xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision 001c68bd94f75ce9270b69227c4199fbf34ee396)
1 /*	$NetBSD: nfs_vfsops.c,v 1.131 2003/06/29 22:32:19 fvdl 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 1.131 2003/06/29 22:32:19 fvdl 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 /*
131  * nfs statfs call
132  */
133 int
134 nfs_statfs(mp, sbp, p)
135 	struct mount *mp;
136 	struct statfs *sbp;
137 	struct proc *p;
138 {
139 	struct vnode *vp;
140 	struct nfs_statfs *sfp;
141 	caddr_t cp;
142 	u_int32_t *tl;
143 	int32_t t1, t2;
144 	caddr_t bpos, dpos, cp2;
145 	struct nfsmount *nmp = VFSTONFS(mp);
146 	int error = 0, retattr;
147 #ifdef NFS_V2_ONLY
148 	const int v3 = 0;
149 #else
150 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
151 #endif
152 	struct mbuf *mreq, *mrep = NULL, *md, *mb;
153 	struct ucred *cred;
154 	u_quad_t tquad;
155 	struct nfsnode *np;
156 
157 #ifndef nolint
158 	sfp = (struct nfs_statfs *)0;
159 #endif
160 	vp = nmp->nm_vnode;
161 	np = VTONFS(vp);
162 	cred = crget();
163 	cred->cr_ngroups = 0;
164 #ifndef NFS_V2_ONLY
165 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
166 		(void)nfs_fsinfo(nmp, vp, cred, p);
167 #endif
168 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
169 	nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3));
170 	nfsm_fhtom(np, v3);
171 	nfsm_request(np, NFSPROC_FSSTAT, p, cred);
172 	if (v3)
173 		nfsm_postop_attr(vp, retattr, 0);
174 	if (error) {
175 		if (mrep != NULL)
176 			m_free(mrep);
177 		goto nfsmout;
178 	}
179 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
180 #ifdef COMPAT_09
181 	sbp->f_type = 2;
182 #else
183 	sbp->f_type = 0;
184 #endif
185 	sbp->f_flags = nmp->nm_flag;
186 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
187 	if (v3) {
188 		sbp->f_bsize = NFS_FABLKSIZE;
189 		tquad = fxdr_hyper(&sfp->sf_tbytes);
190 		sbp->f_blocks = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
191 		tquad = fxdr_hyper(&sfp->sf_fbytes);
192 		sbp->f_bfree = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
193 		tquad = fxdr_hyper(&sfp->sf_abytes);
194 		sbp->f_bavail = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
195 		tquad = fxdr_hyper(&sfp->sf_tfiles);
196 		sbp->f_files = (long)tquad;
197 		tquad = fxdr_hyper(&sfp->sf_ffiles);
198 		sbp->f_ffree = (long)tquad;
199 	} else {
200 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
201 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
202 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
203 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
204 		sbp->f_files = 0;
205 		sbp->f_ffree = 0;
206 	}
207 	copy_statfs_info(sbp, mp);
208 	nfsm_reqdone;
209 	crfree(cred);
210 	return (error);
211 }
212 
213 #ifndef NFS_V2_ONLY
214 /*
215  * nfs version 3 fsinfo rpc call
216  */
217 int
218 nfs_fsinfo(nmp, vp, cred, p)
219 	struct nfsmount *nmp;
220 	struct vnode *vp;
221 	struct ucred *cred;
222 	struct proc *p;
223 {
224 	struct nfsv3_fsinfo *fsp;
225 	caddr_t cp;
226 	int32_t t1, t2;
227 	u_int32_t *tl, pref, max;
228 	caddr_t bpos, dpos, cp2;
229 	int error = 0, retattr;
230 	struct mbuf *mreq, *mrep, *md, *mb;
231 	u_int64_t maxfsize;
232 	struct nfsnode *np = VTONFS(vp);
233 
234 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
235 	nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1));
236 	nfsm_fhtom(np, 1);
237 	nfsm_request(np, NFSPROC_FSINFO, p, cred);
238 	nfsm_postop_attr(vp, retattr, 0);
239 	if (!error) {
240 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
241 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
242 		if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 &&
243 		    pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
244 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
