xref: /openbsd-src/sys/nfs/nfs_vfsops.c (revision 99fd087599a8791921855f21bd7e36130f39aadc)
1 /*	$OpenBSD: nfs_vfsops.c,v 1.125 2020/01/10 10:33:35 bluhm Exp $	*/
2 /*	$NetBSD: nfs_vfsops.c,v 1.46.4.1 1996/05/25 22:40:35 fvdl Exp $	*/
3 
4 /*
5  * Copyright (c) 1989, 1993, 1995
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * Rick Macklem at The University of Guelph.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)nfs_vfsops.c	8.12 (Berkeley) 5/20/95
36  */
37 
38 #include <sys/param.h>
39 #include <sys/conf.h>
40 #include <sys/ioctl.h>
41 #include <sys/signal.h>
42 #include <sys/proc.h>
43 #include <sys/namei.h>
44 #include <sys/vnode.h>
45 #include <sys/lock.h>
46 #include <sys/kernel.h>
47 #include <sys/mount.h>
48 #include <sys/swap.h>
49 #include <sys/buf.h>
50 #include <sys/mbuf.h>
51 #include <sys/dirent.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/systm.h>
55 #include <sys/sysctl.h>
56 #include <sys/queue.h>
57 
58 #include <netinet/in.h>
59 
60 #include <nfs/rpcv2.h>
61 #include <nfs/nfsproto.h>
62 #include <nfs/nfsnode.h>
63 #include <nfs/nfs.h>
64 #include <nfs/nfsmount.h>
65 #include <nfs/xdr_subs.h>
66 #include <nfs/nfsm_subs.h>
67 #include <nfs/nfsdiskless.h>
68 #include <nfs/nfs_var.h>
69 
70 extern struct nfsstats nfsstats;
71 extern int nfs_ticks;
72 extern u_int32_t nfs_procids[NFS_NPROCS];
73 
74 int	nfs_sysctl(int *, u_int, void *, size_t *, void *, size_t,
75 	    struct proc *);
76 int	nfs_checkexp(struct mount *, struct mbuf *, int *, struct ucred **);
77 struct	mount *nfs_mount_diskless(struct nfs_dlmount *, char *, int,
78 	    struct vnode **, struct proc *p);
79 int	mountnfs(struct nfs_args *, struct mount *, struct mbuf *,
80 	    const char *, char *, struct vnode **, struct proc *p);
81 int	nfs_quotactl(struct mount *, int, uid_t, caddr_t, struct proc *);
82 int	nfs_root(struct mount *, struct vnode **);
83 int	nfs_start(struct mount *, int, struct proc *);
84 int	nfs_statfs(struct mount *, struct statfs *, struct proc *);
85 int	nfs_sync(struct mount *, int, int, struct ucred *, struct proc *);
86 int	nfs_unmount(struct mount *, int, struct proc *);
87 void	nfs_reaper(void *);
88 int	nfs_vget(struct mount *, ino_t, struct vnode **);
89 int	nfs_vptofh(struct vnode *, struct fid *);
90 int	nfs_mountroot(void);
91 void	nfs_decode_args(struct nfsmount *, struct nfs_args *,
92 	    struct nfs_args *);
93 int	nfs_fhtovp(struct mount *, struct fid *, struct vnode **);
94 
95 /*
96  * nfs vfs operations.
