xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision 1ffa7b76c40339c17a0fb2a09fac93f287cfc046)
1 /*	$NetBSD: nfs_vfsops.c,v 1.127 2003/05/03 16:28:58 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_vfsops.c	8.12 (Berkeley) 5/20/95
39  */
40 
41 #include <sys/cdefs.h>
42 __KERNEL_RCSID(0, "$NetBSD: nfs_vfsops.c,v 1.127 2003/05/03 16:28:58 yamt Exp $");
43 
44 #if defined(_KERNEL_OPT)
45 #include "opt_compat_netbsd.h"
46 #include "opt_nfs.h"
47 #endif
48 
49 #include <sys/param.h>
50 #include <sys/ioctl.h>
51 #include <sys/signal.h>
52 #include <sys/proc.h>
53 #include <sys/namei.h>
54 #include <sys/device.h>
55 #include <sys/vnode.h>
56 #include <sys/kernel.h>
57 #include <sys/mount.h>
58 #include <sys/buf.h>
59 #include <sys/mbuf.h>
60 #include <sys/socket.h>
61 #include <sys/socketvar.h>
62 #include <sys/sysctl.h>
63 #include <sys/systm.h>
64 
65 #include <net/if.h>
66 #include <net/route.h>
67 #include <netinet/in.h>
68 
69 #include <nfs/rpcv2.h>
70 #include <nfs/nfsproto.h>
71 #include <nfs/nfsnode.h>
72 #include <nfs/nfs.h>
73 #include <nfs/nfsmount.h>
74 #include <nfs/xdr_subs.h>
75 #include <nfs/nfsm_subs.h>
76 #include <nfs/nfsdiskless.h>
77 #include <nfs/nqnfs.h>
78 #include <nfs/nfs_var.h>
79 
80 extern struct nfsstats nfsstats;
81 extern int nfs_ticks;
82 
83 MALLOC_DEFINE(M_NFSMNT, "NFS mount", "NFS mount structure");
84 
85 int nfs_sysctl __P((int *, u_int, void *, size_t *, void *, size_t,
86 		      struct proc *));
87 
88 /*
89  * nfs vfs operations.
90  */
91 
92 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
93 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
94 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
95 
96 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = {
97 	&nfsv2_vnodeop_opv_desc,
98 	&spec_nfsv2nodeop_opv_desc,
99 	&fifo_nfsv2nodeop_opv_desc,
100 	NULL,
101 };
102 
103 struct vfsops nfs_vfsops = {
104 	MOUNT_NFS,
105 	nfs_mount,
106 	nfs_start,
107 	nfs_unmount,
108 	nfs_root,
109 	nfs_quotactl,
110 	nfs_statfs,
111 	nfs_sync,
112 	nfs_vget,
113 	nfs_fhtovp,
114 	nfs_vptofh,
115 	nfs_vfs_init,
116 	nfs_vfs_reinit,
117 	nfs_vfs_done,
118 	nfs_sysctl,
119 	nfs_mountroot,
120 	nfs_checkexp,
121 	nfs_vnodeopv_descs,
122 };
123 
124 extern u_int32_t nfs_procids[NFS_NPROCS];
125 extern u_int32_t nfs_prog, nfs_vers;
126 
127 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
128     struct mount **, struct vnode **, struct proc *));
129 
130 #define TRUE	1
131 #define	FALSE	0
132 
133 /*
134  * nfs statfs call
135  */
136 int
137 nfs_statfs(mp, sbp, p)
138 	struct mount *mp;
139 	struct statfs *sbp;
140 	struct proc *p;
141 {
142 	struct vnode *vp;
143 	struct nfs_statfs *sfp;
144 	caddr_t cp;
145 	u_int32_t *tl;
146 	int32_t t1, t2;
147 	caddr_t bpos, dpos, cp2;
148 	struct nfsmount *nmp = VFSTONFS(mp);
149 	int error = 0, retattr;
150 #ifdef NFS_V2_ONLY
151 	const int v3 = 0;
152 #else
153 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
154 #endif
155 	struct mbuf *mreq, *mrep = NULL, *md, *mb;
156 	struct ucred *cred;
157 	u_quad_t tquad;
158 	struct nfsnode *np;
159 
160 #ifndef nolint
161 	sfp = (struct nfs_statfs *)0;
162 #endif
163 	vp = nmp->nm_vnode;
164 	np = VTONFS(vp);
165 	cred = crget();
166 	cred->cr_ngroups = 0;
167 #ifndef NFS_V2_ONLY
168 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
169 		(void)nfs_fsinfo(nmp, vp, cred, p);
170 #endif
171 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
172 	nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3));
173 	nfsm_fhtom(np, v3);
174 	nfsm_request(np, NFSPROC_FSSTAT, p, cred);
175 	if (v3)
176 		nfsm_postop_attr(vp, retattr, 0);
177 	if (error) {
178 		if (mrep != NULL)
179 			m_free(mrep);
180 		goto nfsmout;
181 	}
182 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
183 #ifdef COMPAT_09
184 	sbp->f_type = 2;
185 #else
