xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision b8c616269f5ebf18ab2e35cb8099d683130a177c)
1 /*	$NetBSD: nfs_vfsops.c,v 1.121 2003/02/01 06:23:49 thorpej 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 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: nfs_vfsops.c,v 1.121 2003/02/01 06:23:49 thorpej Exp $");
43 
44 #if defined(_KERNEL_OPT)
45 #include "opt_compat_netbsd.h"
46 #include "opt_nfs.h"
47 #endif
48 
49 #include <sys/param.h>
50 #include <sys/ioctl.h>
51 #include <sys/signal.h>
52 #include <sys/proc.h>
53 #include <sys/namei.h>
54 #include <sys/device.h>
55 #include <sys/vnode.h>
56 #include <sys/kernel.h>
57 #include <sys/mount.h>
58 #include <sys/buf.h>
59 #include <sys/mbuf.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <sys/sysctl.h>
63 #include <sys/systm.h>
64 
65 #include <net/if.h>
66 #include <net/route.h>
67 #include <netinet/in.h>
68 
69 #include <nfs/rpcv2.h>
70 #include <nfs/nfsproto.h>
71 #include <nfs/nfsnode.h>
72 #include <nfs/nfs.h>
73 #include <nfs/nfsmount.h>
74 #include <nfs/xdr_subs.h>
75 #include <nfs/nfsm_subs.h>
76 #include <nfs/nfsdiskless.h>
77 #include <nfs/nqnfs.h>
78 #include <nfs/nfs_var.h>
79 
80 extern struct nfsstats nfsstats;
81 extern int nfs_ticks;
82 
83 MALLOC_DEFINE(M_NFSMNT, "NFS mount", "NFS mount structure");
84 
85 int nfs_sysctl __P((int *, u_int, void *, size_t *, void *, size_t,
86 		      struct proc *));
87 
88 /*
89  * nfs vfs operations.
90  */
91 
92 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
93 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
94 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
95 
96 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = {
97 	&nfsv2_vnodeop_opv_desc,
98 	&spec_nfsv2nodeop_opv_desc,
99 	&fifo_nfsv2nodeop_opv_desc,
100 	NULL,
101 };
102 
103 struct vfsops nfs_vfsops = {
104 	MOUNT_NFS,
105 	nfs_mount,
106 	nfs_start,
107 	nfs_unmount,
108 	nfs_root,
109 	nfs_quotactl,
110 	nfs_statfs,
111 	nfs_sync,
112 	nfs_vget,
113 	nfs_fhtovp,
114 	nfs_vptofh,
115 	nfs_vfs_init,
116 	nfs_vfs_reinit,
117 	nfs_vfs_done,
118 	nfs_sysctl,
119 	nfs_mountroot,
120 	nfs_checkexp,
121 	nfs_vnodeopv_descs,
122 };
123 
124 extern u_int32_t nfs_procids[NFS_NPROCS];
125 extern u_int32_t nfs_prog, nfs_vers;
126 
127 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
128     struct mount **, struct vnode **, struct proc *));
129 
130 #define TRUE	1
131 #define	FALSE	0
132 
133 /*
134  * nfs statfs call
135  */
136 int
137 nfs_statfs(mp, sbp, p)
138 	struct mount *mp;
139 	struct statfs *sbp;
140 	struct proc *p;
141 {
142 	struct vnode *vp;
143 	struct nfs_statfs *sfp;
144 	caddr_t cp;
145 	u_int32_t *tl;
146 	int32_t t1, t2;
147 	caddr_t bpos, dpos, cp2;
148 	struct nfsmount *nmp = VFSTONFS(mp);
149 	int error = 0, retattr;
150 #ifdef NFS_V2_ONLY
151 	const int v3 = 0;
152 #else
153 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
154 #endif
155 	struct mbuf *mreq, *mrep = NULL, *md, *mb, *mb2;
156 	struct ucred *cred;
157 	u_quad_t tquad;
158 
159 #ifndef nolint
160 	sfp = (struct nfs_statfs *)0;
161 #endif
162 	vp = nmp->nm_vnode;
163 	cred = crget();
164 	cred->cr_ngroups = 0;
165 #ifndef NFS_V2_ONLY
166 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
167 		(void)nfs_fsinfo(nmp, vp, cred, p);
168 #endif
169 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
170 	nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
171 	nfsm_fhtom(vp, v3);
172 	nfsm_request(vp, NFSPROC_FSSTAT, p, cred);
173 	if (v3)
174 		nfsm_postop_attr(vp, retattr, 0);
175 	if (error) {
176 		if (mrep != NULL)
177 			m_free(mrep);
178 		goto nfsmout;
179 	}
180 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
181 #ifdef COMPAT_09
182 	sbp->f_type = 2;
183 #else
184 	sbp->f_type = 0;
185 #endif
186 	sbp->f_flags = nmp->nm_flag;