245 				~(NFS_FABLKSIZE - 1);
246 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
247 		if (max < nmp->nm_wsize && max > 0) {
248 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
249 			if (nmp->nm_wsize == 0)
250 				nmp->nm_wsize = max;
251 		}
252 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
253 		if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 &&
254 		    pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
255 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
256 				~(NFS_FABLKSIZE - 1);
257 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
258 		if (max < nmp->nm_rsize && max > 0) {
259 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
260 			if (nmp->nm_rsize == 0)
261 				nmp->nm_rsize = max;
262 		}
263 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
264 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
265 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
266 				~(NFS_DIRFRAGSIZ - 1);
267 		if (max < nmp->nm_readdirsize && max > 0) {
268 			nmp->nm_readdirsize = max & ~(NFS_DIRFRAGSIZ - 1);
269 			if (nmp->nm_readdirsize == 0)
270 				nmp->nm_readdirsize = max;
271 		}
272 		/* XXX */
273 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
274 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
275 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
276 			nmp->nm_maxfilesize = maxfsize;
277 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
278 	}
279 	nfsm_reqdone;
280 	return (error);
281 }
282 #endif
283 
284 /*
285  * Mount a remote root fs via. NFS.  It goes like this:
286  * - Call nfs_boot_init() to fill in the nfs_diskless struct
287  * - build the rootfs mount point and call mountnfs() to do the rest.
288  */
289 int
290 nfs_mountroot()
291 {
292 	struct nfs_diskless *nd;
293 	struct vattr attr;
294 	struct mount *mp;
295 	struct vnode *vp;
296 	struct proc *procp;
297 	long n;
298 	int error;
299 
300 	procp = curproc; /* XXX */
301 
302 	if (root_device->dv_class != DV_IFNET)
303 		return (ENODEV);
304 
305 	/*
306 	 * XXX time must be non-zero when we init the interface or else
307 	 * the arp code will wedge.  [Fixed now in if_ether.c]
308 	 * However, the NFS attribute cache gives false "hits" when
309 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
310 	 */
311 	if (time.tv_sec < NFS_MAXATTRTIMO)
312 		time.tv_sec = NFS_MAXATTRTIMO;
313 
314 	/*
315 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
316 	 * Side effect:  Finds and configures a network interface.
317 	 */
318 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
319 	memset((caddr_t)nd, 0, sizeof(*nd));
320 	error = nfs_boot_init(nd, procp);
321 	if (error) {
322 		free(nd, M_NFSMNT);
323 		return (error);
324 	}
325 
326 	/*
327 	 * Create the root mount point.
328 	 */
329 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp);
330 	if (error)
331 		goto out;
332 	printf("root on %s\n", nd->nd_root.ndm_host);
333 
334 	/*
335 	 * Link it into the mount list.
336 	 */
337 	simple_lock(&mountlist_slock);
338 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
339 	simple_unlock(&mountlist_slock);
340 	rootvp = vp;
341 	mp->mnt_vnodecovered = NULLVP;
342 	vfs_unbusy(mp);
343 
344 	/* Get root attributes (for the time). */
345 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
346 	if (error)
347 		panic("nfs_mountroot: getattr for root");
348 	n = attr.va_atime.tv_sec;
349 #ifdef	DEBUG
350 	printf("root time: 0x%lx\n", n);
351 #endif
352 	inittodr(n);
353 
354 out:
355 	if (error)
356 		nfs_boot_cleanup(nd, procp);
357 	free(nd, M_NFSMNT);
358 	return (error);
359 }
360 
361 /*
362  * Internal version of mount system call for diskless setup.
363  * Separate function because we used to call it twice.
