Home | History | Annotate | Line # | Download | only in nfs
nfs_vfsops.c revision 1.74
      1 /*	$NetBSD: nfs_vfsops.c,v 1.74 1998/06/22 22:01:06 sommerfe 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 "opt_fifo.h"
     42 
     43 #include <sys/param.h>
     44 #include <sys/conf.h>
     45 #include <sys/ioctl.h>
     46 #include <sys/signal.h>
     47 #include <sys/proc.h>
     48 #include <sys/namei.h>
     49 #include <sys/device.h>
     50 #include <sys/vnode.h>
     51 #include <sys/kernel.h>
     52 #include <sys/mount.h>
     53 #include <sys/buf.h>
     54 #include <sys/mbuf.h>
     55 #include <sys/socket.h>
     56 #include <sys/socketvar.h>
     57 #include <sys/systm.h>
     58 
     59 #include <net/if.h>
     60 #include <net/route.h>
     61 #include <netinet/in.h>
     62 
     63 #include <nfs/rpcv2.h>
     64 #include <nfs/nfsproto.h>
     65 #include <nfs/nfsnode.h>
     66 #include <nfs/nfs.h>
     67 #include <nfs/nfsmount.h>
     68 #include <nfs/xdr_subs.h>
     69 #include <nfs/nfsm_subs.h>
     70 #include <nfs/nfsdiskless.h>
     71 #include <nfs/nqnfs.h>
     72 #include <nfs/nfs_var.h>
     73 
     74 extern struct nfsstats nfsstats;
     75 extern int nfs_ticks;
     76 
     77 int nfs_sysctl __P((int *, u_int, void *, size_t *, void *, size_t,
     78 		      struct proc *));
     79 
     80 /*
     81  * nfs vfs operations.
     82  */
     83 
     84 extern struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
     85 extern struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
     86 #ifdef FIFO
     87 extern struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
     88 #endif
     89 
     90 struct vnodeopv_desc *nfs_vnodeopv_descs[] = {
     91 	&nfsv2_vnodeop_opv_desc,
     92 	&spec_nfsv2nodeop_opv_desc,
     93 #ifdef FIFO
     94 	&fifo_nfsv2nodeop_opv_desc,
     95 #endif
     96 	NULL,
     97 };
     98 
     99 struct vfsops nfs_vfsops = {
    100 	MOUNT_NFS,
    101 	nfs_mount,
    102 	nfs_start,
    103 	nfs_unmount,
    104 	nfs_root,
    105 	nfs_quotactl,
    106 	nfs_statfs,
    107 	nfs_sync,
    108 	nfs_vget,
    109 	nfs_fhtovp,
    110 	nfs_vptofh,
    111 	nfs_vfs_init,
    112 	nfs_sysctl,
    113 	nfs_mountroot,
    114 	nfs_vnodeopv_descs,
    115 };
    116 
    117 extern u_int32_t nfs_procids[NFS_NPROCS];
    118 extern u_int32_t nfs_prog, nfs_vers;
    119 
    120 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
    121     struct mount **, struct vnode **, struct proc *));
    122 
    123 #define TRUE	1
    124 #define	FALSE	0
    125 
    126 /*
    127  * nfs statfs call
    128  */
    129 int
    130 nfs_statfs(mp, sbp, p)
    131 	struct mount *mp;
    132 	register struct statfs *sbp;
    133 	struct proc *p;
    134 {
    135 	register struct vnode *vp;
    136 	register struct nfs_statfs *sfp;
    137 	register caddr_t cp;
    138 	register u_int32_t *tl;
    139 	register int32_t t1, t2;
    140 	caddr_t bpos, dpos, cp2;
    141 	struct nfsmount *nmp = VFSTONFS(mp);
    142 	int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
    143 	struct mbuf *mreq, *mrep = NULL, *md, *mb, *mb2;
    144 	struct ucred *cred;
    145 	struct nfsnode *np;
    146 	u_quad_t tquad;
    147 
    148 #ifndef nolint
    149 	sfp = (struct nfs_statfs *)0;
    150 #endif
    151 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
    152 	if (error)
    153 		return (error);
    154 	vp = NFSTOV(np);
    155 	cred = crget();
    156 	cred->cr_ngroups = 0;
    157 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
    158 		(void)nfs_fsinfo(nmp, vp, cred, p);
    159 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
    160 	nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
    161 	nfsm_fhtom(vp, v3);
    162 	nfsm_request(vp, NFSPROC_FSSTAT, p, cred);
    163 	if (v3)
    164 		nfsm_postop_attr(vp, retattr);
    165 	if (error) {
    166 		if (mrep != NULL)
    167 			m_free(mrep);
    168 		goto nfsmout;
    169 	}
    170 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
    171 #ifdef COMPAT_09
    172 	sbp->f_type = 2;
    173 #else
    174 	sbp->f_type = 0;
    175 #endif
    176 	sbp->f_flags = nmp->nm_flag;
    177 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
