nfs_socket.c

早期freebsd实现

			mlen = hsiz;
	
		/*
		 * Loop through the mbuf list consolidating data.
		 */
		while (m) {
			while (olen > 0) {
				if (mlen == 0) {
					m2->m_flags &= ~M_PKTHDR;
					if (m2->m_flags & M_EXT)
						m2->m_data = m2->m_ext.ext_buf;
					else
						m2->m_data = m2->m_dat;
					m2->m_len = 0;
					mlen = M_TRAILINGSPACE(m2);
					tcp = mtod(m2, caddr_t);
					mnew = m2;
					m2 = m2->m_next;
				}
				siz = min(mlen, olen);
				if (tcp != fcp)
					bcopy(fcp, tcp, siz);
				mnew->m_len += siz;
				mlen -= siz;
				olen -= siz;
				tcp += siz;
				fcp += siz;
			}
			m = m->m_next;
			if (m) {
				olen = m->m_len;
				fcp = mtod(m, caddr_t);
			}
		}
	
		/*
		 * Finally, set m_len == 0 for any trailing mbufs that have
		 * been copied out of.
		 */
		while (m2) {
			m2->m_len = 0;
			m2 = m2->m_next;
		}
		return;
	    }
	    m = m->m_next;
	}
}

/*
 * Socket upcall routine for the nfsd sockets.
 * The caddr_t arg is a pointer to the "struct nfssvc_sock".
 * Essentially do as much as possible non-blocking, else punt and it will
 * be called with M_WAIT from an nfsd.
 */
void
nfsrv_rcv(so, arg, waitflag)
	struct socket *so;
	caddr_t arg;
	int waitflag;
{
	register struct nfssvc_sock *slp = (struct nfssvc_sock *)arg;
	register struct mbuf *m;
	struct mbuf *mp, *nam;
	struct uio auio;
	int flags, error;

	if ((slp->ns_flag & SLP_VALID) == 0)
		return;
#ifdef notdef
	/*
	 * Define this to test for nfsds handling this under heavy load.
	 */
	if (waitflag == M_DONTWAIT) {
		slp->ns_flag |= SLP_NEEDQ; goto dorecs;
	}
#endif
	auio.uio_procp = NULL;
	if (so->so_type == SOCK_STREAM) {
		/*
		 * If there are already records on the queue, defer soreceive()
		 * to an nfsd so that there is feedback to the TCP layer that
		 * the nfs servers are heavily loaded.
		 */
		if (slp->ns_rec && waitflag == M_DONTWAIT) {
			slp->ns_flag |= SLP_NEEDQ;
			goto dorecs;
		}

		/*
		 * Do soreceive().
		 */
		auio.uio_resid = 1000000000;
		flags = MSG_DONTWAIT;
		error = soreceive(so, &nam, &auio, &mp, (struct mbuf **)0, &flags);
		if (error || mp == (struct mbuf *)0) {
			if (error == EWOULDBLOCK)
				slp->ns_flag |= SLP_NEEDQ;
			else
				slp->ns_flag |= SLP_DISCONN;
			goto dorecs;
		}
		m = mp;
		if (slp->ns_rawend) {
			slp->ns_rawend->m_next = m;
			slp->ns_cc += 1000000000 - auio.uio_resid;
		} else {
			slp->ns_raw = m;
			slp->ns_cc = 1000000000 - auio.uio_resid;
		}
		while (m->m_next)
			m = m->m_next;
		slp->ns_rawend = m;

		/*
		 * Now try and parse record(s) out of the raw stream data.
		 */
		if (error = nfsrv_getstream(slp, waitflag)) {
			if (error == EPERM)
				slp->ns_flag |= SLP_DISCONN;
			else
				slp->ns_flag |= SLP_NEEDQ;
		}
	} else {
		do {
			auio.uio_resid = 1000000000;
			flags = MSG_DONTWAIT;
			error = soreceive(so, &nam, &auio, &mp,
						(struct mbuf **)0, &flags);
			if (mp) {
				nfs_realign(mp, 10 * NFSX_UNSIGNED);
				if (nam) {
					m = nam;
					m->m_next = mp;
				} else
					m = mp;
				if (slp->ns_recend)
					slp->ns_recend->m_nextpkt = m;
				else
					slp->ns_rec = m;
				slp->ns_recend = m;
				m->m_nextpkt = (struct mbuf *)0;
			}
			if (error) {
				if ((so->so_proto->pr_flags & PR_CONNREQUIRED)
					&& error != EWOULDBLOCK) {
					slp->ns_flag |= SLP_DISCONN;
					goto dorecs;
				}
			}
		} while (mp);
	}

