Home | History | Annotate | Line # | Download | only in nfs
nfs_vfsops.c revision 1.101.2.5
      1 /*	$NetBSD: nfs_vfsops.c,v 1.101.2.5 2001/10/22 20:42:10 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 #if defined(_KERNEL_OPT)
     42 #include "opt_compat_netbsd.h"
     43 #include "opt_nfs.h"
     44 #endif
     45 
     46 #include <sys/param.h>
     47 #include <sys/ioctl.h>
     48 #include <sys/signal.h>
     49 #include <sys/lwp.h>
     50 #include <sys/proc.h>
     51 #include <sys/namei.h>
     52 #include <sys/device.h>
     53 #include <sys/vnode.h>
     54 #include <sys/kernel.h>
     55 #include <sys/mount.h>
     56 #include <sys/buf.h>
     57 #include <sys/mbuf.h>
     58 #include <sys/socket.h>
     59 #include <sys/socketvar.h>
     60 #include <sys/sysctl.h>
     61 #include <sys/systm.h>
     62 
     63 #include <net/if.h>
     64 #include <net/route.h>
     65 #include <netinet/in.h>
     66 
     67 #include <nfs/rpcv2.h>
     68 #include <nfs/nfsproto.h>
     69 #include <nfs/nfsnode.h>
     70 #include <nfs/nfs.h>
     71 #include <nfs/nfsmount.h>
     72 #include <nfs/xdr_subs.h>
     73 #include <nfs/nfsm_subs.h>
     74 #include <nfs/nfsdiskless.h>
     75 #include <nfs/nqnfs.h>
     76 #include <nfs/nfs_var.h>
     77 
     78 extern struct nfsstats nfsstats;
     79 extern int nfs_ticks;
     80 
     81 int nfs_sysctl __P((int *, u_int, void *, size_t *, void *, size_t,
     82 		      struct proc *));
     83 
     84 /*
     85  * nfs vfs operations.
     86  */
     87 
     88 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc;
     89 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc;
     90 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc;
     91 
     92 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = {
     93 	&nfsv2_vnodeop_opv_desc,
     94 	&spec_nfsv2nodeop_opv_desc,
     95 	&fifo_nfsv2nodeop_opv_desc,
     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_vfs_reinit,
    113 	nfs_vfs_done,
    114 	nfs_sysctl,
    115 	nfs_mountroot,
    116 	nfs_checkexp,
    117 	nfs_vnodeopv_descs,
    118 };
    119 
    120 extern u_int32_t nfs_procids[NFS_NPROCS];
    121 extern u_int32_t nfs_prog, nfs_vers;
    122 
    123 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *,
    124     struct mount **, struct vnode **, struct proc *));
    125 
    126 #define TRUE	1
    127 #define	FALSE	0
    128 
    129 /*
    130  * nfs statfs call
    131  */
    132 int
    133 nfs_statfs(mp, sbp, p)
    134 	struct mount *mp;
    135 	struct statfs *sbp;
    136 	struct proc *p;
    137 {
    138 	struct vnode *vp;
    139 	struct nfs_statfs *sfp;
    140 	caddr_t cp;
    141 	u_int32_t *tl;
    142 	int32_t t1, t2;
    143 	caddr_t bpos, dpos, cp2;
    144 	struct nfsmount *nmp = VFSTONFS(mp);
    145 	int error = 0, retattr;
    146 #ifdef NFS_V2_ONLY
    147 	const int v3 = 0;
    148 #else
    149 	int v3 = (nmp->nm_flag & NFSMNT_NFSV3);
    150 #endif
    151 	struct mbuf *mreq, *mrep = NULL, *md, *mb, *mb2;
    152 	struct ucred *cred;
    153 	u_quad_t tquad;
    154 
    155 #ifndef nolint
    156 	sfp = (struct nfs_statfs *)0;
    157 #endif
    158 	vp = nmp->nm_vnode;
    159 	cred = crget();
    160 	cred->cr_ngroups = 0;
    161 	if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0)
    162 		(void)nfs_fsinfo(nmp, vp, cred, p);
    163 	nfsstats.rpccnt[NFSPROC_FSSTAT]++;
    164 	nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3));
    165 	nfsm_fhtom(vp, v3);
    166 	nfsm_request(vp, NFSPROC_FSSTAT, p, cred);
    167 	if (v3)
    168 		nfsm_postop_attr(vp, retattr);
    169 	if (error) {
    170 		if (mrep != NULL)
    171 			m_free(mrep);
    172 		goto nfsmout;
    173 	}
    174 	nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
    175 #ifdef COMPAT_09
    176 	sbp->f_type = 2;
    177 #else
    178 	sbp->f_type = 0;
    179 #endif
    180 	sbp->f_flags = nmp->nm_flag;
    181 	sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize);
    182 	if (v3) {
    183 		sbp->f_bsize = NFS_FABLKSIZE;
    184 		tquad = fxdr_hyper(&sfp->sf_tbytes);