187 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
188 	if (v3) {
189 		sbp->f_bsize = NFS_FABLKSIZE;
190 		tquad = fxdr_hyper(&sfp->sf_tbytes);
191 		sbp->f_blocks = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
192 		tquad = fxdr_hyper(&sfp->sf_fbytes);
193 		sbp->f_bfree = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
194 		tquad = fxdr_hyper(&sfp->sf_abytes);
195 		sbp->f_bavail = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
196 		tquad = fxdr_hyper(&sfp->sf_tfiles);
197 		sbp->f_files = (long)tquad;
198 		tquad = fxdr_hyper(&sfp->sf_ffiles);
199 		sbp->f_ffree = (long)tquad;
200 	} else {
201 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
202 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
203 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
204 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
205 		sbp->f_files = 0;
206 		sbp->f_ffree = 0;
207 	}
208 	if (sbp != &mp->mnt_stat) {
209 		memcpy(sbp->f_mntonname, mp->mnt_stat.f_mntonname, MNAMELEN);
210 		memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, MNAMELEN);
211 	}
212 	strncpy(&sbp->f_fstypename[0], mp->mnt_op->vfs_name, MFSNAMELEN);
213 	nfsm_reqdone;
214 	crfree(cred);
215 	return (error);
216 }
217 
218 #ifndef NFS_V2_ONLY
219 /*
220  * nfs version 3 fsinfo rpc call
221  */
222 int
223 nfs_fsinfo(nmp, vp, cred, p)
224 	struct nfsmount *nmp;
225 	struct vnode *vp;
226 	struct ucred *cred;
227 	struct proc *p;
228 {
229 	struct nfsv3_fsinfo *fsp;
230 	caddr_t cp;
231 	int32_t t1, t2;
232 	u_int32_t *tl, pref, max;
233 	caddr_t bpos, dpos, cp2;
234 	int error = 0, retattr;
235 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
236 	u_int64_t maxfsize;
237 
238 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
239 	nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
240 	nfsm_fhtom(vp, 1);
241 	nfsm_request(vp, NFSPROC_FSINFO, p, cred);
242 	nfsm_postop_attr(vp, retattr, 0);
243 	if (!error) {
244 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
245 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
246 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
247 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
248 				~(NFS_FABLKSIZE - 1);
249 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
250 		if (max < nmp->nm_wsize && max > 0) {
251 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
252 			if (nmp->nm_wsize == 0)
253 				nmp->nm_wsize = max;
254 		}
255 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
256 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
257 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
258 				~(NFS_FABLKSIZE - 1);
259 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
260 		if (max < nmp->nm_rsize && max > 0) {
261 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
262 			if (nmp->nm_rsize == 0)
263 				nmp->nm_rsize = max;
264 		}
265 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
266 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
267 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
268 				~(NFS_DIRFRAGSIZ - 1);
269 		if (max < nmp->nm_readdirsize && max > 0) {
270 			nmp->nm_readdirsize = max & ~(NFS_DIRFRAGSIZ - 1);
271 			if (nmp->nm_readdirsize == 0)
272 				nmp->nm_readdirsize = max;
273 		}
274 		/* XXX */
275 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
276 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
277 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
278 			nmp->nm_maxfilesize = maxfsize;
279 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
280 	}
281 	nfsm_reqdone;
282 	return (error);
283 }
284 #endif
285 
286 /*
287  * Mount a remote root fs via. NFS.  It goes like this:
288  * - Call nfs_boot_init() to fill in the nfs_diskless struct
289  * - build the rootfs mount point and call mountnfs() to do the rest.
