net/colo-compare.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 "qemu/error-report.h"
  17 #include "trace.h"
  18 #include "qemu-common.h"
  19 #include "qapi/qmp/qerror.h"
  20 #include "qapi/error.h"
  21 #include "net/net.h"
  22 #include "net/eth.h"
  23 #include "qom/object_interfaces.h"
  24 #include "qemu/iov.h"
  25 #include "qom/object.h"
  26 #include "qemu/typedefs.h"
  27 #include "net/queue.h"
  28 #include "chardev/char-fe.h"
  29 #include "qemu/sockets.h"
  30 #include "qapi-visit.h"
  31 #include "net/colo.h"
  32
  33 #define TYPE_COLO_COMPARE "colo-compare"
  34 #define COLO_COMPARE(obj) \
  35     OBJECT_CHECK(CompareState, (obj), TYPE_COLO_COMPARE)
  36
  37 #define COMPARE_READ_LEN_MAX NET_BUFSIZE
  38 #define MAX_QUEUE_SIZE 1024
  39
  40 /* TODO: Should be configurable */
  41 #define REGULAR_PACKET_CHECK_MS 3000
  42
  43 /*
  44   + CompareState ++
  45   |               |
  46   +---------------+   +---------------+         +---------------+
  47   |conn list      +--->conn           +--------->conn           |
  48   +---------------+   +---------------+         +---------------+
  49   |               |     |           |             |          |
  50   +---------------+ +---v----+  +---v----+    +---v----+ +---v----+
  51                     |primary |  |secondary    |primary | |secondary
  52                     |packet  |  |packet  +    |packet  | |packet  +
  53                     +--------+  +--------+    +--------+ +--------+
  54                         |           |             |          |
  55                     +---v----+  +---v----+    +---v----+ +---v----+
  56                     |primary |  |secondary    |primary | |secondary
  57                     |packet  |  |packet  +    |packet  | |packet  +
  58                     +--------+  +--------+    +--------+ +--------+
  59                         |           |             |          |
  60                     +---v----+  +---v----+    +---v----+ +---v----+
  61                     |primary |  |secondary    |primary | |secondary
  62                     |packet  |  |packet  +    |packet  | |packet  +
  63                     +--------+  +--------+    +--------+ +--------+
  64 */
  65 typedef struct CompareState {
  66     Object parent;
  67
  68     char *pri_indev;
  69     char *sec_indev;
  70     char *outdev;
  71     CharBackend chr_pri_in;
  72     CharBackend chr_sec_in;
  73     CharBackend chr_out;
  74     SocketReadState pri_rs;
  75     SocketReadState sec_rs;
  76
  77     /* connection list: the connections belonged to this NIC could be found
  78      * in this list.
  79      * element type: Connection
  80      */
  81     GQueue conn_list;
  82     /* hashtable to save connection */
  83     GHashTable *connection_track_table;
  84     /* compare thread, a thread for each NIC */
  85     QemuThread thread;
  86
  87     GMainContext *worker_context;
  88     GMainLoop *compare_loop;
  89 } CompareState;
  90
  91 typedef struct CompareClass {
  92     ObjectClass parent_class;
  93 } CompareClass;
  94
  95 enum {
  96     PRIMARY_IN = 0,
  97     SECONDARY_IN,
  98 };
  99
 100 static int compare_chr_send(CharBackend *out,
 101                             const uint8_t *buf,
 102                             uint32_t size);
 103
 104 static gint seq_sorter(Packet *a, Packet *b, gpointer data)
 105 {
 106     struct tcphdr *atcp, *btcp;
 107
 108     atcp = (struct tcphdr *)(a->transport_header);
 109     btcp = (struct tcphdr *)(b->transport_header);
 110     return ntohl(atcp->th_seq) - ntohl(btcp->th_seq);
 111 }
 112
 113 /*
 114  * Return 0 on success, if return -1 means the pkt
 115  * is unsupported(arp and ipv6) and will be sent later
 116  */
 117 static int packet_enqueue(CompareState *s, int mode)
 118 {
 119     ConnectionKey key;
 120     Packet *pkt = NULL;
 121     Connection *conn;
 122
 123     if (mode == PRIMARY_IN) {
 124         pkt = packet_new(s->pri_rs.buf, s->pri_rs.packet_len);
 125     } else {
 126         pkt = packet_new(s->sec_rs.buf, s->sec_rs.packet_len);
 127     }
 128
 129     if (parse_packet_early(pkt)) {
 130         packet_destroy(pkt, NULL);
 131         pkt = NULL;
 132         return -1;
 133     }
 134     fill_connection_key(pkt, &key);
