xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision 5f7096188587a2c7c95fa3c69b78e1ec9c7923d0)
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 1.5 1993/11/19 02:32:34 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, curproc))
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 			struct vattr attr;
229 
230 			if (nfs_dogetattr(vp,&attr,NOCRED,0,0)) {
231 			    panic("nfs swap");
232 			}
233 			swdevt[0].sw_nblks = attr.va_size / DEV_BSIZE;
234 		}
235 	}
236 
237 	/*
238 	 * Create the rootfs mount point.
239 	 */
240 	mp = (struct mount *)malloc((u_long)sizeof(struct mount),
241 		M_MOUNT, M_NOWAIT);
242 	if (mp == NULL)
243 		panic("nfs root mount2");
244 	mp->mnt_op = &nfs_vfsops;
245 	mp->mnt_flag = MNT_RDONLY;
246 	mp->mnt_exroot = 0;
247 	mp->mnt_mounth = NULLVP;
248 
249 	/*
250 	 * Set up the root fs args and call mountnfs() to do the rest.
251 	 */
252 	nfs_diskless.root_args.fh = (nfsv2fh_t *)nfs_diskless.root_fh;
253 	MGET(m, MT_SONAME, M_DONTWAIT);
254 	if (m == NULL)
255 		panic("nfs root mbuf2");
256 	bcopy((caddr_t)&nfs_diskless.root_saddr, mtod(m, caddr_t),
257 		nfs_diskless.root_saddr.sa_len);
258 	m->m_len = nfs_diskless.root_saddr.sa_len;
259 	if (mountnfs(&nfs_diskless.root_args, mp, m, "/",
260 		nfs_diskless.root_hostnam, &vp))
261 		panic("nfs root");
262 	if (vfs_lock(mp))
263 		panic("nfs root2");
264 	rootfs = mp;
265 	mp->mnt_next = mp;
266 	mp->mnt_prev = mp;
267 	mp->mnt_vnodecovered = NULLVP;
268 	vfs_unlock(mp);
269 	rootvp = vp;
270 	inittodr((time_t)0);	/* There is no time in the nfs fsstat so ?? */
271 	return (0);
272 }
273 
274 /*
275  * VFS Operations.
276  *
277  * mount system call
278  * It seems a bit dumb to copyinstr() the host and path here and then
279  * bcopy() them in mountnfs(), but I wanted to detect errors before
280  * doing the sockargs() call because sockargs() allocates an mbuf and
281  * an error after that means that I have to release the mbuf.
282  */
283 /* ARGSUSED */
284 nfs_mount(mp, path, data, ndp, p)
285 	struct mount *mp;
286 	char *path;
287 	caddr_t data;
288 	struct nameidata *ndp;
289 	struct proc *p;
290 {
291 	int error;
292 	struct nfs_args args;
293 	struct mbuf *nam;
294 	struct vnode *vp;
295 	char pth[MNAMELEN], hst[MNAMELEN];
296 	u_int len;
297 	nfsv2fh_t nfh;
298 
299 	if (mp->mnt_flag & MNT_UPDATE)
300 		return (0);
301 	if (error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)))
302 		return (error);
303 	if (error = copyin((caddr_t)args.fh, (caddr_t)&nfh, sizeof (nfsv2fh_t)))
304 		return (error);
305 	if (error = copyinstr(path, pth, MNAMELEN-1, &len))
306 		return (error);
307 	bzero(&pth[len], MNAMELEN - len);
308 	if (error = copyinstr(args.hostname, hst, MNAMELEN-1, &len))
309 		return (error);
310 	bzero(&hst[len], MNAMELEN - len);
311 	/* sockargs() call must be after above copyin() calls */
312 	if (error = sockargs(&nam, (caddr_t)args.addr,
313 		sizeof (struct sockaddr), MT_SONAME))
314 		return (error);
315 	args.fh = &nfh;
316 	error = mountnfs(&args, mp, nam, pth, hst, &vp);
317 	return (error);
318 }
319 
320 /*
321  * Common code for mount and mountroot