364  * (once for root and once for swap)
365  */
366 static int
367 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p)
368 	struct nfs_dlmount *ndmntp;
369 	const char *mntname;	/* mount point name */
370 	struct mount **mpp;
371 	struct vnode **vpp;
372 	struct proc *p;
373 {
374 	struct mount *mp;
375 	struct mbuf *m;
376 	int error;
377 
378 	vfs_rootmountalloc(MOUNT_NFS, (char *)mntname, &mp);
379 
380 	mp->mnt_op = &nfs_vfsops;
381 
382 	/*
383 	 * Historical practice expects NFS root file systems to
384 	 * be initially mounted r/w.
385 	 */
386 	mp->mnt_flag &= ~MNT_RDONLY;
387 
388 	/* Get mbuf for server sockaddr. */
389 	m = m_get(M_WAIT, MT_SONAME);
390 	if (m == NULL)
391 		panic("nfs_mountroot: mget soname for %s", mntname);
392 	MCLAIM(m, &nfs_mowner);
393 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
394 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
395 
396 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
397 			 ndmntp->ndm_args.hostname, vpp, p);
398 	if (error) {
399 		mp->mnt_op->vfs_refcount--;
400 		vfs_unbusy(mp);
401 		printf("nfs_mountroot: mount %s failed: %d\n",
402 		       mntname, error);
403 		free(mp, M_MOUNT);
404 	} else
405 		*mpp = mp;
406 
407 	return (error);
408 }
409 
410 void
411 nfs_decode_args(nmp, argp)
412 	struct nfsmount *nmp;
413 	struct nfs_args *argp;
414 {
415 	int s;
416 	int adjsock;
417 	int maxio;
418 
419 	s = splsoftnet();
420 
421 	/*
422 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
423 	 * no sense in that context.
424 	 */
425 	if (argp->sotype == SOCK_STREAM)
426 		argp->flags &= ~NFSMNT_NOCONN;
427 
428 	/*
429 	 * Cookie translation is not needed for v2, silently ignore it.
430 	 */
431 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
432 	    NFSMNT_XLATECOOKIE)
433 		argp->flags &= ~NFSMNT_XLATECOOKIE;
434 
435 	/* Re-bind if rsrvd port requested and wasn't on one */
436 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
437 		  && (argp->flags & NFSMNT_RESVPORT);
438 	/* Also re-bind if we're switching to/from a connected UDP socket */
439 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
440 		    (argp->flags & NFSMNT_NOCONN));
441 
442 	/* Update flags. */
443 	nmp->nm_flag = argp->flags;
444 	splx(s);
445 
446 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
447 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
448 		if (nmp->nm_timeo < NFS_MINTIMEO)
449 			nmp->nm_timeo = NFS_MINTIMEO;
450 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
451 			nmp->nm_timeo = NFS_MAXTIMEO;
452 	}
453 
454 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
455 		nmp->nm_retry = argp->retrans;
456 		if (nmp->nm_retry > NFS_MAXREXMIT)
457 			nmp->nm_retry = NFS_MAXREXMIT;
458 	}
459 
460 #ifndef NFS_V2_ONLY
461 	if (argp->flags & NFSMNT_NFSV3) {
462 		if (argp->sotype == SOCK_DGRAM)
463 			maxio = NFS_MAXDGRAMDATA;
464 		else
465 			maxio = NFS_MAXDATA;
466 	} else
467 #endif
468 		maxio = NFS_V2MAXDATA;
469 
470 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
471 		int osize = nmp->nm_wsize;
472 		nmp->nm_wsize = argp->wsize;
473 		/* Round down to multiple of blocksize */
474 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
475 		if (nmp->nm_wsize <= 0)
476 			nmp->nm_wsize = NFS_FABLKSIZE;
477 		adjsock |= (nmp->nm_wsize != osize);
478 	}
479 	if (nmp->nm_wsize > maxio)
480 		nmp->nm_wsize = maxio;
481 	if (nmp->nm_wsize > MAXBSIZE)
482 		nmp->nm_wsize = MAXBSIZE;
483 
484 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
485 		int osize = nmp->nm_rsize;
486 		nmp->nm_rsize = argp->rsize;