 135
 136     conn = connection_get(s->connection_track_table,
 137                           &key,
 138                           &s->conn_list);
 139
 140     if (!conn->processing) {
 141         g_queue_push_tail(&s->conn_list, conn);
 142         conn->processing = true;
 143     }
 144
 145     if (mode == PRIMARY_IN) {
 146         if (g_queue_get_length(&conn->primary_list) <=
 147                                MAX_QUEUE_SIZE) {
 148             g_queue_push_tail(&conn->primary_list, pkt);
 149             if (conn->ip_proto == IPPROTO_TCP) {
 150                 g_queue_sort(&conn->primary_list,
 151                              (GCompareDataFunc)seq_sorter,
 152                              NULL);
 153             }
 154         } else {
 155             error_report("colo compare primary queue size too big,"
 156                          "drop packet");
 157         }
 158     } else {
 159         if (g_queue_get_length(&conn->secondary_list) <=
 160                                MAX_QUEUE_SIZE) {
 161             g_queue_push_tail(&conn->secondary_list, pkt);
 162             if (conn->ip_proto == IPPROTO_TCP) {
 163                 g_queue_sort(&conn->secondary_list,
 164                              (GCompareDataFunc)seq_sorter,
 165                              NULL);
 166             }
 167         } else {
 168             error_report("colo compare secondary queue size too big,"
 169                          "drop packet");
 170         }
 171     }
 172
 173     return 0;
 174 }
 175
 176 /*
 177  * The IP packets sent by primary and secondary
 178  * will be compared in here
 179  * TODO support ip fragment, Out-Of-Order
 180  * return:    0  means packet same
 181  *            > 0 || < 0 means packet different
 182  */
 183 static int colo_packet_compare_common(Packet *ppkt, Packet *spkt, int offset)
 184 {
 185     if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) {
 186         char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20];
 187
 188         strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src));
 189         strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst));
 190         strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src));
 191         strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst));
 192
 193         trace_colo_compare_ip_info(ppkt->size, pri_ip_src,
 194                                    pri_ip_dst, spkt->size,
 195                                    sec_ip_src, sec_ip_dst);
 196     }
 197
 198     if (ppkt->size == spkt->size) {
 199         return memcmp(ppkt->data + offset, spkt->data + offset,
 200                       spkt->size - offset);
 201     } else {
 202         trace_colo_compare_main("Net packet size are not the same");
 203         return -1;
 204     }
 205 }
 206
 207 /*
 208  * Called from the compare thread on the primary
 209  * for compare tcp packet
 210  * compare_tcp copied from Dr. David Alan Gilbert's branch
 211  */
 212 static int colo_packet_compare_tcp(Packet *spkt, Packet *ppkt)
 213 {
 214     struct tcphdr *ptcp, *stcp;
 215     int res;
 216
 217     trace_colo_compare_main("compare tcp");
 218
 219     ptcp = (struct tcphdr *)ppkt->transport_header;
 220     stcp = (struct tcphdr *)spkt->transport_header;
 221
 222     /*
 223      * The 'identification' field in the IP header is *very* random
 224      * it almost never matches.  Fudge this by ignoring differences in
 225      * unfragmented packets; they'll normally sort themselves out if different
 226      * anyway, and it should recover at the TCP level.
 227      * An alternative would be to get both the primary and secondary to rewrite
 228      * somehow; but that would need some sync traffic to sync the state
 229      */
 230     if (ntohs(ppkt->ip->ip_off) & IP_DF) {
 231         spkt->ip->ip_id = ppkt->ip->ip_id;
 232         /* and the sum will be different if the IDs were different */
 233         spkt->ip->ip_sum = ppkt->ip->ip_sum;
 234     }
 235
 236     /*
 237      * Check tcp header length for tcp option field.
 238      * th_off > 5 means this tcp packet have options field.
 239      * The tcp options maybe always different.
 240      * for example:
 241      * From RFC 7323.
