diff options
Diffstat (limited to 'plugins/lb-plugin/lb')
| -rw-r--r-- | plugins/lb-plugin/lb/node.c | 325 |
1 files changed, 163 insertions, 162 deletions
diff --git a/plugins/lb-plugin/lb/node.c b/plugins/lb-plugin/lb/node.c index 4f71616120e..c51a2108a0f 100644 --- a/plugins/lb-plugin/lb/node.c +++ b/plugins/lb-plugin/lb/node.c @@ -108,168 +108,169 @@ lb_node_fn (vlib_main_t * vm, u8 is_input_v4, //Compile-time parameter stating that is input is v4 (or v6) u8 is_encap_v4) //Compile-time parameter stating that is GRE encap is v4 (or v6) { - ip_lookup_main_t *lm = (is_input_v4)?&ip4_main.lookup_main:&ip6_main.lookup_main; - lb_main_t *lbm = &lb_main; - vlib_node_runtime_t *error_node = node; - u32 n_left_from, *from, next_index, *to_next, n_left_to_next; - u32 cpu_index = os_get_cpu_number(); - u32 lb_time = lb_hash_time_now(vm); - - lb_hash_t *sticky_ht = lb_get_sticky_table(cpu_index); - from = vlib_frame_vector_args (frame); - n_left_from = frame->n_vectors; - next_index = node->cached_next_index; - - while (n_left_from > 0) - { - vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); - while (n_left_from > 0 && n_left_to_next > 0) - { - u32 pi0; - vlib_buffer_t *p0; - ip_adjacency_t *adj0; - lb_adj_data_t *ad0; - lb_vip_t *vip0; - lb_as_t *as0; - gre_header_t *gre0; - u16 len0; - u32 value0, available_index0, hash0; - u64 key0[5]; - lb_error_t error0 = LB_ERROR_NONE; - lb_next_t next0 = LB_NEXT_LOOKUP; - - if (PREDICT_TRUE(n_left_from > 1)) - { - vlib_buffer_t *p2; - p2 = vlib_get_buffer(vm, from[1]); - vlib_prefetch_buffer_header(p2, STORE); - /* IPv4 + 8 = 28. possibly plus -40 */ - CLIB_PREFETCH (vlib_buffer_get_current(p2) - 40, 128, STORE); - } - - pi0 = to_next[0] = from[0]; - from += 1; - n_left_from -= 1; - to_next += 1; - n_left_to_next -= 1; - - p0 = vlib_get_buffer (vm, pi0); - adj0 = ip_get_adjacency (lm, vnet_buffer (p0)->ip.adj_index[VLIB_TX]); - ad0 = (lb_adj_data_t *) &adj0->opaque; - vip0 = pool_elt_at_index (lbm->vips, ad0->vip_index); - - if (is_input_v4) { - ip4_header_t *ip40; - ip40 = vlib_buffer_get_current (p0); - len0 = clib_net_to_host_u16(ip40->length); - key0[0] = (u64) ip40->src_address.as_u32; - key0[1] = (u64) ip40->dst_address.as_u32; - key0[2] = 0; - key0[3] = 0; - key0[4] = ((u64)((udp_header_t *)(ip40 + 1))->src_port << 32) | - ((u64)((udp_header_t *)(ip40 + 1))->dst_port << 16); - - hash0 = lb_hash_hash(key0); - } else { - ip6_header_t *ip60; - ip60 = vlib_buffer_get_current (p0); - len0 = clib_net_to_host_u16(ip60->payload_length) + sizeof(ip6_header_t); - key0[0] = ip60->src_address.as_u64[0]; - key0[1] = ip60->src_address.as_u64[1]; - key0[2] = ip60->dst_address.as_u64[0]; - key0[3] = ip60->dst_address.as_u64[1]; - key0[4] = ((u64)((udp_header_t *)(ip60 + 1))->src_port << 32) | - ((u64)((udp_header_t *)(ip60 + 1))->dst_port << 16); - - hash0 = lb_hash_hash(key0); - } - - //NOTE: This is an ugly trick to not include the VIP index in the hash calculation - //but actually use it in the key determination. - key0[4] |= ((vip0 - lbm->vips)); - - lb_hash_get(sticky_ht, key0, hash0, lb_time, &available_index0, &value0); - if (PREDICT_TRUE(value0 != ~0)) { - //Found an existing entry - as0 = &lbm->ass[value0]; - } else if (PREDICT_TRUE(available_index0 != ~0)) { - //There is an available slot for a new flow - as0 = &lbm->ass[vip0->new_flow_table[hash0 & vip0->new_flow_table_mask].as_index]; - if (PREDICT_FALSE(as0 == lbm->ass)) { //Special first element - error0 = LB_ERROR_NO_SERVER; - next0 = LB_NEXT_DROP; - } else { - vlib_increment_simple_counter(&lbm->vip_counters[LB_VIP_COUNTER_TRACKED_SESSION], - cpu_index, vip0 - lbm->vips, 1); - } - - //TODO: There are race conditions with as0 and vip0 manipulation. - //Configuration may be changed, vectors resized, etc... - - //Dereference previously used - vlib_refcount_add(&lbm->as_refcount, cpu_index, lb_hash_available_value(sticky_ht, available_index0), -1); - vlib_refcount_add(&lbm->as_refcount, cpu_index, as0 - lbm->ass, 1); - - //Add sticky entry - //Note that when there is no AS configured, an entry is configured anyway. - //But no configured AS is not something that should happen - lb_hash_put(sticky_ht, key0, as0 - lbm->ass, available_index0, lb_time); - } else { - //Could not store new entry in the table - as0 = &lbm->ass[vip0->new_flow_table[hash0 & vip0->new_flow_table_mask].as_index]; - vlib_increment_simple_counter(&lbm->vip_counters[LB_VIP_COUNTER_UNTRACKED_PACKET], - cpu_index, vip0 - lbm->vips, 1); - } - - //Now let's encap - if (is_encap_v4) { - ip4_header_t *ip40; - vlib_buffer_advance(p0, - sizeof(ip4_header_t) - sizeof(gre_header_t)); - ip40 = vlib_buffer_get_current(p0); - gre0 = (gre_header_t *)(ip40 + 1); - ip40->src_address = lbm->ip4_src_address; - ip40->dst_address = as0->address.ip4; - ip40->ip_version_and_header_length = 0x45; - ip40->ttl = 128; - ip40->length = clib_host_to_net_u16(len0 + sizeof(gre_header_t) + sizeof(ip4_header_t)); - ip40->protocol = IP_PROTOCOL_GRE; - ip40->checksum = ip4_header_checksum (ip40); - } else { - ip6_header_t *ip60; - vlib_buffer_advance(p0, - sizeof(ip6_header_t) - sizeof(gre_header_t)); - ip60 = vlib_buffer_get_current(p0); - gre0 = (gre_header_t *)(ip60 + 1); - ip60->dst_address = as0->address.ip6; - ip60->src_address = lbm->ip6_src_address; - ip60->hop_limit = 128; - ip60->ip_version_traffic_class_and_flow_label = clib_host_to_net_u32 (0x6<<28); - ip60->payload_length = clib_host_to_net_u16(len0 + sizeof(gre_header_t)); - ip60->protocol = IP_PROTOCOL_GRE; - } - - gre0->flags_and_version = 0; - gre0->protocol = (is_input_v4)? - clib_host_to_net_u16(0x0800): - clib_host_to_net_u16(0x86DD); - - vnet_buffer(p0)->ip.adj_index[VLIB_TX] = as0->adj_index; - next0 = (as0->adj_index != ~0)?LB_NEXT_REWRITE:next0; - - if (PREDICT_FALSE (p0->flags & VLIB_BUFFER_IS_TRACED)) - { - lb_trace_t *tr = vlib_add_trace (vm, node, p0, sizeof (*tr)); - tr->as_index = as0 - lbm->ass; - tr->vip_index = ad0->vip_index; - } - - p0->error = error_node->errors[error0]; - vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, - n_left_to_next, pi0, next0); - } - vlib_put_next_frame (vm, node, next_index, n_left_to_next); - } - - return frame->n_vectors; + /* ip_lookup_main_t *lm = (is_input_v4)?