2 * COarse-grain LOck-stepping Virtual Machines for Non-stop Service (COLO)
3 * (a.k.a. Fault Tolerance or Continuous Replication)
5 * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
6 * Copyright (c) 2016 FUJITSU LIMITED
7 * Copyright (c) 2016 Intel Corporation
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.
15 #include "qemu/osdep.h"
16 #include "qemu/error-report.h"
18 #include "qemu-common.h"
19 #include "qapi/qmp/qerror.h"
20 #include "qapi/error.h"
23 #include "qom/object_interfaces.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"
33 #define TYPE_COLO_COMPARE "colo-compare"
34 #define COLO_COMPARE(obj) \
35 OBJECT_CHECK(CompareState, (obj), TYPE_COLO_COMPARE)
37 #define COMPARE_READ_LEN_MAX NET_BUFSIZE
38 #define MAX_QUEUE_SIZE 1024
40 /* TODO: Should be configurable */
41 #define REGULAR_PACKET_CHECK_MS 3000
46 +---------------+ +---------------+ +---------------+
47 |conn list +--->conn +--------->conn |
48 +---------------+ +---------------+ +---------------+
50 +---------------+ +---v----+ +---v----+ +---v----+ +---v----+
51 |primary | |secondary |primary | |secondary
52 |packet | |packet + |packet | |packet +
53 +--------+ +--------+ +--------+ +--------+
55 +---v----+ +---v----+ +---v----+ +---v----+
56 |primary | |secondary |primary | |secondary
57 |packet | |packet + |packet | |packet +
58 +--------+ +--------+ +--------+ +--------+
60 +---v----+ +---v----+ +---v----+ +---v----+
61 |primary | |secondary |primary | |secondary
62 |packet | |packet + |packet | |packet +
63 +--------+ +--------+ +--------+ +--------+
65 typedef struct CompareState {
71 CharBackend chr_pri_in;
72 CharBackend chr_sec_in;
74 SocketReadState pri_rs;
75 SocketReadState sec_rs;
77 /* connection list: the connections belonged to this NIC could be found
79 * element type: Connection
82 /* hashtable to save connection */
83 GHashTable *connection_track_table;
84 /* compare thread, a thread for each NIC */
87 GMainContext *worker_context;
88 GMainLoop *compare_loop;
91 typedef struct CompareClass {
92 ObjectClass parent_class;
100 static int compare_chr_send(CharBackend *out,
104 static gint seq_sorter(Packet *a, Packet *b, gpointer data)
106 struct tcphdr *atcp, *btcp;
108 atcp = (struct tcphdr *)(a->transport_header);
109 btcp = (struct tcphdr *)(b->transport_header);
110 return ntohl(atcp->th_seq) - ntohl(btcp->th_seq);
114 * Return 0 on success, if return -1 means the pkt
115 * is unsupported(arp and ipv6) and will be sent later
117 static int packet_enqueue(CompareState *s, int mode)
123 if (mode == PRIMARY_IN) {
124 pkt = packet_new(s->pri_rs.buf,
125 s->pri_rs.packet_len,
126 s->pri_rs.vnet_hdr_len);
128 pkt = packet_new(s->sec_rs.buf,
129 s->sec_rs.packet_len,
130 s->sec_rs.vnet_hdr_len);
133 if (parse_packet_early(pkt)) {
134 packet_destroy(pkt, NULL);
138 fill_connection_key(pkt, &key);
140 conn = connection_get(s->connection_track_table,
144 if (!conn->processing) {
145 g_queue_push_tail(&s->conn_list, conn);
146 conn->processing = true;
149 if (mode == PRIMARY_IN) {
150 if (g_queue_get_length(&conn->primary_list) <=
152 g_queue_push_tail(&conn->primary_list, pkt);
153 if (conn->ip_proto == IPPROTO_TCP) {
154 g_queue_sort(&conn->primary_list,
155 (GCompareDataFunc)seq_sorter,
159 error_report("colo compare primary queue size too big,"
163 if (g_queue_get_length(&conn->secondary_list) <=
165 g_queue_push_tail(&conn->secondary_list, pkt);
166 if (conn->ip_proto == IPPROTO_TCP) {
167 g_queue_sort(&conn->secondary_list,
168 (GCompareDataFunc)seq_sorter,
172 error_report("colo compare secondary queue size too big,"
181 * The IP packets sent by primary and secondary
182 * will be compared in here
183 * TODO support ip fragment, Out-Of-Order
184 * return: 0 means packet same
185 * > 0 || < 0 means packet different
187 static int colo_packet_compare_common(Packet *ppkt, Packet *spkt, int offset)
