits. */
493 	nmp->nm_flag = argp->flags | nmp->nm_flag;
494 	splx(s);
495 
496 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
497 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
498 		if (nmp->nm_timeo < NFS_MINTIMEO)
499 			nmp->nm_timeo = NFS_MINTIMEO;
500 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
501 			nmp->nm_timeo = NFS_MAXTIMEO;
502 	}
503 
504 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
505 		nmp->nm_retry = argp->retrans;
506 		if (nmp->nm_retry > NFS_MAXREXMIT)
507 			nmp->nm_retry = NFS_MAXREXMIT;
508 	}
509 
510 	if (argp->flags & NFSMNT_NFSV3) {
511 		if (argp->sotype == SOCK_DGRAM)
512 			maxio = NFS_MAXDGRAMDATA;
513 		else
514 			maxio = NFS_MAXDATA;
515 	} else
516 		maxio = NFS_V2MAXDATA;
517 
518 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
519 		int osize = nmp->nm_wsize;
520 		nmp->nm_wsize = argp->wsize;
521 		/* Round down to multiple of blocksize */
522 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
523 		if (nmp->nm_wsize <= 0)
524 			nmp->nm_wsize = NFS_FABLKSIZE;
525 		adjsock |= (nmp->nm_wsize != osize);
526 	}
527 	if (nmp->nm_wsize > maxio)
528 		nmp->nm_wsize = maxio;
529 	if (nmp->nm_wsize > MAXBSIZE)
530 		nmp->nm_wsize = MAXBSIZE;
531 
532 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
533 		int osize = nmp->nm_rsize;
534 		nmp->nm_rsize = argp->rsize;
535 		/* Round down to multiple of blocksize */
536 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
537 		if (nmp->nm_rsize <= 0)
538 			nmp->nm_rsize = NFS_FABLKSIZE;
539 		adjsock |= (nmp->nm_rsize != osize);
540 	}
541 	if (nmp->nm_rsize > maxio)
542 		nmp->nm_rsize = maxio;
543 	if (nmp->nm_rsize > MAXBSIZE)
544 		nmp->nm_rsize = MAXBSIZE;
545 
546 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
547 		nmp->nm_readdirsize = argp->readdirsize;
548 		/* Round down to multiple of minimum blocksize */
549 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
550 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
551 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
552 		/* Bigger than buffer size makes no sense */
553 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
554 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
555 	} else if (argp->flags & NFSMNT_RSIZE)
556 		nmp->nm_readdirsize = nmp->nm_rsize;
557 
558 	if (nmp->nm_readdirsize > maxio)
559 		nmp->nm_readdirsize = maxio;
560 
561 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
562 		argp->maxgrouplist <= NFS_MAXGRPS)
563 		nmp->nm_numgrps = argp->maxgrouplist;
564 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
565 		argp->readahead <= NFS_MAXRAHEAD)
566 		nmp->nm_readahead = argp->readahead;
567 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
568 		argp->leaseterm <= NQ_MAXLEASE)
569 		nmp->nm_leaseterm = argp->leaseterm;
570 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
571 		argp->deadthresh <= NQ_NEVERDEAD)
572 		nmp->nm_deadthresh = argp->deadthresh;
573 
574 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
575 		    (nmp->nm_soproto != argp->proto));
576 	nmp->nm_sotype = argp->sotype;
577 	nmp->nm_soproto = argp->proto;
578 
579 	if (nmp->nm_so && adjsock) {
580 		nfs_safedisconnect(nmp);
581 		if (nmp->nm_sotype == SOCK_DGRAM)
582 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
583 				printf("nfs_args: retrying connect\n");
584 				(void) tsleep((caddr_t)&lbolt,
585 					      PSOCK, "nfscn3", 0);
586 			}
587 	}
588 }
589 
590 /*
591  * VFS Operations.
592  *
593  * mount system call
594  * It seems a bit dumb to copyinstr() the host and path here and then
595  * memcpy() them in mountnfs(), but I wanted to detect errors before
596  * doing the sockargs() call because sockargs() allocates an mbuf and
597  * an error after that means that I have to release the mbuf.
