1 /* 2 * COarse-grain LOck-stepping Virtual Machines for Non-stop Service (COLO) 3 * (a.k.a. Fault Tolerance or Continuous Replication) 4 * 5 * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD. 6 * Copyright (c) 2016 FUJITSU LIMITED 7 * Copyright (c) 2016 Intel Corporation 8 * 9 * Author: Zhang Chen <zhangchen.fnst@cn.fujitsu.com> 10 * 11 * This work is licensed under the terms of the GNU GPL, version 2 or 12 * later. See the COPYING file in the top-level directory. 13 */ 14 15 #include "qemu/osdep.h" 16 #include "trace.h" 17 #include "colo.h" 18 19 uint32_t connection_key_hash(const void *opaque) 20 { 21 const ConnectionKey *key = opaque; 22 uint32_t a, b, c; 23 24 /* Jenkins hash */ 25 a = b = c = JHASH_INITVAL + sizeof(*key); 26 a += key->src.s_addr; 27 b += key->dst.s_addr; 28 c += (key->src_port | key->dst_port << 16); 29 __jhash_mix(a, b, c); 30 31 a += key->ip_proto; 32 __jhash_final(a, b, c); 33 34 return c; 35 } 36 37 int connection_key_equal(const void *key1, const void *key2) 38 { 39 return memcmp(key1, key2, sizeof(ConnectionKey)) == 0; 40 } 41 42 int parse_packet_early(Packet *pkt) 43 { 44 int network_length; 45 static const uint8_t vlan[] = {0x81, 0x00}; 46 uint8_t *data = pkt->data + pkt->vnet_hdr_len; 47 uint16_t l3_proto; 48 ssize_t l2hdr_len = eth_get_l2_hdr_length(data); 49 50 if (pkt->size < ETH_HLEN + pkt->vnet_hdr_len) { 51 trace_colo_proxy_main("pkt->size < ETH_HLEN"); 52 return 1; 53 } 54 55 /* 56 * TODO: support vlan. 57 */ 58 if (!memcmp(&data[12], vlan, sizeof(vlan))) { 59 trace_colo_proxy_main("COLO-proxy don't support vlan"); 60 return 1; 61 } 62 63 pkt->network_header = data + l2hdr_len; 64 65 const struct iovec l2vec = { 66 .iov_base = (void *) data, 67 .iov_len = l2hdr_len 68 }; 69 l3_proto = eth_get_l3_proto(&l2vec, 1, l2hdr_len); 70 71 if (l3_proto != ETH_P_IP) { 72 return 1; 73 } 74 75 network_length = pkt->ip->ip_hl * 4; 76 if (pkt->size < l2hdr_len + network_length + pkt->vnet_hdr_len) { 77 trace_colo_proxy_main("pkt->size < network_header + network_length"); 78 return 1; 79 } 80 pkt->transport_header = pkt->network_header + network_length; 81 82 return 0; 83 } 84 85 void extract_ip_and_port(uint32_t tmp_ports, ConnectionKey *key, Packet *pkt) 86 { 87 key->src = pkt->ip->ip_src; 88 key->dst = pkt->ip->ip_dst; 89 key->src_port = ntohs(tmp_ports >> 16); 90 key->dst_port = ntohs(tmp_ports & 0xffff); 91 } 92 93 void fill_connection_key(Packet *pkt, ConnectionKey *key) 94 { 95 uint32_t tmp_ports; 96 97 memset(key, 0, sizeof(*key)); 98 key->ip_proto = pkt->ip->ip_p; 99 100 switch (key->ip_proto) { 101 case IPPROTO_TCP: 102 case IPPROTO_UDP: 103 case IPPROTO_DCCP: 104 case IPPROTO_ESP: 105 case IPPROTO_SCTP: 106 case IPPROTO_UDPLITE: 107 tmp_ports = *(uint32_t *)(pkt->transport_header); 108 extract_ip_and_port(tmp_ports, key, pkt); 109 break; 110 case IPPROTO_AH: 111 tmp_ports = *(uint32_t *)(pkt->transport_header + 4); 112 extract_ip_and_port(tmp_ports, key, pkt); 113 break; 114 default: 115 break; 116 } 117 } 118 119 void reverse_connection_key(ConnectionKey *key) 120 { 121 struct in_addr tmp_ip; 122 uint16_t tmp_port; 123 124 tmp_ip = key->src; 125 key->src = key->dst; 126 key->dst = tmp_ip; 127 128 tmp_port = key->src_port; 129 key->src_port = key->dst_port; 130 key->dst_port = tmp_port; 131 } 132 133 Connection *connection_new(ConnectionKey *key) 134 { 135 Connection *conn = g_slice_new(Connection); 136 137 conn->ip_proto = key->ip_proto; 138 conn->processing = false; 139 conn->offset = 0; 140 conn->syn_flag = 0; 141 conn->pack = 0; 142 conn->sack = 0; 143 g_queue_init(&conn->primary_list); 144 g_queue_init(&conn->secondary_list); 145 146 return conn; 147 } 148 149 void connection_destroy(void *opaque) 150 { 151 Connection *conn = opaque; 152 153 g_queue_foreach(&conn->primary_list, packet_destroy, NULL); 154 g_queue_clear(&conn->primary_list); 155 g_queue_foreach(&conn->secondary_list, packet_destroy, NULL); 156 g_queue_clear(&conn->secondary_list); 157 g_slice_free(Connection, conn); 158 } 159 160 Packet *packet_new(const void *data, int size, int vnet_hdr_len) 161 { 162 Packet *pkt = g_slice_new(Packet); 163 164 pkt->data = g_memdup(data, size); 165 pkt->size = size; 166 pkt->creation_ms = qemu_clock_get_ms(QEMU_CLOCK_HOST); 167 pkt->vnet_hdr_len = vnet_hdr_len; 168 pkt->tcp_seq = 0; 169 pkt->tcp_ack = 0; 170 pkt->seq_end = 0; 171 pkt->header_size = 0; 172 pkt->payload_size = 0; 173 pkt->offset = 0; 174 pkt->flags = 0; 175 176 return pkt; 177 } 178 179 void packet_destroy(void *opaque, void *user_data) 180 { 181 Packet *pkt = opaque; 182 183 g_free(pkt->data); 184 g_slice_free(Packet, pkt); 185 } 186 187 /* 188 * Clear hashtable, stop this hash growing really huge 189 */ 190 void connection_hashtable_reset(GHashTable *connection_track_table) 191 { 192 g_hash_table_remove_all(connection_track_table); 193 } 194 195 /* if not found, create a new connection and add to hash table */ 196 Connection *connection_get(GHashTable *connection_track_table, 197 ConnectionKey *key, 198 GQueue *conn_list) 199 { 200 Connection *conn = g_hash_table_lookup(connection_track_table, key); 201 202 if (conn == NULL) { 203 ConnectionKey *new_key = g_memdup(key, sizeof(*key)); 204 205 conn = connection_new(key); 206 207 if (g_hash_table_size(connection_track_table) > HASHTABLE_MAX_SIZE) { 208 trace_colo_proxy_main("colo proxy connection hashtable full," 209 " clear it"); 210 connection_hashtable_reset(connection_track_table); 211 /* 212 * clear the conn_list 213 */ 214 while (!g_queue_is_empty(conn_list)) { 215 connection_destroy(g_queue_pop_head(conn_list)); 216 } 217 } 218 219 g_hash_table_insert(connection_track_table, new_key, conn); 220 } 221 222 return conn; 223 } 224