290  */
291 int
292 nfs_mountroot()
293 {
294 	struct nfs_diskless *nd;
295 	struct vattr attr;
296 	struct mount *mp;
297 	struct vnode *vp;
298 	struct proc *procp;
299 	long n;
300 	int error;
301 
302 	procp = curproc; /* XXX */
303 
304 	if (root_device->dv_class != DV_IFNET)
305 		return (ENODEV);
306 
307 	/*
308 	 * XXX time must be non-zero when we init the interface or else
309 	 * the arp code will wedge.  [Fixed now in if_ether.c]
310 	 * However, the NFS attribute cache gives false "hits" when
311 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
312 	 */
313 	if (time.tv_sec < NFS_MAXATTRTIMO)
314 		time.tv_sec = NFS_MAXATTRTIMO;
315 
316 	/*
317 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
318 	 * Side effect:  Finds and configures a network interface.
319 	 */
320 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
321 	memset((caddr_t)nd, 0, sizeof(*nd));
322 	error = nfs_boot_init(nd, procp);
323 	if (error) {
324 		free(nd, M_NFSMNT);
325 		return (error);
326 	}
327 
328 	/*
329 	 * Create the root mount point.
330 	 */
331 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp);
332 	if (error)
333 		goto out;
334 	printf("root on %s\n", nd->nd_root.ndm_host);
335 
336 	/*
337 	 * Link it into the mount list.
338 	 */
339 	simple_lock(&mountlist_slock);
340 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
341 	simple_unlock(&mountlist_slock);
342 	rootvp = vp;
343 	mp->mnt_vnodecovered = NULLVP;
344 	vfs_unbusy(mp);
345 
346 	/* Get root attributes (for the time). */
347 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
348 	if (error)
349 		panic("nfs_mountroot: getattr for root");
350 	n = attr.va_atime.tv_sec;
351 #ifdef	DEBUG
352 	printf("root time: 0x%lx\n", n);
353 #endif
354 	inittodr(n);
355 
356 out:
357 	if (error)
358 		nfs_boot_cleanup(nd, procp);
359 	free(nd, M_NFSMNT);
360 	return (error);
361 }
362 
363 /*
364  * Internal version of mount system call for diskless setup.
365  * Separate function because we used to call it twice.
366  * (once for root and once for swap)
367  */
368 static int
369 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p)
370 	struct nfs_dlmount *ndmntp;
371 	const char *mntname;	/* mount point name */
372 	struct mount **mpp;
373 	struct vnode **vpp;
374 	struct proc *p;
375 {
376 	struct mount *mp;
377 	struct mbuf *m;
378 	int error;
379 
380 	vfs_rootmountalloc(MOUNT_NFS, (char *)mntname, &mp);
381 
382 	mp->mnt_op = &nfs_vfsops;
383 
384 	/*
385 	 * Historical practice expects NFS root file systems to
386 	 * be initially mounted r/w.
387 	 */
388 	mp->mnt_flag &= ~MNT_RDONLY;
389 
390 	/* Get mbuf for server sockaddr. */
391 	m = m_get(M_WAIT, MT_SONAME);
392 	if (m == NULL)
393 		panic("nfs_mountroot: mget soname for %s", mntname);
394 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
395 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
396 
397 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
398 			 ndmntp->ndm_args.hostname, vpp, p);
399 	if (error) {
400 		mp->mnt_op->vfs_refcount--;
401 		vfs_unbusy(mp);
402 		printf("nfs_mountroot: mount %s failed: %d\n",
403 		       mntname, error);
404 		free(mp, M_MOUNT);
405 	} else
406 		*mpp = mp;
407 
408 	return (error);
409 }
410 
411 void
412 nfs_decode_args(nmp, argp)
413 	struct nfsmount *nmp;
414 	struct nfs_args *argp;
415 {
416 	int s;
417 	int adjsock;
418 	int maxio;
419 
420 	s = splsoftnet();
421 
422 	/*
423 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
424 	 * no sense in that context.
425 	 */
426 	if (argp->sotype == SOCK_STREAM)
427 		argp->flags &= ~NFSMNT_NOCONN;
428 
429 	/*
430 	 * Cookie translation is not needed for v2, silently ignore it.
