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 #include "sysemu/iothread.h" 33 34 #define TYPE_COLO_COMPARE "colo-compare" 35 #define COLO_COMPARE(obj) \ 36 OBJECT_CHECK(CompareState, (obj), TYPE_COLO_COMPARE) 37 38 #define COMPARE_READ_LEN_MAX NET_BUFSIZE 39 #define MAX_QUEUE_SIZE 1024 40 41 /* TODO: Should be configurable */ 42 #define REGULAR_PACKET_CHECK_MS 3000 43 44 /* 45 * + CompareState ++ 46 * | | 47 * +---------------+ +---------------+ +---------------+ 48 * | conn list + - > conn + ------- > conn + -- > ...... 49 * +---------------+ +---------------+ +---------------+ 50 * | | | | | | 51 * +---------------+ +---v----+ +---v----+ +---v----+ +---v----+ 52 * |primary | |secondary |primary | |secondary 53 * |packet | |packet + |packet | |packet + 54 * +--------+ +--------+ +--------+ +--------+ 55 * | | | | 56 * +---v----+ +---v----+ +---v----+ +---v----+ 57 * |primary | |secondary |primary | |secondary 58 * |packet | |packet + |packet | |packet + 59 * +--------+ +--------+ +--------+ +--------+ 60 * | | | | 61 * +---v----+ +---v----+ +---v----+ +---v----+ 62 * |primary | |secondary |primary | |secondary 63 * |packet | |packet + |packet | |packet + 64 * +--------+ +--------+ +--------+ +--------+ 65 */ 66 typedef struct CompareState { 67 Object parent; 68 69 char *pri_indev; 70 char *sec_indev; 71 char *outdev; 72 CharBackend chr_pri_in; 73 CharBackend chr_sec_in; 74 CharBackend chr_out; 75 SocketReadState pri_rs; 76 SocketReadState sec_rs; 77 bool vnet_hdr; 78 79 /* 80 * Record the connection that through the NIC 81 * Element type: Connection 82 */ 83 GQueue conn_list; 84 /* Record the connection without repetition */ 85 GHashTable *connection_track_table; 86 87 IOThread *iothread; 88 GMainContext *worker_context; 89 QEMUTimer *packet_check_timer; 90 } CompareState; 91 92 typedef struct CompareClass { 93 ObjectClass parent_class; 94 } CompareClass; 95 96 enum { 97 PRIMARY_IN = 0, 98 SECONDARY_IN, 99 }; 100 101 static int compare_chr_send(CompareState *s, 102 const uint8_t *buf, 103 uint32_t size, 104 uint32_t vnet_hdr_len); 105 106 static gint seq_sorter(Packet *a, Packet *b, gpointer data) 107 { 108 struct tcphdr *atcp, *btcp; 109 110 atcp = (struct tcphdr *)(a->transport_header); 111 btcp = (struct tcphdr *)(b->transport_header); 112 return ntohl(atcp->th_seq) - ntohl(btcp->th_seq); 113 } 114 115 /* 116 * Return 1 on success, if return 0 means the 117 * packet will be dropped 118 */ 119 static int colo_insert_packet(GQueue *queue, Packet *pkt) 120 { 121 if (g_queue_get_length(queue) <= MAX_QUEUE_SIZE) { 122 if (pkt->ip->ip_p == IPPROTO_TCP) { 123 g_queue_insert_sorted(queue, 124 pkt, 125 (GCompareDataFunc)seq_sorter, 126 NULL); 127 } else { 128 g_queue_push_tail(queue, pkt); 129 } 130 return 1; 131 } 132 return 0; 133 } 134 135 /* 136 * Return 0 on success, if return -1 means the pkt 137 * is unsupported(arp and ipv6) and will be sent later 138 */ 139 static int packet_enqueue(CompareState *s, int mode, Connection **con) 140 { 141 ConnectionKey key; 142 Packet *pkt = NULL; 143 Connection *conn; 144 145 if (mode == PRIMARY_IN) { 146 pkt = packet_new(s->pri_rs.buf, 147 s->pri_rs.packet_len, 148 s->pri_rs.vnet_hdr_len); 149 } else { 150 pkt = packet_new(s->sec_rs.buf, 151 