xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision 27578b9aac214cc7796ead81dcc5427e79d5f2a0)
1 /*	$NetBSD: nfs_vfsops.c,v 1.109 2001/09/20 08:22:04 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_OPT)
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 const struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
88 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
89 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
90 
91 const struct vnodeopv_desc * const 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_reinit,
112 	nfs_vfs_done,
113 	nfs_sysctl,
114 	nfs_mountroot,
115 	nfs_checkexp,
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 #define TRUE	1
126 #define	FALSE	0
127 
128 /*
129  * nfs statfs call
130  */
131 int
132 nfs_statfs(mp, sbp, p)
133 	struct mount *mp;
134 	struct statfs *sbp;
135 	struct proc *p;
136 {
137 	struct vnode *vp;
138 	struct nfs_statfs *sfp;
139 	caddr_t cp;
140 	u_int32_t *tl;
141 	int32_t t1, t2;
142 	caddr_t bpos, dpos, cp2;
143 	struct nfsmount *nmp = VFSTONFS(mp);
144 	int error = 0, retattr;
145 #ifdef NFS_V2_ONLY
146 	const int v3 = 0;
147 #else
148 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
149 #endif
150 	struct mbuf *mreq, *mrep = NULL, *md, *mb, *mb2;
151 	struct ucred *cred;
152 	u_quad_t tquad;
153 
154 #ifndef nolint
155 	sfp = (struct nfs_statfs *)0;
156 #endif
157 	vp = nmp->nm_vnode;
158 	cred = crget();
159 	cred->cr_ngroups = 0;
160 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
161 		(void)nfs_fsinfo(nmp, vp, cred, p);
162 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
163 	nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
164 	nfsm_fhtom(vp, v3);
165 	nfsm_request(vp, NFSPROC_FSSTAT, p, cred);
166 	if (v3)
167 		nfsm_postop_attr(vp, retattr);
168 	if (error) {
169 		if (mrep != NULL)
170 			m_free(mrep);
171 		goto nfsmout;
172 	}
173 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
174 #ifdef COMPAT_09
175 	sbp->f_type = 2;
176 #else
177 	sbp->f_type = 0;
178 #endif
179 	sbp->f_flags = nmp->nm_flag;
180 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
181 	if (v3) {
182 		sbp->f_bsize = NFS_FABLKSIZE;
183 		tquad = fxdr_hyper(&sfp->sf_tbytes);
184 		sbp->f_blocks = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
185 		tquad = fxdr_hyper(&sfp->sf_fbytes);
186 		sbp->f_bfree = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
187 		tquad = fxdr_hyper(&sfp->sf_abytes);
188 		sbp->f_bavail = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
189 		tquad = fxdr_hyper(&sfp->sf_tfiles);
190 		sbp->f_files = (long)tquad;
191 		tquad = fxdr_hyper(&sfp->sf_ffiles);
192 		sbp->f_ffree = (long)tquad;
193 	} else {
194 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
195 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
196 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
197 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
198 		sbp->f_files = 0;
199 		sbp->f_ffree = 0;
200 	}
201 	if (sbp != &mp->mnt_stat) {
202 		memcpy(sbp->f_mntonname, mp->mnt_stat.f_mntonname, MNAMELEN);
203 		memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, MNAMELEN);
204 	}
205 	strncpy(&sbp->f_fstypename[0], mp->mnt_op->vfs_name, MFSNAMELEN);
206 	nfsm_reqdone;
207 	crfree(cred);
208 	return (error);
209 }
210 
211 #ifndef NFS_V2_ONLY
212 /*
213  * nfs version 3 fsinfo rpc call
214  */
215 int
216 nfs_fsinfo(nmp, vp, cred, p)
217 	struct nfsmount *nmp;
218 	struct vnode *vp;
219 	struct ucred *cred;
220 	struct proc *p;
221 {
222 	struct nfsv3_fsinfo *fsp;
223 	caddr_t cp;
224 	int32_t t1, t2;
225 	u_int32_t *tl, pref, max;
226 	caddr_t bpos, dpos, cp2;
227 	int error = 0, retattr;
228 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
229 	u_int64_t maxfsize;
230 
231 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
232 	nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
233 	nfsm_fhtom(vp, 1);
234 	nfsm_request(vp, NFSPROC_FSINFO, p, cred);
235 	nfsm_postop_attr(vp, retattr);
236 	if (!error) {
237 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
238 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
239 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
240 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
