xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision c71562d660be5e4ad22016bce45e96f08af190cc)
1 /*	$NetBSD: nfs_vfsops.c,v 1.154 2006/04/20 12:13:53 blymn 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 1.154 2006/04/20 12:13:53 blymn 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 /*
81  * keep a count of the nfs mounts to generate ficticious drive names
82  * for the per drive stats.
83  */
84 unsigned int nfs_mount_count = 0;
85 
86 MALLOC_DEFINE(M_NFSMNT, "NFS mount", "NFS mount structure");
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_statvfs,
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_mountroot,
119 	(int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp,
120 	vfs_stdextattrctl,
121 	nfs_vnodeopv_descs,
122 };
123 VFS_ATTACH(nfs_vfsops);
124 
125 extern u_int32_t nfs_procids[NFS_NPROCS];
126 extern u_int32_t nfs_prog, nfs_vers;
127 
128 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
129     struct mount **, struct vnode **, struct lwp *));
130 
131 /*
132  * nfs statvfs call
133  */
134 int
135 nfs_statvfs(mp, sbp, l)
136 	struct mount *mp;
137 	struct statvfs *sbp;
138 	struct lwp *l;
139 {
140 	struct vnode *vp;
141 	struct nfs_statfs *sfp;
142 	caddr_t cp;
143 	u_int32_t *tl;
144 	int32_t t1, t2;
145 	caddr_t bpos, dpos, cp2;
146 	struct nfsmount *nmp = VFSTONFS(mp);
147 	int error = 0, retattr;
148 #ifdef NFS_V2_ONLY
149 	const int v3 = 0;
150 #else
151 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
152 #endif
153 	struct mbuf *mreq, *mrep = NULL, *md, *mb;
154 	struct ucred *cred;
155 	u_quad_t tquad;
156 	struct nfsnode *np;
157 
158 #ifndef nolint
159 	sfp = (struct nfs_statfs *)0;
160 #endif
161 	vp = nmp->nm_vnode;
162 	np = VTONFS(vp);
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, l);
168 #endif
169 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
170 	nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3));
171 	nfsm_fhtom(np, v3);
172 	nfsm_request(np, NFSPROC_FSSTAT, l, cred);
173 	if (v3)
174 		nfsm_postop_attr(vp, retattr, 0);
175 	if (error) {
176 		if (mrep != NULL) {
177 			if (mrep->m_next != NULL)
178 				printf("nfs_vfsops: nfs_statvfs would lose buffers\n");
179 			m_freem(mrep);
180 		}
181 		goto nfsmout;
182 	}
183 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
184 	sbp->f_flag = nmp->nm_flag;
185 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
186 	if (v3) {
187 		sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE;
188 		tquad = fxdr_hyper(&sfp->sf_tbytes);
189 		sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
190 		tquad = fxdr_hyper(&sfp->sf_fbytes);
191 		sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
192 		tquad = fxdr_hyper(&sfp->sf_abytes);
193 		tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
194 		sbp->f_bresvd = sbp->f_bfree - tquad;
195 		sbp->f_bavail = tquad;
196 #ifdef COMPAT_20
197 		/* Handle older NFS servers returning negative values */
198 		if ((quad_t)sbp->f_bavail < 0)
199 			sbp->f_bavail = 0;
200 #endif
201 		tquad = fxdr_hyper(&sfp->sf_tfiles);
202 		sbp->f_files = tquad;
203 		tquad = fxdr_hyper(&sfp->sf_ffiles);
204 		sbp->f_ffree = tquad;
205 		sbp->f_favail = tquad;
206 		sbp->f_fresvd = 0;
207 		sbp->f_namemax = MAXNAMLEN;
208 	} else {
209 		sbp->f_bsize = NFS_FABLKSIZE;
210 		sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
211 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
212 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
213 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
214 		sbp->f_fresvd = 0;
215 		sbp->f_files = 0;
216 		sbp->f_ffree = 0;
217 		sbp->f_favail = 0;
218 		sbp->f_fresvd = 0;
219 		sbp->f_namemax = MAXNAMLEN;
220 	}
221 	copy_statvfs_info(sbp, mp);
222 	nfsm_reqdone;
223 	crfree(cred);
224 	return (error);
225 }
226 
227 #ifndef NFS_V2_ONLY
228 /*
229  * nfs version 3 fsinfo rpc call
230  */
231 int
232 nfs_fsinfo(nmp, vp, cred, l)
233 	struct nfsmount *nmp;
234 	struct vnode *vp;