97  */
98 const struct vfsops nfs_vfsops = {
99 	.vfs_mount	= nfs_mount,
100 	.vfs_start	= nfs_start,
101 	.vfs_unmount	= nfs_unmount,
102 	.vfs_root	= nfs_root,
103 	.vfs_quotactl	= nfs_quotactl,
104 	.vfs_statfs	= nfs_statfs,
105 	.vfs_sync	= nfs_sync,
106 	.vfs_vget	= nfs_vget,
107 	.vfs_fhtovp	= nfs_fhtovp,
108 	.vfs_vptofh	= nfs_vptofh,
109 	.vfs_init	= nfs_vfs_init,
110 	.vfs_sysctl	= nfs_sysctl,
111 	.vfs_checkexp	= nfs_checkexp,
112 };
113 
114 /*
115  * nfs statfs call
116  */
117 int
118 nfs_statfs(struct mount *mp, struct statfs *sbp, struct proc *p)
119 {
120 	struct vnode *vp;
121 	struct nfs_statfs *sfp = NULL;
122 	struct nfsm_info	info;
123 	u_int32_t *tl;
124 	int32_t t1;
125 	caddr_t cp2;
126 	struct nfsmount *nmp = VFSTONFS(mp);
127 	int error = 0, retattr;
128 	struct ucred *cred;
129 	u_quad_t tquad;
130 
131 	info.nmi_v3 = (nmp->nm_flag & NFSMNT_NFSV3);
132 
133 	error = nfs_root(mp, &vp);
134 	if (error)
135 		return (error);
136 	cred = crget();
137 	cred->cr_ngroups = 0;
138 	if (info.nmi_v3 && (nmp->nm_flag & NFSMNT_GOTFSINFO) == 0)
139 		(void)nfs_fsinfo(nmp, vp, cred, p);
140 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
141 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(info.nmi_v3));
142 	nfsm_fhtom(&info, vp, info.nmi_v3);
143 
144 	info.nmi_procp = p;
145 	info.nmi_cred = cred;
146 	error = nfs_request(vp, NFSPROC_FSSTAT, &info);
147 	if (info.nmi_v3)
148 		nfsm_postop_attr(vp, retattr);
149 	if (error) {
150 		m_freem(info.nmi_mrep);
151 		goto nfsmout;
152 	}
153 
154 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(info.nmi_v3));
155 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
156 	if (info.nmi_v3) {
157 		sbp->f_bsize = NFS_FABLKSIZE;
158 		tquad = fxdr_hyper(&sfp->sf_tbytes);
159 		sbp->f_blocks = tquad / (u_quad_t)NFS_FABLKSIZE;
160 		tquad = fxdr_hyper(&sfp->sf_fbytes);
161 		sbp->f_bfree = tquad / (u_quad_t)NFS_FABLKSIZE;
162 		tquad = fxdr_hyper(&sfp->sf_abytes);
163 		sbp->f_bavail = (quad_t)tquad / (quad_t)NFS_FABLKSIZE;
164 
165 		tquad = fxdr_hyper(&sfp->sf_tfiles);
166 		sbp->f_files = tquad;
167 		tquad = fxdr_hyper(&sfp->sf_ffiles);
168 		sbp->f_ffree = tquad;
169 		sbp->f_favail = tquad;
170 	} else {
171 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
172 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
173 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
174 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
175 		sbp->f_files = 0;
176 		sbp->f_ffree = 0;
177 		sbp->f_favail = 0;
178 	}
179 	copy_statfs_info(sbp, mp);
180 	m_freem(info.nmi_mrep);
181 nfsmout:
182 	vput(vp);
183 	crfree(cred);
184 	return (error);
185 }
186 
187 /*
188  * nfs version 3 fsinfo rpc call
189  */
190 int
191 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct ucred *cred,
192     struct proc *p)
193 {
194 	struct nfsv3_fsinfo *fsp;
195 	struct nfsm_info	info;
196 	int32_t t1;
197 	u_int32_t *tl, pref, max;
198 	caddr_t cp2;
199 	int error = 0, retattr;
200 
201 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
202 	info.nmi_mb = info.nmi_mreq = nfsm_reqhead(NFSX_FH(1));
203 	nfsm_fhtom(&info, vp, 1);
204 
205 	info.nmi_procp = p;
206 	info.nmi_cred = cred;
207 	error = nfs_request(vp, NFSPROC_FSINFO, &info);
208 
209 	nfsm_postop_attr(vp, retattr);
210 	if (error) {
211 		m_freem(info.nmi_mrep);
212 		goto nfsmout;
213 	}
214 
215 	nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
216 	pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
217 	if (pref < nmp->nm_wsize)
218 		nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
219 			~(NFS_FABLKSIZE - 1);
220 	max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
221 	if (max < nmp->nm_wsize) {
222 		nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
223 		if (nmp->nm_wsize == 0)
224 			nmp->nm_wsize = max;
225 	}
226 	pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
227 	if (pref < nmp->nm_rsize)
228 		nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
229 			~(NFS_FABLKSIZE - 1);
230 	max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
231 	if (max < nmp->nm_rsize) {
232 		nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
233 		if (nmp->nm_rsize == 0)
234 			nmp->nm_rsize = max;
235 	}
236 	pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
237 	if (pref < nmp->nm_readdirsize)
238 		nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
239 			~(NFS_DIRBLKSIZ - 1);
240 	if (max < nmp->nm_readdirsize) {
241 		nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
242 		if (nmp->nm_readdirsize == 0)
243 			nmp->nm_readdirsize = max;
244 	}
245 	nmp->nm_flag |= NFSMNT_GOTFSINFO;
246 
247 	m_freem(info.nmi_mrep);
248 nfsmout:
249 	return (error);
250 }
251 
252 struct nfs_diskless nfs_diskless;
253 
254 /*
255  * Mount a remote root fs via. NFS.  It goes like this:
256  * - Call nfs_boot_init() to fill in the nfs_diskless struct
257  *   (using RARP, bootparam RPC, mountd RPC)
258  * - hand craft the swap nfs vnode hanging off a fake mount point
259  *	if swdevt[0].sw_dev == NODEV
260  * - build the rootfs mount point and call mountnfs() to do the rest.
