xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision df0caa2637da0538ecdf6b878c4d08e684b43d8f)
1 /*	$NetBSD: nfs_vfsops.c,v 1.148 2005/06/09 02:19:59 atatat 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 1.148 2005/06/09 02:19:59 atatat 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 VFS_ATTACH(nfs_vfsops);
120 
121 extern u_int32_t nfs_procids[NFS_NPROCS];
122 extern u_int32_t nfs_prog, nfs_vers;
123 
124 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
125     struct mount **, struct vnode **, struct proc *));
126 
127 /*
128  * nfs statvfs call
129  */
130 int
131 nfs_statvfs(mp, sbp, p)
132 	struct mount *mp;
133 	struct statvfs *sbp;
134 	struct proc *p;
135 {
136 	struct vnode *vp;
137 	struct nfs_statfs *sfp;
138 	caddr_t cp;
139 	u_int32_t *tl;
140 	int32_t t1, t2;
141 	caddr_t bpos, dpos, cp2;
142 	struct nfsmount *nmp = VFSTONFS(mp);
143 	int error = 0, retattr;
144 #ifdef NFS_V2_ONLY
145 	const int v3 = 0;
146 #else
147 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
148 #endif
149 	struct mbuf *mreq, *mrep = NULL, *md, *mb;
150 	struct ucred *cred;
151 	u_quad_t tquad;
152 	struct nfsnode *np;
153 
154 #ifndef nolint
155 	sfp = (struct nfs_statfs *)0;
156 #endif
157 	vp = nmp->nm_vnode;
158 	np = VTONFS(vp);
159 	cred = crget();
160 	cred->cr_ngroups = 0;
161 #ifndef NFS_V2_ONLY
162 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
163 		(void)nfs_fsinfo(nmp, vp, cred, p);
164 #endif
165 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
166 	nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3));
167 	nfsm_fhtom(np, v3);
168 	nfsm_request(np, NFSPROC_FSSTAT, p, cred);
169 	if (v3)
170 		nfsm_postop_attr(vp, retattr, 0);
171 	if (error) {
172 		if (mrep != NULL) {
173 			if (mrep->m_next != NULL)
174 				printf("nfs_vfsops: nfs_statvfs would lose buffers\n");
175 			m_freem(mrep);
176 		}
177 		goto nfsmout;
178 	}
179 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
180 	sbp->f_flag = nmp->nm_flag;
181 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
182 	if (v3) {
183 		sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE;
184 		tquad = fxdr_hyper(&sfp->sf_tbytes);
185 		sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
186 		tquad = fxdr_hyper(&sfp->sf_fbytes);
187 		sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
188 		tquad = fxdr_hyper(&sfp->sf_abytes);
189 		tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
190 		sbp->f_bresvd = sbp->f_bfree - tquad;
191 		sbp->f_bavail = tquad;
192 #ifdef COMPAT_20
193 		/* Handle older NFS servers returning negative values */
194 		if ((quad_t)sbp->f_bavail < 0)
195 			sbp->f_bavail = 0;
196 #endif
197 		tquad = fxdr_hyper(&sfp->sf_tfiles);
198 		sbp->f_files = tquad;
199 		tquad = fxdr_hyper(&sfp->sf_ffiles);
200 		sbp->f_ffree = tquad;
201 		sbp->f_favail = tquad;
202 		sbp->f_fresvd = 0;
203 		sbp->f_namemax = MAXNAMLEN;
204 	} else {
205 		sbp->f_bsize = NFS_FABLKSIZE;
206 		sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
207 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
208 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
209 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
210 		sbp->f_fresvd = 0;
211 		sbp->f_files = 0;
212 		sbp->f_ffree = 0;
213 		sbp->f_favail = 0;
214 		sbp->f_fresvd = 0;
215 		sbp->f_namemax = MAXNAMLEN;
216 	}
217 	copy_statvfs_info(sbp, mp);
218 	nfsm_reqdone;
219 	crfree(cred);
220 	return (error);
221 }
222 
223 #ifndef NFS_V2_ONLY
224 /*
225  * nfs version 3 fsinfo rpc call
226  */
227 int
228 nfs_fsinfo(nmp, vp, cred, p)
229 	struct nfsmount *nmp;
230 	struct vnode *vp;
231 	struct ucred *cred;
232 	struct proc *p;
233 {
234 	struct nfsv3_fsinfo *fsp;
235 	caddr_t cp;
236 	int32_t t1, t2;
237 	u_int32_t *tl, pref, xmax;
238 	caddr_t bpos, dpos, cp2;
