Home | History | Annotate | Line # | Download | only in nfs
nfs_vfsops.c revision 1.63
      1 /*	$NetBSD: nfs_vfsops.c,v 1.63 1997/08/29 16:12:51 gwr 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/param.h>
     42 #include <sys/conf.h>
     43 #include <sys/ioctl.h>
     44 #include <sys/signal.h>
     45 #include <sys/proc.h>
     46 #include <sys/namei.h>
     47 #include <sys/device.h>
     48 #include <sys/vnode.h>
     49 #include <sys/kernel.h>
     50 #include <sys/mount.h>
     51 #include <sys/buf.h>
     52 #include <sys/mbuf.h>
     53 #include <sys/socket.h>
     54 #include <sys/socketvar.h>
     55 #include <sys/systm.h>
     56 
     57 #include <net/if.h>
     58 #include <net/route.h>
     59 #include <netinet/in.h>
     60 
     61 #include <nfs/rpcv2.h>
     62 #include <nfs/nfsproto.h>
     63 #include <nfs/nfsnode.h>
     64 #include <nfs/nfs.h>
     65 #include <nfs/nfsmount.h>
     66 #include <nfs/xdr_subs.h>
     67 #include <nfs/nfsm_subs.h>
     68 #include <nfs/nfsdiskless.h>
     69 #include <nfs/nqnfs.h>
     70 #include <nfs/nfs_var.h>
     71 
     72 extern struct nfsstats nfsstats;
     73 extern int nfs_ticks;
     74 
     75 #ifdef notyet
     76 static int nfs_sysctl(int *, u_int, void *, size_t *, void *, size_t,
     77 		      struct proc *);
     78 #endif
     79 
     80 /*
     81  * nfs vfs operations.
     82  */
     83 struct vfsops nfs_vfsops = {
     84 	MOUNT_NFS,
     85 	nfs_mount,
     86 	nfs_start,
     87 	nfs_unmount,
     88 	nfs_root,
     89 	nfs_quotactl,
     90 	nfs_statfs,
     91 	nfs_sync,
     92 	nfs_vget,
     93 	nfs_fhtovp,
     94 	nfs_vptofh,
     95 	nfs_vfs_init,
     96 	nfs_mountroot,
     97 #ifdef notyet
     98 	nfs_sysctl
     99 #endif
    100 };
    101 
    102 extern u_int32_t nfs_procids[NFS_NPROCS];
    103 extern u_int32_t nfs_prog, nfs_vers;
    104 
    105 static int
    106 nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
    107     struct mount **, struct vnode **));
    108 
    109 #define TRUE	1
    110 #define	FALSE	0
    111 
    112 /*
    113  * nfs statfs call
    114  */
    115 int
    116 nfs_statfs(mp, sbp, p)
    117 	struct mount *mp;
    118 	register struct statfs *sbp;
    119 	struct proc *p;
    120 {
    121 	register struct vnode *vp;
    122 	register struct nfs_statfs *sfp;
    123 	register caddr_t cp;
    124 	register u_int32_t *tl;
    125 	register int32_t t1, t2;
    126 	caddr_t bpos, dpos, cp2;
    127 	struct nfsmount *nmp = VFSTONFS(mp);
    128 	int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr;
    129 	struct mbuf *mreq, *mrep = NULL, *md, *mb, *mb2;
    130 	struct ucred *cred;
    131 	struct nfsnode *np;
    132 	u_quad_t tquad;
    133 
    134 #ifndef nolint
    135 	sfp = (struct nfs_statfs *)0;
    136 #endif
    137 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
    138 	if (error)
    139 		return (error);
    140 	vp = NFSTOV(np);
    141 	cred = crget();
    142 	cred->cr_ngroups = 0;
    143 	if (v3 && (nmp->nm_flag & NFSMNT_GOTFSINFO) == 0)
    144 		(void)nfs_fsinfo(nmp, vp, cred, p);
    145 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
    146 	nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
    147 	nfsm_fhtom(vp, v3);
    148 	nfsm_request(vp, NFSPROC_FSSTAT, p, cred);
    149 	if (v3)
    150 		nfsm_postop_attr(vp, retattr);
    151 	if (error) {
    152 		if (mrep != NULL)
    153 			m_free(mrep);
    154 		goto nfsmout;
    155 	}
    156 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
    157 #ifdef COMPAT_09
    158 	sbp->f_type = 2;
    159 #else
    160 	sbp->f_type = 0;
    161 #endif
    162 	sbp->f_flags = nmp->nm_flag;
    163 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
    164 	if (v3) {
    165 		sbp->f_bsize = NFS_FABLKSIZE;
    166 		fxdr_hyper(&sfp->sf_tbytes, &tquad);
    167 		sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
    168 		fxdr_hyper(&sfp->sf_fbytes, &tquad);
    169 		sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
    170 		fxdr_hyper(&sfp->sf_abytes, &tquad);
    171 		sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE));
    172 		sbp->f_files = (fxdr_unsigned(int32_t,
    173 		    sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff);
    174 		sbp->f_ffree = (fxdr_unsigned(int32_t,
    175 		    sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff);
    176 	} else {
