xref: /netbsd-src/sys/nfs/nfs_vfsops.c (revision 08c81a9c2dc8c7300e893321eb65c0925d60871c)
1 /*	$NetBSD: nfs_vfsops.c,v 1.115 2002/07/30 07:40:13 soren 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 1.115 2002/07/30 07:40:13 soren 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 int nfs_sysctl __P((int *, u_int, void *, size_t *, void *, size_t,
84 		      struct proc *));
85 
86 /*
87  * nfs vfs operations.
88  */
89 
90 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
91 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
92 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
93 
94 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = {
95 	&nfsv2_vnodeop_opv_desc,
96 	&spec_nfsv2nodeop_opv_desc,
97 	&fifo_nfsv2nodeop_opv_desc,
98 	NULL,
99 };
100 
101 struct vfsops nfs_vfsops = {
102 	MOUNT_NFS,
103 	nfs_mount,
104 	nfs_start,
105 	nfs_unmount,
106 	nfs_root,
107 	nfs_quotactl,
108 	nfs_statfs,
109 	nfs_sync,
110 	nfs_vget,
111 	nfs_fhtovp,
112 	nfs_vptofh,
113 	nfs_vfs_init,
114 	nfs_vfs_reinit,
115 	nfs_vfs_done,
116 	nfs_sysctl,
117 	nfs_mountroot,
118 	nfs_checkexp,
119 	nfs_vnodeopv_descs,
120 };
121 
122 extern u_int32_t nfs_procids[NFS_NPROCS];
123 extern u_int32_t nfs_prog, nfs_vers;
124 
125 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
126     struct mount **, struct vnode **, struct proc *));
127 
128 #define TRUE	1
129 #define	FALSE	0
130 
131 /*
132  * nfs statfs call
133  */
134 int
135 nfs_statfs(mp, sbp, p)
136 	struct mount *mp;
137 	struct statfs *sbp;
138 	struct proc *p;
139 {
140 	struct vnode *vp;
141 	struct nfs_statfs *sfp;
142 	caddr_t cp;
143 	u_int32_t *tl;
144 	int32_t t1, t2;
145 	caddr_t bpos, dpos, cp2;
146 	struct nfsmount *nmp = VFSTONFS(mp);
147 	int error = 0, retattr;
148 #ifdef NFS_V2_ONLY
149 	const int v3 = 0;
150 #else
151 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
152 #endif
153 	struct mbuf *mreq, *mrep = NULL, *md, *mb, *mb2;
154 	struct ucred *cred;
155 	u_quad_t tquad;
156 
157 #ifndef nolint
158 	sfp = (struct nfs_statfs *)0;
159 #endif
160 	vp = nmp->nm_vnode;
161 	cred = crget();
162 	cred->cr_ngroups = 0;
163 #ifndef NFS_V2_ONLY
164 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
165 		(void)nfs_fsinfo(nmp, vp, cred, p);
166 #endif
167 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
168 	nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
169 	nfsm_fhtom(vp, v3);
170 	nfsm_request(vp, NFSPROC_FSSTAT, p, cred);
171 	if (v3)
172 		nfsm_postop_attr(vp, retattr);
173 	if (error) {
174 		if (mrep != NULL)
175 			m_free(mrep);
176 		goto nfsmout;
177 	}
178 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
179 #ifdef COMPAT_09
180 	sbp->f_type = 2;
181 #else
182 	sbp->f_type = 0;
183 #endif
184 	sbp->f_flags = nmp->nm_flag;
185 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
186 	if (v3) {
187 		sbp->f_bsize = NFS_FABLKSIZE;
188 		tquad = fxdr_hyper(&sfp->sf_tbytes);
189 		sbp->f_blocks = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
190 		tquad = fxdr_hyper(&sfp->sf_fbytes);
191 		sbp->f_bfree = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
192 		tquad = fxdr_hyper(&sfp->sf_abytes);
193 		sbp->f_bavail = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
194 		tquad = fxdr_hyper(&sfp->sf_tfiles);
195 		sbp->f_files = (long)tquad;
196 		tquad = fxdr_hyper(&sfp->sf_ffiles);
197 		sbp->f_ffree = (long)tquad;
198 	} else {
199 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
200 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
201 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
202 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
203 		sbp->f_files = 0;
204 		sbp->f_ffree = 0;
205 	}
206 	if (sbp != &mp->mnt_stat) {
207 		memcpy(sbp->f_mntonname, mp->mnt_stat.f_mntonname, MNAMELEN);
208 		memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, MNAMELEN);
209 	}
210 	strncpy(&sbp->f_fstypename[0], mp->mnt_op->vfs_name, MFSNAMELEN);
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, *mb2;
234 	u_int64_t maxfsize;
235 
236 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
237 	nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
238 	nfsm_fhtom(vp, 1);
239 	nfsm_request(vp, NFSPROC_FSINFO, p, cred);
240 	nfsm_postop_attr(vp, retattr);
241 	if (!error) {
