xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision ce63d6c20fc4ec8ddc95c84bb229e3c4ecf82b69)
1 /*
2  * Copyright (c) 1989 The Regents of the University of California.
3  * All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	from: @(#)nfs_vfsops.c	7.31 (Berkeley) 5/6/91
37  *	$Id: nfs_vfsops.c,v 1.2 1993/05/20 03:18:53 cgd Exp $
38  */
39 
40 #include "param.h"
41 #include "conf.h"
42 #include "ioctl.h"
43 #include "signal.h"
44 #include "proc.h"
45 #include "namei.h"
46 #include "vnode.h"
47 #include "mount.h"
48 #include "buf.h"
49 #include "mbuf.h"
50 #include "socket.h"
51 #include "systm.h"
52 
53 #include "../net/if.h"
54 #include "../net/route.h"
55 #include "../netinet/in.h"
56 
57 #include "nfsv2.h"
58 #include "nfsnode.h"
59 #include "nfsmount.h"
60 #include "nfs.h"
61 #include "xdr_subs.h"
62 #include "nfsm_subs.h"
63 #include "nfsdiskless.h"
64 
65 /*
66  * nfs vfs operations.
67  */
68 struct vfsops nfs_vfsops = {
69 	nfs_mount,
70 	nfs_start,
71 	nfs_unmount,
72 	nfs_root,
73 	nfs_quotactl,
74 	nfs_statfs,
75 	nfs_sync,
76 	nfs_fhtovp,
77 	nfs_vptofh,
78 	nfs_init,
79 };
80 
81 static u_char nfs_mntid;
82 extern u_long nfs_procids[NFS_NPROCS];
83 extern u_long nfs_prog, nfs_vers;
84 struct nfs_diskless nfs_diskless;
85 void nfs_disconnect();
86 
87 #define TRUE	1
88 #define	FALSE	0
89 
90 /*
91  * nfs statfs call
92  */
93 nfs_statfs(mp, sbp, p)
94 	struct mount *mp;
95 	register struct statfs *sbp;
96 	struct proc *p;
97 {
98 	register struct vnode *vp;
99 	register struct nfsv2_statfs *sfp;
100 	register caddr_t cp;
101 	register long t1;
102 	caddr_t bpos, dpos, cp2;
103 	u_long xid;
104 	int error = 0;
105 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
106 	struct nfsmount *nmp;
107 	struct ucred *cred;
108 	struct nfsnode *np;
109 
110 	nmp = VFSTONFS(mp);
111 	if (error = nfs_nget(mp, &nmp->nm_fh, &np))
112 		return (error);
113 	vp = NFSTOV(np);
114 	nfsstats.rpccnt[NFSPROC_STATFS]++;
115 	cred = crget();
116 	cred->cr_ngroups = 1;
117 	nfsm_reqhead(nfs_procids[NFSPROC_STATFS], cred, NFSX_FH);
118 	nfsm_fhtom(vp);
119 	nfsm_request(vp, NFSPROC_STATFS, p, 0);
120 	nfsm_disect(sfp, struct nfsv2_statfs *, NFSX_STATFS);
121 	sbp->f_type = MOUNT_NFS;
122 	sbp->f_flags = nmp->nm_flag;
123 	sbp->f_bsize = fxdr_unsigned(long, sfp->sf_tsize);
124 	sbp->f_fsize = fxdr_unsigned(long, sfp->sf_bsize);
125 	sbp->f_blocks = fxdr_unsigned(long, sfp->sf_blocks);
126 	sbp->f_bfree = fxdr_unsigned(long, sfp->sf_bfree);
127 	sbp->f_bavail = fxdr_unsigned(long, sfp->sf_bavail);
128 	sbp->f_files = 0;
129 	sbp->f_ffree = 0;
130 	if (sbp != &mp->mnt_stat) {
131 		bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
132 		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
133 	}
134 	nfsm_reqdone;
135 	nfs_nput(vp);
136 	crfree(cred);
137 	return (error);
138 }
139 
140 /*
141  * Mount a remote root fs via. nfs. This depends on the info in the
142  * nfs_diskless structure that has been filled in properly by some primary
143  * bootstrap.
144  * It goes something like this:
145  * - do enough of "ifconfig" by calling ifioctl() so that the system
146  *   can talk to the server
147  * - If nfs_diskless.mygateway is filled in, use that address as
148  *   a default gateway.