322  */
323 mountnfs(argp, mp, nam, pth, hst, vpp)
324 	register struct nfs_args *argp;
325 	register struct mount *mp;
326 	struct mbuf *nam;
327 	char *pth, *hst;
328 	struct vnode **vpp;
329 {
330 	register struct nfsmount *nmp;
331 	struct proc *p = curproc;		/* XXX */
332 	struct nfsnode *np;
333 	int error;
334 	fsid_t tfsid;
335 
336 	MALLOC(nmp, struct nfsmount *, sizeof *nmp, M_NFSMNT, M_WAITOK);
337 	bzero((caddr_t)nmp, sizeof *nmp);
338 	mp->mnt_data = (qaddr_t)nmp;
339 
340 	getnewfsid(mp, MOUNT_NFS);
341 	nmp->nm_mountp = mp;
342 	nmp->nm_flag = argp->flags;
343 	nmp->nm_rto = NFS_TIMEO;
344 	nmp->nm_rtt = -1;
345 	nmp->nm_rttvar = nmp->nm_rto << 1;
346 	nmp->nm_retry = NFS_RETRANS;
347 	nmp->nm_wsize = NFS_WSIZE;
348 	nmp->nm_rsize = NFS_RSIZE;
349 	bcopy((caddr_t)argp->fh, (caddr_t)&nmp->nm_fh, sizeof(nfsv2fh_t));
350 	bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
351 	bcopy(pth, mp->mnt_stat.f_mntonname, MNAMELEN);
352 	nmp->nm_nam = nam;
353 
354 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
355 		nmp->nm_rto = argp->timeo;
356 		/* NFS timeouts are specified in 1/10 sec. */
357 		nmp->nm_rto = (nmp->nm_rto * 10) / NFS_HZ;
358 		if (nmp->nm_rto < NFS_MINTIMEO)
359 			nmp->nm_rto = NFS_MINTIMEO;
360 		else if (nmp->nm_rto > NFS_MAXTIMEO)
361 			nmp->nm_rto = NFS_MAXTIMEO;
362 		nmp->nm_rttvar = nmp->nm_rto << 1;
363 	}
364 
365 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
366 		nmp->nm_retry = argp->retrans;
367 		if (nmp->nm_retry > NFS_MAXREXMIT)
368 			nmp->nm_retry = NFS_MAXREXMIT;
369 	}
370 
371 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
372 		nmp->nm_wsize = argp->wsize;
373 		/* Round down to multiple of blocksize */
374 		nmp->nm_wsize &= ~0x1ff;
375 		if (nmp->nm_wsize <= 0)
376 			nmp->nm_wsize = 512;
377 		else if (nmp->nm_wsize > NFS_MAXDATA)
378 			nmp->nm_wsize = NFS_MAXDATA;
379 	}
380 	if (nmp->nm_wsize > MAXBSIZE)
381 		nmp->nm_wsize = MAXBSIZE;
382 
383 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
384 		nmp->nm_rsize = argp->rsize;
385 		/* Round down to multiple of blocksize */
386 		nmp->nm_rsize &= ~0x1ff;
387 		if (nmp->nm_rsize <= 0)
388 			nmp->nm_rsize = 512;
389 		else if (nmp->nm_rsize > NFS_MAXDATA)
390 			nmp->nm_rsize = NFS_MAXDATA;
391 	}
392 	if (nmp->nm_rsize > MAXBSIZE)
393 		nmp->nm_rsize = MAXBSIZE;
394 	/* Set up the sockets and per-host congestion */
395 	nmp->nm_sotype = argp->sotype;
396 	nmp->nm_soproto = argp->proto;
397 	if (error = nfs_connect(nmp))
398 		goto bad;
399 
400 	if (error = nfs_statfs(mp, &mp->mnt_stat, p))
401 		goto bad;
402 	/*
403 	 * A reference count is needed on the nfsnode representing the
404 	 * remote root.  If this object is not persistent, then backward
405 	 * traversals of the mount point (i.e. "..") will not work if
406 	 * the nfsnode gets flushed out of the cache. Ufs does not have
407 	 * this problem, because one can identify root inodes by their
408 	 * number == ROOTINO (2).
409 	 */
410 	if (error = nfs_nget(mp, &nmp->nm_fh, &np))
411 		goto bad;
412 	/*
413 	 * Unlock it, but keep the reference count.