 242      * TCP Timestamps option (TSopt):
 243      * Kind: 8
 244      *
 245      * Length: 10 bytes
 246      *
 247      *    +-------+-------+---------------------+---------------------+
 248      *    |Kind=8 |  10   |   TS Value (TSval)  |TS Echo Reply (TSecr)|
 249      *    +-------+-------+---------------------+---------------------+
 250      *       1       1              4                     4
 251      *
 252      * In this case the primary guest's timestamp always different with
 253      * the secondary guest's timestamp. COLO just focus on payload,
 254      * so we just need skip this field.
 255      */
 256     if (ptcp->th_off > 5) {
 257         ptrdiff_t tcp_offset;
 258         tcp_offset = ppkt->transport_header - (uint8_t *)ppkt->data
 259                      + (ptcp->th_off * 4);
 260         res = colo_packet_compare_common(ppkt, spkt, tcp_offset);
 261     } else if (ptcp->th_sum == stcp->th_sum) {
 262         res = colo_packet_compare_common(ppkt, spkt, ETH_HLEN);
 263     } else {
 264         res = -1;
 265     }
 266
 267     if (res != 0 && trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) {
 268         char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20];
 269
 270         strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src));
 271         strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst));
 272         strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src));
 273         strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst));
 274
 275         trace_colo_compare_ip_info(ppkt->size, pri_ip_src,
 276                                    pri_ip_dst, spkt->size,
 277                                    sec_ip_src, sec_ip_dst);
 278
 279         trace_colo_compare_tcp_info("pri tcp packet",
 280                                     ntohl(ptcp->th_seq),
 281                                     ntohl(ptcp->th_ack),
 282                                     res, ptcp->th_flags,
 283                                     ppkt->size);
 284
 285         trace_colo_compare_tcp_info("sec tcp packet",
 286                                     ntohl(stcp->th_seq),
 287                                     ntohl(stcp->th_ack),
 288                                     res, stcp->th_flags,
 289                                     spkt->size);
 290
 291         qemu_hexdump((char *)ppkt->data, stderr,
 292                      "colo-compare ppkt", ppkt->size);
 293         qemu_hexdump((char *)spkt->data, stderr,
 294                      "colo-compare spkt", spkt->size);
 295     }
 296
 297     return res;
 298 }
 299
 300 /*
 301  * Called from the compare thread on the primary
 302  * for compare udp packet
 303  */
 304 static int colo_packet_compare_udp(Packet *spkt, Packet *ppkt)
 305 {
 306     int ret;
 307     int network_header_length = ppkt->ip->ip_hl * 4;
 308
 309     trace_colo_compare_main("compare udp");
 310
 311     /*
 312      * Because of ppkt and spkt are both in the same connection,
 313      * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are
 314      * same with spkt. In addition, IP header's Identification is a random
 315      * field, we can handle it in IP fragmentation function later.
 316      * COLO just concern the response net packet payload from primary guest
 317      * and secondary guest are same or not, So we ignored all IP header include
 318      * other field like TOS,TTL,IP Checksum. we only need to compare
 319      * the ip payload here.
 320      */
 321     ret = colo_packet_compare_common(ppkt, spkt,
 322                                      network_header_length + ETH_HLEN);
 323
 324     if (ret) {
 325         trace_colo_compare_udp_miscompare("primary pkt size", ppkt->size);
 326         trace_colo_compare_udp_miscompare("Secondary pkt size", spkt->size);
 327         if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) {
 328             qemu_hexdump((char *)ppkt->data, stderr, "colo-compare pri pkt",
 329                          ppkt->size);
 330             qemu_hexdump((char *)spkt->data, stderr, "colo-compare sec pkt",
 331                          spkt->size);
 332         }
 333     }
 334
 335     return ret;
 336 }
 337
 338 /*
 339  * Called from the compare thread on the primary
 340  * for compare icmp packet
 341  */
 342 static int colo_packet_compare_icmp(Packet *spkt, Packet *ppkt)
 343 {
 344     int network_header_length = ppkt->ip->ip_hl * 4;
 345
 346     trace_colo_compare_main("compare icmp");
 347
 348     /*
 349      * Because of ppkt and spkt are both in the same connection,
 350      * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are
 351      * same with spkt. In addition, IP header's Identification is a random
 352      * field, we can handle it in IP fragmentation function later.
 353      * COLO just concern the response net packet payload from primary guest
 354      * and secondary guest are same or not, So we ignored all IP header include
 355      * other field like TOS,TTL,IP Checksum. we only need to compare
 356      * the ip payload here.
