xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision fd5cb0acea84d278e04e640d37ca2398f894991f)
1 /*	$NetBSD: nfs_vfsops.c,v 1.144 2005/01/02 16:08:30 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. 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 1.144 2005/01/02 16:08:30 thorpej 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 	vfs_stdextattrctl,
117 	nfs_vnodeopv_descs,
118 };
119 
120 extern u_int32_t nfs_procids[NFS_NPROCS];
121 extern u_int32_t nfs_prog, nfs_vers;
122 
123 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
124     struct mount **, struct vnode **, struct proc *));
125 
126 /*
127  * nfs statvfs call
128  */
129 int
130 nfs_statvfs(mp, sbp, p)
131 	struct mount *mp;
132 	struct statvfs *sbp;
133 	struct proc *p;
134 {
135 	struct vnode *vp;
136 	struct nfs_statfs *sfp;
137 	caddr_t cp;
138 	u_int32_t *tl;
139 	int32_t t1, t2;
140 	caddr_t bpos, dpos, cp2;
141 	struct nfsmount *nmp = VFSTONFS(mp);
142 	int error = 0, retattr;
143 #ifdef NFS_V2_ONLY
144 	const int v3 = 0;
145 #else
146 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
147 #endif
148 	struct mbuf *mreq, *mrep = NULL, *md, *mb;
149 	struct ucred *cred;
150 	u_quad_t tquad;
151 	struct nfsnode *np;
152 
153 #ifndef nolint
154 	sfp = (struct nfs_statfs *)0;
155 #endif
156 	vp = nmp->nm_vnode;
157 	np = VTONFS(vp);
158 	cred = crget();
159 	cred->cr_ngroups = 0;
160 #ifndef NFS_V2_ONLY
161 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
162 		(void)nfs_fsinfo(nmp, vp, cred, p);
163 #endif
164 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
165 	nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3));
166 	nfsm_fhtom(np, v3);
167 	nfsm_request(np, NFSPROC_FSSTAT, p, cred);
168 	if (v3)
169 		nfsm_postop_attr(vp, retattr, 0);
170 	if (error) {
171 		if (mrep != NULL) {
172 			if (mrep->m_next != NULL)
173 				printf("nfs_vfsops: nfs_statvfs would lose buffers\n");
174 			m_freem(mrep);
175 		}
176 		goto nfsmout;
177 	}
178 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
179 	sbp->f_flag = nmp->nm_flag;
180 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
181 	if (v3) {
182 		sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE;
183 		tquad = fxdr_hyper(&sfp->sf_tbytes);
184 		sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
185 		tquad = fxdr_hyper(&sfp->sf_fbytes);
186 		sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
187 		tquad = fxdr_hyper(&sfp->sf_abytes);
188 		tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
189 		sbp->f_bresvd = sbp->f_bfree - tquad;
190 		sbp->f_bavail = tquad;
191 #ifdef COMPAT_20
192 		/* Handle older NFS servers returning negative values */
193 		if ((quad_t)sbp->f_bavail < 0)
194 			sbp->f_bavail = 0;
195 #endif
196 		tquad = fxdr_hyper(&sfp->sf_tfiles);
197 		sbp->f_files = tquad;
198 		tquad = fxdr_hyper(&sfp->sf_ffiles);
199 		sbp->f_ffree = tquad;
200 		sbp->f_favail = tquad;
201 		sbp->f_fresvd = 0;
202 		sbp->f_namemax = MAXNAMLEN;
203 	} else {
204 		sbp->f_bsize = NFS_FABLKSIZE;
205 		sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
206 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
207 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
208 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
209 		sbp->f_fresvd = 0;
210 		sbp->f_files = 0;
211 		sbp->f_ffree = 0;
212 		sbp->f_favail = 0;
213 		sbp->f_fresvd = 0;
214 		sbp->f_namemax = MAXNAMLEN;
215 	}
216 	copy_statvfs_info(sbp, mp);
217 	nfsm_reqdone;
218 	crfree(cred);
219 	return (error);
220 }
221 
222 #ifndef NFS_V2_ONLY
223 /*
224  * nfs version 3 fsinfo rpc call
225  */
226 int
227 nfs_fsinfo(nmp, vp, cred, p)
228 	struct nfsmount *nmp;
229 	struct vnode *vp;
230 	struct ucred *cred;
231 	struct proc *p;
232 {
233 	struct nfsv3_fsinfo *fsp;
234 	caddr_t cp;
235 	int32_t t1, t2;
