Home | History | Annotate | Line # | Download | only in nfs
nfs_vfsops.c revision 1.101.2.8
      1 /*	$NetBSD: nfs_vfsops.c,v 1.101.2.8 2002/06/24 22:12:11 nathanw 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 1.101.2.8 2002/06/24 22:12:11 nathanw 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/lwp.h>
     53 #include <sys/proc.h>
     54 #include <sys/namei.h>
     55 #include <sys/device.h>
     56 #include <sys/vnode.h>
     57 #include <sys/kernel.h>
     58 #include <sys/mount.h>
     59 #include <sys/buf.h>
     60 #include <sys/mbuf.h>
     61 #include <sys/socket.h>
     62 #include <sys/socketvar.h>
     63 #include <sys/sysctl.h>
     64 #include <sys/systm.h>
     65 
     66 #include <net/if.h>
     67 #include <net/route.h>
     68 #include <netinet/in.h>
     69 
     70 #include <nfs/rpcv2.h>
     71 #include <nfs/nfsproto.h>
     72 #include <nfs/nfsnode.h>
     73 #include <nfs/nfs.h>
     74 #include <nfs/nfsmount.h>
     75 #include <nfs/xdr_subs.h>
     76 #include <nfs/nfsm_subs.h>
     77 #include <nfs/nfsdiskless.h>
     78 #include <nfs/nqnfs.h>
     79 #include <nfs/nfs_var.h>
     80 
     81 extern struct nfsstats nfsstats;
     82 extern int nfs_ticks;
     83 
     84 int nfs_sysctl __P((int *, u_int, void *, size_t *, void *, size_t,
     85 		      struct proc *));
     86 
     87 /*
     88  * nfs vfs operations.
     89  */
     90 
     91 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
     92 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
     93 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
     94 
     95 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = {
     96 	&nfsv2_vnodeop_opv_desc,
     97 	&spec_nfsv2nodeop_opv_desc,
     98 	&fifo_nfsv2nodeop_opv_desc,
     99 	NULL,
    100 };
    101 
    102 struct vfsops nfs_vfsops = {
    103 	MOUNT_NFS,
    104 	nfs_mount,
    105 	nfs_start,
    106 	nfs_unmount,
    107 	nfs_root,
    108 	nfs_quotactl,
    109 	nfs_statfs,
    110 	nfs_sync,
    111 	nfs_vget,
    112 	nfs_fhtovp,
    113 	nfs_vptofh,
    114 	nfs_vfs_init,
    115 	nfs_vfs_reinit,
    116 	nfs_vfs_done,
    117 	nfs_sysctl,
    118 	nfs_mountroot,
    119 	nfs_checkexp,
    120 	nfs_vnodeopv_descs,
    121 };
    122 
    123 extern u_int32_t nfs_procids[NFS_NPROCS];
    124 extern u_int32_t nfs_prog, nfs_vers;
    125 
    126 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
    127     struct mount **, struct vnode **, struct proc *));
    128 
    129 #define TRUE	1
    130 #define	FALSE	0
    131 
    132 /*
    133  * nfs statfs call
    134  */
    135 int
    136 nfs_statfs(mp, sbp, p)
    137 	struct mount *mp;
    138 	struct statfs *sbp;
    139 	struct proc *p;
    140 {
    141 	struct vnode *vp;
    142 	struct nfs_statfs *sfp;
    143 	caddr_t cp;
    144 	u_int32_t *tl;
    145 	int32_t t1, t2;
    146 	caddr_t bpos, dpos, cp2;
    147 	struct nfsmount *nmp = VFSTONFS(mp);
    148 	int error = 0, retattr;
    149 #ifdef NFS_V2_ONLY
    150 	const int v3 = 0;
    151 #else
    152 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
    153 #endif
    154 	struct mbuf *mreq, *mrep = NULL, *md, *mb, *mb2;
    155 	struct ucred *cred;
    156 	u_quad_t tquad;
    157 
    158 #ifndef nolint
    159 	sfp = (struct nfs_statfs *)0;
    160 #endif
    161 	vp = nmp->nm_vnode;
    162 	cred = crget();
    163 	cred->cr_ngroups = 0;
    164 #ifndef NFS_V2_ONLY
    165 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
    166 		(void)nfs_fsinfo(nmp, vp, cred, p);
    167 #endif
    168 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
    169 	nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
    170 	nfsm_fhtom(vp, v3);
    171 	nfsm_request(vp, NFSPROC_FSSTAT, p, cred);
    172 	if (v3)
    173 		nfsm_postop_attr(vp, retattr);
    174 	if (error) {
    175 		if (mrep != NULL)
    176 			m_free(mrep);
    177 		goto nfsmout;
    178 	}
    179 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
    180 #ifdef COMPAT_09
    181 	sbp->f_type = 2;
    182 #else
    183 	sbp->f_type = 0;
    184 #endif
    185 	sbp->f_flags = nmp->nm_flag;
    186 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