186 	sbp->f_type = 0;
187 #endif
188 	sbp->f_flags = nmp->nm_flag;
189 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
190 	if (v3) {
191 		sbp->f_bsize = NFS_FABLKSIZE;
192 		tquad = fxdr_hyper(&sfp->sf_tbytes);
193 		sbp->f_blocks = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
194 		tquad = fxdr_hyper(&sfp->sf_fbytes);
195 		sbp->f_bfree = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
196 		tquad = fxdr_hyper(&sfp->sf_abytes);
197 		sbp->f_bavail = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
198 		tquad = fxdr_hyper(&sfp->sf_tfiles);
199 		sbp->f_files = (long)tquad;
200 		tquad = fxdr_hyper(&sfp->sf_ffiles);
201 		sbp->f_ffree = (long)tquad;
202 	} else {
203 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
204 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
205 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
206 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
207 		sbp->f_files = 0;
208 		sbp->f_ffree = 0;
209 	}
210 	copy_statfs_info(sbp, mp);
211 	nfsm_reqdone;
212 	crfree(cred);
213 	return (error);
214 }
215 
216 #ifndef NFS_V2_ONLY
217 /*
218  * nfs version 3 fsinfo rpc call
219  */
220 int
221 nfs_fsinfo(nmp, vp, cred, p)
222 	struct nfsmount *nmp;
223 	struct vnode *vp;
224 	struct ucred *cred;
225 	struct proc *p;
226 {
227 	struct nfsv3_fsinfo *fsp;
228 	caddr_t cp;
229 	int32_t t1, t2;
230 	u_int32_t *tl, pref, max;
231 	caddr_t bpos, dpos, cp2;
232 	int error = 0, retattr;
233 	struct mbuf *mreq, *mrep, *md, *mb;
234 	u_int64_t maxfsize;
235 	struct nfsnode *np = VTONFS(vp);
236 
237 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
238 	nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1));
239 	nfsm_fhtom(np, 1);
240 	nfsm_request(np, NFSPROC_FSINFO, p, cred);
241 	nfsm_postop_attr(vp, retattr, 0);
242 	if (!error) {
243 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
244 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
245 		if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 &&
246 		    pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
247 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
248 				~(NFS_FABLKSIZE - 1);
249 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
250 		if (max < nmp->nm_wsize && max > 0) {
251 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
252 			if (nmp->nm_wsize == 0)
253 				nmp->nm_wsize = max;
254 		}
255 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
256 		if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 &&
257 		    pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
258 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
259 				~(NFS_FABLKSIZE - 1);
260 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
261 		if (max < nmp->nm_rsize && max > 0) {
262 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
263 			if (nmp->nm_rsize == 0)
264 				nmp->nm_rsize = max;
265 		}
266 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
267 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
268 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
269 				~(NFS_DIRFRAGSIZ - 1);
270 		if (max < nmp->nm_readdirsize && max > 0) {
271 			nmp->nm_readdirsize = max & ~(NFS_DIRFRAGSIZ - 1);
272 			if (nmp->nm_readdirsize == 0)
273 				nmp->nm_readdirsize = max;
274 		}
275 		/* XXX */
276 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
277 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
278 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
279 			nmp->nm_maxfilesize = maxfsize;
280 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
281 	}
282 	nfsm_reqdone;
283 	return (error);
284 }
285 #endif
286 
287 /*
288  * Mount a remote root fs via. NFS.  It goes like this:
289  * - Call nfs_boot_init() to fill in the nfs_diskless struct
290  * - build the rootfs mount point and call mountnfs() to do the rest.
