예제 #1
0
static int
udp6_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr6,
    struct mbuf *control, struct thread *td)
{
	u_int32_t ulen = m->m_pkthdr.len;
	u_int32_t plen = sizeof(struct udphdr) + ulen;
	struct ip6_hdr *ip6;
	struct udphdr *udp6;
	struct in6_addr *laddr, *faddr, in6a;
	struct sockaddr_in6 *sin6 = NULL;
	int cscov_partial = 0;
	int scope_ambiguous = 0;
	u_short fport;
	int error = 0;
	uint8_t nxt;
	uint16_t cscov = 0;
	struct ip6_pktopts *optp, opt;
	int af = AF_INET6, hlen = sizeof(struct ip6_hdr);
	int flags;
	struct sockaddr_in6 tmp;

	INP_WLOCK_ASSERT(inp);
	INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);

	if (addr6) {
		/* addr6 has been validated in udp6_send(). */
		sin6 = (struct sockaddr_in6 *)addr6;

		/* protect *sin6 from overwrites */
		tmp = *sin6;
		sin6 = &tmp;

		/*
		 * Application should provide a proper zone ID or the use of
		 * default zone IDs should be enabled.  Unfortunately, some
		 * applications do not behave as it should, so we need a
		 * workaround.  Even if an appropriate ID is not determined,
		 * we'll see if we can determine the outgoing interface.  If we
		 * can, determine the zone ID based on the interface below.
		 */
		if (sin6->sin6_scope_id == 0 && !V_ip6_use_defzone)
			scope_ambiguous = 1;
		if ((error = sa6_embedscope(sin6, V_ip6_use_defzone)) != 0)
			return (error);
	}

	nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
	    IPPROTO_UDP : IPPROTO_UDPLITE;
	if (control) {
		if ((error = ip6_setpktopts(control, &opt,
		    inp->in6p_outputopts, td->td_ucred, nxt)) != 0)
			goto release;
		optp = &opt;
	} else
		optp = inp->in6p_outputopts;

	if (sin6) {
		faddr = &sin6->sin6_addr;

		/*
		 * Since we saw no essential reason for calling in_pcbconnect,
		 * we get rid of such kind of logic, and call in6_selectsrc
		 * and in6_pcbsetport in order to fill in the local address
		 * and the local port.
		 */
		if (sin6->sin6_port == 0) {
			error = EADDRNOTAVAIL;
			goto release;
		}

		if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
			/* how about ::ffff:0.0.0.0 case? */
			error = EISCONN;
			goto release;
		}

		fport = sin6->sin6_port; /* allow 0 port */

		if (IN6_IS_ADDR_V4MAPPED(faddr)) {
			if ((inp->inp_flags & IN6P_IPV6_V6ONLY)) {
				/*
				 * I believe we should explicitly discard the
				 * packet when mapped addresses are disabled,
				 * rather than send the packet as an IPv6 one.
				 * If we chose the latter approach, the packet
				 * might be sent out on the wire based on the
				 * default route, the situation which we'd
				 * probably want to avoid.
				 * (20010421 [email protected])
				 */
				error = EINVAL;
				goto release;
			}
			if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) &&
			    !IN6_IS_ADDR_V4MAPPED(&inp->in6p_laddr)) {
				/*
				 * when remote addr is an IPv4-mapped address,
				 * local addr should not be an IPv6 address,
				 * since you cannot determine how to map IPv6
				 * source address to IPv4.
				 */
				error = EINVAL;
				goto release;
			}

			af = AF_INET;
		}

		if (!IN6_IS_ADDR_V4MAPPED(faddr)) {
			error = in6_selectsrc_socket(sin6, optp, inp,
			    td->td_ucred, scope_ambiguous, &in6a, NULL);
			if (error)
				goto release;
			laddr = &in6a;
		} else
			laddr = &inp->in6p_laddr;	/* XXX */
		if (laddr == NULL) {
			if (error == 0)
				error = EADDRNOTAVAIL;
			goto release;
		}
		if (inp->inp_lport == 0 &&
		    (error = in6_pcbsetport(laddr, inp, td->td_ucred)) != 0) {
			/* Undo an address bind that may have occurred. */
			inp->in6p_laddr = in6addr_any;
			goto release;
		}
	} else {
		if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
			error = ENOTCONN;
			goto release;
		}
		if (IN6_IS_ADDR_V4MAPPED(&inp->in6p_faddr)) {
			if ((inp->inp_flags & IN6P_IPV6_V6ONLY)) {
				/*
				 * XXX: this case would happen when the
				 * application sets the V6ONLY flag after
				 * connecting the foreign address.
				 * Such applications should be fixed,
				 * so we bark here.
				 */
				log(LOG_INFO, "udp6_output: IPV6_V6ONLY "
				    "option was set for a connected socket\n");
				error = EINVAL;
				goto release;
			} else
				af = AF_INET;
		}
		laddr = &inp->in6p_laddr;
		faddr = &inp->in6p_faddr;
		fport = inp->inp_fport;
	}

	if (af == AF_INET)
		hlen = sizeof(struct ip);