    187 	if (v3) {
    188 		sbp->f_bsize = NFS_FABLKSIZE;
    189 		tquad = fxdr_hyper(&sfp->sf_tbytes);
    190 		sbp->f_blocks = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
    191 		tquad = fxdr_hyper(&sfp->sf_fbytes);
    192 		sbp->f_bfree = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
    193 		tquad = fxdr_hyper(&sfp->sf_abytes);
    194 		sbp->f_bavail = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
    195 		tquad = fxdr_hyper(&sfp->sf_tfiles);
    196 		sbp->f_files = (long)tquad;
    197 		tquad = fxdr_hyper(&sfp->sf_ffiles);
    198 		sbp->f_ffree = (long)tquad;
    199 	} else {
    200 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
    201 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
    202 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
    203 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
    204 		sbp->f_files = 0;
    205 		sbp->f_ffree = 0;
    206 	}
    207 	if (sbp != &mp->mnt_stat) {
    208 		memcpy(sbp->f_mntonname, mp->mnt_stat.f_mntonname, MNAMELEN);
    209 		memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, MNAMELEN);
    210 	}
    211 	strncpy(&sbp->f_fstypename[0], mp->mnt_op->vfs_name, MFSNAMELEN);
    212 	nfsm_reqdone;
    213 	crfree(cred);
    214 	return (error);
    215 }
    216 
    217 #ifndef NFS_V2_ONLY
    218 /*
    219  * nfs version 3 fsinfo rpc call
    220  */
    221 int
    222 nfs_fsinfo(nmp, vp, cred, p)
    223 	struct nfsmount *nmp;
    224 	struct vnode *vp;
    225 	struct ucred *cred;
    226 	struct proc *p;
    227 {
    228 	struct nfsv3_fsinfo *fsp;
    229 	caddr_t cp;
    230 	int32_t t1, t2;
    231 	u_int32_t *tl, pref, max;
    232 	caddr_t bpos, dpos, cp2;
    233 	int error = 0, retattr;
    234 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
    235 	u_int64_t maxfsize;
    236 
    237 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
    238 	nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
    239 	nfsm_fhtom(vp, 1);
    240 	nfsm_request(vp, NFSPROC_FSINFO, p, cred);
    241 	nfsm_postop_attr(vp, retattr);
    242 	if (!error) {
    243 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
    244 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
    245 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
    246 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
    247 				~(NFS_FABLKSIZE - 1);
    248 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
    249 		if (max < nmp->nm_wsize && max > 0) {
    250 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
    251 			if (nmp->nm_wsize == 0)
    252 				nmp->nm_wsize = max;
    253 		}
    254 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
    255 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
    256 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
    257 				~(NFS_FABLKSIZE - 1);
    258 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
    259 		if (max < nmp->nm_rsize && max > 0) {
    260 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
    261 			if (nmp->nm_rsize == 0)
    262 				nmp->nm_rsize = max;
    263 		}
    264 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
    265 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
    266 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
    267 				~(NFS_DIRFRAGSIZ - 1);
    268 		if (max < nmp->nm_readdirsize && max > 0) {
    269 			nmp->nm_readdirsize = max & ~(NFS_DIRFRAGSIZ - 1);
    270 			if (nmp->nm_readdirsize == 0)
    271 				nmp->nm_readdirsize = max;
    272 		}
    273 		/* XXX */
    274 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
    275 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
    276 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
    277 			nmp->nm_maxfilesize = maxfsize;
    278 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
    279 	}
    280 	nfsm_reqdone;
    281 	return (error);
    282 }
    283 #endif
    284 
    285 /*
    286  * Mount a remote root fs via. NFS.  It goes like this:
    287  * - Call nfs_boot_init() to fill in the nfs_diskless struct
    288  * - build the rootfs mount point and call mountnfs() to do the rest.
