/* * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Rick Macklem at The University of Guelph. * * %sccs.include.redist.c% * * @(#)nfs_vfsops.c 8.4 (Berkeley) 08/18/94 */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * nfs vfs operations. */ struct vfsops nfs_vfsops = { nfs_mount, nfs_start, nfs_unmount, nfs_root, nfs_quotactl, nfs_statfs, nfs_sync, nfs_vget, nfs_fhtovp, nfs_vptofh, nfs_init, }; /* * This structure must be filled in by a primary bootstrap or bootstrap * server for a diskless/dataless machine. It is initialized below just * to ensure that it is allocated to initialized data (.data not .bss). */ struct nfs_diskless nfs_diskless = { 0 }; extern u_long nfs_procids[NFS_NPROCS]; extern u_long nfs_prog, nfs_vers; void nfs_disconnect __P((struct nfsmount *)); void nfsargs_ntoh __P((struct nfs_args *)); static struct mount *nfs_mountdiskless __P((char *, char *, int, struct sockaddr_in *, struct nfs_args *, register struct vnode **)); #define TRUE 1 #define FALSE 0 /* * nfs statfs call */ int nfs_statfs(mp, sbp, p) struct mount *mp; register struct statfs *sbp; struct proc *p; { register struct vnode *vp; register struct nfsv2_statfs *sfp; register caddr_t cp; register long t1; caddr_t bpos, dpos, cp2; int error = 0, isnq; struct mbuf *mreq, *mrep, *md, *mb, *mb2; struct nfsmount *nmp; struct ucred *cred; struct nfsnode *np; nmp = VFSTONFS(mp); isnq = (nmp->nm_flag & NFSMNT_NQNFS); if (error = nfs_nget(mp, &nmp->nm_fh, &np)) return (error); vp = NFSTOV(np); nfsstats.rpccnt[NFSPROC_STATFS]++; cred = crget(); cred->cr_ngroups = 1; nfsm_reqhead(vp, NFSPROC_STATFS, NFSX_FH); nfsm_fhtom(vp); nfsm_request(vp, NFSPROC_STATFS, p, cred); nfsm_dissect(sfp, struct nfsv2_statfs *, NFSX_STATFS(isnq)); sbp->f_type = MOUNT_NFS; sbp->f_flags = nmp->nm_flag; sbp->f_iosize = NFS_MAXDGRAMDATA; sbp->f_bsize = fxdr_unsigned(long, sfp->sf_bsize); sbp->f_blocks = fxdr_unsigned(long, sfp->sf_blocks); sbp->f_bfree = fxdr_unsigned(long, sfp->sf_bfree); sbp->f_bavail = fxdr_unsigned(long, sfp->sf_bavail); if (isnq) { sbp->f_files = fxdr_unsigned(long, sfp->sf_files); sbp->f_ffree = fxdr_unsigned(long, sfp->sf_ffree); } else { sbp->f_files = 0; sbp->f_ffree = 0; } if (sbp != &mp->mnt_stat) { bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN); bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN); } nfsm_reqdone; vrele(vp); crfree(cred); return (error); } /* * Mount a remote root fs via. nfs. This depends on the info in the * nfs_diskless structure that has been filled in properly by some primary * bootstrap. * It goes something like this: * - do enough of "ifconfig" by calling ifioctl() so that the system * can talk to the server * - If nfs_diskless.mygateway is filled in, use that address as * a default gateway. * - hand craft the swap nfs vnode hanging off a fake mount point * if swdevt[0].sw_dev == NODEV * - build the rootfs mount point and call mountnfs() to do the rest. */ int nfs_mountroot() { register struct mount *mp; register struct nfs_diskless *nd = &nfs_diskless; struct socket *so; struct vnode *vp; struct proc *p = curproc; /* XXX */ int error, i; /* * XXX time must be non-zero when we init the interface or else * the arp code will wedge... */ if (time.tv_sec == 0) time.tv_sec = 1; #ifdef