	/*
	 * Calculate data length and get a mbuf
	 * for UDP and IP6 headers.
	 */
	M_PREPEND(m, hlen + sizeof(struct udphdr), M_NOWAIT);
	if (m == NULL) {
		error = ENOBUFS;
		goto release;
	}

	/*
	 * Stuff checksum and output datagram.
	 */
	udp6 = (struct udphdr *)(mtod(m, caddr_t) + hlen);
	udp6->uh_sport = inp->inp_lport; /* lport is always set in the PCB */
	udp6->uh_dport = fport;
	if (nxt == IPPROTO_UDPLITE) {
		struct udpcb *up;

		up = intoudpcb(inp);
		cscov = up->u_txcslen;
		if (cscov >= plen)
			cscov = 0;
		udp6->uh_ulen = htons(cscov);
		/*
		 * For UDP-Lite, checksum coverage length of zero means
		 * the entire UDPLite packet is covered by the checksum.
		 */
		cscov_partial = (cscov == 0) ? 0 : 1;
	} else if (plen <= 0xffff)
		udp6->uh_ulen = htons((u_short)plen);
	else
		udp6->uh_ulen = 0;
	udp6->uh_sum = 0;

	switch (af) {
	case AF_INET6:
		ip6 = mtod(m, struct ip6_hdr *);
		ip6->ip6_flow	= inp->inp_flow & IPV6_FLOWINFO_MASK;
		ip6->ip6_vfc	&= ~IPV6_VERSION_MASK;
		ip6->ip6_vfc	|= IPV6_VERSION;
		ip6->ip6_plen	= htons((u_short)plen);
		ip6->ip6_nxt	= nxt;
		ip6->ip6_hlim	= in6_selecthlim(inp, NULL);
		ip6->ip6_src	= *laddr;
		ip6->ip6_dst	= *faddr;

		if (cscov_partial) {
			if ((udp6->uh_sum = in6_cksum_partial(m, nxt,
			    sizeof(struct ip6_hdr), plen, cscov)) == 0)
				udp6->uh_sum = 0xffff;
		} else {
			udp6->uh_sum = in6_cksum_pseudo(ip6, plen, nxt, 0);
			m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
		}

#ifdef	RSS
		{
			uint32_t hash_val, hash_type;
			uint8_t pr;

			pr = inp->inp_socket->so_proto->pr_protocol;
			/*
			 * Calculate an appropriate RSS hash for UDP and
			 * UDP Lite.
			 *
			 * The called function will take care of figuring out
			 * whether a 2-tuple or 4-tuple hash is required based
			 * on the currently configured scheme.
			 *
			 * Later later on connected socket values should be
			 * cached in the inpcb and reused, rather than constantly
			 * re-calculating it.
			 *
			 * UDP Lite is a different protocol number and will
			 * likely end up being hashed as a 2-tuple until
			 * RSS / NICs grow UDP Lite protocol awareness.
			 */
			if (rss_proto_software_hash_v6(faddr, laddr, fport,
			    inp->inp_lport, pr, &hash_val, &hash_type) == 0) {
				m->m_pkthdr.flowid = hash_val;
				M_HASHTYPE_SET(m, hash_type);
			}
		}
#endif
		flags = 0;
#ifdef	RSS
		/*
		 * Don't override with the inp cached flowid.
		 *
		 * Until the whole UDP path is vetted, it may actually
		 * be incorrect.
		 */
		flags |= IP_NODEFAULTFLOWID;
#endif

		UDP_PROBE(send, NULL, inp, ip6, inp, udp6);
		UDPSTAT_INC(udps_opackets);
		error = ip6_output(m, optp, &inp->inp_route6, flags,
		    inp->in6p_moptions, NULL, inp);
		break;
	case AF_INET:
		error = EAFNOSUPPORT;
		goto release;
	}
	goto releaseopt;

release:
	m_freem(m);

releaseopt:
	if (control) {
		ip6_clearpktopts(&opt, -1);
		m_freem(control);
	}
	return (error);
}
예제 #2
0
/*
 * Do a software calculation of the RSS for the given mbuf.
 *
 * This is typically used by the input path to recalculate the RSS after
 * some form of packet processing (eg de-capsulation, IP fragment reassembly.)
 *
 * dir is the packet direction - RSS_HASH_PKT_INGRESS for incoming and
 * RSS_HASH_PKT_EGRESS for outgoing.
 *
 * Returns 0 if a hash was done, -1 if no hash was done, +1 if
 * the mbuf already had a valid RSS flowid.
 *
 * This function doesn't modify the mbuf.  It's up to the caller to
 * assign flowid/flowtype as appropriate.
 */
int
rss_mbuf_software_hash_v6(const struct mbuf *m, int dir, uint32_t *hashval,
    uint32_t *hashtype)
{
	const struct ip6_hdr *ip6;
	const struct tcphdr *th;
	const struct udphdr *uh;
	uint32_t flowtype;
	uint8_t proto;
	int off, newoff;
	int nxt;