s->sec_rs.packet_len, 152 s->sec_rs.vnet_hdr_len); 153 } 154 155 if (parse_packet_early(pkt)) { 156 packet_destroy(pkt, NULL); 157 pkt = NULL; 158 return -1; 159 } 160 fill_connection_key(pkt, &key); 161 162 conn = connection_get(s->connection_track_table, 163 &key, 164 &s->conn_list); 165 166 if (!conn->processing) { 167 g_queue_push_tail(&s->conn_list, conn); 168 conn->processing = true; 169 } 170 171 if (mode == PRIMARY_IN) { 172 if (!colo_insert_packet(&conn->primary_list, pkt)) { 173 error_report("colo compare primary queue size too big," 174 "drop packet"); 175 } 176 } else { 177 if (!colo_insert_packet(&conn->secondary_list, pkt)) { 178 error_report("colo compare secondary queue size too big," 179 "drop packet"); 180 } 181 } 182 con = &conn; 183 184 return 0; 185 } 186 187 /* 188 * The IP packets sent by primary and secondary 189 * will be compared in here 190 * TODO support ip fragment, Out-Of-Order 191 * return: 0 means packet same 192 * > 0 || < 0 means packet different 193 */ 194 static int colo_packet_compare_common(Packet *ppkt, 195 Packet *spkt, 196 int poffset, 197 int soffset) 198 { 199 if (trace_event_get_state_backends(TRACE_COLO_COMPARE_MISCOMPARE)) { 200 char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20]; 201 202 strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src)); 203 strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst)); 204 strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src)); 205 strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst)); 206 207 trace_colo_compare_ip_info(ppkt->size, pri_ip_src, 208 pri_ip_dst, spkt->size, 209 sec_ip_src, sec_ip_dst); 210 } 211 212 poffset = ppkt->vnet_hdr_len + poffset; 213 soffset = ppkt->vnet_hdr_len + soffset; 214 215 if (ppkt->size - poffset == spkt->size - soffset) { 216 return memcmp(ppkt->data + poffset, 217 spkt->data + soffset, 218 spkt->size - soffset); 219 } else { 220 trace_colo_compare_main("Net packet size are not the same"); 221 return -1; 222 } 223 } 224 225 /* 226 * Called from the compare thread on the primary 227 * for compare tcp packet 228 * compare_tcp copied from Dr. David Alan Gilbert's branch 229 */ 230 static int colo_packet_compare_tcp(Packet *spkt, Packet *ppkt) 231 { 232 struct tcphdr *ptcp, *stcp; 233 int res; 234 235 trace_colo_compare_main("compare tcp"); 236 237 ptcp = (struct tcphdr *)ppkt->transport_header; 238 stcp = (struct tcphdr *)spkt->transport_header; 239 240 /* 241 * The 'identification' field in the IP header is *very* random 242 * it almost never matches. Fudge this by ignoring differences in 243 * unfragmented packets; they'll normally sort themselves out if different 244 * anyway, and it should recover at the TCP level. 245 * An alternative would be to get both the primary and secondary to rewrite 246 * somehow; but that would need some sync traffic to sync the state 247 */ 248 if (ntohs(ppkt->ip->ip_off) & IP_DF) { 249 spkt->ip->ip_id = ppkt->ip->ip_id; 250 /* and the sum will be different if the IDs were different */ 251 spkt->ip->ip_sum = ppkt->ip->ip_sum; 252 } 253 254 /* 255 * Check tcp header length for tcp option field. 256 * th_off > 5 means this tcp packet have options field. 257 * The tcp options maybe always different. 258 * for example: 259 * From RFC 7323. 