236 	u_int32_t *tl, pref, max;
237 	caddr_t bpos, dpos, cp2;
238 	int error = 0, retattr;
239 	struct mbuf *mreq, *mrep, *md, *mb;
240 	u_int64_t maxfsize;
241 	struct nfsnode *np = VTONFS(vp);
242 
243 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
244 	nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1));
245 	nfsm_fhtom(np, 1);
246 	nfsm_request(np, NFSPROC_FSINFO, p, cred);
247 	nfsm_postop_attr(vp, retattr, 0);
248 	if (!error) {
249 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
250 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
251 		if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 &&
252 		    pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
253 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
254 				~(NFS_FABLKSIZE - 1);
255 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
256 		if (max < nmp->nm_wsize && max > 0) {
257 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
258 			if (nmp->nm_wsize == 0)
259 				nmp->nm_wsize = max;
260 		}
261 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
262 		if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 &&
263 		    pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
264 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
265 				~(NFS_FABLKSIZE - 1);
266 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
267 		if (max < nmp->nm_rsize && max > 0) {
268 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
269 			if (nmp->nm_rsize == 0)
270 				nmp->nm_rsize = max;
271 		}
272 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
273 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
274 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
275 				~(NFS_DIRFRAGSIZ - 1);
276 		if (max < nmp->nm_readdirsize && max > 0) {
277 			nmp->nm_readdirsize = max & ~(NFS_DIRFRAGSIZ - 1);
278 			if (nmp->nm_readdirsize == 0)
279 				nmp->nm_readdirsize = max;
280 		}
281 		/* XXX */
282 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
283 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
284 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
285 			nmp->nm_maxfilesize = maxfsize;
286 		nmp->nm_mountp->mnt_fs_bshift =
287 		    ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
288 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
289 	}
290 	nfsm_reqdone;
291 	return (error);
292 }
293 #endif
294 
295 /*
296  * Mount a remote root fs via. NFS.  It goes like this:
297  * - Call nfs_boot_init() to fill in the nfs_diskless struct
298  * - build the rootfs mount point and call mountnfs() to do the rest.
299  */
300 int
301 nfs_mountroot()
302 {
303 	struct nfs_diskless *nd;
304 	struct vattr attr;
305 	struct mount *mp;
306 	struct vnode *vp;
307 	struct proc *procp;
308 	long n;
309 	int error;
310 
311 	procp = curproc; /* XXX */
312 
313 	if (root_device->dv_class != DV_IFNET)
314 		return (ENODEV);
315 
316 	/*
317 	 * XXX time must be non-zero when we init the interface or else
318 	 * the arp code will wedge.  [Fixed now in if_ether.c]
319 	 * However, the NFS attribute cache gives false "hits" when
320 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
321 	 */
322 	if (time.tv_sec < NFS_MAXATTRTIMO)
323 		time.tv_sec = NFS_MAXATTRTIMO;
324 
325 	/*
326 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
327 	 * Side effect:  Finds and configures a network interface.
328 	 */
329 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
330 	memset((caddr_t)nd, 0, sizeof(*nd));
331 	error = nfs_boot_init(nd, procp);
332 	if (error) {
333 		free(nd, M_NFSMNT);
334 		return (error);
335 	}
336 
337 	/*
338 	 * Create the root mount point.
339 	 */
340 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp);
341 	if (error)
342 		goto out;
343 	printf("root on %s\n", nd->nd_root.ndm_host);
344 
345 	/*
346 	 * Link it into the mount list.