241 				~(NFS_FABLKSIZE - 1);
242 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
243 		if (max < nmp->nm_wsize && max > 0) {
244 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
245 			if (nmp->nm_wsize == 0)
246 				nmp->nm_wsize = max;
247 		}
248 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
249 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
250 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
251 				~(NFS_FABLKSIZE - 1);
252 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
253 		if (max < nmp->nm_rsize && max > 0) {
254 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
255 			if (nmp->nm_rsize == 0)
256 				nmp->nm_rsize = max;
257 		}
258 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
259 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
260 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
261 				~(NFS_DIRFRAGSIZ - 1);
262 		if (max < nmp->nm_readdirsize && max > 0) {
263 			nmp->nm_readdirsize = max & ~(NFS_DIRFRAGSIZ - 1);
264 			if (nmp->nm_readdirsize == 0)
265 				nmp->nm_readdirsize = max;
266 		}
267 		/* XXX */
268 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
269 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
270 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
271 			nmp->nm_maxfilesize = maxfsize;
272 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
273 	}
274 	nfsm_reqdone;
275 	return (error);
276 }
277 #endif
278 
279 /*
280  * Mount a remote root fs via. NFS.  It goes like this:
281  * - Call nfs_boot_init() to fill in the nfs_diskless struct
282  * - build the rootfs mount point and call mountnfs() to do the rest.
283  */
284 int
285 nfs_mountroot()
286 {
287 	struct nfs_diskless *nd;
288 	struct vattr attr;
289 	struct mount *mp;
290 	struct vnode *vp;
291 	struct proc *procp;
292 	long n;
293 	int error;
294 
295 	procp = curproc; /* XXX */
296 
297 	if (root_device->dv_class != DV_IFNET)
298 		return (ENODEV);
299 
300 	/*
301 	 * XXX time must be non-zero when we init the interface or else
302 	 * the arp code will wedge.  [Fixed now in if_ether.c]
303 	 * However, the NFS attribute cache gives false "hits" when
304 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
305 	 */
306 	if (time.tv_sec < NFS_MAXATTRTIMO)
307 		time.tv_sec = NFS_MAXATTRTIMO;
308 
309 	/*
310 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
311 	 * Side effect:  Finds and configures a network interface.
312 	 */
313 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
314 	memset((caddr_t)nd, 0, sizeof(*nd));
315 	error = nfs_boot_init(nd, procp);
316 	if (error) {
317 		free(nd, M_NFSMNT);
318 		return (error);
319 	}
320 
321 	/*
322 	 * Create the root mount point.
323 	 */
324 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp);
325 	if (error)
326 		goto out;
327 	printf("root on %s\n", nd->nd_root.ndm_host);
328 
329 	/*
330 	 * Link it into the mount list.
331 	 */
332 	simple_lock(&mountlist_slock);
333 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
334 	simple_unlock(&mountlist_slock);
335 	rootvp = vp;
336 	mp->mnt_vnodecovered = NULLVP;
337 	vfs_unbusy(mp);
338 
339 	/* Get root attributes (for the time). */
340 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
341 	if (error)
342 		panic("nfs_mountroot: getattr for root");
343 	n = attr.va_atime.tv_sec;
344 #ifdef	DEBUG
345 	printf("root time: 0x%lx\n", n);
346 #endif
347 	inittodr(n);
348 
349 out:
350 	if (error)
351 		nfs_boot_cleanup(nd, procp);
352 	free(nd, M_NFSMNT);
353 	return (error);
354 }
355 
356 /*
357  * Internal version of mount system call for diskless setup.
358  * Separate function because we used to call it twice.
359  * (once for root and once for swap)
360  */
361 static int
362 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p)
363 	struct nfs_dlmount *ndmntp;
364 	const char *mntname;	/* mount point name */
365 	struct mount **mpp;
366 	struct vnode **vpp;
367 	struct proc *p;
368 {
369 	struct mount *mp;
370 	struct mbuf *m;
371 	int error;
372 
373 	vfs_rootmountalloc(MOUNT_NFS, (char *)mntname, &mp);
374 
375 	mp->mnt_op = &nfs_vfsops;
376 
377 	/*
378 	 * Historical practice expects NFS root file systems to
379 	 * be initially mounted r/w.
