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 "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 #include "net/colo.h"
  24 #include "migration/colo.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 #define FAILOVER_MODE_ON  true
  31 #define FAILOVER_MODE_OFF false
  32
  33 typedef struct RewriterState {
  34     NetFilterState parent_obj;
  35     NetQueue *incoming_queue;
  36     /* hashtable to save connection */
  37     GHashTable *connection_track_table;
  38     bool vnet_hdr;
  39     bool failover_mode;
  40 } RewriterState;
  41
  42 static void filter_rewriter_failover_mode(RewriterState *s)
  43 {
  44     s->failover_mode = FAILOVER_MODE_ON;
  45 }
  46
  47 static void filter_rewriter_flush(NetFilterState *nf)
  48 {
  49     RewriterState *s = FILTER_COLO_REWRITER(nf);
  50
  51     if (!qemu_net_queue_flush(s->incoming_queue)) {
  52         /* Unable to empty the queue, purge remaining packets */
  53         qemu_net_queue_purge(s->incoming_queue, nf->netdev);
  54     }
  55 }
  56
  57 /*
  58  * Return 1 on success, if return 0 means the pkt
  59  * is not TCP packet
  60  */
  61 static int is_tcp_packet(Packet *pkt)
  62 {
  63     if (!parse_packet_early(pkt) &&
  64         pkt->ip->ip_p == IPPROTO_TCP) {
  65         return 1;
  66     } else {
  67         return 0;
  68     }
  69 }
  70
  71 /* handle tcp packet from primary guest */
  72 static int handle_primary_tcp_pkt(RewriterState *rf,
  73                                   Connection *conn,
  74                                   Packet *pkt, ConnectionKey *key)
  75 {
  76     struct tcphdr *tcp_pkt;
  77
  78     tcp_pkt = (struct tcphdr *)pkt->transport_header;
  79     if (trace_event_get_state_backends(TRACE_COLO_FILTER_REWRITER_DEBUG)) {
  80         trace_colo_filter_rewriter_pkt_info(__func__,
  81                     inet_ntoa(pkt->ip->ip_src), inet_ntoa(pkt->ip->ip_dst),
  82                     ntohl(tcp_pkt->th_seq), ntohl(tcp_pkt->th_ack),
  83                     tcp_pkt->th_flags);
  84         trace_colo_filter_rewriter_conn_offset(conn->offset);
  85     }
  86
  87     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == (TH_ACK | TH_SYN)) &&
  88         conn->tcp_state == TCPS_SYN_SENT) {
  89         conn->tcp_state = TCPS_ESTABLISHED;
  90     }
  91
  92     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_SYN)) {
  93         /*
  94          * we use this flag update offset func
  95          * run once in independent tcp connection
  96          */
  97         conn->tcp_state = TCPS_SYN_RECEIVED;
  98     }
  99
 100     if (((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_ACK)) {
 101         if (conn->tcp_state == TCPS_SYN_RECEIVED) {
 102             /*
 103              * offset = secondary_seq - primary seq
 104              * ack packet sent by guest from primary node,
 105              * so we use th_ack - 1 get primary_seq
 106              */
 107             conn->offset -= (ntohl(tcp_pkt->th_ack) - 1);
 108             conn->tcp_state = TCPS_ESTABLISHED;
 109         }
 110         if (conn->offset) {
 111             /* handle packets to the secondary from the primary */
 112             tcp_pkt->th_ack = htonl(ntohl(tcp_pkt->th_ack) + conn->offset);
 113
 114             net_checksum_calculate((uint8_t *)pkt->data + pkt->vnet_hdr_len,
 115                                    pkt->size - pkt->vnet_hdr_len);
 116         }
 117
 118         /*
 119          * Passive close step 3
 120          */
 121         if ((conn->tcp_state == TCPS_LAST_ACK) &&
 122             (ntohl(tcp_pkt->th_ack) == (conn->fin_ack_seq + 1))) {
 123             conn->tcp_state = TCPS_CLOSED;
 124             g_hash_table_remove(rf->connection_track_table, key);
 125         }
 126     }
 127
 128     if ((tcp_pkt->th_flags & TH_FIN) == TH_FIN) {
 129         /*
 130          * Passive close.
