xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision 21e37cc72a480a47828990a439cde7ac9ffaf0c6)
1 /*	$NetBSD: nfs_vfsops.c,v 1.140 2004/05/25 14:54:58 hannken 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. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_vfsops.c	8.12 (Berkeley) 5/20/95
35  */
36 
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: nfs_vfsops.c,v 1.140 2004/05/25 14:54:58 hannken Exp $");
39 
40 #if defined(_KERNEL_OPT)
41 #include "opt_compat_netbsd.h"
42 #include "opt_nfs.h"
43 #endif
44 
45 #include <sys/param.h>
46 #include <sys/ioctl.h>
47 #include <sys/signal.h>
48 #include <sys/proc.h>
49 #include <sys/namei.h>
50 #include <sys/device.h>
51 #include <sys/vnode.h>
52 #include <sys/kernel.h>
53 #include <sys/mount.h>
54 #include <sys/buf.h>
55 #include <sys/mbuf.h>
56 #include <sys/dirent.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/sysctl.h>
60 #include <sys/systm.h>
61 
62 #include <net/if.h>
63 #include <net/route.h>
64 #include <netinet/in.h>
65 
66 #include <nfs/rpcv2.h>
67 #include <nfs/nfsproto.h>
68 #include <nfs/nfsnode.h>
69 #include <nfs/nfs.h>
70 #include <nfs/nfsmount.h>
71 #include <nfs/xdr_subs.h>
72 #include <nfs/nfsm_subs.h>
73 #include <nfs/nfsdiskless.h>
74 #include <nfs/nqnfs.h>
75 #include <nfs/nfs_var.h>
76 
77 extern struct nfsstats nfsstats;
78 extern int nfs_ticks;
79 
80 MALLOC_DEFINE(M_NFSMNT, "NFS mount", "NFS mount structure");
81 
82 /*
83  * nfs vfs operations.
84  */
85 
86 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
87 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
88 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
89 
90 const struct vnodeopv_desc * const 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_statvfs,
105 	nfs_sync,
106 	nfs_vget,
107 	nfs_fhtovp,
108 	nfs_vptofh,
109 	nfs_vfs_init,
110 	nfs_vfs_reinit,
111 	nfs_vfs_done,
112 	NULL,
113 	nfs_mountroot,
114 	nfs_checkexp,
115 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
116 	nfs_vnodeopv_descs,
117 };
118 
119 extern u_int32_t nfs_procids[NFS_NPROCS];
120 extern u_int32_t nfs_prog, nfs_vers;
121 
122 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
123     struct mount **, struct vnode **, struct proc *));
124 
125 /*
126  * nfs statvfs call
127  */
128 int
129 nfs_statvfs(mp, sbp, p)
130 	struct mount *mp;
131 	struct statvfs *sbp;
132 	struct proc *p;
133 {
134 	struct vnode *vp;
135 	struct nfs_statfs *sfp;
136 	caddr_t cp;
137 	u_int32_t *tl;
138 	int32_t t1, t2;
139 	caddr_t bpos, dpos, cp2;
140 	struct nfsmount *nmp = VFSTONFS(mp);
141 	int error = 0, retattr;
142 #ifdef NFS_V2_ONLY
143 	const int v3 = 0;
144 #else
145 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
146 #endif
147 	struct mbuf *mreq, *mrep = NULL, *md, *mb;
148 	struct ucred *cred;
149 	u_quad_t tquad;
150 	struct nfsnode *np;
151 
152 #ifndef nolint
153 	sfp = (struct nfs_statfs *)0;
154 #endif
155 	vp = nmp->nm_vnode;
156 	np = VTONFS(vp);
157 	cred = crget();
158 	cred->cr_ngroups = 0;
159 #ifndef NFS_V2_ONLY
160 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
161 		(void)nfs_fsinfo(nmp, vp, cred, p);
162 #endif
163 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
164 	nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3));
165 	nfsm_fhtom(np, v3);
166 	nfsm_request(np, NFSPROC_FSSTAT, p, cred);
167 	if (v3)
168 		nfsm_postop_attr(vp, retattr, 0);
169 	if (error) {
170 		if (mrep != NULL) {
171 			if (mrep->m_next != NULL)
172 				printf("nfs_vfsops: nfs_statvfs would lose buffers\n");
173 			m_freem(mrep);
174 		}
175 		goto nfsmout;
176 	}
177 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
178 	sbp->f_flag = nmp->nm_flag;
179 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
180 	if (v3) {
181 		sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE;
182 		tquad = fxdr_hyper(&sfp->sf_tbytes);
183 		sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
184 		tquad = fxdr_hyper(&sfp->sf_fbytes);