189 if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) {
190 char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20];
192 strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src));
193 strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst));
194 strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src));
195 strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst));
197 trace_colo_compare_ip_info(ppkt->size, pri_ip_src,
198 pri_ip_dst, spkt->size,
199 sec_ip_src, sec_ip_dst);
202 if (ppkt->size == spkt->size) {
203 return memcmp(ppkt->data + offset, spkt->data + offset,
204 spkt->size - offset);
206 trace_colo_compare_main("Net packet size are not the same");
212 * Called from the compare thread on the primary
213 * for compare tcp packet
214 * compare_tcp copied from Dr. David Alan Gilbert's branch
216 static int colo_packet_compare_tcp(Packet *spkt, Packet *ppkt)
218 struct tcphdr *ptcp, *stcp;
221 trace_colo_compare_main("compare tcp");
223 ptcp = (struct tcphdr *)ppkt->transport_header;
224 stcp = (struct tcphdr *)spkt->transport_header;
227 * The 'identification' field in the IP header is *very* random
228 * it almost never matches. Fudge this by ignoring differences in
229 * unfragmented packets; they'll normally sort themselves out if different
230 * anyway, and it should recover at the TCP level.
231 * An alternative would be to get both the primary and secondary to rewrite
232 * somehow; but that would need some sync traffic to sync the state
234 if (ntohs(ppkt->ip->ip_off) & IP_DF) {
235 spkt->ip->ip_id = ppkt->ip->ip_id;
236 /* and the sum will be different if the IDs were different */
237 spkt->ip->ip_sum = ppkt->ip->ip_sum;
241 * Check tcp header length for tcp option field.
242 * th_off > 5 means this tcp packet have options field.
243 * The tcp options maybe always different.
246 * TCP Timestamps option (TSopt):
251 * +-------+-------+---------------------+---------------------+
252 * |Kind=8 | 10 | TS Value (TSval) |TS Echo Reply (TSecr)|
253 * +-------+-------+---------------------+---------------------+
256 * In this case the primary guest's timestamp always different with
257 * the secondary guest's timestamp. COLO just focus on payload,
258 * so we just need skip this field.
260 if (ptcp->th_off > 5) {
261 ptrdiff_t tcp_offset;
262 tcp_offset = ppkt->transport_header - (uint8_t *)ppkt->data
263 + (ptcp->th_off * 4);
264 res = colo_packet_compare_common(ppkt, spkt, tcp_offset);
265 } else if (ptcp->th_sum == stcp->th_sum) {
266 res = colo_packet_compare_common(ppkt, spkt, ETH_HLEN);
271 if (res != 0 && trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) {
272 char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20];
274 strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src));
275 strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst));
276 strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src));
277 strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst));
279 trace_colo_compare_ip_info(ppkt->size, pri_ip_src,
280 pri_ip_dst, spkt->size,
281 sec_ip_src, sec_ip_dst);
283 trace_colo_compare_tcp_info("pri tcp packet",
289 trace_colo_compare_tcp_info("sec tcp packet",
295 qemu_hexdump((char *)ppkt->data, stderr,
296 "colo-compare ppkt", ppkt->size);
297 qemu_hexdump((char *)spkt->data, stderr,
298 "colo-compare spkt", spkt->size);
305 * Called from the compare thread on the primary
306 * for compare udp packet
308 static int colo_packet_compare_udp(Packet *spkt, Packet *ppkt)
311 int network_header_length = ppkt->ip->ip_hl * 4;
313 trace_colo_compare_main("compare udp");
316 * Because of ppkt and spkt are both in the same connection,
317 * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are
318 * same with spkt. In addition, IP header's Identification is a random
319 * field, we can handle it in IP fragmentation function later.