    177 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
    178 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
    179 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
    180 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
    181 		sbp->f_files = 0;
    182 		sbp->f_ffree = 0;
    183 	}
    184 	if (sbp != &mp->mnt_stat) {
    185 		bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
    186 		bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
    187 	}
    188 	strncpy(&sbp->f_fstypename[0], mp->mnt_op->vfs_name, MFSNAMELEN);
    189 	nfsm_reqdone;
    190 	vrele(vp);
    191 	crfree(cred);
    192 	return (error);
    193 }
    194 
    195 /*
    196  * nfs version 3 fsinfo rpc call
    197  */
    198 int
    199 nfs_fsinfo(nmp, vp, cred, p)
    200 	register struct nfsmount *nmp;
    201 	register struct vnode *vp;
    202 	struct ucred *cred;
    203 	struct proc *p;
    204 {
    205 	register struct nfsv3_fsinfo *fsp;
    206 	register caddr_t cp;
    207 	register int32_t t1, t2;
    208 	register u_int32_t *tl, pref, max;
    209 	caddr_t bpos, dpos, cp2;
    210 	int error = 0, retattr;
    211 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
    212 	u_int64_t maxfsize;
    213 
    214 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
    215 	nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
    216 	nfsm_fhtom(vp, 1);
    217 	nfsm_request(vp, NFSPROC_FSINFO, p, cred);
    218 	nfsm_postop_attr(vp, retattr);
    219 	if (!error) {
    220 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
    221 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
    222 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
    223 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
    224 				~(NFS_FABLKSIZE - 1);
    225 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
    226 		if (max < nmp->nm_wsize && max > 0) {
    227 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
    228 			if (nmp->nm_wsize == 0)
    229 				nmp->nm_wsize = max;
    230 		}
    231 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
    232 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
    233 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
    234 				~(NFS_FABLKSIZE - 1);
    235 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
    236 		if (max < nmp->nm_rsize && max > 0) {
    237 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
    238 			if (nmp->nm_rsize == 0)
    239 				nmp->nm_rsize = max;
    240 		}
    241 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
    242 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ)
    243 			nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) &
    244 				~(NFS_DIRBLKSIZ - 1);
    245 		if (max < nmp->nm_readdirsize && max > 0) {
    246 			nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1);
    247 			if (nmp->nm_readdirsize == 0)
    248 				nmp->nm_readdirsize = max;
    249 		}
    250 		/* XXX */
    251 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
    252 		fxdr_hyper(&fsp->fs_maxfilesize, &maxfsize);
    253 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
    254 			nmp->nm_maxfilesize = maxfsize;
    255 		nmp->nm_flag |= NFSMNT_GOTFSINFO;
    256 	}
    257 	nfsm_reqdone;
    258 	return (error);
    259 }
    260 
    261 /*
    262  * Mount a remote root fs via. NFS.  It goes like this:
    263  * - Call nfs_boot_init() to fill in the nfs_diskless struct
    264  * - build the rootfs mount point and call mountnfs() to do the rest.
    265  */
    266 int
    267 nfs_mountroot()
    268 {
    269 	struct nfs_diskless *nd;
    270 	struct vattr attr;
    271 	struct mount *mp;
    272 	struct vnode *vp;
    273 	struct proc *procp;
    274 	long n;
    275 	int error;
    276 
    277 	procp = curproc; /* XXX */
    278 
    279 	if (root_device->dv_class != DV_IFNET)
    280 		return (ENODEV);
    281 
    282 	/*
    283 	 * XXX time must be non-zero when we init the interface or else
    284 	 * the arp code will wedge.  [Fixed now in if_ether.c]
    285 	 * However, the NFS attribute cache gives false "hits" when
    286 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
    287 	 */
    288 	if (time.tv_sec < NFS_MAXATTRTIMO)
    289 		time.tv_sec = NFS_MAXATTRTIMO;
    290 
    291 	/*
    292 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
    293 	 * Side effect:  Finds and configures a network interface.