 357      */
 358     if (colo_packet_compare_common(ppkt, spkt,
 359                                    network_header_length + ETH_HLEN)) {
 360         trace_colo_compare_icmp_miscompare("primary pkt size",
 361                                            ppkt->size);
 362         trace_colo_compare_icmp_miscompare("Secondary pkt size",
 363                                            spkt->size);
 364         if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) {
 365             qemu_hexdump((char *)ppkt->data, stderr, "colo-compare pri pkt",
 366                          ppkt->size);
 367             qemu_hexdump((char *)spkt->data, stderr, "colo-compare sec pkt",
 368                          spkt->size);
 369         }
 370         return -1;
 371     } else {
 372         return 0;
 373     }
 374 }
 375
 376 /*
 377  * Called from the compare thread on the primary
 378  * for compare other packet
 379  */
 380 static int colo_packet_compare_other(Packet *spkt, Packet *ppkt)
 381 {
 382     trace_colo_compare_main("compare other");
 383     if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) {
 384         char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20];
 385
 386         strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src));
 387         strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst));
 388         strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src));
 389         strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst));
 390
 391         trace_colo_compare_ip_info(ppkt->size, pri_ip_src,
 392                                    pri_ip_dst, spkt->size,
 393                                    sec_ip_src, sec_ip_dst);
 394     }
 395
 396     return colo_packet_compare_common(ppkt, spkt, 0);
 397 }
 398
 399 static int colo_old_packet_check_one(Packet *pkt, int64_t *check_time)
 400 {
 401     int64_t now = qemu_clock_get_ms(QEMU_CLOCK_HOST);
 402
 403     if ((now - pkt->creation_ms) > (*check_time)) {
 404         trace_colo_old_packet_check_found(pkt->creation_ms);
 405         return 0;
 406     } else {
 407         return 1;
 408     }
 409 }
 410
 411 static int colo_old_packet_check_one_conn(Connection *conn,
 412                                           void *user_data)
 413 {
 414     GList *result = NULL;
 415     int64_t check_time = REGULAR_PACKET_CHECK_MS;
 416
 417     result = g_queue_find_custom(&conn->primary_list,
 418                                  &check_time,
 419                                  (GCompareFunc)colo_old_packet_check_one);
 420
 421     if (result) {
 422         /* do checkpoint will flush old packet */
 423         /* TODO: colo_notify_checkpoint();*/
 424         return 0;
 425     }
 426
 427     return 1;
 428 }
 429
 430 /*
 431  * Look for old packets that the secondary hasn't matched,
 432  * if we have some then we have to checkpoint to wake
 433  * the secondary up.
 434  */
 435 static void colo_old_packet_check(void *opaque)
 436 {
 437     CompareState *s = opaque;
 438
 439     /*
 440      * If we find one old packet, stop finding job and notify
 441      * COLO frame do checkpoint.
 442      */
 443     g_queue_find_custom(&s->conn_list, NULL,
 444                         (GCompareFunc)colo_old_packet_check_one_conn);
 445 }
 446
 447 /*
 448  * Called from the compare thread on the primary
 449  * for compare connection
 450  */
 451 static void colo_compare_connection(void *opaque, void *user_data)
 452 {
 453     CompareState *s = user_data;
 454     Connection *conn = opaque;
 455     Packet *pkt = NULL;
 456     GList *result = NULL;
 457     int ret;
 458
 459     while (!g_queue_is_empty(&conn->primary_list) &&
 460            !g_queue_is_empty(&conn->secondary_list)) {
 461         pkt = g_queue_pop_tail(&conn->primary_list);
 462         switch (conn->ip_proto) {
 463         case IPPROTO_TCP:
 464             result = g_queue_find_custom(&conn->secondary_list,
 465                      pkt, (GCompareFunc)colo_packet_compare_tcp);
 466             break;
 467         case IPPROTO_UDP:
 468             result = g_queue_find_custom(&conn->secondary_list,
 469                      pkt, (GCompareFunc)colo_packet_compare_udp);
 470             break;
 471         case IPPROTO_ICMP:
 472             result = g_queue_find_custom(&conn->secondary_list,
 473                      pkt, (GCompareFunc)colo_packet_compare_icmp);
 474             break;
 475         default:
 476             result = g_queue_find_custom(&conn->secondary_list,
 477                      pkt, (GCompareFunc)colo_packet_compare_other);
 478             break;
 479         }
 480
 481         if (result) {
 482             ret = compare_chr_send(&s->chr_out, pkt->data, pkt->size);
 483             if (ret < 0) {
 484                 error_report("colo_send_primary_packet failed");
 485             }
 486             trace_colo_compare_main("packet same and release packet");
 487             g_queue_remove(&conn->secondary_list, result->data);
 488             packet_destroy(pkt, NULL);
 489         } else {
 490             /*
 491              * If one packet arrive late, the secondary_list or
 492              * primary_list will be empty, so we can't compare it
 493              * until next comparison.