 131          * Step 1:
 132          * The *server* side of this connect is VM, *client* tries to close
 133          * the connection. We will into CLOSE_WAIT status.
 134          *
 135          * Step 2:
 136          * In this step we will into LAST_ACK status.
 137          *
 138          * We got 'fin=1, ack=1' packet from server side, we need to
 139          * record the seq of 'fin=1, ack=1' packet.
 140          *
 141          * Step 3:
 142          * We got 'ack=1' packets from client side, it acks 'fin=1, ack=1'
 143          * packet from server side. From this point, we can ensure that there
 144          * will be no packets in the connection, except that, some errors
 145          * happen between the path of 'filter object' and vNIC, if this rare
 146          * case really happen, we can still create a new connection,
 147          * So it is safe to remove the connection from connection_track_table.
 148          *
 149          */
 150         if (conn->tcp_state == TCPS_ESTABLISHED) {
 151             conn->tcp_state = TCPS_CLOSE_WAIT;
 152         }
 153
 154         /*
 155          * Active close step 2.
 156          */
 157         if (conn->tcp_state == TCPS_FIN_WAIT_1) {
 158             conn->tcp_state = TCPS_TIME_WAIT;
 159             /*
 160              * For simplify implementation, we needn't wait 2MSL time
 161              * in filter rewriter. Because guest kernel will track the
 162              * TCP status and wait 2MSL time, if client resend the FIN
 163              * packet, guest will apply the last ACK too.
 164              */
 165             conn->tcp_state = TCPS_CLOSED;
 166             g_hash_table_remove(rf->connection_track_table, key);
 167         }
 168     }
 169
 170     return 0;
 171 }
 172
 173 /* handle tcp packet from secondary guest */
 174 static int handle_secondary_tcp_pkt(RewriterState *rf,
 175                                     Connection *conn,
 176                                     Packet *pkt, ConnectionKey *key)
 177 {
 178     struct tcphdr *tcp_pkt;
 179
 180     tcp_pkt = (struct tcphdr *)pkt->transport_header;
 181
 182     if (trace_event_get_state_backends(TRACE_COLO_FILTER_REWRITER_DEBUG)) {
 183         trace_colo_filter_rewriter_pkt_info(__func__,
 184                     inet_ntoa(pkt->ip->ip_src), inet_ntoa(pkt->ip->ip_dst),
 185                     ntohl(tcp_pkt->th_seq), ntohl(tcp_pkt->th_ack),
 186                     tcp_pkt->th_flags);
 187         trace_colo_filter_rewriter_conn_offset(conn->offset);
 188     }
 189
 190     if (conn->tcp_state == TCPS_SYN_RECEIVED &&
 191         ((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == (TH_ACK | TH_SYN))) {
 192         /*
 193          * save offset = secondary_seq and then
 194          * in handle_primary_tcp_pkt make offset
 195          * = secondary_seq - primary_seq
 196          */
 197         conn->offset = ntohl(tcp_pkt->th_seq);
 198     }
 199
 200     /* VM active connect */
 201     if (conn->tcp_state == TCPS_CLOSED &&
 202         ((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_SYN)) {
 203         conn->tcp_state = TCPS_SYN_SENT;
 204     }
 205
 206     if ((tcp_pkt->th_flags & (TH_ACK | TH_SYN)) == TH_ACK) {
 207         /* Only need to adjust seq while offset is Non-zero */
 208         if (conn->offset) {
 209             /* handle packets to the primary from the secondary*/
 210             tcp_pkt->th_seq = htonl(ntohl(tcp_pkt->th_seq) - conn->offset);
 211
 212             net_checksum_calculate((uint8_t *)pkt->data + pkt->vnet_hdr_len,
 213                                    pkt->size - pkt->vnet_hdr_len);
 214         }
 215     }
 216
 217     /*
 218      * Passive close step 2:
 219      */
 220     if (conn->tcp_state == TCPS_CLOSE_WAIT &&
 221         (tcp_pkt->th_flags & (TH_ACK | TH_FIN)) == (TH_ACK | TH_FIN)) {
 222         conn->fin_ack_seq = ntohl(tcp_pkt->th_seq);
 223         conn->tcp_state = TCPS_LAST_ACK;
 224     }
 225
 226     /*
 227      * Active close
 228      *
 229      * Step 1:
 230      * The *server* side of this connect is VM, *server* tries to close
 231      * the connection.