    294 	 */
    295 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
    296 	bzero((caddr_t)nd, sizeof(*nd));
    297 	error = nfs_boot_init(nd, procp);
    298 	if (error)
    299 		goto out;
    300 
    301 	/*
    302 	 * Create the root mount point.
    303 	 */
    304 	error = nfs_boot_getfh(&nd->nd_root);
    305 	if (error)
    306 		goto out;
    307 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp);
    308 	if (error)
    309 		goto out;
    310 	printf("root on %s\n", nd->nd_root.ndm_host);
    311 
    312 	/*
    313 	 * Link it into the mount list.
    314 	 */
    315 #ifdef Lite2_integrated
    316 	simple_lock(&mountlist_slock);
    317 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
    318 	simple_unlock(&mountlist_slock);
    319 	rootvp = vp;
    320 	vfs_unbusy(mp, procp);
    321 #else
    322 	if (vfs_lock(mp))
    323 		panic("nfs_mountroot: vfs_lock");
    324 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
    325 	mp->mnt_vnodecovered = NULLVP;
    326 	vfs_unlock(mp);
    327 	rootvp = vp;
    328 #endif
    329 
    330 	/* Get root attributes (for the time). */
    331 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
    332 	if (error)
    333 		panic("nfs_mountroot: getattr for root");
    334 	n = attr.va_mtime.tv_sec;
    335 #ifdef	DEBUG
    336 	printf("root time: 0x%lx\n", n);
    337 #endif
    338 	inittodr(n);
    339 
    340 #if 0
    341 	/*
    342 	 * XXX splnet, so networks will receive...
    343 	 * XXX Which port needed this hack?
    344 	 * XXX Should already be at spl0.
    345 	 */
    346 	splnet();
    347 #endif
    348 
    349 #if 0	/* swap now comes in from swapctl(2) */
    350 #ifdef notyet
    351 	/* Set up swap credentials. */
    352 	proc0.p_ucred->cr_uid = 0;
    353 	proc0.p_ucred->cr_gid = 0;
    354 	proc0.p_ucred->cr_ngroups = 0;
    355 #endif
    356 
    357 	/*
    358 	 * "Mount" the swap device.
    359 	 *
    360 	 * On a "dataless" configuration (swap on disk) we will have:
    361 	 *	(swdevt[0].sw_dev != NODEV) identifying the swap device.
    362 	 */
    363 	if (bdevvp(swapdev, &swapdev_vp))
    364 		panic("nfs_mountroot: can't setup swap vp");
    365 	if (swdevt[0].sw_dev != NODEV) {
    366 		printf("swap on device 0x%x\n", swdevt[0].sw_dev);
    367 		error = 0;
    368 		goto out;
    369 	}
    370 
    371 	/*
    372 	 * If swapping to an nfs node:  (swdevt[0].sw_dev == NODEV)
    373 	 * Create a fake mount point just for the swap vnode so that the
    374 	 * swap file can be on a different server from the rootfs.
    375 	 */
    376 	error = nfs_boot_getfh(&nd->nd_swap);
    377 	if (error) {
    378 		printf("nfs_boot: warning: getfh(swap), error=%d\n", error);
    379 		error = 0;
    380 		goto out;
    381 	}
    382 	error = nfs_mount_diskless(&nd->nd_swap, "/swap", &mp, &vp);
    383 	if (error) {
    384 		printf("nfs_boot: warning: mount(swap), error=%d\n", error);
    385 		error = 0;
    386 		goto out;
    387 	}
    388 #ifdef Lite2_integrated
    389 	vfs_unbusy(mp, procp);
    390 #endif
    391 	printf("swap on %s\n", nd->nd_swap.ndm_host);
    392 
    393 	/*
    394 	 * Since the swap file is not the root dir of a file system,
    395 	 * hack it to a regular file.
    396 	 */
    397 	vp->v_type = VREG;
    398 	vp->v_flag = 0;
    399 	swdevt[0].sw_vp = vp;
    400 
    401 	/*
    402 	 * Find out how large the swap file is.
    403 	 */
    404 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
    405 	if (error)
    406 		panic("nfs_mountroot: getattr for swap");
    407 	n = (long) (attr.va_size >> DEV_BSHIFT);
    408 #ifdef	DEBUG
    409 	printf("swap size: 0x%lx (blocks)\n", n);
    410 #endif
    411 	swdevt[0].sw_nblks = n;
    412 #endif
    413 
    414 out:
    415 	free(nd, M_NFSMNT);
    416 	return (error);
    417 }
    418 
    419 /*
    420  * Internal version of mount system call for diskless setup.