380 	 */
381 	mp->mnt_flag &= ~MNT_RDONLY;
382 
383 	/* Get mbuf for server sockaddr. */
384 	m = m_get(M_WAIT, MT_SONAME);
385 	if (m == NULL)
386 		panic("nfs_mountroot: mget soname for %s", mntname);
387 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
388 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
389 
390 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
391 			 ndmntp->ndm_args.hostname, vpp, p);
392 	if (error) {
393 		mp->mnt_op->vfs_refcount--;
394 		vfs_unbusy(mp);
395 		printf("nfs_mountroot: mount %s failed: %d\n",
396 		       mntname, error);
397 		free(mp, M_MOUNT);
398 	} else
399 		*mpp = mp;
400 
401 	return (error);
402 }
403 
404 void
405 nfs_decode_args(nmp, argp)
406 	struct nfsmount *nmp;
407 	struct nfs_args *argp;
408 {
409 	int s;
410 	int adjsock;
411 	int maxio;
412 
413 	s = splsoftnet();
414 
415 	/*
416 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
417 	 * no sense in that context.
418 	 */
419 	if (argp->sotype == SOCK_STREAM)
420 		argp->flags &= ~NFSMNT_NOCONN;
421 
422 	/*
423 	 * Cookie translation is not needed for v2, silently ignore it.
424 	 */
425 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
426 	    NFSMNT_XLATECOOKIE)
427 		argp->flags &= ~NFSMNT_XLATECOOKIE;
428 
429 	/* Re-bind if rsrvd port requested and wasn't on one */
430 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
431 		  && (argp->flags & NFSMNT_RESVPORT);
432 	/* Also re-bind if we're switching to/from a connected UDP socket */
433 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
434 		    (argp->flags & NFSMNT_NOCONN));
435 
436 	/* Update flags. */
437 	nmp->nm_flag = argp->flags;
438 	splx(s);
439 
440 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
441 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
442 		if (nmp->nm_timeo < NFS_MINTIMEO)
443 			nmp->nm_timeo = NFS_MINTIMEO;
444 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
445 			nmp->nm_timeo = NFS_MAXTIMEO;
446 	}
447 
448 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
449 		nmp->nm_retry = argp->retrans;
450 		if (nmp->nm_retry > NFS_MAXREXMIT)
451 			nmp->nm_retry = NFS_MAXREXMIT;
452 	}
453 
454 #ifndef NFS_V2_ONLY
455 	if (argp->flags & NFSMNT_NFSV3) {
456 		if (argp->sotype == SOCK_DGRAM)
457 			maxio = NFS_MAXDGRAMDATA;
458 		else
459 			maxio = NFS_MAXDATA;
460 	} else
461 #endif
462 		maxio = NFS_V2MAXDATA;
463 
464 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
465 		int osize = nmp->nm_wsize;
466 		nmp->nm_wsize = argp->wsize;
467 		/* Round down to multiple of blocksize */
468 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
469 		if (nmp->nm_wsize <= 0)
470 			nmp->nm_wsize = NFS_FABLKSIZE;
471 		adjsock |= (nmp->nm_wsize != osize);
472 	}
473 	if (nmp->nm_wsize > maxio)
474 		nmp->nm_wsize = maxio;
475 	if (nmp->nm_wsize > MAXBSIZE)
476 		nmp->nm_wsize = MAXBSIZE;
477 
478 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
479 		int osize = nmp->nm_rsize;
480 		nmp->nm_rsize = argp->rsize;
481 		/* Round down to multiple of blocksize */
482 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
483 		if (nmp->nm_rsize <= 0)
484 			nmp->nm_rsize = NFS_FABLKSIZE;
485 		adjsock |= (nmp->nm_rsize != osize);
486 	}
487 	if (nmp->nm_rsize > maxio)
488 		nmp->nm_rsize = maxio;
489 	if (nmp->nm_rsize > MAXBSIZE)
490 		nmp->nm_rsize = MAXBSIZE;
491 
492 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
493 		nmp->nm_readdirsize = argp->readdirsize;
494 		/* Round down to multiple of minimum blocksize */
495 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
496 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
497 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
498 		/* Bigger than buffer size makes no sense */
499 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
500 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
501 	} else if (argp->flags & NFSMNT_RSIZE)
502 		nmp->nm_readdirsize = nmp->nm_rsize;