291  */
292 int
293 nfs_mountroot()
294 {
295 	struct nfs_diskless *nd;
296 	struct vattr attr;
297 	struct mount *mp;
298 	struct vnode *vp;
299 	struct proc *procp;
300 	long n;
301 	int error;
302 
303 	procp = curproc; /* XXX */
304 
305 	if (root_device->dv_class != DV_IFNET)
306 		return (ENODEV);
307 
308 	/*
309 	 * XXX time must be non-zero when we init the interface or else
310 	 * the arp code will wedge.  [Fixed now in if_ether.c]
311 	 * However, the NFS attribute cache gives false "hits" when
312 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
313 	 */
314 	if (time.tv_sec < NFS_MAXATTRTIMO)
315 		time.tv_sec = NFS_MAXATTRTIMO;
316 
317 	/*
318 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
319 	 * Side effect:  Finds and configures a network interface.
320 	 */
321 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
322 	memset((caddr_t)nd, 0, sizeof(*nd));
323 	error = nfs_boot_init(nd, procp);
324 	if (error) {
325 		free(nd, M_NFSMNT);
326 		return (error);
327 	}
328 
329 	/*
330 	 * Create the root mount point.
331 	 */
332 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp);
333 	if (error)
334 		goto out;
335 	printf("root on %s\n", nd->nd_root.ndm_host);
336 
337 	/*
338 	 * Link it into the mount list.
339 	 */
340 	simple_lock(&mountlist_slock);
341 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
342 	simple_unlock(&mountlist_slock);
343 	rootvp = vp;
344 	mp->mnt_vnodecovered = NULLVP;
345 	vfs_unbusy(mp);
346 
347 	/* Get root attributes (for the time). */
348 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
349 	if (error)
350 		panic("nfs_mountroot: getattr for root");
351 	n = attr.va_atime.tv_sec;
352 #ifdef	DEBUG
353 	printf("root time: 0x%lx\n", n);
354 #endif
355 	inittodr(n);
356 
357 out:
358 	if (error)
359 		nfs_boot_cleanup(nd, procp);
360 	free(nd, M_NFSMNT);
361 	return (error);
362 }
363 
364 /*
365  * Internal version of mount system call for diskless setup.
366  * Separate function because we used to call it twice.
367  * (once for root and once for swap)
368  */
369 static int
370 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p)
371 	struct nfs_dlmount *ndmntp;
372 	const char *mntname;	/* mount point name */
373 	struct mount **mpp;
374 	struct vnode **vpp;
375 	struct proc *p;
376 {
377 	struct mount *mp;
378 	struct mbuf *m;
379 	int error;
380 
381 	vfs_rootmountalloc(MOUNT_NFS, (char *)mntname, &mp);
382 
383 	mp->mnt_op = &nfs_vfsops;
384 
385 	/*
386 	 * Historical practice expects NFS root file systems to
387 	 * be initially mounted r/w.
388 	 */
389 	mp->mnt_flag &= ~MNT_RDONLY;
390 
391 	/* Get mbuf for server sockaddr. */
392 	m = m_get(M_WAIT, MT_SONAME);
393 	if (m == NULL)
394 		panic("nfs_mountroot: mget soname for %s", mntname);
395 	MCLAIM(m, &nfs_mowner);
396 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
397 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
398 
399 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
400 			 ndmntp->ndm_args.hostname, vpp, p);
401 	if (error) {
402 		mp->mnt_op->vfs_refcount--;
403 		vfs_unbusy(mp);
404 		printf("nfs_mountroot: mount %s failed: %d\n",
405 		       mntname, error);
406 		free(mp, M_MOUNT);
407 	} else
408 		*mpp = mp;
409 
410 	return (error);
411 }
412 
413 void
414 nfs_decode_args(nmp, argp)
415 	struct nfsmount *nmp;
416 	struct nfs_args *argp;
417 {
418 	int s;
419 	int adjsock;
420 	int maxio;
421 
422 	s = splsoftnet();
423 
424 	/*
425 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
426 	 * no sense in that context.