185 		sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
186 		tquad = fxdr_hyper(&sfp->sf_abytes);
187 		tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
188 		sbp->f_bresvd = sbp->f_bfree - tquad;
189 		sbp->f_bavail = tquad;
190 #ifdef COMPAT_20
191 		/* Handle older NFS servers returning negative values */
192 		if ((quad_t)sbp->f_bavail < 0)
193 			sbp->f_bavail = 0;
194 #endif
195 		tquad = fxdr_hyper(&sfp->sf_tfiles);
196 		sbp->f_files = tquad;
197 		tquad = fxdr_hyper(&sfp->sf_ffiles);
198 		sbp->f_ffree = tquad;
199 		sbp->f_favail = tquad;
200 		sbp->f_fresvd = 0;
201 		sbp->f_namemax = MAXNAMLEN;
202 	} else {
203 		sbp->f_bsize = NFS_FABLKSIZE;
204 		sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
205 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
206 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
207 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
208 		sbp->f_fresvd = 0;
209 		sbp->f_files = 0;
210 		sbp->f_ffree = 0;
211 		sbp->f_favail = 0;
212 		sbp->f_fresvd = 0;
213 		sbp->f_namemax = MAXNAMLEN;
214 	}
215 	copy_statvfs_info(sbp, mp);
216 	nfsm_reqdone;
217 	crfree(cred);
218 	return (error);
219 }
220 
221 #ifndef NFS_V2_ONLY
222 /*
223  * nfs version 3 fsinfo rpc call
224  */
225 int
226 nfs_fsinfo(nmp, vp, cred, p)
227 	struct nfsmount *nmp;
228 	struct vnode *vp;
229 	struct ucred *cred;
230 	struct proc *p;
231 {
232 	struct nfsv3_fsinfo *fsp;
233 	caddr_t cp;
234 	int32_t t1, t2;
235 	u_int32_t *tl, pref, max;
236 	caddr_t bpos, dpos, cp2;
237 	int error = 0, retattr;
238 	struct mbuf *mreq, *mrep, *md, *mb;
239 	u_int64_t maxfsize;
240 	struct nfsnode *np = VTONFS(vp);
241 
242 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
243 	nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1));
244 	nfsm_fhtom(np, 1);
245 	nfsm_request(np, NFSPROC_FSINFO, p, cred);
246 	nfsm_postop_attr(vp, retattr, 0);
247 	if (!error) {
248 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
249 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
250 		if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 &&
251 		    pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
252 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
253 				~(NFS_FABLKSIZE - 1);
254 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
255 		if (max < nmp->nm_wsize && max > 0) {
256 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
257 			if (nmp->nm_wsize == 0)
258 				nmp->nm_wsize = max;
259 		}
260 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
261 		if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 &&
262 		    pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
263 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
264 				~(NFS_FABLKSIZE - 1);
265 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
266 		if (max < nmp->nm_rsize && max > 0) {
267 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
268 			if (nmp->nm_rsize == 0)
269 				nmp->nm_rsize = max;
270 		}
271 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
272 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
273 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
274 				~(NFS_DIRFRAGSIZ - 1);
275 		if (max < nmp->nm_readdirsize && max > 0) {
276 			nmp->nm_readdirsize = max & ~(NFS_DIRFRAGSIZ - 1);
277 			if (nmp->nm_readdirsize == 0)
278 				nmp->nm_readdirsize = max;
279 		}
280 		/* XXX */
281 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
282 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
283 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
284 			nmp->nm_maxfilesize = maxfsize;
285 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
286 	}
287 	nfsm_reqdone;
288 	return (error);
289 }
290 #endif
291 
292 /*
293  * Mount a remote root fs via. NFS.  It goes like this:
294  * - Call nfs_boot_init() to fill in the nfs_diskless struct
295  * - build the rootfs mount point and call mountnfs() to do the rest.