    421  * Separate function because we used to call it twice.
    422  * (once for root and once for swap)
    423  */
    424 static int
    425 nfs_mount_diskless(ndmntp, mntname, mpp, vpp)
    426 	struct nfs_dlmount *ndmntp;
    427 	const char *mntname;	/* mount point name */
    428 	struct mount **mpp;
    429 	struct vnode **vpp;
    430 {
    431 	struct mount *mp;
    432 	struct mbuf *m;
    433 	int error;
    434 
    435 #ifdef Lite2_integrated
    436 	vfs_rootmountalloc("nfs", mntname, &mp);
    437 #else
    438 	/* Create the mount point. */
    439 	mp = (struct mount *)malloc((u_long)sizeof(struct mount),
    440 	    M_MOUNT, M_WAITOK);
    441 	bzero((char *)mp, (u_long)sizeof(struct mount));
    442 #endif
    443 	mp->mnt_op = &nfs_vfsops;
    444 	/* mp->mnt_flag = 0 */
    445 
    446 	/* Get mbuf for server sockaddr. */
    447 	m = m_get(M_WAIT, MT_SONAME);
    448 	if (m == NULL)
    449 		panic("nfs_mountroot: mget soname for %s", mntname);
    450 	bcopy((caddr_t)ndmntp->ndm_args.addr, mtod(m, caddr_t),
    451 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
    452 
    453 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
    454 			 ndmntp->ndm_args.hostname, vpp);
    455 	if (error) {
    456 		printf("nfs_mountroot: mount %s failed: %d\n",
    457 		       mntname, error);
    458 		free(mp, M_MOUNT);
    459 	} else
    460 		*mpp = mp;
    461 
    462 	return (error);
    463 }
    464 
    465 void
    466 nfs_decode_args(nmp, argp)
    467 	struct nfsmount *nmp;
    468 	struct nfs_args *argp;
    469 {
    470 	int s;
    471 	int adjsock;
    472 	int maxio;
    473 
    474 	s = splsoftnet();
    475 
    476 	/*
    477 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
    478 	 * no sense in that context.
    479 	 */
    480 	if (argp->sotype == SOCK_STREAM)
    481 		argp->flags &= ~NFSMNT_NOCONN;
    482 
    483 	/* Re-bind if rsrvd port requested and wasn't on one */
    484 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
    485 		  && (argp->flags & NFSMNT_RESVPORT);
    486 	/* Also re-bind if we're switching to/from a connected UDP socket */
    487 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
    488 		    (argp->flags & NFSMNT_NOCONN));
    489 
    490 	/* Update flags atomically.  Don't change the lock bits. */
    491 	nmp->nm_flag =
    492 	    (argp->flags & ~NFSMNT_INTERNAL) | (nmp->nm_flag & NFSMNT_INTERNAL);
    493 	splx(s);
    494 
    495 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
    496 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
    497 		if (nmp->nm_timeo < NFS_MINTIMEO)
    498 			nmp->nm_timeo = NFS_MINTIMEO;
    499 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
    500 			nmp->nm_timeo = NFS_MAXTIMEO;
    501 	}
    502 
    503 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
    504 		nmp->nm_retry = argp->retrans;
    505 		if (nmp->nm_retry > NFS_MAXREXMIT)
    506 			nmp->nm_retry = NFS_MAXREXMIT;
    507 	}
    508 
    509 	if (argp->flags & NFSMNT_NFSV3) {
    510 		if (argp->sotype == SOCK_DGRAM)
    511 			maxio = NFS_MAXDGRAMDATA;
    512 		else
    513 			maxio = NFS_MAXDATA;
    514 	} else
    515 		maxio = NFS_V2MAXDATA;
    516 
    517 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
    518 		int osize = nmp->nm_wsize;
    519 		nmp->nm_wsize = argp->wsize;
    520 		/* Round down to multiple of blocksize */
    521 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
    522 		if (nmp->nm_wsize <= 0)
    523 			nmp->nm_wsize = NFS_FABLKSIZE;
    524 		adjsock |= (nmp->nm_wsize != osize);
    525 	}
    526 	if (nmp->nm_wsize > maxio)