 232      *
 233      * Step 2:
 234      * We will into CLOSE_WAIT status.
 235      * We simplify the TCPS_FIN_WAIT_2, TCPS_TIME_WAIT and
 236      * CLOSING status.
 237      */
 238     if (conn->tcp_state == TCPS_ESTABLISHED &&
 239         (tcp_pkt->th_flags & (TH_ACK | TH_FIN)) == TH_FIN) {
 240         conn->tcp_state = TCPS_FIN_WAIT_1;
 241     }
 242
 243     return 0;
 244 }
 245
 246 static ssize_t colo_rewriter_receive_iov(NetFilterState *nf,
 247                                          NetClientState *sender,
 248                                          unsigned flags,
 249                                          const struct iovec *iov,
 250                                          int iovcnt,
 251                                          NetPacketSent *sent_cb)
 252 {
 253     RewriterState *s = FILTER_COLO_REWRITER(nf);
 254     Connection *conn;
 255     ConnectionKey key;
 256     Packet *pkt;
 257     ssize_t size = iov_size(iov, iovcnt);
 258     ssize_t vnet_hdr_len = 0;
 259     char *buf = g_malloc0(size);
 260
 261     iov_to_buf(iov, iovcnt, 0, buf, size);
 262
 263     if (s->vnet_hdr) {
 264         vnet_hdr_len = nf->netdev->vnet_hdr_len;
 265     }
 266
 267     pkt = packet_new(buf, size, vnet_hdr_len);
 268     g_free(buf);
 269
 270     /*
 271      * if we get tcp packet
 272      * we will rewrite it to make secondary guest's
 273      * connection established successfully
 274      */
 275     if (pkt && is_tcp_packet(pkt)) {
 276
 277         fill_connection_key(pkt, &key);
 278
 279         if (sender == nf->netdev) {
 280             /*
 281              * We need make tcp TX and RX packet
 282              * into one connection.
 283              */
 284             reverse_connection_key(&key);
 285         }
 286
 287         /* After failover we needn't change new TCP packet */
 288         if (s->failover_mode &&
 289             !connection_has_tracked(s->connection_track_table, &key)) {
 290             goto out;
 291         }
 292
 293         conn = connection_get(s->connection_track_table,
 294                               &key,
 295                               NULL);
 296
 297         if (sender == nf->netdev) {
 298             /* NET_FILTER_DIRECTION_TX */
 299             if (!handle_primary_tcp_pkt(s, conn, pkt, &key)) {
 300                 qemu_net_queue_send(s->incoming_queue, sender, 0,
 301                 (const uint8_t *)pkt->data, pkt->size, NULL);
 302                 packet_destroy(pkt, NULL);
 303                 pkt = NULL;
 304                 /*
 305                  * We block the packet here,after rewrite pkt
 306                  * and will send it
 307                  */
 308                 return 1;
 309             }
 310         } else {
 311             /* NET_FILTER_DIRECTION_RX */
 312             if (!handle_secondary_tcp_pkt(s, conn, pkt, &key)) {
 313                 qemu_net_queue_send(s->incoming_queue, sender, 0,
 314                 (const uint8_t *)pkt->data, pkt->size, NULL);
 315                 packet_destroy(pkt, NULL);
 316                 pkt = NULL;
 317                 /*
 318                  * We block the packet here,after rewrite pkt
 319                  * and will send it
 320                  */
 321                 return 1;
 322             }
 323         }
 324     }
 325
 326 out:
 327     packet_destroy(pkt, NULL);
 328     pkt = NULL;
 329     return 0;
 330 }
 331
 332 static void reset_seq_offset(gpointer key, gpointer value, gpointer user_data)
 333 {