487 		/* Round down to multiple of blocksize */
488 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
489 		if (nmp->nm_rsize <= 0)
490 			nmp->nm_rsize = NFS_FABLKSIZE;
491 		adjsock |= (nmp->nm_rsize != osize);
492 	}
493 	if (nmp->nm_rsize > maxio)
494 		nmp->nm_rsize = maxio;
495 	if (nmp->nm_rsize > MAXBSIZE)
496 		nmp->nm_rsize = MAXBSIZE;
497 
498 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
499 		nmp->nm_readdirsize = argp->readdirsize;
500 		/* Round down to multiple of minimum blocksize */
501 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
502 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
503 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
504 		/* Bigger than buffer size makes no sense */
505 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
506 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
507 	} else if (argp->flags & NFSMNT_RSIZE)
508 		nmp->nm_readdirsize = nmp->nm_rsize;
509 
510 	if (nmp->nm_readdirsize > maxio)
511 		nmp->nm_readdirsize = maxio;
512 
513 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
514 		argp->maxgrouplist <= NFS_MAXGRPS)
515 		nmp->nm_numgrps = argp->maxgrouplist;
516 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
517 		argp->readahead <= NFS_MAXRAHEAD)
518 		nmp->nm_readahead = argp->readahead;
519 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
520 		argp->leaseterm <= NQ_MAXLEASE)
521 		nmp->nm_leaseterm = argp->leaseterm;
522 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
523 		argp->deadthresh <= NQ_NEVERDEAD)
524 		nmp->nm_deadthresh = argp->deadthresh;
525 
526 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
527 		    (nmp->nm_soproto != argp->proto));
528 	nmp->nm_sotype = argp->sotype;
529 	nmp->nm_soproto = argp->proto;
530 
531 	if (nmp->nm_so && adjsock) {
532 		nfs_safedisconnect(nmp);
533 		if (nmp->nm_sotype == SOCK_DGRAM)
534 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
535 				printf("nfs_args: retrying connect\n");
536 				(void) tsleep((caddr_t)&lbolt,
537 					      PSOCK, "nfscn3", 0);
538 			}
539 	}
540 }
541 
542 /*
543  * VFS Operations.
544  *
545  * mount system call
546  * It seems a bit dumb to copyinstr() the host and path here and then
547  * memcpy() them in mountnfs(), but I wanted to detect errors before
548  * doing the sockargs() call because sockargs() allocates an mbuf and
549  * an error after that means that I have to release the mbuf.
550  */
551 /* ARGSUSED */
552 int
553 nfs_mount(mp, path, data, ndp, p)
554 	struct mount *mp;
555 	const char *path;
556 	void *data;
557 	struct nameidata *ndp;
558 	struct proc *p;
559 {
560 	int error;
561 	struct nfs_args args;
562 	struct mbuf *nam;
563 	struct nfsmount *nmp = VFSTONFS(mp);
564 	struct sockaddr *sa;
565 	struct vnode *vp;
566 	char *pth, *hst;
567 	size_t len;
568 	u_char *nfh;
569 
570 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
571 	if (error)
572 		return (error);
573 
574 	if (mp->mnt_flag & MNT_GETARGS) {
575 
576 		if (nmp == NULL)
577 			return (EIO);
578 		if (args.addr != NULL) {
579 			sa = mtod(nmp->nm_nam, struct sockaddr *);
580 			error = copyout(sa, args.addr, sa->sa_len);
581 			if (error)
582 				return (error);
583 			args.addrlen = sa->sa_len;
584 		} else
585 			args.addrlen = 0;
586 
587 		args.version = NFS_ARGSVERSION;
588 		args.sotype = nmp->nm_sotype;
589 		args.proto = nmp->nm_soproto;
590 		args.fh = NULL;
591 		args.fhsize = 0;
592 		args.flags = nmp->nm_flag;
593 		args.wsize = nmp->nm_wsize;
594 		args.rsize = nmp->nm_rsize;
595 		args.readdirsize = nmp->nm_readdirsize;
