diff options
Diffstat (limited to 'src/plugins/snat/in2out.c')
| -rw-r--r-- | src/plugins/snat/in2out.c | 538 |
1 files changed, 512 insertions, 26 deletions
diff --git a/src/plugins/snat/in2out.c b/src/plugins/snat/in2out.c index e9bc5384..5970588b 100644 --- a/src/plugins/snat/in2out.c +++ b/src/plugins/snat/in2out.c @@ -23,6 +23,7 @@ #include <vnet/fib/ip4_fib.h> #include <snat/snat.h> #include <snat/snat_ipfix_logging.h> +#include <snat/snat_det.h> #include <vppinfra/hash.h> #include <vppinfra/error.h> @@ -86,6 +87,7 @@ vlib_node_registration_t snat_in2out_node; vlib_node_registration_t snat_in2out_slowpath_node; vlib_node_registration_t snat_in2out_fast_node; vlib_node_registration_t snat_in2out_worker_handoff_node; +vlib_node_registration_t snat_det_in2out_node; #define foreach_snat_in2out_error \ _(UNSUPPORTED_PROTOCOL, "Unsupported protocol") \ @@ -1325,6 +1327,510 @@ VLIB_REGISTER_NODE (snat_in2out_slowpath_node) = { VLIB_NODE_FUNCTION_MULTIARCH (snat_in2out_slowpath_node, snat_in2out_slow_path_fn); +/**************************/ +/*** deterministic mode ***/ +/**************************/ +static uword +snat_det_in2out_node_fn (vlib_main_t * vm, + vlib_node_runtime_t * node, + vlib_frame_t * frame) +{ + u32 n_left_from, * from, * to_next; + snat_in2out_next_t next_index; + u32 pkts_processed = 0; + snat_main_t * sm = &snat_main; + u32 now = (u32) vlib_time_now (vm); + + from = vlib_frame_vector_args (frame); + n_left_from = frame->n_vectors; + next_index = node->cached_next_index; + + while (n_left_from > 0) + { + u32 n_left_to_next; + + vlib_get_next_frame (vm, node, next_index, + to_next, n_left_to_next); + + while (n_left_from >= 4 && n_left_to_next >= 2) + { + u32 bi0, bi1; + vlib_buffer_t * b0, * b1; + u32 next0, next1; + u32 sw_if_index0, sw_if_index1; + ip4_header_t * ip0, * ip1; + ip_csum_t sum0, sum1; + ip4_address_t new_addr0, old_addr0, new_addr1, old_addr1; + u16 old_port0, new_port0, lo_port0, i; + u16 old_port1, new_port1, lo_port1; + udp_header_t * udp0, * udp1; + tcp_header_t * tcp0, * tcp1; + u32 proto0, proto1; + snat_det_out_key_t key0, key1; + snat_det_map_t * dm0, * dm1; + snat_det_session_t * ses0 = 0, * ses1 = 0; + + /* Prefetch next iteration. */ + { + vlib_buffer_t * p2, * p3; + + p2 = vlib_get_buffer (vm, from[2]); + p3 = vlib_get_buffer (vm, from[3]); + + vlib_prefetch_buffer_header (p2, LOAD); + vlib_prefetch_buffer_header (p3, LOAD); + + CLIB_PREFETCH (p2->data, CLIB_CACHE_LINE_BYTES, STORE); + CLIB_PREFETCH (p3->data, CLIB_CACHE_LINE_BYTES, STORE); + } + + /* speculatively enqueue b0 and b1 to the current next frame */ + to_next[0] = bi0 = from[0]; + to_next[1] = bi1 = from[1]; + from += 2; + to_next += 2; + n_left_from -= 2; + n_left_to_next -= 2; + + b0 = vlib_get_buffer (vm, bi0); + b1 = vlib_get_buffer (vm, bi1); + + next0 = SNAT_IN2OUT_NEXT_LOOKUP; + next1 = SNAT_IN2OUT_NEXT_LOOKUP; + + ip0 = vlib_buffer_get_current (b0); + udp0 = ip4_next_header (ip0); + tcp0 = (tcp_header_t *) udp0; + + sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_RX]; + + dm0 = snat_det_map_by_user(sm, &ip0->src_address); + if (PREDICT_FALSE(!dm0)) + { + clib_warning("no match for internal host %U", + format_ip4_address, &ip0->src_address); + goto trace0; + } + + snat_det_forward(dm0, &ip0->src_address, &new_addr0, &lo_port0); + + ses0 = snat_det_find_ses_by_in(dm0, &ip0->src_address, tcp0->src); + if (PREDICT_FALSE(!ses0)) + { + key0.ext_host_addr = ip0->dst_address; + key0.ext_host_port = tcp0->dst; + for (i = 0; i < dm0->ports_per_host; i++) + { + key0.out_port = clib_host_to_net_u16 (lo_port0 + i + + (clib_net_to_host_u16 (tcp0->src) % dm0->ports_per_host)); + + if (snat_det_get_ses_by_out (dm0, &ip0->src_address, key0.as_u64)) + continue; + + ses0 = snat_det_ses_create(dm0, &ip0->src_address, tcp0->src, &key0); + break; + } + if (PREDICT_FALSE(!ses0)) + { + next0 = SNAT_IN2OUT_NEXT_DROP; + goto trace0; + } + } + + new_port0 = ses0->out.out_port; + proto0 = ip_proto_to_snat_proto (ip0->protocol); + + old_addr0.as_u32 = ip0->src_address.as_u32; + ip0->src_address.as_u32 = new_addr0.as_u32; + vnet_buffer(b0)->sw_if_index[VLIB_TX] = sm->outside_fib_index; + + sum0 = ip0->checksum; + sum0 = ip_csum_update (sum0, old_addr0.as_u32, new_addr0.as_u32, + ip4_header_t, + src_address /* changed member */); + ip0->checksum = ip_csum_fold (sum0); + + if (PREDICT_TRUE(proto0 == SNAT_PROTOCOL_TCP)) + { + if (tcp0->flags & TCP_FLAG_SYN) + ses0->state = SNAT_SESSION_TCP_SYN_SENT; + else if (tcp0->flags & TCP_FLAG_ACK && ses0->state == SNAT_SESSION_TCP_SYN_SENT) + ses0->state = SNAT_SESSION_TCP_ESTABLISHED; + else if (tcp0->flags & TCP_FLAG_FIN && ses0->state == SNAT_SESSION_TCP_ESTABLISHED) + ses0->state = SNAT_SESSION_TCP_FIN_WAIT; + else if (tcp0->flags & TCP_FLAG_ACK && ses0->state == SNAT_SESSION_TCP_FIN_WAIT) + snat_det_ses_close(dm0, ses0); + else if (tcp0->flags & TCP_FLAG_FIN && ses0->state == SNAT_SESSION_TCP_CLOSE_WAIT) + ses0->state = SNAT_SESSION_TCP_LAST_ACK; + else if (tcp0->flags == 0 && ses0->state == SNAT_SESSION_UNKNOWN) + ses0->state = SNAT_SESSION_TCP_ESTABLISHED; + + old_port0 = tcp0->src; + tcp0->src = new_port0; + + sum0 = tcp0->checksum; + sum0 = ip_csum_update (sum0, old_addr0.as_u32, new_addr0.as_u32, + ip4_header_t, + dst_address /* changed member */); + sum0 = ip_csum_update (sum0, old_port0, new_port0, + ip4_header_t /* cheat */, + length /* changed member */); + tcp0->checksum = ip_csum_fold(sum0); + } + else + { + ses0->state = SNAT_SESSION_UDP_ACTIVE; + old_port0 = udp0->src_port; + udp0->src_port = new_port0; + udp0->checksum = 0; + } + + switch(ses0->state) + { + case SNAT_SESSION_UDP_ACTIVE: + ses0->expire = now + SNAT_UDP_TIMEOUT; + break; + case SNAT_SESSION_TCP_SYN_SENT: + case SNAT_SESSION_TCP_FIN_WAIT: + case SNAT_SESSION_TCP_CLOSE_WAIT: + case SNAT_SESSION_TCP_LAST_ACK: + ses0->expire = now + SNAT_TCP_TRANSITORY_TIMEOUT; + break; + case SNAT_SESSION_TCP_ESTABLISHED: + ses0->expire = now + SNAT_TCP_ESTABLISHED_TIMEOUT; + break; + } + + trace0: + if (PREDICT_FALSE((node->flags & VLIB_NODE_FLAG_TRACE) + && (b0->flags & VLIB_BUFFER_IS_TRACED))) + { + snat_in2out_trace_t *t = + vlib_add_trace (vm, node, b0, sizeof (*t)); + t->is_slow_path = 0; + t->sw_if_index = sw_if_index0; + t->next_index = next0; + t->session_index = ~0; + if (ses0) + t->session_index = ses0 - dm0->sessions; + } + + pkts_processed += next0 != SNAT_IN2OUT_NEXT_DROP; + + ip1 = vlib_buffer_get_current (b1); + udp1 = ip4_next_header (ip1); + tcp1 = (tcp_header_t *) udp1; + + sw_if_index1 = vnet_buffer(b1)->sw_if_index[VLIB_RX]; + + dm1 = snat_det_map_by_user(sm, &ip1->src_address); + if (PREDICT_FALSE(!dm1)) + { + clib_warning("no match for internal host %U", + format_ip4_address, &ip0->src_address); + goto trace1; + } + + snat_det_forward(dm1, &ip1->src_address, &new_addr1, &lo_port1); + + + ses1 = snat_det_find_ses_by_in(dm1, &ip1->src_address, tcp1->src); + if (PREDICT_FALSE(!ses1)) + { + key1.ext_host_addr = ip1->dst_address; + key1.ext_host_port = tcp1->dst; + for (i = 0; i < dm1->ports_per_host; i++) + { + key1.out_port = clib_host_to_net_u16 (lo_port1 + i + + (clib_net_to_host_u16 (tcp1->src) % dm1->ports_per_host)); + + if (snat_det_get_ses_by_out (dm1, &ip1->src_address, key1.as_u64)) + continue; + + ses1 = snat_det_ses_create(dm1, &ip1->src_address, tcp1->src, &key1); + break; + } + if (PREDICT_FALSE(!ses1)) + { + next1 = SNAT_IN2OUT_NEXT_DROP; + goto trace1; + } + } + + new_port1 = ses1->out.out_port; + proto1 = ip_proto_to_snat_proto (ip1->protocol); + + old_addr1.as_u32 = ip1->src_address.as_u32; + ip1->src_address.as_u32 = new_addr1.as_u32; + vnet_buffer(b1)->sw_if_index[VLIB_TX] = sm->outside_fib_index; + + sum1 = ip1->checksum; + sum1 = ip_csum_update (sum1, old_addr1.as_u32, new_addr1.as_u32, + ip4_header_t, + src_address /* changed member */); + ip1->checksum = ip_csum_fold (sum1); + + if (PREDICT_TRUE(proto1 == SNAT_PROTOCOL_TCP)) + { + if (tcp1->flags & TCP_FLAG_SYN) + ses1->state = SNAT_SESSION_TCP_SYN_SENT; + else if (tcp1->flags & TCP_FLAG_ACK && ses1->state == SNAT_SESSION_TCP_SYN_SENT) + ses1->state = SNAT_SESSION_TCP_ESTABLISHED; + else if (tcp1->flags & TCP_FLAG_FIN && ses1->state == SNAT_SESSION_TCP_ESTABLISHED) + ses1->state = SNAT_SESSION_TCP_FIN_WAIT; + else if (tcp1->flags & TCP_FLAG_ACK && ses1->state == SNAT_SESSION_TCP_FIN_WAIT) + snat_det_ses_close(dm1, ses1); + else if (tcp1->flags & TCP_FLAG_FIN && ses1->state == SNAT_SESSION_TCP_CLOSE_WAIT) + ses1->state = SNAT_SESSION_TCP_LAST_ACK; + else if (tcp1->flags == 0 && ses1->state == SNAT_SESSION_UNKNOWN) + ses1->state = SNAT_SESSION_TCP_ESTABLISHED; + + old_port1 = tcp1->src; + tcp1->src = new_port1; + + sum1 = tcp1->checksum; + sum1 = ip_csum_update (sum1, old_addr1.as_u32, new_addr1.as_u32, + ip4_header_t, + dst_address /* changed member */); + sum1 = ip_csum_update (sum1, old_port1, new_port1, + ip4_header_t /* cheat */, + length /* changed member */); + tcp1->checksum = ip_csum_fold(sum1); + } + else + { + ses1->state = SNAT_SESSION_UDP_ACTIVE; + old_port1 = udp1->src_port; + udp1->src_port = new_port1; + udp1->checksum = 0; + } + + switch(ses1->state) + { + case SNAT_SESSION_UDP_ACTIVE: + ses1->expire = now + SNAT_UDP_TIMEOUT; + break; + case SNAT_SESSION_TCP_SYN_SENT: + case SNAT_SESSION_TCP_FIN_WAIT: + case SNAT_SESSION_TCP_CLOSE_WAIT: + case SNAT_SESSION_TCP_LAST_ACK: + ses1->expire = now + SNAT_TCP_TRANSITORY_TIMEOUT; + break; + case SNAT_SESSION_TCP_ESTABLISHED: + ses1->expire = now + SNAT_TCP_ESTABLISHED_TIMEOUT; + break; + } + + trace1: + if (PREDICT_FALSE((node->flags & VLIB_NODE_FLAG_TRACE) + && (b1->flags & VLIB_BUFFER_IS_TRACED))) + { + snat_in2out_trace_t *t = + vlib_add_trace (vm, node, b1, sizeof (*t)); + t->is_slow_path = 0; + t->sw_if_index = sw_if_index1; + t->next_index = next1; + t->session_index = ~0; + if (ses1) + t->session_index = ses1 - dm1->sessions; + } + + pkts_processed += next1 != SNAT_IN2OUT_NEXT_DROP; + + /* verify speculative enqueues, maybe switch current next frame */ + vlib_validate_buffer_enqueue_x2 (vm, node, next_index, + to_next, n_left_to_next, + bi0, bi1, next0, next1); + } + + while (n_left_from > 0 && n_left_to_next > 0) + { + u32 bi0; + vlib_buffer_t * b0; + u32 next0; + u32 sw_if_index0; + ip4_header_t * ip0; + ip_csum_t sum0; + ip4_address_t new_addr0, old_addr0; + u16 old_port0, new_port0, lo_port0, i; + udp_header_t * udp0; + tcp_header_t * tcp0; + u32 proto0; + snat_det_out_key_t key0; + snat_det_map_t * dm0; + snat_det_session_t * ses0 = 0; + + /* speculatively enqueue b0 to the current next frame */ + bi0 = from[0]; + to_next[0] = bi0; + from += 1; + to_next += 1; + n_left_from -= 1; + n_left_to_next -= 1; + + b0 = vlib_get_buffer (vm, bi0); + next0 = SNAT_IN2OUT_NEXT_LOOKUP; + + ip0 = vlib_buffer_get_current (b0); + udp0 = ip4_next_header (ip0); + tcp0 = (tcp_header_t *) udp0; + + sw_if_index0 = vnet_buffer(b0)->sw_if_index[VLIB_RX]; + + dm0 = snat_det_map_by_user(sm, &ip0->src_address); + if (PREDICT_FALSE(!dm0)) + { + clib_warning("no match for internal host %U", + format_ip4_address, &ip0->src_address); + goto trace00; + } + + snat_det_forward(dm0, &ip0->src_address, &new_addr0, &lo_port0); + + ses0 = snat_det_find_ses_by_in(dm0, &ip0->src_address, tcp0->src); + if (PREDICT_FALSE(!ses0)) + { + key0.ext_host_addr = ip0->dst_address; + key0.ext_host_port = tcp0->dst; + for (i = 0; i < dm0->ports_per_host; i++) + { + key0.out_port = clib_host_to_net_u16 (lo_port0 + i + + (clib_net_to_host_u16 (tcp0->src) % dm0->ports_per_host)); + + if (snat_det_get_ses_by_out (dm0, &ip0->src_address, key0.as_u64)) + continue; + + ses0 = snat_det_ses_create(dm0, &ip0->src_address, tcp0->src, &key0); + break; + } + if (PREDICT_FALSE(!ses0)) + { + next0 = SNAT_IN2OUT_NEXT_DROP; + goto trace00; + } + } + + new_port0 = ses0->out.out_port; + proto0 = ip_proto_to_snat_proto (ip0->protocol); + + old_addr0.as_u32 = ip0->src_address.as_u32; + ip0->src_address.as_u32 = new_addr0.as_u32; + vnet_buffer(b0)->sw_if_index[VLIB_TX] = sm->outside_fib_index; + + sum0 = ip0->checksum; + sum0 = ip_csum_update (sum0, old_addr0.as_u32, new_addr0.as_u32, + ip4_header_t, + src_address /* changed member */); + ip0->checksum = ip_csum_fold (sum0); + + if (PREDICT_TRUE(proto0 == SNAT_PROTOCOL_TCP)) + { + if (tcp0->flags & TCP_FLAG_SYN) + ses0->state = SNAT_SESSION_TCP_SYN_SENT; + else if (tcp0->flags & TCP_FLAG_ACK && ses0->state == SNAT_SESSION_TCP_SYN_SENT) + ses0->state = SNAT_SESSION_TCP_ESTABLISHED; + else if (tcp0->flags & TCP_FLAG_FIN && ses0->state == SNAT_SESSION_TCP_ESTABLISHED) + ses0->state = SNAT_SESSION_TCP_FIN_WAIT; + else if (tcp0->flags & TCP_FLAG_ACK && ses0->state == SNAT_SESSION_TCP_FIN_WAIT) + snat_det_ses_close(dm0, ses0); + else if (tcp0->flags & TCP_FLAG_FIN && ses0->state == SNAT_SESSION_TCP_CLOSE_WAIT) + ses0->state = SNAT_SESSION_TCP_LAST_ACK; + else if (tcp0->flags == 0 && ses0->state == SNAT_SESSION_UNKNOWN) + ses0->state = SNAT_SESSION_TCP_ESTABLISHED; + + old_port0 = tcp0->src; + tcp0->src = new_port0; + + sum0 = tcp0->checksum; + sum0 = ip_csum_update (sum0, old_addr0.as_u32, new_addr0.as_u32, + ip4_header_t, + dst_address /* changed member */); + sum0 = ip_csum_update (sum0, old_port0, new_port0, + ip4_header_t /* cheat */, + length /* changed member */); + tcp0->checksum = ip_csum_fold(sum0); + } + else + { + ses0->state = SNAT_SESSION_UDP_ACTIVE; + old_port0 = udp0->src_port; + udp0->src_port = new_port0; + udp0->checksum = 0; + } + + switch(ses0->state) + { + case SNAT_SESSION_UDP_ACTIVE: + ses0->expire = now + SNAT_UDP_TIMEOUT; + break; + case SNAT_SESSION_TCP_SYN_SENT: + case SNAT_SESSION_TCP_FIN_WAIT: + case SNAT_SESSION_TCP_CLOSE_WAIT: + case SNAT_SESSION_TCP_LAST_ACK: + ses0->expire = now + SNAT_TCP_TRANSITORY_TIMEOUT; + break; + case SNAT_SESSION_TCP_ESTABLISHED: + ses0->expire = now + SNAT_TCP_ESTABLISHED_TIMEOUT; + break; + } + + trace00: + if (PREDICT_FALSE((node->flags & VLIB_NODE_FLAG_TRACE) + && (b0->flags & VLIB_BUFFER_IS_TRACED))) + { + snat_in2out_trace_t *t = + vlib_add_trace (vm, node, b0, sizeof (*t)); + t->is_slow_path = 0; + t->sw_if_index = sw_if_index0; + t->next_index = next0; + t->session_index = ~0; + if (ses0) + t->session_index = ses0 - dm0->sessions; + } + + pkts_processed += next0 != SNAT_IN2OUT_NEXT_DROP; + + /* verify speculative enqueue, maybe switch current next frame */ + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, + to_next, n_left_to_next, + bi0, next0); + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + + vlib_node_increment_counter (vm, snat_det_in2out_node.index, + SNAT_IN2OUT_ERROR_IN2OUT_PACKETS, + pkts_processed); + return frame->n_vectors; +} + +VLIB_REGISTER_NODE (snat_det_in2out_node) = { + .function = snat_det_in2out_node_fn, + .name = "snat-det-in2out", + .vector_size = sizeof (u32), + .format_trace = format_snat_in2out_trace, + .type = VLIB_NODE_TYPE_INTERNAL, + + .n_errors = ARRAY_LEN(snat_in2out_error_strings), + .error_strings = snat_in2out_error_strings, + + .runtime_data_bytes = sizeof (snat_runtime_t), + + .n_next_nodes = 2, + + /* edit / add dispositions here */ + .next_nodes = { + [SNAT_IN2OUT_NEXT_DROP] = "error-drop", + [SNAT_IN2OUT_NEXT_LOOKUP] = "ip4-lookup", + }, +}; + +VLIB_NODE_FUNCTION_MULTIARCH (snat_det_in2out_node, snat_det_in2out_node_fn); + +/**********************/ +/*** worker handoff ***/ +/**********************/ static uword snat_in2out_worker_handoff_fn (vlib_main_t * vm, vlib_node_runtime_t * node, @@ -1365,8 +1871,6 @@ snat_in2out_worker_handoff_fn (vlib_main_t * vm, u32 sw_if_index0; u32 rx_fib_index0; ip4_header_t * ip0; - snat_user_key_t key0; - clib_bihash_kv_8_8_t kv0, value0; u8 do_handoff; bi0 = from[0]; @@ -1380,28 +1884,7 @@ snat_in2out_worker_handoff_fn (vlib_main_t * vm, ip0 = vlib_buffer_get_current (b0); - key0.addr = ip0->src_address; - key0.fib_index = rx_fib_index0; - - kv0.key = key0.as_u64; - - /* Ever heard of of the "user" before? */ - if (clib_bihash_search_8_8 (&sm->worker_by_in, &kv0, &value0)) - { - /* No, assign next available worker (RR) */ - next_worker_index = sm->first_worker_index; - if (vec_len (sm->workers)) - { - next_worker_index += - sm->workers[sm->next_worker++ % _vec_len (sm->workers)]; - } - - /* add non-traslated packets worker lookup */ - kv0.value = next_worker_index; - clib_bihash_add_del_8_8 (&sm->worker_by_in, &kv0, 1); - } - else - next_worker_index = value0.value; + next_worker_index = sm->worker_in2out_cb(ip0, rx_fib_index0); if (PREDICT_FALSE (next_worker_index != cpu_index)) { @@ -1441,7 +1924,7 @@ snat_in2out_worker_handoff_fn (vlib_main_t * vm, /* if this is 1st frame */ if (!f) { - f = vlib_get_frame_to_node (vm, snat_in2out_node.index); + f = vlib_get_frame_to_node (vm, sm->in2out_node_index); to_next = vlib_frame_vector_args (f); } @@ -1461,7 +1944,7 @@ snat_in2out_worker_handoff_fn (vlib_main_t * vm, } if (f) - vlib_put_frame_to_node (vm, snat_in2out_node.index, f); + vlib_put_frame_to_node (vm, sm->in2out_node_index, f); if (hf) hf->n_vectors = VLIB_FRAME_SIZE - n_left_to_next_worker; @@ -1508,6 +1991,9 @@ VLIB_REGISTER_NODE (snat_in2out_worker_handoff_node) = { VLIB_NODE_FUNCTION_MULTIARCH (snat_in2out_worker_handoff_node, snat_in2out_worker_handoff_fn); +/********************************/ +/*** static mapping only mode ***/ +/********************************/ static inline u32 icmp_in2out_static_map (snat_main_t *sm, vlib_buffer_t * b0, ip4_header_t * ip0, |