149  *   (This is done the 4.3 way with rtioctl() and should be changed)
150  * - hand craft the swap nfs vnode hanging off a fake mount point
151  * - build the rootfs mount point and call mountnfs() to do the rest.
152  */
153 nfs_mountroot()
154 {
155 	register struct mount *mp;
156 	register struct mbuf *m;
157 	struct socket *so;
158 	struct vnode *vp;
159 	int error;
160 
161 	/*
162 	 * Do enough of ifconfig(8) so that critical net interface can
163 	 * talk to the server.
164 	 */
165 	if (socreate(nfs_diskless.myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0))
166 		panic("nfs ifconf");
167 	if (ifioctl(so, SIOCAIFADDR, &nfs_diskless.myif))
168 		panic("nfs ifconf2");
169 	soclose(so);
170 
171 	/*
172 	 * If the gateway field is filled in, set it as the default route.
173 	 */
174 #ifdef COMPAT_43
175 	if (nfs_diskless.mygateway.sa_family == AF_INET) {
176 		struct ortentry rt;
177 		struct sockaddr_in *sin;
178 
179 		sin = (struct sockaddr_in *) &rt.rt_dst;
180 		sin->sin_len = sizeof (struct sockaddr_in);
181 		sin->sin_family = AF_INET;
182 		sin->sin_addr.s_addr = 0;	/* default */
183 		bcopy((caddr_t)&nfs_diskless.mygateway, (caddr_t)&rt.rt_gateway,
184 			sizeof (struct sockaddr_in));
185 		rt.rt_flags = (RTF_UP | RTF_GATEWAY);
186 		if (rtioctl(SIOCADDRT, (caddr_t)&rt))
187 			panic("nfs root route");
188 	}
189 #endif	/* COMPAT_43 */
190 
191 	/*
192 	 * If swapping to an nfs node (indicated by swdevt[0].sw_dev == NODEV):
193 	 * Create a fake mount point just for the swap vnode so that the
194 	 * swap file can be on a different server from the rootfs.
195 	 */
196 	if (swdevt[0].sw_dev == NODEV) {
197 		mp = (struct mount *)malloc((u_long)sizeof(struct mount),
198 			M_MOUNT, M_NOWAIT);
199 		if (mp == NULL)
200 			panic("nfs root mount");
201 		mp->mnt_op = &nfs_vfsops;
202 		mp->mnt_flag = 0;
203 		mp->mnt_exroot = 0;
204 		mp->mnt_mounth = NULLVP;
205 
206 		/*
207 		 * Set up the diskless nfs_args for the swap mount point
208 		 * and then call mountnfs() to mount it.
209 		 * Since the swap file is not the root dir of a file system,
210 		 * hack it to a regular file.
211 		 */
212 		nfs_diskless.swap_args.fh = (nfsv2fh_t *)nfs_diskless.swap_fh;
213 		MGET(m, MT_SONAME, M_DONTWAIT);
214 		if (m == NULL)
215 			panic("nfs root mbuf");
216 		bcopy((caddr_t)&nfs_diskless.swap_saddr, mtod(m, caddr_t),
217 			nfs_diskless.swap_saddr.sa_len);
218 		m->m_len = nfs_diskless.swap_saddr.sa_len;
219 		if (mountnfs(&nfs_diskless.swap_args, mp, m, "/swap",
220 			nfs_diskless.swap_hostnam, &vp))
221 			panic("nfs swap");
222 		vp->v_type = VREG;
223 		vp->v_flag = 0;
224 		swapdev_vp = vp;
225 		VREF(vp);
226 		swdevt[0].sw_vp = vp;
227 	}
228 
229 	/*
230 	 * Create the rootfs mount point.
231 	 */
232 	mp = (struct mount *)malloc((u_long)sizeof(struct mount),
233 		M_MOUNT, M_NOWAIT);
234 	if (mp == NULL)
235 		panic("nfs root mount2");
236 	mp->mnt_op = &nfs_vfsops;
237 	mp->mnt_flag = MNT_RDONLY;
238 	mp->mnt_exroot = 0;
239 	mp->mnt_mounth = NULLVP;
240 
241 	/*
242 	 * Set up the root fs args and call mountnfs() to do the rest.