    185 		sbp->f_blocks = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
    186 		tquad = fxdr_hyper(&sfp->sf_fbytes);
    187 		sbp->f_bfree = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
    188 		tquad = fxdr_hyper(&sfp->sf_abytes);
    189 		sbp->f_bavail = (long)((quad_t)tquad / (quad_t)NFS_FABLKSIZE);
    190 		tquad = fxdr_hyper(&sfp->sf_tfiles);
    191 		sbp->f_files = (long)tquad;
    192 		tquad = fxdr_hyper(&sfp->sf_ffiles);
    193 		sbp->f_ffree = (long)tquad;
    194 	} else {
    195 		sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize);
    196 		sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks);
    197 		sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree);
    198 		sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail);
    199 		sbp->f_files = 0;
    200 		sbp->f_ffree = 0;
    201 	}
    202 	if (sbp != &mp->mnt_stat) {
    203 		memcpy(sbp->f_mntonname, mp->mnt_stat.f_mntonname, MNAMELEN);
    204 		memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, MNAMELEN);
    205 	}
    206 	strncpy(&sbp->f_fstypename[0], mp->mnt_op->vfs_name, MFSNAMELEN);
    207 	nfsm_reqdone;
    208 	crfree(cred);
    209 	return (error);
    210 }
    211 
    212 #ifndef NFS_V2_ONLY
    213 /*
    214  * nfs version 3 fsinfo rpc call
    215  */
    216 int
    217 nfs_fsinfo(nmp, vp, cred, p)
    218 	struct nfsmount *nmp;
    219 	struct vnode *vp;
    220 	struct ucred *cred;
    221 	struct proc *p;
    222 {
    223 	struct nfsv3_fsinfo *fsp;
    224 	caddr_t cp;
    225 	int32_t t1, t2;
    226 	u_int32_t *tl, pref, max;
    227 	caddr_t bpos, dpos, cp2;
    228 	int error = 0, retattr;
    229 	struct mbuf *mreq, *mrep, *md, *mb, *mb2;
    230 	u_int64_t maxfsize;
    231 
    232 	nfsstats.rpccnt[NFSPROC_FSINFO]++;
    233 	nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1));
    234 	nfsm_fhtom(vp, 1);
    235 	nfsm_request(vp, NFSPROC_FSINFO, p, cred);
    236 	nfsm_postop_attr(vp, retattr);
    237 	if (!error) {
    238 		nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
    239 		pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref);
    240 		if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE)
    241 			nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) &
    242 				~(NFS_FABLKSIZE - 1);
    243 		max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax);
    244 		if (max < nmp->nm_wsize && max > 0) {
    245 			nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1);
    246 			if (nmp->nm_wsize == 0)
    247 				nmp->nm_wsize = max;
    248 		}
    249 		pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref);
    250 		if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE)
    251 			nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) &
    252 				~(NFS_FABLKSIZE - 1);
    253 		max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax);
    254 		if (max < nmp->nm_rsize && max > 0) {
    255 			nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1);
    256 			if (nmp->nm_rsize == 0)
    257 				nmp->nm_rsize = max;
    258 		}
    259 		pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref);
    260 		if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ)
    261 			nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) &
    262 				~(NFS_DIRFRAGSIZ - 1);
    263 		if (max < nmp->nm_readdirsize && max > 0) {
    264 			nmp->nm_readdirsize = max & ~(NFS_DIRFRAGSIZ - 1);
    265 			if (nmp->nm_readdirsize == 0)
    266 				nmp->nm_readdirsize = max;
    267 		}
    268 		/* XXX */
    269 		nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1;
    270 		maxfsize = fxdr_hyper(&fsp->fs_maxfilesize);
    271 		if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize)
    272 			nmp->nm_maxfilesize = maxfsize;
    273 		nmp->nm_iflag |= NFSMNT_GOTFSINFO;
    274 	}
    275 	nfsm_reqdone;
    276 	return (error);
    277 }
    278 #endif
    279 
    280 /*
    281  * Mount a remote root fs via. NFS.  It goes like this:
    282  * - Call nfs_boot_init() to fill in the nfs_diskless struct
    283  * - build the rootfs mount point and call mountnfs() to do the rest.