431 	 */
432 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
433 	    NFSMNT_XLATECOOKIE)
434 		argp->flags &= ~NFSMNT_XLATECOOKIE;
435 
436 	/* Re-bind if rsrvd port requested and wasn't on one */
437 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
438 		  && (argp->flags & NFSMNT_RESVPORT);
439 	/* Also re-bind if we're switching to/from a connected UDP socket */
440 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
441 		    (argp->flags & NFSMNT_NOCONN));
442 
443 	/* Update flags. */
444 	nmp->nm_flag = argp->flags;
445 	splx(s);
446 
447 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
448 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
449 		if (nmp->nm_timeo < NFS_MINTIMEO)
450 			nmp->nm_timeo = NFS_MINTIMEO;
451 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
452 			nmp->nm_timeo = NFS_MAXTIMEO;
453 	}
454 
455 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
456 		nmp->nm_retry = argp->retrans;
457 		if (nmp->nm_retry > NFS_MAXREXMIT)
458 			nmp->nm_retry = NFS_MAXREXMIT;
459 	}
460 
461 #ifndef NFS_V2_ONLY
462 	if (argp->flags & NFSMNT_NFSV3) {
463 		if (argp->sotype == SOCK_DGRAM)
464 			maxio = NFS_MAXDGRAMDATA;
465 		else
466 			maxio = NFS_MAXDATA;
467 	} else
468 #endif
469 		maxio = NFS_V2MAXDATA;
470 
471 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
472 		int osize = nmp->nm_wsize;
473 		nmp->nm_wsize = argp->wsize;
474 		/* Round down to multiple of blocksize */
475 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
476 		if (nmp->nm_wsize <= 0)
477 			nmp->nm_wsize = NFS_FABLKSIZE;
478 		adjsock |= (nmp->nm_wsize != osize);
479 	}
480 	if (nmp->nm_wsize > maxio)
481 		nmp->nm_wsize = maxio;
482 	if (nmp->nm_wsize > MAXBSIZE)
483 		nmp->nm_wsize = MAXBSIZE;
484 
485 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
486 		int osize = nmp->nm_rsize;
487 		nmp->nm_rsize = argp->rsize;
488 		/* Round down to multiple of blocksize */
489 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
490 		if (nmp->nm_rsize <= 0)
491 			nmp->nm_rsize = NFS_FABLKSIZE;
492 		adjsock |= (nmp->nm_rsize != osize);
493 	}
494 	if (nmp->nm_rsize > maxio)
495 		nmp->nm_rsize = maxio;
496 	if (nmp->nm_rsize > MAXBSIZE)
497 		nmp->nm_rsize = MAXBSIZE;
498 
499 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
500 		nmp->nm_readdirsize = argp->readdirsize;
501 		/* Round down to multiple of minimum blocksize */
502 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
503 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
504 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
505 		/* Bigger than buffer size makes no sense */
506 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
507 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
508 	} else if (argp->flags & NFSMNT_RSIZE)
509 		nmp->nm_readdirsize = nmp->nm_rsize;
510 
511 	if (nmp->nm_readdirsize > maxio)
512 		nmp->nm_readdirsize = maxio;
513 
514 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
515 		argp->maxgrouplist <= NFS_MAXGRPS)
516 		nmp->nm_numgrps = argp->maxgrouplist;
517 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
518 		argp->readahead <= NFS_MAXRAHEAD)
519 		nmp->nm_readahead = argp->readahead;
520 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
521 		argp->leaseterm <= NQ_MAXLEASE)
522 		nmp->nm_leaseterm = argp->leaseterm;
523 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
524 		argp->deadthresh <= NQ_NEVERDEAD)
525 		nmp->nm_deadthresh = argp->deadthresh;
526 
527 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
528 		    (nmp->nm_soproto != argp->proto));
529 	nmp->nm_sotype = argp->sotype;
530 	nmp->nm_soproto = argp->proto;
531 
532 	if (nmp->nm_so && adjsock) {
533 		nfs_safedisconnect(nmp);
534 		if (nmp->nm_sotype == SOCK_DGRAM)
535 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
536 				printf("nfs_args: retrying connect\n");
537 				(void) tsleep((caddr_t)&lbolt,
538 					      PSOCK, "nfscn3", 0);
539 			}
540 	}
541 }
542 
543 /*
544  * VFS Operations.
