| 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 | #include "util.h" |
| 19 | |
| 20 | uint32_t connection_key_hash(const void *opaque) |
| 21 | { |
| 22 | const ConnectionKey *key = opaque; |
| 23 | uint32_t a, b, c; |
| 24 | |
| 25 | /* Jenkins hash */ |
| 26 | a = b = c = JHASH_INITVAL + sizeof(*key); |
| 27 | a += key->src.s_addr; |
| 28 | b += key->dst.s_addr; |
| 29 | c += (key->src_port | key->dst_port << 16); |
| 30 | __jhash_mix(a, b, c); |
| 31 | |
| 32 | a += key->ip_proto; |
| 33 | __jhash_final(a, b, c); |
| 34 | |
| 35 | return c; |
| 36 | } |
| 37 | |
| 38 | int connection_key_equal(const void *key1, const void *key2) |
| 39 | { |
| 40 | return memcmp(key1, key2, sizeof(ConnectionKey)) == 0; |
| 41 | } |
| 42 | |
| 43 | int parse_packet_early(Packet *pkt) |
| 44 | { |
| 45 | int network_length; |
| 46 | static const uint8_t vlan[] = {0x81, 0x00}; |
| 47 | uint8_t *data = pkt->data; |
| 48 | uint16_t l3_proto; |
| 49 | ssize_t l2hdr_len; |
| 50 | |
| 51 | assert(data); |
| 52 | |
| 53 | /* Check the received vnet_hdr_len then add the offset */ |
| 54 | if ((pkt->vnet_hdr_len > sizeof(struct virtio_net_hdr_v1_hash)) || |
| 55 | (pkt->size < sizeof(struct eth_header) + sizeof(struct vlan_header) + |
| 56 | pkt->vnet_hdr_len)) { |
| 57 | /* |
| 58 | * The received remote packet maybe misconfiguration here, |
| 59 | * Please enable/disable filter module's the vnet_hdr flag at |
| 60 | * the same time. |
| 61 | */ |
| 62 | trace_colo_proxy_main_vnet_info("This received packet load wrong ", |
| 63 | pkt->vnet_hdr_len, pkt->size); |
| 64 | return 1; |
| 65 | } |
| 66 | data += pkt->vnet_hdr_len; |
| 67 | |
| 68 | l2hdr_len = eth_get_l2_hdr_length(data); |
| 69 | |
| 70 | /* |
| 71 | * TODO: support vlan. |
| 72 | */ |
| 73 | if (!memcmp(&data[12], vlan, sizeof(vlan))) { |
| 74 | trace_colo_proxy_main("COLO-proxy don't support vlan"); |
| 75 | return 1; |
| 76 | } |
| 77 | |
| 78 | pkt->network_header = data + l2hdr_len; |
| 79 | |
| 80 | const struct iovec l2vec = { |
| 81 | .iov_base = (void *) data, |
| 82 | .iov_len = l2hdr_len |
| 83 | }; |
| 84 | l3_proto = eth_get_l3_proto(&l2vec, 1, l2hdr_len); |
| 85 | |
| 86 | if (l3_proto != ETH_P_IP) { |
| 87 | return 1; |
| 88 | } |
| 89 | |
| 90 | network_length = pkt->ip->ip_hl * 4; |
| 91 | if (pkt->size < l2hdr_len + network_length + pkt->vnet_hdr_len) { |
| 92 | trace_colo_proxy_main("pkt->size < network_header + network_length"); |
| 93 | return 1; |
| 94 | } |
| 95 | pkt->transport_header = pkt->network_header + network_length; |
| 96 | |
| 97 | return 0; |
| 98 | } |
| 99 | |
| 100 | void extract_ip_and_port(uint32_t tmp_ports, ConnectionKey *key, |
| 101 | Packet *pkt, bool reverse) |
| 102 | { |
| 103 | if (reverse) { |
| 104 | key->src = pkt->ip->ip_dst; |
| 105 | key->dst = pkt->ip->ip_src; |
| 106 | key->src_port = ntohs(tmp_ports & 0xffff); |
| 107 | key->dst_port = ntohs(tmp_ports >> 16); |
| 108 | } else { |
| 109 | key->src = pkt->ip->ip_src; |
| 110 | key->dst = pkt->ip->ip_dst; |
| 111 | key->src_port = ntohs(tmp_ports >> 16); |
| 112 | key->dst_port = ntohs(tmp_ports & 0xffff); |
| 113 | } |
| 114 | } |
| 115 | |
| 116 | void fill_connection_key(Packet *pkt, ConnectionKey *key, bool reverse) |
| 117 | { |
| 118 | uint32_t tmp_ports = 0; |
| 119 | |
| 120 | key->ip_proto = pkt->ip->ip_p; |
| 121 | |
| 122 | switch (key->ip_proto) { |
| 123 | case IPPROTO_TCP: |
| 124 | case IPPROTO_UDP: |
| 125 | case IPPROTO_DCCP: |
| 126 | case IPPROTO_ESP: |
| 127 | case IPPROTO_SCTP: |
| 128 | case IPPROTO_UDPLITE: |
| 129 | tmp_ports = *(uint32_t *)(pkt->transport_header); |
| 130 | break; |
| 131 | case IPPROTO_AH: |
| 132 | tmp_ports = *(uint32_t *)(pkt->transport_header + 4); |
| 133 | break; |
| 134 | default: |
| 135 | break; |