	/*
	 * Now try and process the request records, non-blocking.
	 */
dorecs:
	if (waitflag == M_DONTWAIT &&
		(slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))))
		nfsrv_wakenfsd(slp);
}

/*
 * Try and extract an RPC request from the mbuf data list received on a
 * stream socket. The "waitflag" argument indicates whether or not it
 * can sleep.
 */
nfsrv_getstream(slp, waitflag)
	register struct nfssvc_sock *slp;
	int waitflag;
{
	register struct mbuf *m;
	register char *cp1, *cp2;
	register int len;
	struct mbuf *om, *m2, *recm;
	u_long recmark;

	if (slp->ns_flag & SLP_GETSTREAM)
		panic("nfs getstream");
	slp->ns_flag |= SLP_GETSTREAM;
	for (;;) {
	    if (slp->ns_reclen == 0) {
		if (slp->ns_cc < NFSX_UNSIGNED) {
			slp->ns_flag &= ~SLP_GETSTREAM;
			return (0);
		}
		m = slp->ns_raw;
		if (m->m_len >= NFSX_UNSIGNED) {
			bcopy(mtod(m, caddr_t), (caddr_t)&recmark, NFSX_UNSIGNED);
			m->m_data += NFSX_UNSIGNED;
			m->m_len -= NFSX_UNSIGNED;
		} else {
			cp1 = (caddr_t)&recmark;
			cp2 = mtod(m, caddr_t);
			while (cp1 < ((caddr_t)&recmark) + NFSX_UNSIGNED) {
				while (m->m_len == 0) {
					m = m->m_next;
					cp2 = mtod(m, caddr_t);
				}
				*cp1++ = *cp2++;
				m->m_data++;
				m->m_len--;
			}
		}
		slp->ns_cc -= NFSX_UNSIGNED;
		slp->ns_reclen = ntohl(recmark) & ~0x80000000;
		if (slp->ns_reclen < NFS_MINPACKET || slp->ns_reclen > NFS_MAXPACKET) {
			slp->ns_flag &= ~SLP_GETSTREAM;
			return (EPERM);
		}
	    }

	    /*
	     * Now get the record part.
	     */
	    if (slp->ns_cc == slp->ns_reclen) {
		recm = slp->ns_raw;
		slp->ns_raw = slp->ns_rawend = (struct mbuf *)0;
		slp->ns_cc = slp->ns_reclen = 0;
	    } else if (slp->ns_cc > slp->ns_reclen) {
		len = 0;
		m = slp->ns_raw;
		om = (struct mbuf *)0;
		while (len < slp->ns_reclen) {
			if ((len + m->m_len) > slp->ns_reclen) {
				m2 = m_copym(m, 0, slp->ns_reclen - len,
					waitflag);
				if (m2) {
					if (om) {
						om->m_next = m2;
						recm = slp->ns_raw;
					} else
						recm = m2;
					m->m_data += slp->ns_reclen - len;
					m->m_len -= slp->ns_reclen - len;
					len = slp->ns_reclen;
				} else {
					slp->ns_flag &= ~SLP_GETSTREAM;
					return (EWOULDBLOCK);
				}
			} else if ((len + m->m_len) == slp->ns_reclen) {
				om = m;
				len += m->m_len;
				m = m->m_next;
				recm = slp->ns_raw;
				om->m_next = (struct mbuf *)0;
			} else {
				om = m;
				len += m->m_len;
				m = m->m_next;
			}
		}
		slp->ns_raw = m;
		slp->ns_cc -= len;
		slp->ns_reclen = 0;
	    } else {
		slp->ns_flag &= ~SLP_GETSTREAM;
		return (0);
	    }
	    nfs_realign(recm, 10 * NFSX_UNSIGNED);
	    if (slp->ns_recend)
		slp->ns_recend->m_nextpkt = recm;
	    else
		slp->ns_rec = recm;
	    slp->ns_recend = recm;
	}
}