296  */
297 int
298 nfs_mountroot()
299 {
300 	struct nfs_diskless *nd;
301 	struct vattr attr;
302 	struct mount *mp;
303 	struct vnode *vp;
304 	struct proc *procp;
305 	long n;
306 	int error;
307 
308 	procp = curproc; /* XXX */
309 
310 	if (root_device->dv_class != DV_IFNET)
311 		return (ENODEV);
312 
313 	/*
314 	 * XXX time must be non-zero when we init the interface or else
315 	 * the arp code will wedge.  [Fixed now in if_ether.c]
316 	 * However, the NFS attribute cache gives false "hits" when
317 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
318 	 */
319 	if (time.tv_sec < NFS_MAXATTRTIMO)
320 		time.tv_sec = NFS_MAXATTRTIMO;
321 
322 	/*
323 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
324 	 * Side effect:  Finds and configures a network interface.
325 	 */
326 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
327 	memset((caddr_t)nd, 0, sizeof(*nd));
328 	error = nfs_boot_init(nd, procp);
329 	if (error) {
330 		free(nd, M_NFSMNT);
331 		return (error);
332 	}
333 
334 	/*
335 	 * Create the root mount point.
336 	 */
337 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp);
338 	if (error)
339 		goto out;
340 	printf("root on %s\n", nd->nd_root.ndm_host);
341 
342 	/*
343 	 * Link it into the mount list.
344 	 */
345 	simple_lock(&mountlist_slock);
346 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
347 	simple_unlock(&mountlist_slock);
348 	rootvp = vp;
349 	mp->mnt_vnodecovered = NULLVP;
350 	vfs_unbusy(mp);
351 
352 	/* Get root attributes (for the time). */
353 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
354 	if (error)
355 		panic("nfs_mountroot: getattr for root");
356 	n = attr.va_atime.tv_sec;
357 #ifdef	DEBUG
358 	printf("root time: 0x%lx\n", n);
359 #endif
360 	inittodr(n);
361 
362 out:
363 	if (error)
364 		nfs_boot_cleanup(nd, procp);
365 	free(nd, M_NFSMNT);
366 	return (error);
367 }
368 
369 /*
370  * Internal version of mount system call for diskless setup.
371  * Separate function because we used to call it twice.
372  * (once for root and once for swap)
373  */
374 static int
375 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p)
376 	struct nfs_dlmount *ndmntp;
377 	const char *mntname;	/* mount point name */
378 	struct mount **mpp;
379 	struct vnode **vpp;
380 	struct proc *p;
381 {
382 	struct mount *mp;
383 	struct mbuf *m;
384 	int error;
385 
386 	vfs_rootmountalloc(MOUNT_NFS, (char *)mntname, &mp);
387 
388 	mp->mnt_op = &nfs_vfsops;
389 
390 	/*
391 	 * Historical practice expects NFS root file systems to
392 	 * be initially mounted r/w.
393 	 */
394 	mp->mnt_flag &= ~MNT_RDONLY;
395 
396 	/* Get mbuf for server sockaddr. */
397 	m = m_get(M_WAIT, MT_SONAME);
398 	if (m == NULL)
399 		panic("nfs_mountroot: mget soname for %s", mntname);
400 	MCLAIM(m, &nfs_mowner);
401 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
402 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
403 
404 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
405 			 ndmntp->ndm_args.hostname, vpp, p);
406 	if (error) {
407 		mp->mnt_op->vfs_refcount--;
408 		vfs_unbusy(mp);
409 		printf("nfs_mountroot: mount %s failed: %d\n",
410 		       mntname, error);
411 		free(mp, M_MOUNT);
412 	} else
413 		*mpp = mp;
414 
415 	return (error);
416 }
417 
418 void
419 nfs_decode_args(nmp, argp, p)
420 	struct nfsmount *nmp;
421 	struct nfs_args *argp;
422 	struct proc *p;
423 {
424 	int s;
425 	int adjsock;
426 	int maxio;
427 
428 	s = splsoftnet();
429 
430 	/*
431 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
432 	 * no sense in that context.
433 	 */
434 	if (argp->sotype == SOCK_STREAM)
435 		argp->flags &= ~NFSMNT_NOCONN;
436 
437 	/*
438 	 * Cookie translation is not needed for v2, silently ignore it.