260 * TCP Timestamps option (TSopt): 261 * Kind: 8 262 * 263 * Length: 10 bytes 264 * 265 * +-------+-------+---------------------+---------------------+ 266 * |Kind=8 | 10 | TS Value (TSval) |TS Echo Reply (TSecr)| 267 * +-------+-------+---------------------+---------------------+ 268 * 1 1 4 4 269 * 270 * In this case the primary guest's timestamp always different with 271 * the secondary guest's timestamp. COLO just focus on payload, 272 * so we just need skip this field. 273 */ 274 if (ptcp->th_off > 5) { 275 ptrdiff_t ptcp_offset, stcp_offset; 276 277 ptcp_offset = ppkt->transport_header - (uint8_t *)ppkt->data 278 + (ptcp->th_off * 4) - ppkt->vnet_hdr_len; 279 stcp_offset = spkt->transport_header - (uint8_t *)spkt->data 280 + (stcp->th_off * 4) - spkt->vnet_hdr_len; 281 282 /* 283 * When network is busy, some tcp options(like sack) will unpredictable 284 * occur in primary side or secondary side. it will make packet size 285 * not same, but the two packet's payload is identical. colo just 286 * care about packet payload, so we skip the option field. 287 */ 288 res = colo_packet_compare_common(ppkt, spkt, ptcp_offset, stcp_offset); 289 } else if (ptcp->th_sum == stcp->th_sum) { 290 res = colo_packet_compare_common(ppkt, spkt, ETH_HLEN, ETH_HLEN); 291 } else { 292 res = -1; 293 } 294 295 if (res != 0 && 296 trace_event_get_state_backends(TRACE_COLO_COMPARE_MISCOMPARE)) { 297 char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20]; 298 299 strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src)); 300 strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst)); 301 strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src)); 302 strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst)); 303 304 trace_colo_compare_ip_info(ppkt->size, pri_ip_src, 305 pri_ip_dst, spkt->size, 306 sec_ip_src, sec_ip_dst); 307 308 trace_colo_compare_tcp_info("pri tcp packet", 309 ntohl(ptcp->th_seq), 310 ntohl(ptcp->th_ack), 311 res, ptcp->th_flags, 312 ppkt->size); 313 314 trace_colo_compare_tcp_info("sec tcp packet", 315 ntohl(stcp->th_seq), 316 ntohl(stcp->th_ack), 317 res, stcp->th_flags, 318 spkt->size); 319 320 qemu_hexdump((char *)ppkt->data, stderr, 321 "colo-compare ppkt", ppkt->size); 322 qemu_hexdump((char *)spkt->data, stderr, 323 "colo-compare spkt", spkt->size); 324 } 325 326 return res; 327 } 328 329 /* 330 * Called from the compare thread on the primary 331 * for compare udp packet 332 */ 333 static int colo_packet_compare_udp(Packet *spkt, Packet *ppkt) 334 { 335 int ret; 336 int network_header_length = ppkt->ip->ip_hl * 4; 337 338 trace_colo_compare_main("compare udp"); 339 340 /* 341 * Because of ppkt and spkt are both in the same connection, 342 * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are 343 * same with spkt. In addition, IP header's Identification is a random 344 * field, we can handle it in IP fragmentation function later. 