    284  */
    285 int
    286 nfs_mountroot()
    287 {
    288 	struct nfs_diskless *nd;
    289 	struct vattr attr;
    290 	struct mount *mp;
    291 	struct vnode *vp;
    292 	struct proc *procp;
    293 	long n;
    294 	int error;
    295 
    296 	procp = curproc->l_proc; /* XXX */
    297 
    298 	if (root_device->dv_class != DV_IFNET)
    299 		return (ENODEV);
    300 
    301 	/*
    302 	 * XXX time must be non-zero when we init the interface or else
    303 	 * the arp code will wedge.  [Fixed now in if_ether.c]
    304 	 * However, the NFS attribute cache gives false "hits" when
    305 	 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now.
    306 	 */
    307 	if (time.tv_sec < NFS_MAXATTRTIMO)
    308 		time.tv_sec = NFS_MAXATTRTIMO;
    309 
    310 	/*
    311 	 * Call nfs_boot_init() to fill in the nfs_diskless struct.
    312 	 * Side effect:  Finds and configures a network interface.
    313 	 */
    314 	nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK);
    315 	memset((caddr_t)nd, 0, sizeof(*nd));
    316 	error = nfs_boot_init(nd, procp);
    317 	if (error) {
    318 		free(nd, M_NFSMNT);
    319 		return (error);
    320 	}
    321 
    322 	/*
    323 	 * Create the root mount point.
    324 	 */
    325 	error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp);
    326 	if (error)
    327 		goto out;
    328 	printf("root on %s\n", nd->nd_root.ndm_host);
    329 
    330 	/*
    331 	 * Link it into the mount list.
    332 	 */
    333 	simple_lock(&mountlist_slock);
    334 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
    335 	simple_unlock(&mountlist_slock);
    336 	rootvp = vp;
    337 	mp->mnt_vnodecovered = NULLVP;
    338 	vfs_unbusy(mp);
    339 
    340 	/* Get root attributes (for the time). */
    341 	error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp);
    342 	if (error)
    343 		panic("nfs_mountroot: getattr for root");
    344 	n = attr.va_atime.tv_sec;
    345 #ifdef	DEBUG
    346 	printf("root time: 0x%lx\n", n);
    347 #endif
    348 	inittodr(n);
    349 
    350 out:
    351 	if (error)
    352 		nfs_boot_cleanup(nd, procp);
    353 	free(nd, M_NFSMNT);
    354 	return (error);
    355 }
    356 
    357 /*
    358  * Internal version of mount system call for diskless setup.
    359  * Separate function because we used to call it twice.
    360  * (once for root and once for swap)
    361  */
    362 static int
    363 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p)
    364 	struct nfs_dlmount *ndmntp;
    365 	const char *mntname;	/* mount point name */
    366 	struct mount **mpp;
    367 	struct vnode **vpp;
    368 	struct proc *p;
    369 {
    370 	struct mount *mp;
    371 	struct mbuf *m;
    372 	int error;
    373 
    374 	vfs_rootmountalloc(MOUNT_NFS, (char *)mntname, &mp);
    375 
    376 	mp->mnt_op = &nfs_vfsops;
    377 
    378 	/*
    379 	 * Historical practice expects NFS root file systems to
    380 	 * be initially mounted r/w.
    381 	 */
    382 	mp->mnt_flag &= ~MNT_RDONLY;
    383 
    384 	/* Get mbuf for server sockaddr. */
    385 	m = m_get(M_WAIT, MT_SONAME);
    386 	if (m == NULL)
    387 		panic("nfs_mountroot: mget soname for %s", mntname);
    388 	memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr,
    389 	      (m->m_len = ndmntp->ndm_args.addr->sa_len));
    390 
    391 	error = mountnfs(&ndmntp->ndm_args, mp, m, mntname,
    392 			 ndmntp->ndm_args.hostname, vpp, p);
    393 	if (error) {
    394 		mp->mnt_op->vfs_refcount--;
    395 		vfs_unbusy(mp);
    396 		printf("nfs_mountroot: mount %s failed: %d\n",
    397 		       mntname, error);
    398 		free(mp, M_MOUNT);
    399 	} else
    400 		*mpp = mp;
    401 
    402 	return (error);
    403 }
    404 
    405 void
    406 nfs_decode_args(nmp, argp)
    407 	struct nfsmount *nmp;
    408 	struct nfs_args *argp;
    409 {
    410 	int s;
    411 	int adjsock;
    412 	int maxio;
    413 
    414 	s = splsoftnet();
    415 
    416 	/*
    417 	 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes
    418 	 * no sense in that context.
    419 	 */
    420 	if (argp->sotype == SOCK_STREAM)
    421 		argp->flags &= ~NFSMNT_NOCONN;
    422 
    423 	/*
    424 	 * Cookie translation is not needed for v2, silently ignore it.