545  *
546  * mount system call
547  * It seems a bit dumb to copyinstr() the host and path here and then
548  * memcpy() them 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(mp, path, data, ndp, p)
555 	struct mount *mp;
556 	const char *path;
557 	void *data;
558 	struct nameidata *ndp;
559 	struct proc *p;
560 {
561 	int error;
562 	struct nfs_args args;
563 	struct mbuf *nam;
564 	struct nfsmount *nmp = VFSTONFS(mp);
565 	struct sockaddr *sa;
566 	struct vnode *vp;
567 	char *pth, *hst;
568 	size_t len;
569 	u_char *nfh;
570 
571 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
572 	if (error)
573 		return (error);
574 
575 	if (mp->mnt_flag & MNT_GETARGS) {
576 
577 		if (nmp == NULL)
578 			return (EIO);
579 		if (args.addr != NULL) {
580 			sa = mtod(nmp->nm_nam, struct sockaddr *);
581 			error = copyout(sa, args.addr, sa->sa_len);
582 			if (error)
583 				return (error);
584 			args.addrlen = sa->sa_len;
585 		} else
586 			args.addrlen = 0;
587 
588 		args.version = NFS_ARGSVERSION;
589 		args.sotype = nmp->nm_sotype;
590 		args.proto = nmp->nm_soproto;
591 		args.fh = NULL;
592 		args.fhsize = 0;
593 		args.flags = nmp->nm_flag;
594 		args.wsize = nmp->nm_wsize;
595 		args.rsize = nmp->nm_rsize;
596 		args.readdirsize = nmp->nm_readdirsize;
597 		args.timeo = nmp->nm_timeo;
598 		args.retrans = nmp->nm_retry;
599 		args.maxgrouplist = nmp->nm_numgrps;
600 		args.readahead = nmp->nm_readahead;
601 		args.leaseterm = nmp->nm_leaseterm;
602 		args.deadthresh = nmp->nm_deadthresh;
603 		args.hostname = NULL;
604 		return (copyout(&args, data, sizeof(args)));
605 	}
606 
607 	if (args.version != NFS_ARGSVERSION)
608 		return (EPROGMISMATCH);
609 #ifdef NFS_V2_ONLY
610 	if (args.flags & NFSMNT_NQNFS)
611 		return (EPROGUNAVAIL);
612 	if (args.flags & NFSMNT_NFSV3)
613 		return (EPROGMISMATCH);
614 #endif
615 	if (mp->mnt_flag & MNT_UPDATE) {
616 		if (nmp == NULL)
617 			return (EIO);
618 		/*
619 		 * When doing an update, we can't change from or to
620 		 * v3 and/or nqnfs, or change cookie translation
621 		 */
622 		args.flags = (args.flags &
623 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
624 		    (nmp->nm_flag &
625 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
626 		nfs_decode_args(nmp, &args);
627 		return (0);
628 	}
629 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
630 		return (EINVAL);
631 	MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK);
632 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
633 	if (error)
634 		return (error);
635 	MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK);
636 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
637 	if (error)
638 		goto free_nfh;
639 	memset(&pth[len], 0, MNAMELEN - len);
640 	MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK);
641 	error = copyinstr(args.hostname, hst, MNAMELEN - 1, &len);
642 	if (error)
643 		goto free_pth;
644 	memset(&hst[len], 0, MNAMELEN - len);
645 	/* sockargs() call must be after above copyin() calls */
646 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
647 	if (error)
648 		goto free_hst;
649 	args.fh = nfh;
650 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
651 
652 free_hst:
653 	FREE(hst, M_TEMP);
654 free_pth:
655 	FREE(pth, M_TEMP);
656 free_nfh:
657 	FREE(nfh, M_TEMP);
658 
659 	return (error);
660 }
661 
662 /*
663  * Common code for mount and mountroot
664  */
665 int
666 mountnfs(argp, mp, nam, pth, hst, vpp, p)
667 	struct nfs_args *argp;
668 	struct mount *mp;
669 	struct mbuf *nam;
670 	const char *pth, *hst;
671 	struct vnode **vpp;
672 	struct proc *p;
673 {
674 	struct nfsmount *nmp;
675 	struct nfsnode *np;
676 	int error;
677 	struct vattr *attrs;
678 	struct ucred *cr;
679 
680 	/*
681 	 * If the number of nfs iothreads to use has never
682 	 * been set, create a reasonable number of them.