235 	struct ucred *cred;
236 	struct lwp *l;
237 {
238 	struct nfsv3_fsinfo *fsp;
239 	caddr_t cp;
240 	int32_t t1, t2;
241 	u_int32_t *tl, pref, xmax;
242 	caddr_t bpos, dpos, cp2;
243 	int error = 0, retattr;
244 	struct mbuf *mreq, *mrep, *md, *mb;
245 	u_int64_t maxfsize;
246 	struct nfsnode *np = VTONFS(vp);
247 
248 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
249 	nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1));
250 	nfsm_fhtom(np, 1);
251 	nfsm_request(np, NFSPROC_FSINFO, l, cred);
252 	nfsm_postop_attr(vp, retattr, 0);
253 	if (!error) {
254 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
255 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
256 		if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 &&
257 		    pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
258 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
259 				~(NFS_FABLKSIZE - 1);
260 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
261 		if (xmax < nmp->nm_wsize && xmax > 0) {
262 			nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1);
263 			if (nmp->nm_wsize == 0)
264 				nmp->nm_wsize = xmax;
265 		}
266 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
267 		if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 &&
268 		    pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
269 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
270 				~(NFS_FABLKSIZE - 1);
271 		xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
272 		if (xmax < nmp->nm_rsize && xmax > 0) {
273 			nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1);
274 			if (nmp->nm_rsize == 0)
275 				nmp->nm_rsize = xmax;
276 		}
277 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
278 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
279 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
280 				~(NFS_DIRFRAGSIZ - 1);
281 		if (xmax < nmp->nm_readdirsize && xmax > 0) {
282 			nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1);
283 			if (nmp->nm_readdirsize == 0)
284 				nmp->nm_readdirsize = xmax;
285 		}
286 		/* XXX */
287 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
288 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
289 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
290 			nmp->nm_maxfilesize = maxfsize;
291 		nmp->nm_mountp->mnt_fs_bshift =
292 		    ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
293 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
294 	}
295 	nfsm_reqdone;
296 	return (error);
297 }
298 #endif
299 
300 /*
301  * Mount a remote root fs via. NFS.  It goes like this:
302  * - Call nfs_boot_init() to fill in the nfs_diskless struct
303  * - build the rootfs mount point and call mountnfs() to do the rest.
304  */
305 int
306 nfs_mountroot()
307 {
308 	struct nfs_diskless *nd;
309 	struct vattr attr;
310 	struct mount *mp;
311 	struct vnode *vp;
312 	struct lwp *l;
313 	long n;
314 	int error;
315 
316 	l = curlwp; /* XXX */
317 
318 	if (device_class(root_device) != DV_IFNET)
319 		return (ENODEV);
320 
321 	/*
322 	 * XXX time must be non-zero when we init the interface or else
323 	 * the arp code will wedge.  [Fixed now in if_ether.c]
324 	 * However, the NFS attribute cache gives false "hits" when
325 	 * time.tv_sec < NFS_ATTRTIMEO(nmp, np) so keep this in for now.
326 	 */
327 	if (time.tv_sec < NFS_MAXATTRTIMO)
328 		time.tv_sec = NFS_MAXATTRTIMO;
329 
330 	/*
331 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
332 	 * Side effect:  Finds and configures a network interface.
333 	 */
334 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
335 	memset((caddr_t)nd, 0, sizeof(*nd));
336 	error = nfs_boot_init(nd, l);
337 	if (error) {
338 		free(nd, M_NFSMNT);
339 		return (error);
340 	}
341 
342 	/*
343 	 * Create the root mount point.
344 	 */
345 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l);
346 	if (error)
347 		goto out;
348 	printf("root on %s\n", nd->nd_root.ndm_host);
349 
350 	/*
351 	 * Link it into the mount list.