261  */
262 int
263 nfs_mountroot(void)
264 {
265 	struct vattr attr;
266 	struct mount *mp;
267 	struct vnode *vp;
268 	struct proc *procp;
269 	long n;
270 	int error;
271 
272 	procp = curproc; /* XXX */
273 
274 	/*
275 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
276 	 * Side effect:	 Finds and configures a network interface.
277 	 */
278 	nfs_boot_init(&nfs_diskless, procp);
279 
280 	/*
281 	 * Create the root mount point.
282 	 */
283 	if (nfs_boot_getfh(&nfs_diskless.nd_boot, "root", &nfs_diskless.nd_root, -1))
284 		panic("nfs_mountroot: root");
285 	mp = nfs_mount_diskless(&nfs_diskless.nd_root, "/", 0, &vp, procp);
286 	printf("root on %s\n", nfs_diskless.nd_root.ndm_host);
287 
288 	/*
289 	 * Link it into the mount list.
290 	 */
291 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
292 	rootvp = vp;
293 	vfs_unbusy(mp);
294 
295 	/* Get root attributes (for the time). */
296 	error = VOP_GETATTR(rootvp, &attr, procp->p_ucred, procp);
297 	if (error) panic("nfs_mountroot: getattr for root");
298 	n = attr.va_atime.tv_sec;
299 #ifdef	DEBUG
300 	printf("root time: 0x%lx\n", n);
301 #endif
302 	inittodr(n);
303 
304 #ifdef notyet
305 	/* Set up swap credentials. */
306 	proc0.p_ucred->cr_uid = ntohl(nfs_diskless.swap_ucred.cr_uid);
307 	proc0.p_ucred->cr_gid = ntohl(nfs_diskless.swap_ucred.cr_gid);
308 	if ((proc0.p_ucred->cr_ngroups = ntohs(nfs_diskless.swap_ucred.cr_ngroups)) >
309 		NGROUPS_MAX)
310 		proc0.p_ucred->cr_ngroups = NGROUPS_MAX;
311 	for (i = 0; i < proc0.p_ucred->cr_ngroups; i++)
312 	    proc0.p_ucred->cr_groups[i] = ntohl(nfs_diskless.swap_ucred.cr_groups[i]);
313 #endif
314 
315 	/*
316 	 * "Mount" the swap device.
317 	 *
318 	 * On a "dataless" configuration (swap on disk) we will have:
319 	 *	(swdevt[0].sw_dev != NODEV) identifying the swap device.
320 	 */
321 	if (swdevt[0].sw_dev != NODEV) {
322 		if (bdevvp(swapdev, &swapdev_vp))
323 			panic("nfs_mountroot: can't setup swap vp");
324 		printf("swap on device 0x%x\n", swdevt[0].sw_dev);
325 		return (0);
326 	}
327 
328 	/*
329 	 * If swapping to an nfs node:	(swdevt[0].sw_dev == NODEV)
330 	 * Create a fake mount point just for the swap vnode so that the
331 	 * swap file can be on a different server from the rootfs.
332 	 *
333 	 * Wait 5 retries, finally no swap is cool. -mickey
334 	 */
335 	error = nfs_boot_getfh(&nfs_diskless.nd_boot, "swap", &nfs_diskless.nd_swap, 5);
336 	if (!error) {
337 		mp = nfs_mount_diskless(&nfs_diskless.nd_swap, "/swap", 0, &vp,
338 		    procp);
339 		vfs_unbusy(mp);
340 
341 		/*
342 		 * Since the swap file is not the root dir of a file system,
343 		 * hack it to a regular file.