notyet /* Set up swap credentials. */ proc0.p_ucred->cr_uid = ntohl(nd->swap_ucred.cr_uid); proc0.p_ucred->cr_gid = ntohl(nd->swap_ucred.cr_gid); if ((proc0.p_ucred->cr_ngroups = ntohs(nd->swap_ucred.cr_ngroups)) > NGROUPS) proc0.p_ucred->cr_ngroups = NGROUPS; for (i = 0; i < proc0.p_ucred->cr_ngroups; i++) proc0.p_ucred->cr_groups[i] = ntohl(nd->swap_ucred.cr_groups[i]); #endif /* * Do enough of ifconfig(8) so that the critical net interface can * talk to the server. */ if (error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0)) panic("nfs_mountroot: socreate: %d", error); if (error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, p)) panic("nfs_mountroot: SIOCAIFADDR: %d", error); soclose(so); /* * If the gateway field is filled in, set it as the default route. */ if (nd->mygateway.sin_len != 0) { struct sockaddr_in mask, sin; bzero((caddr_t)&mask, sizeof(mask)); sin = mask; sin.sin_family = AF_INET; sin.sin_len = sizeof(sin); if (error = rtrequest(RTM_ADD, (struct sockaddr *)&sin, (struct sockaddr *)&nd->mygateway, (struct sockaddr *)&mask, RTF_UP | RTF_GATEWAY, (struct rtentry **)0)) panic("nfs_mountroot: RTM_ADD: %d", error); } /* * If swapping to an nfs node (indicated by swdevt[0].sw_dev == NODEV): * Create a fake mount point just for the swap vnode so that the * swap file can be on a different server from the rootfs. */ if (swdevt[0].sw_dev == NODEV) { nd->swap_args.fh = (nfsv2fh_t *)nd->swap_fh; (void) nfs_mountdiskless(nd->swap_hostnam, "/swap", 0, &nd->swap_saddr, &nd->swap_args, &vp); /* * Since the swap file is not the root dir of a file system, * hack it to a regular file. */ vp->v_type = VREG; vp->v_flag = 0; swapdev_vp = vp; VREF(vp); swdevt[0].sw_vp = vp; swdevt[0].sw_nblks = ntohl(nd->swap_nblks); } else if (bdevvp(swapdev, &swapdev_vp)) panic("nfs_mountroot: can't setup swapdev_vp"); /* * Create the rootfs mount point. */ nd->root_args.fh = (nfsv2fh_t *)nd->root_fh; mp = nfs_mountdiskless(nd->root_hostnam, "/", MNT_RDONLY, &nd->root_saddr, &nd->root_args, &vp); if (vfs_lock(mp)) panic("nfs_mountroot: vfs_lock"); TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); mp->mnt_flag |= MNT_ROOTFS; mp->mnt_vnodecovered = NULLVP; vfs_unlock(mp); rootvp = vp; /* * This is not really an nfs issue, but it is much easier to * set hostname here and then let the "/etc/rc.xxx" files * mount the right /var based upon its preset value. */ bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN); hostname[MAXHOSTNAMELEN - 1] = '\0'; for (i = 0; i < MAXHOSTNAMELEN; i++) if (hostname[i] == '\0') break; hostnamelen = i; inittodr(ntohl(nd->root_time)); return (0); } /* * Internal version of mount system call for diskless setup. */ static struct mount * nfs_mountdiskless(path, which, mountflag, sin, args, vpp) char *path; char *which; int mountflag; struct sockaddr_in *sin; struct nfs_args *args; register struct vnode **vpp; { register struct mount *mp; register struct mbuf *m; register int error; mp = (struct mount *)malloc((u_long)sizeof(struct mount), M_MOUNT, M_NOWAIT); if (mp == NULL) panic("nfs_mountroot: %s mount malloc", which); bzero((char *)mp, (u_long)sizeof(struct mount)); mp->mnt_op = &nfs_vfsops; mp->mnt_flag = mountflag; MGET(m, MT_SONAME, M_DONTWAIT); if (m == NULL) panic("nfs_mountroot: %s mount mbuf", which); bcopy((caddr_t)sin, mtod(m, caddr_t), sin->sin_len); m->m_len = sin->sin_len; nfsargs_ntoh(args); if (error = mountnfs(args, mp, m, which, path, vpp)) panic("nfs_mountroot: mount %s on %s: %d", path, which, error); return (mp); } /* * Convert the integer fields of the nfs_args structure from net byte order * to host byte order. Called by nfs_mountroot() above. */ void nfsargs_ntoh(nfsp) register struct nfs_args *nfsp; { NTOHL(nfsp->sotype); NTOHL(nfsp->proto); NTOHL(nfsp->flags); NTOHL(nfsp->wsize); NTOHL(nfsp->rsize); NTOHL(nfsp->timeo); NTOHL(nfsp->retrans); NTOHL(nfsp->maxgrouplist); NTOHL(nfsp->readahead); NTOHL(nfsp->leaseterm); NTOHL(nfsp->deadthresh); } /* * VFS Operations. * * mount system call * It seems a bit dumb to copyinstr() the host and path here and then * bcopy() them in mountnfs(), but I wanted to detect errors before * doing the sockargs() call because sockargs() allocates an mbuf and * an error after that means that I have to release the mbuf. */ /* ARGSUSED */ int nfs_mount(mp, path, data, ndp, p) struct mount *mp; char *path; caddr_t data; struct nameidata *ndp; struct proc *p; { int error; struct nfs_args args; struct mbuf *nam; struct vnode *vp; char pth[MNAMELEN], hst[MNAMELEN]; u_int len; nfsv2fh_t nfh; if (error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args))) return (error); if (error = copyin((caddr_t)args.fh, (caddr_t)&nfh, sizeof (nfsv2fh_t))) return (error); if (error = copyinstr(path, pth, MNAMELEN-1, &len)) return (error); bzero(&pth[len], MNAMELEN - len); if (error = copyinstr(args.hostname, hst, MNAMELEN-1, &len)) return (error); bzero(&hst[len], MNAMELEN - len); /* sockargs() call must be after above copyin() calls */ if (error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME)) return (error); args.fh = &nfh; error = mountnfs(&args, mp, nam, pth, hst, &vp); return (error); } /* * Common code for mount and mountroot */ int mountnfs(argp, mp, nam, pth, hst, vpp) register struct nfs_args *argp; register struct mount *mp; struct mbuf *nam; char *pth, *hst; struct vnode **vpp; { register struct nfsmount *nmp; struct nfsnode *np; int error; if (mp->mnt_flag & MNT_UPDATE) { nmp = VFSTONFS(mp); /* update paths, file handles, etc, here XXX */ m_freem(nam); return (0); } else { MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount), M_NFSMNT, M_WAITOK); bzero((caddr_t)nmp, sizeof (struct nfsmount)); mp->mnt_data = (qaddr_t)nmp; } getnewfsid(mp, MOUNT_NFS); nmp->nm_mountp = mp; nmp->nm_flag = argp->flags; if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_MYWRITE)) == (NFSMNT_NQNFS | NFSMNT_MYWRITE)) { error = EPERM; goto bad; } if (nmp->nm_flag & NFSMNT_NQNFS) /* * We have to set mnt_maxsymlink to a non-zero value so * that COMPAT_43 routines will know that we are setting * the d_type field in directories (and can zero it for * unsuspecting binaries). */ mp->mnt_maxsymlinklen = 1; nmp->nm_timeo = NFS_TIMEO; nmp->nm_retry = NFS_RETRANS; nmp->nm_wsize = NFS_WSIZE; nmp->nm_rsize = NFS_RSIZE; nmp->nm_numgrps = NFS_MAXGRPS; nmp->nm_readahead = NFS_DEFRAHEAD; nmp->nm_leaseterm = NQ_DEFLEASE; nmp->nm_deadthresh = NQ_DEADTHRESH; CIRCLEQ_INIT(&nmp->nm_timerhead); nmp->nm_inprog = NULLVP; bcopy((caddr_t)argp->fh, (caddr_t)&nmp->nm_fh, sizeof(nfsv2fh_t)); mp->mnt_stat.f_type = MOUNT_NFS; bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN); bcopy(pth, mp->mnt_stat.f_mntonname, MNAMELEN); nmp->nm_nam = nam; if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; if (nmp->nm_timeo < NFS_MINTIMEO) nmp->nm_timeo = NFS_MINTIMEO; else if (nmp->nm_timeo > NFS_MAXTIMEO) nmp->nm_timeo = NFS_MAXTIMEO; } if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { nmp->nm_retry = argp->retrans; if (nmp->nm_retry > NFS_MAXREXMIT) nmp->nm_retry = NFS_MAXREXMIT; } if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { nmp->nm_wsize = argp->wsize; /* Round down to multiple of blocksize */ nmp->nm_wsize &= ~0x1ff; if (nmp->nm_wsize <= 0) nmp->nm_wsize = 512; else if (nmp->nm_wsize > NFS_MAXDATA) nmp->nm_wsize = NFS_MAXDATA; } if (nmp->nm_wsize > MAXBSIZE) nmp->nm_wsize = MAXBSIZE; if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { nmp->nm_rsize = argp->rsize; /* Round down to multiple of blocksize */ nmp->nm_rsize &= ~0x1ff; if (nmp->nm_rsize <= 0) nmp->nm_rsize = 512; else if (nmp->nm_rsize > NFS_MAXDATA) nmp->nm_rsize = NFS_MAXDATA; } if (nmp->nm_rsize > MAXBSIZE) nmp->nm_rsize = MAXBSIZE; if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 && argp->maxgrouplist <= NFS_MAXGRPS) nmp->nm_numgrps = argp->maxgrouplist; if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 && argp->readahead <= NFS_MAXRAHEAD) nmp->nm_readahead = argp->readahead; if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 && argp->leaseterm <= NQ_MAXLEASE) nmp->nm_leaseterm = argp->leaseterm; if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 && argp->deadthresh <= NQ_NEVERDEAD) nmp->nm_deadthresh = argp->deadthresh; /* Set up the sockets and per-host congestion */ nmp->nm_sotype = argp->sotype; nmp->nm_soproto = argp->proto; /* * For Connection based sockets (TCP,...) defer the connect until * the first request, in case the server is not responding. */ if (nmp->nm_sotype == SOCK_DGRAM && (error = nfs_connect(nmp, (struct nfsreq *)0))) goto bad; /* * This is silly, but it has to be set so that vinifod() works. * We do not want to do an nfs_statfs() here since we can get * stuck on a dead server and we are holding a lock on the mount * point. */ mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA; /* * A reference count is needed on the nfsnode representing the * remote root. If this object is not persistent, then backward * traversals of the mount point (i.e. "..") will not work if * the nfsnode gets flushed out of the cache. Ufs does not have * this problem, because one can identify root inodes by their * number == ROOTINO (2). */ if (error = nfs_nget(mp, &nmp->nm_fh, &np)) goto bad; *vpp = NFSTOV(np); return (0); bad: nfs_disconnect(nmp); free((caddr_t)nmp, M_NFSMNT); m_freem(nam); return (error); } /* * unmount system call */ int nfs_unmount(mp, mntflags, p) struct mount *mp; int mntflags; struct proc *p; { register struct nfsmount *nmp; struct nfsnode *np; struct vnode *vp; int error, flags = 0; extern int doforce; if (mntflags & MNT_FORCE) { if (!doforce || (mp->mnt_flag & MNT_ROOTFS)) return (EINVAL); flags |= FORCECLOSE; } nmp = VFSTONFS(mp); /* * Goes something like this.. * - Check for activity on the root vnode (other than