347 	 */
348 	simple_lock(&mountlist_slock);
349 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
350 	simple_unlock(&mountlist_slock);
351 	rootvp = vp;
352 	mp->mnt_vnodecovered = NULLVP;
353 	vfs_unbusy(mp);
354 
355 	/* Get root attributes (for the time). */
356 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
357 	if (error)
358 		panic("nfs_mountroot: getattr for root");
359 	n = attr.va_atime.tv_sec;
360 #ifdef	DEBUG
361 	printf("root time: 0x%lx\n", n);
362 #endif
363 	setrootfstime(n);
364 
365 out:
366 	if (error)
367 		nfs_boot_cleanup(nd, procp);
368 	free(nd, M_NFSMNT);
369 	return (error);
370 }
371 
372 /*
373  * Internal version of mount system call for diskless setup.
374  * Separate function because we used to call it twice.
375  * (once for root and once for swap)
376  */
377 static int
378 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p)
379 	struct nfs_dlmount *ndmntp;
380 	const char *mntname;	/* mount point name */
381 	struct mount **mpp;
382 	struct vnode **vpp;
383 	struct proc *p;
384 {
385 	struct mount *mp;
386 	struct mbuf *m;
387 	int error;
388 
389 	vfs_rootmountalloc(MOUNT_NFS, (char *)mntname, &mp);
390 
391 	mp->mnt_op = &nfs_vfsops;
392 
393 	/*
394 	 * Historical practice expects NFS root file systems to
395 	 * be initially mounted r/w.
396 	 */
397 	mp->mnt_flag &= ~MNT_RDONLY;
398 
399 	/* Get mbuf for server sockaddr. */
400 	m = m_get(M_WAIT, MT_SONAME);
401 	if (m == NULL)
402 		panic("nfs_mountroot: mget soname for %s", mntname);
403 	MCLAIM(m, &nfs_mowner);
404 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
405 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
406 
407 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
408 			 ndmntp->ndm_args.hostname, vpp, p);
409 	if (error) {
410 		mp->mnt_op->vfs_refcount--;
411 		vfs_unbusy(mp);
412 		printf("nfs_mountroot: mount %s failed: %d\n",
413 		       mntname, error);
414 		free(mp, M_MOUNT);
415 	} else
416 		*mpp = mp;
417 
418 	return (error);
419 }
420 
421 void
422 nfs_decode_args(nmp, argp, p)
423 	struct nfsmount *nmp;
424 	struct nfs_args *argp;
425 	struct proc *p;
426 {
427 	int s;
428 	int adjsock;
429 	int maxio;
430 
431 	s = splsoftnet();
432 
433 	/*
434 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
435 	 * no sense in that context.
436 	 */
437 	if (argp->sotype == SOCK_STREAM)
438 		argp->flags &= ~NFSMNT_NOCONN;
439 
440 	/*
441 	 * Cookie translation is not needed for v2, silently ignore it.