239 	int error = 0, retattr;
240 	struct mbuf *mreq, *mrep, *md, *mb;
241 	u_int64_t maxfsize;
242 	struct nfsnode *np = VTONFS(vp);
243 
244 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
245 	nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1));
246 	nfsm_fhtom(np, 1);
247 	nfsm_request(np, NFSPROC_FSINFO, p, cred);
248 	nfsm_postop_attr(vp, retattr, 0);
249 	if (!error) {
250 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
251 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
252 		if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 &&
253 		    pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
254 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
255 				~(NFS_FABLKSIZE - 1);
256 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
257 		if (xmax < nmp->nm_wsize && xmax > 0) {
258 			nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1);
259 			if (nmp->nm_wsize == 0)
260 				nmp->nm_wsize = xmax;
261 		}
262 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
263 		if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 &&
264 		    pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
265 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
266 				~(NFS_FABLKSIZE - 1);
267 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
268 		if (xmax < nmp->nm_rsize && xmax > 0) {
269 			nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1);
270 			if (nmp->nm_rsize == 0)
271 				nmp->nm_rsize = xmax;
272 		}
273 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
274 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
275 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
276 				~(NFS_DIRFRAGSIZ - 1);
277 		if (xmax < nmp->nm_readdirsize && xmax > 0) {
278 			nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1);
279 			if (nmp->nm_readdirsize == 0)
280 				nmp->nm_readdirsize = xmax;
281 		}
282 		/* XXX */
283 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
284 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
285 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
286 			nmp->nm_maxfilesize = maxfsize;
287 		nmp->nm_mountp->mnt_fs_bshift =
288 		    ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
289 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
290 	}
291 	nfsm_reqdone;
292 	return (error);
293 }
294 #endif
295 
296 /*
297  * Mount a remote root fs via. NFS.  It goes like this:
298  * - Call nfs_boot_init() to fill in the nfs_diskless struct
299  * - build the rootfs mount point and call mountnfs() to do the rest.
300  */
301 int
302 nfs_mountroot()
303 {
304 	struct nfs_diskless *nd;
305 	struct vattr attr;
306 	struct mount *mp;
307 	struct vnode *vp;
308 	struct proc *procp;
309 	long n;
310 	int error;
311 
312 	procp = curproc; /* XXX */
313 
314 	if (root_device->dv_class != DV_IFNET)
315 		return (ENODEV);
316 
317 	/*
318 	 * XXX time must be non-zero when we init the interface or else
319 	 * the arp code will wedge.  [Fixed now in if_ether.c]
320 	 * However, the NFS attribute cache gives false "hits" when
321 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
322 	 */
323 	if (time.tv_sec < NFS_MAXATTRTIMO)
324 		time.tv_sec = NFS_MAXATTRTIMO;
325 
326 	/*
327 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
328 	 * Side effect:  Finds and configures a network interface.
329 	 */
330 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
331 	memset((caddr_t)nd, 0, sizeof(*nd));
332 	error = nfs_boot_init(nd, procp);
333 	if (error) {
334 		free(nd, M_NFSMNT);
335 		return (error);
336 	}
337 
338 	/*
339 	 * Create the root mount point.
340 	 */
341 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp);
342 	if (error)
343 		goto out;
344 	printf("root on %s\n", nd->nd_root.ndm_host);
345 
346 	/*
347 	 * Link it into the mount list.