503 
504 	if (nmp->nm_readdirsize > maxio)
505 		nmp->nm_readdirsize = maxio;
506 
507 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
508 		argp->maxgrouplist <= NFS_MAXGRPS)
509 		nmp->nm_numgrps = argp->maxgrouplist;
510 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
511 		argp->readahead <= NFS_MAXRAHEAD)
512 		nmp->nm_readahead = argp->readahead;
513 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
514 		argp->leaseterm <= NQ_MAXLEASE)
515 		nmp->nm_leaseterm = argp->leaseterm;
516 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
517 		argp->deadthresh <= NQ_NEVERDEAD)
518 		nmp->nm_deadthresh = argp->deadthresh;
519 
520 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
521 		    (nmp->nm_soproto != argp->proto));
522 	nmp->nm_sotype = argp->sotype;
523 	nmp->nm_soproto = argp->proto;
524 
525 	if (nmp->nm_so && adjsock) {
526 		nfs_safedisconnect(nmp);
527 		if (nmp->nm_sotype == SOCK_DGRAM)
528 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
529 				printf("nfs_args: retrying connect\n");
530 				(void) tsleep((caddr_t)&lbolt,
531 					      PSOCK, "nfscn3", 0);
532 			}
533 	}
534 }
535 
536 /*
537  * VFS Operations.
538  *
539  * mount system call
540  * It seems a bit dumb to copyinstr() the host and path here and then
541  * memcpy() them in mountnfs(), but I wanted to detect errors before
542  * doing the sockargs() call because sockargs() allocates an mbuf and
543  * an error after that means that I have to release the mbuf.
544  */
545 /* ARGSUSED */
546 int
547 nfs_mount(mp, path, data, ndp, p)
548 	struct mount *mp;
549 	const char *path;
550 	void *data;
551 	struct nameidata *ndp;
552 	struct proc *p;
553 {
554 	int error;
555 	struct nfs_args args;
556 	struct mbuf *nam;
557 	struct vnode *vp;
558 	char pth[MNAMELEN], hst[MNAMELEN];
559 	size_t len;
560 	u_char nfh[NFSX_V3FHMAX];
561 
562 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
563 	if (error)
564 		return (error);
565 	if (args.version != NFS_ARGSVERSION)
566 		return (EPROGMISMATCH);
567 #ifdef NFS_V2_ONLY
568 	if (args.flags & NFSMNT_NQNFS)
569 		return (EPROGUNAVAIL);
570 	if (args.flags & NFSMNT_NFSV3)
571 		return (EPROGMISMATCH);
572 #endif
573 	if (mp->mnt_flag & MNT_UPDATE) {
574 		struct nfsmount *nmp = VFSTONFS(mp);
575 
576 		if (nmp == NULL)
577 			return (EIO);
578 		/*
579 		 * When doing an update, we can't change from or to
580 		 * v3 and/or nqnfs, or change cookie translation
581 		 */
582 		args.flags = (args.flags &
583 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
584 		    (nmp->nm_flag &
585 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
586 		nfs_decode_args(nmp, &args);
587 		return (0);
588 	}
589 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
590 		return (EINVAL);
591 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
592 	if (error)
593 		return (error);
594 	error = copyinstr(path, pth, MNAMELEN-1, &len);
595 	if (error)
596 		return (error);
597 	memset(&pth[len], 0, MNAMELEN - len);
598 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
599 	if (error)
600 		return (error);
601 	memset(&hst[len], 0, MNAMELEN - len);
602 	/* sockargs() call must be after above copyin() calls */
603 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
604 	if (error)
605 		return (error);
606 	args.fh = nfh;
607 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
608 	return (error);
609 }
610 
611 /*
612  * Common code for mount and mountroot
613  */
614 int
615 mountnfs(argp, mp, nam, pth, hst, vpp, p)
616 	struct nfs_args *argp;
617 	struct mount *mp;
618 	struct mbuf *nam;
619 	const char *pth, *hst;
620 	struct vnode **vpp;
621 	struct proc *p;
622 {
623 	struct nfsmount *nmp;
624 	struct nfsnode *np;
625 	int error;
626 	struct vattr attrs;
627 	struct ucred *cr;
628 
629 	/*
630 	 * If the number of nfs iothreads to use has never
631 	 * been set, create a reasonable number of them.