442 	 */
443 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
444 	    NFSMNT_XLATECOOKIE)
445 		argp->flags &= ~NFSMNT_XLATECOOKIE;
446 
447 	/* Re-bind if rsrvd port requested and wasn't on one */
448 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
449 		  && (argp->flags & NFSMNT_RESVPORT);
450 	/* Also re-bind if we're switching to/from a connected UDP socket */
451 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
452 		    (argp->flags & NFSMNT_NOCONN));
453 
454 	/* Update flags. */
455 	nmp->nm_flag = argp->flags;
456 	splx(s);
457 
458 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
459 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
460 		if (nmp->nm_timeo < NFS_MINTIMEO)
461 			nmp->nm_timeo = NFS_MINTIMEO;
462 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
463 			nmp->nm_timeo = NFS_MAXTIMEO;
464 	}
465 
466 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
467 		nmp->nm_retry = argp->retrans;
468 		if (nmp->nm_retry > NFS_MAXREXMIT)
469 			nmp->nm_retry = NFS_MAXREXMIT;
470 	}
471 
472 #ifndef NFS_V2_ONLY
473 	if (argp->flags & NFSMNT_NFSV3) {
474 		if (argp->sotype == SOCK_DGRAM)
475 			maxio = NFS_MAXDGRAMDATA;
476 		else
477 			maxio = NFS_MAXDATA;
478 	} else
479 #endif
480 		maxio = NFS_V2MAXDATA;
481 
482 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
483 		int osize = nmp->nm_wsize;
484 		nmp->nm_wsize = argp->wsize;
485 		/* Round down to multiple of blocksize */
486 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
487 		if (nmp->nm_wsize <= 0)
488 			nmp->nm_wsize = NFS_FABLKSIZE;
489 		adjsock |= (nmp->nm_wsize != osize);
490 	}
491 	if (nmp->nm_wsize > maxio)
492 		nmp->nm_wsize = maxio;
493 	if (nmp->nm_wsize > MAXBSIZE)
494 		nmp->nm_wsize = MAXBSIZE;
495 
496 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
497 		int osize = nmp->nm_rsize;
498 		nmp->nm_rsize = argp->rsize;
499 		/* Round down to multiple of blocksize */
500 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
501 		if (nmp->nm_rsize <= 0)
502 			nmp->nm_rsize = NFS_FABLKSIZE;
503 		adjsock |= (nmp->nm_rsize != osize);
504 	}
505 	if (nmp->nm_rsize > maxio)
506 		nmp->nm_rsize = maxio;
507 	if (nmp->nm_rsize > MAXBSIZE)
508 		nmp->nm_rsize = MAXBSIZE;
509 
510 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
511 		nmp->nm_readdirsize = argp->readdirsize;
512 		/* Round down to multiple of minimum blocksize */
513 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
514 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
515 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
516 		/* Bigger than buffer size makes no sense */
517 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
518 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
519 	} else if (argp->flags & NFSMNT_RSIZE)
520 		nmp->nm_readdirsize = nmp->nm_rsize;
521 
522 	if (nmp->nm_readdirsize > maxio)
523 		nmp->nm_readdirsize = maxio;
524 
525 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
526 		argp->maxgrouplist <= NFS_MAXGRPS)
527 		nmp->nm_numgrps = argp->maxgrouplist;
528 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
529 		argp->readahead <= NFS_MAXRAHEAD)
530 		nmp->nm_readahead = argp->readahead;
531 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
532 		argp->leaseterm <= NQ_MAXLEASE)
533 		nmp->nm_leaseterm = argp->leaseterm;
534 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
535 		argp->deadthresh <= NQ_NEVERDEAD)
536 		nmp->nm_deadthresh = argp->deadthresh;
537 
538 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
539 		    (nmp->nm_soproto != argp->proto));
540 	nmp->nm_sotype = argp->sotype;
541 	nmp->nm_soproto = argp->proto;
542 
543 	if (nmp->nm_so && adjsock) {
544 		nfs_safedisconnect(nmp);
545 		if (nmp->nm_sotype == SOCK_DGRAM)
546 			while (nfs_connect(nmp, (struct nfsreq *)0, p)) {
547 				printf("nfs_args: retrying connect\n");
548 				(void) tsleep((caddr_t)&lbolt,
549 					      PSOCK, "nfscn3", 0);
550 			}
551 	}
552 }
553 
554 /*
555  * VFS Operations.
556  *
557  * mount system call
558  * It seems a bit dumb to copyinstr() the host and path here and then
559  * memcpy() them in mountnfs(), but I wanted to detect errors before
560  * doing the sockargs() call because sockargs() allocates an mbuf and
561  * an error after that means that I have to release the mbuf.