596 		args.timeo = nmp->nm_timeo;
597 		args.retrans = nmp->nm_retry;
598 		args.maxgrouplist = nmp->nm_numgrps;
599 		args.readahead = nmp->nm_readahead;
600 		args.leaseterm = nmp->nm_leaseterm;
601 		args.deadthresh = nmp->nm_deadthresh;
602 		args.hostname = NULL;
603 		return (copyout(&args, data, sizeof(args)));
604 	}
605 
606 	if (args.version != NFS_ARGSVERSION)
607 		return (EPROGMISMATCH);
608 #ifdef NFS_V2_ONLY
609 	if (args.flags & NFSMNT_NQNFS)
610 		return (EPROGUNAVAIL);
611 	if (args.flags & NFSMNT_NFSV3)
612 		return (EPROGMISMATCH);
613 #endif
614 	if (mp->mnt_flag & MNT_UPDATE) {
615 		if (nmp == NULL)
616 			return (EIO);
617 		/*
618 		 * When doing an update, we can't change from or to
619 		 * v3 and/or nqnfs, or change cookie translation
620 		 */
621 		args.flags = (args.flags &
622 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
623 		    (nmp->nm_flag &
624 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
625 		nfs_decode_args(nmp, &args);
626 		return (0);
627 	}
628 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
629 		return (EINVAL);
630 	MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK);
631 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
632 	if (error)
633 		return (error);
634 	MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK);
635 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
636 	if (error)
637 		goto free_nfh;
638 	memset(&pth[len], 0, MNAMELEN - len);
639 	MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK);
640 	error = copyinstr(args.hostname, hst, MNAMELEN - 1, &len);
641 	if (error)
642 		goto free_pth;
643 	memset(&hst[len], 0, MNAMELEN - len);
644 	/* sockargs() call must be after above copyin() calls */
645 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
646 	if (error)
647 		goto free_hst;
648 	MCLAIM(nam, &nfs_mowner);
649 	args.fh = nfh;
650 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
651 
652 free_hst:
653 	FREE(hst, M_TEMP);
654 free_pth:
655 	FREE(pth, M_TEMP);
656 free_nfh:
657 	FREE(nfh, M_TEMP);
658 
659 	return (error);
660 }
661 
662 /*
663  * Common code for mount and mountroot
664  */
665 int
666 mountnfs(argp, mp, nam, pth, hst, vpp, p)
667 	struct nfs_args *argp;
668 	struct mount *mp;
669 	struct mbuf *nam;
670 	const char *pth, *hst;
671 	struct vnode **vpp;
672 	struct proc *p;
673 {
674 	struct nfsmount *nmp;
675 	struct nfsnode *np;
676 	int error;
677 	struct vattr *attrs;
678 	struct ucred *cr;
679 
680 	/*
681 	 * If the number of nfs iothreads to use has never
682 	 * been set, create a reasonable number of them.
683 	 */
684 
685 	if (nfs_niothreads < 0) {
686 		nfs_niothreads = NFS_DEFAULT_NIOTHREADS;
687 		nfs_getset_niothreads(TRUE);
688 	}
689 
690 	if (mp->mnt_flag & MNT_UPDATE) {
691 		nmp = VFSTONFS(mp);
692 		/* update paths, file handles, etc, here	XXX */
693 		m_freem(nam);
694 		return (0);
695 	} else {
696 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
697 		    M_NFSMNT, M_WAITOK);
698 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
699 		mp->mnt_data = nmp;
700 		TAILQ_INIT(&nmp->nm_uidlruhead);
701 		TAILQ_INIT(&nmp->nm_bufq);
702 		lockinit(&nmp->nm_writeverflock, PRIBIO, "nfswverf", 0, 0);
703 		simple_lock_init(&nmp->nm_slock);
704 	}
705 	vfs_getnewfsid(mp);
706 	nmp->nm_mountp = mp;
707 
708 #ifndef NFS_V2_ONLY
709 	if (argp->flags & NFSMNT_NQNFS)
710 		/*
711 		 * We have to set mnt_maxsymlink to a non-zero value so
712 		 * that COMPAT_43 routines will know that we are setting
713 		 * the d_type field in directories (and can zero it for
714 		 * unsuspecting binaries).