 494              */
 495             trace_colo_compare_main("packet different");
 496             g_queue_push_tail(&conn->primary_list, pkt);
 497             /* TODO: colo_notify_checkpoint();*/
 498             break;
 499         }
 500     }
 501 }
 502
 503 static int compare_chr_send(CharBackend *out,
 504                             const uint8_t *buf,
 505                             uint32_t size)
 506 {
 507     int ret = 0;
 508     uint32_t len = htonl(size);
 509
 510     if (!size) {
 511         return 0;
 512     }
 513
 514     ret = qemu_chr_fe_write_all(out, (uint8_t *)&len, sizeof(len));
 515     if (ret != sizeof(len)) {
 516         goto err;
 517     }
 518
 519     ret = qemu_chr_fe_write_all(out, (uint8_t *)buf, size);
 520     if (ret != size) {
 521         goto err;
 522     }
 523
 524     return 0;
 525
 526 err:
 527     return ret < 0 ? ret : -EIO;
 528 }
 529
 530 static int compare_chr_can_read(void *opaque)
 531 {
 532     return COMPARE_READ_LEN_MAX;
 533 }
 534
 535 /*
 536  * Called from the main thread on the primary for packets
 537  * arriving over the socket from the primary.
 538  */
 539 static void compare_pri_chr_in(void *opaque, const uint8_t *buf, int size)
 540 {
 541     CompareState *s = COLO_COMPARE(opaque);
 542     int ret;
 543
 544     ret = net_fill_rstate(&s->pri_rs, buf, size);
 545     if (ret == -1) {
 546         qemu_chr_fe_set_handlers(&s->chr_pri_in, NULL, NULL, NULL, NULL,
 547                                  NULL, NULL, true);
 548         error_report("colo-compare primary_in error");
 549     }
 550 }
 551
 552 /*
 553  * Called from the main thread on the primary for packets
 554  * arriving over the socket from the secondary.
 555  */
 556 static void compare_sec_chr_in(void *opaque, const uint8_t *buf, int size)
 557 {
 558     CompareState *s = COLO_COMPARE(opaque);
 559     int ret;
 560
 561     ret = net_fill_rstate(&s->sec_rs, buf, size);
 562     if (ret == -1) {
 563         qemu_chr_fe_set_handlers(&s->chr_sec_in, NULL, NULL, NULL, NULL,
 564                                  NULL, NULL, true);
 565         error_report("colo-compare secondary_in error");
 566     }
 567 }
 568
 569 /*
 570  * Check old packet regularly so it can watch for any packets
 571  * that the secondary hasn't produced equivalents of.
 572  */
 573 static gboolean check_old_packet_regular(void *opaque)
 574 {
 575     CompareState *s = opaque;
 576
 577     /* if have old packet we will notify checkpoint */
 578     colo_old_packet_check(s);
 579
 580     return TRUE;
 581 }
 582
 583 static void *colo_compare_thread(void *opaque)
 584 {
 585     CompareState *s = opaque;
 586     GSource *timeout_source;
 587
 588     s->worker_context = g_main_context_new();
 589
 590     qemu_chr_fe_set_handlers(&s->chr_pri_in, compare_chr_can_read,
 591                              compare_pri_chr_in, NULL, NULL,
 592                              s, s->worker_context, true);
 593     qemu_chr_fe_set_handlers(&s->chr_sec_in, compare_chr_can_read,
 594                              compare_sec_chr_in, NULL, NULL,
 595                              s, s->worker_context, true);
 596
 597     s->compare_loop = g_main_loop_new(s->worker_context, FALSE);
 598
 599     /* To kick any packets that the secondary doesn't match */