562  */
563 /* ARGSUSED */
564 int
565 nfs_mount(mp, path, data, ndp, p)
566 	struct mount *mp;
567 	const char *path;
568 	void *data;
569 	struct nameidata *ndp;
570 	struct proc *p;
571 {
572 	int error;
573 	struct nfs_args args;
574 	struct mbuf *nam;
575 	struct nfsmount *nmp = VFSTONFS(mp);
576 	struct sockaddr *sa;
577 	struct vnode *vp;
578 	char *pth, *hst;
579 	size_t len;
580 	u_char *nfh;
581 
582 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
583 	if (error)
584 		return (error);
585 
586 	if (mp->mnt_flag & MNT_GETARGS) {
587 
588 		if (nmp == NULL)
589 			return (EIO);
590 		if (args.addr != NULL) {
591 			sa = mtod(nmp->nm_nam, struct sockaddr *);
592 			error = copyout(sa, args.addr, sa->sa_len);
593 			if (error)
594 				return (error);
595 			args.addrlen = sa->sa_len;
596 		} else
597 			args.addrlen = 0;
598 
599 		args.version = NFS_ARGSVERSION;
600 		args.sotype = nmp->nm_sotype;
601 		args.proto = nmp->nm_soproto;
602 		args.fh = NULL;
603 		args.fhsize = 0;
604 		args.flags = nmp->nm_flag;
605 		args.wsize = nmp->nm_wsize;
606 		args.rsize = nmp->nm_rsize;
607 		args.readdirsize = nmp->nm_readdirsize;
608 		args.timeo = nmp->nm_timeo;
609 		args.retrans = nmp->nm_retry;
610 		args.maxgrouplist = nmp->nm_numgrps;
611 		args.readahead = nmp->nm_readahead;
612 		args.leaseterm = nmp->nm_leaseterm;
613 		args.deadthresh = nmp->nm_deadthresh;
614 		args.hostname = NULL;
615 		return (copyout(&args, data, sizeof(args)));
616 	}
617 
618 	if (args.version != NFS_ARGSVERSION)
619 		return (EPROGMISMATCH);
620 #ifdef NFS_V2_ONLY
621 	if (args.flags & NFSMNT_NQNFS)
622 		return (EPROGUNAVAIL);
623 	if (args.flags & NFSMNT_NFSV3)
624 		return (EPROGMISMATCH);
625 #endif
626 	if (mp->mnt_flag & MNT_UPDATE) {
627 		if (nmp == NULL)
628 			return (EIO);
629 		/*
630 		 * When doing an update, we can't change from or to
631 		 * v3 and/or nqnfs, or change cookie translation
632 		 */
633 		args.flags = (args.flags &
634 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
635 		    (nmp->nm_flag &
636 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
637 		nfs_decode_args(nmp, &args, p);
638 		return (0);
639 	}
640 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
641 		return (EINVAL);
642 	MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK);
643 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
644 	if (error)
645 		return (error);
646 	MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK);
647 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
648 	if (error)
649 		goto free_nfh;
650 	memset(&pth[len], 0, MNAMELEN - len);
651 	MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK);
652 	error = copyinstr(args.hostname, hst, MNAMELEN - 1, &len);
653 	if (error)
654 		goto free_pth;
655 	memset(&hst[len], 0, MNAMELEN - len);
656 	/* sockargs() call must be after above copyin() calls */
657 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
658 	if (error)
659 		goto free_hst;
660 	MCLAIM(nam, &nfs_mowner);
661 	args.fh = nfh;
662 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
663 
664 free_hst:
665 	FREE(hst, M_TEMP);
666 free_pth:
667 	FREE(pth, M_TEMP);
668 free_nfh:
669 	FREE(nfh, M_TEMP);
670 
671 	return (error);
672 }
673 
674 /*
675  * Common code for mount and mountroot
676  */
677 int
678 mountnfs(argp, mp, nam, pth, hst, vpp, p)
679 	struct nfs_args *argp;
680 	struct mount *mp;
681 	struct mbuf *nam;
682 	const char *pth, *hst;
683 	struct vnode **vpp;
684 	struct proc *p;
685 {
686 	struct nfsmount *nmp;
687 	struct nfsnode *np;
688 	int error;
689 	struct vattr *attrs;
690 	struct ucred *cr;
691 
692 	/*
693 	 * If the number of nfs iothreads to use has never
694 	 * been set, create a reasonable number of them.