    527 		nmp->nm_wsize = maxio;
    528 	if (nmp->nm_wsize > MAXBSIZE)
    529 		nmp->nm_wsize = MAXBSIZE;
    530 
    531 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
    532 		int osize = nmp->nm_rsize;
    533 		nmp->nm_rsize = argp->rsize;
    534 		/* Round down to multiple of blocksize */
    535 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
    536 		if (nmp->nm_rsize <= 0)
    537 			nmp->nm_rsize = NFS_FABLKSIZE;
    538 		adjsock |= (nmp->nm_rsize != osize);
    539 	}
    540 	if (nmp->nm_rsize > maxio)
    541 		nmp->nm_rsize = maxio;
    542 	if (nmp->nm_rsize > MAXBSIZE)
    543 		nmp->nm_rsize = MAXBSIZE;
    544 
    545 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
    546 		nmp->nm_readdirsize = argp->readdirsize;
    547 		/* Round down to multiple of blocksize */
    548 		nmp->nm_readdirsize &= ~(NFS_DIRBLKSIZ - 1);
    549 		if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
    550 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
    551 	} else if (argp->flags & NFSMNT_RSIZE)
    552 		nmp->nm_readdirsize = nmp->nm_rsize;
    553 
    554 	if (nmp->nm_readdirsize > maxio)
    555 		nmp->nm_readdirsize = maxio;
    556 
    557 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
    558 		argp->maxgrouplist <= NFS_MAXGRPS)
    559 		nmp->nm_numgrps = argp->maxgrouplist;
    560 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
    561 		argp->readahead <= NFS_MAXRAHEAD)
    562 		nmp->nm_readahead = argp->readahead;
    563 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
    564 		argp->leaseterm <= NQ_MAXLEASE)
    565 		nmp->nm_leaseterm = argp->leaseterm;
    566 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
    567 		argp->deadthresh <= NQ_NEVERDEAD)
    568 		nmp->nm_deadthresh = argp->deadthresh;
    569 
    570 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
    571 		    (nmp->nm_soproto != argp->proto));
    572 	nmp->nm_sotype = argp->sotype;
    573 	nmp->nm_soproto = argp->proto;
    574 
    575 	if (nmp->nm_so && adjsock) {
    576 		nfs_disconnect(nmp);
    577 		if (nmp->nm_sotype == SOCK_DGRAM)
    578 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
    579 				printf("nfs_args: retrying connect\n");
    580 				(void) tsleep((caddr_t)&lbolt,
    581 					      PSOCK, "nfscon", 0);
    582 			}
    583 	}
    584 }
    585 
    586 /*
    587  * VFS Operations.
    588  *
    589  * mount system call
    590  * It seems a bit dumb to copyinstr() the host and path here and then
    591  * bcopy() them in mountnfs(), but I wanted to detect errors before
    592  * doing the sockargs() call because sockargs() allocates an mbuf and
    593  * an error after that means that I have to release the mbuf.
    594  */
    595 /* ARGSUSED */
    596 int
    597 nfs_mount(mp, path, data, ndp, p)
    598 	struct mount *mp;
    599 	const char *path;
    600 	void *data;
    601 	struct nameidata *ndp;
    602 	struct proc *p;
    603 {
    604 	int error;
    605 	struct nfs_args args;
    606 	struct mbuf *nam;
    607 	struct vnode *vp;
    608 	char pth[MNAMELEN], hst[MNAMELEN];
    609 	size_t len;
    610 	u_char nfh[NFSX_V3FHMAX];
    611 
    612 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
    613 	if (error)
    614 		return (error);
    615 	if (args.version != NFS_ARGSVERSION)
    616 		return (EPROGMISMATCH);
    617 	if (mp->mnt_flag & MNT_UPDATE) {
    618 		register struct nfsmount *nmp = VFSTONFS(mp);
    619 
    620 		if (nmp == NULL)
    621 			return (EIO);
    622 		/*
    623 		 * When doing an update, we can't change from or to
    624 		 * v3 and/or nqnfs.