344 		 */
345 		vp->v_type = VREG;
346 		vp->v_flag = 0;
347 
348 		/*
349 		 * Next line is a hack to make swapmount() work on NFS
350 		 * swap files.
351 		 */
352 		swdevt[0].sw_dev = NETDEV;
353 		/* end hack */
354 		nfs_diskless.sw_vp = vp;
355 
356 		/*
357 		 * Find out how large the swap file is.
358 		 */
359 		error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
360 		if (error)
361 			printf("nfs_mountroot: getattr for swap\n");
362 		n = (long) (attr.va_size >> DEV_BSHIFT);
363 
364 		printf("swap on %s\n", nfs_diskless.nd_swap.ndm_host);
365 #ifdef	DEBUG
366 		printf("swap size: 0x%lx (blocks)\n", n);
367 #endif
368 		return (0);
369 	}
370 
371 	printf("WARNING: no swap\n");
372 	swdevt[0].sw_dev = NODEV;
373 	return (0);
374 }
375 
376 /*
377  * Internal version of mount system call for diskless setup.
378  */
379 struct mount *
380 nfs_mount_diskless(struct nfs_dlmount *ndmntp, char *mntname, int mntflag,
381     struct vnode **vpp, struct proc *p)
382 {
383 	struct mount *mp;
384 	struct mbuf *m;
385 	int error;
386 
387 	if (vfs_rootmountalloc("nfs", mntname, &mp))
388 		panic("nfs_mount_diskless: vfs_rootmountalloc failed");
389 	mp->mnt_flag |= mntflag;
390 
391 	/* Get mbuf for server sockaddr. */
392 	m = m_get(M_WAIT, MT_SONAME);
393 	bcopy(ndmntp->ndm_args.addr, mtod(m, caddr_t),
394 	    (m->m_len = ndmntp->ndm_args.addr->sa_len));
395 
396 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
397 	    ndmntp->ndm_args.hostname, vpp, p);
398 	if (error)
399 		panic("nfs_mountroot: mount %s failed: %d", mntname, error);
400 
401 	return (mp);
402 }
403 
404 void
405 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp,
406     struct nfs_args *nargp)
407 {
408 	int adjsock = 0;
409 	int maxio;
410 
411 #if 0
412 	/* Re-bind if rsrvd port requested and wasn't on one */
413 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
414 		  && (argp->flags & NFSMNT_RESVPORT);
415 #endif
416 	/* Also re-bind if we're switching to/from a connected UDP socket */
417 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
418 	    (argp->flags & NFSMNT_NOCONN));
419 
420 	nmp->nm_flag =
421 	    (argp->flags & ~NFSMNT_INTERNAL) | (nmp->nm_flag & NFSMNT_INTERNAL);
422 
423 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
424 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
425 		if (nmp->nm_timeo < NFS_MINTIMEO)
426 			nmp->nm_timeo = NFS_MINTIMEO;
427 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
428 			nmp->nm_timeo = NFS_MAXTIMEO;
429 	}
430 
431 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1)
432 		nmp->nm_retry = MIN(argp->retrans, NFS_MAXREXMIT);
433 	if (!(nmp->nm_flag & NFSMNT_SOFT))
434 		nmp->nm_retry = NFS_MAXREXMIT + 1; /* past clip limit */
435 
436 	if (argp->flags & NFSMNT_NFSV3) {
437 		if (argp->sotype == SOCK_DGRAM)
438 			maxio = NFS_MAXDGRAMDATA;
439 		else
440 			maxio = NFS_MAXDATA;
441 	} else
442 		maxio = NFS_V2MAXDATA;
443 
444 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
445 		int osize = nmp->nm_wsize;
446 		nmp->nm_wsize = argp->wsize;
447 		/* Round down to multiple of blocksize */
448 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
449 		if (nmp->nm_wsize <= 0)
450 			nmp->nm_wsize = NFS_FABLKSIZE;
451 		adjsock |= (nmp->nm_wsize != osize);
452 	}
453 	if (nmp->nm_wsize > maxio)
454 		nmp->nm_wsize = maxio;