320 * COLO just concern the response net packet payload from primary guest
321 * and secondary guest are same or not, So we ignored all IP header include
322 * other field like TOS,TTL,IP Checksum. we only need to compare
323 * the ip payload here.
325 ret = colo_packet_compare_common(ppkt, spkt,
326 network_header_length + ETH_HLEN);
329 trace_colo_compare_udp_miscompare("primary pkt size", ppkt->size);
330 trace_colo_compare_udp_miscompare("Secondary pkt size", spkt->size);
331 if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) {
332 qemu_hexdump((char *)ppkt->data, stderr, "colo-compare pri pkt",
334 qemu_hexdump((char *)spkt->data, stderr, "colo-compare sec pkt",
343 * Called from the compare thread on the primary
344 * for compare icmp packet
346 static int colo_packet_compare_icmp(Packet *spkt, Packet *ppkt)
348 int network_header_length = ppkt->ip->ip_hl * 4;
350 trace_colo_compare_main("compare icmp");
353 * Because of ppkt and spkt are both in the same connection,
354 * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are
355 * same with spkt. In addition, IP header's Identification is a random
356 * field, we can handle it in IP fragmentation function later.
357 * COLO just concern the response net packet payload from primary guest
358 * and secondary guest are same or not, So we ignored all IP header include
359 * other field like TOS,TTL,IP Checksum. we only need to compare
360 * the ip payload here.
362 if (colo_packet_compare_common(ppkt, spkt,
363 network_header_length + ETH_HLEN)) {
364 trace_colo_compare_icmp_miscompare("primary pkt size",
366 trace_colo_compare_icmp_miscompare("Secondary pkt size",
368 if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) {
369 qemu_hexdump((char *)ppkt->data, stderr, "colo-compare pri pkt",
371 qemu_hexdump((char *)spkt->data, stderr, "colo-compare sec pkt",
381 * Called from the compare thread on the primary
382 * for compare other packet
384 static int colo_packet_compare_other(Packet *spkt, Packet *ppkt)
386 trace_colo_compare_main("compare other");
387 if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) {
388 char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20];
390 strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src));
391 strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst));
392 strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src));
393 strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst));
395 trace_colo_compare_ip_info(ppkt->size, pri_ip_src,
396 pri_ip_dst, spkt->size,
397 sec_ip_src, sec_ip_dst);
400 return colo_packet_compare_common(ppkt, spkt, 0);
403 static int colo_old_packet_check_one(Packet *pkt, int64_t *check_time)
405 int64_t now = qemu_clock_get_ms(QEMU_CLOCK_HOST);
407 if ((now - pkt->creation_ms) > (*check_time)) {
408 trace_colo_old_packet_check_found(pkt->creation_ms);
415 static int colo_old_packet_check_one_conn(Connection *conn,
418 GList *result = NULL;
419 int64_t check_time = REGULAR_PACKET_CHECK_MS;
421 result = g_queue_find_custom(&conn->primary_list,
423 (GCompareFunc)colo_old_packet_check_one);
426 /* do checkpoint will flush old packet */
427 /* TODO: colo_notify_checkpoint();*/
435 * Look for old packets that the secondary hasn't matched,
436 * if we have some then we have to checkpoint to wake
439 static void colo_old_packet_check(void *opaque)
441 CompareState *s = opaque;
444 * If we find one old packet, stop finding job and notify
445 * COLO frame do checkpoint.