242 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
243 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
244 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
245 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
246 				~(NFS_FABLKSIZE - 1);
247 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
248 		if (max < nmp->nm_wsize && max > 0) {
249 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
250 			if (nmp->nm_wsize == 0)
251 				nmp->nm_wsize = max;
252 		}
253 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
254 		if (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 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
393 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
394 
395 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
396 			 ndmntp->ndm_args.hostname, vpp, p);
397 	if (error) {
398 		mp->mnt_op->vfs_refcount--;
399 		vfs_unbusy(mp);
400 		printf("nfs_mountroot: mount %s failed: %d\n",
401 		       mntname, error);
402 		free(mp, M_MOUNT);
403 	} else
404 		*mpp = mp;
405 
406 	return (error);
407 }
408 
409 void
410 nfs_decode_args(nmp, argp)
411 	struct nfsmount *nmp;
412 	struct nfs_args *argp;
413 {
414 	int s;
415 	int adjsock;
416 	int maxio;
417 
418 	s = splsoftnet();
419 
420 	/*
421 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
422 	 * no sense in that context.
423 	 */
424 	if (argp->sotype == SOCK_STREAM)
425 		argp->flags &= ~NFSMNT_NOCONN;
426 
427 	/*
428 	 * Cookie translation is not needed for v2, silently ignore it.
429 	 */
430 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
431 	    NFSMNT_XLATECOOKIE)
432 		argp->flags &= ~NFSMNT_XLATECOOKIE;
433 
434 	/* Re-bind if rsrvd port requested and wasn't on one */
435 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
436 		  && (argp->flags & NFSMNT_RESVPORT);
437 	/* Also re-bind if we're switching to/from a connected UDP socket */
438 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
439 		    (argp->flags & NFSMNT_NOCONN));
440 
441 	/* Update flags. */
442 	nmp->nm_flag = argp->flags;
443 	splx(s);
444 
445 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
446 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
447 		if (nmp->nm_timeo < NFS_MINTIMEO)
448 			nmp->nm_timeo = NFS_MINTIMEO;
449 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
450 			nmp->nm_timeo = NFS_MAXTIMEO;
451 	}
452 
453 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
454 		nmp->nm_retry = argp->retrans;
455 		if (nmp->nm_retry > NFS_MAXREXMIT)
456 			nmp->nm_retry = NFS_MAXREXMIT;
457 	}
458 
459 #ifndef NFS_V2_ONLY
460 	if (argp->flags & NFSMNT_NFSV3) {
461 		if (argp->sotype == SOCK_DGRAM)
462 			maxio = NFS_MAXDGRAMDATA;
463 		else
464 			maxio = NFS_MAXDATA;
465 	} else
466 #endif
467 		maxio = NFS_V2MAXDATA;
468 
469 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
470 		int osize = nmp->nm_wsize;
471 		nmp->nm_wsize = argp->wsize;
472 		/* Round down to multiple of blocksize */
473 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
474 		if (nmp->nm_wsize <= 0)
475 			nmp->nm_wsize = NFS_FABLKSIZE;
476 		adjsock |= (nmp->nm_wsize != osize);
477 	}
478 	if (nmp->nm_wsize > maxio)
479 		nmp->nm_wsize = maxio;
480 	if (nmp->nm_wsize > MAXBSIZE)
481 		nmp->nm_wsize = MAXBSIZE;
482 
483 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
484 		int osize = nmp->nm_rsize;
485 		nmp->nm_rsize = argp->rsize;
486 		/* Round down to multiple of blocksize */
487 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
488 		if (nmp->nm_rsize <= 0)
489 			nmp->nm_rsize = NFS_FABLKSIZE;
490 		adjsock |= (nmp->nm_rsize != osize);
491 	}
492 	if (nmp->nm_rsize > maxio)
493 		nmp->nm_rsize = maxio;
494 	if (nmp->nm_rsize > MAXBSIZE)
495 		nmp->nm_rsize = MAXBSIZE;
496 
497 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
498 		nmp->nm_readdirsize = argp->readdirsize;
499 		/* Round down to multiple of minimum blocksize */
500 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
501 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
502 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
503 		/* Bigger than buffer size makes no sense */
504 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
505 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