352 	 */
353 	simple_lock(&mountlist_slock);
354 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
355 	simple_unlock(&mountlist_slock);
356 	rootvp = vp;
357 	mp->mnt_vnodecovered = NULLVP;
358 	vfs_unbusy(mp);
359 
360 	/* Get root attributes (for the time). */
361 	error = VOP_GETATTR(vp, &attr, l->l_proc->p_ucred, l);
362 	if (error)
363 		panic("nfs_mountroot: getattr for root");
364 	n = attr.va_atime.tv_sec;
365 #ifdef	DEBUG
366 	printf("root time: 0x%lx\n", n);
367 #endif
368 	setrootfstime(n);
369 
370 out:
371 	if (error)
372 		nfs_boot_cleanup(nd, l);
373 	free(nd, M_NFSMNT);
374 	return (error);
375 }
376 
377 /*
378  * Internal version of mount system call for diskless setup.
379  * Separate function because we used to call it twice.
380  * (once for root and once for swap)
381  */
382 static int
383 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, l)
384 	struct nfs_dlmount *ndmntp;
385 	const char *mntname;	/* mount point name */
386 	struct mount **mpp;
387 	struct vnode **vpp;
388 	struct lwp *l;
389 {
390 	struct mount *mp;
391 	struct mbuf *m;
392 	int error;
393 
394 	vfs_rootmountalloc(MOUNT_NFS, mntname, &mp);
395 
396 	mp->mnt_op = &nfs_vfsops;
397 
398 	/*
399 	 * Historical practice expects NFS root file systems to
400 	 * be initially mounted r/w.
401 	 */
402 	mp->mnt_flag &= ~MNT_RDONLY;
403 
404 	/* Get mbuf for server sockaddr. */
405 	m = m_get(M_WAIT, MT_SONAME);
406 	if (m == NULL)
407 		panic("nfs_mountroot: mget soname for %s", mntname);
408 	MCLAIM(m, &nfs_mowner);
409 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
410 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
411 
412 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
413 			 ndmntp->ndm_args.hostname, vpp, l);
414 	if (error) {
415 		mp->mnt_op->vfs_refcount--;
416 		vfs_unbusy(mp);
417 		printf("nfs_mountroot: mount %s failed: %d\n",
418 		       mntname, error);
419 		free(mp, M_MOUNT);
420 	} else
421 		*mpp = mp;
422 
423 	return (error);
424 }
425 
426 void
427 nfs_decode_args(nmp, argp, l)
428 	struct nfsmount *nmp;
429 	struct nfs_args *argp;
430 	struct lwp *l;
431 {
432 	int s;
433 	int adjsock;
434 	int maxio;
435 
436 	s = splsoftnet();
437 
438 	/*
439 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
440 	 * no sense in that context.
441 	 */
442 	if (argp->sotype == SOCK_STREAM)
443 		argp->flags &= ~NFSMNT_NOCONN;
444 
445 	/*
446 	 * Cookie translation is not needed for v2, silently ignore it.
447 	 */
448 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
449 	    NFSMNT_XLATECOOKIE)
450 		argp->flags &= ~NFSMNT_XLATECOOKIE;
451 
452 	/* Re-bind if rsrvd port requested and wasn't on one */
453 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
454 		  && (argp->flags & NFSMNT_RESVPORT);
455 	/* Also re-bind if we're switching to/from a connected UDP socket */
456 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
457 		    (argp->flags & NFSMNT_NOCONN));
458 
459 	/* Update flags. */
460 	nmp->nm_flag = argp->flags;
461 	splx(s);
462 
463 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
464 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
465 		if (nmp->nm_timeo < NFS_MINTIMEO)
466 			nmp->nm_timeo = NFS_MINTIMEO;
467 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
468 			nmp->nm_timeo = NFS_MAXTIMEO;
469 	}
470 
471 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