348 	 */
349 	simple_lock(&mountlist_slock);
350 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
351 	simple_unlock(&mountlist_slock);
352 	rootvp = vp;
353 	mp->mnt_vnodecovered = NULLVP;
354 	vfs_unbusy(mp);
355 
356 	/* Get root attributes (for the time). */
357 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
358 	if (error)
359 		panic("nfs_mountroot: getattr for root");
360 	n = attr.va_atime.tv_sec;
361 #ifdef	DEBUG
362 	printf("root time: 0x%lx\n", n);
363 #endif
364 	setrootfstime(n);
365 
366 out:
367 	if (error)
368 		nfs_boot_cleanup(nd, procp);
369 	free(nd, M_NFSMNT);
370 	return (error);
371 }
372 
373 /*
374  * Internal version of mount system call for diskless setup.
375  * Separate function because we used to call it twice.
376  * (once for root and once for swap)
377  */
378 static int
379 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p)
380 	struct nfs_dlmount *ndmntp;
381 	const char *mntname;	/* mount point name */
382 	struct mount **mpp;
383 	struct vnode **vpp;
384 	struct proc *p;
385 {
386 	struct mount *mp;
387 	struct mbuf *m;
388 	int error;
389 
390 	vfs_rootmountalloc(MOUNT_NFS, mntname, &mp);
391 
392 	mp->mnt_op = &nfs_vfsops;
393 
394 	/*
395 	 * Historical practice expects NFS root file systems to
396 	 * be initially mounted r/w.
397 	 */
398 	mp->mnt_flag &= ~MNT_RDONLY;
399 
400 	/* Get mbuf for server sockaddr. */
401 	m = m_get(M_WAIT, MT_SONAME);
402 	if (m == NULL)
403 		panic("nfs_mountroot: mget soname for %s", mntname);
404 	MCLAIM(m, &nfs_mowner);
405 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
406 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
407 
408 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
409 			 ndmntp->ndm_args.hostname, vpp, p);
410 	if (error) {
411 		mp->mnt_op->vfs_refcount--;
412 		vfs_unbusy(mp);
413 		printf("nfs_mountroot: mount %s failed: %d\n",
414 		       mntname, error);
415 		free(mp, M_MOUNT);
416 	} else
417 		*mpp = mp;
418 
419 	return (error);
420 }
421 
422 void
423 nfs_decode_args(nmp, argp, p)
424 	struct nfsmount *nmp;
425 	struct nfs_args *argp;
426 	struct proc *p;
427 {
428 	int s;
429 	int adjsock;
430 	int maxio;
431 
432 	s = splsoftnet();
433 
434 	/*
435 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
436 	 * no sense in that context.
437 	 */
438 	if (argp->sotype == SOCK_STREAM)
439 		argp->flags &= ~NFSMNT_NOCONN;
440 
441 	/*
442 	 * Cookie translation is not needed for v2, silently ignore it.
443 	 */
444 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
445 	    NFSMNT_XLATECOOKIE)
446 		argp->flags &= ~NFSMNT_XLATECOOKIE;
447 
448 	/* Re-bind if rsrvd port requested and wasn't on one */
449 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
450 		  && (argp->flags & NFSMNT_RESVPORT);
451 	/* Also re-bind if we're switching to/from a connected UDP socket */
452 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
453 		    (argp->flags & NFSMNT_NOCONN));
454 
455 	/* Update flags. */
456 	nmp->nm_flag = argp->flags;
457 	splx(s);
458 
459 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
460 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
461 		if (nmp->nm_timeo < NFS_MINTIMEO)
462 			nmp->nm_timeo = NFS_MINTIMEO;
463 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
464 			nmp->nm_timeo = NFS_MAXTIMEO;
465 	}
466 
467 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