345 * COLO just concern the response net packet payload from primary guest 346 * and secondary guest are same or not, So we ignored all IP header include 347 * other field like TOS,TTL,IP Checksum. we only need to compare 348 * the ip payload here. 349 */ 350 ret = colo_packet_compare_common(ppkt, spkt, 351 network_header_length + ETH_HLEN, 352 network_header_length + ETH_HLEN); 353 354 if (ret) { 355 trace_colo_compare_udp_miscompare("primary pkt size", ppkt->size); 356 trace_colo_compare_udp_miscompare("Secondary pkt size", spkt->size); 357 if (trace_event_get_state_backends(TRACE_COLO_COMPARE_MISCOMPARE)) { 358 qemu_hexdump((char *)ppkt->data, stderr, "colo-compare pri pkt", 359 ppkt->size); 360 qemu_hexdump((char *)spkt->data, stderr, "colo-compare sec pkt", 361 spkt->size); 362 } 363 } 364 365 return ret; 366 } 367 368 /* 369 * Called from the compare thread on the primary 370 * for compare icmp packet 371 */ 372 static int colo_packet_compare_icmp(Packet *spkt, Packet *ppkt) 373 { 374 int network_header_length = ppkt->ip->ip_hl * 4; 375 376 trace_colo_compare_main("compare icmp"); 377 378 /* 379 * Because of ppkt and spkt are both in the same connection, 380 * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are 381 * same with spkt. In addition, IP header's Identification is a random 382 * field, we can handle it in IP fragmentation function later. 383 * COLO just concern the response net packet payload from primary guest 384 * and secondary guest are same or not, So we ignored all IP header include 385 * other field like TOS,TTL,IP Checksum. we only need to compare 386 * the ip payload here. 387 */ 388 if (colo_packet_compare_common(ppkt, spkt, 389 network_header_length + ETH_HLEN, 390 network_header_length + ETH_HLEN)) { 391 trace_colo_compare_icmp_miscompare("primary pkt size", 392 ppkt->size); 393 trace_colo_compare_icmp_miscompare("Secondary pkt size", 394 spkt->size); 395 if (trace_event_get_state_backends(TRACE_COLO_COMPARE_MISCOMPARE)) { 396 qemu_hexdump((char *)ppkt->data, stderr, "colo-compare pri pkt", 397 ppkt->size); 398 qemu_hexdump((char *)spkt->data, stderr, "colo-compare sec pkt", 399 spkt->size); 400 } 401 return -1; 402 } else { 403 return 0; 404 } 405 } 406 407 /* 408 * Called from the compare thread on the primary 409 * for compare other packet 410 */ 411 static int colo_packet_compare_other(Packet *spkt, Packet *ppkt) 412 { 413 trace_colo_compare_main("compare other"); 414 if (trace_event_get_state_backends(TRACE_COLO_COMPARE_MISCOMPARE)) { 415 char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20]; 416 417 strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src)); 418 strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst)); 419 strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src)); 420 strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst)); 421 422 trace_colo_compare_ip_info(ppkt->size, pri_ip_src, 423 pri_ip_dst, spkt->size, 424 sec_ip_src, sec_ip_dst); 425 } 426 427 return colo_packet_compare_common(ppkt, spkt, 0, 0); 428 } 429 430 static int colo_old_packet_check_one(Packet *pkt, int64_t *check_time) 431 { 432 int64_t now = qemu_clock_get_ms(QEMU_CLOCK_HOST); 433 434 if ((now - pkt->creation_ms) > (*check_time)) { 435 trace_colo_old_packet_check_found(pkt->creation_ms); 436 return 0; 437 } else { 438 return 1; 439 } 440 } 441 442 static int colo_old_packet_check_one_conn(Connection *conn, 443 void *user_data) 444 { 445 GList *result = NULL; 446 int64_t check_time = REGULAR_PACKET_CHECK_MS; 447 448 result = g_queue_find_custom(&conn->primary_list, 449 &check_time, 450 (GCompareFunc)colo_old_packet_check_one); 451 452 if (result) { 453 /* Do checkpoint will flush old packet */ 454 /* 455 * TODO: Notify colo frame to do checkpoint. 