    178 	if (v3) {
    179 		sbp->f_bsize = NFS_FABLKSIZE;
    180 		fxdr_hyper(&sfp->sf_tbytes, &tquad);
    181 		sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
    182 		fxdr_hyper(&sfp->sf_fbytes, &tquad);
    183 		sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
    184 		fxdr_hyper(&sfp->sf_abytes, &tquad);
    185 		sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
    186 		sbp->f_files = (fxdr_unsigned(int32_t,
    187 		    sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
    188 		sbp->f_ffree = (fxdr_unsigned(int32_t,
    189 		    sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
    190 	} else {
    191 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
    192 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
    193 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
    194 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
    195 		sbp->f_files = 0;
    196 		sbp->f_ffree = 0;
    197 	}
    198 	if (sbp != &mp->mnt_stat) {
    199 		bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
    200 		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
    201 	}
    202 	strncpy(&sbp->f_fstypename[0], mp->mnt_op->vfs_name, MFSNAMELEN);
    203 	nfsm_reqdone;
    204 	vrele(vp);
    205 	crfree(cred);
    206 	return (error);
    207 }
    208 
    209 /*
    210  * nfs version 3 fsinfo rpc call
    211  */
    212 int
    213 nfs_fsinfo(nmp, vp, cred, p)
    214 	register struct nfsmount *nmp;
    215 	register struct vnode *vp;
    216 	struct ucred *cred;
    217 	struct proc *p;
    218 {
    219 	register struct nfsv3_fsinfo *fsp;
    220 	register caddr_t cp;
    221 	register int32_t t1, t2;
    222 	register u_int32_t *tl, pref, max;
    223 	caddr_t bpos, dpos, cp2;
    224 	int error = 0, retattr;
    225 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
    226 	u_int64_t maxfsize;
    227 
    228 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
    229 	nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
    230 	nfsm_fhtom(vp, 1);
    231 	nfsm_request(vp, NFSPROC_FSINFO, p, cred);
    232 	nfsm_postop_attr(vp, retattr);
    233 	if (!error) {
    234 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
    235 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
    236 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
    237 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
    238 				~(NFS_FABLKSIZE - 1);
    239 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
    240 		if (max < nmp->nm_wsize && max > 0) {
    241 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
    242 			if (nmp->nm_wsize == 0)
    243 				nmp->nm_wsize = max;
    244 		}
    245 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
    246 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
    247 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
    248 				~(NFS_FABLKSIZE - 1);
    249 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
    250 		if (max < nmp->nm_rsize && max > 0) {
    251 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
    252 			if (nmp->nm_rsize == 0)
    253 				nmp->nm_rsize = max;
    254 		}
    255 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
    256 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
    257 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
    258 				~(NFS_DIRFRAGSIZ - 1);
    259 		if (max < nmp->nm_readdirsize && max > 0) {
    260 			nmp->nm_readdirsize = max & ~(NFS_DIRFRAGSIZ - 1);
    261 			if (nmp->nm_readdirsize == 0)
    262 				nmp->nm_readdirsize = max;
    263 		}
    264 		/* XXX */
    265 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
    266 		fxdr_hyper(&fsp->fs_maxfilesize, &maxfsize);
    267 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
    268 			nmp->nm_maxfilesize = maxfsize;
    269 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
    270 	}
    271 	nfsm_reqdone;
    272 	return (error);
    273 }
    274 
    275 /*
    276  * Mount a remote root fs via. NFS.  It goes like this:
    277  * - Call nfs_boot_init() to fill in the nfs_diskless struct
    278  * - build the rootfs mount point and call mountnfs() to do the rest.