243 	 */
244 	nfs_diskless.root_args.fh = (nfsv2fh_t *)nfs_diskless.root_fh;
245 	MGET(m, MT_SONAME, M_DONTWAIT);
246 	if (m == NULL)
247 		panic("nfs root mbuf2");
248 	bcopy((caddr_t)&nfs_diskless.root_saddr, mtod(m, caddr_t),
249 		nfs_diskless.root_saddr.sa_len);
250 	m->m_len = nfs_diskless.root_saddr.sa_len;
251 	if (mountnfs(&nfs_diskless.root_args, mp, m, "/",
252 		nfs_diskless.root_hostnam, &vp))
253 		panic("nfs root");
254 	if (vfs_lock(mp))
255 		panic("nfs root2");
256 	rootfs = mp;
257 	mp->mnt_next = mp;
258 	mp->mnt_prev = mp;
259 	mp->mnt_vnodecovered = NULLVP;
260 	vfs_unlock(mp);
261 	rootvp = vp;
262 	inittodr((time_t)0);	/* There is no time in the nfs fsstat so ?? */
263 	return (0);
264 }
265 
266 /*
267  * VFS Operations.
268  *
269  * mount system call
270  * It seems a bit dumb to copyinstr() the host and path here and then
271  * bcopy() them in mountnfs(), but I wanted to detect errors before
272  * doing the sockargs() call because sockargs() allocates an mbuf and
273  * an error after that means that I have to release the mbuf.
274  */
275 /* ARGSUSED */
276 nfs_mount(mp, path, data, ndp, p)
277 	struct mount *mp;
278 	char *path;
279 	caddr_t data;
280 	struct nameidata *ndp;
281 	struct proc *p;
282 {
283 	int error;
284 	struct nfs_args args;
285 	struct mbuf *nam;
286 	struct vnode *vp;
287 	char pth[MNAMELEN], hst[MNAMELEN];
288 	u_int len;
289 	nfsv2fh_t nfh;
290 
291 	if (mp->mnt_flag & MNT_UPDATE)
292 		return (0);
293 	if (error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)))
294 		return (error);
295 	if (error = copyin((caddr_t)args.fh, (caddr_t)&nfh, sizeof (nfsv2fh_t)))
296 		return (error);
297 	if (error = copyinstr(path, pth, MNAMELEN-1, &len))
298 		return (error);
299 	bzero(&pth[len], MNAMELEN - len);
300 	if (error = copyinstr(args.hostname, hst, MNAMELEN-1, &len))
301 		return (error);
302 	bzero(&hst[len], MNAMELEN - len);
303 	/* sockargs() call must be after above copyin() calls */
304 	if (error = sockargs(&nam, (caddr_t)args.addr,
305 		sizeof (struct sockaddr), MT_SONAME))
306 		return (error);
307 	args.fh = &nfh;
308 	error = mountnfs(&args, mp, nam, pth, hst, &vp);
309 	return (error);
310 }
311 
312 /*
313  * Common code for mount and mountroot
314  */
315 mountnfs(argp, mp, nam, pth, hst, vpp)
316 	register struct nfs_args *argp;
317 	register struct mount *mp;
318 	struct mbuf *nam;
319 	char *pth, *hst;
320 	struct vnode **vpp;
321 {
322 	register struct nfsmount *nmp;
323 	struct proc *p = curproc;		/* XXX */
324 	struct nfsnode *np;
325 	int error;
326 	fsid_t tfsid;
327 
328 	MALLOC(nmp, struct nfsmount *, sizeof *nmp, M_NFSMNT, M_WAITOK);
329 	bzero((caddr_t)nmp, sizeof *nmp);
330 	mp->mnt_data = (qaddr_t)nmp;
331 	/*
332 	 * Generate a unique nfs mount id. The problem is that a dev number
333 	 * is not unique across multiple systems. The techique is as follows:
334 	 * 1) Set to nblkdev,0 which will never be used otherwise
335 	 * 2) Generate a first guess as nblkdev,nfs_mntid where nfs_mntid is
336 	 *	NOT 0
337 	 * 3) Loop searching the mount list for another one with same id
338 	 *	If a match, increment val[0] and try again
339 	 * NB: I increment val[0] { a long } instead of nfs_mntid { a u_char }
340 	 *	so that nfs is not limited to 255 mount points