439 	 */
440 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
441 	    NFSMNT_XLATECOOKIE)
442 		argp->flags &= ~NFSMNT_XLATECOOKIE;
443 
444 	/* Re-bind if rsrvd port requested and wasn't on one */
445 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
446 		  && (argp->flags & NFSMNT_RESVPORT);
447 	/* Also re-bind if we're switching to/from a connected UDP socket */
448 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
449 		    (argp->flags & NFSMNT_NOCONN));
450 
451 	/* Update flags. */
452 	nmp->nm_flag = argp->flags;
453 	splx(s);
454 
455 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
456 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
457 		if (nmp->nm_timeo < NFS_MINTIMEO)
458 			nmp->nm_timeo = NFS_MINTIMEO;
459 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
460 			nmp->nm_timeo = NFS_MAXTIMEO;
461 	}
462 
463 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
464 		nmp->nm_retry = argp->retrans;
465 		if (nmp->nm_retry > NFS_MAXREXMIT)
466 			nmp->nm_retry = NFS_MAXREXMIT;
467 	}
468 
469 #ifndef NFS_V2_ONLY
470 	if (argp->flags & NFSMNT_NFSV3) {
471 		if (argp->sotype == SOCK_DGRAM)
472 			maxio = NFS_MAXDGRAMDATA;
473 		else
474 			maxio = NFS_MAXDATA;
475 	} else
476 #endif
477 		maxio = NFS_V2MAXDATA;
478 
479 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
480 		int osize = nmp->nm_wsize;
481 		nmp->nm_wsize = argp->wsize;
482 		/* Round down to multiple of blocksize */
483 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
484 		if (nmp->nm_wsize <= 0)
485 			nmp->nm_wsize = NFS_FABLKSIZE;
486 		adjsock |= (nmp->nm_wsize != osize);
487 	}
488 	if (nmp->nm_wsize > maxio)
489 		nmp->nm_wsize = maxio;
490 	if (nmp->nm_wsize > MAXBSIZE)
491 		nmp->nm_wsize = MAXBSIZE;
492 
493 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
494 		int osize = nmp->nm_rsize;
495 		nmp->nm_rsize = argp->rsize;
496 		/* Round down to multiple of blocksize */
497 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
498 		if (nmp->nm_rsize <= 0)
499 			nmp->nm_rsize = NFS_FABLKSIZE;
500 		adjsock |= (nmp->nm_rsize != osize);
501 	}
502 	if (nmp->nm_rsize > maxio)
503 		nmp->nm_rsize = maxio;
504 	if (nmp->nm_rsize > MAXBSIZE)
505 		nmp->nm_rsize = MAXBSIZE;
506 
507 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
508 		nmp->nm_readdirsize = argp->readdirsize;
509 		/* Round down to multiple of minimum blocksize */
510 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
511 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
512 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
513 		/* Bigger than buffer size makes no sense */
514 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
515 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
516 	} else if (argp->flags & NFSMNT_RSIZE)
517 		nmp->nm_readdirsize = nmp->nm_rsize;
518 
519 	if (nmp->nm_readdirsize > maxio)
520 		nmp->nm_readdirsize = maxio;
521 
522 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
523 		argp->maxgrouplist <= NFS_MAXGRPS)
524 		nmp->nm_numgrps = argp->maxgrouplist;
525 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
526 		argp->readahead <= NFS_MAXRAHEAD)
527 		nmp->nm_readahead = argp->readahead;
528 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
529 		argp->leaseterm <= NQ_MAXLEASE)
530 		nmp->nm_leaseterm = argp->leaseterm;
531 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
532 		argp->deadthresh <= NQ_NEVERDEAD)
533 		nmp->nm_deadthresh = argp->deadthresh;
534 
535 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
536 		    (nmp->nm_soproto != argp->proto));
537 	nmp->nm_sotype = argp->sotype;
538 	nmp->nm_soproto = argp->proto;
539 
540 	if (nmp->nm_so && adjsock) {
541 		nfs_safedisconnect(nmp);
542 		if (nmp->nm_sotype == SOCK_DGRAM)
543 			while (nfs_connect(nmp, (struct nfsreq *)0, p)) {
544 				printf("nfs_args: retrying connect\n");
545 				(void) tsleep((caddr_t)&lbolt,
546 					      PSOCK, "nfscn3", 0);
547 			}
548 	}
549 }
550 
551 /*
552  * VFS Operations.