    279  */
    280 int
    281 nfs_mountroot()
    282 {
    283 	struct nfs_diskless *nd;
    284 	struct vattr attr;
    285 	struct mount *mp;
    286 	struct vnode *vp;
    287 	struct proc *procp;
    288 	long n;
    289 	int error;
    290 
    291 	procp = curproc; /* XXX */
    292 
    293 	if (root_device->dv_class != DV_IFNET)
    294 		return (ENODEV);
    295 
    296 	/*
    297 	 * XXX time must be non-zero when we init the interface or else
    298 	 * the arp code will wedge.  [Fixed now in if_ether.c]
    299 	 * However, the NFS attribute cache gives false "hits" when
    300 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
    301 	 */
    302 	if (time.tv_sec < NFS_MAXATTRTIMO)
    303 		time.tv_sec = NFS_MAXATTRTIMO;
    304 
    305 	/*
    306 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
    307 	 * Side effect:  Finds and configures a network interface.
    308 	 */
    309 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
    310 	bzero((caddr_t)nd, sizeof(*nd));
    311 	error = nfs_boot_init(nd, procp);
    312 	if (error)
    313 		goto out;
    314 
    315 	/*
    316 	 * Create the root mount point.
    317 	 */
    318 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp);
    319 	if (error)
    320 		goto out;
    321 	printf("root on %s\n", nd->nd_root.ndm_host);
    322 
    323 	/*
    324 	 * Link it into the mount list.
    325 	 */
    326 	simple_lock(&mountlist_slock);
    327 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
    328 	simple_unlock(&mountlist_slock);
    329 	rootvp = vp;
    330 	mp->mnt_vnodecovered = NULLVP;
    331 	vfs_unbusy(mp);
    332 
    333 	/* Get root attributes (for the time). */
    334 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
    335 	if (error)
    336 		panic("nfs_mountroot: getattr for root");
    337 	n = attr.va_mtime.tv_sec;
    338 #ifdef	DEBUG
    339 	printf("root time: 0x%lx\n", n);
    340 #endif
    341 	inittodr(n);
    342 
    343 #if 0
    344 	/*
    345 	 * XXX splnet, so networks will receive...
    346 	 * XXX Which port needed this hack?
    347 	 * XXX Should already be at spl0.
    348 	 */
    349 	splnet();
    350 #endif
    351 
    352 #if 0	/* swap now comes in from swapctl(2) */
    353 #ifdef notyet
    354 	/* Set up swap credentials. */
    355 	proc0.p_ucred->cr_uid = 0;
    356 	proc0.p_ucred->cr_gid = 0;
    357 	proc0.p_ucred->cr_ngroups = 0;
    358 #endif
    359 
    360 	/*
    361 	 * "Mount" the swap device.
    362 	 *
    363 	 * On a "dataless" configuration (swap on disk) we will have:
    364 	 *	(swdevt[0].sw_dev != NODEV) identifying the swap device.
    365 	 */
    366 	if (bdevvp(swapdev, &swapdev_vp))
    367 		panic("nfs_mountroot: can't setup swap vp");
    368 	if (swdevt[0].sw_dev != NODEV) {
    369 		printf("swap on device 0x%x\n", swdevt[0].sw_dev);
    370 		error = 0;
    371 		goto out;
    372 	}
    373 
    374 	/*
    375 	 * If swapping to an nfs node:  (swdevt[0].sw_dev == NODEV)
    376 	 * Create a fake mount point just for the swap vnode so that the
    377 	 * swap file can be on a different server from the rootfs.
    378 	 */
    379 	error = nfs_mount_diskless(&nd->nd_swap, "/swap", &mp, &vp, procp);
    380 	if (error) {
    381 		printf("nfs_boot: warning: mount(swap), error=%d\n", error);
    382 		error = 0;
    383 		goto out;
    384 	}
    385 	vfs_unbusy(mp);
    386 	printf("swap on %s\n", nd->nd_swap.ndm_host);
    387 
    388 	/*
    389 	 * Since the swap file is not the root dir of a file system,
    390 	 * hack it to a regular file.
    391 	 */
    392 	vp->v_type = VREG;
    393 	vp->v_flag = 0;
    394 	swdevt[0].sw_vp = vp;
    395 
    396 	/*
    397 	 * Find out how large the swap file is.
    398 	 */
    399 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
    400 	if (error)
    401 		panic("nfs_mountroot: getattr for swap");
    402 	n = (long) (attr.va_size >> DEV_BSHIFT);
    403 #ifdef	DEBUG
    404 	printf("swap size: 0x%lx (blocks)\n", n);
    405 #endif
    406 	swdevt[0].sw_nblks = n;
    407 #endif
    408 
    409 out:
    410 	free(nd, M_NFSMNT);
    411 	return (error);
    412 }
    413 
    414 /*
    415  * Internal version of mount system call for diskless setup.