468 		nmp->nm_retry = argp->retrans;
469 		if (nmp->nm_retry > NFS_MAXREXMIT)
470 			nmp->nm_retry = NFS_MAXREXMIT;
471 	}
472 
473 #ifndef NFS_V2_ONLY
474 	if (argp->flags & NFSMNT_NFSV3) {
475 		if (argp->sotype == SOCK_DGRAM)
476 			maxio = NFS_MAXDGRAMDATA;
477 		else
478 			maxio = NFS_MAXDATA;
479 	} else
480 #endif
481 		maxio = NFS_V2MAXDATA;
482 
483 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
484 		int osize = nmp->nm_wsize;
485 		nmp->nm_wsize = argp->wsize;
486 		/* Round down to multiple of blocksize */
487 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
488 		if (nmp->nm_wsize <= 0)
489 			nmp->nm_wsize = NFS_FABLKSIZE;
490 		adjsock |= (nmp->nm_wsize != osize);
491 	}
492 	if (nmp->nm_wsize > maxio)
493 		nmp->nm_wsize = maxio;
494 	if (nmp->nm_wsize > MAXBSIZE)
495 		nmp->nm_wsize = MAXBSIZE;
496 
497 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
498 		int osize = nmp->nm_rsize;
499 		nmp->nm_rsize = argp->rsize;
500 		/* Round down to multiple of blocksize */
501 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
502 		if (nmp->nm_rsize <= 0)
503 			nmp->nm_rsize = NFS_FABLKSIZE;
504 		adjsock |= (nmp->nm_rsize != osize);
505 	}
506 	if (nmp->nm_rsize > maxio)
507 		nmp->nm_rsize = maxio;
508 	if (nmp->nm_rsize > MAXBSIZE)
509 		nmp->nm_rsize = MAXBSIZE;
510 
511 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
512 		nmp->nm_readdirsize = argp->readdirsize;
513 		/* Round down to multiple of minimum blocksize */
514 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
515 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
516 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
517 		/* Bigger than buffer size makes no sense */
518 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
519 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
520 	} else if (argp->flags & NFSMNT_RSIZE)
521 		nmp->nm_readdirsize = nmp->nm_rsize;
522 
523 	if (nmp->nm_readdirsize > maxio)
524 		nmp->nm_readdirsize = maxio;
525 
526 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
527 		argp->maxgrouplist <= NFS_MAXGRPS)
528 		nmp->nm_numgrps = argp->maxgrouplist;
529 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
530 		argp->readahead <= NFS_MAXRAHEAD)
531 		nmp->nm_readahead = argp->readahead;
532 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
533 		argp->leaseterm <= NQ_MAXLEASE)
534 		nmp->nm_leaseterm = argp->leaseterm;
535 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
536 		argp->deadthresh <= NQ_NEVERDEAD)
537 		nmp->nm_deadthresh = argp->deadthresh;
538 
539 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
540 		    (nmp->nm_soproto != argp->proto));
541 	nmp->nm_sotype = argp->sotype;
542 	nmp->nm_soproto = argp->proto;
543 
544 	if (nmp->nm_so && adjsock) {
545 		nfs_safedisconnect(nmp);
546 		if (nmp->nm_sotype == SOCK_DGRAM)
547 			while (nfs_connect(nmp, (struct nfsreq *)0, p)) {
548 				printf("nfs_args: retrying connect\n");
549 				(void) tsleep((caddr_t)&lbolt,
550 					      PSOCK, "nfscn3", 0);
551 			}
552 	}
553 }
554 
555 /*
556  * VFS Operations.
557  *
558  * mount system call
559  * It seems a bit dumb to copyinstr() the host and path here and then
560  * memcpy() them in mountnfs(), but I wanted to detect errors before
561  * doing the sockargs() call because sockargs() allocates an mbuf and
562  * an error after that means that I have to release the mbuf.
