/*
* Calculate an appropriate ipv6 2-tuple or 4-tuple given the given
* IPv6 source/destination address, UDP or TCP source/destination ports
* and the protocol type.
*
* The protocol code may wish to do a software hash of the given
* tuple. This depends upon the currently configured RSS hash types.
*
* This assumes that the packet in question isn't a fragment.
*
* It also assumes the packet source/destination address
* are in "incoming" packet order (ie, source is "far" address.)
*/
int
rss_proto_software_hash_v6(const struct in6_addr *s, const struct in6_addr *d,
u_short sp, u_short dp, int proto,
uint32_t *hashval, uint32_t *hashtype)
{
uint32_t hash;
/*
* Next, choose the hash type depending upon the protocol
* identifier.
*/
if ((proto == IPPROTO_TCP) &&
(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) {
hash = rss_hash_ip6_4tuple(s, sp, d, dp);
*hashval = hash;
*hashtype = M_HASHTYPE_RSS_TCP_IPV6;
return (0);
} else if ((proto == IPPROTO_UDP) &&
(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) {
hash = rss_hash_ip6_4tuple(s, sp, d, dp);
*hashval = hash;
*hashtype = M_HASHTYPE_RSS_UDP_IPV6;
return (0);
} else if (rss_gethashconfig() & RSS_HASHTYPE_RSS_IPV6) {
/* RSS doesn't hash on other protocols like SCTP; so 2-tuple */
hash = rss_hash_ip6_2tuple(s, d);
*hashval = hash;
*hashtype = M_HASHTYPE_RSS_IPV6;
return (0);
}
/* No configured available hashtypes! */
RSS_DEBUG("no available hashtypes!\n");
return (-1);
}
struct inpcbgroup *
in6_pcbgroup_bytuple(struct inpcbinfo *pcbinfo, const struct in6_addr *laddrp,
u_short lport, const struct in6_addr *faddrp, u_short fport)
{
uint32_t hash;
/*
* RSS note: we pass foreign addr/port as source, and local addr/port
* as destination, as we want to align with what the hardware is
* doing.
*/
switch (pcbinfo->ipi_hashfields) {
case IPI_HASHFIELDS_4TUPLE:
#ifdef RSS
hash = rss_hash_ip6_4tuple(faddrp, fport, laddrp, lport);
#else
hash = faddrp->s6_addr32[3] ^ fport;
#endif
break;
case IPI_HASHFIELDS_2TUPLE:
#ifdef RSS
hash = rss_hash_ip6_2tuple(faddrp, laddrp);
#else
hash = faddrp->s6_addr32[3] ^ laddrp->s6_addr32[3];
#endif
break;
default:
hash = 0;
}
return (&pcbinfo->ipi_pcbgroups[in6_pcbgroup_getbucket(pcbinfo,
hash)]);
}
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;
struct ifnet *oifp = 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);
}
if (control) {
if ((error = ip6_setpktopts(control, &opt,
inp->in6p_outputopts, td->td_ucred, IPPROTO_UDP)) != 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(sin6, optp, inp, NULL,
td->td_ucred, &oifp, &in6a);
if (error)
goto release;
if (oifp && scope_ambiguous &&
(error = in6_setscope(&sin6->sin6_addr,
oifp, NULL))) {
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 == 0) {
error = ENOBUFS;
goto release;
}
/*
* Stuff checksum and output datagram.
*/
nxt = (inp->inp_socket->so_proto->pr_protocol == IPPROTO_UDP) ?
IPPROTO_UDP : IPPROTO_UDPLITE;
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);
}
/*
* XXX for now assume UDP is 2-tuple.
* Later on this may become configurable as 4-tuple;
* we should support that.
*
* XXX .. and we should likely cache this in the inpcb.
*/
#ifdef RSS
m->m_pkthdr.flowid = rss_hash_ip6_2tuple(*faddr, *laddr);
M_HASHTYPE_SET(m, M_HASHTYPE_RSS_IPV6);
#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, NULL, 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);
}