695 	 */
696 
697 	if (nfs_niothreads < 0) {
698 		nfs_niothreads = NFS_DEFAULT_NIOTHREADS;
699 		nfs_getset_niothreads(TRUE);
700 	}
701 
702 	if (mp->mnt_flag & MNT_UPDATE) {
703 		nmp = VFSTONFS(mp);
704 		/* update paths, file handles, etc, here	XXX */
705 		m_freem(nam);
706 		return (0);
707 	} else {
708 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
709 		    M_NFSMNT, M_WAITOK);
710 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
711 		mp->mnt_data = nmp;
712 		TAILQ_INIT(&nmp->nm_uidlruhead);
713 		TAILQ_INIT(&nmp->nm_bufq);
714 		lockinit(&nmp->nm_writeverflock, PRIBIO, "nfswverf", 0, 0);
715 		simple_lock_init(&nmp->nm_slock);
716 	}
717 	vfs_getnewfsid(mp);
718 	nmp->nm_mountp = mp;
719 
720 #ifndef NFS_V2_ONLY
721 	if (argp->flags & NFSMNT_NQNFS)
722 		mp->mnt_iflag |= IMNT_DTYPE;
723 #endif
724 
725 #ifndef NFS_V2_ONLY
726 	if ((argp->flags & NFSMNT_NFSV3) == 0)
727 #endif
728 		/*
729 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
730 		 * will fill this in.
731 		 */
732 		nmp->nm_maxfilesize = 0xffffffffLL;
733 
734 	nmp->nm_timeo = NFS_TIMEO;
735 	nmp->nm_retry = NFS_RETRANS;
736 	nmp->nm_wsize = NFS_WSIZE;
737 	nmp->nm_rsize = NFS_RSIZE;
738 	nmp->nm_readdirsize = NFS_READDIRSIZE;
739 	nmp->nm_numgrps = NFS_MAXGRPS;
740 	nmp->nm_readahead = NFS_DEFRAHEAD;
741 	nmp->nm_leaseterm = NQ_DEFLEASE;
742 	nmp->nm_deadthresh = NQ_DEADTHRESH;
743 	CIRCLEQ_INIT(&nmp->nm_timerhead);
744 	nmp->nm_inprog = NULLVP;
745 	error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, mp, p);
746 	if (error)
747 		goto bad;
748 	nmp->nm_nam = nam;
749 
750 	/* Set up the sockets and per-host congestion */
751 	nmp->nm_sotype = argp->sotype;
752 	nmp->nm_soproto = argp->proto;
753 
754 	nfs_decode_args(nmp, argp, p);
755 
756 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
757 	mp->mnt_dev_bshift = DEV_BSHIFT;
758 
759 	/*
760 	 * For Connection based sockets (TCP,...) defer the connect until
761 	 * the first request, in case the server is not responding.
762 	 */
763 	if (nmp->nm_sotype == SOCK_DGRAM &&
764 		(error = nfs_connect(nmp, (struct nfsreq *)0, p)))
765 		goto bad;
766 
767 	/*
768 	 * This is silly, but it has to be set so that vinifod() works.
769 	 * We do not want to do an nfs_statvfs() here since we can get
770 	 * stuck on a dead server and we are holding a lock on the mount
771 	 * point.
772 	 */
773 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
774 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
775 	if (error)
776 		goto bad;
777 	*vpp = NFSTOV(np);
778 	MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK);
779 	VOP_GETATTR(*vpp, attrs, p->p_ucred, p);
780 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
781 		cr = crget();
782 		cr->cr_uid = attrs->va_uid;
783 		cr->cr_gid = attrs->va_gid;
784 		cr->cr_ngroups = 0;
785 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
786 		crfree(cr);
787 	}
788 	FREE(attrs, M_TEMP);
789 
790 	/*
791 	 * A reference count is needed on the nfsnode representing the
792 	 * remote root.  If this object is not persistent, then backward
793 	 * traversals of the mount point (i.e. "..") will not work if
794 	 * the nfsnode gets flushed out of the cache. Ufs does not have
795 	 * this problem, because one can identify root inodes by their
796 	 * number == ROOTINO (2). So, just unlock, but no rele.
