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