456 * colo_compare_inconsistent_notify(); 457 */ 458 return 0; 459 } 460 461 return 1; 462 } 463 464 /* 465 * Look for old packets that the secondary hasn't matched, 466 * if we have some then we have to checkpoint to wake 467 * the secondary up. 468 */ 469 static void colo_old_packet_check(void *opaque) 470 { 471 CompareState *s = opaque; 472 473 /* 474 * If we find one old packet, stop finding job and notify 475 * COLO frame do checkpoint. 476 */ 477 g_queue_find_custom(&s->conn_list, NULL, 478 (GCompareFunc)colo_old_packet_check_one_conn); 479 } 480 481 /* 482 * Called from the compare thread on the primary 483 * for compare packet with secondary list of the 484 * specified connection when a new packet was 485 * queued to it. 486 */ 487 static void colo_compare_connection(void *opaque, void *user_data) 488 { 489 CompareState *s = user_data; 490 Connection *conn = opaque; 491 Packet *pkt = NULL; 492 GList *result = NULL; 493 int ret; 494 495 while (!g_queue_is_empty(&conn->primary_list) && 496 !g_queue_is_empty(&conn->secondary_list)) { 497 pkt = g_queue_pop_head(&conn->primary_list); 498 switch (conn->ip_proto) { 499 case IPPROTO_TCP: 500 result = g_queue_find_custom(&conn->secondary_list, 501 pkt, (GCompareFunc)colo_packet_compare_tcp); 502 break; 503 case IPPROTO_UDP: 504 result = g_queue_find_custom(&conn->secondary_list, 505 pkt, (GCompareFunc)colo_packet_compare_udp); 506 break; 507 case IPPROTO_ICMP: 508 result = g_queue_find_custom(&conn->secondary_list, 509 pkt, (GCompareFunc)colo_packet_compare_icmp); 510 break; 511 default: 512 result = g_queue_find_custom(&conn->secondary_list, 513 pkt, (GCompareFunc)colo_packet_compare_other); 514 break; 515 } 516 517 if (result) { 518 ret = compare_chr_send(s, 519 pkt->data, 520 pkt->size, 521 pkt->vnet_hdr_len); 522 if (ret < 0) { 523 error_report("colo_send_primary_packet failed"); 524 } 525 trace_colo_compare_main("packet same and release packet"); 526 g_queue_remove(&conn->secondary_list, result->data); 527 packet_destroy(pkt, NULL); 528 } else { 529 /* 530 * If one packet arrive late, the secondary_list or 531 * primary_list will be empty, so we can't compare it 532 * until next comparison. 533 */ 534 trace_colo_compare_main("packet different"); 535 g_queue_push_head(&conn->primary_list, pkt); 536 /* TODO: colo_notify_checkpoint();*/ 537 break; 538 } 539 } 540 } 541 542 static int compare_chr_send(CompareState *s, 543 const uint8_t *buf, 544 uint32_t size, 545 uint32_t vnet_hdr_len) 546 { 547 int ret = 0; 548 uint32_t len = htonl(size); 549 550 if (!size) { 551 return 0; 552 } 553 554 ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)&len, sizeof(len)); 555 if (ret != sizeof(len)) { 556 goto err; 557 } 558 559 if (s->vnet_hdr) { 560 /* 561 * We send vnet header len make other module(like filter-redirector) 562 * know how to parse net packet correctly. 