455 	if (nmp->nm_wsize > MAXBSIZE)
456 		nmp->nm_wsize = MAXBSIZE;
457 
458 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
459 		int osize = nmp->nm_rsize;
460 		nmp->nm_rsize = argp->rsize;
461 		/* Round down to multiple of blocksize */
462 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
463 		if (nmp->nm_rsize <= 0)
464 			nmp->nm_rsize = NFS_FABLKSIZE;
465 		adjsock |= (nmp->nm_rsize != osize);
466 	}
467 	if (nmp->nm_rsize > maxio)
468 		nmp->nm_rsize = maxio;
469 	if (nmp->nm_rsize > MAXBSIZE)
470 		nmp->nm_rsize = MAXBSIZE;
471 
472 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
473 		nmp->nm_readdirsize = argp->readdirsize;
474 		/* Round down to multiple of blocksize */
475 		nmp->nm_readdirsize &= ~(NFS_DIRBLKSIZ - 1);
476 		if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
477 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
478 	} else if (argp->flags & NFSMNT_RSIZE)
479 		nmp->nm_readdirsize = nmp->nm_rsize;
480 
481 	if (nmp->nm_readdirsize > maxio)
482 		nmp->nm_readdirsize = maxio;
483 
484 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
485 		argp->maxgrouplist <= NFS_MAXGRPS)
486 		nmp->nm_numgrps = argp->maxgrouplist;
487 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
488 		argp->readahead <= NFS_MAXRAHEAD)
489 		nmp->nm_readahead = argp->readahead;
490 	if (argp->flags & NFSMNT_ACREGMIN && argp->acregmin >= 0) {
491 		if (argp->acregmin > 0xffff)
492 			nmp->nm_acregmin = 0xffff;
493 		else
494 			nmp->nm_acregmin = argp->acregmin;
495 	}
496 	if (argp->flags & NFSMNT_ACREGMAX && argp->acregmax >= 0) {
497 		if (argp->acregmax > 0xffff)
498 			nmp->nm_acregmax = 0xffff;
499 		else
500 			nmp->nm_acregmax = argp->acregmax;
501 	}
502 	if (nmp->nm_acregmin > nmp->nm_acregmax)
503 	  nmp->nm_acregmin = nmp->nm_acregmax;
504 
505 	if (argp->flags & NFSMNT_ACDIRMIN && argp->acdirmin >= 0) {
506 		if (argp->acdirmin > 0xffff)
507 			nmp->nm_acdirmin = 0xffff;
508 		else
509 			nmp->nm_acdirmin = argp->acdirmin;
510 	}
511 	if (argp->flags & NFSMNT_ACDIRMAX && argp->acdirmax >= 0) {
512 		if (argp->acdirmax > 0xffff)
513 			nmp->nm_acdirmax = 0xffff;
514 		else
515 			nmp->nm_acdirmax = argp->acdirmax;
516 	}
517 	if (nmp->nm_acdirmin > nmp->nm_acdirmax)
518 	  nmp->nm_acdirmin = nmp->nm_acdirmax;
519 
520 	if (nmp->nm_so && adjsock) {
521 		nfs_disconnect(nmp);
522 		if (nmp->nm_sotype == SOCK_DGRAM)
523 			while (nfs_connect(nmp, NULL)) {
524 				printf("nfs_args: retrying connect\n");
525 				tsleep_nsec(&lbolt, PSOCK, "nfscon", INFSLP);
526 			}
527 	}
528 
529 	/* Update nargp based on nmp */
530 	nargp->wsize = nmp->nm_wsize;
531 	nargp->rsize = nmp->nm_rsize;
532 	nargp->readdirsize = nmp->nm_readdirsize;
533 	nargp->timeo = nmp->nm_timeo;
534 	nargp->retrans = nmp->nm_retry;
535 	nargp->maxgrouplist = nmp->nm_numgrps;
536 	nargp->readahead = nmp->nm_readahead;
537 	nargp->acregmin = nmp->nm_acregmin;
538 	nargp->acregmax = nmp->nm_acregmax;
539 	nargp->acdirmin = nmp->nm_acdirmin;
540 	nargp->acdirmax = nmp->nm_acdirmax;
541 }
542 
543 /*
544  * VFS Operations.
545  *
546  * mount system call
547  * It seems a bit dumb to copyinstr() the host here and then
548  * bcopy() it in mountnfs(), but I wanted to detect errors before
549  * doing the sockargs() call because sockargs() allocates an mbuf and
550  * an error after that means that I have to release the mbuf.