563  */
564 /* ARGSUSED */
565 int
566 nfs_mount(mp, path, data, ndp, p)
567 	struct mount *mp;
568 	const char *path;
569 	void *data;
570 	struct nameidata *ndp;
571 	struct proc *p;
572 {
573 	int error;
574 	struct nfs_args args;
575 	struct mbuf *nam;
576 	struct nfsmount *nmp = VFSTONFS(mp);
577 	struct sockaddr *sa;
578 	struct vnode *vp;
579 	char *pth, *hst;
580 	size_t len;
581 	u_char *nfh;
582 
583 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
584 	if (error)
585 		return (error);
586 
587 	if (mp->mnt_flag & MNT_GETARGS) {
588 
589 		if (nmp == NULL)
590 			return (EIO);
591 		if (args.addr != NULL) {
592 			sa = mtod(nmp->nm_nam, struct sockaddr *);
593 			error = copyout(sa, args.addr, sa->sa_len);
594 			if (error)
595 				return (error);
596 			args.addrlen = sa->sa_len;
597 		} else
598 			args.addrlen = 0;
599 
600 		args.version = NFS_ARGSVERSION;
601 		args.sotype = nmp->nm_sotype;
602 		args.proto = nmp->nm_soproto;
603 		args.fh = NULL;
604 		args.fhsize = 0;
605 		args.flags = nmp->nm_flag;
606 		args.wsize = nmp->nm_wsize;
607 		args.rsize = nmp->nm_rsize;
608 		args.readdirsize = nmp->nm_readdirsize;
609 		args.timeo = nmp->nm_timeo;
610 		args.retrans = nmp->nm_retry;
611 		args.maxgrouplist = nmp->nm_numgrps;
612 		args.readahead = nmp->nm_readahead;
613 		args.leaseterm = nmp->nm_leaseterm;
614 		args.deadthresh = nmp->nm_deadthresh;
615 		args.hostname = NULL;
616 		return (copyout(&args, data, sizeof(args)));
617 	}
618 
619 	if (args.version != NFS_ARGSVERSION)
620 		return (EPROGMISMATCH);
621 #ifdef NFS_V2_ONLY
622 	if (args.flags & NFSMNT_NQNFS)
623 		return (EPROGUNAVAIL);
624 	if (args.flags & NFSMNT_NFSV3)
625 		return (EPROGMISMATCH);
626 #endif
627 	if (mp->mnt_flag & MNT_UPDATE) {
628 		if (nmp == NULL)
629 			return (EIO);
630 		/*
631 		 * When doing an update, we can't change from or to
632 		 * v3 and/or nqnfs, or change cookie translation
633 		 */
634 		args.flags = (args.flags &
635 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
636 		    (nmp->nm_flag &
637 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
638 		nfs_decode_args(nmp, &args, p);
639 		return (0);
640 	}
641 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
642 		return (EINVAL);
643 	MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK);
644 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
645 	if (error)
646 		return (error);
647 	MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK);
648 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
649 	if (error)
650 		goto free_nfh;
651 	memset(&pth[len], 0, MNAMELEN - len);
652 	MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK);
653 	error = copyinstr(args.hostname, hst, MNAMELEN - 1, &len);
654 	if (error)
655 		goto free_pth;
656 	memset(&hst[len], 0, MNAMELEN - len);
657 	/* sockargs() call must be after above copyin() calls */
658 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
659 	if (error)
660 		goto free_hst;
661 	MCLAIM(nam, &nfs_mowner);
662 	args.fh = nfh;
663 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
664 
665 free_hst:
666 	FREE(hst, M_TEMP);
667 free_pth:
668 	FREE(pth, M_TEMP);
669 free_nfh:
670 	FREE(nfh, M_TEMP);
671 
672 	return (error);
673 }
674 
675 /*
676  * Common code for mount and mountroot
677  */
678 int
679 mountnfs(argp, mp, nam, pth, hst, vpp, p)
680 	struct nfs_args *argp;
681 	struct mount *mp;
682 	struct mbuf *nam;
683 	const char *pth, *hst;
684 	struct vnode **vpp;
685 	struct proc *p;
686 {
687 	struct nfsmount *nmp;
688 	struct nfsnode *np;
689 	int error;
690 	struct vattr *attrs;
691 	struct ucred *cr;
692 
693 	/*
694 	 * If the number of nfs iothreads to use has never
695 	 * been set, create a reasonable number of them.
