xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision 5aefcfdc06931dd97e76246d2fe0302f7b3fe094)
1 /*	$NetBSD: nfs_vfsops.c,v 1.98 2000/12/10 19:36:32 chs 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 #include "opt_nfs.h"
44 #endif
45 
46 #include <sys/param.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/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 int nfs_sysctl __P((int *, u_int, void *, size_t *, void *, size_t,
81 		      struct proc *));
82 
83 /*
84  * nfs vfs operations.
85  */
86 
87 extern struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
88 extern struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
89 extern struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
90 
91 struct vnodeopv_desc *nfs_vnodeopv_descs[] = {
92 	&nfsv2_vnodeop_opv_desc,
93 	&spec_nfsv2nodeop_opv_desc,
94 	&fifo_nfsv2nodeop_opv_desc,
95 	NULL,
96 };
97 
98 struct vfsops nfs_vfsops = {
99 	MOUNT_NFS,
100 	nfs_mount,
101 	nfs_start,
102 	nfs_unmount,
103 	nfs_root,
104 	nfs_quotactl,
105 	nfs_statfs,
106 	nfs_sync,
107 	nfs_vget,
108 	nfs_fhtovp,
109 	nfs_vptofh,
110 	nfs_vfs_init,
111 	nfs_vfs_done,
112 	nfs_sysctl,
113 	nfs_mountroot,
114 	nfs_checkexp,
115 	nfs_vnodeopv_descs,
116 };
117 
118 extern u_int32_t nfs_procids[NFS_NPROCS];
119 extern u_int32_t nfs_prog, nfs_vers;
120 
121 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
122     struct mount **, struct vnode **, struct proc *));
123 
124 #define TRUE	1
125 #define	FALSE	0
126 
127 /*
128  * nfs statfs call
129  */
130 int
131 nfs_statfs(mp, sbp, p)
132 	struct mount *mp;
133 	struct statfs *sbp;
134 	struct proc *p;
135 {
136 	struct vnode *vp;
137 	struct nfs_statfs *sfp;
138 	caddr_t cp;
139 	u_int32_t *tl;
140 	int32_t t1, t2;
141 	caddr_t bpos, dpos, cp2;
142 	struct nfsmount *nmp = VFSTONFS(mp);
143 	int error = 0, retattr;
144 #ifdef NFS_V2_ONLY
145 	const int v3 = 0;
146 #else
147 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
148 #endif
149 	struct mbuf *mreq, *mrep = NULL, *md, *mb, *mb2;
150 	struct ucred *cred;
151 	struct nfsnode *np;
152 	u_quad_t tquad;
153 
154 #ifndef nolint
155 	sfp = (struct nfs_statfs *)0;
156 #endif
157 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
158 	if (error)
159 		return (error);
160 	vp = NFSTOV(np);
161 	cred = crget();
162 	cred->cr_ngroups = 0;
163 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
164 		(void)nfs_fsinfo(nmp, vp, cred, p);
165 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
166 	nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
167 	nfsm_fhtom(vp, v3);
168 	nfsm_request(vp, NFSPROC_FSSTAT, p, cred);
169 	if (v3)
170 		nfsm_postop_attr(vp, retattr);
171 	if (error) {
172 		if (mrep != NULL)
173 			m_free(mrep);
174 		goto nfsmout;
175 	}
176 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
177 #ifdef COMPAT_09
178 	sbp->f_type = 2;
179 #else
180 	sbp->f_type = 0;
181 #endif
182 	sbp->f_flags = nmp->nm_flag;
183 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
184 	if (v3) {
185 		sbp->f_bsize = NFS_FABLKSIZE;
186 		tquad = fxdr_hyper(&sfp->sf_tbytes);
187 		sbp->f_blocks = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
188 		tquad = fxdr_hyper(&sfp->sf_fbytes);
189 		sbp->f_bfree = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
190 		tquad = fxdr_hyper(&sfp->sf_abytes);
191 		sbp->f_bavail = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
192 		tquad = fxdr_hyper(&sfp->sf_tfiles);
193 		sbp->f_files = (long)tquad;
194 		tquad = fxdr_hyper(&sfp->sf_ffiles);
195 		sbp->f_ffree = (long)tquad;
196 	} else {
197 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
198 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
199 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
200 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
201 		sbp->f_files = 0;
202 		sbp->f_ffree = 0;
203 	}
204 	if (sbp != &mp->mnt_stat) {
205 		memcpy(sbp->f_mntonname, mp->mnt_stat.f_mntonname, MNAMELEN);
206 		memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, MNAMELEN);