563 */ 564 len = htonl(vnet_hdr_len); 565 ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)&len, sizeof(len)); 566 if (ret != sizeof(len)) { 567 goto err; 568 } 569 } 570 571 ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)buf, size); 572 if (ret != size) { 573 goto err; 574 } 575 576 return 0; 577 578 err: 579 return ret < 0 ? ret : -EIO; 580 } 581 582 static int compare_chr_can_read(void *opaque) 583 { 584 return COMPARE_READ_LEN_MAX; 585 } 586 587 /* 588 * Called from the main thread on the primary for packets 589 * arriving over the socket from the primary. 590 */ 591 static void compare_pri_chr_in(void *opaque, const uint8_t *buf, int size) 592 { 593 CompareState *s = COLO_COMPARE(opaque); 594 int ret; 595 596 ret = net_fill_rstate(&s->pri_rs, buf, size); 597 if (ret == -1) { 598 qemu_chr_fe_set_handlers(&s->chr_pri_in, NULL, NULL, NULL, NULL, 599 NULL, NULL, true); 600 error_report("colo-compare primary_in error"); 601 } 602 } 603 604 /* 605 * Called from the main thread on the primary for packets 606 * arriving over the socket from the secondary. 607 */ 608 static void compare_sec_chr_in(void *opaque, const uint8_t *buf, int size) 609 { 610 CompareState *s = COLO_COMPARE(opaque); 611 int ret; 612 613 ret = net_fill_rstate(&s->sec_rs, buf, size); 614 if (ret == -1) { 615 qemu_chr_fe_set_handlers(&s->chr_sec_in, NULL, NULL, NULL, NULL, 616 NULL, NULL, true); 617 error_report("colo-compare secondary_in error"); 618 } 619 } 620 621 /* 622 * Check old packet regularly so it can watch for any packets 623 * that the secondary hasn't produced equivalents of. 624 */ 625 static void check_old_packet_regular(void *opaque) 626 { 627 CompareState *s = opaque; 628 629 /* if have old packet we will notify checkpoint */ 630 colo_old_packet_check(s); 631 timer_mod(s->packet_check_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 632 REGULAR_PACKET_CHECK_MS); 633 } 634 635 static void colo_compare_timer_init(CompareState *s) 636 { 637 AioContext *ctx = iothread_get_aio_context(s->iothread); 638 639 s->packet_check_timer = aio_timer_new(ctx, QEMU_CLOCK_VIRTUAL, 640 SCALE_MS, check_old_packet_regular, 641 s); 642 timer_mod(s->packet_check_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 643 REGULAR_PACKET_CHECK_MS); 644 } 645 646 static void colo_compare_timer_del(CompareState *s) 647 { 648 if (s->packet_check_timer) { 649 timer_del(s->packet_check_timer); 650 timer_free(s->packet_check_timer); 651 s->packet_check_timer = NULL; 652 } 653 } 654 655 static void colo_compare_iothread(CompareState *s) 656 { 657 object_ref(OBJECT(s->iothread)); 658 s->worker_context = iothread_get_g_main_context(s->iothread); 659 660 qemu_chr_fe_set_handlers(&s->chr_pri_in, compare_chr_can_read, 661 compare_pri_chr_in, NULL, NULL, 662 s, s->worker_context, true); 663 qemu_chr_fe_set_handlers(&s->chr_sec_in, compare_chr_can_read, 664 compare_sec_chr_in, NULL, NULL, 665 s, s->worker_context, true); 666 667 colo_compare_timer_init(s); 668 } 669 670 static char *compare_get_pri_indev(Object *obj, Error **errp) 671 { 672 CompareState *s = COLO_COMPARE(obj); 673 674 return g_strdup(s->pri_indev); 675 } 676 677 static void compare_set_pri_indev(Object *obj, const char *value, Error **errp) 678 { 679 CompareState *s = COLO_COMPARE(obj); 680 681 g_free(s->pri_indev); 682 s->pri_indev = g_strdup(value); 683 } 684 685 static char *compare_get_sec_indev(Object *obj, Error **errp) 686 { 687 CompareState *s = COLO_COMPARE(obj); 688 689 