	/*
	 * XXX For now this only handles hashing on incoming mbufs.
	 */
	if (dir != RSS_HASH_PKT_INGRESS) {
		RSS_DEBUG("called on EGRESS packet!\n");
		return (-1);
	}

	off = sizeof(struct ip6_hdr);

	/*
	 * First, validate that the mbuf we have is long enough
	 * to have an IPv6 header in it.
	 */
	if (m->m_pkthdr.len < off) {
		RSS_DEBUG("short mbuf pkthdr\n");
		return (-1);
	}
	if (m->m_len < off) {
		RSS_DEBUG("short mbuf len\n");
		return (-1);
	}

	/* Ok, let's dereference that */
	ip6 = mtod(m, struct ip6_hdr *);
	proto = ip6->ip6_nxt;

	/*
	 * Find the beginning of the TCP/UDP header.
	 *
	 * If this is a fragment then it shouldn't be four-tuple
	 * hashed just yet.  Once it's reassembled into a full
	 * frame it should be re-hashed.
	 */
	while (proto != IPPROTO_FRAGMENT) {
		newoff = ip6_nexthdr(m, off, proto, &nxt);
		if (newoff < 0)
			break;
		off = newoff;
		proto = nxt;
	}

	/*
	 * If the mbuf flowid/flowtype matches the packet type,
	 * and we don't support the 4-tuple version of the given protocol,
	 * then signal to the owner that it can trust the flowid/flowtype
	 * details.
	 *
	 * This is a little picky - eg, if TCPv6 / UDPv6 hashing
	 * is supported but we got a TCP/UDP frame only 2-tuple hashed,
	 * then we shouldn't just "trust" the 2-tuple hash.  We need
	 * a 4-tuple hash.
	 */
	flowtype = M_HASHTYPE_GET(m);

	if (flowtype != M_HASHTYPE_NONE) {
		switch (proto) {
		case IPPROTO_UDP:
			if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) &&
			    (flowtype == M_HASHTYPE_RSS_UDP_IPV6)) {
				return (1);
			}
			/*
			 * Only allow 2-tuple for UDP frames if we don't also
			 * support 4-tuple for UDP.
			 */
			if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) &&
			    ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) == 0) &&
			    flowtype == M_HASHTYPE_RSS_IPV6) {
				return (1);
			}
			break;
		case IPPROTO_TCP:
			if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) &&
			    (flowtype == M_HASHTYPE_RSS_TCP_IPV6)) {
				return (1);
			}
			/*
			 * Only allow 2-tuple for TCP frames if we don't also
			 * support 4-tuple for TCP.
			 */
			if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) &&
			    ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) == 0) &&
			    flowtype == M_HASHTYPE_RSS_IPV6) {
				return (1);
			}
			break;
		default:
			if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) &&
			    flowtype == M_HASHTYPE_RSS_IPV6) {
				return (1);
			}
			break;
		}
	}

	/*
	 * Decode enough information to make a hash decision.
	 */
	if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6) &&
	    (proto == IPPROTO_TCP)) {
		if (m->m_len < off + sizeof(struct tcphdr)) {
			RSS_DEBUG("short TCP frame?\n");
			return (-1);
		}
		th = (const struct tcphdr *)((c_caddr_t)ip6 + off);
		return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst,
		    th->th_sport,
		    th->th_dport,
		    proto,
		    hashval,
		    hashtype);
	} else if ((rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6) &&
	    (proto == IPPROTO_UDP)) {
		if (m->m_len < off + sizeof(struct udphdr)) {
			RSS_DEBUG("short UDP frame?\n");
			return (-1);
		}
		uh = (const struct udphdr *)((c_caddr_t)ip6 + off);
		return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst,
		    uh->uh_sport,
		    uh->uh_dport,
		    proto,
		    hashval,
		    hashtype);
	} else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) {
		/* Default to 2-tuple hash */
		return rss_proto_software_hash_v6(&ip6->ip6_src, &ip6->ip6_dst,
		    0,	/* source port */
		    0,	/* destination port */
		    0,	/* IPPROTO_IP */
		    hashval,
		    hashtype);
	} else {
		RSS_DEBUG("no available hashtypes!\n");
		return (-1);
	}
}