207 	}
208 	strncpy(&sbp->f_fstypename[0], mp->mnt_op->vfs_name, MFSNAMELEN);
209 	nfsm_reqdone;
210 	vrele(vp);
211 	crfree(cred);
212 	return (error);
213 }
214 
215 #ifndef NFS_V2_ONLY
216 /*
217  * nfs version 3 fsinfo rpc call
218  */
219 int
220 nfs_fsinfo(nmp, vp, cred, p)
221 	struct nfsmount *nmp;
222 	struct vnode *vp;
223 	struct ucred *cred;
224 	struct proc *p;
225 {
226 	struct nfsv3_fsinfo *fsp;
227 	caddr_t cp;
228 	int32_t t1, t2;
229 	u_int32_t *tl, pref, max;
230 	caddr_t bpos, dpos, cp2;
231 	int error = 0, retattr;
232 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
233 	u_int64_t maxfsize;
234 
235 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
236 	nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
237 	nfsm_fhtom(vp, 1);
238 	nfsm_request(vp, NFSPROC_FSINFO, p, cred);
239 	nfsm_postop_attr(vp, retattr);
240 	if (!error) {
241 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
242 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
243 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
244 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
245 				~(NFS_FABLKSIZE - 1);
246 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
247 		if (max < nmp->nm_wsize && max > 0) {
248 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
249 			if (nmp->nm_wsize == 0)
250 				nmp->nm_wsize = max;
251 		}
252 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
253 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
254 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
255 				~(NFS_FABLKSIZE - 1);
256 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
257 		if (max < nmp->nm_rsize && max > 0) {
258 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
259 			if (nmp->nm_rsize == 0)
260 				nmp->nm_rsize = max;
261 		}
262 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
263 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
264 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
265 				~(NFS_DIRFRAGSIZ - 1);
266 		if (max < nmp->nm_readdirsize && max > 0) {
267 			nmp->nm_readdirsize = max & ~(NFS_DIRFRAGSIZ - 1);
268 			if (nmp->nm_readdirsize == 0)
269 				nmp->nm_readdirsize = max;
270 		}
271 		/* XXX */
272 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
273 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
274 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
275 			nmp->nm_maxfilesize = maxfsize;
276 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
277 	}
278 	nfsm_reqdone;
279 	return (error);
280 }
281 #endif
282 
283 /*
284  * Mount a remote root fs via. NFS.  It goes like this:
285  * - Call nfs_boot_init() to fill in the nfs_diskless struct
286  * - build the rootfs mount point and call mountnfs() to do the rest.
287  */
288 int
289 nfs_mountroot()
290 {
291 	struct nfs_diskless *nd;
292 	struct vattr attr;
293 	struct mount *mp;
294 	struct vnode *vp;
295 	struct proc *procp;
296 	long n;
297 	int error;
298 
299 	procp = curproc; /* XXX */
300 
301 	if (root_device->dv_class != DV_IFNET)
302 		return (ENODEV);
303 
304 	/*
305 	 * XXX time must be non-zero when we init the interface or else
306 	 * the arp code will wedge.  [Fixed now in if_ether.c]
307 	 * However, the NFS attribute cache gives false "hits" when
308 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
309 	 */
310 	if (time.tv_sec < NFS_MAXATTRTIMO)
311 		time.tv_sec = NFS_MAXATTRTIMO;
312 
313 	/*
314 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
315 	 * Side effect:  Finds and configures a network interface.
316 	 */
317 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
318 	memset((caddr_t)nd, 0, sizeof(*nd));
319 	error = nfs_boot_init(nd, procp);
320 	if (error) {
321 		free(nd, M_NFSMNT);
322 		return (error);
323 	}
324 
325 	/*
326 	 * Create the root mount point.
327 	 */
328 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp);
329 	if (error)
330 		goto out;
331 	printf("root on %s\n", nd->nd_root.ndm_host);
332 
333 	/*
334 	 * Link it into the mount list.