551  */
552 /* ARGSUSED */
553 int
554 nfs_mount(struct mount *mp, const char *path, void *data,
555     struct nameidata *ndp, struct proc *p)
556 {
557 	int error;
558 	struct nfs_args *args = data;
559 	struct mbuf *nam;
560 	struct vnode *vp;
561 	char hst[MNAMELEN];
562 	size_t len;
563 	u_char nfh[NFSX_V3FHMAX];
564 
565 	if (args &&
566 	    (args->flags & (NFSMNT_NFSV3|NFSMNT_RDIRPLUS)) == NFSMNT_RDIRPLUS)
567 		return (EINVAL);
568 
569 	if (nfs_niothreads < 0) {
570 		nfs_niothreads = 4;
571 		nfs_getset_niothreads(1);
572 	}
573 
574 	if (mp->mnt_flag & MNT_UPDATE) {
575 		struct nfsmount *nmp = VFSTONFS(mp);
576 
577 		if (nmp == NULL)
578 			return (EIO);
579 		/*
580 		 * When doing an update, we can't change from or to
581 		 * v3.
582 		 */
583 		if (args) {
584 			args->flags = (args->flags & ~(NFSMNT_NFSV3)) |
585 			    (nmp->nm_flag & (NFSMNT_NFSV3));
586 			nfs_decode_args(nmp, args, &mp->mnt_stat.mount_info.nfs_args);
587 		}
588 		return (0);
589 	}
590 	if (args->fhsize < 0 || args->fhsize > NFSX_V3FHMAX)
591 		return (EINVAL);
592 	error = copyin(args->fh, nfh, args->fhsize);
593 	if (error)
594 		return (error);
595 	error = copyinstr(args->hostname, hst, MNAMELEN-1, &len);
596 	if (error)
597 		return (error);
598 	memset(&hst[len], 0, MNAMELEN - len);
599 	/* sockargs() call must be after above copyin() calls */
600 	error = sockargs(&nam, args->addr, args->addrlen, MT_SONAME);
601 	if (error)
602 		return (error);
603 	args->fh = nfh;
604 	error = mountnfs(args, mp, nam, path, hst, &vp, p);
605 	return (error);
606 }
607 
608 /*
609  * Common code for mount and mountroot
610  */
611 int
612 mountnfs(struct nfs_args *argp, struct mount *mp, struct mbuf *nam,
613     const char *pth, char *hst, struct vnode **vpp, struct proc *p)
614 {
615 	struct nfsmount *nmp;
616 	struct nfsnode *np;
617 	struct vnode *vp;
618 	struct vattr attr;
619 	int error;
620 
621 	if (mp->mnt_flag & MNT_UPDATE) {
622 		nmp = VFSTONFS(mp);
623 		/* update paths, file handles, etc, here	XXX */
624 		m_freem(nam);
625 		return (0);
626 	} else {
627 		nmp = malloc(sizeof(*nmp), M_NFSMNT,
628 		    M_WAITOK|M_ZERO);
629 		mp->mnt_data = nmp;
630 	}
631 
632 	vfs_getnewfsid(mp);
633 	nmp->nm_mountp = mp;
634 	nmp->nm_timeo = NFS_TIMEO;
635 	nmp->nm_retry = NFS_RETRANS;
636 	nmp->nm_wsize = NFS_WSIZE;
637 	nmp->nm_rsize = NFS_RSIZE;
638 	nmp->nm_readdirsize = NFS_READDIRSIZE;
639 	nmp->nm_numgrps = NFS_MAXGRPS;
640 	nmp->nm_readahead = NFS_DEFRAHEAD;
641 	nmp->nm_acregmin = NFS_MINATTRTIMO;
642 	nmp->nm_acregmax = NFS_MAXATTRTIMO;
643 	nmp->nm_acdirmin = NFS_MINATTRTIMO;
644 	nmp->nm_acdirmax = NFS_MAXATTRTIMO;
645 	mp->mnt_stat.f_namemax = MAXNAMLEN;
646 	memset(mp->mnt_stat.f_mntonname, 0, MNAMELEN);
647 	strlcpy(mp->mnt_stat.f_mntonname, pth, MNAMELEN);
648 	memset(mp->mnt_stat.f_mntfromname, 0, MNAMELEN);
649 	strlcpy(mp->mnt_stat.f_mntfromname, hst, MNAMELEN);
650 	memset(mp->mnt_stat.f_mntfromspec, 0, MNAMELEN);
651 	strlcpy(mp->mnt_stat.f_mntfromspec, hst, MNAMELEN);
652 	bcopy(argp, &mp->mnt_stat.mount_info.nfs_args, sizeof(*argp));
653 	nmp->nm_nam = nam;
654 	nfs_decode_args(nmp, argp, &mp->mnt_stat.mount_info.nfs_args);
655 
656 	nfs_ninit(nmp);
657 	TAILQ_INIT(&nmp->nm_reqsq);
658 	timeout_set_proc(&nmp->nm_rtimeout, nfs_timer, nmp);
659 
660 	/* Set up the sockets and per-host congestion */
661 	nmp->nm_sotype = argp->sotype;
662 	nmp->nm_soproto = argp->proto;
663 
664 	/*
665 	 * For Connection based sockets (TCP,...) defer the connect until
666 	 * the first request, in case the server is not responding.