    416  * Separate function because we used to call it twice.
    417  * (once for root and once for swap)
    418  */
    419 static int
    420 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p)
    421 	struct nfs_dlmount *ndmntp;
    422 	const char *mntname;	/* mount point name */
    423 	struct mount **mpp;
    424 	struct vnode **vpp;
    425 	struct proc *p;
    426 {
    427 	struct mount *mp;
    428 	struct mbuf *m;
    429 	int error;
    430 
    431 	vfs_rootmountalloc(MOUNT_NFS, (char *)mntname, &mp);
    432 
    433 	mp->mnt_op = &nfs_vfsops;
    434 
    435 	/*
    436 	 * Historical practice expects NFS root file systems to
    437 	 * be initially mounted r/w.
    438 	 */
    439 	mp->mnt_flag &= ~MNT_RDONLY;
    440 
    441 	/* Get mbuf for server sockaddr. */
    442 	m = m_get(M_WAIT, MT_SONAME);
    443 	if (m == NULL)
    444 		panic("nfs_mountroot: mget soname for %s", mntname);
    445 	bcopy((caddr_t)ndmntp->ndm_args.addr, mtod(m, caddr_t),
    446 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
    447 
    448 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
    449 			 ndmntp->ndm_args.hostname, vpp, p);
    450 	if (error) {
    451 		mp->mnt_op->vfs_refcount--;
    452 		vfs_unbusy(mp);
    453 		printf("nfs_mountroot: mount %s failed: %d\n",
    454 		       mntname, error);
    455 		free(mp, M_MOUNT);
    456 	} else
    457 		*mpp = mp;
    458 
    459 	return (error);
    460 }
    461 
    462 void
    463 nfs_decode_args(nmp, argp)
    464 	struct nfsmount *nmp;
    465 	struct nfs_args *argp;
    466 {
    467 	int s;
    468 	int adjsock;
    469 	int maxio;
    470 
    471 	s = splsoftnet();
    472 
    473 	/*
    474 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
    475 	 * no sense in that context.
    476 	 */
    477 	if (argp->sotype == SOCK_STREAM)
    478 		argp->flags &= ~NFSMNT_NOCONN;
    479 
    480 	/*
    481 	 * Cookie translation is not needed for v2, silently ignore it.
    482 	 */
    483 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
    484 	    NFSMNT_XLATECOOKIE)
    485 		argp->flags &= ~NFSMNT_XLATECOOKIE;
    486 
    487 	/* Re-bind if rsrvd port requested and wasn't on one */
    488 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
    489 		  && (argp->flags & NFSMNT_RESVPORT);
    490 	/* Also re-bind if we're switching to/from a connected UDP socket */
    491 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
    492 		    (argp->flags & NFSMNT_NOCONN));
    493 
    494 	/* Update flags atomically.  Don't change the lock bits. */
    495 	nmp->nm_flag = argp->flags | nmp->nm_flag;
    496 	splx(s);
    497 
    498 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
    499 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
    500 		if (nmp->nm_timeo < NFS_MINTIMEO)
    501 			nmp->nm_timeo = NFS_MINTIMEO;
    502 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
    503 			nmp->nm_timeo = NFS_MAXTIMEO;
    504 	}
    505 
    506 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
    507 		nmp->nm_retry = argp->retrans;
    508 		if (nmp->nm_retry > NFS_MAXREXMIT)
    509 			nmp->nm_retry = NFS_MAXREXMIT;
    510 	}
    511 
    512 	if (argp->flags & NFSMNT_NFSV3) {
    513 		if (argp->sotype == SOCK_DGRAM)
    514 			maxio = NFS_MAXDGRAMDATA;
    515 		else
    516 			maxio = NFS_MAXDATA;
    517 	} else
    518 		maxio = NFS_V2MAXDATA;
    519 
    520 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
    521 		int osize = nmp->nm_wsize;
    522 		nmp->nm_wsize = argp->wsize;
    523 		/* Round down to multiple of blocksize */
    524 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
    525 		if (nmp->nm_wsize <= 0)
    526 			nmp->nm_wsize = NFS_FABLKSIZE;
    527 		adjsock |= (nmp->nm_wsize != osize);
    528 	}
    529 	if (nmp->nm_wsize > maxio)
    530 		nmp->nm_wsize = maxio;
    531 	if (nmp->nm_wsize > MAXBSIZE)
    532 		nmp->nm_wsize = MAXBSIZE;
    533 