335 	 */
336 	simple_lock(&mountlist_slock);
337 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
338 	simple_unlock(&mountlist_slock);
339 	rootvp = vp;
340 	mp->mnt_vnodecovered = NULLVP;
341 	vfs_unbusy(mp);
342 
343 	/* Get root attributes (for the time). */
344 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
345 	if (error)
346 		panic("nfs_mountroot: getattr for root");
347 	n = attr.va_atime.tv_sec;
348 #ifdef	DEBUG
349 	printf("root time: 0x%lx\n", n);
350 #endif
351 	inittodr(n);
352 
353 out:
354 	if (error)
355 		nfs_boot_cleanup(nd, procp);
356 	free(nd, M_NFSMNT);
357 	return (error);
358 }
359 
360 /*
361  * Internal version of mount system call for diskless setup.
362  * Separate function because we used to call it twice.
363  * (once for root and once for swap)
364  */
365 static int
366 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p)
367 	struct nfs_dlmount *ndmntp;
368 	const char *mntname;	/* mount point name */
369 	struct mount **mpp;
370 	struct vnode **vpp;
371 	struct proc *p;
372 {
373 	struct mount *mp;
374 	struct mbuf *m;
375 	int error;
376 
377 	vfs_rootmountalloc(MOUNT_NFS, (char *)mntname, &mp);
378 
379 	mp->mnt_op = &nfs_vfsops;
380 
381 	/*
382 	 * Historical practice expects NFS root file systems to
383 	 * be initially mounted r/w.
384 	 */
385 	mp->mnt_flag &= ~MNT_RDONLY;
386 
387 	/* Get mbuf for server sockaddr. */
388 	m = m_get(M_WAIT, MT_SONAME);
389 	if (m == NULL)
390 		panic("nfs_mountroot: mget soname for %s", mntname);
391 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
392 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
393 
394 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
395 			 ndmntp->ndm_args.hostname, vpp, p);
396 	if (error) {
397 		mp->mnt_op->vfs_refcount--;
398 		vfs_unbusy(mp);
399 		printf("nfs_mountroot: mount %s failed: %d\n",
400 		       mntname, error);
401 		free(mp, M_MOUNT);
402 	} else
403 		*mpp = mp;
404 
405 	return (error);
406 }
407 
408 void
409 nfs_decode_args(nmp, argp)
410 	struct nfsmount *nmp;
411 	struct nfs_args *argp;
412 {
413 	int s;
414 	int adjsock;
415 	int maxio;
416 
417 	s = splsoftnet();
418 
419 	/*
420 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
421 	 * no sense in that context.
422 	 */
423 	if (argp->sotype == SOCK_STREAM)
424 		argp->flags &= ~NFSMNT_NOCONN;
425 
426 	/*
427 	 * Cookie translation is not needed for v2, silently ignore it.
428 	 */
429 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
430 	    NFSMNT_XLATECOOKIE)
431 		argp->flags &= ~NFSMNT_XLATECOOKIE;
432 
433 	/* Re-bind if rsrvd port requested and wasn't on one */
434 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
435 		  && (argp->flags & NFSMNT_RESVPORT);
436 	/* Also re-bind if we're switching to/from a connected UDP socket */
437 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
438 		    (argp->flags & NFSMNT_NOCONN));
439 
440 	/* Update flags. */
441 	nmp->nm_flag = argp->flags;
442 	splx(s);
443 
444 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
445 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
446 		if (nmp->nm_timeo < NFS_MINTIMEO)
447 			nmp->nm_timeo = NFS_MINTIMEO;
448 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
449 			nmp->nm_timeo = NFS_MAXTIMEO;
450 	}
451 
452 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
453 		nmp->nm_retry = argp->retrans;
454 		if (nmp->nm_retry > NFS_MAXREXMIT)
455 			nmp->nm_retry = NFS_MAXREXMIT;
456 	}
457 
458 #ifndef NFS_V2_ONLY
459 	if (argp->flags & NFSMNT_NFSV3) {
460 		if (argp->sotype == SOCK_DGRAM)
461 			maxio = NFS_MAXDGRAMDATA;
462 		else
463 			maxio = NFS_MAXDATA;
464 	} else
465 #endif
466 		maxio = NFS_V2MAXDATA;
467 
468 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
469 		int osize = nmp->nm_wsize;
470 		nmp->nm_wsize = argp->wsize;
471 		/* Round down to multiple of blocksize */
472 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
473 		if (nmp->nm_wsize <= 0)
474 			nmp->nm_wsize = NFS_FABLKSIZE;
475 		adjsock |= (nmp->nm_wsize != osize);
476 	}
477 	if (nmp->nm_wsize > maxio)
478 		nmp->nm_wsize = maxio;
479 	if (nmp->nm_wsize > MAXBSIZE)
480 		nmp->nm_wsize = MAXBSIZE;
481 
482 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
483 		int osize = nmp->nm_rsize;
484 		nmp->nm_rsize = argp->rsize;
485 		/* Round down to multiple of blocksize */
486 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
