net/colo.c

   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 "net/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 fill_connection_key(Packet *pkt, ConnectionKey *key)
  86 {
  87     uint32_t tmp_ports;
  88
  89     memset(key, 0, sizeof(*key));
  90     key->ip_proto = pkt->ip->ip_p;
  91
  92     switch (key->ip_proto) {
  93     case IPPROTO_TCP:
  94     case IPPROTO_UDP:
  95     case IPPROTO_DCCP:
  96     case IPPROTO_ESP:
  97     case IPPROTO_SCTP:
  98     case IPPROTO_UDPLITE:
  99         tmp_ports = *(uint32_t *)(pkt->transport_header);
 100         key->src = pkt->ip->ip_src;
 101         key->dst = pkt->ip->ip_dst;
 102         key->src_port = ntohs(tmp_ports & 0xffff);
 103         key->dst_port = ntohs(tmp_ports >> 16);
 104         break;
 105     case IPPROTO_AH:
 106         tmp_ports = *(uint32_t *)(pkt->transport_header + 4);
 107         key->src = pkt->ip->ip_src;
 108         key->dst = pkt->ip->ip_dst;
 109         key->src_port = ntohs(tmp_ports & 0xffff);
 110         key->dst_port = ntohs(tmp_ports >> 16);
 111         break;
 112     default:
 113         break;
 114     }
 115 }
 116
 117 void reverse_connection_key(ConnectionKey *key)
 118 {
 119     struct in_addr tmp_ip;
 120     uint16_t tmp_port;
 121
 122     tmp_ip = key->src;
 123     key->src = key->dst;
 124     key->dst = tmp_ip;
 125
 126     tmp_port = key->src_port;
 127     key->src_port = key->dst_port;
 128     key->dst_port = tmp_port;
 129 }
 130
 131 Connection *connection_new(ConnectionKey *key)
 132 {
 133     Connection *conn = g_slice_new(Connection);
 134
 135     conn->ip_proto = key->ip_proto;
 136     conn->processing = false;
 137     conn->offset = 0;
 138     conn->syn_flag = 0;
 139     g_queue_init(&conn->primary_list);
 140     g_queue_init(&conn->secondary_list);
 141
 142     return conn;
 143 }
 144
 145 void connection_destroy(void *opaque)
 146 {
 147     Connection *conn = opaque;
 148
 149     g_queue_foreach(&conn->primary_list, packet_destroy, NULL);
 150     g_queue_clear(&conn->primary_list);
 151     g_queue_foreach(&conn->secondary_list, packet_destroy, NULL);
 152     g_queue_clear(&conn->secondary_list);
 153     g_slice_free(Connection, conn);
 154 }
 155
 156 Packet *packet_new(const void *data, int size, int vnet_hdr_len)
 157 {
 158     Packet *pkt = g_slice_new(Packet);
 159
 160     pkt->data = g_memdup(data, size);
 161     pkt->size = size;
 162     pkt->creation_ms = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 163     pkt->vnet_hdr_len = vnet_hdr_len;
 164
 165     return pkt;
 166 }
 167
 168 void packet_destroy(void *opaque, void *user_data)
 169 {
 170     Packet *pkt = opaque;
 171
 172     g_free(pkt->data);
 173     g_slice_free(Packet, pkt);
 174 }
 175
 176 /*
 177  * Clear hashtable, stop this hash growing really huge
 178  */
 179 void connection_hashtable_reset(GHashTable *connection_track_table)
 180 {
 181     g_hash_table_remove_all(connection_track_table);
 182 }
 183
 184 /* if not found, create a new connection and add to hash table */
 185 Connection *connection_get(GHashTable *connection_track_table,
 186                            ConnectionKey *key,
 187                            GQueue *conn_list)
 188 {
 189     Connection *conn = g_hash_table_lookup(connection_track_table, key);
 190
 191     if (conn == NULL) {
 192         ConnectionKey *new_key = g_memdup(key, sizeof(*key));
 193
 194         conn = connection_new(key);
 195
 196         if (g_hash_table_size(connection_track_table) > HASHTABLE_MAX_SIZE) {
 197             trace_colo_proxy_main("colo proxy connection hashtable full,"
 198                                   " clear it");
 199             connection_hashtable_reset(connection_track_table);
 200             /*
 201              * clear the conn_list
 202              */
 203             while (!g_queue_is_empty(conn_list)) {
 204                 connection_destroy(g_queue_pop_head(conn_list));
 205             }
 206         }
 207
 208         g_hash_table_insert(connection_track_table, new_key, conn);
 209     }
 210
 211     return conn;
 212 }