return g_strdup(s->sec_indev); 690 } 691 692 static void compare_set_sec_indev(Object *obj, const char *value, Error **errp) 693 { 694 CompareState *s = COLO_COMPARE(obj); 695 696 g_free(s->sec_indev); 697 s->sec_indev = g_strdup(value); 698 } 699 700 static char *compare_get_outdev(Object *obj, Error **errp) 701 { 702 CompareState *s = COLO_COMPARE(obj); 703 704 return g_strdup(s->outdev); 705 } 706 707 static void compare_set_outdev(Object *obj, const char *value, Error **errp) 708 { 709 CompareState *s = COLO_COMPARE(obj); 710 711 g_free(s->outdev); 712 s->outdev = g_strdup(value); 713 } 714 715 static bool compare_get_vnet_hdr(Object *obj, Error **errp) 716 { 717 CompareState *s = COLO_COMPARE(obj); 718 719 return s->vnet_hdr; 720 } 721 722 static void compare_set_vnet_hdr(Object *obj, 723 bool value, 724 Error **errp) 725 { 726 CompareState *s = COLO_COMPARE(obj); 727 728 s->vnet_hdr = value; 729 } 730 731 static void compare_pri_rs_finalize(SocketReadState *pri_rs) 732 { 733 CompareState *s = container_of(pri_rs, CompareState, pri_rs); 734 Connection *conn = NULL; 735 736 if (packet_enqueue(s, PRIMARY_IN, &conn)) { 737 trace_colo_compare_main("primary: unsupported packet in"); 738 compare_chr_send(s, 739 pri_rs->buf, 740 pri_rs->packet_len, 741 pri_rs->vnet_hdr_len); 742 } else { 743 /* compare packet in the specified connection */ 744 colo_compare_connection(conn, s); 745 } 746 } 747 748 static void compare_sec_rs_finalize(SocketReadState *sec_rs) 749 { 750 CompareState *s = container_of(sec_rs, CompareState, sec_rs); 751 Connection *conn = NULL; 752 753 if (packet_enqueue(s, SECONDARY_IN, &conn)) { 754 trace_colo_compare_main("secondary: unsupported packet in"); 755 } else { 756 /* compare packet in the specified connection */ 757 colo_compare_connection(conn, s); 758 } 759 } 760 761 762 /* 763 * Return 0 is success. 764 * Return 1 is failed. 765 */ 766 static int find_and_check_chardev(Chardev **chr, 767 char *chr_name, 768 Error **errp) 769 { 770 *chr = qemu_chr_find(chr_name); 771 if (*chr == NULL) { 772 error_setg(errp, "Device '%s' not found", 773 chr_name); 774 return 1; 775 } 776 777 if (!qemu_chr_has_feature(*chr, QEMU_CHAR_FEATURE_RECONNECTABLE)) { 778 error_setg(errp, "chardev \"%s\" is not reconnectable", 779 chr_name); 780 return 1; 781 } 782 783 return 0; 784 } 785 786 /* 787 * Called from the main thread on the primary 788 * to setup colo-compare. 789 */ 790 static void colo_compare_complete(UserCreatable *uc, Error **errp) 791 { 792 CompareState *s = COLO_COMPARE(uc); 793 Chardev *chr; 794 795 if (!s->pri_indev || !s->sec_indev || !s->outdev || !s->iothread) { 796 error_setg(errp, "colo compare needs 'primary_in' ," 797 "'secondary_in','outdev','iothread' property set"); 798 return; 799 } else if (!strcmp(s->pri_indev, s->outdev) || 800 !strcmp(s->sec_indev, s->outdev) || 801 !strcmp(s->pri_indev, s->sec_indev)) { 802 error_setg(errp, "'indev' and 'outdev' could not be same " 803 "for compare module"); 804 return; 805 } 806 807 if (find_and_check_chardev(&chr, s->pri_indev, errp) || 808 !qemu_chr_fe_init(&s->chr_pri_in, chr, errp)) { 809 return; 810 } 811 812 if (find_and_check_chardev(&chr, s->sec_indev, errp) || 813 !qemu_chr_fe_init(&s->chr_sec_in, chr, errp)) { 814 