715 		 */
716 		mp->mnt_maxsymlinklen = 1;
717 #endif
718 
719 #ifndef NFS_V2_ONLY
720 	if ((argp->flags & NFSMNT_NFSV3) == 0)
721 #endif
722 		/*
723 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
724 		 * will fill this in.
725 		 */
726 		nmp->nm_maxfilesize = 0xffffffffLL;
727 
728 	nmp->nm_timeo = NFS_TIMEO;
729 	nmp->nm_retry = NFS_RETRANS;
730 	nmp->nm_wsize = NFS_WSIZE;
731 	nmp->nm_rsize = NFS_RSIZE;
732 	nmp->nm_readdirsize = NFS_READDIRSIZE;
733 	nmp->nm_numgrps = NFS_MAXGRPS;
734 	nmp->nm_readahead = NFS_DEFRAHEAD;
735 	nmp->nm_leaseterm = NQ_DEFLEASE;
736 	nmp->nm_deadthresh = NQ_DEADTHRESH;
737 	CIRCLEQ_INIT(&nmp->nm_timerhead);
738 	nmp->nm_inprog = NULLVP;
739 #ifdef COMPAT_09
740 	mp->mnt_stat.f_type = 2;
741 #else
742 	mp->mnt_stat.f_type = 0;
743 #endif
744 	error = set_statfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, mp, p);
745 	if (error)
746 		goto bad;
747 	nmp->nm_nam = nam;
748 
749 	/* Set up the sockets and per-host congestion */
750 	nmp->nm_sotype = argp->sotype;
751 	nmp->nm_soproto = argp->proto;
752 
753 	nfs_decode_args(nmp, argp);
754 
755 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
756 	mp->mnt_dev_bshift = DEV_BSHIFT;
757 
758 	/*
759 	 * For Connection based sockets (TCP,...) defer the connect until
760 	 * the first request, in case the server is not responding.
761 	 */
762 	if (nmp->nm_sotype == SOCK_DGRAM &&
763 		(error = nfs_connect(nmp, (struct nfsreq *)0)))
764 		goto bad;
765 
766 	/*
767 	 * This is silly, but it has to be set so that vinifod() works.
768 	 * We do not want to do an nfs_statfs() here since we can get
769 	 * stuck on a dead server and we are holding a lock on the mount
770 	 * point.
771 	 */
772 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
773 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
774 	if (error)
775 		goto bad;
776 	*vpp = NFSTOV(np);
777 	MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK);
778 	VOP_GETATTR(*vpp, attrs, p->p_ucred, p);
779 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
780 		cr = crget();
781 		cr->cr_uid = attrs->va_uid;
782 		cr->cr_gid = attrs->va_gid;
783 		cr->cr_ngroups = 0;
784 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
785 		crfree(cr);
786 	}
787 	FREE(attrs, M_TEMP);
788 
789 	/*
790 	 * A reference count is needed on the nfsnode representing the
791 	 * remote root.  If this object is not persistent, then backward
792 	 * traversals of the mount point (i.e. "..") will not work if
793 	 * the nfsnode gets flushed out of the cache. Ufs does not have
794 	 * this problem, because one can identify root inodes by their
795 	 * number == ROOTINO (2). So, just unlock, but no rele.