ourselves). * - Call vflush() to clear out vnodes for this file system, * except for the root vnode. * - Decrement reference on the vnode representing remote root. * - Close the socket * - Free up the data structures */ /* * We need to decrement the ref. count on the nfsnode representing * the remote root. See comment in mountnfs(). The VFS unmount() * has done vput on this vnode, otherwise we would get deadlock! */ if (error = nfs_nget(mp, &nmp->nm_fh, &np)) return(error); vp = NFSTOV(np); if (vp->v_usecount > 2) { vput(vp); return (EBUSY); } /* * Must handshake with nqnfs_clientd() if it is active. */ nmp->nm_flag |= NFSMNT_DISMINPROG; while (nmp->nm_inprog != NULLVP) (void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0); if (error = vflush(mp, vp, flags)) { vput(vp); nmp->nm_flag &= ~NFSMNT_DISMINPROG; return (error); } /* * We are now committed to the unmount. * For NQNFS, let the server daemon free the nfsmount structure. */ if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) nmp->nm_flag |= NFSMNT_DISMNT; /* * There are two reference counts to get rid of here. */ vrele(vp); vrele(vp); vgone(vp); nfs_disconnect(nmp); m_freem(nmp->nm_nam); if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0) free((caddr_t)nmp, M_NFSMNT); return (0); } /* * Return root of a filesystem */ int nfs_root(mp, vpp) struct mount *mp; struct vnode **vpp; { register struct vnode *vp; struct nfsmount *nmp; struct nfsnode *np; int error; nmp = VFSTONFS(mp); if (error = nfs_nget(mp, &nmp->nm_fh, &np)) return (error); vp = NFSTOV(np); vp->v_type = VDIR; vp->v_flag = VROOT; *vpp = vp; return (0); } extern int syncprt; /* * Flush out the buffer cache */ /* ARGSUSED */ int nfs_sync(mp, waitfor, cred, p) struct mount *mp; int waitfor; struct ucred *cred; struct proc *p; { register struct vnode *vp; int error, allerror = 0; /* * Force stale buffer cache information to be flushed. */ loop: for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = vp->v_mntvnodes.le_next) { /* * If the vnode that we are about to sync is no longer * associated with this mount point, start over. */ if (vp->v_mount != mp) goto loop; if (VOP_ISLOCKED(vp) || vp->v_dirtyblkhd.lh_first == NULL) continue; if (vget(vp, 1)) goto loop; if (error = VOP_FSYNC(vp, cred, waitfor, p)) allerror = error; vput(vp); } return (allerror); } /* * NFS flat namespace lookup. * Currently unsupported. */ /* ARGSUSED */ int nfs_vget(mp, ino, vpp) struct mount *mp; ino_t ino; struct vnode **vpp; { return (EOPNOTSUPP); } /* * At this point, this should never happen */ /* ARGSUSED */ int nfs_fhtovp(mp, fhp, nam, vpp, exflagsp, credanonp) register struct mount *mp; struct fid *fhp; struct mbuf *nam; struct vnode **vpp; int *exflagsp; struct ucred **credanonp; { return (EINVAL); } /* * Vnode pointer to File handle, should never happen either */ /* ARGSUSED */ int nfs_vptofh(vp, fhp) struct vnode *vp; struct fid *fhp; { return (EINVAL); } /* * Vfs start routine, a no-op. */ /* ARGSUSED */ int nfs_start(mp, flags, p) struct mount *mp; int flags; struct proc *p; { return (0); } /* * Do operations associated with quotas, not supported */ /* ARGSUSED */ int nfs_quotactl(mp, cmd, uid, arg, p) struct mount *mp; int cmd; uid_t uid; caddr_t arg; struct proc *p; { return (EOPNOTSUPP); }