    425 	 */
    426 	if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) ==
    427 	    NFSMNT_XLATECOOKIE)
    428 		argp->flags &= ~NFSMNT_XLATECOOKIE;
    429 
    430 	/* Re-bind if rsrvd port requested and wasn't on one */
    431 	adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT)
    432 		  && (argp->flags & NFSMNT_RESVPORT);
    433 	/* Also re-bind if we're switching to/from a connected UDP socket */
    434 	adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) !=
    435 		    (argp->flags & NFSMNT_NOCONN));
    436 
    437 	/* Update flags. */
    438 	nmp->nm_flag = argp->flags;
    439 	splx(s);
    440 
    441 	if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) {
    442 		nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10;
    443 		if (nmp->nm_timeo < NFS_MINTIMEO)
    444 			nmp->nm_timeo = NFS_MINTIMEO;
    445 		else if (nmp->nm_timeo > NFS_MAXTIMEO)
    446 			nmp->nm_timeo = NFS_MAXTIMEO;
    447 	}
    448 
    449 	if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) {
    450 		nmp->nm_retry = argp->retrans;
    451 		if (nmp->nm_retry > NFS_MAXREXMIT)
    452 			nmp->nm_retry = NFS_MAXREXMIT;
    453 	}
    454 
    455 #ifndef NFS_V2_ONLY
    456 	if (argp->flags & NFSMNT_NFSV3) {
    457 		if (argp->sotype == SOCK_DGRAM)
    458 			maxio = NFS_MAXDGRAMDATA;
    459 		else
    460 			maxio = NFS_MAXDATA;
    461 	} else
    462 #endif
    463 		maxio = NFS_V2MAXDATA;
    464 
    465 	if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) {
    466 		int osize = nmp->nm_wsize;
    467 		nmp->nm_wsize = argp->wsize;
    468 		/* Round down to multiple of blocksize */
    469 		nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
    470 		if (nmp->nm_wsize <= 0)
    471 			nmp->nm_wsize = NFS_FABLKSIZE;
    472 		adjsock |= (nmp->nm_wsize != osize);
    473 	}
    474 	if (nmp->nm_wsize > maxio)
    475 		nmp->nm_wsize = maxio;
    476 	if (nmp->nm_wsize > MAXBSIZE)
    477 		nmp->nm_wsize = MAXBSIZE;
    478 
    479 	if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) {
    480 		int osize = nmp->nm_rsize;
    481 		nmp->nm_rsize = argp->rsize;
    482 		/* Round down to multiple of blocksize */
    483 		nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
    484 		if (nmp->nm_rsize <= 0)
    485 			nmp->nm_rsize = NFS_FABLKSIZE;
    486 		adjsock |= (nmp->nm_rsize != osize);
    487 	}
    488 	if (nmp->nm_rsize > maxio)
    489 		nmp->nm_rsize = maxio;
    490 	if (nmp->nm_rsize > MAXBSIZE)
    491 		nmp->nm_rsize = MAXBSIZE;
    492 
    493 	if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) {
    494 		nmp->nm_readdirsize = argp->readdirsize;
    495 		/* Round down to multiple of minimum blocksize */
    496 		nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1);
    497 		if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ)
    498 			nmp->nm_readdirsize = NFS_DIRFRAGSIZ;
    499 		/* Bigger than buffer size makes no sense */
    500 		if (nmp->nm_readdirsize > NFS_DIRBLKSIZ)
    501 			nmp->nm_readdirsize = NFS_DIRBLKSIZ;
    502 	} else if (argp->flags & NFSMNT_RSIZE)
    503 		nmp->nm_readdirsize = nmp->nm_rsize;
    504 
    505 	if (nmp->nm_readdirsize > maxio)
    506 		nmp->nm_readdirsize = maxio;
    507 
    508 	if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 &&
    509 		argp->maxgrouplist <= NFS_MAXGRPS)
    510 		nmp->nm_numgrps = argp->maxgrouplist;
    511 	if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 &&
    512 		argp->readahead <= NFS_MAXRAHEAD)
    513 		nmp->nm_readahead = argp->readahead;
    514 	if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 &&
    515 		argp->leaseterm <= NQ_MAXLEASE)
    516 		nmp->nm_leaseterm = argp->leaseterm;
    517 	if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 &&
    518 		argp->deadthresh <= NQ_NEVERDEAD)
    519 		nmp->nm_deadthresh = argp->deadthresh;
    520 
    521 	adjsock |= ((nmp->nm_sotype != argp->sotype) ||
    522 		    (nmp->nm_soproto != argp->proto));
    523 	nmp->nm_sotype = argp->sotype;
    524 	nmp->nm_soproto = argp->proto;
    525 
    526 	if (nmp->nm_so && adjsock) {
    527 		nfs_safedisconnect(nmp);
    528 		if (nmp->nm_sotype == SOCK_DGRAM)
    529 			while (nfs_connect(nmp, (struct nfsreq *)0)) {
    530 				printf("nfs_args: retrying connect\n");
    531 				(void) tsleep((caddr_t)&lbolt,
    532 					      PSOCK, "nfscn3", 0);
    533 			}
    534 	}
    535 }
    536 
    537 /*
    538  * VFS Operations.