553  *
554  * mount system call
555  * It seems a bit dumb to copyinstr() the host and path here and then
556  * memcpy() them in mountnfs(), but I wanted to detect errors before
557  * doing the sockargs() call because sockargs() allocates an mbuf and
558  * an error after that means that I have to release the mbuf.
559  */
560 /* ARGSUSED */
561 int
562 nfs_mount(mp, path, data, ndp, p)
563 	struct mount *mp;
564 	const char *path;
565 	void *data;
566 	struct nameidata *ndp;
567 	struct proc *p;
568 {
569 	int error;
570 	struct nfs_args args;
571 	struct mbuf *nam;
572 	struct nfsmount *nmp = VFSTONFS(mp);
573 	struct sockaddr *sa;
574 	struct vnode *vp;
575 	char *pth, *hst;
576 	size_t len;
577 	u_char *nfh;
578 
579 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
580 	if (error)
581 		return (error);
582 
583 	if (mp->mnt_flag & MNT_GETARGS) {
584 
585 		if (nmp == NULL)
586 			return (EIO);
587 		if (args.addr != NULL) {
588 			sa = mtod(nmp->nm_nam, struct sockaddr *);
589 			error = copyout(sa, args.addr, sa->sa_len);
590 			if (error)
591 				return (error);
592 			args.addrlen = sa->sa_len;
593 		} else
594 			args.addrlen = 0;
595 
596 		args.version = NFS_ARGSVERSION;
597 		args.sotype = nmp->nm_sotype;
598 		args.proto = nmp->nm_soproto;
599 		args.fh = NULL;
600 		args.fhsize = 0;
601 		args.flags = nmp->nm_flag;
602 		args.wsize = nmp->nm_wsize;
603 		args.rsize = nmp->nm_rsize;
604 		args.readdirsize = nmp->nm_readdirsize;
605 		args.timeo = nmp->nm_timeo;
606 		args.retrans = nmp->nm_retry;
607 		args.maxgrouplist = nmp->nm_numgrps;
608 		args.readahead = nmp->nm_readahead;
609 		args.leaseterm = nmp->nm_leaseterm;
610 		args.deadthresh = nmp->nm_deadthresh;
611 		args.hostname = NULL;
612 		return (copyout(&args, data, sizeof(args)));
613 	}
614 
615 	if (args.version != NFS_ARGSVERSION)
616 		return (EPROGMISMATCH);
617 #ifdef NFS_V2_ONLY
618 	if (args.flags & NFSMNT_NQNFS)
619 		return (EPROGUNAVAIL);
620 	if (args.flags & NFSMNT_NFSV3)
621 		return (EPROGMISMATCH);
622 #endif
623 	if (mp->mnt_flag & MNT_UPDATE) {
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, p);
635 		return (0);
636 	}
637 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
638 		return (EINVAL);
639 	MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK);
640 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
641 	if (error)
642 		return (error);
643 	MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK);
644 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
645 	if (error)
646 		goto free_nfh;
647 	memset(&pth[len], 0, MNAMELEN - len);
648 	MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK);
649 	error = copyinstr(args.hostname, hst, MNAMELEN - 1, &len);
650 	if (error)
651 		goto free_pth;
652 	memset(&hst[len], 0, MNAMELEN - len);
653 	/* sockargs() call must be after above copyin() calls */
654 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
655 	if (error)
656 		goto free_hst;
657 	MCLAIM(nam, &nfs_mowner);
658 	args.fh = nfh;
659 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
660 
661 free_hst:
662 	FREE(hst, M_TEMP);
663 free_pth:
664 	FREE(pth, M_TEMP);
665 free_nfh:
666 	FREE(nfh, M_TEMP);
667 
668 	return (error);
669 }
670 
671 /*
672  * Common code for mount and mountroot
673  */
674 int
675 mountnfs(argp, mp, nam, pth, hst, vpp, p)
676 	struct nfs_args *argp;
677 	struct mount *mp;
678 	struct mbuf *nam;
679 	const char *pth, *hst;
680 	struct vnode **vpp;
681 	struct proc *p;
682 {
683 	struct nfsmount *nmp;
684 	struct nfsnode *np;
685 	int error;
686 	struct vattr *attrs;
687 	struct ucred *cr;
688 
689 	/*
690 	 * If the number of nfs iothreads to use has never
691 	 * been set, create a reasonable number of them.