    289  */
    290 int
    291 nfs_mountroot()
    292 {
    293 	struct nfs_diskless *nd;
    294 	struct vattr attr;
    295 	struct mount *mp;
    296 	struct vnode *vp;
    297 	struct proc *procp;
    298 	long n;
    299 	int error;
    300 
    301 	procp = curproc; /* XXX */
    302 
    303 	if (root_device->dv_class != DV_IFNET)
    304 		return (ENODEV);
    305 
    306 	/*
    307 	 * XXX time must be non-zero when we init the interface or else
    308 	 * the arp code will wedge.  [Fixed now in if_ether.c]
    309 	 * However, the NFS attribute cache gives false "hits" when
    310 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
    311 	 */
    312 	if (time.tv_sec < NFS_MAXATTRTIMO)
    313 		time.tv_sec = NFS_MAXATTRTIMO;
    314 
    315 	/*
    316 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
    317 	 * Side effect:  Finds and configures a network interface.
    318 	 */
    319 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
    320 	memset((caddr_t)nd, 0, sizeof(*nd));
    321 	error = nfs_boot_init(nd, procp);
    322 	if (error) {
    323 		free(nd, M_NFSMNT);
    324 		return (error);
    325 	}
    326 
    327 	/*
    328 	 * Create the root mount point.
    329 	 */
    330 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp);
    331 	if (error)
    332 		goto out;
    333 	printf("root on %s\n", nd->nd_root.ndm_host);
    334 
    335 	/*
    336 	 * Link it into the mount list.
    337 	 */
    338 	simple_lock(&mountlist_slock);
    339 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
    340 	simple_unlock(&mountlist_slock);
    341 	rootvp = vp;
    342 	mp->mnt_vnodecovered = NULLVP;
    343 	vfs_unbusy(mp);
    344 
    345 	/* Get root attributes (for the time). */
    346 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
    347 	if (error)
    348 		panic("nfs_mountroot: getattr for root");
    349 	n = attr.va_atime.tv_sec;
    350 #ifdef	DEBUG
    351 	printf("root time: 0x%lx\n", n);
    352 #endif
    353 	inittodr(n);
    354 
    355 out:
    356 	if (error)
    357 		nfs_boot_cleanup(nd, procp);
    358 	free(nd, M_NFSMNT);
    359 	return (error);
    360 }
    361 
    362 /*
    363  * Internal version of mount system call for diskless setup.
    364  * Separate function because we used to call it twice.
    365  * (once for root and once for swap)
    366  */
    367 static int
    368 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p)
    369 	struct nfs_dlmount *ndmntp;
    370 	const char *mntname;	/* mount point name */
    371 	struct mount **mpp;
    372 	struct vnode **vpp;
    373 	struct proc *p;
    374 {
    375 	struct mount *mp;
    376 	struct mbuf *m;
    377 	int error;
    378 
    379 	vfs_rootmountalloc(MOUNT_NFS, (char *)mntname, &mp);
    380 
    381 	mp->mnt_op = &nfs_vfsops;
    382 
    383 	/*
    384 	 * Historical practice expects NFS root file systems to
    385 	 * be initially mounted r/w.
    386 	 */
    387 	mp->mnt_flag &= ~MNT_RDONLY;
    388 
    389 	/* Get mbuf for server sockaddr. */
    390 	m = m_get(M_WAIT, MT_SONAME);
    391 	if (m == NULL)
    392 		panic("nfs_mountroot: mget soname for %s", mntname);
    393 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
    394 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
    395 
    396 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
    397 			 ndmntp->ndm_args.hostname, vpp, p);
    398 	if (error) {
    399 		mp->mnt_op->vfs_refcount--;
    400 		vfs_unbusy(mp);
    401 		printf("nfs_mountroot: mount %s failed: %d\n",
    402 		       mntname, error);
    403 		free(mp, M_MOUNT);
    404 	} else
    405 		*mpp = mp;
    406 
    407 	return (error);
    408 }
    409 
    410 void
    411 nfs_decode_args(nmp, argp)
    412 	struct nfsmount *nmp;
    413 	struct nfs_args *argp;
    414 {
    415 	int s;
    416 	int adjsock;
    417 	int maxio;
    418 
    419 	s = splsoftnet();
    420 
    421 	/*
    422 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
    423 	 * no sense in that context.