632 	 */
633 
634 	if (nfs_niothreads < 0) {
635 		nfs_niothreads = NFS_DEFAULT_NIOTHREADS;
636 		nfs_getset_niothreads(TRUE);
637 	}
638 
639 	if (mp->mnt_flag & MNT_UPDATE) {
640 		nmp = VFSTONFS(mp);
641 		/* update paths, file handles, etc, here	XXX */
642 		m_freem(nam);
643 		return (0);
644 	} else {
645 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
646 		    M_NFSMNT, M_WAITOK);
647 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
648 		mp->mnt_data = (qaddr_t)nmp;
649 		TAILQ_INIT(&nmp->nm_uidlruhead);
650 		TAILQ_INIT(&nmp->nm_bufq);
651 	}
652 	vfs_getnewfsid(mp);
653 	nmp->nm_mountp = mp;
654 
655 #ifndef NFS_V2_ONLY
656 	if (argp->flags & NFSMNT_NQNFS)
657 		/*
658 		 * We have to set mnt_maxsymlink to a non-zero value so
659 		 * that COMPAT_43 routines will know that we are setting
660 		 * the d_type field in directories (and can zero it for
661 		 * unsuspecting binaries).
662 		 */
663 		mp->mnt_maxsymlinklen = 1;
664 #endif
665 
666 #ifndef NFS_V2_ONLY
667 	if ((argp->flags & NFSMNT_NFSV3) == 0)
668 #endif
669 		/*
670 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
671 		 * will fill this in.
672 		 */
673 		nmp->nm_maxfilesize = 0xffffffffLL;
674 
675 	nmp->nm_timeo = NFS_TIMEO;
676 	nmp->nm_retry = NFS_RETRANS;
677 	nmp->nm_wsize = NFS_WSIZE;
678 	nmp->nm_rsize = NFS_RSIZE;
679 	nmp->nm_readdirsize = NFS_READDIRSIZE;
680 	nmp->nm_numgrps = NFS_MAXGRPS;
681 	nmp->nm_readahead = NFS_DEFRAHEAD;
682 	nmp->nm_leaseterm = NQ_DEFLEASE;
683 	nmp->nm_deadthresh = NQ_DEADTHRESH;
684 	CIRCLEQ_INIT(&nmp->nm_timerhead);
685 	nmp->nm_inprog = NULLVP;
686 #ifdef COMPAT_09
687 	mp->mnt_stat.f_type = 2;
688 #else
689 	mp->mnt_stat.f_type = 0;
690 #endif
691 	strncpy(&mp->mnt_stat.f_fstypename[0], mp->mnt_op->vfs_name,
692 	    MFSNAMELEN);
693 	memcpy(mp->mnt_stat.f_mntfromname, hst, MNAMELEN);
694 	memcpy(mp->mnt_stat.f_mntonname, pth, MNAMELEN);
695 	nmp->nm_nam = nam;
696 
697 	/* Set up the sockets and per-host congestion */
698 	nmp->nm_sotype = argp->sotype;
699 	nmp->nm_soproto = argp->proto;
700 
701 	nfs_decode_args(nmp, argp);
702 
703 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
704 	mp->mnt_dev_bshift = DEV_BSHIFT;
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 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
722 	if (error)
723 		goto bad;
724 	*vpp = NFSTOV(np);
725 	VOP_GETATTR(*vpp, &attrs, p->p_ucred, p);
726 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
727 		cr = crget();
728 		cr->cr_uid = attrs.va_uid;
729 		cr->cr_gid = attrs.va_gid;
730 		cr->cr_ngroups = 0;
731 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
732 		crfree(cr);
733 	}
734 
735 	/*
736 	 * A reference count is needed on the nfsnode representing the
737 	 * remote root.  If this object is not persistent, then backward
738 	 * traversals of the mount point (i.e. "..") will not work if
739 	 * the nfsnode gets flushed out of the cache. Ufs does not have
740 	 * this problem, because one can identify root inodes by their
741 	 * number == ROOTINO (2). So, just unlock, but no rele.