683 	 */
684 
685 	if (nfs_niothreads < 0) {
686 		nfs_niothreads = NFS_DEFAULT_NIOTHREADS;
687 		nfs_getset_niothreads(TRUE);
688 	}
689 
690 	if (mp->mnt_flag & MNT_UPDATE) {
691 		nmp = VFSTONFS(mp);
692 		/* update paths, file handles, etc, here	XXX */
693 		m_freem(nam);
694 		return (0);
695 	} else {
696 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
697 		    M_NFSMNT, M_WAITOK);
698 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
699 		mp->mnt_data = nmp;
700 		TAILQ_INIT(&nmp->nm_uidlruhead);
701 		TAILQ_INIT(&nmp->nm_bufq);
702 	}
703 	vfs_getnewfsid(mp);
704 	nmp->nm_mountp = mp;
705 
706 #ifndef NFS_V2_ONLY
707 	if (argp->flags & NFSMNT_NQNFS)
708 		/*
709 		 * We have to set mnt_maxsymlink to a non-zero value so
710 		 * that COMPAT_43 routines will know that we are setting
711 		 * the d_type field in directories (and can zero it for
712 		 * unsuspecting binaries).
713 		 */
714 		mp->mnt_maxsymlinklen = 1;
715 #endif
716 
717 #ifndef NFS_V2_ONLY
718 	if ((argp->flags & NFSMNT_NFSV3) == 0)
719 #endif
720 		/*
721 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
722 		 * will fill this in.
723 		 */
724 		nmp->nm_maxfilesize = 0xffffffffLL;
725 
726 	nmp->nm_timeo = NFS_TIMEO;
727 	nmp->nm_retry = NFS_RETRANS;
728 	nmp->nm_wsize = NFS_WSIZE;
729 	nmp->nm_rsize = NFS_RSIZE;
730 	nmp->nm_readdirsize = NFS_READDIRSIZE;
731 	nmp->nm_numgrps = NFS_MAXGRPS;
732 	nmp->nm_readahead = NFS_DEFRAHEAD;
733 	nmp->nm_leaseterm = NQ_DEFLEASE;
734 	nmp->nm_deadthresh = NQ_DEADTHRESH;
735 	CIRCLEQ_INIT(&nmp->nm_timerhead);
736 	nmp->nm_inprog = NULLVP;
737 #ifdef COMPAT_09
738 	mp->mnt_stat.f_type = 2;
739 #else
740 	mp->mnt_stat.f_type = 0;
741 #endif
742 	strncpy(&mp->mnt_stat.f_fstypename[0], mp->mnt_op->vfs_name,
743 	    MFSNAMELEN);
744 	memcpy(mp->mnt_stat.f_mntfromname, hst, MNAMELEN);
745 	memcpy(mp->mnt_stat.f_mntonname, pth, MNAMELEN);
746 	nmp->nm_nam = nam;
747 
748 	/* Set up the sockets and per-host congestion */
749 	nmp->nm_sotype = argp->sotype;
750 	nmp->nm_soproto = argp->proto;
751 
752 	nfs_decode_args(nmp, argp);
753 
754 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
755 	mp->mnt_dev_bshift = DEV_BSHIFT;
756 
757 	/*
758 	 * For Connection based sockets (TCP,...) defer the connect until
759 	 * the first request, in case the server is not responding.
760 	 */
761 	if (nmp->nm_sotype == SOCK_DGRAM &&
762 		(error = nfs_connect(nmp, (struct nfsreq *)0)))
763 		goto bad;
764 
765 	/*
766 	 * This is silly, but it has to be set so that vinifod() works.
767 	 * We do not want to do an nfs_statfs() here since we can get
768 	 * stuck on a dead server and we are holding a lock on the mount
769 	 * point.
770 	 */
771 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
772 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
773 	if (error)
774 		goto bad;
775 	*vpp = NFSTOV(np);
776 	MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK);
777 	VOP_GETATTR(*vpp, attrs, p->p_ucred, p);
778 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
779 		cr = crget();
780 		cr->cr_uid = attrs->va_uid;
781 		cr->cr_gid = attrs->va_gid;
782 		cr->cr_ngroups = 0;
783 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
784 		crfree(cr);
785 	}
786 	FREE(attrs, M_TEMP);
787 
788 	/*
789 	 * A reference count is needed on the nfsnode representing the
790 	 * remote root.  If this object is not persistent, then backward
791 	 * traversals of the mount point (i.e. "..") will not work if
792 	 * the nfsnode gets flushed out of the cache. Ufs does not have
793 	 * this problem, because one can identify root inodes by their
794 	 * number == ROOTINO (2). So, just unlock, but no rele.