    424 	 */
    425 	if (argp->sotype == SOCK_STREAM)
    426 		argp->flags &= ~NFSMNT_NOCONN;
    427 
    428 	/*
    429 	 * Cookie translation is not needed for v2, silently ignore it.
    430 	 */
    431 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
    432 	    NFSMNT_XLATECOOKIE)
    433 		argp->flags &= ~NFSMNT_XLATECOOKIE;
    434 
    435 	/* Re-bind if rsrvd port requested and wasn't on one */
    436 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
    437 		  && (argp->flags & NFSMNT_RESVPORT);
    438 	/* Also re-bind if we're switching to/from a connected UDP socket */
    439 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
    440 		    (argp->flags & NFSMNT_NOCONN));
    441 
    442 	/* Update flags. */
    443 	nmp->nm_flag = argp->flags;
    444 	splx(s);
    445 
    446 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
    447 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
    448 		if (nmp->nm_timeo < NFS_MINTIMEO)
    449 			nmp->nm_timeo = NFS_MINTIMEO;
    450 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
    451 			nmp->nm_timeo = NFS_MAXTIMEO;
    452 	}
    453 
    454 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
    455 		nmp->nm_retry = argp->retrans;
    456 		if (nmp->nm_retry > NFS_MAXREXMIT)
    457 			nmp->nm_retry = NFS_MAXREXMIT;
    458 	}
    459 
    460 #ifndef NFS_V2_ONLY
    461 	if (argp->flags & NFSMNT_NFSV3) {
    462 		if (argp->sotype == SOCK_DGRAM)
    463 			maxio = NFS_MAXDGRAMDATA;
    464 		else
    465 			maxio = NFS_MAXDATA;
    466 	} else
    467 #endif
    468 		maxio = NFS_V2MAXDATA;
    469 
    470 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
    471 		int osize = nmp->nm_wsize;
    472 		nmp->nm_wsize = argp->wsize;
    473 		/* Round down to multiple of blocksize */
    474 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
    475 		if (nmp->nm_wsize <= 0)
    476 			nmp->nm_wsize = NFS_FABLKSIZE;
    477 		adjsock |= (nmp->nm_wsize != osize);
    478 	}
    479 	if (nmp->nm_wsize > maxio)
    480 		nmp->nm_wsize = maxio;
    481 	if (nmp->nm_wsize > MAXBSIZE)
    482 		nmp->nm_wsize = MAXBSIZE;
    483 
    484 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
    485 		int osize = nmp->nm_rsize;
    486 		nmp->nm_rsize = argp->rsize;
    487 		/* Round down to multiple of blocksize */
    488 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
    489 		if (nmp->nm_rsize <= 0)
    490 			nmp->nm_rsize = NFS_FABLKSIZE;
    491 		adjsock |= (nmp->nm_rsize != osize);
    492 	}
    493 	if (nmp->nm_rsize > maxio)
    494 		nmp->nm_rsize = maxio;
    495 	if (nmp->nm_rsize > MAXBSIZE)
    496 		nmp->nm_rsize = MAXBSIZE;
    497 
    498 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
    499 		nmp->nm_readdirsize = argp->readdirsize;
    500 		/* Round down to multiple of minimum blocksize */
    501 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
    502 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
    503 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
    504 		/* Bigger than buffer size makes no sense */
    505 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
    506 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
    507 	} else if (argp->flags & NFSMNT_RSIZE)
    508 		nmp->nm_readdirsize = nmp->nm_rsize;
    509 
    510 	if (nmp->nm_readdirsize > maxio)
    511 		nmp->nm_readdirsize = maxio;
    512 
    513 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
    514 		argp->maxgrouplist <= NFS_MAXGRPS)
    515 		nmp->nm_numgrps = argp->maxgrouplist;
    516 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
    517 		argp->readahead <= NFS_MAXRAHEAD)
    518 		nmp->nm_readahead = argp->readahead;
    519 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
    520 		argp->leaseterm <= NQ_MAXLEASE)
    521 		nmp->nm_leaseterm = argp->leaseterm;
    522 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
    523 		argp->deadthresh <= NQ_NEVERDEAD)
    524 		nmp->nm_deadthresh = argp->deadthresh;
    525 
    526 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
    527 		    (nmp->nm_soproto != argp->proto));
    528 	nmp->nm_sotype = argp->sotype;
    529 	nmp->nm_soproto = argp->proto;
    530 
    531 	if (nmp->nm_so && adjsock) {
    532 		nfs_safedisconnect(nmp);
    533 		if (nmp->nm_sotype == SOCK_DGRAM)
    534 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
    535 				printf("nfs_args: retrying connect\n");
    536 				(void) tsleep((caddr_t)&lbolt,
    537 					      PSOCK, "nfscn3", 0);
    538 			}
    539 	}
    540 }
    541 
    542 /*
    543  * VFS Operations.