    534 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
    535 		int osize = nmp->nm_rsize;
    536 		nmp->nm_rsize = argp->rsize;
    537 		/* Round down to multiple of blocksize */
    538 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
    539 		if (nmp->nm_rsize <= 0)
    540 			nmp->nm_rsize = NFS_FABLKSIZE;
    541 		adjsock |= (nmp->nm_rsize != osize);
    542 	}
    543 	if (nmp->nm_rsize > maxio)
    544 		nmp->nm_rsize = maxio;
    545 	if (nmp->nm_rsize > MAXBSIZE)
    546 		nmp->nm_rsize = MAXBSIZE;
    547 
    548 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
    549 		nmp->nm_readdirsize = argp->readdirsize;
    550 		/* Round down to multiple of minimum blocksize */
    551 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
    552 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
    553 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
    554 		/* Bigger than buffer size makes no sense */
    555 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
    556 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
    557 	} else if (argp->flags & NFSMNT_RSIZE)
    558 		nmp->nm_readdirsize = nmp->nm_rsize;
    559 
    560 	if (nmp->nm_readdirsize > maxio)
    561 		nmp->nm_readdirsize = maxio;
    562 
    563 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
    564 		argp->maxgrouplist <= NFS_MAXGRPS)
    565 		nmp->nm_numgrps = argp->maxgrouplist;
    566 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
    567 		argp->readahead <= NFS_MAXRAHEAD)
    568 		nmp->nm_readahead = argp->readahead;
    569 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
    570 		argp->leaseterm <= NQ_MAXLEASE)
    571 		nmp->nm_leaseterm = argp->leaseterm;
    572 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
    573 		argp->deadthresh <= NQ_NEVERDEAD)
    574 		nmp->nm_deadthresh = argp->deadthresh;
    575 
    576 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
    577 		    (nmp->nm_soproto != argp->proto));
    578 	nmp->nm_sotype = argp->sotype;
    579 	nmp->nm_soproto = argp->proto;
    580 
    581 	if (nmp->nm_so && adjsock) {
    582 		nfs_safedisconnect(nmp);
    583 		if (nmp->nm_sotype == SOCK_DGRAM)
    584 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
    585 				printf("nfs_args: retrying connect\n");
    586 				(void) tsleep((caddr_t)&lbolt,
    587 					      PSOCK, "nfscon", 0);
    588 			}
    589 	}
    590 }
    591 
    592 /*
    593  * VFS Operations.
    594  *
    595  * mount system call
    596  * It seems a bit dumb to copyinstr() the host and path here and then
    597  * bcopy() them in mountnfs(), but I wanted to detect errors before
    598  * doing the sockargs() call because sockargs() allocates an mbuf and
    599  * an error after that means that I have to release the mbuf.
    600  */
    601 /* ARGSUSED */
    602 int
    603 nfs_mount(mp, path, data, ndp, p)
    604 	struct mount *mp;
    605 	const char *path;
    606 	void *data;
    607 	struct nameidata *ndp;
    608 	struct proc *p;
    609 {
    610 	int error;
    611 	struct nfs_args args;
    612 	struct mbuf *nam;
    613 	struct vnode *vp;
    614 	char pth[MNAMELEN], hst[MNAMELEN];
    615 	size_t len;
    616 	u_char nfh[NFSX_V3FHMAX];
    617 
    618 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
    619 	if (error)
    620 		return (error);
    621 	if (args.version != NFS_ARGSVERSION)
    622 		return (EPROGMISMATCH);
    623 	if (mp->mnt_flag & MNT_UPDATE) {
    624 		register struct nfsmount *nmp = VFSTONFS(mp);
    625 
    626 		if (nmp == NULL)
    627 			return (EIO);
    628 		/*
    629 		 * When doing an update, we can't change from or to
    630 		 * v3 and/or nqnfs, or change cookie translation
    631 		 */
    632 		args.flags = (args.flags &
    633 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
    634 		    (nmp->nm_flag &
    635 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
    636 		nfs_decode_args(nmp, &args);
    637 		return (0);