 600     timeout_source = g_timeout_source_new(REGULAR_PACKET_CHECK_MS);
 601     g_source_set_callback(timeout_source,
 602                           (GSourceFunc)check_old_packet_regular, s, NULL);
 603     g_source_attach(timeout_source, s->worker_context);
 604
 605     g_main_loop_run(s->compare_loop);
 606
 607     g_source_unref(timeout_source);
 608     g_main_loop_unref(s->compare_loop);
 609     g_main_context_unref(s->worker_context);
 610     return NULL;
 611 }
 612
 613 static char *compare_get_pri_indev(Object *obj, Error **errp)
 614 {
 615     CompareState *s = COLO_COMPARE(obj);
 616
 617     return g_strdup(s->pri_indev);
 618 }
 619
 620 static void compare_set_pri_indev(Object *obj, const char *value, Error **errp)
 621 {
 622     CompareState *s = COLO_COMPARE(obj);
 623
 624     g_free(s->pri_indev);
 625     s->pri_indev = g_strdup(value);
 626 }
 627
 628 static char *compare_get_sec_indev(Object *obj, Error **errp)
 629 {
 630     CompareState *s = COLO_COMPARE(obj);
 631
 632     return g_strdup(s->sec_indev);
 633 }
 634
 635 static void compare_set_sec_indev(Object *obj, const char *value, Error **errp)
 636 {
 637     CompareState *s = COLO_COMPARE(obj);
 638
 639     g_free(s->sec_indev);
 640     s->sec_indev = g_strdup(value);
 641 }
 642
 643 static char *compare_get_outdev(Object *obj, Error **errp)
 644 {
 645     CompareState *s = COLO_COMPARE(obj);
 646
 647     return g_strdup(s->outdev);
 648 }
 649
 650 static void compare_set_outdev(Object *obj, const char *value, Error **errp)
 651 {
 652     CompareState *s = COLO_COMPARE(obj);
 653
 654     g_free(s->outdev);
 655     s->outdev = g_strdup(value);
 656 }
 657
 658 static void compare_pri_rs_finalize(SocketReadState *pri_rs)
 659 {
 660     CompareState *s = container_of(pri_rs, CompareState, pri_rs);
 661
 662     if (packet_enqueue(s, PRIMARY_IN)) {
 663         trace_colo_compare_main("primary: unsupported packet in");
 664         compare_chr_send(&s->chr_out, pri_rs->buf, pri_rs->packet_len);
 665     } else {
 666         /* compare connection */
 667         g_queue_foreach(&s->conn_list, colo_compare_connection, s);
 668     }
 669 }
 670
 671 static void compare_sec_rs_finalize(SocketReadState *sec_rs)
 672 {
 673     CompareState *s = container_of(sec_rs, CompareState, sec_rs);
 674
 675     if (packet_enqueue(s, SECONDARY_IN)) {
 676         trace_colo_compare_main("secondary: unsupported packet in");
 677     } else {
 678         /* compare connection */
 679         g_queue_foreach(&s->conn_list, colo_compare_connection, s);
 680     }
 681 }
 682
 683
 684 /*
 685  * Return 0 is success.
 686  * Return 1 is failed.
 687  */
 688 static int find_and_check_chardev(Chardev **chr,
 689                                   char *chr_name,
 690                                   Error **errp)
 691 {
 692     *chr = qemu_chr_find(chr_name);
 693     if (*chr == NULL) {
 694         error_setg(errp, "Device '%s' not found",
 695                    chr_name);
 696         return 1;
 697     }
 698
 699     if (!qemu_chr_has_feature(*chr, QEMU_CHAR_FEATURE_RECONNECTABLE)) {
 700         error_setg(errp, "chardev \"%s\" is not reconnectable",
 701                    chr_name);
 702         return 1;
 703     }
 704
 705     return 0;
 706 }
 707
 708 /*
 709  * Called from the main thread on the primary
 710  * to setup colo-compare.