    544  *
    545  * mount system call
    546  * It seems a bit dumb to copyinstr() the host and path here and then
    547  * memcpy() them in mountnfs(), but I wanted to detect errors before
    548  * doing the sockargs() call because sockargs() allocates an mbuf and
    549  * an error after that means that I have to release the mbuf.
    550  */
    551 /* ARGSUSED */
    552 int
    553 nfs_mount(mp, path, data, ndp, p)
    554 	struct mount *mp;
    555 	const char *path;
    556 	void *data;
    557 	struct nameidata *ndp;
    558 	struct proc *p;
    559 {
    560 	int error;
    561 	struct nfs_args args;
    562 	struct mbuf *nam;
    563 	struct vnode *vp;
    564 	char pth[MNAMELEN], hst[MNAMELEN];
    565 	size_t len;
    566 	u_char nfh[NFSX_V3FHMAX];
    567 
    568 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
    569 	if (error)
    570 		return (error);
    571 	if (args.version != NFS_ARGSVERSION)
    572 		return (EPROGMISMATCH);
    573 #ifdef NFS_V2_ONLY
    574 	if (args.flags & NFSMNT_NQNFS)
    575 		return (EPROGUNAVAIL);
    576 	if (args.flags & NFSMNT_NFSV3)
    577 		return (EPROGMISMATCH);
    578 #endif
    579 	if (mp->mnt_flag & MNT_UPDATE) {
    580 		struct nfsmount *nmp = VFSTONFS(mp);
    581 
    582 		if (nmp == NULL)
    583 			return (EIO);
    584 		/*
    585 		 * When doing an update, we can't change from or to
    586 		 * v3 and/or nqnfs, or change cookie translation
    587 		 */
    588 		args.flags = (args.flags &
    589 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
    590 		    (nmp->nm_flag &
    591 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
    592 		nfs_decode_args(nmp, &args);
    593 		return (0);
    594 	}
    595 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
    596 		return (EINVAL);
    597 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
    598 	if (error)
    599 		return (error);
    600 	error = copyinstr(path, pth, MNAMELEN-1, &len);
    601 	if (error)
    602 		return (error);
    603 	memset(&pth[len], 0, MNAMELEN - len);
    604 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
    605 	if (error)
    606 		return (error);
    607 	memset(&hst[len], 0, MNAMELEN - len);
    608 	/* sockargs() call must be after above copyin() calls */
    609 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
    610 	if (error)
    611 		return (error);
    612 	args.fh = nfh;
    613 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
    614 	return (error);
    615 }
    616 
    617 /*
    618  * Common code for mount and mountroot
    619  */
    620 int
    621 mountnfs(argp, mp, nam, pth, hst, vpp, p)
    622 	struct nfs_args *argp;
    623 	struct mount *mp;
    624 	struct mbuf *nam;
    625 	const char *pth, *hst;
    626 	struct vnode **vpp;
    627 	struct proc *p;
    628 {
    629 	struct nfsmount *nmp;
    630 	struct nfsnode *np;
    631 	int error;
    632 	struct vattr attrs;
    633 	struct ucred *cr;
    634 
    635 	/*
    636 	 * If the number of nfs iothreads to use has never
    637 	 * been set, create a reasonable number of them.
    638 	 */
    639 
    640 	if (nfs_niothreads < 0) {
    641 		nfs_niothreads = NFS_DEFAULT_NIOTHREADS;
    642 		nfs_getset_niothreads(TRUE);
    643 	}
    644 
    645 	if (mp->mnt_flag & MNT_UPDATE) {
    646 		nmp = VFSTONFS(mp);
    647 		/* update paths, file handles, etc, here	XXX */
    648 		m_freem(nam);
    649 		return (0);
    650 	} else {
    651 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
    652 		    M_NFSMNT, M_WAITOK);
    653 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
    654 		mp->mnt_data = (qaddr_t)nmp;
    655 		TAILQ_INIT(&nmp->nm_uidlruhead);
    656 		TAILQ_INIT(&nmp->nm_bufq);
    657 	}
    658 	vfs_getnewfsid(mp);
    659 	nmp->nm_mountp = mp;
    660 
    661 #ifndef NFS_V2_ONLY
    662 	if (argp->flags & NFSMNT_NQNFS)
    663 		/*
    664 		 * We have to set mnt_maxsymlink to a non-zero value so
    665 		 * that COMPAT_43 routines will know that we are setting
    666 		 * the d_type field in directories (and can zero it for
    667 		 * unsuspecting binaries).
