/**
* Filtering and updating done during the CLOSED state of the TCP state machine.
* Part of RFC 6146 section 3.5.2.2.
*/
static verdict tcp_closed_state_handle(struct packet *pkt, struct tuple *tuple)
{
struct bib_entry *bib;
verdict result;
int error;
switch (pkt_l3_proto(pkt)) {
case L3PROTO_IPV6:
if (pkt_tcp_hdr(pkt)->syn) {
result = is_error(tcp_closed_v6_syn(pkt, tuple)) ? VERDICT_DROP : VERDICT_CONTINUE;
goto syn_out;
}
break;
case L3PROTO_IPV4:
if (pkt_tcp_hdr(pkt)->syn) {
result = tcp_closed_v4_syn(pkt, tuple);
goto syn_out;
}
break;
}
error = bibdb_get(tuple, &bib);
if (error) {
log_debug("Closed state: Packet is not SYN and there is no BIB entry, so discarding. "
"ERRcode %d", error);
inc_stats(pkt, IPSTATS_MIB_INNOROUTES);
return VERDICT_DROP;
}
bib_return(bib);
return VERDICT_CONTINUE;
syn_out:
if (result == VERDICT_DROP)
inc_stats(pkt, IPSTATS_MIB_INDISCARDS);
return result;
}
/**
* This is just a wrapper. Its sole intent is to minimize mess below.
*/
static int find_mask_domain(struct xlation *state,
struct ipv4_transport_addr *dst,
struct mask_domain **masks)
{
struct ipv6hdr *hdr6 = pkt_ip6_hdr(&state->in);
struct route4_args args = {
.ns = state->jool.ns,
.daddr = dst->l3,
.tos = ttp64_xlat_tos(&state->jool.global->cfg, hdr6),
.proto = ttp64_xlat_proto(hdr6),
.mark = state->in.skb->mark,
};
*masks = mask_domain_find(state->jool.nat64.pool4, &state->in.tuple,
state->jool.global->cfg.nat64.f_args, &args);
if (*masks)
return 0;
log_debug("There is no mask domain mapped to mark %u.",
state->in.skb->mark);
return -EINVAL;
}
/**
* Assumes that "tuple" represents a IPv6-UDP or ICMP packet, and filters and
* updates based on it.
*
* This is RFC 6146, first halves of both sections 3.5.1 and 3.5.3.
*
* @pkt: tuple's packet. This is actually only used for error reporting.
* @tuple: summary of the packet Jool is currently translating.
*/
static verdict ipv6_simple(struct xlation *state)
{
struct ipv4_transport_addr dst4;
struct mask_domain *masks;
int error;
if (xlat_dst_6to4(state, &dst4))
return drop(state, JSTAT_UNTRANSLATABLE_DST6);
if (find_mask_domain(state, &dst4, &masks))
return drop(state, JSTAT_MASK_DOMAIN_NOT_FOUND);
error = bib_add6(state, masks, &state->in.tuple, &dst4);
mask_domain_put(masks);
switch (error) {
case 0:
return succeed(state);
default:
/*
* Error msg already printed, but since bib_add6() sprawls
* messily, let's leave this here just in case.
*/
log_debug("bib_add6() threw error code %d.", error);
return drop(state, JSTAT_BIB6_NOT_FOUND);
}
}
/**
* Assumes that "tuple" represents a IPv4-UDP or ICMP packet, and filters and
* updates based on it.
*
* This is RFC 6146, second halves of both sections 3.5.1 and 3.5.3.
*
* @pkt skb tuple's packet. This is actually only used for error reporting.
* @tuple4 tuple summary of the packet Jool is currently translating.
* @return VER_CONTINUE if everything went OK, VER_DROP otherwise.
*/
static verdict ipv4_simple(struct xlation *state)
{
/*
* Because this is the IPv4->IPv6 direction, what the tuple labels
* "source" is what the BIB entry labels "destination."
* We're inheriting this naming quirk from the RFC.