598  */
599 /* ARGSUSED */
600 int
601 nfs_mount(mp, path, data, ndp, p)
602 	struct mount *mp;
603 	const char *path;
604 	void *data;
605 	struct nameidata *ndp;
606 	struct proc *p;
607 {
608 	int error;
609 	struct nfs_args args;
610 	struct mbuf *nam;
611 	struct vnode *vp;
612 	char pth[MNAMELEN], hst[MNAMELEN];
613 	size_t len;
614 	u_char nfh[NFSX_V3FHMAX];
615 
616 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
617 	if (error)
618 		return (error);
619 	if (args.version != NFS_ARGSVERSION)
620 		return (EPROGMISMATCH);
621 	if (mp->mnt_flag & MNT_UPDATE) {
622 		register struct nfsmount *nmp = VFSTONFS(mp);
623 
624 		if (nmp == NULL)
625 			return (EIO);
626 		/*
627 		 * When doing an update, we can't change from or to
628 		 * v3 and/or nqnfs, or change cookie translation
629 		 */
630 		args.flags = (args.flags &
631 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
632 		    (nmp->nm_flag &
633 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
634 		nfs_decode_args(nmp, &args);
635 		return (0);
636 	}
637 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
638 	if (error)
639 		return (error);
640 	error = copyinstr(path, pth, MNAMELEN-1, &len);
641 	if (error)
642 		return (error);
643 	memset(&pth[len], 0, MNAMELEN - len);
644 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
645 	if (error)
646 		return (error);
647 	memset(&hst[len], 0, MNAMELEN - len);
648 	/* sockargs() call must be after above copyin() calls */
649 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
650 	if (error)
651 		return (error);
652 	args.fh = nfh;
653 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
654 	return (error);
655 }
656 
657 /*
658  * Common code for mount and mountroot
659  */
660 int
661 mountnfs(argp, mp, nam, pth, hst, vpp, p)
662 	register struct nfs_args *argp;
663 	register struct mount *mp;
664 	struct mbuf *nam;
665 	const char *pth, *hst;
666 	struct vnode **vpp;
667 	struct proc *p;
668 {
669 	register struct nfsmount *nmp;
670 	struct nfsnode *np;
671 	int error;
672 	struct vattr attrs;
673 	struct ucred *cr;
674 
675 	if (mp->mnt_flag & MNT_UPDATE) {
676 		nmp = VFSTONFS(mp);
677 		/* update paths, file handles, etc, here	XXX */
678 		m_freem(nam);
679 		return (0);
680 	} else {
681 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
682 		    M_NFSMNT, M_WAITOK);
683 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
684 		mp->mnt_data = (qaddr_t)nmp;
685 		TAILQ_INIT(&nmp->nm_uidlruhead);
686 		TAILQ_INIT(&nmp->nm_bufq);
687 	}
688 	vfs_getnewfsid(mp, MOUNT_NFS);
689 	nmp->nm_mountp = mp;
690 
691 	if (argp->flags & NFSMNT_NQNFS)
692 		/*
693 		 * We have to set mnt_maxsymlink to a non-zero value so
694 		 * that COMPAT_43 routines will know that we are setting
695 		 * the d_type field in directories (and can zero it for
696 		 * unsuspecting binaries).
697 		 */
698 		mp->mnt_maxsymlinklen = 1;
699 
700 	if ((argp->flags & NFSMNT_NFSV3) == 0)
701 		/*
702 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
703 		 * will fill this in.
704 		 */
705 		nmp->nm_maxfilesize = 0xffffffffLL;
706 
707 	nmp->nm_timeo = NFS_TIMEO;
708 	nmp->nm_retry = NFS_RETRANS;
709 	nmp->nm_wsize = NFS_WSIZE;
710 	nmp->nm_rsize = NFS_RSIZE;
711 	nmp->nm_readdirsize = NFS_READDIRSIZE;
712 	nmp->nm_numgrps = NFS_MAXGRPS;
713 	nmp->nm_readahead = NFS_DEFRAHEAD;
714 	nmp->nm_leaseterm = NQ_DEFLEASE;
715 	nmp->nm_deadthresh = NQ_DEADTHRESH;
716 	CIRCLEQ_INIT(&nmp->nm_timerhead);
717 	nmp->nm_inprog = NULLVP;
718 	nmp->nm_fhsize = argp->fhsize;
719 	memcpy((caddr_t)nmp->nm_fh, (caddr_t)argp->fh, argp->fhsize);
720 #ifdef COMPAT_09
721 	mp->mnt_stat.f_type = 2;
722 #else
723 	mp->mnt_stat.f_type = 0;
724 #endif
725 	strncpy(&mp->mnt_stat.f_fstypename[0], mp->mnt_op->vfs_name,
726 	    MFSNAMELEN);
727 	memcpy(mp->mnt_stat.f_mntfromname, hst, MNAMELEN);
728 	memcpy(mp->mnt_stat.f_mntonname, pth, MNAMELEN);
729 	nmp->nm_nam = nam;
730 
731 	/* Set up the sockets and per-host congestion */
732 	nmp->nm_sotype = argp->sotype;
733 	nmp->nm_soproto = argp->proto;
734 
735 	nfs_decode_args(nmp, argp);
736 
737 	/*
738 	 * For Connection based sockets (TCP,...) defer the connect until
739 	 * the first request, in case the server is not responding.