692 	 */
693 
694 	if (nfs_niothreads < 0) {
695 		nfs_niothreads = NFS_DEFAULT_NIOTHREADS;
696 		nfs_getset_niothreads(TRUE);
697 	}
698 
699 	if (mp->mnt_flag & MNT_UPDATE) {
700 		nmp = VFSTONFS(mp);
701 		/* update paths, file handles, etc, here	XXX */
702 		m_freem(nam);
703 		return (0);
704 	} else {
705 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
706 		    M_NFSMNT, M_WAITOK);
707 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
708 		mp->mnt_data = nmp;
709 		TAILQ_INIT(&nmp->nm_uidlruhead);
710 		TAILQ_INIT(&nmp->nm_bufq);
711 		lockinit(&nmp->nm_writeverflock, PRIBIO, "nfswverf", 0, 0);
712 		simple_lock_init(&nmp->nm_slock);
713 	}
714 	vfs_getnewfsid(mp);
715 	nmp->nm_mountp = mp;
716 
717 #ifndef NFS_V2_ONLY
718 	if (argp->flags & NFSMNT_NQNFS)
719 		/*
720 		 * We have to set mnt_maxsymlink to a non-zero value so
721 		 * that COMPAT_43 routines will know that we are setting
722 		 * the d_type field in directories (and can zero it for
723 		 * unsuspecting binaries).
724 		 */
725 		mp->mnt_maxsymlinklen = 1;
726 #endif
727 
728 #ifndef NFS_V2_ONLY
729 	if ((argp->flags & NFSMNT_NFSV3) == 0)
730 #endif
731 		/*
732 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
733 		 * will fill this in.
734 		 */
735 		nmp->nm_maxfilesize = 0xffffffffLL;
736 
737 	nmp->nm_timeo = NFS_TIMEO;
738 	nmp->nm_retry = NFS_RETRANS;
739 	nmp->nm_wsize = NFS_WSIZE;
740 	nmp->nm_rsize = NFS_RSIZE;
741 	nmp->nm_readdirsize = NFS_READDIRSIZE;
742 	nmp->nm_numgrps = NFS_MAXGRPS;
743 	nmp->nm_readahead = NFS_DEFRAHEAD;
744 	nmp->nm_leaseterm = NQ_DEFLEASE;
745 	nmp->nm_deadthresh = NQ_DEADTHRESH;
746 	CIRCLEQ_INIT(&nmp->nm_timerhead);
747 	nmp->nm_inprog = NULLVP;
748 	error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, mp, p);
749 	if (error)
750 		goto bad;
751 	nmp->nm_nam = nam;
752 
753 	/* Set up the sockets and per-host congestion */
754 	nmp->nm_sotype = argp->sotype;
755 	nmp->nm_soproto = argp->proto;
756 
757 	nfs_decode_args(nmp, argp, p);
758 
759 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
760 	mp->mnt_dev_bshift = DEV_BSHIFT;
761 
762 	/*
763 	 * For Connection based sockets (TCP,...) defer the connect until
764 	 * the first request, in case the server is not responding.
765 	 */
766 	if (nmp->nm_sotype == SOCK_DGRAM &&
767 		(error = nfs_connect(nmp, (struct nfsreq *)0, p)))
768 		goto bad;
769 
770 	/*
771 	 * This is silly, but it has to be set so that vinifod() works.
772 	 * We do not want to do an nfs_statvfs() here since we can get
773 	 * stuck on a dead server and we are holding a lock on the mount
774 	 * point.
775 	 */
776 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
777 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
778 	if (error)
779 		goto bad;
780 	*vpp = NFSTOV(np);
781 	MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK);
782 	VOP_GETATTR(*vpp, attrs, p->p_ucred, p);
783 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
784 		cr = crget();
785 		cr->cr_uid = attrs->va_uid;
786 		cr->cr_gid = attrs->va_gid;
787 		cr->cr_ngroups = 0;
788 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
789 		crfree(cr);
790 	}
791 	FREE(attrs, M_TEMP);
792 
793 	/*
794 	 * A reference count is needed on the nfsnode representing the
795 	 * remote root.  If this object is not persistent, then backward
796 	 * traversals of the mount point (i.e. "..") will not work if
797 	 * the nfsnode gets flushed out of the cache. Ufs does not have
798 	 * this problem, because one can identify root inodes by their
799 	 * number == ROOTINO (2). So, just unlock, but no rele.