795 	 */
796 
797 	nmp->nm_vnode = *vpp;
798 	VOP_UNLOCK(*vpp, 0);
799 
800 	return (0);
801 bad:
802 	nfs_disconnect(nmp);
803 	free((caddr_t)nmp, M_NFSMNT);
804 	m_freem(nam);
805 	return (error);
806 }
807 
808 /*
809  * unmount system call
810  */
811 int
812 nfs_unmount(mp, mntflags, p)
813 	struct mount *mp;
814 	int mntflags;
815 	struct proc *p;
816 {
817 	struct nfsmount *nmp;
818 	struct vnode *vp;
819 	int error, flags = 0;
820 
821 	if (mntflags & MNT_FORCE)
822 		flags |= FORCECLOSE;
823 	nmp = VFSTONFS(mp);
824 	/*
825 	 * Goes something like this..
826 	 * - Check for activity on the root vnode (other than ourselves).
827 	 * - Call vflush() to clear out vnodes for this file system,
828 	 *   except for the root vnode.
829 	 * - Decrement reference on the vnode representing remote root.
830 	 * - Close the socket
831 	 * - Free up the data structures
832 	 */
833 	/*
834 	 * We need to decrement the ref. count on the nfsnode representing
835 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
836 	 * has done vput on this vnode, otherwise we would get deadlock!
837 	 */
838 	vp = nmp->nm_vnode;
839 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
840 	if (error != 0)
841 		return error;
842 
843 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
844 		vput(vp);
845 		return (EBUSY);
846 	}
847 
848 	/*
849 	 * Must handshake with nqnfs_clientd() if it is active.
850 	 */
851 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
852 	while (nmp->nm_inprog != NULLVP)
853 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
854 	error = vflush(mp, vp, flags);
855 	if (error) {
856 		vput(vp);
857 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
858 		return (error);
859 	}
860 
861 	/*
862 	 * We are now committed to the unmount; mark the mount structure
863 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
864 	 * will go away cleanly.
865 	 */
866 	nmp->nm_iflag |= NFSMNT_DISMNT;
867 
868 	/*
869 	 * There are two reference counts to get rid of here
870 	 * (see comment in mountnfs()).
871 	 */
872 	vrele(vp);
873 	vput(vp);
874 	vgone(vp);
875 	nfs_disconnect(nmp);
876 	m_freem(nmp->nm_nam);
877 
878 	/*
879 	 * For NQNFS, let the server daemon free the nfsmount structure.
880 	 */
881 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
882 		free((caddr_t)nmp, M_NFSMNT);
883 	return (0);
884 }
885 
886 /*
887  * Return root of a filesystem
888  */
889 int
890 nfs_root(mp, vpp)
891 	struct mount *mp;
892 	struct vnode **vpp;
893 {
894 	struct vnode *vp;
895 	struct nfsmount *nmp;
896 	int error;
897 
898 	nmp = VFSTONFS(mp);
899 	vp = nmp->nm_vnode;
900 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
901 	if (error != 0)
902 		return error;
903 	if (vp->v_type == VNON)
904 		vp->v_type = VDIR;
905 	vp->v_flag = VROOT;
906 	*vpp = vp;
907 	return (0);
908 }
909 
910 extern int syncprt;
911 
912 /*
913  * Flush out the buffer cache
914  */
915 /* ARGSUSED */
916 int
917 nfs_sync(mp, waitfor, cred, p)
918 	struct mount *mp;
919 	int waitfor;
920 	struct ucred *cred;
921 	struct proc *p;
922 {
923 	struct vnode *vp;
924 	int error, allerror = 0;
925 
926 	/*
927 	 * Force stale buffer cache information to be flushed.
928 	 */
929 loop:
930 	for (vp = mp->mnt_vnodelist.lh_first;
931 	     vp != NULL;
932 	     vp = vp->v_mntvnodes.le_next) {
933 		/*
934 		 * If the vnode that we are about to sync is no longer
935 		 * associated with this mount point, start over.