427 	 */
428 	if (argp->sotype == SOCK_STREAM)
429 		argp->flags &= ~NFSMNT_NOCONN;
430 
431 	/*
432 	 * Cookie translation is not needed for v2, silently ignore it.
433 	 */
434 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
435 	    NFSMNT_XLATECOOKIE)
436 		argp->flags &= ~NFSMNT_XLATECOOKIE;
437 
438 	/* Re-bind if rsrvd port requested and wasn't on one */
439 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
440 		  && (argp->flags & NFSMNT_RESVPORT);
441 	/* Also re-bind if we're switching to/from a connected UDP socket */
442 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
443 		    (argp->flags & NFSMNT_NOCONN));
444 
445 	/* Update flags. */
446 	nmp->nm_flag = argp->flags;
447 	splx(s);
448 
449 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
450 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
451 		if (nmp->nm_timeo < NFS_MINTIMEO)
452 			nmp->nm_timeo = NFS_MINTIMEO;
453 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
454 			nmp->nm_timeo = NFS_MAXTIMEO;
455 	}
456 
457 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
458 		nmp->nm_retry = argp->retrans;
459 		if (nmp->nm_retry > NFS_MAXREXMIT)
460 			nmp->nm_retry = NFS_MAXREXMIT;
461 	}
462 
463 #ifndef NFS_V2_ONLY
464 	if (argp->flags & NFSMNT_NFSV3) {
465 		if (argp->sotype == SOCK_DGRAM)
466 			maxio = NFS_MAXDGRAMDATA;
467 		else
468 			maxio = NFS_MAXDATA;
469 	} else
470 #endif
471 		maxio = NFS_V2MAXDATA;
472 
473 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
474 		int osize = nmp->nm_wsize;
475 		nmp->nm_wsize = argp->wsize;
476 		/* Round down to multiple of blocksize */
477 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
478 		if (nmp->nm_wsize <= 0)
479 			nmp->nm_wsize = NFS_FABLKSIZE;
480 		adjsock |= (nmp->nm_wsize != osize);
481 	}
482 	if (nmp->nm_wsize > maxio)
483 		nmp->nm_wsize = maxio;
484 	if (nmp->nm_wsize > MAXBSIZE)
485 		nmp->nm_wsize = MAXBSIZE;
486 
487 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
488 		int osize = nmp->nm_rsize;
489 		nmp->nm_rsize = argp->rsize;
490 		/* Round down to multiple of blocksize */
491 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
492 		if (nmp->nm_rsize <= 0)
493 			nmp->nm_rsize = NFS_FABLKSIZE;
494 		adjsock |= (nmp->nm_rsize != osize);
495 	}
496 	if (nmp->nm_rsize > maxio)
497 		nmp->nm_rsize = maxio;
498 	if (nmp->nm_rsize > MAXBSIZE)
499 		nmp->nm_rsize = MAXBSIZE;
500 
501 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
502 		nmp->nm_readdirsize = argp->readdirsize;
503 		/* Round down to multiple of minimum blocksize */
504 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
505 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
506 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
507 		/* Bigger than buffer size makes no sense */
508 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
509 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
510 	} else if (argp->flags & NFSMNT_RSIZE)
511 		nmp->nm_readdirsize = nmp->nm_rsize;
512 
513 	if (nmp->nm_readdirsize > maxio)
514 		nmp->nm_readdirsize = maxio;
515 
516 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
517 		argp->maxgrouplist <= NFS_MAXGRPS)
518 		nmp->nm_numgrps = argp->maxgrouplist;
519 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
520 		argp->readahead <= NFS_MAXRAHEAD)
521 		nmp->nm_readahead = argp->readahead;
522 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
523 		argp->leaseterm <= NQ_MAXLEASE)
524 		nmp->nm_leaseterm = argp->leaseterm;
525 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
526 		argp->deadthresh <= NQ_NEVERDEAD)
527 		nmp->nm_deadthresh = argp->deadthresh;
528 
529 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
530 		    (nmp->nm_soproto != argp->proto));
531 	nmp->nm_sotype = argp->sotype;
532 	nmp->nm_soproto = argp->proto;
533 
534 	if (nmp->nm_so && adjsock) {
535 		nfs_safedisconnect(nmp);
536 		if (nmp->nm_sotype == SOCK_DGRAM)
537 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
538 				printf("nfs_args: retrying connect\n");
539 				(void) tsleep((caddr_t)&lbolt,
540 					      PSOCK, "nfscn3", 0);
541 			}
542 	}
543 }
544 
545 /*
546  * VFS Operations.