472 		nmp->nm_retry = argp->retrans;
473 		if (nmp->nm_retry > NFS_MAXREXMIT)
474 			nmp->nm_retry = NFS_MAXREXMIT;
475 	}
476 
477 #ifndef NFS_V2_ONLY
478 	if (argp->flags & NFSMNT_NFSV3) {
479 		if (argp->sotype == SOCK_DGRAM)
480 			maxio = NFS_MAXDGRAMDATA;
481 		else
482 			maxio = NFS_MAXDATA;
483 	} else
484 #endif
485 		maxio = NFS_V2MAXDATA;
486 
487 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
488 		int osize = nmp->nm_wsize;
489 		nmp->nm_wsize = argp->wsize;
490 		/* Round down to multiple of blocksize */
491 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
492 		if (nmp->nm_wsize <= 0)
493 			nmp->nm_wsize = NFS_FABLKSIZE;
494 		adjsock |= (nmp->nm_wsize != osize);
495 	}
496 	if (nmp->nm_wsize > maxio)
497 		nmp->nm_wsize = maxio;
498 	if (nmp->nm_wsize > MAXBSIZE)
499 		nmp->nm_wsize = MAXBSIZE;
500 
501 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
502 		int osize = nmp->nm_rsize;
503 		nmp->nm_rsize = argp->rsize;
504 		/* Round down to multiple of blocksize */
505 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
506 		if (nmp->nm_rsize <= 0)
507 			nmp->nm_rsize = NFS_FABLKSIZE;
508 		adjsock |= (nmp->nm_rsize != osize);
509 	}
510 	if (nmp->nm_rsize > maxio)
511 		nmp->nm_rsize = maxio;
512 	if (nmp->nm_rsize > MAXBSIZE)
513 		nmp->nm_rsize = MAXBSIZE;
514 
515 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
516 		nmp->nm_readdirsize = argp->readdirsize;
517 		/* Round down to multiple of minimum blocksize */
518 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
519 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
520 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
521 		/* Bigger than buffer size makes no sense */
522 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
523 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
524 	} else if (argp->flags & NFSMNT_RSIZE)
525 		nmp->nm_readdirsize = nmp->nm_rsize;
526 
527 	if (nmp->nm_readdirsize > maxio)
528 		nmp->nm_readdirsize = maxio;
529 
530 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
531 		argp->maxgrouplist <= NFS_MAXGRPS)
532 		nmp->nm_numgrps = argp->maxgrouplist;
533 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
534 		argp->readahead <= NFS_MAXRAHEAD)
535 		nmp->nm_readahead = argp->readahead;
536 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
537 		argp->leaseterm <= NQ_MAXLEASE)
538 		nmp->nm_leaseterm = argp->leaseterm;
539 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
540 		argp->deadthresh <= NQ_NEVERDEAD)
541 		nmp->nm_deadthresh = argp->deadthresh;
542 
543 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
544 		    (nmp->nm_soproto != argp->proto));
545 	nmp->nm_sotype = argp->sotype;
546 	nmp->nm_soproto = argp->proto;
547 
548 	if (nmp->nm_so && adjsock) {
549 		nfs_safedisconnect(nmp);
550 		if (nmp->nm_sotype == SOCK_DGRAM)
551 			while (nfs_connect(nmp, (struct nfsreq *)0, l)) {
552 				printf("nfs_args: retrying connect\n");
553 				(void) tsleep((caddr_t)&lbolt,
554 					      PSOCK, "nfscn3", 0);
555 			}
556 	}
557 }
558 
559 /*
560  * VFS Operations.
561  *
562  * mount system call
563  * It seems a bit dumb to copyinstr() the host and path here and then
564  * memcpy() them in mountnfs(), but I wanted to detect errors before
565  * doing the sockargs() call because sockargs() allocates an mbuf and
566  * an error after that means that I have to release the mbuf.