return; 815 } 816 817 if (find_and_check_chardev(&chr, s->outdev, errp) || 818 !qemu_chr_fe_init(&s->chr_out, chr, errp)) { 819 return; 820 } 821 822 net_socket_rs_init(&s->pri_rs, compare_pri_rs_finalize, s->vnet_hdr); 823 net_socket_rs_init(&s->sec_rs, compare_sec_rs_finalize, s->vnet_hdr); 824 825 g_queue_init(&s->conn_list); 826 827 s->connection_track_table = g_hash_table_new_full(connection_key_hash, 828 connection_key_equal, 829 g_free, 830 connection_destroy); 831 832 colo_compare_iothread(s); 833 return; 834 } 835 836 static void colo_flush_packets(void *opaque, void *user_data) 837 { 838 CompareState *s = user_data; 839 Connection *conn = opaque; 840 Packet *pkt = NULL; 841 842 while (!g_queue_is_empty(&conn->primary_list)) { 843 pkt = g_queue_pop_head(&conn->primary_list); 844 compare_chr_send(s, 845 pkt->data, 846 pkt->size, 847 pkt->vnet_hdr_len); 848 packet_destroy(pkt, NULL); 849 } 850 while (!g_queue_is_empty(&conn->secondary_list)) { 851 pkt = g_queue_pop_head(&conn->secondary_list); 852 packet_destroy(pkt, NULL); 853 } 854 } 855 856 static void colo_compare_class_init(ObjectClass *oc, void *data) 857 { 858 UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc); 859 860 ucc->complete = colo_compare_complete; 861 } 862 863 static void colo_compare_init(Object *obj) 864 { 865 CompareState *s = COLO_COMPARE(obj); 866 867 object_property_add_str(obj, "primary_in", 868 compare_get_pri_indev, compare_set_pri_indev, 869 NULL); 870 object_property_add_str(obj, "secondary_in", 871 compare_get_sec_indev, compare_set_sec_indev, 872 NULL); 873 object_property_add_str(obj, "outdev", 874 compare_get_outdev, compare_set_outdev, 875 NULL); 876 object_property_add_link(obj, "iothread", TYPE_IOTHREAD, 877 (Object **)&s->iothread, 878 object_property_allow_set_link, 879 OBJ_PROP_LINK_UNREF_ON_RELEASE, NULL); 880 881 s->vnet_hdr = false; 882 object_property_add_bool(obj, "vnet_hdr_support", compare_get_vnet_hdr, 883 compare_set_vnet_hdr, NULL); 884 } 885 886 static void colo_compare_finalize(Object *obj) 887 { 888 CompareState *s = COLO_COMPARE(obj); 889 890 qemu_chr_fe_deinit(&s->chr_pri_in, false); 891 qemu_chr_fe_deinit(&s->chr_sec_in, false); 892 qemu_chr_fe_deinit(&s->chr_out, false); 893 if (s->iothread) { 894 colo_compare_timer_del(s); 895 } 896 /* Release all unhandled packets after compare thead exited */ 897 g_queue_foreach(&s->conn_list, colo_flush_packets, s); 898 899 g_queue_clear(&s->conn_list); 900 901 if (s->connection_track_table) { 902 g_hash_table_destroy(s->connection_track_table); 903 } 904 905 if (s->iothread) { 906 object_unref(OBJECT(s->iothread)); 907 } 908 g_free(s->pri_indev); 909 g_free(s->sec_indev); 910 g_free(s->outdev); 911 } 912 913 static const TypeInfo colo_compare_info = { 914 .name = TYPE_COLO_COMPARE, 915 .parent = TYPE_OBJECT, 916 .instance_size = sizeof(CompareState), 917 .instance_init = colo_compare_init, 918 .instance_finalize = colo_compare_finalize, 919 .class_size = sizeof(CompareClass), 920 .class_init = colo_compare_class_init, 921 .interfaces = (InterfaceInfo[]) { 922 { TYPE_USER_CREATABLE }, 923 { } 924 } 925 }; 926 927 static void register_types(void) 928 { 929 type_register_static(&colo_compare_info); 930 } 931 932 type_init(register_types); 933