696 	 */
697 
698 	if (nfs_niothreads < 0) {
699 		nfs_niothreads = NFS_DEFAULT_NIOTHREADS;
700 		nfs_getset_niothreads(TRUE);
701 	}
702 
703 	if (mp->mnt_flag & MNT_UPDATE) {
704 		nmp = VFSTONFS(mp);
705 		/* update paths, file handles, etc, here	XXX */
706 		m_freem(nam);
707 		return (0);
708 	} else {
709 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
710 		    M_NFSMNT, M_WAITOK);
711 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
712 		mp->mnt_data = nmp;
713 		TAILQ_INIT(&nmp->nm_uidlruhead);
714 		TAILQ_INIT(&nmp->nm_bufq);
715 		lockinit(&nmp->nm_writeverflock, PRIBIO, "nfswverf", 0, 0);
716 		simple_lock_init(&nmp->nm_slock);
717 	}
718 	vfs_getnewfsid(mp);
719 	nmp->nm_mountp = mp;
720 
721 #ifndef NFS_V2_ONLY
722 	if (argp->flags & NFSMNT_NQNFS)
723 		mp->mnt_iflag |= IMNT_DTYPE;
724 #endif
725 
726 #ifndef NFS_V2_ONLY
727 	if ((argp->flags & NFSMNT_NFSV3) == 0)
728 #endif
729 		/*
730 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
731 		 * will fill this in.
732 		 */
733 		nmp->nm_maxfilesize = 0xffffffffLL;
734 
735 	nmp->nm_timeo = NFS_TIMEO;
736 	nmp->nm_retry = NFS_RETRANS;
737 	nmp->nm_wsize = NFS_WSIZE;
738 	nmp->nm_rsize = NFS_RSIZE;
739 	nmp->nm_readdirsize = NFS_READDIRSIZE;
740 	nmp->nm_numgrps = NFS_MAXGRPS;
741 	nmp->nm_readahead = NFS_DEFRAHEAD;
742 	nmp->nm_leaseterm = NQ_DEFLEASE;
743 	nmp->nm_deadthresh = NQ_DEADTHRESH;
744 	CIRCLEQ_INIT(&nmp->nm_timerhead);
745 	nmp->nm_inprog = NULLVP;
746 	error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, mp, p);
747 	if (error)
748 		goto bad;
749 	nmp->nm_nam = nam;
750 
751 	/* Set up the sockets and per-host congestion */
752 	nmp->nm_sotype = argp->sotype;
753 	nmp->nm_soproto = argp->proto;
754 
755 	nfs_decode_args(nmp, argp, p);
756 
757 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
758 	mp->mnt_dev_bshift = DEV_BSHIFT;
759 
760 	/*
761 	 * For Connection based sockets (TCP,...) defer the connect until
762 	 * the first request, in case the server is not responding.
763 	 */
764 	if (nmp->nm_sotype == SOCK_DGRAM &&
765 		(error = nfs_connect(nmp, (struct nfsreq *)0, p)))
766 		goto bad;
767 
768 	/*
769 	 * This is silly, but it has to be set so that vinifod() works.
770 	 * We do not want to do an nfs_statvfs() here since we can get
771 	 * stuck on a dead server and we are holding a lock on the mount
772 	 * point.
773 	 */
774 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
775 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
776 	if (error)
777 		goto bad;
778 	*vpp = NFSTOV(np);
779 	MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK);
780 	VOP_GETATTR(*vpp, attrs, p->p_ucred, p);
781 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
782 		cr = crget();
783 		cr->cr_uid = attrs->va_uid;
784 		cr->cr_gid = attrs->va_gid;
785 		cr->cr_ngroups = 0;
786 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
787 		crfree(cr);
788 	}
789 	FREE(attrs, M_TEMP);
790 
791 	/*
792 	 * A reference count is needed on the nfsnode representing the
793 	 * remote root.  If this object is not persistent, then backward
794 	 * traversals of the mount point (i.e. "..") will not work if
795 	 * the nfsnode gets flushed out of the cache. Ufs does not have
796 	 * this problem, because one can identify root inodes by their
797 	 * number == ROOTINO (2). So, just unlock, but no rele.