341 	 *     Incrementing the high order bits does no real harm, since it
342 	 *     simply makes the major dev number tick up. The upper bound is
343 	 *     set to major dev 127 to avoid any sign extention problems
344 	 */
345 	mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev, 0);
346 	mp->mnt_stat.f_fsid.val[1] = MOUNT_NFS;
347 	if (++nfs_mntid == 0)
348 		++nfs_mntid;
349 	tfsid.val[0] = makedev(nblkdev, nfs_mntid);
350 	tfsid.val[1] = MOUNT_NFS;
351 	while (rootfs && getvfs(&tfsid)) {
352 		tfsid.val[0]++;
353 		nfs_mntid++;
354 	}
355 	if (major(tfsid.val[0]) > 127) {
356 		error = ENOENT;
357 		goto bad;
358 	}
359 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
360 	nmp->nm_mountp = mp;
361 	nmp->nm_flag = argp->flags;
362 	nmp->nm_rto = NFS_TIMEO;
363 	nmp->nm_rtt = -1;
364 	nmp->nm_rttvar = nmp->nm_rto << 1;
365 	nmp->nm_retry = NFS_RETRANS;
366 	nmp->nm_wsize = NFS_WSIZE;
367 	nmp->nm_rsize = NFS_RSIZE;
368 	bcopy((caddr_t)argp->fh, (caddr_t)&nmp->nm_fh, sizeof(nfsv2fh_t));
369 	bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
370 	bcopy(pth, mp->mnt_stat.f_mntonname, MNAMELEN);
371 	nmp->nm_nam = nam;
372 
373 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
374 		nmp->nm_rto = argp->timeo;
375 		/* NFS timeouts are specified in 1/10 sec. */
376 		nmp->nm_rto = (nmp->nm_rto * 10) / NFS_HZ;
377 		if (nmp->nm_rto < NFS_MINTIMEO)
378 			nmp->nm_rto = NFS_MINTIMEO;
379 		else if (nmp->nm_rto > NFS_MAXTIMEO)
380 			nmp->nm_rto = NFS_MAXTIMEO;
381 		nmp->nm_rttvar = nmp->nm_rto << 1;
382 	}
383 
384 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
385 		nmp->nm_retry = argp->retrans;
386 		if (nmp->nm_retry > NFS_MAXREXMIT)
387 			nmp->nm_retry = NFS_MAXREXMIT;
388 	}
389 
390 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
391 		nmp->nm_wsize = argp->wsize;
392 		/* Round down to multiple of blocksize */
393 		nmp->nm_wsize &= ~0x1ff;
394 		if (nmp->nm_wsize <= 0)
395 			nmp->nm_wsize = 512;
396 		else if (nmp->nm_wsize > NFS_MAXDATA)
397 			nmp->nm_wsize = NFS_MAXDATA;
398 	}
399 	if (nmp->nm_wsize > MAXBSIZE)
400 		nmp->nm_wsize = MAXBSIZE;
401 
402 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
403 		nmp->nm_rsize = argp->rsize;
404 		/* Round down to multiple of blocksize */
405 		nmp->nm_rsize &= ~0x1ff;
406 		if (nmp->nm_rsize <= 0)
407 			nmp->nm_rsize = 512;
408 		else if (nmp->nm_rsize > NFS_MAXDATA)
409 			nmp->nm_rsize = NFS_MAXDATA;
410 	}
411 	if (nmp->nm_rsize > MAXBSIZE)
412 		nmp->nm_rsize = MAXBSIZE;
413 	/* Set up the sockets and per-host congestion */
414 	nmp->nm_sotype = argp->sotype;
415 	nmp->nm_soproto = argp->proto;
416 	if (error = nfs_connect(nmp))
417 		goto bad;
418 
419 	if (error = nfs_statfs(mp, &mp->mnt_stat, p))
420 		goto bad;
421 	/*
422 	 * A reference count is needed on the nfsnode representing the
423 	 * remote root.  If this object is not persistent, then backward
424 	 * traversals of the mount point (i.e. "..") will not work if
425 	 * the nfsnode gets flushed out of the cache. Ufs does not have
426 	 * this problem, because one can identify root inodes by their
427 	 * number == ROOTINO (2).
428 	 */
429 	if (error = nfs_nget(mp, &nmp->nm_fh, &np))
430 		goto bad;
431 	/*
432 	 * Unlock it, but keep the reference count.