&ip4_main.lookup_main:&ip6_main.lookup_main; */ + /* lb_main_t *lbm = &lb_main; */ + /* vlib_node_runtime_t *error_node = node; */ + /* u32 n_left_from, *from, next_index, *to_next, n_left_to_next; */ + /* u32 cpu_index = os_get_cpu_number(); */ + /* u32 lb_time = lb_hash_time_now(vm); */ + + /* lb_hash_t *sticky_ht = lb_get_sticky_table(cpu_index); */ + /* from = vlib_frame_vector_args (frame); */ + /* n_left_from = frame->n_vectors; */ + /* next_index = node->cached_next_index; */ + + /* while (n_left_from > 0) */ + /* { */ + /* vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); */ + /* while (n_left_from > 0 && n_left_to_next > 0) */ + /* { */ + /* u32 pi0; */ + /* vlib_buffer_t *p0; */ + /* ip_adjacency_t *adj0; */ + /* lb_adj_data_t *ad0; */ + /* lb_vip_t *vip0; */ + /* lb_as_t *as0; */ + /* gre_header_t *gre0; */ + /* u16 len0; */ + /* u32 value0, available_index0, hash0; */ + /* u64 key0[5]; */ + /* lb_error_t error0 = LB_ERROR_NONE; */ + /* lb_next_t next0 = LB_NEXT_LOOKUP; */ + + /* if (PREDICT_TRUE(n_left_from > 1)) */ + /* { */ + /* vlib_buffer_t *p2; */ + /* p2 = vlib_get_buffer(vm, from[1]); */ + /* vlib_prefetch_buffer_header(p2, STORE); */ + /* /\* IPv4 + 8 = 28. possibly plus -40 *\/ */ + /* CLIB_PREFETCH (vlib_buffer_get_current(p2) - 40, 128, STORE); */ + /* } */ + + /* pi0 = to_next[0] = from[0]; */ + /* from += 1; */ + /* n_left_from -= 1; */ + /* to_next += 1; */ + /* n_left_to_next -= 1; */ + + /* p0 = vlib_get_buffer (vm, pi0); */ + /* adj0 = ip_get_adjacency (lm, vnet_buffer (p0)->ip.adj_index[VLIB_TX]); */ + /* ad0 = (lb_adj_data_t *) &adj0->opaque; */ + /* vip0 = pool_elt_at_index (lbm->vips, ad0->vip_index); */ + + /* if (is_input_v4) { */ + /* ip4_header_t *ip40; */ + /* ip40 = vlib_buffer_get_current (p0); */ + /* len0 = clib_net_to_host_u16(ip40->length); */ + /* key0[0] = (u64) ip40->src_address.as_u32; */ + /* key0[1] = (u64) ip40->dst_address.as_u32; */ + /* key0[2] = 0; */ + /* key0[3] = 0; */ + /* key0[4] = ((u64)((udp_header_t *)(ip40 + 1))->src_port << 32) | */ + /* ((u64)((udp_header_t *)(ip40 + 1))->dst_port << 16); */ + + /* hash0 = lb_hash_hash(key0); */ + /* } else { */ + /* ip6_header_t *ip60; */ + /* ip60 = vlib_buffer_get_current (p0); */ + /* len0 = clib_net_to_host_u16(ip60->payload_length) + sizeof(ip6_header_t); */ + /* key0[0] = ip60->src_address.as_u64[0]; */ + /* key0[1] = ip60->src_address.as_u64[1]; */ + /* key0[2] = ip60->dst_address.as_u64[0]; */ + /* key0[3] = ip60->dst_address.as_u64[1]; */ + /* key0[4] = ((u64)((udp_header_t *)(ip60 + 1))->src_port << 32) | */ + /* ((u64)((udp_header_t *)(ip60 + 1))->dst_port << 16); */ + + /* hash0 = lb_hash_hash(key0); */ + /* } */ + + /* //NOTE: This is an ugly trick to not include the VIP index in the hash calculation */ + /* //but actually use it in the key determination. */ + /* key0[4] |= ((vip0 - lbm->vips)); */ + + /* lb_hash_get(sticky_ht, key0, hash0, lb_time, &available_index0, &value0); */ + /* if (PREDICT_TRUE(value0 != ~0)) { */ + /* //Found an existing entry */ + /* as0 = &lbm->ass[value0]; */ + /* } else if (PREDICT_TRUE(available_index0 != ~0)) { */ + /* //There is an available slot for a new flow */ + /* as0 = &lbm->ass[vip0->new_flow_table[hash0 & vip0->new_flow_table_mask].as_index]; */ + /* if (PREDICT_FALSE(as0 == lbm->ass)) { //Special