 334     Connection *conn = (Connection *)value;
 335
 336     conn->offset = 0;
 337 }
 338
 339 static gboolean offset_is_nonzero(gpointer key,
 340                                   gpointer value,
 341                                   gpointer user_data)
 342 {
 343     Connection *conn = (Connection *)value;
 344
 345     return conn->offset ? true : false;
 346 }
 347
 348 static void colo_rewriter_handle_event(NetFilterState *nf, int event,
 349                                        Error **errp)
 350 {
 351     RewriterState *rs = FILTER_COLO_REWRITER(nf);
 352
 353     switch (event) {
 354     case COLO_EVENT_CHECKPOINT:
 355         g_hash_table_foreach(rs->connection_track_table,
 356                             reset_seq_offset, NULL);
 357         break;
 358     case COLO_EVENT_FAILOVER:
 359         if (!g_hash_table_find(rs->connection_track_table,
 360                               offset_is_nonzero, NULL)) {
 361             filter_rewriter_failover_mode(rs);
 362         }
 363         break;
 364     default:
 365         break;
 366     }
 367 }
 368
 369 static void colo_rewriter_cleanup(NetFilterState *nf)
 370 {
 371     RewriterState *s = FILTER_COLO_REWRITER(nf);
 372
 373     /* flush packets */
 374     if (s->incoming_queue) {
 375         filter_rewriter_flush(nf);
 376         g_free(s->incoming_queue);
 377     }
 378 }
 379
 380 static void colo_rewriter_setup(NetFilterState *nf, Error **errp)
 381 {
 382     RewriterState *s = FILTER_COLO_REWRITER(nf);
 383
 384     s->connection_track_table = g_hash_table_new_full(connection_key_hash,
 385                                                       connection_key_equal,
 386                                                       g_free,
 387                                                       connection_destroy);
 388     s->incoming_queue = qemu_new_net_queue(qemu_netfilter_pass_to_next, nf);
 389 }
 390
 391 static bool filter_rewriter_get_vnet_hdr(Object *obj, Error **errp)
 392 {
 393     RewriterState *s = FILTER_COLO_REWRITER(obj);
 394
 395     return s->vnet_hdr;
 396 }
 397
 398 static void filter_rewriter_set_vnet_hdr(Object *obj,
 399                                          bool value,
 400                                          Error **errp)
 401 {
 402     RewriterState *s = FILTER_COLO_REWRITER(obj);
 403
 404     s->vnet_hdr = value;
 405 }
 406
 407 static void filter_rewriter_init(Object *obj)
 408 {
 409     RewriterState *s = FILTER_COLO_REWRITER(obj);
 410
 411     s->vnet_hdr = false;
 412     s->failover_mode = FAILOVER_MODE_OFF;
 413     object_property_add_bool(obj, "vnet_hdr_support",
 414                              filter_rewriter_get_vnet_hdr,
 415                              filter_rewriter_set_vnet_hdr, NULL);
 416 }
 417
 418 static void colo_rewriter_class_init(ObjectClass *oc, void *data)
 419 {
 420     NetFilterClass *nfc = NETFILTER_CLASS(oc);
 421
 422     nfc->setup = colo_rewriter_setup;
 423     nfc->cleanup = colo_rewriter_cleanup;
 424     nfc->receive_iov = colo_rewriter_receive_iov;
 425     nfc->handle_event = colo_rewriter_handle_event;
 426 }
 427
 428 static const TypeInfo colo_rewriter_info = {
 429     .name = TYPE_FILTER_REWRITER,
 430     .parent = TYPE_NETFILTER,
 431     .class_init = colo_rewriter_class_init,
 432     .instance_init = filter_rewriter_init,
 433     .instance_size = sizeof(RewriterState),
 434 };
 435
 436 static void register_types(void)
 437 {
 438     type_register_static(&colo_rewriter_info);
 439 }
 440
 441 type_init(register_types);