    668 		 */
    669 		mp->mnt_maxsymlinklen = 1;
    670 #endif
    671 
    672 #ifndef NFS_V2_ONLY
    673 	if ((argp->flags & NFSMNT_NFSV3) == 0)
    674 #endif
    675 		/*
    676 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
    677 		 * will fill this in.
    678 		 */
    679 		nmp->nm_maxfilesize = 0xffffffffLL;
    680 
    681 	nmp->nm_timeo = NFS_TIMEO;
    682 	nmp->nm_retry = NFS_RETRANS;
    683 	nmp->nm_wsize = NFS_WSIZE;
    684 	nmp->nm_rsize = NFS_RSIZE;
    685 	nmp->nm_readdirsize = NFS_READDIRSIZE;
    686 	nmp->nm_numgrps = NFS_MAXGRPS;
    687 	nmp->nm_readahead = NFS_DEFRAHEAD;
    688 	nmp->nm_leaseterm = NQ_DEFLEASE;
    689 	nmp->nm_deadthresh = NQ_DEADTHRESH;
    690 	CIRCLEQ_INIT(&nmp->nm_timerhead);
    691 	nmp->nm_inprog = NULLVP;
    692 #ifdef COMPAT_09
    693 	mp->mnt_stat.f_type = 2;
    694 #else
    695 	mp->mnt_stat.f_type = 0;
    696 #endif
    697 	strncpy(&mp->mnt_stat.f_fstypename[0], mp->mnt_op->vfs_name,
    698 	    MFSNAMELEN);
    699 	memcpy(mp->mnt_stat.f_mntfromname, hst, MNAMELEN);
    700 	memcpy(mp->mnt_stat.f_mntonname, pth, MNAMELEN);
    701 	nmp->nm_nam = nam;
    702 
    703 	/* Set up the sockets and per-host congestion */
    704 	nmp->nm_sotype = argp->sotype;
    705 	nmp->nm_soproto = argp->proto;
    706 
    707 	nfs_decode_args(nmp, argp);
    708 
    709 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
    710 	mp->mnt_dev_bshift = DEV_BSHIFT;
    711 
    712 	/*
    713 	 * For Connection based sockets (TCP,...) defer the connect until
    714 	 * the first request, in case the server is not responding.
    715 	 */
    716 	if (nmp->nm_sotype == SOCK_DGRAM &&
    717 		(error = nfs_connect(nmp, (struct nfsreq *)0)))
    718 		goto bad;
    719 
    720 	/*
    721 	 * This is silly, but it has to be set so that vinifod() works.
    722 	 * We do not want to do an nfs_statfs() here since we can get
    723 	 * stuck on a dead server and we are holding a lock on the mount
    724 	 * point.
    725 	 */
    726 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
    727 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
    728 	if (error)
    729 		goto bad;
    730 	*vpp = NFSTOV(np);
    731 	VOP_GETATTR(*vpp, &attrs, p->p_ucred, p);
    732 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
    733 		cr = crget();
    734 		cr->cr_uid = attrs.va_uid;
    735 		cr->cr_gid = attrs.va_gid;
    736 		cr->cr_ngroups = 0;
    737 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
    738 		crfree(cr);
    739 	}
    740 
    741 	/*
    742 	 * A reference count is needed on the nfsnode representing the
    743 	 * remote root.  If this object is not persistent, then backward
    744 	 * traversals of the mount point (i.e. "..") will not work if
    745 	 * the nfsnode gets flushed out of the cache. Ufs does not have
    746 	 * this problem, because one can identify root inodes by their
    747 	 * number == ROOTINO (2). So, just unlock, but no rele.