506 	} else if (argp->flags & NFSMNT_RSIZE)
507 		nmp->nm_readdirsize = nmp->nm_rsize;
508 
509 	if (nmp->nm_readdirsize > maxio)
510 		nmp->nm_readdirsize = maxio;
511 
512 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
513 		argp->maxgrouplist <= NFS_MAXGRPS)
514 		nmp->nm_numgrps = argp->maxgrouplist;
515 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
516 		argp->readahead <= NFS_MAXRAHEAD)
517 		nmp->nm_readahead = argp->readahead;
518 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
519 		argp->leaseterm <= NQ_MAXLEASE)
520 		nmp->nm_leaseterm = argp->leaseterm;
521 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
522 		argp->deadthresh <= NQ_NEVERDEAD)
523 		nmp->nm_deadthresh = argp->deadthresh;
524 
525 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
526 		    (nmp->nm_soproto != argp->proto));
527 	nmp->nm_sotype = argp->sotype;
528 	nmp->nm_soproto = argp->proto;
529 
530 	if (nmp->nm_so && adjsock) {
531 		nfs_safedisconnect(nmp);
532 		if (nmp->nm_sotype == SOCK_DGRAM)
533 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
534 				printf("nfs_args: retrying connect\n");
535 				(void) tsleep((caddr_t)&lbolt,
536 					      PSOCK, "nfscn3", 0);
537 			}
538 	}
539 }
540 
541 /*
542  * VFS Operations.
543  *
544  * mount system call
545  * It seems a bit dumb to copyinstr() the host and path here and then
546  * memcpy() them in mountnfs(), but I wanted to detect errors before
547  * doing the sockargs() call because sockargs() allocates an mbuf and
548  * an error after that means that I have to release the mbuf.
549  */
550 /* ARGSUSED */
551 int
552 nfs_mount(mp, path, data, ndp, p)
553 	struct mount *mp;
554 	const char *path;
555 	void *data;
556 	struct nameidata *ndp;
557 	struct proc *p;
558 {
559 	int error;
560 	struct nfs_args args;
561 	struct mbuf *nam;
562 	struct vnode *vp;
563 	char *pth, *hst;
564 	size_t len;
565 	u_char *nfh;
566 
567 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
568 	if (error)
569 		return (error);
570 	if (args.version != NFS_ARGSVERSION)
571 		return (EPROGMISMATCH);
572 #ifdef NFS_V2_ONLY
573 	if (args.flags & NFSMNT_NQNFS)
574 		return (EPROGUNAVAIL);
575 	if (args.flags & NFSMNT_NFSV3)
576 		return (EPROGMISMATCH);
577 #endif
578 	if (mp->mnt_flag & MNT_UPDATE) {
579 		struct nfsmount *nmp = VFSTONFS(mp);
580 
581 		if (nmp == NULL)
582 			return (EIO);
583 		/*
584 		 * When doing an update, we can't change from or to
585 		 * v3 and/or nqnfs, or change cookie translation
586 		 */
587 		args.flags = (args.flags &
588 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
589 		    (nmp->nm_flag &
590 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
591 		nfs_decode_args(nmp, &args);
592 		return (0);
593 	}
594 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
595 		return (EINVAL);
596 	MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK);
597 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
598 	if (error)
599 		return (error);
600 	MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK);
601 	error = copyinstr(path, pth, MNAMELEN - 1, &len);
602 	if (error)
603 		goto free_nfh;
604 	memset(&pth[len], 0, MNAMELEN - len);
605 	MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK);
606 	error = copyinstr(args.hostname, hst, MNAMELEN - 1, &len);
607 	if (error)
608 		goto free_pth;
609 	memset(&hst[len], 0, MNAMELEN - len);
610 	/* sockargs() call must be after above copyin() calls */
611 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
612 	if (error)
613 		goto free_hst;
614 	args.fh = nfh;
615 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
616 
617 free_hst:
618 	FREE(hst, M_TEMP);
619 free_pth:
620 	FREE(pth, M_TEMP);
621 free_nfh:
622 	FREE(nfh, M_TEMP);
623 
624 	return (error);
625 }
626 
627 /*
628  * Common code for mount and mountroot
629  */
630 int
631 mountnfs(argp, mp, nam, pth, hst, vpp, p)
632 	struct nfs_args *argp;
633 	struct mount *mp;
634 	struct mbuf *nam;
635 	const char *pth, *hst;
636 	struct vnode **vpp;
637 	struct proc *p;
638 {
639 	struct nfsmount *nmp;
640 	struct nfsnode *np;
641 	int error;
642 	struct vattr *attrs;
643 	struct ucred *cr;
644 
645 	/*
646 	 * If the number of nfs iothreads to use has never
647 	 * been set, create a reasonable number of them.