567  */
568 /* ARGSUSED */
569 int
570 nfs_mount(mp, path, data, ndp, l)
571 	struct mount *mp;
572 	const char *path;
573 	void *data;
574 	struct nameidata *ndp;
575 	struct lwp *l;
576 {
577 	int error;
578 	struct nfs_args args;
579 	struct mbuf *nam;
580 	struct nfsmount *nmp = VFSTONFS(mp);
581 	struct sockaddr *sa;
582 	struct vnode *vp;
583 	char *pth, *hst;
584 	struct proc *p;
585 	size_t len;
586 	u_char *nfh;
587 
588 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
589 	if (error)
590 		return (error);
591 
592 	p = l->l_proc;
593 	if (mp->mnt_flag & MNT_GETARGS) {
594 
595 		if (nmp == NULL)
596 			return (EIO);
597 		if (args.addr != NULL) {
598 			sa = mtod(nmp->nm_nam, struct sockaddr *);
599 			error = copyout(sa, args.addr, sa->sa_len);
600 			if (error)
601 				return (error);
602 			args.addrlen = sa->sa_len;
603 		} else
604 			args.addrlen = 0;
605 
606 		args.version = NFS_ARGSVERSION;
607 		args.sotype = nmp->nm_sotype;
608 		args.proto = nmp->nm_soproto;
609 		args.fh = NULL;
610 		args.fhsize = 0;
611 		args.flags = nmp->nm_flag;
612 		args.wsize = nmp->nm_wsize;
613 		args.rsize = nmp->nm_rsize;
614 		args.readdirsize = nmp->nm_readdirsize;
615 		args.timeo = nmp->nm_timeo;
616 		args.retrans = nmp->nm_retry;
617 		args.maxgrouplist = nmp->nm_numgrps;
618 		args.readahead = nmp->nm_readahead;
619 		args.leaseterm = nmp->nm_leaseterm;
620 		args.deadthresh = nmp->nm_deadthresh;
621 		args.hostname = NULL;
622 		return (copyout(&args, data, sizeof(args)));
623 	}
624 
625 	if (args.version != NFS_ARGSVERSION)
626 		return (EPROGMISMATCH);
627 #ifdef NFS_V2_ONLY
628 	if (args.flags & NFSMNT_NQNFS)
629 		return (EPROGUNAVAIL);
630 	if (args.flags & NFSMNT_NFSV3)
631 		return (EPROGMISMATCH);
632 #endif
633 	if (mp->mnt_flag & MNT_UPDATE) {
634 		if (nmp == NULL)
635 			return (EIO);
636 		/*
637 		 * When doing an update, we can't change from or to
638 		 * v3 and/or nqnfs, or change cookie translation
639 		 */
640 		args.flags = (args.flags &
641 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
642 		    (nmp->nm_flag &
643 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
644 		nfs_decode_args(nmp, &args, l);
645 		return (0);
646 	}
647 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
648 		return (EINVAL);
649 	MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK);
650 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
651 	if (error)
652 		return (error);
653 	MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK);
654 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
655 	if (error)
656 		goto free_nfh;
657 	memset(&pth[len], 0, MNAMELEN - len);
658 	MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK);
659 	error = copyinstr(args.hostname, hst, MNAMELEN - 1, &len);
660 	if (error)
661 		goto free_pth;
662 	memset(&hst[len], 0, MNAMELEN - len);
663 	/* sockargs() call must be after above copyin() calls */
664 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
665 	if (error)
666 		goto free_hst;
667 	MCLAIM(nam, &nfs_mowner);
668 	args.fh = nfh;
669 	error = mountnfs(&args, mp, nam, pth, hst, &vp, l);
670 
671 free_hst:
672 	FREE(hst, M_TEMP);
673 free_pth:
674 	FREE(pth, M_TEMP);
675 free_nfh:
676 	FREE(nfh, M_TEMP);
677 
678 	return (error);
679 }
680 
681 /*
682  * Common code for mount and mountroot
683  */
684 int
685 mountnfs(argp, mp, nam, pth, hst, vpp, l)
686 	struct nfs_args *argp;
687 	struct mount *mp;
688 	struct mbuf *nam;
689 	const char *pth, *hst;
690 	struct vnode **vpp;
691 	struct lwp *l;
692 {
693 	struct nfsmount *nmp;
694 	struct nfsnode *np;
695 	int error;
696 	struct vattr *attrs;
697 	struct ucred *cr;
698 
699 	/*
700 	 * If the number of nfs iothreads to use has never
701 	 * been set, create a reasonable number of them.