 711  */
 712 static void colo_compare_complete(UserCreatable *uc, Error **errp)
 713 {
 714     CompareState *s = COLO_COMPARE(uc);
 715     Chardev *chr;
 716     char thread_name[64];
 717     static int compare_id;
 718
 719     if (!s->pri_indev || !s->sec_indev || !s->outdev) {
 720         error_setg(errp, "colo compare needs 'primary_in' ,"
 721                    "'secondary_in','outdev' property set");
 722         return;
 723     } else if (!strcmp(s->pri_indev, s->outdev) ||
 724                !strcmp(s->sec_indev, s->outdev) ||
 725                !strcmp(s->pri_indev, s->sec_indev)) {
 726         error_setg(errp, "'indev' and 'outdev' could not be same "
 727                    "for compare module");
 728         return;
 729     }
 730
 731     if (find_and_check_chardev(&chr, s->pri_indev, errp) ||
 732         !qemu_chr_fe_init(&s->chr_pri_in, chr, errp)) {
 733         return;
 734     }
 735
 736     if (find_and_check_chardev(&chr, s->sec_indev, errp) ||
 737         !qemu_chr_fe_init(&s->chr_sec_in, chr, errp)) {
 738         return;
 739     }
 740
 741     if (find_and_check_chardev(&chr, s->outdev, errp) ||
 742         !qemu_chr_fe_init(&s->chr_out, chr, errp)) {
 743         return;
 744     }
 745
 746     net_socket_rs_init(&s->pri_rs, compare_pri_rs_finalize);
 747     net_socket_rs_init(&s->sec_rs, compare_sec_rs_finalize);
 748
 749     g_queue_init(&s->conn_list);
 750
 751     s->connection_track_table = g_hash_table_new_full(connection_key_hash,
 752                                                       connection_key_equal,
 753                                                       g_free,
 754                                                       connection_destroy);
 755
 756     sprintf(thread_name, "colo-compare %d", compare_id);
 757     qemu_thread_create(&s->thread, thread_name,
 758                        colo_compare_thread, s,
 759                        QEMU_THREAD_JOINABLE);
 760     compare_id++;
 761
 762     return;
 763 }
 764
 765 static void colo_flush_packets(void *opaque, void *user_data)
 766 {
 767     CompareState *s = user_data;
 768     Connection *conn = opaque;
 769     Packet *pkt = NULL;
 770
 771     while (!g_queue_is_empty(&conn->primary_list)) {
 772         pkt = g_queue_pop_head(&conn->primary_list);
 773         compare_chr_send(&s->chr_out, pkt->data, pkt->size);
 774         packet_destroy(pkt, NULL);
 775     }
 776     while (!g_queue_is_empty(&conn->secondary_list)) {
 777         pkt = g_queue_pop_head(&conn->secondary_list);
 778         packet_destroy(pkt, NULL);
 779     }
 780 }
 781
 782 static void colo_compare_class_init(ObjectClass *oc, void *data)
 783 {
 784     UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
 785
 786     ucc->complete = colo_compare_complete;
 787 }
 788
 789 static void colo_compare_init(Object *obj)
 790 {
 791     object_property_add_str(obj, "primary_in",
 792                             compare_get_pri_indev, compare_set_pri_indev,
 793                             NULL);
 794     object_property_add_str(obj, "secondary_in",
 795                             compare_get_sec_indev, compare_set_sec_indev,
 796                             NULL);
 797     object_property_add_str(obj, "outdev",
 798                             compare_get_outdev, compare_set_outdev,
 799                             NULL);
 800 }
 801
 802 static void colo_compare_finalize(Object *obj)
 803 {
 804     CompareState *s = COLO_COMPARE(obj);
 805
 806     qemu_chr_fe_deinit(&s->chr_pri_in, false);
 807     qemu_chr_fe_deinit(&s->chr_sec_in, false);
 808     qemu_chr_fe_deinit(&s->chr_out, false);
 809
 810     g_main_loop_quit(s->compare_loop);
 811     qemu_thread_join(&s->thread);
 812
 813     /* Release all unhandled packets after compare thead exited */
 814     g_queue_foreach(&s->conn_list, colo_flush_packets, s);
 815
 816     g_queue_clear(&s->conn_list);
 817
 818     g_hash_table_destroy(s->connection_track_table);
 819     g_free(s->pri_indev);
 820     g_free(s->sec_indev);
 821     g_free(s->outdev);
 822 }
 823
 824 static const TypeInfo colo_compare_info = {
 825     .name = TYPE_COLO_COMPARE,
 826     .parent = TYPE_OBJECT,
 827     .instance_size = sizeof(CompareState),
 828     .instance_init = colo_compare_init,
 829     .instance_finalize = colo_compare_finalize,
 830     .class_size = sizeof(CompareClass),
 831     .class_init = colo_compare_class_init,
 832     .interfaces = (InterfaceInfo[]) {
 833         { TYPE_USER_CREATABLE },
 834         { }
 835     }
 836 };
 837
 838 static void register_types(void)
 839 {
 840     type_register_static(&colo_compare_info);
 841 }
 842
 843 type_init(register_types);