static int
udp_output(struct inpcb *inp, struct mbuf *m, struct sockaddr *addr,
struct mbuf *control, struct thread *td)
{
struct udpiphdr *ui;
int len = m->m_pkthdr.len;
struct in_addr faddr, laddr;
struct cmsghdr *cm;
struct inpcbinfo *pcbinfo;
struct sockaddr_in *sin, src;
int cscov_partial = 0;
int error = 0;
int ipflags;
u_short fport, lport;
int unlock_udbinfo;
u_char tos;
uint8_t pr;
uint16_t cscov = 0;
/*
* udp_output() may need to temporarily bind or connect the current
* inpcb. As such, we don't know up front whether we will need the
* pcbinfo lock or not. Do any work to decide what is needed up
* front before acquiring any locks.
*/
if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) {
if (control)
m_freem(control);
m_freem(m);
return (EMSGSIZE);
}
src.sin_family = 0;
INP_RLOCK(inp);
tos = inp->inp_ip_tos;
if (control != NULL) {
/*
* XXX: Currently, we assume all the optional information is
* stored in a single mbuf.
*/
if (control->m_next) {
INP_RUNLOCK(inp);
m_freem(control);
m_freem(m);
return (EINVAL);
}
for (; control->m_len > 0;
control->m_data += CMSG_ALIGN(cm->cmsg_len),
control->m_len -= CMSG_ALIGN(cm->cmsg_len)) {
cm = mtod(control, struct cmsghdr *);
if (control->m_len < sizeof(*cm) || cm->cmsg_len == 0
|| cm->cmsg_len > control->m_len) {
error = EINVAL;
break;
}
if (cm->cmsg_level != IPPROTO_IP)
continue;
switch (cm->cmsg_type) {
case IP_SENDSRCADDR:
if (cm->cmsg_len !=
CMSG_LEN(sizeof(struct in_addr))) {
error = EINVAL;
break;
}
bzero(&src, sizeof(src));
src.sin_family = AF_INET;
src.sin_len = sizeof(src);
src.sin_port = inp->inp_lport;
src.sin_addr =
*(struct in_addr *)CMSG_DATA(cm);
break;
case IP_TOS:
if (cm->cmsg_len != CMSG_LEN(sizeof(u_char))) {
error = EINVAL;
break;
}
tos = *(u_char *)CMSG_DATA(cm);
break;
default:
error = ENOPROTOOPT;
break;
}
if (error)
break;
}
m_freem(control);
}
if (error) {
INP_RUNLOCK(inp);
m_freem(m);
return (error);
}
/*
* Depending on whether or not the application has bound or connected
* the socket, we may have to do varying levels of work. The optimal
* case is for a connected UDP socket, as a global lock isn't
* required at all.
*
* In order to decide which we need, we require stability of the
* inpcb binding, which we ensure by acquiring a read lock on the
* inpcb. This doesn't strictly follow the lock order, so we play
* the trylock and retry game; note that we may end up with more
* conservative locks than required the second time around, so later
* assertions have to accept that. Further analysis of the number of
* misses under contention is required.
*
* XXXRW: Check that hash locking update here is correct.
*/
pr = inp->inp_socket->so_proto->pr_protocol;
pcbinfo = get_inpcbinfo(pr);
sin = (struct sockaddr_in *)addr;
if (sin != NULL &&
(inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0)) {
INP_RUNLOCK(inp);
INP_WLOCK(inp);
INP_HASH_WLOCK(pcbinfo);
unlock_udbinfo = UH_WLOCKED;
} else if ((sin != NULL && (
(sin->sin_addr.s_addr == INADDR_ANY) ||
(sin->sin_addr.s_addr == INADDR_BROADCAST) ||
(inp->inp_laddr.s_addr == INADDR_ANY) ||
(inp->inp_lport == 0))) ||
(src.sin_family == AF_INET)) {
INP_HASH_RLOCK(pcbinfo);
unlock_udbinfo = UH_RLOCKED;
} else
unlock_udbinfo = UH_UNLOCKED;
/*
* If the IP_SENDSRCADDR control message was specified, override the
* source address for this datagram. Its use is invalidated if the
* address thus specified is incomplete or clobbers other inpcbs.
*/
laddr = inp->inp_laddr;
lport = inp->inp_lport;
if (src.sin_family == AF_INET) {
INP_HASH_LOCK_ASSERT(pcbinfo);
if ((lport == 0) ||
(laddr.s_addr == INADDR_ANY &&
src.sin_addr.s_addr == INADDR_ANY)) {
error = EINVAL;
goto release;
}
error = in_pcbbind_setup(inp, (struct sockaddr *)&src,
&laddr.s_addr, &lport, td->td_ucred);
if (error)
goto release;
}
/*
* If a UDP socket has been connected, then a local address/port will
* have been selected and bound.
*
* If a UDP socket has not been connected to, then an explicit
* destination address must be used, in which case a local
* address/port may not have been selected and bound.
*/
if (sin != NULL) {
INP_LOCK_ASSERT(inp);
if (inp->inp_faddr.s_addr != INADDR_ANY) {
error = EISCONN;
goto release;
}
/*
* Jail may rewrite the destination address, so let it do
* that before we use it.
*/
error = prison_remote_ip4(td->td_ucred, &sin->sin_addr);
if (error)
goto release;
/*
* If a local address or port hasn't yet been selected, or if
* the destination address needs to be rewritten due to using
* a special INADDR_ constant, invoke in_pcbconnect_setup()
* to do the heavy lifting. Once a port is selected, we
* commit the binding back to the socket; we also commit the
* binding of the address if in jail.