702 	 */
703 
704 	if (nfs_niothreads < 0) {
705 		nfs_niothreads = NFS_DEFAULT_NIOTHREADS;
706 		nfs_getset_niothreads(TRUE);
707 	}
708 
709 	if (mp->mnt_flag & MNT_UPDATE) {
710 		nmp = VFSTONFS(mp);
711 		/* update paths, file handles, etc, here	XXX */
712 		m_freem(nam);
713 		return (0);
714 	} else {
715 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
716 		    M_NFSMNT, M_WAITOK);
717 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
718 		mp->mnt_data = nmp;
719 		TAILQ_INIT(&nmp->nm_uidlruhead);
720 		TAILQ_INIT(&nmp->nm_bufq);
721 		lockinit(&nmp->nm_writeverflock, PRIBIO, "nfswverf", 0, 0);
722 		simple_lock_init(&nmp->nm_slock);
723 	}
724 	vfs_getnewfsid(mp);
725 	nmp->nm_mountp = mp;
726 
727 #ifndef NFS_V2_ONLY
728 	if (argp->flags & NFSMNT_NQNFS)
729 		mp->mnt_iflag |= IMNT_DTYPE;
730 #endif
731 
732 #ifndef NFS_V2_ONLY
733 	if ((argp->flags & NFSMNT_NFSV3) == 0)
734 #endif
735 		/*
736 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
737 		 * will fill this in.
738 		 */
739 		nmp->nm_maxfilesize = 0xffffffffLL;
740 
741 	nmp->nm_timeo = NFS_TIMEO;
742 	nmp->nm_retry = NFS_RETRANS;
743 	nmp->nm_wsize = NFS_WSIZE;
744 	nmp->nm_rsize = NFS_RSIZE;
745 	nmp->nm_readdirsize = NFS_READDIRSIZE;
746 	nmp->nm_numgrps = NFS_MAXGRPS;
747 	nmp->nm_readahead = NFS_DEFRAHEAD;
748 	nmp->nm_leaseterm = NQ_DEFLEASE;
749 	nmp->nm_deadthresh = NQ_DEADTHRESH;
750 	CIRCLEQ_INIT(&nmp->nm_timerhead);
751 	nmp->nm_inprog = NULLVP;
752 	error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, mp, l);
753 	if (error)
754 		goto bad;
755 	nmp->nm_nam = nam;
756 
757 	/* Set up the sockets and per-host congestion */
758 	nmp->nm_sotype = argp->sotype;
759 	nmp->nm_soproto = argp->proto;
760 
761 	nfs_decode_args(nmp, argp, l);
762 
763 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
764 	mp->mnt_dev_bshift = DEV_BSHIFT;
765 
766 	/*
767 	 * For Connection based sockets (TCP,...) defer the connect until
768 	 * the first request, in case the server is not responding.
769 	 */
770 	if (nmp->nm_sotype == SOCK_DGRAM &&
771 		(error = nfs_connect(nmp, (struct nfsreq *)0, l)))
772 		goto bad;
773 
774 	/*
775 	 * This is silly, but it has to be set so that vinifod() works.
776 	 * We do not want to do an nfs_statvfs() here since we can get
777 	 * stuck on a dead server and we are holding a lock on the mount
778 	 * point.
779 	 */
780 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
781 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
782 	if (error)
783 		goto bad;
784 	*vpp = NFSTOV(np);
785 	MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK);
786 	VOP_GETATTR(*vpp, attrs, l->l_proc->p_ucred, l);
787 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
788 		cr = crget();
789 		cr->cr_uid = attrs->va_uid;
790 		cr->cr_gid = attrs->va_gid;
791 		cr->cr_ngroups = 0;
792 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, l, cr);
793 		crfree(cr);
794 	}
795 	FREE(attrs, M_TEMP);
796 
797 	/*
798 	 * A reference count is needed on the nfsnode representing the
799 	 * remote root.  If this object is not persistent, then backward
800 	 * traversals of the mount point (i.e. "..") will not work if
801 	 * the nfsnode gets flushed out of the cache. Ufs does not have
802 	 * this problem, because one can identify root inodes by their
803 	 * number == ROOTINO (2). So, just unlock, but no rele.
