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 <zhangchen.fnst@cn.fujitsu.com>
   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 "qapi/error.h"
  19 #include "qapi/qmp/qerror.h"
  20 #include "qapi-visit.h"
  21 #include "qom/object.h"
  22 #include "qemu/main-loop.h"
  23 #include "qemu/iov.h"
  24 #include "net/checksum.h"
  25
  26 #define FILTER_COLO_REWRITER(obj) \
  27     OBJECT_CHECK(RewriterState, (obj), TYPE_FILTER_REWRITER)
  28
  29 #define TYPE_FILTER_REWRITER "filter-rewriter"
  30
  31 typedef struct RewriterState {
  32     NetFilterState parent_obj;
  33     NetQueue *incoming_queue;
  34     /* hashtable to save connection */
  35     GHashTable *connection_track_table;
  36 } RewriterState;
  37
  38 static void filter_rewriter_flush(NetFilterState *nf)
  39 {
  40     RewriterState *s = FILTER_COLO_REWRITER(nf);
  41
  42     if (!qemu_net_queue_flush(s->incoming_queue)) {
  43         /* Unable to empty the queue, purge remaining packets */
  44         qemu_net_queue_purge(s->incoming_queue, nf->netdev);
  45     }
  46 }
  47
  48 /*
  49  * Return 1 on success, if return 0 means the pkt
  50  * is not TCP packet
  51  */
  52 static int is_tcp_packet(Packet *pkt)
  53 {
  54     if (!parse_packet_early(pkt) &&
  55         pkt->ip->ip_p == IPPROTO_TCP) {
  56         return 1;
  57     } else {
  58         return 0;
  59     }
  60 }
  61
  62 /* handle tcp packet from primary guest */
  63 static int handle_primary_tcp_pkt(NetFilterState *nf,
  64                                   Connection *conn,
  65                                   Packet *pkt)
  66 {
  67     struct tcphdr *tcp_pkt;
  68
  69     tcp_pkt = (struct tcphdr *)pkt->transport_header;
  70     if (trace_event_get_state(TRACE_COLO_FILTER_REWRITER_DEBUG)) {
  71         trace_colo_filter_rewriter_pkt_info(__func__,
  72                     inet_ntoa(pkt->ip->ip_src), inet_ntoa(pkt->ip->ip_dst),
  73                     ntohl(tcp_pkt->th_seq), ntohl(tcp_pkt->th_ack),
  74                     tcp_pkt->th_flags);
  75         trace_colo_filter_rewriter_conn_offset(conn->offset);
  76     }
  77
  78     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_SYN)) {
  79         /*
  80          * we use this flag update offset func
  81          * run once in independent tcp connection
  82          */
  83         conn->syn_flag = 1;
  84     }
  85
  86     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_ACK)) {
  87         if (conn->syn_flag) {
  88             /*
  89              * offset = secondary_seq - primary seq
  90              * ack packet sent by guest from primary node,
  91              * so we use th_ack - 1 get primary_seq
  92              */
  93             conn->offset -= (ntohl(tcp_pkt->th_ack) - 1);
  94             conn->syn_flag = 0;
  95         }
  96         if (conn->offset) {
  97             /* handle packets to the secondary from the primary */
  98             tcp_pkt->th_ack = htonl(ntohl(tcp_pkt->th_ack) + conn->offset);
  99
 100             net_checksum_calculate((uint8_t *)pkt->data, pkt->size);
 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(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->size);
 140         }
 141     }
 142
 143     return 0;
 144 }
 145
 146 static ssize_t colo_rewriter_receive_iov(NetFilterState *nf,
 147                                          NetClientState *sender,
 148                                          unsigned flags,
 149                                          const struct iovec *iov,
 150                                          int iovcnt,
 151                                          NetPacketSent *sent_cb)
 152 {
 153     RewriterState *s = FILTER_COLO_REWRITER(nf);
 154     Connection *conn;
 155     ConnectionKey key;
 156     Packet *pkt;
 157     ssize_t size = iov_size(iov, iovcnt);
 158     char *buf = g_malloc0(size);
 159
 160     iov_to_buf(iov, iovcnt, 0, buf, size);
 161     pkt = packet_new(buf, size);
 162     g_free(buf);
 163
 164     /*
 165      * if we get tcp packet
 166      * we will rewrite it to make secondary guest's
 167      * connection established successfully
 168      */
 169     if (pkt && is_tcp_packet(pkt)) {
 170
 171         fill_connection_key(pkt, &key);
 172
 173         if (sender == nf->netdev) {
 174             /*
 175              * We need make tcp TX and RX packet
 176              * into one connection.
 177              */
 178             reverse_connection_key(&key);
 179         }
 180         conn = connection_get(s->connection_track_table,
 181                               &key,
 182                               NULL);
 183
 184         if (sender == nf->netdev) {
 185             /* NET_FILTER_DIRECTION_TX */
 186             if (!handle_primary_tcp_pkt(nf, conn, pkt)) {
 187                 qemu_net_queue_send(s->incoming_queue, sender, 0,
 188                 (const uint8_t *)pkt->data, pkt->size, NULL);
 189                 packet_destroy(pkt, NULL);
 190                 pkt = NULL;
 191                 /*
 192                  * We block the packet here,after rewrite pkt
 193                  * and will send it
 194                  */
 195                 return 1;
 196             }
 197         } else {
 198             /* NET_FILTER_DIRECTION_RX */
 199             if (!handle_secondary_tcp_pkt(nf, conn, pkt)) {
 200                 qemu_net_queue_send(s->incoming_queue, sender, 0,
 201                 (const uint8_t *)pkt->data, pkt->size, NULL);
 202                 packet_destroy(pkt, NULL);
 203                 pkt = NULL;
 204                 /*
 205                  * We block the packet here,after rewrite pkt
 206                  * and will send it
 207                  */
 208                 return 1;
 209             }
 210         }
 211     }
 212
 213     packet_destroy(pkt, NULL);
 214     pkt = NULL;
 215     return 0;
 216 }
 217
 218 static void colo_rewriter_cleanup(NetFilterState *nf)
 219 {
 220     RewriterState *s = FILTER_COLO_REWRITER(nf);
 221
 222     /* flush packets */
 223     if (s->incoming_queue) {
 224         filter_rewriter_flush(nf);
 225         g_free(s->incoming_queue);
 226     }
 227 }
 228
 229 static void colo_rewriter_setup(NetFilterState *nf, Error **errp)
 230 {
 231     RewriterState *s = FILTER_COLO_REWRITER(nf);
 232
 233     s->connection_track_table = g_hash_table_new_full(connection_key_hash,
 234                                                       connection_key_equal,
 235                                                       g_free,
 236                                                       connection_destroy);
 237     s->incoming_queue = qemu_new_net_queue(qemu_netfilter_pass_to_next, nf);
 238 }
 239
 240 static void colo_rewriter_class_init(ObjectClass *oc, void *data)
 241 {
 242     NetFilterClass *nfc = NETFILTER_CLASS(oc);
 243
 244     nfc->setup = colo_rewriter_setup;
 245     nfc->cleanup = colo_rewriter_cleanup;
 246     nfc->receive_iov = colo_rewriter_receive_iov;
 247 }
 248
 249 static const TypeInfo colo_rewriter_info = {
 250     .name = TYPE_FILTER_REWRITER,
 251     .parent = TYPE_NETFILTER,
 252     .class_init = colo_rewriter_class_init,
 253     .instance_size = sizeof(RewriterState),
 254 };
 255
 256 static void register_types(void)
 257 {
 258     type_register_static(&colo_rewriter_info);
 259 }
 260
 261 type_init(register_types);