first element */ + /* error0 = LB_ERROR_NO_SERVER; */ + /* next0 = LB_NEXT_DROP; */ + /* } else { */ + /* vlib_increment_simple_counter(&lbm->vip_counters[LB_VIP_COUNTER_TRACKED_SESSION], */ + /* cpu_index, vip0 - lbm->vips, 1); */ + /* } */ + + /* //TODO: There are race conditions with as0 and vip0 manipulation. */ + /* //Configuration may be changed, vectors resized, etc... */ + + /* //Dereference previously used */ + /* vlib_refcount_add(&lbm->as_refcount, cpu_index, lb_hash_available_value(sticky_ht, available_index0), -1); */ + /* vlib_refcount_add(&lbm->as_refcount, cpu_index, as0 - lbm->ass, 1); */ + + /* //Add sticky entry */ + /* //Note that when there is no AS configured, an entry is configured anyway. */ + /* //But no configured AS is not something that should happen */ + /* lb_hash_put(sticky_ht, key0, as0 - lbm->ass, available_index0, lb_time); */ + /* } else { */ + /* //Could not store new entry in the table */ + /* as0 = &lbm->ass[vip0->new_flow_table[hash0 & vip0->new_flow_table_mask].as_index]; */ + /* vlib_increment_simple_counter(&lbm->vip_counters[LB_VIP_COUNTER_UNTRACKED_PACKET], */ + /* cpu_index, vip0 - lbm->vips, 1); */ + /* } */ + + /* //Now let's encap */ + /* if (is_encap_v4) { */ + /* ip4_header_t *ip40; */ + /* vlib_buffer_advance(p0, - sizeof(ip4_header_t) - sizeof(gre_header_t)); */ + /* ip40 = vlib_buffer_get_current(p0); */ + /* gre0 = (gre_header_t *)(ip40 + 1); */ + /* ip40->src_address = lbm->ip4_src_address; */ + /* ip40->dst_address = as0->address.ip4; */ + /* ip40->ip_version_and_header_length = 0x45; */ + /* ip40->ttl = 128; */ + /* ip40->length = clib_host_to_net_u16(len0 + sizeof(gre_header_t) + sizeof(ip4_header_t)); */ + /* ip40->protocol = IP_PROTOCOL_GRE; */ + /* ip40->checksum = ip4_header_checksum (ip40); */ + /* } else { */ + /* ip6_header_t *ip60; */ + /* vlib_buffer_advance(p0, - sizeof(ip6_header_t) - sizeof(gre_header_t)); */ + /* ip60 = vlib_buffer_get_current(p0); */ + /* gre0 = (gre_header_t *)(ip60 + 1); */ + /* ip60->dst_address = as0->address.ip6; */ + /* ip60->src_address = lbm->ip6_src_address; */ + /* ip60->hop_limit = 128; */ + /* ip60->ip_version_traffic_class_and_flow_label = clib_host_to_net_u32 (0x6<<28); */ + /* ip60->payload_length = clib_host_to_net_u16(len0 + sizeof(gre_header_t)); */ + /* ip60->protocol = IP_PROTOCOL_GRE; */ + /* } */ + + /* gre0->flags_and_version = 0; */ + /* gre0->protocol = (is_input_v4)? */ + /* clib_host_to_net_u16(0x0800): */ + /* clib_host_to_net_u16(0x86DD); */ + + /* vnet_buffer(p0)->ip.adj_index[VLIB_TX] = as0->adj_index; */ + /* next0 = (as0->adj_index != ~0)?LB_NEXT_REWRITE:next0; */ + + /* if (PREDICT_FALSE (p0->flags & VLIB_BUFFER_IS_TRACED)) */ + /* { */ + /* lb_trace_t *tr = vlib_add_trace (vm, node, p0, sizeof (*tr)); */ + /* tr->as_index = as0 - lbm->ass; */ + /* tr->vip_index = ad0->vip_index; */ + /* } */ + + /* p0->error = error_node->errors[error0]; */ + /* vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next, */ + /* n_left_to_next, pi0, next0); */ + /* } */ + /* vlib_put_next_frame (vm, node, next_index, n_left_to_next); */ + /* } */ + + /* return frame->n_vectors; */ + return 0; } static uword |