798 	 */
799 
800 	nmp->nm_vnode = *vpp;
801 	VOP_UNLOCK(*vpp, 0);
802 
803 	return (0);
804 bad:
805 	nfs_disconnect(nmp);
806 	free((caddr_t)nmp, M_NFSMNT);
807 	m_freem(nam);
808 	return (error);
809 }
810 
811 /*
812  * unmount system call
813  */
814 int
815 nfs_unmount(mp, mntflags, p)
816 	struct mount *mp;
817 	int mntflags;
818 	struct proc *p;
819 {
820 	struct nfsmount *nmp;
821 	struct vnode *vp;
822 	int error, flags = 0;
823 
824 	if (mntflags & MNT_FORCE)
825 		flags |= FORCECLOSE;
826 	nmp = VFSTONFS(mp);
827 	/*
828 	 * Goes something like this..
829 	 * - Check for activity on the root vnode (other than ourselves).
830 	 * - Call vflush() to clear out vnodes for this file system,
831 	 *   except for the root vnode.
832 	 * - Decrement reference on the vnode representing remote root.
833 	 * - Close the socket
834 	 * - Free up the data structures
835 	 */
836 	/*
837 	 * We need to decrement the ref. count on the nfsnode representing
838 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
839 	 * has done vput on this vnode, otherwise we would get deadlock!
840 	 */
841 	vp = nmp->nm_vnode;
842 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
843 	if (error != 0)
844 		return error;
845 
846 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
847 		vput(vp);
848 		return (EBUSY);
849 	}
850 
851 	/*
852 	 * Must handshake with nqnfs_clientd() if it is active.
853 	 */
854 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
855 	while (nmp->nm_inprog != NULLVP)
856 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
857 	error = vflush(mp, vp, flags);
858 	if (error) {
859 		vput(vp);
860 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
861 		return (error);
862 	}
863 
864 	/*
865 	 * We are now committed to the unmount; mark the mount structure
866 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
867 	 * will go away cleanly.
868 	 */
869 	nmp->nm_iflag |= NFSMNT_DISMNT;
870 
871 	/*
872 	 * There are two reference counts to get rid of here
873 	 * (see comment in mountnfs()).
874 	 */
875 	vrele(vp);
876 	vput(vp);
877 	vgone(vp);
878 	nfs_disconnect(nmp);
879 	m_freem(nmp->nm_nam);
880 
881 	/*
882 	 * For NQNFS, let the server daemon free the nfsmount structure.
883 	 */
884 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
885 		free((caddr_t)nmp, M_NFSMNT);
886 	return (0);
887 }
888 
889 /*
890  * Return root of a filesystem
891  */
892 int
893 nfs_root(mp, vpp)
894 	struct mount *mp;
895 	struct vnode **vpp;
896 {
897 	struct vnode *vp;
898 	struct nfsmount *nmp;
899 	int error;
900 
901 	nmp = VFSTONFS(mp);
902 	vp = nmp->nm_vnode;
903 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
904 	if (error != 0)
905 		return error;
906 	if (vp->v_type == VNON)
907 		vp->v_type = VDIR;
908 	vp->v_flag = VROOT;
909 	*vpp = vp;
910 	return (0);
911 }
912 
913 extern int syncprt;
914 
915 /*
916  * Flush out the buffer cache
917  */
918 /* ARGSUSED */
919 int
920 nfs_sync(mp, waitfor, cred, p)
921 	struct mount *mp;
922 	int waitfor;
923 	struct ucred *cred;
924 	struct proc *p;
925 {
926 	struct vnode *vp;
927 	int error, allerror = 0;
928 
929 	/*
930 	 * Force stale buffer cache information to be flushed.
931 	 */
932 loop:
933 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
934 		/*
935 		 * If the vnode that we are about to sync is no longer
936 		 * associated with this mount point, start over.
937 		 */
938 		if (vp->v_mount != mp)
939 			goto loop;
940 		if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) ||
941 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
942 		     vp->v_uobj.uo_npages == 0))
943 			continue;
944 		if (vget(vp, LK_EXCLUSIVE))
945 			goto loop;
946 		error = VOP_FSYNC(vp, cred,
947 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p);
948 		if (error)
949 			allerror = error;
950 		vput(vp);
951 	}
952 	return (allerror);
953 }
954 
955 /*
956  * NFS flat namespace lookup.
957  * Currently unsupported.