796 	 */
797 
798 	nmp->nm_vnode = *vpp;
799 	VOP_UNLOCK(*vpp, 0);
800 
801 	return (0);
802 bad:
803 	nfs_disconnect(nmp);
804 	free((caddr_t)nmp, M_NFSMNT);
805 	m_freem(nam);
806 	return (error);
807 }
808 
809 /*
810  * unmount system call
811  */
812 int
813 nfs_unmount(mp, mntflags, p)
814 	struct mount *mp;
815 	int mntflags;
816 	struct proc *p;
817 {
818 	struct nfsmount *nmp;
819 	struct vnode *vp;
820 	int error, flags = 0;
821 
822 	if (mntflags & MNT_FORCE)
823 		flags |= FORCECLOSE;
824 	nmp = VFSTONFS(mp);
825 	/*
826 	 * Goes something like this..
827 	 * - Check for activity on the root vnode (other than ourselves).
828 	 * - Call vflush() to clear out vnodes for this file system,
829 	 *   except for the root vnode.
830 	 * - Decrement reference on the vnode representing remote root.
831 	 * - Close the socket
832 	 * - Free up the data structures
833 	 */
834 	/*
835 	 * We need to decrement the ref. count on the nfsnode representing
836 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
837 	 * has done vput on this vnode, otherwise we would get deadlock!
838 	 */
839 	vp = nmp->nm_vnode;
840 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
841 	if (error != 0)
842 		return error;
843 
844 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
845 		vput(vp);
846 		return (EBUSY);
847 	}
848 
849 	/*
850 	 * Must handshake with nqnfs_clientd() if it is active.
851 	 */
852 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
853 	while (nmp->nm_inprog != NULLVP)
854 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
855 	error = vflush(mp, vp, flags);
856 	if (error) {
857 		vput(vp);
858 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
859 		return (error);
860 	}
861 
862 	/*
863 	 * We are now committed to the unmount; mark the mount structure
864 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
865 	 * will go away cleanly.
866 	 */
867 	nmp->nm_iflag |= NFSMNT_DISMNT;
868 
869 	/*
870 	 * There are two reference counts to get rid of here
871 	 * (see comment in mountnfs()).
872 	 */
873 	vrele(vp);
874 	vput(vp);
875 	vgone(vp);
876 	nfs_disconnect(nmp);
877 	m_freem(nmp->nm_nam);
878 
879 	/*
880 	 * For NQNFS, let the server daemon free the nfsmount structure.
881 	 */
882 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
883 		free((caddr_t)nmp, M_NFSMNT);
884 	return (0);
885 }
886 
887 /*
888  * Return root of a filesystem
889  */
890 int
891 nfs_root(mp, vpp)
892 	struct mount *mp;
893 	struct vnode **vpp;
894 {
895 	struct vnode *vp;
896 	struct nfsmount *nmp;
897 	int error;
898 
899 	nmp = VFSTONFS(mp);
900 	vp = nmp->nm_vnode;
901 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
902 	if (error != 0)
903 		return error;
904 	if (vp->v_type == VNON)
905 		vp->v_type = VDIR;
906 	vp->v_flag = VROOT;
907 	*vpp = vp;
908 	return (0);
909 }
910 
911 extern int syncprt;
912 
913 /*
914  * Flush out the buffer cache
915  */
916 /* ARGSUSED */
917 int
918 nfs_sync(mp, waitfor, cred, p)
919 	struct mount *mp;
920 	int waitfor;
921 	struct ucred *cred;
922 	struct proc *p;
923 {
924 	struct vnode *vp;
925 	int error, allerror = 0;
926 
927 	/*
928 	 * Force stale buffer cache information to be flushed.
929 	 */
930 loop:
931 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
932 		/*
933 		 * If the vnode that we are about to sync is no longer
934 		 * associated with this mount point, start over.
935 		 */
936 		if (vp->v_mount != mp)
937 			goto loop;
938 		if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) ||
939 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
940 		     vp->v_uobj.uo_npages == 0))
941 			continue;
942 		if (vget(vp, LK_EXCLUSIVE))
943 			goto loop;
944 		error = VOP_FSYNC(vp, cred,
945 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p);
946 		if (error)
947 			allerror = error;
948 		vput(vp);
949 	}
950 	return (allerror);
951 }
952 
953 /*
954  * NFS flat namespace lookup.