433 	 */
434 	nfs_unlock(NFSTOV(np));
435 	*vpp = NFSTOV(np);
436 
437 	return (0);
438 bad:
439 	nfs_disconnect(nmp);
440 	FREE(nmp, M_NFSMNT);
441 	m_freem(nam);
442 	return (error);
443 }
444 
445 /*
446  * unmount system call
447  */
448 nfs_unmount(mp, mntflags, p)
449 	struct mount *mp;
450 	int mntflags;
451 	struct proc *p;
452 {
453 	register struct nfsmount *nmp;
454 	struct nfsnode *np;
455 	struct vnode *vp;
456 	int error, flags = 0;
457 	extern int doforce;
458 
459 	if (mntflags & MNT_FORCE) {
460 		if (!doforce || mp == rootfs)
461 			return (EINVAL);
462 		flags |= FORCECLOSE;
463 	}
464 	nmp = VFSTONFS(mp);
465 	/*
466 	 * Clear out the buffer cache
467 	 */
468 	mntflushbuf(mp, 0);
469 	if (mntinvalbuf(mp))
470 		return (EBUSY);
471 	/*
472 	 * Goes something like this..
473 	 * - Check for activity on the root vnode (other than ourselves).
474 	 * - Call vflush() to clear out vnodes for this file system,
475 	 *   except for the root vnode.
476 	 * - Decrement reference on the vnode representing remote root.
477 	 * - Close the socket
478 	 * - Free up the data structures
479 	 */
480 	/*
481 	 * We need to decrement the ref. count on the nfsnode representing
482 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
483 	 * has done vput on this vnode, otherwise we would get deadlock!
484 	 */
485 	if (error = nfs_nget(mp, &nmp->nm_fh, &np))
486 		return(error);
487 	vp = NFSTOV(np);
488 	if (vp->v_usecount > 2) {
489 		vput(vp);
490 		return (EBUSY);
491 	}
492 	if (error = vflush(mp, vp, flags)) {
493 		vput(vp);
494 		return (error);
495 	}
496 	/*
497 	 * Get rid of two reference counts, and unlock it on the second.
498 	 */
499 	vrele(vp);
500 	vput(vp);
501 	nfs_disconnect(nmp);
502 	m_freem(nmp->nm_nam);
503 	free((caddr_t)nmp, M_NFSMNT);
504 	return (0);
505 }
506 
507 /*
508  * Return root of a filesystem
509  */
510 nfs_root(mp, vpp)
511 	struct mount *mp;
512 	struct vnode **vpp;
513 {
514 	register struct vnode *vp;
515 	struct nfsmount *nmp;
516 	struct nfsnode *np;
517 	int error;
518 
519 	nmp = VFSTONFS(mp);
520 	if (error = nfs_nget(mp, &nmp->nm_fh, &np))
521 		return (error);
522 	vp = NFSTOV(np);
523 	vp->v_type = VDIR;
524 	vp->v_flag = VROOT;
525 	*vpp = vp;
526 	return (0);
527 }
528 
529 extern int syncprt;
530 
531 /*
532  * Flush out the buffer cache
533  */
534 /* ARGSUSED */
535 nfs_sync(mp, waitfor)
536 	struct mount *mp;
537 	int waitfor;
538 {
539 	if (syncprt)
540 		bufstats();
541 	/*
542 	 * Force stale buffer cache information to be flushed.
543 	 */
544 	mntflushbuf(mp, waitfor == MNT_WAIT ? B_SYNC : 0);
545 	return (0);
546 }
547 
548 /*
549  * At this point, this should never happen
550  */
551 /* ARGSUSED */
552 nfs_fhtovp(mp, fhp, vpp)
553 	struct mount *mp;
554 	struct fid *fhp;
555 	struct vnode **vpp;
556 {
557 
558 	return (EINVAL);
559 }
560 
561 /*
562  * Vnode pointer to File handle, should never happen either
563  */
564 /* ARGSUSED */
565 nfs_vptofh(vp, fhp)
566 	struct vnode *vp;
567 	struct fid *fhp;
568 {
569 
570 	return (EINVAL);
571 }
572 
573 /*
574  * Vfs start routine, a no-op.
575  */
576 /* ARGSUSED */
577 nfs_start(mp, flags, p)
578 	struct mount *mp;
579 	int flags;
580 	struct proc *p;
581 {
582 
583 	return (0);
584 }
585 
586 /*
587  * Do operations associated with quotas, not supported
588  */
589 nfs_quotactl(mp, cmd, uid, arg, p)
590 	struct mount *mp;
591 	int cmd;
592 	uid_t uid;
593 	caddr_t arg;
594 	struct proc *p;
595 {
596 #ifdef lint
597 	mp = mp; cmd = cmd; uid = uid; arg = arg;
598 #endif /* lint */
599 	return (EOPNOTSUPP);
600 }
601