    539  *
    540  * mount system call
    541  * It seems a bit dumb to copyinstr() the host and path here and then
    542  * memcpy() them in mountnfs(), but I wanted to detect errors before
    543  * doing the sockargs() call because sockargs() allocates an mbuf and
    544  * an error after that means that I have to release the mbuf.
    545  */
    546 /* ARGSUSED */
    547 int
    548 nfs_mount(mp, path, data, ndp, p)
    549 	struct mount *mp;
    550 	const char *path;
    551 	void *data;
    552 	struct nameidata *ndp;
    553 	struct proc *p;
    554 {
    555 	int error;
    556 	struct nfs_args args;
    557 	struct mbuf *nam;
    558 	struct vnode *vp;
    559 	char pth[MNAMELEN], hst[MNAMELEN];
    560 	size_t len;
    561 	u_char nfh[NFSX_V3FHMAX];
    562 
    563 	error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args));
    564 	if (error)
    565 		return (error);
    566 	if (args.version != NFS_ARGSVERSION)
    567 		return (EPROGMISMATCH);
    568 #ifdef NFS_V2_ONLY
    569 	if (args.flags & NFSMNT_NQNFS)
    570 		return (EPROGUNAVAIL);
    571 	if (args.flags & NFSMNT_NFSV3)
    572 		return (EPROGMISMATCH);
    573 #endif
    574 	if (mp->mnt_flag & MNT_UPDATE) {
    575 		struct nfsmount *nmp = VFSTONFS(mp);
    576 
    577 		if (nmp == NULL)
    578 			return (EIO);
    579 		/*
    580 		 * When doing an update, we can't change from or to
    581 		 * v3 and/or nqnfs, or change cookie translation
    582 		 */
    583 		args.flags = (args.flags &
    584 		    ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) |
    585 		    (nmp->nm_flag &
    586 			(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE));
    587 		nfs_decode_args(nmp, &args);
    588 		return (0);
    589 	}
    590 	if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX)
    591 		return (EINVAL);
    592 	error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize);
    593 	if (error)
    594 		return (error);
    595 	error = copyinstr(path, pth, MNAMELEN-1, &len);
    596 	if (error)
    597 		return (error);
    598 	memset(&pth[len], 0, MNAMELEN - len);
    599 	error = copyinstr(args.hostname, hst, MNAMELEN-1, &len);
    600 	if (error)
    601 		return (error);
    602 	memset(&hst[len], 0, MNAMELEN - len);
    603 	/* sockargs() call must be after above copyin() calls */
    604 	error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME);
    605 	if (error)
    606 		return (error);
    607 	args.fh = nfh;
    608 	error = mountnfs(&args, mp, nam, pth, hst, &vp, p);
    609 	return (error);
    610 }
    611 
    612 /*
    613  * Common code for mount and mountroot
    614  */
    615 int
    616 mountnfs(argp, mp, nam, pth, hst, vpp, p)
    617 	struct nfs_args *argp;
    618 	struct mount *mp;
    619 	struct mbuf *nam;
    620 	const char *pth, *hst;
    621 	struct vnode **vpp;
    622 	struct proc *p;
    623 {
    624 	struct nfsmount *nmp;
    625 	struct nfsnode *np;
    626 	int error;
    627 	struct vattr attrs;
    628 	struct ucred *cr;
    629 
    630 	/*
    631 	 * If the number of nfs iothreads to use has never
    632 	 * been set, create a reasonable number of them.