740 	 */
741 	if (nmp->nm_sotype == SOCK_DGRAM &&
742 		(error = nfs_connect(nmp, (struct nfsreq *)0)))
743 		goto bad;
744 
745 	/*
746 	 * This is silly, but it has to be set so that vinifod() works.
747 	 * We do not want to do an nfs_statfs() here since we can get
748 	 * stuck on a dead server and we are holding a lock on the mount
749 	 * point.
750 	 */
751 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
752 	/*
753 	 * A reference count is needed on the nfsnode representing the
754 	 * remote root.  If this object is not persistent, then backward
755 	 * traversals of the mount point (i.e. "..") will not work if
756 	 * the nfsnode gets flushed out of the cache. Ufs does not have
757 	 * this problem, because one can identify root inodes by their
758 	 * number == ROOTINO (2).
759 	 */
760 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
761 	if (error)
762 		goto bad;
763 	*vpp = NFSTOV(np);
764 	VOP_GETATTR(*vpp, &attrs, p->p_ucred, p);
765 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
766 		cr = crget();
767 		cr->cr_uid = attrs.va_uid;
768 		cr->cr_gid = attrs.va_gid;
769 		cr->cr_ngroups = 0;
770 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
771 		crfree(cr);
772 	}
773 
774 	return (0);
775 bad:
776 	nfs_disconnect(nmp);
777 	free((caddr_t)nmp, M_NFSMNT);
778 	m_freem(nam);
779 	return (error);
780 }
781 
782 /*
783  * unmount system call
784  */
785 int
786 nfs_unmount(mp, mntflags, p)
787 	struct mount *mp;
788 	int mntflags;
789 	struct proc *p;
790 {
791 	register struct nfsmount *nmp;
792 	struct nfsnode *np;
793 	struct vnode *vp;
794 	int error, flags = 0;
795 
796 	if (mntflags & MNT_FORCE)
797 		flags |= FORCECLOSE;
798 	nmp = VFSTONFS(mp);
799 	/*
800 	 * Goes something like this..
801 	 * - Check for activity on the root vnode (other than ourselves).
802 	 * - Call vflush() to clear out vnodes for this file system,
803 	 *   except for the root vnode.
804 	 * - Decrement reference on the vnode representing remote root.
805 	 * - Close the socket
806 	 * - Free up the data structures
807 	 */
808 	/*
809 	 * We need to decrement the ref. count on the nfsnode representing
810 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
811 	 * has done vput on this vnode, otherwise we would get deadlock!
812 	 */
813 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
814 	if (error)
815 		return(error);
816 	vp = NFSTOV(np);
817 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
818 		vput(vp);
819 		return (EBUSY);
820 	}
821 
822 	/*
823 	 * Must handshake with nqnfs_clientd() if it is active.
824 	 */
825 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
826 	while (nmp->nm_inprog != NULLVP)
827 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
828 	error = vflush(mp, vp, flags);
829 	if (error) {
830 		vput(vp);
831 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
832 		return (error);
833 	}
834 
835 	/*
836 	 * We are now committed to the unmount.
837 	 * For NQNFS, let the server daemon free the nfsmount structure.
838 	 */
839 	if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB))
840 		nmp->nm_iflag |= NFSMNT_DISMNT;
841 
842 	/*
843 	 * There are two reference counts to get rid of here.
844 	 */
845 	vrele(vp);
846 	vrele(vp);
847 	vgone(vp);
848 	nfs_disconnect(nmp);
849 	m_freem(nmp->nm_nam);
850 
851 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
852 		free((caddr_t)nmp, M_NFSMNT);
853 	return (0);
854 }
855 
856 /*
857  * Return root of a filesystem
858  */
859 int
860 nfs_root(mp, vpp)
861 	struct mount *mp;
862 	struct vnode **vpp;
863 {
864 	register struct vnode *vp;
865 	struct nfsmount *nmp;
866 	struct nfsnode *np;
867 	int error;
868 
869 	nmp = VFSTONFS(mp);
870 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
871 	if (error)
872 		return (error);
873 	vp = NFSTOV(np);
874 	if (vp->v_type == VNON)
875 		vp->v_type = VDIR;
876 	vp->v_flag = VROOT;
877 	*vpp = vp;
878 	return (0);
879 }
880 
881 extern int syncprt;
882 
883 /*
884  * Flush out the buffer cache
885  */
886 /* ARGSUSED */
887 int
888 nfs_sync(mp, waitfor, cred, p)
889 	struct mount *mp;
890 	int waitfor;
891 	struct ucred *cred;
892 	struct proc *p;
893 {
894 	register struct vnode *vp;
895 	int error, allerror = 0;
896 
897 	/*
898 	 * Force stale buffer cache information to be flushed.