800 	 */
801 
802 	nmp->nm_vnode = *vpp;
803 	VOP_UNLOCK(*vpp, 0);
804 
805 	return (0);
806 bad:
807 	nfs_disconnect(nmp);
808 	free((caddr_t)nmp, M_NFSMNT);
809 	m_freem(nam);
810 	return (error);
811 }
812 
813 /*
814  * unmount system call
815  */
816 int
817 nfs_unmount(mp, mntflags, p)
818 	struct mount *mp;
819 	int mntflags;
820 	struct proc *p;
821 {
822 	struct nfsmount *nmp;
823 	struct vnode *vp;
824 	int error, flags = 0;
825 
826 	if (mntflags & MNT_FORCE)
827 		flags |= FORCECLOSE;
828 	nmp = VFSTONFS(mp);
829 	/*
830 	 * Goes something like this..
831 	 * - Check for activity on the root vnode (other than ourselves).
832 	 * - Call vflush() to clear out vnodes for this file system,
833 	 *   except for the root vnode.
834 	 * - Decrement reference on the vnode representing remote root.
835 	 * - Close the socket
836 	 * - Free up the data structures
837 	 */
838 	/*
839 	 * We need to decrement the ref. count on the nfsnode representing
840 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
841 	 * has done vput on this vnode, otherwise we would get deadlock!
842 	 */
843 	vp = nmp->nm_vnode;
844 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
845 	if (error != 0)
846 		return error;
847 
848 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
849 		vput(vp);
850 		return (EBUSY);
851 	}
852 
853 	/*
854 	 * Must handshake with nqnfs_clientd() if it is active.
855 	 */
856 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
857 	while (nmp->nm_inprog != NULLVP)
858 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
859 	error = vflush(mp, vp, flags);
860 	if (error) {
861 		vput(vp);
862 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
863 		return (error);
864 	}
865 
866 	/*
867 	 * We are now committed to the unmount; mark the mount structure
868 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
869 	 * will go away cleanly.
870 	 */
871 	nmp->nm_iflag |= NFSMNT_DISMNT;
872 
873 	/*
874 	 * There are two reference counts to get rid of here
875 	 * (see comment in mountnfs()).
876 	 */
877 	vrele(vp);
878 	vput(vp);
879 	vgone(vp);
880 	nfs_disconnect(nmp);
881 	m_freem(nmp->nm_nam);
882 
883 	/*
884 	 * For NQNFS, let the server daemon free the nfsmount structure.
885 	 */
886 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
887 		free((caddr_t)nmp, M_NFSMNT);
888 	return (0);
889 }
890 
891 /*
892  * Return root of a filesystem
893  */
894 int
895 nfs_root(mp, vpp)
896 	struct mount *mp;
897 	struct vnode **vpp;
898 {
899 	struct vnode *vp;
900 	struct nfsmount *nmp;
901 	int error;
902 
903 	nmp = VFSTONFS(mp);
904 	vp = nmp->nm_vnode;
905 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
906 	if (error != 0)
907 		return error;
908 	if (vp->v_type == VNON)
909 		vp->v_type = VDIR;
910 	vp->v_flag = VROOT;
911 	*vpp = vp;
912 	return (0);
913 }
914 
915 extern int syncprt;
916 
917 /*
918  * Flush out the buffer cache
919  */
920 /* ARGSUSED */
921 int
922 nfs_sync(mp, waitfor, cred, p)
923 	struct mount *mp;
924 	int waitfor;
925 	struct ucred *cred;
926 	struct proc *p;
927 {
928 	struct vnode *vp;
929 	int error, allerror = 0;
930 
931 	/*
932 	 * Force stale buffer cache information to be flushed.
933 	 */
934 loop:
935 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
936 		/*
937 		 * If the vnode that we are about to sync is no longer
938 		 * associated with this mount point, start over.
939 		 */
940 		if (vp->v_mount != mp)
941 			goto loop;
942 		if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) ||
943 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
944 		     vp->v_uobj.uo_npages == 0))
945 			continue;
946 		if (vget(vp, LK_EXCLUSIVE))
947 			goto loop;
948 		error = VOP_FSYNC(vp, cred,
949 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p);
950 		if (error)
951 			allerror = error;
952 		vput(vp);
953 	}
954 	return (allerror);
955 }
956 
957 /*
958  * NFS flat namespace lookup.