742 	 */
743 
744 	nmp->nm_vnode = *vpp;
745 	VOP_UNLOCK(*vpp, 0);
746 
747 	return (0);
748 bad:
749 	nfs_disconnect(nmp);
750 	free((caddr_t)nmp, M_NFSMNT);
751 	m_freem(nam);
752 	return (error);
753 }
754 
755 /*
756  * unmount system call
757  */
758 int
759 nfs_unmount(mp, mntflags, p)
760 	struct mount *mp;
761 	int mntflags;
762 	struct proc *p;
763 {
764 	struct nfsmount *nmp;
765 	struct vnode *vp;
766 	int error, flags = 0;
767 
768 	if (mntflags & MNT_FORCE)
769 		flags |= FORCECLOSE;
770 	nmp = VFSTONFS(mp);
771 	/*
772 	 * Goes something like this..
773 	 * - Check for activity on the root vnode (other than ourselves).
774 	 * - Call vflush() to clear out vnodes for this file system,
775 	 *   except for the root vnode.
776 	 * - Decrement reference on the vnode representing remote root.
777 	 * - Close the socket
778 	 * - Free up the data structures
779 	 */
780 	/*
781 	 * We need to decrement the ref. count on the nfsnode representing
782 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
783 	 * has done vput on this vnode, otherwise we would get deadlock!
784 	 */
785 	vp = nmp->nm_vnode;
786 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
787 	if (error != 0)
788 		return error;
789 
790 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
791 		vput(vp);
792 		return (EBUSY);
793 	}
794 
795 	/*
796 	 * Must handshake with nqnfs_clientd() if it is active.
797 	 */
798 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
799 	while (nmp->nm_inprog != NULLVP)
800 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
801 	error = vflush(mp, vp, flags);
802 	if (error) {
803 		vput(vp);
804 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
805 		return (error);
806 	}
807 
808 	/*
809 	 * We are now committed to the unmount; mark the mount structure
810 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
811 	 * will go away cleanly.
812 	 */
813 	nmp->nm_iflag |= NFSMNT_DISMNT;
814 
815 	/*
816 	 * There are two reference counts to get rid of here
817 	 * (see comment in mountnfs()).
818 	 */
819 	vrele(vp);
820 	vput(vp);
821 	vgone(vp);
822 	nfs_disconnect(nmp);
823 	m_freem(nmp->nm_nam);
824 
825 	/*
826 	 * For NQNFS, let the server daemon free the nfsmount structure.
827 	 */
828 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
829 		free((caddr_t)nmp, M_NFSMNT);
830 	return (0);
831 }
832 
833 /*
834  * Return root of a filesystem
835  */
836 int
837 nfs_root(mp, vpp)
838 	struct mount *mp;
839 	struct vnode **vpp;
840 {
841 	struct vnode *vp;
842 	struct nfsmount *nmp;
843 	int error;
844 
845 	nmp = VFSTONFS(mp);
846 	vp = nmp->nm_vnode;
847 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
848 	if (error != 0)
849 		return error;
850 	if (vp->v_type == VNON)
851 		vp->v_type = VDIR;
852 	vp->v_flag = VROOT;
853 	*vpp = vp;
854 	return (0);
855 }
856 
857 extern int syncprt;
858 
859 /*
860  * Flush out the buffer cache
861  */
862 /* ARGSUSED */
863 int
864 nfs_sync(mp, waitfor, cred, p)
865 	struct mount *mp;
866 	int waitfor;
867 	struct ucred *cred;
868 	struct proc *p;
869 {
870 	struct vnode *vp;
871 	int error, allerror = 0;
872 
873 	/*
874 	 * Force stale buffer cache information to be flushed.
875 	 */
876 loop:
877 	for (vp = mp->mnt_vnodelist.lh_first;
878 	     vp != NULL;
879 	     vp = vp->v_mntvnodes.le_next) {
880 		/*
881 		 * If the vnode that we are about to sync is no longer
882 		 * associated with this mount point, start over.