797 	 */
798 
799 	nmp->nm_vnode = *vpp;
800 	VOP_UNLOCK(*vpp, 0);
801 
802 	return (0);
803 bad:
804 	nfs_disconnect(nmp);
805 	free((caddr_t)nmp, M_NFSMNT);
806 	m_freem(nam);
807 	return (error);
808 }
809 
810 /*
811  * unmount system call
812  */
813 int
814 nfs_unmount(mp, mntflags, p)
815 	struct mount *mp;
816 	int mntflags;
817 	struct proc *p;
818 {
819 	struct nfsmount *nmp;
820 	struct vnode *vp;
821 	int error, flags = 0;
822 
823 	if (mntflags & MNT_FORCE)
824 		flags |= FORCECLOSE;
825 	nmp = VFSTONFS(mp);
826 	/*
827 	 * Goes something like this..
828 	 * - Check for activity on the root vnode (other than ourselves).
829 	 * - Call vflush() to clear out vnodes for this file system,
830 	 *   except for the root vnode.
831 	 * - Decrement reference on the vnode representing remote root.
832 	 * - Close the socket
833 	 * - Free up the data structures
834 	 */
835 	/*
836 	 * We need to decrement the ref. count on the nfsnode representing
837 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
838 	 * has done vput on this vnode, otherwise we would get deadlock!
839 	 */
840 	vp = nmp->nm_vnode;
841 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
842 	if (error != 0)
843 		return error;
844 
845 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
846 		vput(vp);
847 		return (EBUSY);
848 	}
849 
850 	/*
851 	 * Must handshake with nqnfs_clientd() if it is active.
852 	 */
853 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
854 	while (nmp->nm_inprog != NULLVP)
855 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
856 	error = vflush(mp, vp, flags);
857 	if (error) {
858 		vput(vp);
859 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
860 		return (error);
861 	}
862 
863 	/*
864 	 * We are now committed to the unmount; mark the mount structure
865 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
866 	 * will go away cleanly.
867 	 */
868 	nmp->nm_iflag |= NFSMNT_DISMNT;
869 
870 	/*
871 	 * There are two reference counts to get rid of here
872 	 * (see comment in mountnfs()).
873 	 */
874 	vrele(vp);
875 	vput(vp);
876 	vgone(vp);
877 	nfs_disconnect(nmp);
878 	m_freem(nmp->nm_nam);
879 
880 	/*
881 	 * For NQNFS, let the server daemon free the nfsmount structure.
882 	 */
883 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
884 		free((caddr_t)nmp, M_NFSMNT);
885 	return (0);
886 }
887 
888 /*
889  * Return root of a filesystem
890  */
891 int
892 nfs_root(mp, vpp)
893 	struct mount *mp;
894 	struct vnode **vpp;
895 {
896 	struct vnode *vp;
897 	struct nfsmount *nmp;
898 	int error;
899 
900 	nmp = VFSTONFS(mp);
901 	vp = nmp->nm_vnode;
902 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
903 	if (error != 0)
904 		return error;
905 	if (vp->v_type == VNON)
906 		vp->v_type = VDIR;
907 	vp->v_flag = VROOT;
908 	*vpp = vp;
909 	return (0);
910 }
911 
912 extern int syncprt;
913 
914 /*
915  * Flush out the buffer cache
916  */
917 /* ARGSUSED */
918 int
919 nfs_sync(mp, waitfor, cred, p)
920 	struct mount *mp;
921 	int waitfor;
922 	struct ucred *cred;
923 	struct proc *p;
924 {
925 	struct vnode *vp;
926 	int error, allerror = 0;
927 
928 	/*
929 	 * Force stale buffer cache information to be flushed.