    638 	}
    639 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
    640 	if (error)
    641 		return (error);
    642 	error = copyinstr(path, pth, MNAMELEN-1, &len);
    643 	if (error)
    644 		return (error);
    645 	bzero(&pth[len], MNAMELEN - len);
    646 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
    647 	if (error)
    648 		return (error);
    649 	bzero(&hst[len], MNAMELEN - len);
    650 	/* sockargs() call must be after above copyin() calls */
    651 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
    652 	if (error)
    653 		return (error);
    654 	args.fh = nfh;
    655 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
    656 	return (error);
    657 }
    658 
    659 /*
    660  * Common code for mount and mountroot
    661  */
    662 int
    663 mountnfs(argp, mp, nam, pth, hst, vpp, p)
    664 	register struct nfs_args *argp;
    665 	register struct mount *mp;
    666 	struct mbuf *nam;
    667 	const char *pth, *hst;
    668 	struct vnode **vpp;
    669 	struct proc *p;
    670 {
    671 	register struct nfsmount *nmp;
    672 	struct nfsnode *np;
    673 	int error;
    674 	struct vattr attrs;
    675 	struct ucred *cr;
    676 
    677 	if (mp->mnt_flag & MNT_UPDATE) {
    678 		nmp = VFSTONFS(mp);
    679 		/* update paths, file handles, etc, here	XXX */
    680 		m_freem(nam);
    681 		return (0);
    682 	} else {
    683 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
    684 		    M_NFSMNT, M_WAITOK);
    685 		bzero((caddr_t)nmp, sizeof (struct nfsmount));
    686 		mp->mnt_data = (qaddr_t)nmp;
    687 		TAILQ_INIT(&nmp->nm_uidlruhead);
    688 		TAILQ_INIT(&nmp->nm_bufq);
    689 	}
    690 	vfs_getnewfsid(mp, MOUNT_NFS);
    691 	nmp->nm_mountp = mp;
    692 
    693 	if (argp->flags & NFSMNT_NQNFS)
    694 		/*
    695 		 * We have to set mnt_maxsymlink to a non-zero value so
    696 		 * that COMPAT_43 routines will know that we are setting
    697 		 * the d_type field in directories (and can zero it for
    698 		 * unsuspecting binaries).
    699 		 */
    700 		mp->mnt_maxsymlinklen = 1;
    701 
    702 	if ((argp->flags & NFSMNT_NFSV3) == 0)
    703 		/*
    704 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
    705 		 * will fill this in.
    706 		 */
    707 		nmp->nm_maxfilesize = 0xffffffffLL;
    708 
    709 	nmp->nm_timeo = NFS_TIMEO;
    710 	nmp->nm_retry = NFS_RETRANS;
    711 	nmp->nm_wsize = NFS_WSIZE;
    712 	nmp->nm_rsize = NFS_RSIZE;
    713 	nmp->nm_readdirsize = NFS_READDIRSIZE;
    714 	nmp->nm_numgrps = NFS_MAXGRPS;
    715 	nmp->nm_readahead = NFS_DEFRAHEAD;
    716 	nmp->nm_leaseterm = NQ_DEFLEASE;
    717 	nmp->nm_deadthresh = NQ_DEADTHRESH;
    718 	CIRCLEQ_INIT(&nmp->nm_timerhead);
    719 	nmp->nm_inprog = NULLVP;
    720 	nmp->nm_fhsize = argp->fhsize;
    721 	bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
    722 #ifdef COMPAT_09
    723 	mp->mnt_stat.f_type = 2;
    724 #else
    725 	mp->mnt_stat.f_type = 0;
    726 #endif
    727 	strncpy(&mp->mnt_stat.f_fstypename[0], mp->mnt_op->vfs_name,
    728 	    MFSNAMELEN);
    729 	bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
    730 	bcopy(pth, mp->mnt_stat.f_mntonname, MNAMELEN);
    731 	nmp->nm_nam = nam;
    732 
    733 	/* Set up the sockets and per-host congestion */
    734 	nmp->nm_sotype = argp->sotype;
    735 	nmp->nm_soproto = argp->proto;
    736 
    737 	nfs_decode_args(nmp, argp);
    738 
    739 	/*
    740 	 * For Connection based sockets (TCP,...) defer the connect until
    741 	 * the first request, in case the server is not responding.
    742 	 */
    743 	if (nmp->nm_sotype == SOCK_DGRAM &&
    744 		(error = nfs_connect(nmp, (struct nfsreq *)0)))
    745 		goto bad;
    746 
    747 	/*
    748 	 * This is silly, but it has to be set so that vinifod() works.
    749 	 * We do not want to do an nfs_statfs() here since we can get
    750 	 * stuck on a dead server and we are holding a lock on the mount
    751 	 * point.