648 	 */
649 
650 	if (nfs_niothreads < 0) {
651 		nfs_niothreads = NFS_DEFAULT_NIOTHREADS;
652 		nfs_getset_niothreads(TRUE);
653 	}
654 
655 	if (mp->mnt_flag & MNT_UPDATE) {
656 		nmp = VFSTONFS(mp);
657 		/* update paths, file handles, etc, here	XXX */
658 		m_freem(nam);
659 		return (0);
660 	} else {
661 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
662 		    M_NFSMNT, M_WAITOK);
663 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
664 		mp->mnt_data = nmp;
665 		TAILQ_INIT(&nmp->nm_uidlruhead);
666 		TAILQ_INIT(&nmp->nm_bufq);
667 	}
668 	vfs_getnewfsid(mp);
669 	nmp->nm_mountp = mp;
670 
671 #ifndef NFS_V2_ONLY
672 	if (argp->flags & NFSMNT_NQNFS)
673 		/*
674 		 * We have to set mnt_maxsymlink to a non-zero value so
675 		 * that COMPAT_43 routines will know that we are setting
676 		 * the d_type field in directories (and can zero it for
677 		 * unsuspecting binaries).
678 		 */
679 		mp->mnt_maxsymlinklen = 1;
680 #endif
681 
682 #ifndef NFS_V2_ONLY
683 	if ((argp->flags & NFSMNT_NFSV3) == 0)
684 #endif
685 		/*
686 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
687 		 * will fill this in.
688 		 */
689 		nmp->nm_maxfilesize = 0xffffffffLL;
690 
691 	nmp->nm_timeo = NFS_TIMEO;
692 	nmp->nm_retry = NFS_RETRANS;
693 	nmp->nm_wsize = NFS_WSIZE;
694 	nmp->nm_rsize = NFS_RSIZE;
695 	nmp->nm_readdirsize = NFS_READDIRSIZE;
696 	nmp->nm_numgrps = NFS_MAXGRPS;
697 	nmp->nm_readahead = NFS_DEFRAHEAD;
698 	nmp->nm_leaseterm = NQ_DEFLEASE;
699 	nmp->nm_deadthresh = NQ_DEADTHRESH;
700 	CIRCLEQ_INIT(&nmp->nm_timerhead);
701 	nmp->nm_inprog = NULLVP;
702 #ifdef COMPAT_09
703 	mp->mnt_stat.f_type = 2;
704 #else
705 	mp->mnt_stat.f_type = 0;
706 #endif
707 	strncpy(&mp->mnt_stat.f_fstypename[0], mp->mnt_op->vfs_name,
708 	    MFSNAMELEN);
709 	memcpy(mp->mnt_stat.f_mntfromname, hst, MNAMELEN);
710 	memcpy(mp->mnt_stat.f_mntonname, pth, MNAMELEN);
711 	nmp->nm_nam = nam;
712 
713 	/* Set up the sockets and per-host congestion */
714 	nmp->nm_sotype = argp->sotype;
715 	nmp->nm_soproto = argp->proto;
716 
717 	nfs_decode_args(nmp, argp);
718 
719 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
720 	mp->mnt_dev_bshift = DEV_BSHIFT;
721 
722 	/*
723 	 * For Connection based sockets (TCP,...) defer the connect until
724 	 * the first request, in case the server is not responding.
725 	 */
726 	if (nmp->nm_sotype == SOCK_DGRAM &&
727 		(error = nfs_connect(nmp, (struct nfsreq *)0)))
728 		goto bad;
729 
730 	/*
731 	 * This is silly, but it has to be set so that vinifod() works.
732 	 * We do not want to do an nfs_statfs() here since we can get
733 	 * stuck on a dead server and we are holding a lock on the mount
734 	 * point.