547  *
548  * mount system call
549  * It seems a bit dumb to copyinstr() the host and path here and then
550  * memcpy() them in mountnfs(), but I wanted to detect errors before
551  * doing the sockargs() call because sockargs() allocates an mbuf and
552  * an error after that means that I have to release the mbuf.
553  */
554 /* ARGSUSED */
555 int
556 nfs_mount(mp, path, data, ndp, p)
557 	struct mount *mp;
558 	const char *path;
559 	void *data;
560 	struct nameidata *ndp;
561 	struct proc *p;
562 {
563 	int error;
564 	struct nfs_args args;
565 	struct mbuf *nam;
566 	struct nfsmount *nmp = VFSTONFS(mp);
567 	struct sockaddr *sa;
568 	struct vnode *vp;
569 	char *pth, *hst;
570 	size_t len;
571 	u_char *nfh;
572 
573 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
574 	if (error)
575 		return (error);
576 
577 	if (mp->mnt_flag & MNT_GETARGS) {
578 
579 		if (nmp == NULL)
580 			return (EIO);
581 		if (args.addr != NULL) {
582 			sa = mtod(nmp->nm_nam, struct sockaddr *);
583 			error = copyout(sa, args.addr, sa->sa_len);
584 			if (error)
585 				return (error);
586 			args.addrlen = sa->sa_len;
587 		} else
588 			args.addrlen = 0;
589 
590 		args.version = NFS_ARGSVERSION;
591 		args.sotype = nmp->nm_sotype;
592 		args.proto = nmp->nm_soproto;
593 		args.fh = NULL;
594 		args.fhsize = 0;
595 		args.flags = nmp->nm_flag;
596 		args.wsize = nmp->nm_wsize;
597 		args.rsize = nmp->nm_rsize;
598 		args.readdirsize = nmp->nm_readdirsize;
599 		args.timeo = nmp->nm_timeo;
600 		args.retrans = nmp->nm_retry;
601 		args.maxgrouplist = nmp->nm_numgrps;
602 		args.readahead = nmp->nm_readahead;
603 		args.leaseterm = nmp->nm_leaseterm;
604 		args.deadthresh = nmp->nm_deadthresh;
605 		args.hostname = NULL;
606 		return (copyout(&args, data, sizeof(args)));
607 	}
608 
609 	if (args.version != NFS_ARGSVERSION)
610 		return (EPROGMISMATCH);
611 #ifdef NFS_V2_ONLY
612 	if (args.flags & NFSMNT_NQNFS)
613 		return (EPROGUNAVAIL);
614 	if (args.flags & NFSMNT_NFSV3)
615 		return (EPROGMISMATCH);
616 #endif
617 	if (mp->mnt_flag & MNT_UPDATE) {
618 		if (nmp == NULL)
619 			return (EIO);
620 		/*
621 		 * When doing an update, we can't change from or to
622 		 * v3 and/or nqnfs, or change cookie translation
623 		 */
624 		args.flags = (args.flags &
625 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
626 		    (nmp->nm_flag &
627 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
628 		nfs_decode_args(nmp, &args);
629 		return (0);
630 	}
631 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
632 		return (EINVAL);
633 	MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK);
634 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
635 	if (error)
636 		return (error);
637 	MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK);
638 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
639 	if (error)
640 		goto free_nfh;
641 	memset(&pth[len], 0, MNAMELEN - len);
642 	MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK);
643 	error = copyinstr(args.hostname, hst, MNAMELEN - 1, &len);
644 	if (error)
645 		goto free_pth;
646 	memset(&hst[len], 0, MNAMELEN - len);
647 	/* sockargs() call must be after above copyin() calls */
648 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
649 	if (error)
650 		goto free_hst;
651 	MCLAIM(nam, &nfs_mowner);
652 	args.fh = nfh;
653 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
654 
655 free_hst:
656 	FREE(hst, M_TEMP);
657 free_pth:
658 	FREE(pth, M_TEMP);
659 free_nfh:
660 	FREE(nfh, M_TEMP);
661 
662 	return (error);
663 }
664 
665 /*
666  * Common code for mount and mountroot
667  */
668 int
669 mountnfs(argp, mp, nam, pth, hst, vpp, p)
670 	struct nfs_args *argp;
671 	struct mount *mp;
672 	struct mbuf *nam;
673 	const char *pth, *hst;
674 	struct vnode **vpp;
675 	struct proc *p;
676 {
677 	struct nfsmount *nmp;
678 	struct nfsnode *np;
679 	int error;
680 	struct vattr *attrs;
681 	struct ucred *cr;
682 
683 	/*
684 	 * If the number of nfs iothreads to use has never
685 	 * been set, create a reasonable number of them.