*
* If we already have a valid binding and we're not
* requesting a destination address rewrite, use a fast path.
*/
if (inp->inp_laddr.s_addr == INADDR_ANY ||
inp->inp_lport == 0 ||
sin->sin_addr.s_addr == INADDR_ANY ||
sin->sin_addr.s_addr == INADDR_BROADCAST) {
INP_HASH_LOCK_ASSERT(pcbinfo);
error = in_pcbconnect_setup(inp, addr, &laddr.s_addr,
&lport, &faddr.s_addr, &fport, NULL,
td->td_ucred);
if (error)
goto release;
/*
* XXXRW: Why not commit the port if the address is
* !INADDR_ANY?
*/
/* Commit the local port if newly assigned. */
if (inp->inp_laddr.s_addr == INADDR_ANY &&
inp->inp_lport == 0) {
INP_WLOCK_ASSERT(inp);
INP_HASH_WLOCK_ASSERT(pcbinfo);
/*
* Remember addr if jailed, to prevent
* rebinding.
*/
if (prison_flag(td->td_ucred, PR_IP4))
inp->inp_laddr = laddr;
inp->inp_lport = lport;
if (in_pcbinshash(inp) != 0) {
inp->inp_lport = 0;
error = EAGAIN;
goto release;
}
inp->inp_flags |= INP_ANONPORT;
}
} else {
faddr = sin->sin_addr;
fport = sin->sin_port;
}
} else {
INP_LOCK_ASSERT(inp);
faddr = inp->inp_faddr;
fport = inp->inp_fport;
if (faddr.s_addr == INADDR_ANY) {
error = ENOTCONN;
goto release;
}
}
/*
* Calculate data length and get a mbuf for UDP, IP, and possible
* link-layer headers. Immediate slide the data pointer back forward
* since we won't use that space at this layer.
*/
M_PREPEND(m, sizeof(struct udpiphdr) + max_linkhdr, M_NOWAIT);
if (m == NULL) {
error = ENOBUFS;
goto release;
}
m->m_data += max_linkhdr;
m->m_len -= max_linkhdr;
m->m_pkthdr.len -= max_linkhdr;
/*
* Fill in mbuf with extended UDP header and addresses and length put
* into network format.
*/
ui = mtod(m, struct udpiphdr *);
bzero(ui->ui_x1, sizeof(ui->ui_x1)); /* XXX still needed? */
ui->ui_pr = pr;
ui->ui_src = laddr;
ui->ui_dst = faddr;
ui->ui_sport = lport;
ui->ui_dport = fport;
ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr));
if (pr == IPPROTO_UDPLITE) {
struct udpcb *up;
uint16_t plen;
up = intoudpcb(inp);
cscov = up->u_txcslen;
plen = (u_short)len + sizeof(struct udphdr);
if (cscov >= plen)
cscov = 0;
ui->ui_len = htons(plen);
ui->ui_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
ui->ui_v = IPVERSION << 4;
/*
* Set the Don't Fragment bit in the IP header.
*/
if (inp->inp_flags & INP_DONTFRAG) {
struct ip *ip;
ip = (struct ip *)&ui->ui_i;
ip->ip_off |= htons(IP_DF);
}
ipflags = 0;
if (inp->inp_socket->so_options & SO_DONTROUTE)
ipflags |= IP_ROUTETOIF;
if (inp->inp_socket->so_options & SO_BROADCAST)
ipflags |= IP_ALLOWBROADCAST;
if (inp->inp_flags & INP_ONESBCAST)
ipflags |= IP_SENDONES;
#ifdef MAC
mac_inpcb_create_mbuf(inp, m);
#endif
/*
* Set up checksum and output datagram.
*/
ui->ui_sum = 0;
if (cscov_partial) {
if (inp->inp_flags & INP_ONESBCAST)
faddr.s_addr = INADDR_BROADCAST;
if ((ui->ui_sum = in_cksum(m, sizeof(struct ip) + cscov)) == 0)
ui->ui_sum = 0xffff;
} else if (V_udp_cksum || !cscov_partial) {
if (inp->inp_flags & INP_ONESBCAST)
faddr.s_addr = INADDR_BROADCAST;
ui->ui_sum = in_pseudo(ui->ui_src.s_addr, faddr.s_addr,
htons((u_short)len + sizeof(struct udphdr) + pr));
m->m_pkthdr.csum_flags = CSUM_UDP;
m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
}
((struct ip *)ui)->ip_len = htons(sizeof(struct udpiphdr) + len);
((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */
((struct ip *)ui)->ip_tos = tos; /* XXX */
UDPSTAT_INC(udps_opackets);
if (unlock_udbinfo == UH_WLOCKED)
INP_HASH_WUNLOCK(pcbinfo);
else if (unlock_udbinfo == UH_RLOCKED)
INP_HASH_RUNLOCK(pcbinfo);
UDP_PROBE(send, NULL, inp, &ui->ui_i, inp, &ui->ui_u);
error = ip_output(m, inp->inp_options, NULL, ipflags,
inp->inp_moptions, inp);
if (unlock_udbinfo == UH_WLOCKED)
INP_WUNLOCK(inp);
else
INP_RUNLOCK(inp);
return (error);
release:
if (unlock_udbinfo == UH_WLOCKED) {
INP_HASH_WUNLOCK(pcbinfo);
INP_WUNLOCK(inp);
} else if (unlock_udbinfo == UH_RLOCKED) {
INP_HASH_RUNLOCK(pcbinfo);
INP_RUNLOCK(inp);
} else
INP_RUNLOCK(inp);
m_freem(m);
return (error);
}
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);
}