*/
struct ipv4_transport_addr *dst4 = &state->in.tuple.src.addr4;
struct ipv6_transport_addr dst6;
int error;
if (RFC6052_4TO6(state, &dst4->l3, &dst6.l3))
return drop(state, JSTAT_UNTRANSLATABLE_DST4);
dst6.l4 = dst4->l4;
error = bib_add4(state, &dst6, &state->in.tuple);
switch (error) {
case 0:
return succeed(state);
case -ESRCH:
log_debug("There is no BIB entry for the IPv4 packet.");
return untranslatable_icmp(state, JSTAT_BIB4_NOT_FOUND,
ICMPERR_ADDR_UNREACHABLE, 0);
case -EPERM:
log_debug("Packet was blocked by Address-Dependent Filtering.");
return drop_icmp(state, JSTAT_ADF, ICMPERR_FILTER, 0);
default:
log_debug("Errcode %d while finding a BIB entry.", error);
return drop(state, JSTAT_UNKNOWN);
}
}
/**
* Filtering and updating during the V4 INIT state of the TCP state machine.
* Part of RFC 6146 section 3.5.2.2.
*/
static enum session_fate tcp_v4_init_state(struct session_entry *session,
struct xlation *state)
{
struct packet *pkt = &state->in;
switch (pkt_l3_proto(pkt)) {
case L3PROTO_IPV6:
if (pkt_tcp_hdr(pkt)->syn) {
if (session->has_stored)
log_debug("Simultaneous Open!");
session->state = ESTABLISHED;
session->has_stored = false;
return FATE_TIMER_EST;
}
break;
/**
* "OMG WHAT IS THIS?!!!!1!1oneone"
*
* Well, basically, they don't seem to have tested the packet storage
* thing all that well while writing the RFC.
*
* This is a patch that helps type 2 packets work. This is the problem:
*
* - IPv4 node n4 writes a TCP SYN. Let's call this packet "A".
* A arrives to the NAT64.
* - Let's say there is a BIB entry but no session that matches A, and
* also, ADF is active, so the NAT64 decides to store A.
* To this end, it creates and stores session entry [src6=a, dst6=b,
* src4=c, dst4=d, proto=TCP, state=V4 INIT, stored=A].
* A is not translated.
*
* The intent is that the NAT64 is now waiting for an IPv6 packet "B"
* that is the Simultaneous Open counterpart to A. If B arrives within 6
* seconds, A is allowed, and if it doesn't, then A is not allowed and
* will be ICMP errored.
* So far so good, right?
*
* Wrong.
*
* The problem is that A created a fully valid session that corresponds
* to itself. Because n4 doesn't receive an answer, it retries A. It
* does so before the 6-second timeout because sockets are impatient
* like that. So A2 arrives at the NAT64 and is translated successfully
* because there's now a valid session that matches it. In other words,
* A authorized itself despite ADF.
*
* One might argue that this would be a reason to not treat type 1 and 2
* packets differently: Simply store these bogus sessions away from the
* main database and the A2 session lookup will fail. This doesn't work
* either, because the whole thing is that this session needs to be
* lookupable in the 6-to-4 direction, otherwise B cannot cancel the
* ICMP error.
*
* Also, these sessions are mapped to a valid BIB entry, and as such
* need to prevent this entry from dying. This is hard to enforce when
* storing these sessions in another database.
*
* So the core of the issue is that the V4 INIT state lets v4 packets
* through even when ADF is active. Hence this switch case.
* (Because this only handles type 2 packets, ADF active = packet stored
* in this case.)
*
* Type 1 packets don't suffer from this problem because they aren't
* associated with a valid BIB entry.
*
* Similar to type 1 packets, we will assume that this retry is not
* entitled to a session timeout update. Or any session updates, for
* that matter. (See pktqueue_add())
*/
case L3PROTO_IPV4:
if (session->has_stored) {
log_debug("Simultaneous Open already exists.");
return FATE_DROP;
}
break;
}
return FATE_PRESERVE;
}
/**
* Filtering and updating during the V6 INIT state of the TCP state machine.
* Part of RFC 6146 section 3.5.2.2.
*/
static enum session_fate tcp_v6_init_state(struct session_entry *session,
struct xlation *state)
{
struct packet *pkt = &state->in;
if (pkt_tcp_hdr(pkt)->syn) {
switch (pkt_l3_proto(pkt)) {
case L3PROTO_IPV4:
session->state = ESTABLISHED;
return FATE_TIMER_EST;
case L3PROTO_IPV6:
return FATE_TIMER_TRANS;
}
}
return FATE_PRESERVE;
}
/**
* Filtering and updating during the ESTABLISHED state of the TCP state machine.