804 	 */
805 
806 	nmp->nm_vnode = *vpp;
807 	VOP_UNLOCK(*vpp, 0);
808 
809 	nmp->nm_stats = iostat_alloc(IOSTAT_NFS);
810 	nmp->nm_stats->io_parent = nmp;
811 	  /* generate a ficticious drive name for the nfs mount */
812 	MALLOC(nmp->nm_stats->io_name, char *, IOSTATNAMELEN, M_NFSMNT,
813 	       M_WAITOK);
814 	snprintf(nmp->nm_stats->io_name, IOSTATNAMELEN, "nfs%u",
815 		 nfs_mount_count);
816 	nfs_mount_count++;
817 
818 	return (0);
819 bad:
820 	nfs_disconnect(nmp);
821 	free((caddr_t)nmp, M_NFSMNT);
822 	m_freem(nam);
823 	return (error);
824 }
825 
826 /*
827  * unmount system call
828  */
829 int
830 nfs_unmount(mp, mntflags, l)
831 	struct mount *mp;
832 	int mntflags;
833 	struct lwp *l;
834 {
835 	struct nfsmount *nmp;
836 	struct vnode *vp;
837 	int error, flags = 0;
838 
839 	if (mntflags & MNT_FORCE)
840 		flags |= FORCECLOSE;
841 	nmp = VFSTONFS(mp);
842 	/*
843 	 * Goes something like this..
844 	 * - Check for activity on the root vnode (other than ourselves).
845 	 * - Call vflush() to clear out vnodes for this file system,
846 	 *   except for the root vnode.
847 	 * - Decrement reference on the vnode representing remote root.
848 	 * - Close the socket
849 	 * - Free up the data structures
850 	 */
851 	/*
852 	 * We need to decrement the ref. count on the nfsnode representing
853 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
854 	 * has done vput on this vnode, otherwise we would get deadlock!
855 	 */
856 	vp = nmp->nm_vnode;
857 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
858 	if (error != 0)
859 		return error;
860 
861 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
862 		vput(vp);
863 		return (EBUSY);
864 	}
865 
866 	/*
867 	 * Must handshake with nqnfs_clientd() if it is active.
868 	 */
869 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
870 	while (nmp->nm_inprog != NULLVP)
871 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
872 	error = vflush(mp, vp, flags);
873 	if (error) {
874 		vput(vp);
875 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
876 		return (error);
877 	}
878 
879 	/*
880 	 * We are now committed to the unmount; mark the mount structure
881 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
882 	 * will go away cleanly.
883 	 */
884 	nmp->nm_iflag |= NFSMNT_DISMNT;
885 
886 	/*
887 	 * Clean up the stats... note that we carefully avoid decrementing
888 	 * nfs_mount_count here for good reason - we may not be unmounting
889 	 * the last thing mounted.
890 	 */
891 	FREE(nmp->nm_stats->io_name, M_NFSMNT);
892 	iostat_free(nmp->nm_stats);
893 
894 	/*
895 	 * There are two reference counts to get rid of here
896 	 * (see comment in mountnfs()).
897 	 */
898 	vrele(vp);
899 	vput(vp);
900 	vgone(vp);
901 	nfs_disconnect(nmp);
902 	m_freem(nmp->nm_nam);
903 
904 	/*
905 	 * For NQNFS, let the server daemon free the nfsmount structure.
906 	 */
907 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
908 		free((caddr_t)nmp, M_NFSMNT);
909 	return (0);
910 }
911 
912 /*
913  * Return root of a filesystem
914  */
915 int
916 nfs_root(mp, vpp)
917 	struct mount *mp;
918 	struct vnode **vpp;
919 {
920 	struct vnode *vp;
921 	struct nfsmount *nmp;
922 	int error;
923 
924 	nmp = VFSTONFS(mp);
925 	vp = nmp->nm_vnode;
926 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
927 	if (error != 0)
928 		return error;
929 	if (vp->v_type == VNON)
930 		vp->v_type = VDIR;
931 	vp->v_flag = VROOT;
932 	*vpp = vp;
933 	return (0);
934 }
935 
936 extern int syncprt;
937 
938 /*
939  * Flush out the buffer cache
940  */
941 /* ARGSUSED */
942 int
943 nfs_sync(mp, waitfor, cred, l)
944 	struct mount *mp;
945 	int waitfor;
946 	struct ucred *cred;
947 	struct lwp *l;
948 {
949 	struct vnode *vp;
950 	int error, allerror = 0;
951 
952 	/*
953 	 * Force stale buffer cache information to be flushed.
954 	 */
955 loop:
956 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
957 		/*
958 		 * If the vnode that we are about to sync is no longer
959 		 * associated with this mount point, start over.