/*
 * Parse an RPC header.
 */
nfsrv_dorec(slp, nd)
	register struct nfssvc_sock *slp;
	register struct nfsd *nd;
{
	register struct mbuf *m;
	int error;

	if ((slp->ns_flag & SLP_VALID) == 0 ||
	    (m = slp->ns_rec) == (struct mbuf *)0)
		return (ENOBUFS);
	if (slp->ns_rec = m->m_nextpkt)
		m->m_nextpkt = (struct mbuf *)0;
	else
		slp->ns_recend = (struct mbuf *)0;
	if (m->m_type == MT_SONAME) {
		nd->nd_nam = m;
		nd->nd_md = nd->nd_mrep = m->m_next;
		m->m_next = (struct mbuf *)0;
	} else {
		nd->nd_nam = (struct mbuf *)0;
		nd->nd_md = nd->nd_mrep = m;
	}
	nd->nd_dpos = mtod(nd->nd_md, caddr_t);
	if (error = nfs_getreq(nd, TRUE)) {
		m_freem(nd->nd_nam);
		return (error);
	}
	return (0);
}

/*
 * Parse an RPC request
 * - verify it
 * - fill in the cred struct.
 */
nfs_getreq(nd, has_header)
	register struct nfsd *nd;
	int has_header;
{
	register int len, i;
	register u_long *tl;
	register long t1;
	struct uio uio;
	struct iovec iov;
	caddr_t dpos, cp2;
	u_long nfsvers, auth_type;
	int error = 0, nqnfs = 0;
	struct mbuf *mrep, *md;

	mrep = nd->nd_mrep;
	md = nd->nd_md;
	dpos = nd->nd_dpos;
	if (has_header) {
		nfsm_dissect(tl, u_long *, 10*NFSX_UNSIGNED);
		nd->nd_retxid = *tl++;
		if (*tl++ != rpc_call) {
			m_freem(mrep);
			return (EBADRPC);
		}
	} else {
		nfsm_dissect(tl, u_long *, 8*NFSX_UNSIGNED);
	}
	nd->nd_repstat = 0;
	if (*tl++ != rpc_vers) {
		nd->nd_repstat = ERPCMISMATCH;
		nd->nd_procnum = NFSPROC_NOOP;
		return (0);
	}
	nfsvers = nfs_vers;
	if (*tl != nfs_prog) {
		if (*tl == nqnfs_prog) {
			nqnfs++;
			nfsvers = nqnfs_vers;
		} else {
			nd->nd_repstat = EPROGUNAVAIL;
			nd->nd_procnum = NFSPROC_NOOP;
			return (0);
		}
	}
	tl++;
	if (*tl++ != nfsvers) {
		nd->nd_repstat = EPROGMISMATCH;
		nd->nd_procnum = NFSPROC_NOOP;
		return (0);
	}
	nd->nd_procnum = fxdr_unsigned(u_long, *tl++);
	if (nd->nd_procnum == NFSPROC_NULL)
		return (0);
	if (nd->nd_procnum >= NFS_NPROCS ||
		(!nqnfs && nd->nd_procnum > NFSPROC_STATFS) ||
		(*tl != rpc_auth_unix && *tl != rpc_auth_kerb)) {
		nd->nd_repstat = EPROCUNAVAIL;
		nd->nd_procnum = NFSPROC_NOOP;
		return (0);
	}
	auth_type = *tl++;
	len = fxdr_unsigned(int, *tl++);
	if (len < 0 || len > RPCAUTH_MAXSIZ) {
		m_freem(mrep);
		return (EBADRPC);
	}