    625 		 */
    626 		args.flags = (args.flags & ~(NFSMNT_NFSV3|NFSMNT_NQNFS)) |
    627 		    (nmp->nm_flag & (NFSMNT_NFSV3|NFSMNT_NQNFS));
    628 		nfs_decode_args(nmp, &args);
    629 		return (0);
    630 	}
    631 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
    632 	if (error)
    633 		return (error);
    634 	error = copyinstr(path, pth, MNAMELEN-1, &len);
    635 	if (error)
    636 		return (error);
    637 	bzero(&pth[len], MNAMELEN - len);
    638 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
    639 	if (error)
    640 		return (error);
    641 	bzero(&hst[len], MNAMELEN - len);
    642 	/* sockargs() call must be after above copyin() calls */
    643 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
    644 	if (error)
    645 		return (error);
    646 	args.fh = nfh;
    647 	error = mountnfs(&args, mp, nam, pth, hst, &vp);
    648 	return (error);
    649 }
    650 
    651 /*
    652  * Common code for mount and mountroot
    653  */
    654 int
    655 mountnfs(argp, mp, nam, pth, hst, vpp)
    656 	register struct nfs_args *argp;
    657 	register struct mount *mp;
    658 	struct mbuf *nam;
    659 	const char *pth, *hst;
    660 	struct vnode **vpp;
    661 {
    662 	register struct nfsmount *nmp;
    663 	struct nfsnode *np;
    664 	int error;
    665 	struct vattr attrs;
    666 
    667 	if (mp->mnt_flag & MNT_UPDATE) {
    668 		nmp = VFSTONFS(mp);
    669 		/* update paths, file handles, etc, here	XXX */
    670 		m_freem(nam);
    671 		return (0);
    672 	} else {
    673 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
    674 		    M_NFSMNT, M_WAITOK);
    675 		bzero((caddr_t)nmp, sizeof (struct nfsmount));
    676 		mp->mnt_data = (qaddr_t)nmp;
    677 		TAILQ_INIT(&nmp->nm_uidlruhead);
    678 		TAILQ_INIT(&nmp->nm_bufq);
    679 	}
    680 #ifdef Lite2_integrated
    681 	vfs_getnewfsid(mp, makefstype(MOUNT_NFS));
    682 #else
    683 	getnewfsid(mp, makefstype(MOUNT_NFS));
    684 #endif
    685 	nmp->nm_mountp = mp;
    686 
    687 	if (argp->flags & NFSMNT_NQNFS)
    688 		/*
    689 		 * We have to set mnt_maxsymlink to a non-zero value so
    690 		 * that COMPAT_43 routines will know that we are setting
    691 		 * the d_type field in directories (and can zero it for
    692 		 * unsuspecting binaries).
    693 		 */
    694 		mp->mnt_maxsymlinklen = 1;
    695 
    696 	if ((argp->flags & NFSMNT_NFSV3) == 0)
    697 		/*
    698 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
    699 		 * will fill this in.
    700 		 */
    701 		nmp->nm_maxfilesize = 0xffffffffLL;
    702 
    703 	nmp->nm_timeo = NFS_TIMEO;
    704 	nmp->nm_retry = NFS_RETRANS;
    705 	nmp->nm_wsize = NFS_WSIZE;
    706 	nmp->nm_rsize = NFS_RSIZE;
    707 	nmp->nm_readdirsize = NFS_READDIRSIZE;
    708 	nmp->nm_numgrps = NFS_MAXGRPS;
    709 	nmp->nm_readahead = NFS_DEFRAHEAD;
    710 	nmp->nm_leaseterm = NQ_DEFLEASE;
    711 	nmp->nm_deadthresh = NQ_DEADTHRESH;
    712 	CIRCLEQ_INIT(&nmp->nm_timerhead);
    713 	nmp->nm_inprog = NULLVP;
    714 	nmp->nm_fhsize = argp->fhsize;
    715 	bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize);
    716 #ifdef COMPAT_09
    717 	mp->mnt_stat.f_type = 2;
    718 #else
    719 	mp->mnt_stat.f_type = 0;
    720 #endif
    721 	strncpy(&mp->mnt_stat.f_fstypename[0], mp->mnt_op->vfs_name, MFSNAMELEN);
    722 	bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN);
    723 	bcopy(pth, mp->mnt_stat.f_mntonname, MNAMELEN);
    724 	nmp->nm_nam = nam;
    725 
    726 	/* Set up the sockets and per-host congestion */
    727 	nmp->nm_sotype = argp->sotype;
    728 	nmp->nm_soproto = argp->proto;
    729 
    730 	nfs_decode_args(nmp, argp);
    731 
    732 	/*
    733 	 * For Connection based sockets (TCP,...) defer the connect until
    734 	 * the first request, in case the server is not responding.
    735 	 */
    736 	if (nmp->nm_sotype == SOCK_DGRAM &&
    737 		(error = nfs_connect(nmp, (struct nfsreq *)0)))
    738 		goto bad;
    739 
    740 	/*
    741 	 * This is silly, but it has to be set so that vinifod() works.