    633 	 */
    634 
    635 	if (nfs_niothreads < 0) {
    636 		nfs_niothreads = NFS_DEFAULT_NIOTHREADS;
    637 		nfs_getset_niothreads(TRUE);
    638 	}
    639 
    640 	if (mp->mnt_flag & MNT_UPDATE) {
    641 		nmp = VFSTONFS(mp);
    642 		/* update paths, file handles, etc, here	XXX */
    643 		m_freem(nam);
    644 		return (0);
    645 	} else {
    646 		MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount),
    647 		    M_NFSMNT, M_WAITOK);
    648 		memset((caddr_t)nmp, 0, sizeof (struct nfsmount));
    649 		mp->mnt_data = (qaddr_t)nmp;
    650 		TAILQ_INIT(&nmp->nm_uidlruhead);
    651 		TAILQ_INIT(&nmp->nm_bufq);
    652 	}
    653 	vfs_getnewfsid(mp);
    654 	nmp->nm_mountp = mp;
    655 
    656 #ifndef NFS_V2_ONLY
    657 	if (argp->flags & NFSMNT_NQNFS)
    658 		/*
    659 		 * We have to set mnt_maxsymlink to a non-zero value so
    660 		 * that COMPAT_43 routines will know that we are setting
    661 		 * the d_type field in directories (and can zero it for
    662 		 * unsuspecting binaries).
    663 		 */
    664 		mp->mnt_maxsymlinklen = 1;
    665 #endif
    666 
    667 #ifndef NFS_V2_ONLY
    668 	if ((argp->flags & NFSMNT_NFSV3) == 0)
    669 #endif
    670 		/*
    671 		 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo
    672 		 * will fill this in.
    673 		 */
    674 		nmp->nm_maxfilesize = 0xffffffffLL;
    675 
    676 	nmp->nm_timeo = NFS_TIMEO;
    677 	nmp->nm_retry = NFS_RETRANS;
    678 	nmp->nm_wsize = NFS_WSIZE;
    679 	nmp->nm_rsize = NFS_RSIZE;
    680 	nmp->nm_readdirsize = NFS_READDIRSIZE;
    681 	nmp->nm_numgrps = NFS_MAXGRPS;
    682 	nmp->nm_readahead = NFS_DEFRAHEAD;
    683 	nmp->nm_leaseterm = NQ_DEFLEASE;
    684 	nmp->nm_deadthresh = NQ_DEADTHRESH;
    685 	CIRCLEQ_INIT(&nmp->nm_timerhead);
    686 	nmp->nm_inprog = NULLVP;
    687 #ifdef COMPAT_09
    688 	mp->mnt_stat.f_type = 2;
    689 #else
    690 	mp->mnt_stat.f_type = 0;
    691 #endif
    692 	strncpy(&mp->mnt_stat.f_fstypename[0], mp->mnt_op->vfs_name,
    693 	    MFSNAMELEN);
    694 	memcpy(mp->mnt_stat.f_mntfromname, hst, MNAMELEN);
    695 	memcpy(mp->mnt_stat.f_mntonname, pth, MNAMELEN);
    696 	nmp->nm_nam = nam;
    697 
    698 	/* Set up the sockets and per-host congestion */
    699 	nmp->nm_sotype = argp->sotype;
    700 	nmp->nm_soproto = argp->proto;
    701 
    702 	nfs_decode_args(nmp, argp);
    703 
    704 	mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1;
    705 	mp->mnt_dev_bshift = DEV_BSHIFT;
    706 
    707 	/*
    708 	 * For Connection based sockets (TCP,...) defer the connect until
    709 	 * the first request, in case the server is not responding.
    710 	 */
    711 	if (nmp->nm_sotype == SOCK_DGRAM &&
    712 		(error = nfs_connect(nmp, (struct nfsreq *)0)))
    713 		goto bad;
    714 
    715 	/*
    716 	 * This is silly, but it has to be set so that vinifod() works.
    717 	 * We do not want to do an nfs_statfs() here since we can get
    718 	 * stuck on a dead server and we are holding a lock on the mount
    719 	 * point.
    720 	 */
    721 	mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA;
    722 	error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np);
    723 	if (error)
    724 		goto bad;
    725 	*vpp = NFSTOV(np);
    726 	VOP_GETATTR(*vpp, &attrs, p->p_ucred, p);
    727 	if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) {
    728 		cr = crget();
    729 		cr->cr_uid = attrs.va_uid;
    730 		cr->cr_gid = attrs.va_gid;
    731 		cr->cr_ngroups = 0;
    732 		nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr);
    733 		crfree(cr);
    734 	}
    735 
    736 	/*
    737 	 * A reference count is needed on the nfsnode representing the
    738 	 * remote root.  If this object is not persistent, then backward
    739 	 * traversals of the mount point (i.e. "..") will not work if
    740 	 * the nfsnode gets flushed out of the cache. Ufs does not have
    741 	 * this problem, because one can identify root inodes by their
    742 	 * number == ROOTINO (2). So, just unlock, but no rele.