| 136 | } |
| 137 | |
| 138 | extract_ip_and_port(tmp_ports, key, pkt, reverse); |
| 139 | } |
| 140 | |
| 141 | Connection *connection_new(ConnectionKey *key) |
| 142 | { |
| 143 | Connection *conn = g_slice_new0(Connection); |
| 144 | |
| 145 | conn->ip_proto = key->ip_proto; |
| 146 | conn->processing = false; |
| 147 | conn->tcp_state = TCPS_CLOSED; |
| 148 | g_queue_init(&conn->primary_list); |
| 149 | g_queue_init(&conn->secondary_list); |
| 150 | |
| 151 | return conn; |
| 152 | } |
| 153 | |
| 154 | void connection_destroy(void *opaque) |
| 155 | { |
| 156 | Connection *conn = opaque; |
| 157 | |
| 158 | g_queue_foreach(&conn->primary_list, packet_destroy, NULL); |
| 159 | g_queue_clear(&conn->primary_list); |
| 160 | g_queue_foreach(&conn->secondary_list, packet_destroy, NULL); |
| 161 | g_queue_clear(&conn->secondary_list); |
| 162 | g_slice_free(Connection, conn); |
| 163 | } |
| 164 | |
| 165 | Packet *packet_new(const void *data, int size, int vnet_hdr_len) |
| 166 | { |
| 167 | Packet *pkt = g_slice_new0(Packet); |
| 168 | |
| 169 | pkt->data = g_memdup(data, size); |
| 170 | pkt->size = size; |
| 171 | pkt->creation_ms = qemu_clock_get_ms(QEMU_CLOCK_HOST); |
| 172 | pkt->vnet_hdr_len = vnet_hdr_len; |
| 173 | |
| 174 | return pkt; |
| 175 | } |
| 176 | |
| 177 | /* |
| 178 | * packet_new_nocopy will not copy data, so the caller can't release |
| 179 | * the data. And it will be released in packet_destroy. |
| 180 | */ |
| 181 | Packet *packet_new_nocopy(void *data, int size, int vnet_hdr_len) |
| 182 | { |
| 183 | Packet *pkt = g_slice_new0(Packet); |
| 184 | |
| 185 | pkt->data = data; |
| 186 | pkt->size = size; |
| 187 | pkt->creation_ms = qemu_clock_get_ms(QEMU_CLOCK_HOST); |
| 188 | pkt->vnet_hdr_len = vnet_hdr_len; |
| 189 | |
| 190 | return pkt; |
| 191 | } |
| 192 | |
| 193 | void packet_destroy(void *opaque, void *user_data) |
| 194 | { |
| 195 | Packet *pkt = opaque; |
| 196 | |
| 197 | g_free(pkt->data); |
| 198 | g_slice_free(Packet, pkt); |
| 199 | } |
| 200 | |
| 201 | void packet_destroy_partial(void *opaque, void *user_data) |
| 202 | { |
| 203 | Packet *pkt = opaque; |
| 204 | |
| 205 | g_slice_free(Packet, pkt); |
| 206 | } |
| 207 | |
| 208 | /* |
| 209 | * Clear hashtable, stop this hash growing really huge |
| 210 | */ |
| 211 | void connection_hashtable_reset(GHashTable *connection_track_table) |
| 212 | { |
| 213 | g_hash_table_remove_all(connection_track_table); |
| 214 | } |
| 215 | |
| 216 | /* if not found, create a new connection and add to hash table */ |
| 217 | Connection *connection_get(GHashTable *connection_track_table, |
| 218 | ConnectionKey *key, |
| 219 | GQueue *conn_list) |
| 220 | { |
| 221 | Connection *conn = g_hash_table_lookup(connection_track_table, key); |
| 222 | |
| 223 | if (conn == NULL) { |
| 224 | ConnectionKey *new_key = g_memdup(key, sizeof(*key)); |
| 225 | |
| 226 | conn = connection_new(key); |
| 227 | |
| 228 | if (g_hash_table_size(connection_track_table) > HASHTABLE_MAX_SIZE) { |
| 229 | trace_colo_proxy_main("colo proxy connection hashtable full," |
| 230 | " clear it"); |
| 231 | connection_hashtable_reset(connection_track_table); |
| 232 | /* |
| 233 | * clear the conn_list |
| 234 | */ |
| 235 | while (conn_list && !g_queue_is_empty(conn_list)) { |
| 236 | connection_destroy(g_queue_pop_head(conn_list)); |
| 237 | } |
| 238 | } |
| 239 | |
| 240 | g_hash_table_insert(connection_track_table, new_key, conn); |
| 241 | } |
| 242 | |
| 243 | return conn; |
| 244 | } |
| 245 | |
| 246 | bool connection_has_tracked(GHashTable *connection_track_table, |
| 247 | ConnectionKey *key) |
| 248 | { |
| 249 | Connection *conn = g_hash_table_lookup(connection_track_table, key); |
| 250 | |
| 251 | return conn ? true : false; |
| 252 | } |