883 		 */
884 		if (vp->v_mount != mp)
885 			goto loop;
886 		if (waitfor == MNT_LAZY ||
887 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
888 		     vp->v_uobj.uo_npages == 0))
889 			continue;
890 		if (vget(vp, LK_EXCLUSIVE | LK_NOWAIT))
891 			goto loop;
892 		error = VOP_FSYNC(vp, cred,
893 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p);
894 		if (error)
895 			allerror = error;
896 		vput(vp);
897 	}
898 	return (allerror);
899 }
900 
901 /*
902  * NFS flat namespace lookup.
903  * Currently unsupported.
904  */
905 /* ARGSUSED */
906 int
907 nfs_vget(mp, ino, vpp)
908 	struct mount *mp;
909 	ino_t ino;
910 	struct vnode **vpp;
911 {
912 
913 	return (EOPNOTSUPP);
914 }
915 
916 /*
917  * Do that sysctl thang...
918  */
919 int
920 nfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
921 	int *name;
922 	u_int namelen;
923 	void *oldp;
924 	size_t *oldlenp;
925 	void *newp;
926 	size_t newlen;
927 	struct proc *p;
928 {
929 	int rv;
930 
931 	/*
932 	 * All names at this level are terminal.
933 	 */
934 	if(namelen > 1)
935 		return ENOTDIR;	/* overloaded */
936 
937 	switch(name[0]) {
938 	case NFS_NFSSTATS:
939 		if(!oldp) {
940 			*oldlenp = sizeof nfsstats;
941 			return 0;
942 		}
943 
944 		if(*oldlenp < sizeof nfsstats) {
945 			*oldlenp = sizeof nfsstats;
946 			return ENOMEM;
947 		}
948 
949 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
950 		if(rv) return rv;
951 
952 		if(newp && newlen != sizeof nfsstats)
953 			return EINVAL;
954 
955 		if(newp) {
956 			return copyin(newp, &nfsstats, sizeof nfsstats);
957 		}
958 		return 0;
959 
960 	case NFS_IOTHREADS:
961 		nfs_getset_niothreads(0);
962 
963 		rv = (sysctl_int(oldp, oldlenp, newp, newlen,
964 		    &nfs_niothreads));
965 
966 		if (newp)
967 			nfs_getset_niothreads(1);
968 
969 		return rv;
970 
971 	default:
972 		return EOPNOTSUPP;
973 	}
974 }
975 
976 
977 /*
978  * At this point, this should never happen
979  */
980 /* ARGSUSED */
981 int
982 nfs_fhtovp(mp, fhp, vpp)
983 	struct mount *mp;
984 	struct fid *fhp;
985 	struct vnode **vpp;
986 {
987 
988 	return (EINVAL);
989 }
990 
991 /* ARGSUSED */
992 int
993 nfs_checkexp(mp, nam, exflagsp, credanonp)
994 	struct mount *mp;
995 	struct mbuf *nam;
996 	int *exflagsp;
997 	struct ucred **credanonp;
998 {
999 
1000 	return (EINVAL);
1001 }
1002 
1003 /*
1004  * Vnode pointer to File handle, should never happen either
1005  */
1006 /* ARGSUSED */
1007 int
1008 nfs_vptofh(vp, fhp)
1009 	struct vnode *vp;
1010 	struct fid *fhp;
1011 {
1012 
1013 	return (EINVAL);
1014 }
1015 
1016 /*
1017  * Vfs start routine, a no-op.
1018  */
1019 /* ARGSUSED */
1020 int
1021 nfs_start(mp, flags, p)
1022 	struct mount *mp;
1023 	int flags;
1024 	struct proc *p;
1025 {
1026 
1027 	return (0);
1028 }
1029 
1030 /*
1031  * Do operations associated with quotas, not supported
1032  */
1033 /* ARGSUSED */
1034 int
1035 nfs_quotactl(mp, cmd, uid, arg, p)
1036 	struct mount *mp;
1037 	int cmd;
1038 	uid_t uid;
1039 	caddr_t arg;
1040 	struct proc *p;
1041 {
1042 
1043 	return (EOPNOTSUPP);
1044 }
1045