667 	 */
668 	if (nmp->nm_sotype == SOCK_DGRAM &&
669 	    (error = nfs_connect(nmp, NULL)))
670 		goto bad;
671 
672 	/*
673 	 * This is silly, but it has to be set so that vinifod() works.
674 	 * We do not want to do an nfs_statfs() here since we can get
675 	 * stuck on a dead server and we are holding a lock on the mount
676 	 * point.
677 	 */
678 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
679 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
680 	if (error)
681 		goto bad;
682 	vp = NFSTOV(np);
683 	error = VOP_GETATTR(vp, &attr, p->p_ucred, p);
684 	if (error) {
685 		vput(vp);
686 		goto bad;
687 	}
688 
689 	/*
690 	 * A reference count is needed on the nfsnode representing the
691 	 * remote root.  If this object is not persistent, then backward
692 	 * traversals of the mount point (i.e. "..") will not work if
693 	 * the nfsnode gets flushed out of the cache. Ufs does not have
694 	 * this problem, because one can identify root inodes by their
695 	 * number == ROOTINO (2).  So, just unlock, but no rele.
696 	 */
697 	nmp->nm_vnode = vp;
698 	if (vp->v_type == VNON)
699 		vp->v_type = VDIR;
700 	vp->v_flag = VROOT;
701 	VOP_UNLOCK(vp);
702 	*vpp = vp;
703 
704 	return (0);
705 bad:
706 	nfs_disconnect(nmp);
707 	free(nmp, M_NFSMNT, sizeof(*nmp));
708 	m_freem(nam);
709 	return (error);
710 }
711 
712 /* unmount system call */
713 int
714 nfs_unmount(struct mount *mp, int mntflags, struct proc *p)
715 {
716 	struct nfsmount *nmp;
717 	struct vnode *vp;
718 	int error, flags = 0;
719 
720 	nmp = VFSTONFS(mp);
721 	error = nfs_root(mp, &vp);
722 	if (error)
723 		return (error);
724 
725 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
726 		vput(vp);
727 		return (EBUSY);
728 	}
729 
730 	if (mntflags & MNT_FORCE)
731 		flags |= FORCECLOSE;
732 
733 	error = vflush(mp, vp, flags);
734 	if (error) {
735 		vput(vp);
736 		return (error);
737 	}
738 
739 	/*
740 	 * There are two references count to get rid of here: one
741 	 * from mountnfs() and one from nfs_root() above.
742 	 */
743 	vrele(vp);
744 	vput(vp);
745 	vgone(vp);
746 	nfs_disconnect(nmp);
747 	m_freem(nmp->nm_nam);
748 	timeout_del(&nmp->nm_rtimeout);
749 	timeout_set_proc(&nmp->nm_rtimeout, nfs_reaper, nmp);
750 	timeout_add(&nmp->nm_rtimeout, 0);
751 	mp->mnt_data = NULL;
752 	return (0);
753 }
754 
755 /*
756  * Delay nfs mount point free until pending or sleeping timeouts have finished.
