net/filter-rewriter.c

   1 /*
   2  * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
   3  * Copyright (c) 2016 FUJITSU LIMITED
   4  * Copyright (c) 2016 Intel Corporation
   5  *
   6  * Author: Zhang Chen <[email protected]>
   7  *
   8  * This work is licensed under the terms of the GNU GPL, version 2 or
   9  * later.  See the COPYING file in the top-level directory.
  10  */
  11
  12 #include "qemu/osdep.h"
  13 #include "trace.h"
  14 #include "net/colo.h"
  15 #include "net/filter.h"
  16 #include "net/net.h"
  17 #include "qemu-common.h"
  18 #include "qemu/error-report.h"
  19 #include "qom/object.h"
  20 #include "qemu/main-loop.h"
  21 #include "qemu/iov.h"
  22 #include "net/checksum.h"
  23
  24 #define FILTER_COLO_REWRITER(obj) \
  25     OBJECT_CHECK(RewriterState, (obj), TYPE_FILTER_REWRITER)
  26
  27 #define TYPE_FILTER_REWRITER "filter-rewriter"
  28
  29 typedef struct RewriterState {
  30     NetFilterState parent_obj;
  31     NetQueue *incoming_queue;
  32     /* hashtable to save connection */
  33     GHashTable *connection_track_table;
  34     bool vnet_hdr;
  35 } RewriterState;
  36
  37 static void filter_rewriter_flush(NetFilterState *nf)
  38 {
  39     RewriterState *s = FILTER_COLO_REWRITER(nf);
  40
  41     if (!qemu_net_queue_flush(s->incoming_queue)) {
  42         /* Unable to empty the queue, purge remaining packets */
  43         qemu_net_queue_purge(s->incoming_queue, nf->netdev);
  44     }
  45 }
  46
  47 /*
  48  * Return 1 on success, if return 0 means the pkt
  49  * is not TCP packet
  50  */
  51 static int is_tcp_packet(Packet *pkt)
  52 {
  53     if (!parse_packet_early(pkt) &&
  54         pkt->ip->ip_p == IPPROTO_TCP) {
  55         return 1;
  56     } else {
  57         return 0;
  58     }
  59 }
  60
  61 /* handle tcp packet from primary guest */
  62 static int handle_primary_tcp_pkt(NetFilterState *nf,
  63                                   Connection *conn,
  64                                   Packet *pkt)
  65 {
  66     struct tcphdr *tcp_pkt;
  67
  68     tcp_pkt = (struct tcphdr *)pkt->transport_header;
  69     if (trace_event_get_state_backends(TRACE_COLO_FILTER_REWRITER_DEBUG)) {
  70         trace_colo_filter_rewriter_pkt_info(__func__,
  71                     inet_ntoa(pkt->ip->ip_src), inet_ntoa(pkt->ip->ip_dst),
  72                     ntohl(tcp_pkt->th_seq), ntohl(tcp_pkt->th_ack),
  73                     tcp_pkt->th_flags);
  74         trace_colo_filter_rewriter_conn_offset(conn->offset);
  75     }
  76
  77     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_SYN)) {
  78         /*
  79          * we use this flag update offset func
  80          * run once in independent tcp connection
  81          */
  82         conn->syn_flag = 1;
  83     }
  84
  85     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_ACK)) {
  86         if (conn->syn_flag) {
  87             /*
  88              * offset = secondary_seq - primary seq
  89              * ack packet sent by guest from primary node,
  90              * so we use th_ack - 1 get primary_seq
  91              */
  92             conn->offset -= (ntohl(tcp_pkt->th_ack) - 1);
  93             conn->syn_flag = 0;
  94         }
  95         if (conn->offset) {
  96             /* handle packets to the secondary from the primary */
  97             tcp_pkt->th_ack = htonl(ntohl(tcp_pkt->th_ack) + conn->offset);
  98
  99             net_checksum_calculate((uint8_t *)pkt->data + pkt->vnet_hdr_len,
 100                                    pkt->size - pkt->vnet_hdr_len);
 101         }
 102     }
 103
 104     return 0;
 105 }
 106
 107 /* handle tcp packet from secondary guest */
 108 static int handle_secondary_tcp_pkt(NetFilterState *nf,
 109                                     Connection *conn,
 110                                     Packet *pkt)
 111 {
 112     struct tcphdr *tcp_pkt;
 113
 114     tcp_pkt = (struct tcphdr *)pkt->transport_header;
 115
 116     if (trace_event_get_state_backends(TRACE_COLO_FILTER_REWRITER_DEBUG)) {
 117         trace_colo_filter_rewriter_pkt_info(__func__,
 118                     inet_ntoa(pkt->ip->ip_src), inet_ntoa(pkt->ip->ip_dst),
 119                     ntohl(tcp_pkt->th_seq), ntohl(tcp_pkt->th_ack),
 120                     tcp_pkt->th_flags);
 121         trace_colo_filter_rewriter_conn_offset(conn->offset);
 122     }
 123
 124     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == (TH_ACK | TH_SYN))) {
 125         /*
 126          * save offset = secondary_seq and then
 127          * in handle_primary_tcp_pkt make offset
 128          * = secondary_seq - primary_seq
 129          */
 130         conn->offset = ntohl(tcp_pkt->th_seq);
 131     }
 132
 133     if ((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_ACK) {
 134         /* Only need to adjust seq while offset is Non-zero */
 135         if (conn->offset) {
 136             /* handle packets to the primary from the secondary*/
 137             tcp_pkt->th_seq = htonl(ntohl(tcp_pkt->th_seq) - conn->offset);
 138
 139             net_checksum_calculate((uint8_t *)pkt->data + pkt->vnet_hdr_len,
 140                                    pkt->size - pkt->vnet_hdr_len);
 141         }
 142     }
 143
 144     return 0;
 145 }
 146
 147 static ssize_t colo_rewriter_receive_iov(NetFilterState *nf,
 148                                          NetClientState *sender,
 149                                          unsigned flags,
 150                                          const struct iovec *iov,
 151                                          int iovcnt,
 152                                          NetPacketSent *sent_cb)
 153 {
 154     RewriterState *s = FILTER_COLO_REWRITER(nf);
 155     Connection *conn;
 156     ConnectionKey key;
 157     Packet *pkt;
 158     ssize_t size = iov_size(iov, iovcnt);
 159     ssize_t vnet_hdr_len = 0;
 160     char *buf = g_malloc0(size);
 161
 162     iov_to_buf(iov, iovcnt, 0, buf, size);
 163
 164     if (s->vnet_hdr) {
 165         vnet_hdr_len = nf->netdev->vnet_hdr_len;
 166     }
 167
 168     pkt = packet_new(buf, size, vnet_hdr_len);
 169     g_free(buf);
 170
 171     /*
 172      * if we get tcp packet
 173      * we will rewrite it to make secondary guest's
 174      * connection established successfully
 175      */
 176     if (pkt && is_tcp_packet(pkt)) {
 177
 178         fill_connection_key(pkt, &key);
 179
 180         if (sender == nf->netdev) {
 181             /*
 182              * We need make tcp TX and RX packet
 183              * into one connection.