936 		 */
937 		if (vp->v_mount != mp)
938 			goto loop;
939 		if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) ||
940 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
941 		     vp->v_uobj.uo_npages == 0))
942 			continue;
943 		if (vget(vp, LK_EXCLUSIVE))
944 			goto loop;
945 		error = VOP_FSYNC(vp, cred,
946 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p);
947 		if (error)
948 			allerror = error;
949 		vput(vp);
950 	}
951 	return (allerror);
952 }
953 
954 /*
955  * NFS flat namespace lookup.
956  * Currently unsupported.
957  */
958 /* ARGSUSED */
959 int
960 nfs_vget(mp, ino, vpp)
961 	struct mount *mp;
962 	ino_t ino;
963 	struct vnode **vpp;
964 {
965 
966 	return (EOPNOTSUPP);
967 }
968 
969 /*
970  * Do that sysctl thang...
971  */
972 int
973 nfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
974 	int *name;
975 	u_int namelen;
976 	void *oldp;
977 	size_t *oldlenp;
978 	void *newp;
979 	size_t newlen;
980 	struct proc *p;
981 {
982 	int rv;
983 
984 	/*
985 	 * All names at this level are terminal.
986 	 */
987 	if(namelen > 1)
988 		return ENOTDIR;	/* overloaded */
989 
990 	switch(name[0]) {
991 	case NFS_NFSSTATS:
992 		if(!oldp) {
993 			*oldlenp = sizeof nfsstats;
994 			return 0;
995 		}
996 
997 		if(*oldlenp < sizeof nfsstats) {
998 			*oldlenp = sizeof nfsstats;
999 			return ENOMEM;
1000 		}
1001 
1002 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
1003 		if(rv) return rv;
1004 
1005 		if(newp && newlen != sizeof nfsstats)
1006 			return EINVAL;
1007 
1008 		if(newp) {
1009 			return copyin(newp, &nfsstats, sizeof nfsstats);
1010 		}
1011 		return 0;
1012 
1013 	case NFS_IOTHREADS:
1014 		nfs_getset_niothreads(0);
1015 
1016 		rv = (sysctl_int(oldp, oldlenp, newp, newlen,
1017 		    &nfs_niothreads));
1018 
1019 		if (newp)
1020 			nfs_getset_niothreads(1);
1021 
1022 		return rv;
1023 
1024 	default:
1025 		return EOPNOTSUPP;
1026 	}
1027 }
1028 
1029 
1030 /*
1031  * At this point, this should never happen
1032  */
1033 /* ARGSUSED */
1034 int
1035 nfs_fhtovp(mp, fhp, vpp)
1036 	struct mount *mp;
1037 	struct fid *fhp;
1038 	struct vnode **vpp;
1039 {
1040 
1041 	return (EINVAL);
1042 }
1043 
1044 /* ARGSUSED */
1045 int
1046 nfs_checkexp(mp, nam, exflagsp, credanonp)
1047 	struct mount *mp;
1048 	struct mbuf *nam;
1049 	int *exflagsp;
1050 	struct ucred **credanonp;
1051 {
1052 
1053 	return (EINVAL);
1054 }
1055 
1056 /*
1057  * Vnode pointer to File handle, should never happen either
1058  */
1059 /* ARGSUSED */
1060 int
1061 nfs_vptofh(vp, fhp)
1062 	struct vnode *vp;
1063 	struct fid *fhp;
1064 {
1065 
1066 	return (EINVAL);
1067 }
1068 
1069 /*
1070  * Vfs start routine, a no-op.
1071  */
1072 /* ARGSUSED */
1073 int
1074 nfs_start(mp, flags, p)
1075 	struct mount *mp;
1076 	int flags;
1077 	struct proc *p;
1078 {
1079 
1080 	return (0);
1081 }
1082 
1083 /*
1084  * Do operations associated with quotas, not supported
1085  */
1086 /* ARGSUSED */
1087 int
1088 nfs_quotactl(mp, cmd, uid, arg, p)
1089 	struct mount *mp;
1090 	int cmd;
1091 	uid_t uid;
1092 	caddr_t arg;
1093 	struct proc *p;
1094 {
1095 
1096 	return (EOPNOTSUPP);
1097 }
1098