    752 	 */
    753 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
    754 	/*
    755 	 * A reference count is needed on the nfsnode representing the
    756 	 * remote root.  If this object is not persistent, then backward
    757 	 * traversals of the mount point (i.e. "..") will not work if
    758 	 * the nfsnode gets flushed out of the cache. Ufs does not have
    759 	 * this problem, because one can identify root inodes by their
    760 	 * number == ROOTINO (2).
    761 	 */
    762 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
    763 	if (error)
    764 		goto bad;
    765 	*vpp = NFSTOV(np);
    766 	VOP_GETATTR(*vpp, &attrs, p->p_ucred, p);
    767 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
    768 		cr = crget();
    769 		cr->cr_uid = attrs.va_uid;
    770 		cr->cr_gid = attrs.va_gid;
    771 		cr->cr_ngroups = 0;
    772 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
    773 		crfree(cr);
    774 	}
    775 
    776 	return (0);
    777 bad:
    778 	nfs_disconnect(nmp);
    779 	free((caddr_t)nmp, M_NFSMNT);
    780 	m_freem(nam);
    781 	return (error);
    782 }
    783 
    784 /*
    785  * unmount system call
    786  */
    787 int
    788 nfs_unmount(mp, mntflags, p)
    789 	struct mount *mp;
    790 	int mntflags;
    791 	struct proc *p;
    792 {
    793 	register struct nfsmount *nmp;
    794 	struct nfsnode *np;
    795 	struct vnode *vp;
    796 	int error, flags = 0;
    797 
    798 	if (mntflags & MNT_FORCE)
    799 		flags |= FORCECLOSE;
    800 	nmp = VFSTONFS(mp);
    801 	/*
    802 	 * Goes something like this..
    803 	 * - Check for activity on the root vnode (other than ourselves).
    804 	 * - Call vflush() to clear out vnodes for this file system,
    805 	 *   except for the root vnode.
    806 	 * - Decrement reference on the vnode representing remote root.
    807 	 * - Close the socket
    808 	 * - Free up the data structures
    809 	 */
    810 	/*
    811 	 * We need to decrement the ref. count on the nfsnode representing
    812 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
    813 	 * has done vput on this vnode, otherwise we would get deadlock!
    814 	 */
    815 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
    816 	if (error)
    817 		return(error);
    818 	vp = NFSTOV(np);
    819 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
    820 		vput(vp);
    821 		return (EBUSY);
    822 	}
    823 
    824 	/*
    825 	 * Must handshake with nqnfs_clientd() if it is active.
    826 	 */
    827 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
    828 	while (nmp->nm_inprog != NULLVP)
    829 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
    830 	error = vflush(mp, vp, flags);
    831 	if (error) {
    832 		vput(vp);
    833 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
    834 		return (error);
    835 	}
    836 
    837 	/*
    838 	 * We are now committed to the unmount.
    839 	 * For NQNFS, let the server daemon free the nfsmount structure.
    840 	 */
    841 	if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB))
    842 		nmp->nm_iflag |= NFSMNT_DISMNT;
    843 
    844 	/*
    845 	 * There are two reference counts to get rid of here.
    846 	 */
    847 	vrele(vp);
    848 	vrele(vp);
    849 	vgone(vp);
    850 	nfs_disconnect(nmp);
    851 	m_freem(nmp->nm_nam);
    852 
    853 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
    854 		free((caddr_t)nmp, M_NFSMNT);
    855 	return (0);
    856 }
    857 
    858 /*
    859  * Return root of a filesystem
    860  */
    861 int
    862 nfs_root(mp, vpp)
    863 	struct mount *mp;
    864 	struct vnode **vpp;
    865 {
    866 	register struct vnode *vp;
    867 	struct nfsmount *nmp;
    868 	struct nfsnode *np;
    869 	int error;
    870 
    871 	nmp = VFSTONFS(mp);
    872 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
    873 	if (error)
    874 		return (error);
    875 	vp = NFSTOV(np);
    876 	if (vp->v_type == VNON)
    877 		vp->v_type = VDIR;
    878 	vp->v_flag = VROOT;
    879 	*vpp = vp;
    880 	return (0);
    881 }
    882 
    883 extern int syncprt;
    884 
    885 /*
    886  * Flush out the buffer cache
    887  */
    888 /* ARGSUSED */
    889 int
    890 nfs_sync(mp, waitfor, cred, p)
    891 	struct mount *mp;
    892 	int waitfor;
    893 	struct ucred *cred;
    894 	struct proc *p;
    895 {
    896 	register struct vnode *vp;
    897 	int error, allerror = 0;
    898 
    899 	/*
    900 	 * Force stale buffer cache information to be flushed.