899 	 */
900 loop:
901 	for (vp = mp->mnt_vnodelist.lh_first;
902 	     vp != NULL;
903 	     vp = vp->v_mntvnodes.le_next) {
904 		/*
905 		 * If the vnode that we are about to sync is no longer
906 		 * associated with this mount point, start over.
907 		 */
908 		if (vp->v_mount != mp)
909 			goto loop;
910 		if (VOP_ISLOCKED(vp) || vp->v_dirtyblkhd.lh_first == NULL)
911 			continue;
912 		if (vget(vp, LK_EXCLUSIVE))
913 			goto loop;
914 		error = VOP_FSYNC(vp, cred,
915 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, p);
916 		if (error)
917 			allerror = error;
918 		vput(vp);
919 	}
920 	return (allerror);
921 }
922 
923 /*
924  * NFS flat namespace lookup.
925  * Currently unsupported.
926  */
927 /* ARGSUSED */
928 int
929 nfs_vget(mp, ino, vpp)
930 	struct mount *mp;
931 	ino_t ino;
932 	struct vnode **vpp;
933 {
934 
935 	return (EOPNOTSUPP);
936 }
937 
938 /*
939  * Do that sysctl thang...
940  */
941 int
942 nfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
943 	int *name;
944 	u_int namelen;
945 	void *oldp;
946 	size_t *oldlenp;
947 	void *newp;
948 	size_t newlen;
949 	struct proc *p;
950 {
951 	int rv;
952 
953 	/*
954 	 * All names at this level are terminal.
955 	 */
956 	if(namelen > 1)
957 		return ENOTDIR;	/* overloaded */
958 
959 	switch(name[0]) {
960 	case NFS_NFSSTATS:
961 		if(!oldp) {
962 			*oldlenp = sizeof nfsstats;
963 			return 0;
964 		}
965 
966 		if(*oldlenp < sizeof nfsstats) {
967 			*oldlenp = sizeof nfsstats;
968 			return ENOMEM;
969 		}
970 
971 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
972 		if(rv) return rv;
973 
974 		if(newp && newlen != sizeof nfsstats)
975 			return EINVAL;
976 
977 		if(newp) {
978 			return copyin(newp, &nfsstats, sizeof nfsstats);
979 		}
980 		return 0;
981 
982 	default:
983 		return EOPNOTSUPP;
984 	}
985 }
986 
987 
988 /*
989  * At this point, this should never happen
990  */
991 /* ARGSUSED */
992 int
993 nfs_fhtovp(mp, fhp, nam, vpp, exflagsp, credanonp)
994 	register struct mount *mp;
995 	struct fid *fhp;
996 	struct mbuf *nam;
997 	struct vnode **vpp;
998 	int *exflagsp;
999 	struct ucred **credanonp;
1000 {
1001 
1002 	return (EINVAL);
1003 }
1004 
1005 /*
1006  * Vnode pointer to File handle, should never happen either
1007  */
1008 /* ARGSUSED */
1009 int
1010 nfs_vptofh(vp, fhp)
1011 	struct vnode *vp;
1012 	struct fid *fhp;
1013 {
1014 
1015 	return (EINVAL);
1016 }
1017 
1018 /*
1019  * Vfs start routine, a no-op.
1020  */
1021 /* ARGSUSED */
1022 int
1023 nfs_start(mp, flags, p)
1024 	struct mount *mp;
1025 	int flags;
1026 	struct proc *p;
1027 {
1028 
1029 	return (0);
1030 }
1031 
1032 /*
1033  * Do operations associated with quotas, not supported
1034  */
1035 /* ARGSUSED */
1036 int
1037 nfs_quotactl(mp, cmd, uid, arg, p)
1038 	struct mount *mp;
1039 	int cmd;
1040 	uid_t uid;
1041 	caddr_t arg;
1042 	struct proc *p;
1043 {
1044 
1045 	return (EOPNOTSUPP);
1046 }
1047