686 	 */
687 
688 	if (nfs_niothreads < 0) {
689 		nfs_niothreads = NFS_DEFAULT_NIOTHREADS;
690 		nfs_getset_niothreads(TRUE);
691 	}
692 
693 	if (mp->mnt_flag & MNT_UPDATE) {
694 		nmp = VFSTONFS(mp);
695 		/* update paths, file handles, etc, here	XXX */
696 		m_freem(nam);
697 		return (0);
698 	} else {
699 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
700 		    M_NFSMNT, M_WAITOK);
701 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
702 		mp->mnt_data = nmp;
703 		TAILQ_INIT(&nmp->nm_uidlruhead);
704 		TAILQ_INIT(&nmp->nm_bufq);
705 		lockinit(&nmp->nm_writeverflock, PRIBIO, "nfswverf", 0, 0);
706 		simple_lock_init(&nmp->nm_slock);
707 	}
708 	vfs_getnewfsid(mp);
709 	nmp->nm_mountp = mp;
710 
711 #ifndef NFS_V2_ONLY
712 	if (argp->flags & NFSMNT_NQNFS)
713 		/*
714 		 * We have to set mnt_maxsymlink to a non-zero value so
715 		 * that COMPAT_43 routines will know that we are setting
716 		 * the d_type field in directories (and can zero it for
717 		 * unsuspecting binaries).
718 		 */
719 		mp->mnt_maxsymlinklen = 1;
720 #endif
721 
722 #ifndef NFS_V2_ONLY
723 	if ((argp->flags & NFSMNT_NFSV3) == 0)
724 #endif
725 		/*
726 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
727 		 * will fill this in.
728 		 */
729 		nmp->nm_maxfilesize = 0xffffffffLL;
730 
731 	nmp->nm_timeo = NFS_TIMEO;
732 	nmp->nm_retry = NFS_RETRANS;
733 	nmp->nm_wsize = NFS_WSIZE;
734 	nmp->nm_rsize = NFS_RSIZE;
735 	nmp->nm_readdirsize = NFS_READDIRSIZE;
736 	nmp->nm_numgrps = NFS_MAXGRPS;
737 	nmp->nm_readahead = NFS_DEFRAHEAD;
738 	nmp->nm_leaseterm = NQ_DEFLEASE;
739 	nmp->nm_deadthresh = NQ_DEADTHRESH;
740 	CIRCLEQ_INIT(&nmp->nm_timerhead);
741 	nmp->nm_inprog = NULLVP;
742 #ifdef COMPAT_09
743 	mp->mnt_stat.f_type = 2;
744 #else
745 	mp->mnt_stat.f_type = 0;
746 #endif
747 	error = set_statfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, mp, p);
748 	if (error)
749 		goto bad;
750 	nmp->nm_nam = nam;
751 
752 	/* Set up the sockets and per-host congestion */
753 	nmp->nm_sotype = argp->sotype;
754 	nmp->nm_soproto = argp->proto;
755 
756 	nfs_decode_args(nmp, argp);
757 
758 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
759 	mp->mnt_dev_bshift = DEV_BSHIFT;
760 
761 	/*
762 	 * For Connection based sockets (TCP,...) defer the connect until
763 	 * the first request, in case the server is not responding.
764 	 */
765 	if (nmp->nm_sotype == SOCK_DGRAM &&
766 		(error = nfs_connect(nmp, (struct nfsreq *)0)))
767 		goto bad;
768 
769 	/*
770 	 * This is silly, but it has to be set so that vinifod() works.