955  * Currently unsupported.
956  */
957 /* ARGSUSED */
958 int
959 nfs_vget(mp, ino, vpp)
960 	struct mount *mp;
961 	ino_t ino;
962 	struct vnode **vpp;
963 {
964 
965 	return (EOPNOTSUPP);
966 }
967 
968 /*
969  * Do that sysctl thang...
970  */
971 int
972 nfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
973 	int *name;
974 	u_int namelen;
975 	void *oldp;
976 	size_t *oldlenp;
977 	void *newp;
978 	size_t newlen;
979 	struct proc *p;
980 {
981 	int rv;
982 
983 	/*
984 	 * All names at this level are terminal.
985 	 */
986 	if(namelen > 1)
987 		return ENOTDIR;	/* overloaded */
988 
989 	switch(name[0]) {
990 	case NFS_NFSSTATS:
991 		if(!oldp) {
992 			*oldlenp = sizeof nfsstats;
993 			return 0;
994 		}
995 
996 		if(*oldlenp < sizeof nfsstats) {
997 			*oldlenp = sizeof nfsstats;
998 			return ENOMEM;
999 		}
1000 
1001 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
1002 		if(rv) return rv;
1003 
1004 		if(newp && newlen != sizeof nfsstats)
1005 			return EINVAL;
1006 
1007 		if(newp) {
1008 			return copyin(newp, &nfsstats, sizeof nfsstats);
1009 		}
1010 		return 0;
1011 
1012 	case NFS_IOTHREADS:
1013 		nfs_getset_niothreads(0);
1014 
1015 		rv = (sysctl_int(oldp, oldlenp, newp, newlen,
1016 		    &nfs_niothreads));
1017 
1018 		if (newp)
1019 			nfs_getset_niothreads(1);
1020 
1021 		return rv;
1022 
1023 	default:
1024 		return EOPNOTSUPP;
1025 	}
1026 }
1027 
1028 
1029 /*
1030  * At this point, this should never happen
1031  */
1032 /* ARGSUSED */
1033 int
1034 nfs_fhtovp(mp, fhp, vpp)
1035 	struct mount *mp;
1036 	struct fid *fhp;
1037 	struct vnode **vpp;
1038 {
1039 
1040 	return (EINVAL);
1041 }
1042 
1043 /* ARGSUSED */
1044 int
1045 nfs_checkexp(mp, nam, exflagsp, credanonp)
1046 	struct mount *mp;
1047 	struct mbuf *nam;
1048 	int *exflagsp;
1049 	struct ucred **credanonp;
1050 {
1051 
1052 	return (EINVAL);
1053 }
1054 
1055 /*
1056  * Vnode pointer to File handle, should never happen either
1057  */
1058 /* ARGSUSED */
1059 int
1060 nfs_vptofh(vp, fhp)
1061 	struct vnode *vp;
1062 	struct fid *fhp;
1063 {
1064 
1065 	return (EINVAL);
1066 }
1067 
1068 /*
1069  * Vfs start routine, a no-op.
1070  */
1071 /* ARGSUSED */
1072 int
1073 nfs_start(mp, flags, p)
1074 	struct mount *mp;
1075 	int flags;
1076 	struct proc *p;
1077 {
1078 
1079 	return (0);
1080 }
1081 
1082 /*
1083  * Do operations associated with quotas, not supported
1084  */
1085 /* ARGSUSED */
1086 int
1087 nfs_quotactl(mp, cmd, uid, arg, p)
1088 	struct mount *mp;
1089 	int cmd;
1090 	uid_t uid;
1091 	caddr_t arg;
1092 	struct proc *p;
1093 {
1094 
1095 	return (EOPNOTSUPP);
1096 }
1097