930 	 */
931 loop:
932 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
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 static int
973 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS)
974 {
975 	int error;
976 
977 	nfs_getset_niothreads(0);
978         error = sysctl_lookup(SYSCTLFN_CALL(rnode));
979 	if (error || newp == NULL)
980 		return (error);
981 	nfs_getset_niothreads(1);
982 
983 	return (0);
984 }
985 
986 SYSCTL_SETUP(sysctl_vfs_nfs_setup, "sysctl vfs.nfs subtree setup")
987 {
988 
989 	sysctl_createv(clog, 0, NULL, NULL,
990 		       CTLFLAG_PERMANENT,
991 		       CTLTYPE_NODE, "vfs", NULL,
992 		       NULL, 0, NULL, 0,
993 		       CTL_VFS, CTL_EOL);
994 	sysctl_createv(clog, 0, NULL, NULL,
995 		       CTLFLAG_PERMANENT,
996 		       CTLTYPE_NODE, "nfs",
997 		       SYSCTL_DESCR("NFS vfs options"),
998 		       NULL, 0, NULL, 0,
999 		       CTL_VFS, 2, CTL_EOL);
1000 	/*
1001 	 * XXX the "2" above could be dynamic, thereby eliminating one
1002 	 * more instance of the "number to vfs" mapping problem, but
1003 	 * "2" is the order as taken from sys/mount.h
1004 	 */
1005 
1006 	sysctl_createv(clog, 0, NULL, NULL,
1007 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1008 		       CTLTYPE_STRUCT, "nfsstats",
1009 		       SYSCTL_DESCR("NFS operation statistics"),
1010 		       NULL, 0, &nfsstats, sizeof(nfsstats),
1011 		       CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL);
1012 	sysctl_createv(clog, 0, NULL, NULL,
1013 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1014 		       CTLTYPE_INT, "iothreads",
1015 		       SYSCTL_DESCR("Number of NFS client processes desired"),
1016 		       sysctl_vfs_nfs_iothreads, 0, &nfs_niothreads, 0,
1017 		       CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL);
1018 }
1019 
1020 /*
1021  * At this point, this should never happen
1022  */
1023 /* ARGSUSED */
1024 int
1025 nfs_fhtovp(mp, fhp, vpp)
1026 	struct mount *mp;
1027 	struct fid *fhp;
1028 	struct vnode **vpp;
1029 {
1030 
1031 	return (EINVAL);
1032 }
1033 
1034 /* ARGSUSED */
1035 int
1036 nfs_checkexp(mp, nam, exflagsp, credanonp)
1037 	struct mount *mp;
1038 	struct mbuf *nam;
1039 	int *exflagsp;
1040 	struct ucred **credanonp;
1041 {
1042 
1043 	return (EINVAL);
1044 }
1045 
1046 /*
1047  * Vnode pointer to File handle, should never happen either
1048  */
1049 /* ARGSUSED */
1050 int
1051 nfs_vptofh(vp, fhp)
1052 	struct vnode *vp;
1053 	struct fid *fhp;
1054 {
1055 
1056 	return (EINVAL);
1057 }
1058 
1059 /*
1060  * Vfs start routine, a no-op.
1061  */
1062 /* ARGSUSED */
1063 int
1064 nfs_start(mp, flags, p)
1065 	struct mount *mp;
1066 	int flags;
1067 	struct proc *p;
1068 {
1069 
1070 	return (0);
1071 }
1072 
1073 /*
1074  * Do operations associated with quotas, not supported
1075  */
1076 /* ARGSUSED */
1077 int
1078 nfs_quotactl(mp, cmd, uid, arg, p)
1079 	struct mount *mp;
1080 	int cmd;
1081 	uid_t uid;
1082 	void *arg;
1083 	struct proc *p;
1084 {
1085 
1086 	return (EOPNOTSUPP);
1087 }
1088