    742 	 * We do not want to do an nfs_statfs() here since we can get
    743 	 * stuck on a dead server and we are holding a lock on the mount
    744 	 * point.
    745 	 */
    746 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
    747 	/*
    748 	 * A reference count is needed on the nfsnode representing the
    749 	 * remote root.  If this object is not persistent, then backward
    750 	 * traversals of the mount point (i.e. "..") will not work if
    751 	 * the nfsnode gets flushed out of the cache. Ufs does not have
    752 	 * this problem, because one can identify root inodes by their
    753 	 * number == ROOTINO (2).
    754 	 */
    755 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
    756 	if (error)
    757 		goto bad;
    758 	*vpp = NFSTOV(np);
    759 	VOP_GETATTR(*vpp, &attrs, curproc->p_ucred, curproc); /* XXX */
    760 
    761 	return (0);
    762 bad:
    763 	nfs_disconnect(nmp);
    764 	free((caddr_t)nmp, M_NFSMNT);
    765 	m_freem(nam);
    766 	return (error);
    767 }
    768 
    769 /*
    770  * unmount system call
    771  */
    772 int
    773 nfs_unmount(mp, mntflags, p)
    774 	struct mount *mp;
    775 	int mntflags;
    776 	struct proc *p;
    777 {
    778 	register struct nfsmount *nmp;
    779 	struct nfsnode *np;
    780 	struct vnode *vp;
    781 	int error, flags = 0;
    782 
    783 	if (mntflags & MNT_FORCE)
    784 		flags |= FORCECLOSE;
    785 	nmp = VFSTONFS(mp);
    786 	/*
    787 	 * Goes something like this..
    788 	 * - Check for activity on the root vnode (other than ourselves).
    789 	 * - Call vflush() to clear out vnodes for this file system,
    790 	 *   except for the root vnode.
    791 	 * - Decrement reference on the vnode representing remote root.
    792 	 * - Close the socket
    793 	 * - Free up the data structures
    794 	 */
    795 	/*
    796 	 * We need to decrement the ref. count on the nfsnode representing
    797 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
    798 	 * has done vput on this vnode, otherwise we would get deadlock!
    799 	 */
    800 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
    801 	if (error)
    802 		return(error);
    803 	vp = NFSTOV(np);
    804 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
    805 		vput(vp);
    806 		return (EBUSY);
    807 	}
    808 
    809 	/*
    810 	 * Must handshake with nqnfs_clientd() if it is active.
    811 	 */
    812 	nmp->nm_flag |= NFSMNT_DISMINPROG;
    813 	while (nmp->nm_inprog != NULLVP)
    814 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
    815 	error = vflush(mp, vp, flags);
    816 	if (error) {
    817 		vput(vp);
    818 		nmp->nm_flag &= ~NFSMNT_DISMINPROG;
    819 		return (error);
    820 	}
    821 
    822 	/*
    823 	 * We are now committed to the unmount.
    824 	 * For NQNFS, let the server daemon free the nfsmount structure.
    825 	 */
    826 	if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB))
    827 		nmp->nm_flag |= NFSMNT_DISMNT;
    828 
    829 	/*
    830 	 * There are two reference counts to get rid of here.
    831 	 */
    832 	vrele(vp);
    833 	vrele(vp);
    834 	vgone(vp);
    835 	nfs_disconnect(nmp);
    836 	m_freem(nmp->nm_nam);
    837 
    838 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
    839 		free((caddr_t)nmp, M_NFSMNT);
    840 	return (0);
    841 }
    842 
    843 /*
    844  * Return root of a filesystem
    845  */
    846 int
    847 nfs_root(mp, vpp)
    848 	struct mount *mp;
    849 	struct vnode **vpp;
    850 {
    851 	register struct vnode *vp;
    852 	struct nfsmount *nmp;
    853 	struct nfsnode *np;
    854 	int error;
    855 
    856 	nmp = VFSTONFS(mp);
    857 	error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np);
    858 	if (error)
    859 		return (error);
    860 	vp = NFSTOV(np);
    861 	if (vp->v_type == VNON)
    862 		vp->v_type = VDIR;
    863 	vp->v_flag = VROOT;
    864 	*vpp = vp;
    865 	return (0);
    866 }
    867 
    868 extern int syncprt;
    869 
    870 /*
    871  * Flush out the buffer cache
    872  */
    873 /* ARGSUSED */
    874 int
    875 nfs_sync(mp, waitfor, cred, p)
    876 	struct mount *mp;
    877 	int waitfor;
    878 	struct ucred *cred;
    879 	struct proc *p;
    880 {
    881 	register struct vnode *vp;
    882 	int error, allerror = 0;
    883 
    884 	/*
    885 	 * Force stale buffer cache information to be flushed.