 184              */
 185             reverse_connection_key(&key);
 186         }
 187         conn = connection_get(s->connection_track_table,
 188                               &key,
 189                               NULL);
 190
 191         if (sender == nf->netdev) {
 192             /* NET_FILTER_DIRECTION_TX */
 193             if (!handle_primary_tcp_pkt(nf, conn, pkt)) {
 194                 qemu_net_queue_send(s->incoming_queue, sender, 0,
 195                 (const uint8_t *)pkt->data, pkt->size, NULL);
 196                 packet_destroy(pkt, NULL);
 197                 pkt = NULL;
 198                 /*
 199                  * We block the packet here,after rewrite pkt
 200                  * and will send it
 201                  */
 202                 return 1;
 203             }
 204         } else {
 205             /* NET_FILTER_DIRECTION_RX */
 206             if (!handle_secondary_tcp_pkt(nf, conn, pkt)) {
 207                 qemu_net_queue_send(s->incoming_queue, sender, 0,
 208                 (const uint8_t *)pkt->data, pkt->size, NULL);
 209                 packet_destroy(pkt, NULL);
 210                 pkt = NULL;
 211                 /*
 212                  * We block the packet here,after rewrite pkt
 213                  * and will send it
 214                  */
 215                 return 1;
 216             }
 217         }
 218     }
 219
 220     packet_destroy(pkt, NULL);
 221     pkt = NULL;
 222     return 0;
 223 }
 224
 225 static void colo_rewriter_cleanup(NetFilterState *nf)
 226 {
 227     RewriterState *s = FILTER_COLO_REWRITER(nf);
 228
 229     /* flush packets */
 230     if (s->incoming_queue) {
 231         filter_rewriter_flush(nf);
 232         g_free(s->incoming_queue);
 233     }
 234 }
 235
 236 static void colo_rewriter_setup(NetFilterState *nf, Error **errp)
 237 {
 238     RewriterState *s = FILTER_COLO_REWRITER(nf);
 239
 240     s->connection_track_table = g_hash_table_new_full(connection_key_hash,
 241                                                       connection_key_equal,
 242                                                       g_free,
 243                                                       connection_destroy);
 244     s->incoming_queue = qemu_new_net_queue(qemu_netfilter_pass_to_next, nf);
 245 }
 246
 247 static bool filter_rewriter_get_vnet_hdr(Object *obj, Error **errp)
 248 {
 249     RewriterState *s = FILTER_COLO_REWRITER(obj);
 250
 251     return s->vnet_hdr;
 252 }
 253
 254 static void filter_rewriter_set_vnet_hdr(Object *obj,
 255                                          bool value,
 256                                          Error **errp)
 257 {
 258     RewriterState *s = FILTER_COLO_REWRITER(obj);
 259
 260     s->vnet_hdr = value;
 261 }
 262
 263 static void filter_rewriter_init(Object *obj)
 264 {
 265     RewriterState *s = FILTER_COLO_REWRITER(obj);
 266
 267     s->vnet_hdr = false;
 268     object_property_add_bool(obj, "vnet_hdr_support",
 269                              filter_rewriter_get_vnet_hdr,
 270                              filter_rewriter_set_vnet_hdr, NULL);
 271 }
 272
 273 static void colo_rewriter_class_init(ObjectClass *oc, void *data)
 274 {
 275     NetFilterClass *nfc = NETFILTER_CLASS(oc);
 276
 277     nfc->setup = colo_rewriter_setup;
 278     nfc->cleanup = colo_rewriter_cleanup;
 279     nfc->receive_iov = colo_rewriter_receive_iov;
 280 }
 281
 282 static const TypeInfo colo_rewriter_info = {
 283     .name = TYPE_FILTER_REWRITER,
 284     .parent = TYPE_NETFILTER,
 285     .class_init = colo_rewriter_class_init,
 286     .instance_init = filter_rewriter_init,
 287     .instance_size = sizeof(RewriterState),
 288 };
 289
 290 static void register_types(void)
 291 {
 292     type_register_static(&colo_rewriter_info);
 293 }
 294
 295 type_init(register_types);