771 	 * We do not want to do an nfs_statfs() here since we can get
772 	 * stuck on a dead server and we are holding a lock on the mount
773 	 * point.
774 	 */
775 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
776 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
777 	if (error)
778 		goto bad;
779 	*vpp = NFSTOV(np);
780 	MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK);
781 	VOP_GETATTR(*vpp, attrs, p->p_ucred, p);
782 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
783 		cr = crget();
784 		cr->cr_uid = attrs->va_uid;
785 		cr->cr_gid = attrs->va_gid;
786 		cr->cr_ngroups = 0;
787 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
788 		crfree(cr);
789 	}
790 	FREE(attrs, M_TEMP);
791 
792 	/*
793 	 * A reference count is needed on the nfsnode representing the
794 	 * remote root.  If this object is not persistent, then backward
795 	 * traversals of the mount point (i.e. "..") will not work if
796 	 * the nfsnode gets flushed out of the cache. Ufs does not have
797 	 * this problem, because one can identify root inodes by their
798 	 * number == ROOTINO (2). So, just unlock, but no rele.
799 	 */
800 
801 	nmp->nm_vnode = *vpp;
802 	VOP_UNLOCK(*vpp, 0);
803 
804 	return (0);
805 bad:
806 	nfs_disconnect(nmp);
807 	free((caddr_t)nmp, M_NFSMNT);
808 	m_freem(nam);
809 	return (error);
810 }
811 
812 /*
813  * unmount system call
814  */
815 int
816 nfs_unmount(mp, mntflags, p)
817 	struct mount *mp;
818 	int mntflags;
819 	struct proc *p;
820 {
821 	struct nfsmount *nmp;
822 	struct vnode *vp;
823 	int error, flags = 0;
824 
825 	if (mntflags & MNT_FORCE)
826 		flags |= FORCECLOSE;
827 	nmp = VFSTONFS(mp);
828 	/*
829 	 * Goes something like this..
830 	 * - Check for activity on the root vnode (other than ourselves).
831 	 * - Call vflush() to clear out vnodes for this file system,
832 	 *   except for the root vnode.
833 	 * - Decrement reference on the vnode representing remote root.
834 	 * - Close the socket
835 	 * - Free up the data structures
836 	 */
837 	/*
838 	 * We need to decrement the ref. count on the nfsnode representing
839 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
840 	 * has done vput on this vnode, otherwise we would get deadlock!
841 	 */
842 	vp = nmp->nm_vnode;
843 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
844 	if (error != 0)
845 		return error;
846 
847 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
848 		vput(vp);
849 		return (EBUSY);
850 	}
851 
852 	/*
853 	 * Must handshake with nqnfs_clientd() if it is active.
854 	 */
855 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
856 	while (nmp->nm_inprog != NULLVP)
857 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
858 	error = vflush(mp, vp, flags);
859 	if (error) {
860 		vput(vp);
861 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
862 		return (error);
863 	}
864 
865 	/*
866 	 * We are now committed to the unmount; mark the mount structure
867 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
868 	 * will go away cleanly.
869 	 */
870 	nmp->nm_iflag |= NFSMNT_DISMNT;
871 
872 	/*
873 	 * There are two reference counts to get rid of here
874 	 * (see comment in mountnfs()).
875 	 */
876 	vrele(vp);
877 	vput(vp);
878 	vgone(vp);
879 	nfs_disconnect(nmp);
880 	m_freem(nmp->nm_nam);
881 
882 	/*
883 	 * For NQNFS, let the server daemon free the nfsmount structure.