487 		if (nmp->nm_rsize <= 0)
488 			nmp->nm_rsize = NFS_FABLKSIZE;
489 		adjsock |= (nmp->nm_rsize != osize);
490 	}
491 	if (nmp->nm_rsize > maxio)
492 		nmp->nm_rsize = maxio;
493 	if (nmp->nm_rsize > MAXBSIZE)
494 		nmp->nm_rsize = MAXBSIZE;
495 
496 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
497 		nmp->nm_readdirsize = argp->readdirsize;
498 		/* Round down to multiple of minimum blocksize */
499 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
500 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
501 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
502 		/* Bigger than buffer size makes no sense */
503 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
504 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
505 	} else if (argp->flags & NFSMNT_RSIZE)
506 		nmp->nm_readdirsize = nmp->nm_rsize;
507 
508 	if (nmp->nm_readdirsize > maxio)
509 		nmp->nm_readdirsize = maxio;
510 
511 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
512 		argp->maxgrouplist <= NFS_MAXGRPS)
513 		nmp->nm_numgrps = argp->maxgrouplist;
514 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
515 		argp->readahead <= NFS_MAXRAHEAD)
516 		nmp->nm_readahead = argp->readahead;
517 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
518 		argp->leaseterm <= NQ_MAXLEASE)
519 		nmp->nm_leaseterm = argp->leaseterm;
520 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
521 		argp->deadthresh <= NQ_NEVERDEAD)
522 		nmp->nm_deadthresh = argp->deadthresh;
523 
524 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
525 		    (nmp->nm_soproto != argp->proto));
526 	nmp->nm_sotype = argp->sotype;
527 	nmp->nm_soproto = argp->proto;
528 
529 	if (nmp->nm_so && adjsock) {
530 		nfs_safedisconnect(nmp);
531 		if (nmp->nm_sotype == SOCK_DGRAM)
532 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
533 				printf("nfs_args: retrying connect\n");
534 				(void) tsleep((caddr_t)&lbolt,
535 					      PSOCK, "nfscn3", 0);
536 			}
537 	}
538 }
539 
540 /*
541  * VFS Operations.
542  *
543  * mount system call
544  * It seems a bit dumb to copyinstr() the host and path here and then
545  * memcpy() them in mountnfs(), but I wanted to detect errors before
546  * doing the sockargs() call because sockargs() allocates an mbuf and
547  * an error after that means that I have to release the mbuf.
548  */
549 /* ARGSUSED */
550 int
551 nfs_mount(mp, path, data, ndp, p)
552 	struct mount *mp;
553 	const char *path;
554 	void *data;
555 	struct nameidata *ndp;
556 	struct proc *p;
557 {
558 	int error;
559 	struct nfs_args args;
560 	struct mbuf *nam;
561 	struct vnode *vp;
562 	char pth[MNAMELEN], hst[MNAMELEN];
563 	size_t len;
564 	u_char nfh[NFSX_V3FHMAX];
565 
566 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
567 	if (error)
568 		return (error);
569 	if (args.version != NFS_ARGSVERSION)
570 		return (EPROGMISMATCH);
571 #ifdef NFS_V2_ONLY
572 	args.flags &= ~(NFSMNT_NFSV3 | NFSMNT_NQNFS);
573 #endif
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 and/or nqnfs, or change cookie translation
582 		 */
583 		args.flags = (args.flags &
584 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
585 		    (nmp->nm_flag &
586 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
587 		nfs_decode_args(nmp, &args);
588 		return (0);
589 	}
590 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
591 	if (error)
592 		return (error);
593 	error = copyinstr(path, pth, MNAMELEN-1, &len);
594 	if (error)
595 		return (error);
596 	memset(&pth[len], 0, MNAMELEN - len);
597 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
598 	if (error)
599 		return (error);
600 	memset(&hst[len], 0, MNAMELEN - len);
601 	/* sockargs() call must be after above copyin() calls */
602 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
603 	if (error)
604 		return (error);
605 	args.fh = nfh;
606 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
607 	return (error);
608 }
609 
610 /*
611  * Common code for mount and mountroot
612  */
613 int
614 mountnfs(argp, mp, nam, pth, hst, vpp, p)
615 	struct nfs_args *argp;
616 	struct mount *mp;
617 	struct mbuf *nam;
618 	const char *pth, *hst;
619 	struct vnode **vpp;
620 	struct proc *p;
621 {
622 	struct nfsmount *nmp;
623 	struct nfsnode *np;
624 	int error;
625 	struct vattr attrs;
626 	struct ucred *cr;
627 
628 	/*
629 	 * If the number of nfs iothreads to use has never
630 	 * been set, create a reasonable number of them.