757  */
758 void
759 nfs_reaper(void *arg)
760 {
761 	struct nfsmount *nmp = arg;
762 
763 	free(nmp, M_NFSMNT, sizeof(*nmp));
764 }
765 
766 /*
767  * Return root of a filesystem
768  */
769 int
770 nfs_root(struct mount *mp, struct vnode **vpp)
771 {
772 	struct vnode *vp;
773 	struct nfsmount *nmp;
774 	int error;
775 
776 	nmp = VFSTONFS(mp);
777 	vp = nmp->nm_vnode;
778 	vref(vp);
779 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
780 	if (error) {
781 		vrele(vp);
782 		return (error);
783 	}
784 	*vpp = vp;
785 	return (0);
786 }
787 
788 /*
789  * Flush out the buffer cache
790  */
791 int
792 nfs_sync(struct mount *mp, int waitfor, int stall, struct ucred *cred, struct proc *p)
793 {
794 	struct vnode *vp;
795 	int error, allerror = 0;
796 
797 	/*
798 	 * Don't traverse the vnode list if we want to skip all of them.
799 	 */
800 	if (waitfor == MNT_LAZY)
801 		return (allerror);
802 
803 	/*
804 	 * Force stale buffer cache information to be flushed.
805 	 */
806 loop:
807 	TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
808 		/*
809 		 * If the vnode that we are about to sync is no longer
810 		 * associated with this mount point, start over.
811 		 */
812 		if (vp->v_mount != mp)
813 			goto loop;
814 		if (VOP_ISLOCKED(vp) || LIST_EMPTY(&vp->v_dirtyblkhd))
815 			continue;
816 		if (vget(vp, LK_EXCLUSIVE))
817 			goto loop;
818 		error = VOP_FSYNC(vp, cred, waitfor, p);
819 		if (error)
820 			allerror = error;
821 		vput(vp);
822 	}
823 
824 	return (allerror);
825 }
826 
827 /*
828  * NFS flat namespace lookup.
829  * Currently unsupported.
830  */
831 /* ARGSUSED */
832 int
833 nfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
834 {
835 
836 	return (EOPNOTSUPP);
837 }
838 
839 /*
840  * Do that sysctl thang...
841  */
842 int
843 nfs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
844     size_t newlen, struct proc *p)
845 {
846 	int rv;
847 
848 	/*
849 	 * All names at this level are terminal.
850 	 */
851 	if(namelen > 1)
852 		return ENOTDIR;	/* overloaded */
853 
854 	switch(name[0]) {
855 	case NFS_NFSSTATS:
856 		if(!oldp) {
857 			*oldlenp = sizeof nfsstats;
858 			return 0;
859 		}
860 
861 		if(*oldlenp < sizeof nfsstats) {
862 			*oldlenp = sizeof nfsstats;
863 			return ENOMEM;
864 		}
865 
866 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
867 		if(rv) return rv;
868 
869 		if(newp && newlen != sizeof nfsstats)
870 			return EINVAL;
871 
872 		if(newp) {
873 			return copyin(newp, &nfsstats, sizeof nfsstats);
874 		}
875 		return 0;
876 
877 	case NFS_NIOTHREADS:
878 		nfs_getset_niothreads(0);
879 
880 		rv = sysctl_int(oldp, oldlenp, newp, newlen, &nfs_niothreads);
881 		if (newp)
882 			nfs_getset_niothreads(1);
883 
884 		return rv;
885 
886 	default:
887 		return EOPNOTSUPP;
888 	}
889 }
890 
891 
892 /*
893  * At this point, this should never happen
894  */
895 /* ARGSUSED */
896 int
897 nfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
898 {
899 	return (EINVAL);
900 }
901 
902 /*
903  * Vnode pointer to File handle, should never happen either
904  */
905 /* ARGSUSED */
906 int
907 nfs_vptofh(struct vnode *vp, struct fid *fhp)
908 {
909 	return (EINVAL);
910 }
911 
912 /*
913  * Vfs start routine, a no-op.
914  */
915 /* ARGSUSED */
916 int
917 nfs_start(struct mount *mp, int flags, struct proc *p)
918 {
919 	return (0);
920 }
921 
922 /*
923  * Do operations associated with quotas, not supported
924  */
925 /* ARGSUSED */
926 int
927 nfs_quotactl(struct mount *mp, int cmd, uid_t uid, caddr_t arg, struct proc *p)
928 {
929 	return (EOPNOTSUPP);
930 }
931 
932 /*
933  * check export permission, not supported
934  */
935 /* ARGUSED */
936 int
937 nfs_checkexp(struct mount *mp, struct mbuf *nam, int *exflagsp,
938     struct ucred **credanonp)
939 {
940 	return (EOPNOTSUPP);
941 }
942 
943