    748 	 */
    749 
    750 	nmp->nm_vnode = *vpp;
    751 	VOP_UNLOCK(*vpp, 0);
    752 
    753 	return (0);
    754 bad:
    755 	nfs_disconnect(nmp);
    756 	free((caddr_t)nmp, M_NFSMNT);
    757 	m_freem(nam);
    758 	return (error);
    759 }
    760 
    761 /*
    762  * unmount system call
    763  */
    764 int
    765 nfs_unmount(mp, mntflags, p)
    766 	struct mount *mp;
    767 	int mntflags;
    768 	struct proc *p;
    769 {
    770 	struct nfsmount *nmp;
    771 	struct vnode *vp;
    772 	int error, flags = 0;
    773 
    774 	if (mntflags & MNT_FORCE)
    775 		flags |= FORCECLOSE;
    776 	nmp = VFSTONFS(mp);
    777 	/*
    778 	 * Goes something like this..
    779 	 * - Check for activity on the root vnode (other than ourselves).
    780 	 * - Call vflush() to clear out vnodes for this file system,
    781 	 *   except for the root vnode.
    782 	 * - Decrement reference on the vnode representing remote root.
    783 	 * - Close the socket
    784 	 * - Free up the data structures
    785 	 */
    786 	/*
    787 	 * We need to decrement the ref. count on the nfsnode representing
    788 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
    789 	 * has done vput on this vnode, otherwise we would get deadlock!
    790 	 */
    791 	vp = nmp->nm_vnode;
    792 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
    793 	if (error != 0)
    794 		return error;
    795 
    796 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
    797 		vput(vp);
    798 		return (EBUSY);
    799 	}
    800 
    801 	/*
    802 	 * Must handshake with nqnfs_clientd() if it is active.
    803 	 */
    804 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
    805 	while (nmp->nm_inprog != NULLVP)
    806 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
    807 	error = vflush(mp, vp, flags);
    808 	if (error) {
    809 		vput(vp);
    810 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
    811 		return (error);
    812 	}
    813 
    814 	/*
    815 	 * We are now committed to the unmount; mark the mount structure
    816 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
    817 	 * will go away cleanly.
    818 	 */
    819 	nmp->nm_iflag |= NFSMNT_DISMNT;
    820 
    821 	/*
    822 	 * There are two reference counts to get rid of here
    823 	 * (see comment in mountnfs()).
    824 	 */
    825 	vrele(vp);
    826 	vput(vp);
    827 	vgone(vp);
    828 	nfs_disconnect(nmp);
    829 	m_freem(nmp->nm_nam);
    830 
    831 	/*
    832 	 * For NQNFS, let the server daemon free the nfsmount structure.
    833 	 */
    834 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
    835 		free((caddr_t)nmp, M_NFSMNT);
    836 	return (0);
    837 }
    838 
    839 /*
    840  * Return root of a filesystem
    841  */
    842 int
    843 nfs_root(mp, vpp)
    844 	struct mount *mp;
    845 	struct vnode **vpp;
    846 {
    847 	struct vnode *vp;
    848 	struct nfsmount *nmp;
    849 	int error;
    850 
    851 	nmp = VFSTONFS(mp);
    852 	vp = nmp->nm_vnode;
    853 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
    854 	if (error != 0)
    855 		return error;
    856 	if (vp->v_type == VNON)
    857 		vp->v_type = VDIR;
    858 	vp->v_flag = VROOT;
    859 	*vpp = vp;
    860 	return (0);
    861 }
    862 
    863 extern int syncprt;
    864 
    865 /*
    866  * Flush out the buffer cache
    867  */
    868 /* ARGSUSED */
    869 int
    870 nfs_sync(mp, waitfor, cred, p)
    871 	struct mount *mp;
    872 	int waitfor;
    873 	struct ucred *cred;
    874 	struct proc *p;
    875 {
    876 	struct vnode *vp;
    877 	int error, allerror = 0;
    878 
    879 	/*
    880 	 * Force stale buffer cache information to be flushed.
    881 	 */
    882 loop:
    883 	for (vp = mp->mnt_vnodelist.lh_first;
    884 	     vp != NULL;
    885 	     vp = vp->v_mntvnodes.le_next) {
    886 		/*
    887 		 * If the vnode that we are about to sync is no longer
    888 		 * associated with this mount point, start over.