* Part of RFC 6146 section 3.5.2.2.
*/
static enum session_fate tcp_established_state(struct session_entry *session,
struct xlation *state)
{
struct packet *pkt = &state->in;
if (pkt_tcp_hdr(pkt)->fin) {
switch (pkt_l3_proto(pkt)) {
case L3PROTO_IPV4:
session->state = V4_FIN_RCV;
break;
case L3PROTO_IPV6:
session->state = V6_FIN_RCV;
break;
}
return FATE_PRESERVE;
} else if (pkt_tcp_hdr(pkt)->rst) {
session->state = TRANS;
return FATE_TIMER_TRANS;
}
return FATE_TIMER_EST;
}
static bool handle_rst_during_fin_rcv(struct xlation *state)
{
return state->jool.global->cfg.nat64.handle_rst_during_fin_rcv;
}
/**
* Filtering and updating during the V4 FIN RCV state of the TCP state machine.
* Part of RFC 6146 section 3.5.2.2.
*/
static enum session_fate tcp_v4_fin_rcv_state(struct session_entry *session,
struct xlation *state)
{
struct packet *pkt = &state->in;
struct tcphdr *hdr;
if (pkt_l3_proto(pkt) == L3PROTO_IPV6) {
hdr = pkt_tcp_hdr(pkt);
if (hdr->fin) {
session->state = V4_FIN_V6_FIN_RCV;
return FATE_TIMER_TRANS;
}
if (hdr->rst && handle_rst_during_fin_rcv(state)) {
/* https://github.com/NICMx/Jool/issues/212 */
return FATE_TIMER_TRANS;
}
}
return FATE_TIMER_EST;
}
/**
* Filtering and updating during the V6 FIN RCV state of the TCP state machine.
* Part of RFC 6146 section 3.5.2.2.
*/
static enum session_fate tcp_v6_fin_rcv_state(struct session_entry *session,
struct xlation *state)
{
struct packet *pkt = &state->in;
struct tcphdr *hdr;
if (pkt_l3_proto(pkt) == L3PROTO_IPV4) {
hdr = pkt_tcp_hdr(pkt);
if (hdr->fin) {
session->state = V4_FIN_V6_FIN_RCV;
return FATE_TIMER_TRANS;
}
if (hdr->rst && handle_rst_during_fin_rcv(state)) {
/* https://github.com/NICMx/Jool/issues/212 */
return FATE_TIMER_TRANS;
}
}
return FATE_TIMER_EST;
}
/**
* Filtering and updating during the V6 FIN + V4 FIN RCV state of the TCP state
* machine.
* Part of RFC 6146 section 3.5.2.2.
*/
static enum session_fate tcp_v4_fin_v6_fin_rcv_state(void)
{
return FATE_PRESERVE; /* Only the timeout can change this state. */
}
/**
* Filtering and updating done during the TRANS state of the TCP state machine.
* Part of RFC 6146 section 3.5.2.2.
*/
static enum session_fate tcp_trans_state(struct session_entry *session,
struct xlation *state)
{
struct packet *pkt = &state->in;
if (!pkt_tcp_hdr(pkt)->rst) {
session->state = ESTABLISHED;
return FATE_TIMER_EST;
}
return FATE_PRESERVE;
}
static enum session_fate tcp_state_machine(struct session_entry *session,
void *arg)
{
switch (session->state) {
case ESTABLISHED:
return tcp_established_state(session, arg);
case V4_INIT:
return tcp_v4_init_state(session, arg);
case V6_INIT:
return tcp_v6_init_state(session, arg);
case V4_FIN_RCV:
return tcp_v4_fin_rcv_state(session, arg);
case V6_FIN_RCV:
return tcp_v6_fin_rcv_state(session, arg);
case V4_FIN_V6_FIN_RCV:
return tcp_v4_fin_v6_fin_rcv_state();
case TRANS:
return tcp_trans_state(session, arg);
}
WARN(true, "Invalid state found: %u.", session->state);
return FATE_RM;
}
/**
* IPv6 half of RFC 6146 section 3.5.2.