959  * Currently unsupported.
960  */
961 /* ARGSUSED */
962 int
963 nfs_vget(mp, ino, vpp)
964 	struct mount *mp;
965 	ino_t ino;
966 	struct vnode **vpp;
967 {
968 
969 	return (EOPNOTSUPP);
970 }
971 
972 /*
973  * Do that sysctl thang...
974  */
975 static int
976 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS)
977 {
978 	int error;
979 
980 	nfs_getset_niothreads(0);
981         error = sysctl_lookup(SYSCTLFN_CALL(rnode));
982 	if (error || newp == NULL)
983 		return (error);
984 	nfs_getset_niothreads(1);
985 
986 	return (0);
987 }
988 
989 SYSCTL_SETUP(sysctl_vfs_nfs_setup, "sysctl vfs.nfs subtree setup")
990 {
991 
992 	sysctl_createv(clog, 0, NULL, NULL,
993 		       CTLFLAG_PERMANENT,
994 		       CTLTYPE_NODE, "vfs", NULL,
995 		       NULL, 0, NULL, 0,
996 		       CTL_VFS, CTL_EOL);
997 	sysctl_createv(clog, 0, NULL, NULL,
998 		       CTLFLAG_PERMANENT,
999 		       CTLTYPE_NODE, "nfs",
1000 		       SYSCTL_DESCR("NFS vfs options"),
1001 		       NULL, 0, NULL, 0,
1002 		       CTL_VFS, 2, CTL_EOL);
1003 	/*
1004 	 * XXX the "2" above could be dynamic, thereby eliminating one
1005 	 * more instance of the "number to vfs" mapping problem, but
1006 	 * "2" is the order as taken from sys/mount.h
1007 	 */
1008 
1009 	sysctl_createv(clog, 0, NULL, NULL,
1010 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1011 		       CTLTYPE_STRUCT, "nfsstats",
1012 		       SYSCTL_DESCR("NFS operation statistics"),
1013 		       NULL, 0, &nfsstats, sizeof(nfsstats),
1014 		       CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL);
1015 	sysctl_createv(clog, 0, NULL, NULL,
1016 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1017 		       CTLTYPE_INT, "iothreads",
1018 		       SYSCTL_DESCR("Number of NFS client processes desired"),
1019 		       sysctl_vfs_nfs_iothreads, 0, &nfs_niothreads, 0,
1020 		       CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL);
1021 }
1022 
1023 /*
1024  * At this point, this should never happen
1025  */
1026 /* ARGSUSED */
1027 int
1028 nfs_fhtovp(mp, fhp, vpp)
1029 	struct mount *mp;
1030 	struct fid *fhp;
1031 	struct vnode **vpp;
1032 {
1033 
1034 	return (EINVAL);
1035 }
1036 
1037 /* ARGSUSED */
1038 int
1039 nfs_checkexp(mp, nam, exflagsp, credanonp)
1040 	struct mount *mp;
1041 	struct mbuf *nam;
1042 	int *exflagsp;
1043 	struct ucred **credanonp;
1044 {
1045 
1046 	return (EINVAL);
1047 }
1048 
1049 /*
1050  * Vnode pointer to File handle, should never happen either
1051  */
1052 /* ARGSUSED */
1053 int
1054 nfs_vptofh(vp, fhp)
1055 	struct vnode *vp;
1056 	struct fid *fhp;
1057 {
1058 
1059 	return (EINVAL);
1060 }
1061 
1062 /*
1063  * Vfs start routine, a no-op.
1064  */
1065 /* ARGSUSED */
1066 int
1067 nfs_start(mp, flags, p)
1068 	struct mount *mp;
1069 	int flags;
1070 	struct proc *p;
1071 {
1072 
1073 	return (0);
1074 }
1075 
1076 /*
1077  * Do operations associated with quotas, not supported
1078  */
1079 /* ARGSUSED */
1080 int
1081 nfs_quotactl(mp, cmd, uid, arg, p)
1082 	struct mount *mp;
1083 	int cmd;
1084 	uid_t uid;
1085 	void *arg;
1086 	struct proc *p;
1087 {
1088 
1089 	return (EOPNOTSUPP);
1090 }
1091