958  */
959 /* ARGSUSED */
960 int
961 nfs_vget(mp, ino, vpp)
962 	struct mount *mp;
963 	ino_t ino;
964 	struct vnode **vpp;
965 {
966 
967 	return (EOPNOTSUPP);
968 }
969 
970 /*
971  * Do that sysctl thang...
972  */
973 static int
974 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS)
975 {
976 	int error;
977 
978 	nfs_getset_niothreads(0);
979         error = sysctl_lookup(SYSCTLFN_CALL(rnode));
980 	if (error || newp == NULL)
981 		return (error);
982 	nfs_getset_niothreads(1);
983 
984 	return (0);
985 }
986 
987 SYSCTL_SETUP(sysctl_vfs_nfs_setup, "sysctl vfs.nfs subtree setup")
988 {
989 
990 	sysctl_createv(clog, 0, NULL, NULL,
991 		       CTLFLAG_PERMANENT,
992 		       CTLTYPE_NODE, "vfs", NULL,
993 		       NULL, 0, NULL, 0,
994 		       CTL_VFS, CTL_EOL);
995 	sysctl_createv(clog, 0, NULL, NULL,
996 		       CTLFLAG_PERMANENT,
997 		       CTLTYPE_NODE, "nfs",
998 		       SYSCTL_DESCR("NFS vfs options"),
999 		       NULL, 0, NULL, 0,
1000 		       CTL_VFS, 2, CTL_EOL);
1001 	/*
1002 	 * XXX the "2" above could be dynamic, thereby eliminating one
1003 	 * more instance of the "number to vfs" mapping problem, but
1004 	 * "2" is the order as taken from sys/mount.h
1005 	 */
1006 
1007 	sysctl_createv(clog, 0, NULL, NULL,
1008 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1009 		       CTLTYPE_STRUCT, "nfsstats",
1010 		       SYSCTL_DESCR("NFS operation statistics"),
1011 		       NULL, 0, &nfsstats, sizeof(nfsstats),
1012 		       CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL);
1013 	sysctl_createv(clog, 0, NULL, NULL,
1014 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1015 		       CTLTYPE_INT, "iothreads",
1016 		       SYSCTL_DESCR("Number of NFS client processes desired"),
1017 		       sysctl_vfs_nfs_iothreads, 0, &nfs_niothreads, 0,
1018 		       CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL);
1019 }
1020 
1021 /*
1022  * At this point, this should never happen
1023  */
1024 /* ARGSUSED */
1025 int
1026 nfs_fhtovp(mp, fhp, vpp)
1027 	struct mount *mp;
1028 	struct fid *fhp;
1029 	struct vnode **vpp;
1030 {
1031 
1032 	return (EINVAL);
1033 }
1034 
1035 /* ARGSUSED */
1036 int
1037 nfs_checkexp(mp, nam, exflagsp, credanonp)
1038 	struct mount *mp;
1039 	struct mbuf *nam;
1040 	int *exflagsp;
1041 	struct ucred **credanonp;
1042 {
1043 
1044 	return (EINVAL);
1045 }
1046 
1047 /*
1048  * Vnode pointer to File handle, should never happen either
1049  */
1050 /* ARGSUSED */
1051 int
1052 nfs_vptofh(vp, fhp)
1053 	struct vnode *vp;
1054 	struct fid *fhp;
1055 {
1056 
1057 	return (EINVAL);
1058 }
1059 
1060 /*
1061  * Vfs start routine, a no-op.
1062  */
1063 /* ARGSUSED */
1064 int
1065 nfs_start(mp, flags, p)
1066 	struct mount *mp;
1067 	int flags;
1068 	struct proc *p;
1069 {
1070 
1071 	return (0);
1072 }
1073 
1074 /*
1075  * Do operations associated with quotas, not supported
1076  */
1077 /* ARGSUSED */
1078 int
1079 nfs_quotactl(mp, cmd, uid, arg, p)
1080 	struct mount *mp;
1081 	int cmd;
1082 	uid_t uid;
1083 	void *arg;
1084 	struct proc *p;
1085 {
1086 
1087 	return (EOPNOTSUPP);
1088 }
1089