*/
static verdict ipv6_tcp(struct xlation *state)
{
struct ipv4_transport_addr dst4;
struct collision_cb cb;
struct mask_domain *masks;
verdict result;
if (xlat_dst_6to4(state, &dst4))
return drop(state, JSTAT_UNTRANSLATABLE_DST6);
if (find_mask_domain(state, &dst4, &masks))
return drop(state, JSTAT_MASK_DOMAIN_NOT_FOUND);
cb.cb = tcp_state_machine;
cb.arg = state;
result = bib_add_tcp6(state, masks, &dst4, &cb);
mask_domain_put(masks);
return (result == VERDICT_CONTINUE) ? succeed(state) : result;
}
/**
* IPv4 half of RFC 6146 section 3.5.2.
*/
static verdict ipv4_tcp(struct xlation *state)
{
struct ipv4_transport_addr *dst4 = &state->in.tuple.src.addr4;
struct ipv6_transport_addr dst6;
struct collision_cb cb;
verdict result;
if (RFC6052_4TO6(state, &dst4->l3, &dst6.l3))
return drop(state, JSTAT_UNTRANSLATABLE_DST4);
dst6.l4 = dst4->l4;
cb.cb = tcp_state_machine;
cb.arg = state;
result = bib_add_tcp4(state, &dst6, &cb);
return (result == VERDICT_CONTINUE) ? succeed(state) : result;
}
/**
* filtering_and_updating - Main F&U routine. Decides if "skb" should be
* processed, updating binding and session information.
*/
verdict filtering_and_updating(struct xlation *state)
{
struct packet *in = &state->in;
struct ipv6hdr *hdr_ip6;
verdict result = VERDICT_CONTINUE;
log_debug("Step 2: Filtering and Updating");
switch (pkt_l3_proto(in)) {
case L3PROTO_IPV6:
/* Get rid of hairpinning loops and unwanted packets. */
hdr_ip6 = pkt_ip6_hdr(in);
if (pool6_contains(state, &hdr_ip6->saddr)) {
log_debug("Hairpinning loop. Dropping...");
return drop(state, JSTAT_HAIRPIN_LOOP);
}
if (!pool6_contains(state, &hdr_ip6->daddr)) {
log_debug("Packet does not belong to pool6.");
return untranslatable(state, JSTAT_POOL6_MISMATCH);
}
/* ICMP errors should not be filtered or affect the tables. */
if (pkt_is_icmp6_error(in)) {
log_debug("Packet is ICMPv6 error; skipping step...");
return VERDICT_CONTINUE;
}
break;
case L3PROTO_IPV4:
/* Get rid of unexpected packets */
if (!pool4db_contains(state->jool.nat64.pool4, state->jool.ns,
in->tuple.l4_proto, &in->tuple.dst.addr4)) {
log_debug("Packet does not belong to pool4.");
return untranslatable(state, JSTAT_POOL4_MISMATCH);
}
/* ICMP errors should not be filtered or affect the tables. */
if (pkt_is_icmp4_error(in)) {
log_debug("Packet is ICMPv4 error; skipping step...");
return VERDICT_CONTINUE;
}
break;
}
/*
* Note: I'm sorry, but the remainder of the Filtering and Updating step
* is not going to be done in the order in which the RFC explains it.
* This is because the BIB has a critical spinlock, and we need to
* take out as much work from it as possible.
*/
switch (pkt_l4_proto(in)) {
case L4PROTO_UDP:
switch (pkt_l3_proto(in)) {
case L3PROTO_IPV6:
result = ipv6_simple(state);
break;
case L3PROTO_IPV4:
result = ipv4_simple(state);
break;
}
break;
case L4PROTO_TCP:
switch (pkt_l3_proto(in)) {
case L3PROTO_IPV6:
result = ipv6_tcp(state);
break;
case L3PROTO_IPV4:
result = ipv4_tcp(state);
break;
}
break;
case L4PROTO_ICMP:
switch (pkt_l3_proto(in)) {
case L3PROTO_IPV6:
if (state->jool.global->cfg.nat64.drop_icmp6_info) {
log_debug("Packet is ICMPv6 info (ping); dropping due to policy.");
return drop(state, JSTAT_ICMP6_FILTER);
}
result = ipv6_simple(state);
break;
case L3PROTO_IPV4:
result = ipv4_simple(state);
break;
}
break;
case L4PROTO_OTHER:
WARN(true, "Unknown layer 4 protocol: %d", pkt_l4_proto(in));
return drop(state, JSTAT_UNKNOWN_L4_PROTO);
}
log_debug("Done: Step 2.");
return result;
}