414 	 */
415 	nfs_unlock(NFSTOV(np));
416 	*vpp = NFSTOV(np);
417 
418 	return (0);
419 bad:
420 	nfs_disconnect(nmp);
421 	FREE(nmp, M_NFSMNT);
422 	m_freem(nam);
423 	return (error);
424 }
425 
426 /*
427  * unmount system call
428  */
429 nfs_unmount(mp, mntflags, p)
430 	struct mount *mp;
431 	int mntflags;
432 	struct proc *p;
433 {
434 	register struct nfsmount *nmp;
435 	struct nfsnode *np;
436 	struct vnode *vp;
437 	int error, flags = 0;
438 	extern int doforce;
439 
440 	if (mntflags & MNT_FORCE) {
441 		if (!doforce || mp == rootfs)
442 			return (EINVAL);
443 		flags |= FORCECLOSE;
444 	}
445 	nmp = VFSTONFS(mp);
446 	/*
447 	 * Clear out the buffer cache
448 	 */
449 	mntflushbuf(mp, 0);
450 	if (mntinvalbuf(mp))
451 		return (EBUSY);
452 	/*
453 	 * Goes something like this..
454 	 * - Check for activity on the root vnode (other than ourselves).
455 	 * - Call vflush() to clear out vnodes for this file system,
456 	 *   except for the root vnode.
457 	 * - Decrement reference on the vnode representing remote root.
458 	 * - Close the socket
459 	 * - Free up the data structures
460 	 */
461 	/*
462 	 * We need to decrement the ref. count on the nfsnode representing
463 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
464 	 * has done vput on this vnode, otherwise we would get deadlock!
465 	 */
466 	if (error = nfs_nget(mp, &nmp->nm_fh, &np))
467 		return(error);
468 	vp = NFSTOV(np);
469 	if (vp->v_usecount > 2) {
470 		vput(vp);
471 		return (EBUSY);
472 	}
473 	if (error = vflush(mp, vp, flags)) {
474 		vput(vp);
475 		return (error);
476 	}
477 	/*
478 	 * Get rid of two reference counts, and unlock it on the second.
479 	 */
480 	vrele(vp);
481 	vput(vp);
482 	nfs_disconnect(nmp);
483 	m_freem(nmp->nm_nam);
484 	free((caddr_t)nmp, M_NFSMNT);
485 	return (0);
486 }
487 
488 /*
489  * Return root of a filesystem
490  */
491 nfs_root(mp, vpp)
492 	struct mount *mp;
493 	struct vnode **vpp;
494 {
495 	register struct vnode *vp;
496 	struct nfsmount *nmp;
497 	struct nfsnode *np;
498 	int error;
499 
500 	nmp = VFSTONFS(mp);
501 	if (error = nfs_nget(mp, &nmp->nm_fh, &np))
502 		return (error);
503 	vp = NFSTOV(np);
504 	vp->v_type = VDIR;
505 	vp->v_flag = VROOT;
506 	*vpp = vp;
507 	return (0);
508 }
509 
510 extern int syncprt;
511 
512 /*
513  * Flush out the buffer cache
514  */
515 /* ARGSUSED */
516 nfs_sync(mp, waitfor)
517 	struct mount *mp;
518 	int waitfor;
519 {
520 	if (syncprt)
521 		bufstats();
522 	/*
523 	 * Force stale buffer cache information to be flushed.
524 	 */
525 	mntflushbuf(mp, waitfor == MNT_WAIT ? B_SYNC : 0);
526 	return (0);
527 }
528 
529 /*
530  * At this point, this should never happen
531  */
532 /* ARGSUSED */
533 nfs_fhtovp(mp, fhp, vpp)
534 	struct mount *mp;
535 	struct fid *fhp;
536 	struct vnode **vpp;
537 {
538 
539 	return (EINVAL);
540 }
541 
542 /*
543  * Vnode pointer to File handle, should never happen either
544  */
545 /* ARGSUSED */
546 nfs_vptofh(vp, fhp)
547 	struct vnode *vp;
548 	struct fid *fhp;
549 {
550 
551 	return (EINVAL);
552 }
553 
554 /*
555  * Vfs start routine, a no-op.
556  */
557 /* ARGSUSED */
558 nfs_start(mp, flags, p)
559 	struct mount *mp;
560 	int flags;
561 	struct proc *p;
562 {
563 
564 	return (0);
565 }
566 
567 /*
568  * Do operations associated with quotas, not supported
569  */
570 nfs_quotactl(mp, cmd, uid, arg, p)
571 	struct mount *mp;
572 	int cmd;
573 	uid_t uid;
574 	caddr_t arg;
575 	struct proc *p;
576 {
577 #ifdef lint
578 	mp = mp; cmd = cmd; uid = uid; arg = arg;
579 #endif /* lint */
580 	return (EOPNOTSUPP);
581 }
582