631 	 */
632 
633 	if (nfs_niothreads < 0) {
634 		nfs_niothreads = 4;
635 		nfs_getset_niothreads(TRUE);
636 	}
637 
638 	if (mp->mnt_flag & MNT_UPDATE) {
639 		nmp = VFSTONFS(mp);
640 		/* update paths, file handles, etc, here	XXX */
641 		m_freem(nam);
642 		return (0);
643 	} else {
644 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
645 		    M_NFSMNT, M_WAITOK);
646 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
647 		mp->mnt_data = (qaddr_t)nmp;
648 		TAILQ_INIT(&nmp->nm_uidlruhead);
649 		TAILQ_INIT(&nmp->nm_bufq);
650 	}
651 	vfs_getnewfsid(mp);
652 	nmp->nm_mountp = mp;
653 
654 #ifndef NFS_V2_ONLY
655 	if (argp->flags & NFSMNT_NQNFS)
656 		/*
657 		 * We have to set mnt_maxsymlink to a non-zero value so
658 		 * that COMPAT_43 routines will know that we are setting
659 		 * the d_type field in directories (and can zero it for
660 		 * unsuspecting binaries).
661 		 */
662 		mp->mnt_maxsymlinklen = 1;
663 #endif
664 
665 #ifndef NFS_V2_ONLY
666 	if ((argp->flags & NFSMNT_NFSV3) == 0)
667 #endif
668 		/*
669 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
670 		 * will fill this in.
671 		 */
672 		nmp->nm_maxfilesize = 0xffffffffLL;
673 
674 	nmp->nm_timeo = NFS_TIMEO;
675 	nmp->nm_retry = NFS_RETRANS;
676 	nmp->nm_wsize = NFS_WSIZE;
677 	nmp->nm_rsize = NFS_RSIZE;
678 	nmp->nm_readdirsize = NFS_READDIRSIZE;
679 	nmp->nm_numgrps = NFS_MAXGRPS;
680 	nmp->nm_readahead = NFS_DEFRAHEAD;
681 	nmp->nm_leaseterm = NQ_DEFLEASE;
682 	nmp->nm_deadthresh = NQ_DEADTHRESH;
683 	CIRCLEQ_INIT(&nmp->nm_timerhead);
684 	nmp->nm_inprog = NULLVP;
685 	nmp->nm_fhsize = argp->fhsize;
686 	memcpy((caddr_t)nmp->nm_fh, (caddr_t)argp->fh, argp->fhsize);
687 #ifdef COMPAT_09
688 	mp->mnt_stat.f_type = 2;
689 #else
690 	mp->mnt_stat.f_type = 0;
691 #endif
692 	mp->mnt_fs_bshift = DEV_BSHIFT;
693 	mp->mnt_dev_bshift = -1;
694 	strncpy(&mp->mnt_stat.f_fstypename[0], mp->mnt_op->vfs_name,
695 	    MFSNAMELEN);
696 	memcpy(mp->mnt_stat.f_mntfromname, hst, MNAMELEN);
697 	memcpy(mp->mnt_stat.f_mntonname, pth, MNAMELEN);
698 	nmp->nm_nam = nam;
699 
700 	/* Set up the sockets and per-host congestion */
701 	nmp->nm_sotype = argp->sotype;
702 	nmp->nm_soproto = argp->proto;
703 
704 	nfs_decode_args(nmp, argp);
705 
706 	/*
707 	 * For Connection based sockets (TCP,...) defer the connect until
708 	 * the first request, in case the server is not responding.
709 	 */
710 	if (nmp->nm_sotype == SOCK_DGRAM &&
711 		(error = nfs_connect(nmp, (struct nfsreq *)0)))
712 		goto bad;
713 
714 	/*
715 	 * This is silly, but it has to be set so that vinifod() works.
716 	 * We do not want to do an nfs_statfs() here since we can get
717 	 * stuck on a dead server and we are holding a lock on the mount
718 	 * point.