884 	 */
885 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
886 		free((caddr_t)nmp, M_NFSMNT);
887 	return (0);
888 }
889 
890 /*
891  * Return root of a filesystem
892  */
893 int
894 nfs_root(mp, vpp)
895 	struct mount *mp;
896 	struct vnode **vpp;
897 {
898 	struct vnode *vp;
899 	struct nfsmount *nmp;
900 	int error;
901 
902 	nmp = VFSTONFS(mp);
903 	vp = nmp->nm_vnode;
904 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
905 	if (error != 0)
906 		return error;
907 	if (vp->v_type == VNON)
908 		vp->v_type = VDIR;
909 	vp->v_flag = VROOT;
910 	*vpp = vp;
911 	return (0);
912 }
913 
914 extern int syncprt;
915 
916 /*
917  * Flush out the buffer cache
918  */
919 /* ARGSUSED */
920 int
921 nfs_sync(mp, waitfor, cred, p)
922 	struct mount *mp;
923 	int waitfor;
924 	struct ucred *cred;
925 	struct proc *p;
926 {
927 	struct vnode *vp;
928 	int error, allerror = 0;
929 
930 	/*
931 	 * Force stale buffer cache information to be flushed.
932 	 */
933 loop:
934 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
935 		/*
936 		 * If the vnode that we are about to sync is no longer
937 		 * associated with this mount point, start over.
938 		 */
939 		if (vp->v_mount != mp)
940 			goto loop;
941 		if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) ||
942 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
943 		     vp->v_uobj.uo_npages == 0))
944 			continue;
945 		if (vget(vp, LK_EXCLUSIVE))
946 			goto loop;
947 		error = VOP_FSYNC(vp, cred,
948 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p);
949 		if (error)
950 			allerror = error;
951 		vput(vp);
952 	}
953 	return (allerror);
954 }
955 
956 /*
957  * NFS flat namespace lookup.
958  * Currently unsupported.
959  */
960 /* ARGSUSED */
961 int
962 nfs_vget(mp, ino, vpp)
963 	struct mount *mp;
964 	ino_t ino;
965 	struct vnode **vpp;
966 {
967 
968 	return (EOPNOTSUPP);
969 }
970 
971 /*
972  * Do that sysctl thang...
973  */
974 int
975 nfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
976 	int *name;
977 	u_int namelen;
978 	void *oldp;
979 	size_t *oldlenp;
980 	void *newp;
981 	size_t newlen;
982 	struct proc *p;
983 {
984 	int rv;
985 
986 	/*
987 	 * All names at this level are terminal.
988 	 */
989 	if(namelen > 1)
990 		return ENOTDIR;	/* overloaded */
991 
992 	switch(name[0]) {
993 	case NFS_NFSSTATS:
994 		if(!oldp) {
995 			*oldlenp = sizeof nfsstats;
996 			return 0;
997 		}
998 
999 		if(*oldlenp < sizeof nfsstats) {
1000 			*oldlenp = sizeof nfsstats;
1001 			return ENOMEM;
1002 		}
1003 
1004 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
1005 		if(rv) return rv;
1006 
1007 		if(newp && newlen != sizeof nfsstats)
1008 			return EINVAL;
1009 
1010 		if(newp) {
1011 			return copyin(newp, &nfsstats, sizeof nfsstats);
1012 		}
1013 		return 0;
1014 
1015 	case NFS_IOTHREADS:
1016 		nfs_getset_niothreads(0);
1017 
1018 		rv = (sysctl_int(oldp, oldlenp, newp, newlen,
1019 		    &nfs_niothreads));
1020 
1021 		if (newp)
1022 			nfs_getset_niothreads(1);
1023 
1024 		return rv;
1025 
1026 	default:
1027 		return EOPNOTSUPP;
1028 	}
1029 }
1030 
1031 
1032 /*
1033  * At this point, this should never happen
1034  */
1035 /* ARGSUSED */
1036 int
1037 nfs_fhtovp(mp, fhp, vpp)
1038 	struct mount *mp;
1039 	struct fid *fhp;
1040 	struct vnode **vpp;
1041 {
1042 
1043 	return (EINVAL);
1044 }
1045 
1046 /* ARGSUSED */
1047 int
1048 nfs_checkexp(mp, nam, exflagsp, credanonp)
1049 	struct mount *mp;
1050 	struct mbuf *nam;
1051 	int *exflagsp;
1052 	struct ucred **credanonp;
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, p)
1077 	struct mount *mp;
1078 	int flags;
1079 	struct proc *p;
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, p)
1091 	struct mount *mp;
1092 	int cmd;
1093 	uid_t uid;
1094 	caddr_t arg;
1095 	struct proc *p;
1096 {
1097 
1098 	return (EOPNOTSUPP);
1099 }
1100