    743 	 */
    744 
    745 	nmp->nm_vnode = *vpp;
    746 	VOP_UNLOCK(*vpp, 0);
    747 
    748 	return (0);
    749 bad:
    750 	nfs_disconnect(nmp);
    751 	free((caddr_t)nmp, M_NFSMNT);
    752 	m_freem(nam);
    753 	return (error);
    754 }
    755 
    756 /*
    757  * unmount system call
    758  */
    759 int
    760 nfs_unmount(mp, mntflags, p)
    761 	struct mount *mp;
    762 	int mntflags;
    763 	struct proc *p;
    764 {
    765 	struct nfsmount *nmp;
    766 	struct vnode *vp;
    767 	int error, flags = 0;
    768 
    769 	if (mntflags & MNT_FORCE)
    770 		flags |= FORCECLOSE;
    771 	nmp = VFSTONFS(mp);
    772 	/*
    773 	 * Goes something like this..
    774 	 * - Check for activity on the root vnode (other than ourselves).
    775 	 * - Call vflush() to clear out vnodes for this file system,
    776 	 *   except for the root vnode.
    777 	 * - Decrement reference on the vnode representing remote root.
    778 	 * - Close the socket
    779 	 * - Free up the data structures
    780 	 */
    781 	/*
    782 	 * We need to decrement the ref. count on the nfsnode representing
    783 	 * the remote root.  See comment in mountnfs().  The VFS unmount()
    784 	 * has done vput on this vnode, otherwise we would get deadlock!
    785 	 */
    786 	vp = nmp->nm_vnode;
    787 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
    788 	if (error != 0)
    789 		return error;
    790 
    791 	if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) {
    792 		vput(vp);
    793 		return (EBUSY);
    794 	}
    795 
    796 	/*
    797 	 * Must handshake with nqnfs_clientd() if it is active.
    798 	 */
    799 	nmp->nm_iflag |= NFSMNT_DISMINPROG;
    800 	while (nmp->nm_inprog != NULLVP)
    801 		(void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0);
    802 	error = vflush(mp, vp, flags);
    803 	if (error) {
    804 		vput(vp);
    805 		nmp->nm_iflag &= ~NFSMNT_DISMINPROG;
    806 		return (error);
    807 	}
    808 
    809 	/*
    810 	 * We are now committed to the unmount; mark the mount structure
    811 	 * as doomed so that any sleepers kicked awake by nfs_disconnect
    812 	 * will go away cleanly.
    813 	 */
    814 	nmp->nm_iflag |= NFSMNT_DISMNT;
    815 
    816 	/*
    817 	 * There are two reference counts to get rid of here
    818 	 * (see comment in mountnfs()).
    819 	 */
    820 	vrele(vp);
    821 	vput(vp);
    822 	vgone(vp);
    823 	nfs_disconnect(nmp);
    824 	m_freem(nmp->nm_nam);
    825 
    826 	/*
    827 	 * For NQNFS, let the server daemon free the nfsmount structure.
    828 	 */
    829 	if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0)
    830 		free((caddr_t)nmp, M_NFSMNT);
    831 	return (0);
    832 }
    833 
    834 /*
    835  * Return root of a filesystem
    836  */
    837 int
    838 nfs_root(mp, vpp)
    839 	struct mount *mp;
    840 	struct vnode **vpp;
    841 {
    842 	struct vnode *vp;
    843 	struct nfsmount *nmp;
    844 	int error;
    845 
    846 	nmp = VFSTONFS(mp);
    847 	vp = nmp->nm_vnode;
    848 	error = vget(vp, LK_EXCLUSIVE | LK_RETRY);
    849 	if (error != 0)
    850 		return error;
    851 	if (vp->v_type == VNON)
    852 		vp->v_type = VDIR;
    853 	vp->v_flag = VROOT;
    854 	*vpp = vp;
    855 	return (0);
    856 }
    857 
    858 extern int syncprt;
    859 
    860 /*
    861  * Flush out the buffer cache
    862  */
    863 /* ARGSUSED */
    864 int
    865 nfs_sync(mp, waitfor, cred, p)
    866 	struct mount *mp;
    867 	int waitfor;
    868 	struct ucred *cred;
    869 	struct proc *p;
    870 {
    871 	struct vnode *vp;
    872 	int error, allerror = 0;
    873 
    874 	/*
    875 	 * Force stale buffer cache information to be flushed.
    876 	 */
    877 loop:
    878 	for (vp = mp->mnt_vnodelist.lh_first;
    879 	     vp != NULL;
    880 	     vp = vp->v_mntvnodes.le_next) {
    881 		/*
    882 		 * If the vnode that we are about to sync is no longer
    883 		 * associated with this mount point, start over.