719 	 */
720 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
721 	/*
722 	 * A reference count is needed on the nfsnode representing the
723 	 * remote root.  If this object is not persistent, then backward
724 	 * traversals of the mount point (i.e. "..") will not work if
725 	 * the nfsnode gets flushed out of the cache. Ufs does not have
726 	 * this problem, because one can identify root inodes by their
727 	 * number == ROOTINO (2).
728 	 */
729 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
730 	if (error)
731 		goto bad;
732 	*vpp = NFSTOV(np);
733 	VOP_GETATTR(*vpp, &attrs, p->p_ucred, p);
734 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
735 		cr = crget();
736 		cr->cr_uid = attrs.va_uid;
737 		cr->cr_gid = attrs.va_gid;
738 		cr->cr_ngroups = 0;
739 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
740 		crfree(cr);
741 	}
742 
743 	return (0);
744 bad:
745 	nfs_disconnect(nmp);
746 	free((caddr_t)nmp, M_NFSMNT);
747 	m_freem(nam);
748 	return (error);
749 }
750 
751 /*
752  * unmount system call
753  */
754 int
755 nfs_unmount(mp, mntflags, p)
756 	struct mount *mp;
757 	int mntflags;
758 	struct proc *p;
759 {
760 	struct nfsmount *nmp;
761 	struct nfsnode *np;
762 	struct vnode *vp;
763 	int error, flags = 0;
764 
765 	if (mntflags & MNT_FORCE)
766 		flags |= FORCECLOSE;
767 	nmp = VFSTONFS(mp);
768 	/*
769 	 * Goes something like this..
770 	 * - Check for activity on the root vnode (other than ourselves).
771 	 * - Call vflush() to clear out vnodes for this file system,
772 	 *   except for the root vnode.
773 	 * - Decrement reference on the vnode representing remote root.
774 	 * - Close the socket
775 	 * - Free up the data structures
776 	 */
777 	/*
778 	 * We need to decrement the ref. count on the nfsnode representing
779 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
780 	 * has done vput on this vnode, otherwise we would get deadlock!
781 	 */
782 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
783 	if (error)
784 		return(error);
785 	vp = NFSTOV(np);
786 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
787 		vput(vp);
788 		return (EBUSY);
789 	}
790 
791 	/*
792 	 * Must handshake with nqnfs_clientd() if it is active.
793 	 */
794 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
795 	while (nmp->nm_inprog != NULLVP)
796 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
797 	error = vflush(mp, vp, flags);
798 	if (error) {
799 		vput(vp);
800 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
801 		return (error);
802 	}
803 
804 	/*
805 	 * We are now committed to the unmount; mark the mount structure
806 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
807 	 * will go away cleanly.
808 	 */
809 	nmp->nm_iflag |= NFSMNT_DISMNT;
810 
811 	/*
812 	 * There are two reference counts to get rid of here.
813 	 */
814 	vrele(vp);
815 	vrele(vp);
816 	vgone(vp);
817 	nfs_disconnect(nmp);
818 	m_freem(nmp->nm_nam);
819 
820 	/*
821 	 * For NQNFS, let the server daemon free the nfsmount structure.
822 	 */
823 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
824 		free((caddr_t)nmp, M_NFSMNT);
825 	return (0);
826 }
827 
828 /*
829  * Return root of a filesystem
830  */
831 int
832 nfs_root(mp, vpp)
833 	struct mount *mp;
834 	struct vnode **vpp;
835 {
836 	struct vnode *vp;
837 	struct nfsmount *nmp;
838 	struct nfsnode *np;
839 	int error;
840 
841 	nmp = VFSTONFS(mp);
842 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
843 	if (error)
844 		return (error);
845 	vp = NFSTOV(np);
846 	if (vp->v_type == VNON)
847 		vp->v_type = VDIR;
848 	vp->v_flag = VROOT;
849 	*vpp = vp;
850 	return (0);
851 }
852 
853 extern int syncprt;
854 
855 /*
856  * Flush out the buffer cache
857  */
858 /* ARGSUSED */
859 int
860 nfs_sync(mp, waitfor, cred, p)
861 	struct mount *mp;
862 	int waitfor;
863 	struct ucred *cred;
864 	struct proc *p;
865 {
866 	struct vnode *vp;
867 	int error, allerror = 0;
868 
869 	/*
870 	 * Force stale buffer cache information to be flushed.