    889 		 */
    890 		if (vp->v_mount != mp)
    891 			goto loop;
    892 		if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) ||
    893 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
    894 		     vp->v_uobj.uo_npages == 0))
    895 			continue;
    896 		if (vget(vp, LK_EXCLUSIVE))
    897 			goto loop;
    898 		error = VOP_FSYNC(vp, cred,
    899 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p);
    900 		if (error)
    901 			allerror = error;
    902 		vput(vp);
    903 	}
    904 	return (allerror);
    905 }
    906 
    907 /*
    908  * NFS flat namespace lookup.
    909  * Currently unsupported.
    910  */
    911 /* ARGSUSED */
    912 int
    913 nfs_vget(mp, ino, vpp)
    914 	struct mount *mp;
    915 	ino_t ino;
    916 	struct vnode **vpp;
    917 {
    918 
    919 	return (EOPNOTSUPP);
    920 }
    921 
    922 /*
    923  * Do that sysctl thang...
    924  */
    925 int
    926 nfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
    927 	int *name;
    928 	u_int namelen;
    929 	void *oldp;
    930 	size_t *oldlenp;
    931 	void *newp;
    932 	size_t newlen;
    933 	struct proc *p;
    934 {
    935 	int rv;
    936 
    937 	/*
    938 	 * All names at this level are terminal.
    939 	 */
    940 	if(namelen > 1)
    941 		return ENOTDIR;	/* overloaded */
    942 
    943 	switch(name[0]) {
    944 	case NFS_NFSSTATS:
    945 		if(!oldp) {
    946 			*oldlenp = sizeof nfsstats;
    947 			return 0;
    948 		}
    949 
    950 		if(*oldlenp < sizeof nfsstats) {
    951 			*oldlenp = sizeof nfsstats;
    952 			return ENOMEM;
    953 		}
    954 
    955 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
    956 		if(rv) return rv;
    957 
    958 		if(newp && newlen != sizeof nfsstats)
    959 			return EINVAL;
    960 
    961 		if(newp) {
    962 			return copyin(newp, &nfsstats, sizeof nfsstats);
    963 		}
    964 		return 0;
    965 
    966 	case NFS_IOTHREADS:
    967 		nfs_getset_niothreads(0);
    968 
    969 		rv = (sysctl_int(oldp, oldlenp, newp, newlen,
    970 		    &nfs_niothreads));
    971 
    972 		if (newp)
    973 			nfs_getset_niothreads(1);
    974 
    975 		return rv;
    976 
    977 	default:
    978 		return EOPNOTSUPP;
    979 	}
    980 }
    981 
    982 
    983 /*
    984  * At this point, this should never happen
    985  */
    986 /* ARGSUSED */
    987 int
    988 nfs_fhtovp(mp, fhp, vpp)
    989 	struct mount *mp;
    990 	struct fid *fhp;
    991 	struct vnode **vpp;
    992 {
    993 
    994 	return (EINVAL);
    995 }
    996 
    997 /* ARGSUSED */
    998 int
    999 nfs_checkexp(mp, nam, exflagsp, credanonp)
   1000 	struct mount *mp;
   1001 	struct mbuf *nam;
   1002 	int *exflagsp;
   1003 	struct ucred **credanonp;
   1004 {
   1005 
   1006 	return (EINVAL);
   1007 }
   1008 
   1009 /*
   1010  * Vnode pointer to File handle, should never happen either
   1011  */
   1012 /* ARGSUSED */
   1013 int
   1014 nfs_vptofh(vp, fhp)
   1015 	struct vnode *vp;
   1016 	struct fid *fhp;
   1017 {
   1018 
   1019 	return (EINVAL);
   1020 }
   1021 
   1022 /*
   1023  * Vfs start routine, a no-op.
   1024  */
   1025 /* ARGSUSED */
   1026 int
   1027 nfs_start(mp, flags, p)
   1028 	struct mount *mp;
   1029 	int flags;
   1030 	struct proc *p;
   1031 {
   1032 
   1033 	return (0);
   1034 }
   1035 
   1036 /*
   1037  * Do operations associated with quotas, not supported
   1038  */
   1039 /* ARGSUSED */
   1040 int
   1041 nfs_quotactl(mp, cmd, uid, arg, p)
   1042 	struct mount *mp;
   1043 	int cmd;
   1044 	uid_t uid;
   1045 	caddr_t arg;
   1046 	struct proc *p;
   1047 {
   1048 
   1049 	return (EOPNOTSUPP);
   1050 }
   1051