    886 	 */
    887 loop:
    888 	for (vp = mp->mnt_vnodelist.lh_first;
    889 	     vp != NULL;
    890 	     vp = vp->v_mntvnodes.le_next) {
    891 		/*
    892 		 * If the vnode that we are about to sync is no longer
    893 		 * associated with this mount point, start over.
    894 		 */
    895 		if (vp->v_mount != mp)
    896 			goto loop;
    897 		if (VOP_ISLOCKED(vp) || vp->v_dirtyblkhd.lh_first == NULL)
    898 			continue;
    899 #ifdef Lite2_integrated
    900 		if (vget(vp, LK_EXCLUSIVE, p))
    901 #else
    902 		if (vget(vp, 1))
    903 #endif
    904 			goto loop;
    905 		error = VOP_FSYNC(vp, cred, waitfor, p);
    906 		if (error)
    907 			allerror = error;
    908 		vput(vp);
    909 	}
    910 	return (allerror);
    911 }
    912 
    913 /*
    914  * NFS flat namespace lookup.
    915  * Currently unsupported.
    916  */
    917 /* ARGSUSED */
    918 int
    919 nfs_vget(mp, ino, vpp)
    920 	struct mount *mp;
    921 	ino_t ino;
    922 	struct vnode **vpp;
    923 {
    924 
    925 	return (EOPNOTSUPP);
    926 }
    927 
    928 #ifdef notyet
    929 /*
    930  * Do that sysctl thang...
    931  */
    932 static int
    933 nfs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
    934 	   size_t newlen, struct proc *p)
    935 {
    936 	int rv;
    937 
    938 	/*
    939 	 * All names at this level are terminal.
    940 	 */
    941 	if(namelen > 1)
    942 		return ENOTDIR;	/* overloaded */
    943 
    944 	switch(name[0]) {
    945 	case NFS_NFSSTATS:
    946 		if(!oldp) {
    947 			*oldlenp = sizeof nfsstats;
    948 			return 0;
    949 		}
    950 
    951 		if(*oldlenp < sizeof nfsstats) {
    952 			*oldlenp = sizeof nfsstats;
    953 			return ENOMEM;
    954 		}
    955 
    956 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
    957 		if(rv) return rv;
    958 
    959 		if(newp && newlen != sizeof nfsstats)
    960 			return EINVAL;
    961 
    962 		if(newp) {
    963 			return copyin(newp, &nfsstats, sizeof nfsstats);
    964 		}
    965 		return 0;
    966 
    967 	default:
    968 		return EOPNOTSUPP;
    969 	}
    970 }
    971 #endif
    972 
    973 
    974 /*
    975  * At this point, this should never happen
    976  */
    977 /* ARGSUSED */
    978 int
    979 nfs_fhtovp(mp, fhp, nam, vpp, exflagsp, credanonp)
    980 	register struct mount *mp;
    981 	struct fid *fhp;
    982 	struct mbuf *nam;
    983 	struct vnode **vpp;
    984 	int *exflagsp;
    985 	struct ucred **credanonp;
    986 {
    987 
    988 	return (EINVAL);
    989 }
    990 
    991 /*
    992  * Vnode pointer to File handle, should never happen either
    993  */
    994 /* ARGSUSED */
    995 int
    996 nfs_vptofh(vp, fhp)
    997 	struct vnode *vp;
    998 	struct fid *fhp;
    999 {
   1000 
   1001 	return (EINVAL);
   1002 }
   1003 
   1004 /*
   1005  * Vfs start routine, a no-op.
   1006  */
   1007 /* ARGSUSED */
   1008 int
   1009 nfs_start(mp, flags, p)
   1010 	struct mount *mp;
   1011 	int flags;
   1012 	struct proc *p;
   1013 {
   1014 
   1015 	return (0);
   1016 }
   1017 
   1018 /*
   1019  * Do operations associated with quotas, not supported
   1020  */
   1021 /* ARGSUSED */
   1022 int
   1023 nfs_quotactl(mp, cmd, uid, arg, p)
   1024 	struct mount *mp;
   1025 	int cmd;
   1026 	uid_t uid;
   1027 	caddr_t arg;
   1028 	struct proc *p;
   1029 {
   1030 
   1031 	return (EOPNOTSUPP);
   1032 }
   1033