735 	 */
736 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
737 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
738 	if (error)
739 		goto bad;
740 	*vpp = NFSTOV(np);
741 	MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK);
742 	VOP_GETATTR(*vpp, attrs, p->p_ucred, p);
743 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
744 		cr = crget();
745 		cr->cr_uid = attrs->va_uid;
746 		cr->cr_gid = attrs->va_gid;
747 		cr->cr_ngroups = 0;
748 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
749 		crfree(cr);
750 	}
751 	FREE(attrs, M_TEMP);
752 
753 	/*
754 	 * A reference count is needed on the nfsnode representing the
755 	 * remote root.  If this object is not persistent, then backward
756 	 * traversals of the mount point (i.e. "..") will not work if
757 	 * the nfsnode gets flushed out of the cache. Ufs does not have
758 	 * this problem, because one can identify root inodes by their
759 	 * number == ROOTINO (2). So, just unlock, but no rele.
760 	 */
761 
762 	nmp->nm_vnode = *vpp;
763 	VOP_UNLOCK(*vpp, 0);
764 
765 	return (0);
766 bad:
767 	nfs_disconnect(nmp);
768 	free((caddr_t)nmp, M_NFSMNT);
769 	m_freem(nam);
770 	return (error);
771 }
772 
773 /*
774  * unmount system call
775  */
776 int
777 nfs_unmount(mp, mntflags, p)
778 	struct mount *mp;
779 	int mntflags;
780 	struct proc *p;
781 {
782 	struct nfsmount *nmp;
783 	struct vnode *vp;
784 	int error, flags = 0;
785 
786 	if (mntflags & MNT_FORCE)
787 		flags |= FORCECLOSE;
788 	nmp = VFSTONFS(mp);
789 	/*
790 	 * Goes something like this..
791 	 * - Check for activity on the root vnode (other than ourselves).
792 	 * - Call vflush() to clear out vnodes for this file system,
793 	 *   except for the root vnode.
794 	 * - Decrement reference on the vnode representing remote root.
795 	 * - Close the socket
796 	 * - Free up the data structures
797 	 */
798 	/*
799 	 * We need to decrement the ref. count on the nfsnode representing
800 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
801 	 * has done vput on this vnode, otherwise we would get deadlock!
802 	 */
803 	vp = nmp->nm_vnode;
804 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
805 	if (error != 0)
806 		return error;
807 
808 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
809 		vput(vp);
810 		return (EBUSY);
811 	}
812 
813 	/*
814 	 * Must handshake with nqnfs_clientd() if it is active.
815 	 */
816 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
817 	while (nmp->nm_inprog != NULLVP)
818 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
819 	error = vflush(mp, vp, flags);
820 	if (error) {
821 		vput(vp);
822 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
823 		return (error);
824 	}
825 
826 	/*
827 	 * We are now committed to the unmount; mark the mount structure
828 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
829 	 * will go away cleanly.
830 	 */
831 	nmp->nm_iflag |= NFSMNT_DISMNT;
832 
833 	/*
834 	 * There are two reference counts to get rid of here
835 	 * (see comment in mountnfs()).
836 	 */
837 	vrele(vp);
838 	vput(vp);
839 	vgone(vp);
840 	nfs_disconnect(nmp);
841 	m_freem(nmp->nm_nam);
842 
843 	/*
844 	 * For NQNFS, let the server daemon free the nfsmount structure.
845 	 */
846 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
847 		free((caddr_t)nmp, M_NFSMNT);
848 	return (0);
849 }
850 
851 /*
852  * Return root of a filesystem
853  */
854 int
855 nfs_root(mp, vpp)
856 	struct mount *mp;
857 	struct vnode **vpp;
858 {
859 	struct vnode *vp;
860 	struct nfsmount *nmp;
861 	int error;
862 
863 	nmp = VFSTONFS(mp);
864 	vp = nmp->nm_vnode;
865 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
866 	if (error != 0)
867 		return error;
868 	if (vp->v_type == VNON)
869 		vp->v_type = VDIR;
870 	vp->v_flag = VROOT;
871 	*vpp = vp;
872 	return (0);
873 }
874 
875 extern int syncprt;
876 
877 /*
878  * Flush out the buffer cache
879  */
880 /* ARGSUSED */
881 int
882 nfs_sync(mp, waitfor, cred, p)
883 	struct mount *mp;
884 	int waitfor;
885 	struct ucred *cred;
886 	struct proc *p;
887 {
888 	struct vnode *vp;
889 	int error, allerror = 0;
890 
891 	/*
892 	 * Force stale buffer cache information to be flushed.