871 	 */
872 loop:
873 	for (vp = mp->mnt_vnodelist.lh_first;
874 	     vp != NULL;
875 	     vp = vp->v_mntvnodes.le_next) {
876 		/*
877 		 * If the vnode that we are about to sync is no longer
878 		 * associated with this mount point, start over.
879 		 */
880 		if (vp->v_mount != mp)
881 			goto loop;
882 		if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) ||
883 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
884 		     vp->v_uvm.u_obj.uo_npages == 0))
885 			continue;
886 		if (vget(vp, LK_EXCLUSIVE))
887 			goto loop;
888 		error = VOP_FSYNC(vp, cred,
889 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p);
890 		if (error)
891 			allerror = error;
892 		vput(vp);
893 	}
894 	return (allerror);
895 }
896 
897 /*
898  * NFS flat namespace lookup.
899  * Currently unsupported.
900  */
901 /* ARGSUSED */
902 int
903 nfs_vget(mp, ino, vpp)
904 	struct mount *mp;
905 	ino_t ino;
906 	struct vnode **vpp;
907 {
908 
909 	return (EOPNOTSUPP);
910 }
911 
912 /*
913  * Do that sysctl thang...
914  */
915 int
916 nfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
917 	int *name;
918 	u_int namelen;
919 	void *oldp;
920 	size_t *oldlenp;
921 	void *newp;
922 	size_t newlen;
923 	struct proc *p;
924 {
925 	int rv;
926 
927 	/*
928 	 * All names at this level are terminal.
929 	 */
930 	if(namelen > 1)
931 		return ENOTDIR;	/* overloaded */
932 
933 	switch(name[0]) {
934 	case NFS_NFSSTATS:
935 		if(!oldp) {
936 			*oldlenp = sizeof nfsstats;
937 			return 0;
938 		}
939 
940 		if(*oldlenp < sizeof nfsstats) {
941 			*oldlenp = sizeof nfsstats;
942 			return ENOMEM;
943 		}
944 
945 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
946 		if(rv) return rv;
947 
948 		if(newp && newlen != sizeof nfsstats)
949 			return EINVAL;
950 
951 		if(newp) {
952 			return copyin(newp, &nfsstats, sizeof nfsstats);
953 		}
954 		return 0;
955 
956 	case NFS_IOTHREADS:
957 		nfs_getset_niothreads(0);
958 
959 		rv = (sysctl_int(oldp, oldlenp, newp, newlen,
960 		    &nfs_niothreads));
961 
962 		if (newp)
963 			nfs_getset_niothreads(1);
964 
965 		return rv;
966 
967 	default:
968 		return EOPNOTSUPP;
969 	}
970 }
971 
972 
973 /*
974  * At this point, this should never happen
975  */
976 /* ARGSUSED */
977 int
978 nfs_fhtovp(mp, fhp, vpp)
979 	struct mount *mp;
980 	struct fid *fhp;
981 	struct vnode **vpp;
982 {
983 
984 	return (EINVAL);
985 }
986 
987 /* ARGSUSED */
988 int
989 nfs_checkexp(mp, nam, exflagsp, credanonp)
990 	struct mount *mp;
991 	struct mbuf *nam;
992 	int *exflagsp;
993 	struct ucred **credanonp;
994 {
995 
996 	return (EINVAL);
997 }
998 
999 /*
1000  * Vnode pointer to File handle, should never happen either
1001  */
1002 /* ARGSUSED */
1003 int
1004 nfs_vptofh(vp, fhp)
1005 	struct vnode *vp;
1006 	struct fid *fhp;
1007 {
1008 
1009 	return (EINVAL);
1010 }
1011 
1012 /*
1013  * Vfs start routine, a no-op.
1014  */
1015 /* ARGSUSED */
1016 int
1017 nfs_start(mp, flags, p)
1018 	struct mount *mp;
1019 	int flags;
1020 	struct proc *p;
1021 {
1022 
1023 	return (0);
1024 }
1025 
1026 /*
1027  * Do operations associated with quotas, not supported
1028  */
1029 /* ARGSUSED */
1030 int
1031 nfs_quotactl(mp, cmd, uid, arg, p)
1032 	struct mount *mp;
1033 	int cmd;
1034 	uid_t uid;
1035 	caddr_t arg;
1036 	struct proc *p;
1037 {
1038 
1039 	return (EOPNOTSUPP);
1040 }
1041