	/*
	 * Handle auth_unix or auth_kerb.
	 */
	if (auth_type == rpc_auth_unix) {
		len = fxdr_unsigned(int, *++tl);
		if (len < 0 || len > NFS_MAXNAMLEN) {
			m_freem(mrep);
			return (EBADRPC);
		}
		nfsm_adv(nfsm_rndup(len));
		nfsm_dissect(tl, u_long *, 3*NFSX_UNSIGNED);
		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
		nd->nd_cr.cr_gid = fxdr_unsigned(gid_t, *tl++);
		len = fxdr_unsigned(int, *tl);
		if (len < 0 || len > RPCAUTH_UNIXGIDS) {
			m_freem(mrep);
			return (EBADRPC);
		}
		nfsm_dissect(tl, u_long *, (len + 2)*NFSX_UNSIGNED);
		for (i = 1; i <= len; i++)
			if (i < NGROUPS)
				nd->nd_cr.cr_groups[i] = fxdr_unsigned(gid_t, *tl++);
			else
				tl++;
		nd->nd_cr.cr_ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1);
	} else if (auth_type == rpc_auth_kerb) {
		nd->nd_cr.cr_uid = fxdr_unsigned(uid_t, *tl++);
		nd->nd_authlen = fxdr_unsigned(int, *tl);
		uio.uio_resid = nfsm_rndup(nd->nd_authlen);
		if (uio.uio_resid > (len - 2 * NFSX_UNSIGNED)) {
			m_freem(mrep);
			return (EBADRPC);
		}
		uio.uio_offset = 0;
		uio.uio_iov = &iov;
		uio.uio_iovcnt = 1;
		uio.uio_segflg = UIO_SYSSPACE;
		iov.iov_base = (caddr_t)nd->nd_authstr;
		iov.iov_len = RPCAUTH_MAXSIZ;
		nfsm_mtouio(&uio, uio.uio_resid);
		nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED);
		nd->nd_flag |= NFSD_NEEDAUTH;
	}

	/*
	 * Do we have any use for the verifier.
	 * According to the "Remote Procedure Call Protocol Spec." it
	 * should be AUTH_NULL, but some clients make it AUTH_UNIX?
	 * For now, just skip over it
	 */
	len = fxdr_unsigned(int, *++tl);
	if (len < 0 || len > RPCAUTH_MAXSIZ) {
		m_freem(mrep);
		return (EBADRPC);
	}
	if (len > 0) {
		nfsm_adv(nfsm_rndup(len));
	}

	/*
	 * For nqnfs, get piggybacked lease request.
	 */
	if (nqnfs && nd->nd_procnum != NQNFSPROC_EVICTED) {
		nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
		nd->nd_nqlflag = fxdr_unsigned(int, *tl);
		if (nd->nd_nqlflag) {
			nfsm_dissect(tl, u_long *, NFSX_UNSIGNED);
			nd->nd_duration = fxdr_unsigned(int, *tl);
		} else
			nd->nd_duration = NQ_MINLEASE;
	} else {
		nd->nd_nqlflag = NQL_NOVAL;
		nd->nd_duration = NQ_MINLEASE;
	}
	nd->nd_md = md;
	nd->nd_dpos = dpos;
	return (0);
nfsmout:
	return (error);
}

/*
 * Search for a sleeping nfsd and wake it up.
 * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the
 * running nfsds will go look for the work in the nfssvc_sock list.
 */
void
nfsrv_wakenfsd(slp)
	struct nfssvc_sock *slp;
{
	register struct nfsd *nd = nfsd_head.nd_next;

	if ((slp->ns_flag & SLP_VALID) == 0)
		return;
	while (nd != (struct nfsd *)&nfsd_head) {
		if (nd->nd_flag & NFSD_WAITING) {
			nd->nd_flag &= ~NFSD_WAITING;
			if (nd->nd_slp)
				panic("nfsd wakeup");
			slp->ns_sref++;
			nd->nd_slp = slp;
			wakeup((caddr_t)nd);
			return;
		}
		nd = nd->nd_next;
	}
	slp->ns_flag |= SLP_DOREC;
	nfsd_head.nd_flag |= NFSD_CHECKSLP;
}

nfs_msg(p, server, msg)
	struct proc *p;
	char *server, *msg;
{
	tpr_t tpr;

	if (p)
		tpr = tprintf_open(p);
	else
		tpr = NULL;
	tprintf(tpr, "nfs server %s: %s\n", server, msg);
	tprintf_close(tpr);
}