893 	 */
894 loop:
895 	for (vp = mp->mnt_vnodelist.lh_first;
896 	     vp != NULL;
897 	     vp = vp->v_mntvnodes.le_next) {
898 		/*
899 		 * If the vnode that we are about to sync is no longer
900 		 * associated with this mount point, start over.
901 		 */
902 		if (vp->v_mount != mp)
903 			goto loop;
904 		if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) ||
905 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
906 		     vp->v_uobj.uo_npages == 0))
907 			continue;
908 		if (vget(vp, LK_EXCLUSIVE))
909 			goto loop;
910 		error = VOP_FSYNC(vp, cred,
911 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p);
912 		if (error)
913 			allerror = error;
914 		vput(vp);
915 	}
916 	return (allerror);
917 }
918 
919 /*
920  * NFS flat namespace lookup.
921  * Currently unsupported.
922  */
923 /* ARGSUSED */
924 int
925 nfs_vget(mp, ino, vpp)
926 	struct mount *mp;
927 	ino_t ino;
928 	struct vnode **vpp;
929 {
930 
931 	return (EOPNOTSUPP);
932 }
933 
934 /*
935  * Do that sysctl thang...
936  */
937 int
938 nfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
939 	int *name;
940 	u_int namelen;
941 	void *oldp;
942 	size_t *oldlenp;
943 	void *newp;
944 	size_t newlen;
945 	struct proc *p;
946 {
947 	int rv;
948 
949 	/*
950 	 * All names at this level are terminal.
951 	 */
952 	if(namelen > 1)
953 		return ENOTDIR;	/* overloaded */
954 
955 	switch(name[0]) {
956 	case NFS_NFSSTATS:
957 		if(!oldp) {
958 			*oldlenp = sizeof nfsstats;
959 			return 0;
960 		}
961 
962 		if(*oldlenp < sizeof nfsstats) {
963 			*oldlenp = sizeof nfsstats;
964 			return ENOMEM;
965 		}
966 
967 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
968 		if(rv) return rv;
969 
970 		if(newp && newlen != sizeof nfsstats)
971 			return EINVAL;
972 
973 		if(newp) {
974 			return copyin(newp, &nfsstats, sizeof nfsstats);
975 		}
976 		return 0;
977 
978 	case NFS_IOTHREADS:
979 		nfs_getset_niothreads(0);
980 
981 		rv = (sysctl_int(oldp, oldlenp, newp, newlen,
982 		    &nfs_niothreads));
983 
984 		if (newp)
985 			nfs_getset_niothreads(1);
986 
987 		return rv;
988 
989 	default:
990 		return EOPNOTSUPP;
991 	}
992 }
993 
994 
995 /*
996  * At this point, this should never happen
997  */
998 /* ARGSUSED */
999 int
1000 nfs_fhtovp(mp, fhp, vpp)
1001 	struct mount *mp;
1002 	struct fid *fhp;
1003 	struct vnode **vpp;
1004 {
1005 
1006 	return (EINVAL);
1007 }
1008 
1009 /* ARGSUSED */
1010 int
1011 nfs_checkexp(mp, nam, exflagsp, credanonp)
1012 	struct mount *mp;
1013 	struct mbuf *nam;
1014 	int *exflagsp;
1015 	struct ucred **credanonp;
1016 {
1017 
1018 	return (EINVAL);
1019 }
1020 
1021 /*
1022  * Vnode pointer to File handle, should never happen either
1023  */
1024 /* ARGSUSED */
1025 int
1026 nfs_vptofh(vp, fhp)
1027 	struct vnode *vp;
1028 	struct fid *fhp;
1029 {
1030 
1031 	return (EINVAL);
1032 }
1033 
1034 /*
1035  * Vfs start routine, a no-op.
1036  */
1037 /* ARGSUSED */
1038 int
1039 nfs_start(mp, flags, p)
1040 	struct mount *mp;
1041 	int flags;
1042 	struct proc *p;
1043 {
1044 
1045 	return (0);
1046 }
1047 
1048 /*
1049  * Do operations associated with quotas, not supported
1050  */
1051 /* ARGSUSED */
1052 int
1053 nfs_quotactl(mp, cmd, uid, arg, p)
1054 	struct mount *mp;
1055 	int cmd;
1056 	uid_t uid;
1057 	caddr_t arg;
1058 	struct proc *p;
1059 {
1060 
1061 	return (EOPNOTSUPP);
1062 }
1063