960 		 */
961 		if (vp->v_mount != mp)
962 			goto loop;
963 		if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) ||
964 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
965 		     vp->v_uobj.uo_npages == 0))
966 			continue;
967 		if (vget(vp, LK_EXCLUSIVE))
968 			goto loop;
969 		error = VOP_FSYNC(vp, cred,
970 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, l);
971 		if (error)
972 			allerror = error;
973 		vput(vp);
974 	}
975 	return (allerror);
976 }
977 
978 /*
979  * NFS flat namespace lookup.
980  * Currently unsupported.
981  */
982 /* ARGSUSED */
983 int
984 nfs_vget(mp, ino, vpp)
985 	struct mount *mp;
986 	ino_t ino;
987 	struct vnode **vpp;
988 {
989 
990 	return (EOPNOTSUPP);
991 }
992 
993 /*
994  * Do that sysctl thang...
995  */
996 static int
997 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS)
998 {
999 	int error;
1000 
1001 	nfs_getset_niothreads(0);
1002         error = sysctl_lookup(SYSCTLFN_CALL(rnode));
1003 	if (error || newp == NULL)
1004 		return (error);
1005 	nfs_getset_niothreads(1);
1006 
1007 	return (0);
1008 }
1009 
1010 SYSCTL_SETUP(sysctl_vfs_nfs_setup, "sysctl vfs.nfs subtree setup")
1011 {
1012 
1013 	sysctl_createv(clog, 0, NULL, NULL,
1014 		       CTLFLAG_PERMANENT,
1015 		       CTLTYPE_NODE, "vfs", NULL,
1016 		       NULL, 0, NULL, 0,
1017 		       CTL_VFS, CTL_EOL);
1018 	sysctl_createv(clog, 0, NULL, NULL,
1019 		       CTLFLAG_PERMANENT,
1020 		       CTLTYPE_NODE, "nfs",
1021 		       SYSCTL_DESCR("NFS vfs options"),
1022 		       NULL, 0, NULL, 0,
1023 		       CTL_VFS, 2, CTL_EOL);
1024 	/*
1025 	 * XXX the "2" above could be dynamic, thereby eliminating one
1026 	 * more instance of the "number to vfs" mapping problem, but
1027 	 * "2" is the order as taken from sys/mount.h
1028 	 */
1029 
1030 	sysctl_createv(clog, 0, NULL, NULL,
1031 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1032 		       CTLTYPE_STRUCT, "nfsstats",
1033 		       SYSCTL_DESCR("NFS operation statistics"),
1034 		       NULL, 0, &nfsstats, sizeof(nfsstats),
1035 		       CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL);
1036 	sysctl_createv(clog, 0, NULL, NULL,
1037 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1038 		       CTLTYPE_INT, "iothreads",
1039 		       SYSCTL_DESCR("Number of NFS client processes desired"),
1040 		       sysctl_vfs_nfs_iothreads, 0, &nfs_niothreads, 0,
1041 		       CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL);
1042 }
1043 
1044 /*
1045  * At this point, this should never happen
1046  */
1047 /* ARGSUSED */
1048 int
1049 nfs_fhtovp(mp, fhp, vpp)
1050 	struct mount *mp;
1051 	struct fid *fhp;
1052 	struct vnode **vpp;
1053 {
1054 
1055 	return (EINVAL);
1056 }
1057 
1058 /*
1059  * Vnode pointer to File handle, should never happen either
1060  */
1061 /* ARGSUSED */
1062 int
1063 nfs_vptofh(vp, fhp)
1064 	struct vnode *vp;
1065 	struct fid *fhp;
1066 {
1067 
1068 	return (EINVAL);
1069 }
1070 
1071 /*
1072  * Vfs start routine, a no-op.
1073  */
1074 /* ARGSUSED */
1075 int
1076 nfs_start(mp, flags, l)
1077 	struct mount *mp;
1078 	int flags;
1079 	struct lwp *l;
1080 {
1081 
1082 	return (0);
1083 }
1084 
1085 /*
1086  * Do operations associated with quotas, not supported
1087  */
1088 /* ARGSUSED */
1089 int
1090 nfs_quotactl(mp, cmd, uid, arg, l)
1091 	struct mount *mp;
1092 	int cmd;
1093 	uid_t uid;
1094 	void *arg;
1095 	struct lwp *l;
1096 {
1097 
1098 	return (EOPNOTSUPP);
1099 }
1100