    884 		 */
    885 		if (vp->v_mount != mp)
    886 			goto loop;
    887 		if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) ||
    888 		    (LIST_EMPTY(&vp->v_dirtyblkhd) &&
    889 		     vp->v_uobj.uo_npages == 0))
    890 			continue;
    891 		if (vget(vp, LK_EXCLUSIVE))
    892 			goto loop;
    893 		error = VOP_FSYNC(vp, cred,
    894 		    waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p);
    895 		if (error)
    896 			allerror = error;
    897 		vput(vp);
    898 	}
    899 	return (allerror);
    900 }
    901 
    902 /*
    903  * NFS flat namespace lookup.
    904  * Currently unsupported.
    905  */
    906 /* ARGSUSED */
    907 int
    908 nfs_vget(mp, ino, vpp)
    909 	struct mount *mp;
    910 	ino_t ino;
    911 	struct vnode **vpp;
    912 {
    913 
    914 	return (EOPNOTSUPP);
    915 }
    916 
    917 /*
    918  * Do that sysctl thang...
    919  */
    920 int
    921 nfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
    922 	int *name;
    923 	u_int namelen;
    924 	void *oldp;
    925 	size_t *oldlenp;
    926 	void *newp;
    927 	size_t newlen;
    928 	struct proc *p;
    929 {
    930 	int rv;
    931 
    932 	/*
    933 	 * All names at this level are terminal.
    934 	 */
    935 	if(namelen > 1)
    936 		return ENOTDIR;	/* overloaded */
    937 
    938 	switch(name[0]) {
    939 	case NFS_NFSSTATS:
    940 		if(!oldp) {
    941 			*oldlenp = sizeof nfsstats;
    942 			return 0;
    943 		}
    944 
    945 		if(*oldlenp < sizeof nfsstats) {
    946 			*oldlenp = sizeof nfsstats;
    947 			return ENOMEM;
    948 		}
    949 
    950 		rv = copyout(&nfsstats, oldp, sizeof nfsstats);
    951 		if(rv) return rv;
    952 
    953 		if(newp && newlen != sizeof nfsstats)
    954 			return EINVAL;
    955 
    956 		if(newp) {
    957 			return copyin(newp, &nfsstats, sizeof nfsstats);
    958 		}
    959 		return 0;
    960 
    961 	case NFS_IOTHREADS:
    962 		nfs_getset_niothreads(0);
    963 
    964 		rv = (sysctl_int(oldp, oldlenp, newp, newlen,
    965 		    &nfs_niothreads));
    966 
    967 		if (newp)
    968 			nfs_getset_niothreads(1);
    969 
    970 		return rv;
    971 
    972 	default:
    973 		return EOPNOTSUPP;
    974 	}
    975 }
    976 
    977 
    978 /*
    979  * At this point, this should never happen
    980  */
    981 /* ARGSUSED */
    982 int
    983 nfs_fhtovp(mp, fhp, vpp)
    984 	struct mount *mp;
    985 	struct fid *fhp;
    986 	struct vnode **vpp;
    987 {
    988 
    989 	return (EINVAL);
    990 }
    991 
    992 /* ARGSUSED */
    993 int
    994 nfs_checkexp(mp, nam, exflagsp, credanonp)
    995 	struct mount *mp;
    996 	struct mbuf *nam;
    997 	int *exflagsp;
    998 	struct ucred **credanonp;
    999 {
   1000 
   1001 	return (EINVAL);
   1002 }
   1003 
   1004 /*
   1005  * Vnode pointer to File handle, should never happen either
   1006  */
   1007 /* ARGSUSED */
   1008 int
   1009 nfs_vptofh(vp, fhp)
   1010 	struct vnode *vp;
   1011 	struct fid *fhp;
   1012 {
   1013 
   1014 	return (EINVAL);
   1015 }
   1016 
   1017 /*
   1018  * Vfs start routine, a no-op.
   1019  */
   1020 /* ARGSUSED */
   1021 int
   1022 nfs_start(mp, flags, p)
   1023 	struct mount *mp;
   1024 	int flags;
   1025 	struct proc *p;
   1026 {
   1027 
   1028 	return (0);
   1029 }
   1030 
   1031 /*
   1032  * Do operations associated with quotas, not supported
   1033  */
   1034 /* ARGSUSED */
   1035 int
   1036 nfs_quotactl(mp, cmd, uid, arg, p)
   1037 	struct mount *mp;
   1038 	int cmd;
   1039 	uid_t uid;
   1040 	caddr_t arg;
   1041 	struct proc *p;
   1042 {
   1043 
   1044 	return (EOPNOTSUPP);
   1045 }
   1046