    901 	 */
    902 loop:
    903 	for (vp = mp->mnt_vnodelist.lh_first;
    904 	     vp != NULL;
    905 	     vp = vp->v_mntvnodes.le_next) {
    906 		/*
    907 		 * If the vnode that we are about to sync is no longer
    908 		 * associated with this mount point, start over.
    909 		 */
    910 		if (vp->v_mount != mp)
    911 			goto loop;
    912 		if (VOP_ISLOCKED(vp) || vp->v_dirtyblkhd.lh_first == NULL)
    913 			continue;
    914 		if (vget(vp, LK_EXCLUSIVE))
    915 			goto loop;
    916 		error = VOP_FSYNC(vp, cred,
    917 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, p);
    918 		if (error)
    919 			allerror = error;
    920 		vput(vp);
    921 	}
    922 	return (allerror);
    923 }
    924 
    925 /*
    926  * NFS flat namespace lookup.
    927  * Currently unsupported.
    928  */
    929 /* ARGSUSED */
    930 int
    931 nfs_vget(mp, ino, vpp)
    932 	struct mount *mp;
    933 	ino_t ino;
    934 	struct vnode **vpp;
    935 {
    936 
    937 	return (EOPNOTSUPP);
    938 }
    939 
    940 /*
    941  * Do that sysctl thang...
    942  */
    943 int
    944 nfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
    945 	int *name;
    946 	u_int namelen;
    947 	void *oldp;
    948 	size_t *oldlenp;
    949 	void *newp;
    950 	size_t newlen;
    951 	struct proc *p;
    952 {
    953 	int rv;
    954 
    955 	/*
    956 	 * All names at this level are terminal.
    957 	 */
    958 	if(namelen > 1)
    959 		return ENOTDIR;	/* overloaded */
    960 
    961 	switch(name[0]) {
    962 	case NFS_NFSSTATS:
    963 		if(!oldp) {
    964 			*oldlenp = sizeof nfsstats;
    965 			return 0;
    966 		}
    967 
    968 		if(*oldlenp < sizeof nfsstats) {
    969 			*oldlenp = sizeof nfsstats;
    970 			return ENOMEM;
    971 		}
    972 
    973 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
    974 		if(rv) return rv;
    975 
    976 		if(newp && newlen != sizeof nfsstats)
    977 			return EINVAL;
    978 
    979 		if(newp) {
    980 			return copyin(newp, &nfsstats, sizeof nfsstats);
    981 		}
    982 		return 0;
    983 
    984 	default:
    985 		return EOPNOTSUPP;
    986 	}
    987 }
    988 
    989 
    990 /*
    991  * At this point, this should never happen
    992  */
    993 /* ARGSUSED */
    994 int
    995 nfs_fhtovp(mp, fhp, nam, vpp, exflagsp, credanonp)
    996 	register struct mount *mp;
    997 	struct fid *fhp;
    998 	struct mbuf *nam;
    999 	struct vnode **vpp;
   1000 	int *exflagsp;
   1001 	struct ucred **credanonp;
   1002 {
   1003 
   1004 	return (EINVAL);
   1005 }
   1006 
   1007 /*
   1008  * Vnode pointer to File handle, should never happen either
   1009  */
   1010 /* ARGSUSED */
   1011 int
   1012 nfs_vptofh(vp, fhp)
   1013 	struct vnode *vp;
   1014 	struct fid *fhp;
   1015 {
   1016 
   1017 	return (EINVAL);
   1018 }
   1019 
   1020 /*
   1021  * Vfs start routine, a no-op.
   1022  */
   1023 /* ARGSUSED */
   1024 int
   1025 nfs_start(mp, flags, p)
   1026 	struct mount *mp;
   1027 	int flags;
   1028 	struct proc *p;
   1029 {
   1030 
   1031 	return (0);
   1032 }
   1033 
   1034 /*
   1035  * Do operations associated with quotas, not supported
   1036  */
   1037 /* ARGSUSED */
   1038 int
   1039 nfs_quotactl(mp, cmd, uid, arg, p)
   1040 	struct mount *mp;
   1041 	int cmd;
   1042 	uid_t uid;
   1043 	caddr_t arg;
   1044 	struct proc *p;
   1045 {
   1046 
   1047 	return (EOPNOTSUPP);
   1048 }
   1049