447 g_queue_find_custom(&s->conn_list, NULL,
448 (GCompareFunc)colo_old_packet_check_one_conn);
452 * Called from the compare thread on the primary
453 * for compare connection
455 static void colo_compare_connection(void *opaque, void *user_data)
457 CompareState *s = user_data;
458 Connection *conn = opaque;
460 GList *result = NULL;
463 while (!g_queue_is_empty(&conn->primary_list) &&
464 !g_queue_is_empty(&conn->secondary_list)) {
465 pkt = g_queue_pop_tail(&conn->primary_list);
466 switch (conn->ip_proto) {
468 result = g_queue_find_custom(&conn->secondary_list,
469 pkt, (GCompareFunc)colo_packet_compare_tcp);
472 result = g_queue_find_custom(&conn->secondary_list,
473 pkt, (GCompareFunc)colo_packet_compare_udp);
476 result = g_queue_find_custom(&conn->secondary_list,
477 pkt, (GCompareFunc)colo_packet_compare_icmp);
480 result = g_queue_find_custom(&conn->secondary_list,
481 pkt, (GCompareFunc)colo_packet_compare_other);
486 ret = compare_chr_send(&s->chr_out, pkt->data, pkt->size);
488 error_report("colo_send_primary_packet failed");
490 trace_colo_compare_main("packet same and release packet");
491 g_queue_remove(&conn->secondary_list, result->data);
492 packet_destroy(pkt, NULL);
495 * If one packet arrive late, the secondary_list or
496 * primary_list will be empty, so we can't compare it
497 * until next comparison.
499 trace_colo_compare_main("packet different");
500 g_queue_push_tail(&conn->primary_list, pkt);
501 /* TODO: colo_notify_checkpoint();*/
507 static int compare_chr_send(CharBackend *out,
512 uint32_t len = htonl(size);
518 ret = qemu_chr_fe_write_all(out, (uint8_t *)&len, sizeof(len));
519 if (ret != sizeof(len)) {
523 ret = qemu_chr_fe_write_all(out, (uint8_t *)buf, size);
531 return ret < 0 ? ret : -EIO;
534 static int compare_chr_can_read(void *opaque)
536 return COMPARE_READ_LEN_MAX;
540 * Called from the main thread on the primary for packets
541 * arriving over the socket from the primary.
543 static void compare_pri_chr_in(void *opaque, const uint8_t *buf, int size)
545 CompareState *s = COLO_COMPARE(opaque);
548 ret = net_fill_rstate(&s->pri_rs, buf, size);
550 qemu_chr_fe_set_handlers(&s->chr_pri_in, NULL, NULL, NULL, NULL,
552 error_report("colo-compare primary_in error");
557 * Called from the main thread on the primary for packets
558 * arriving over the socket from the secondary.
560 static void compare_sec_chr_in(void *opaque, const uint8_t *buf, int size)
562 CompareState *s = COLO_COMPARE(opaque);
565 ret = net_fill_rstate(&s->sec_rs, buf, size);
567 qemu_chr_fe_set_handlers(&s->chr_sec_in, NULL, NULL, NULL, NULL,
569 error_report("colo-compare secondary_in error");
574 * Check old packet regularly so it can watch for any packets
575 * that the secondary hasn't produced equivalents of.
577 static gboolean check_old_packet_regular(void *opaque)
579 CompareState *s = opaque;
581 /* if have old packet we will notify checkpoint */
582 colo_old_packet_check(s);
587 static void *colo_compare_thread(void *opaque)
589 CompareState *s = opaque;
590 GSource *timeout_source;
592 s->worker_context = g_main_context_new();
594 qemu_chr_fe_set_handlers(&s->chr_pri_in, compare_chr_can_read,
595 compare_pri_chr_in, NULL, NULL,
596 s, s->worker_context, true);
597 qemu_chr_fe_set_handlers(&s->chr_sec_in, compare_chr_can_read,
598 compare_sec_chr_in, NULL, NULL,
599 s, s->worker_context, true);
601 s->compare_loop = g_main_loop_new(s->worker_context, FALSE);
603 /* To kick any packets that the secondary doesn't match */
604 timeout_source = g_timeout_source_new(REGULAR_PACKET_CHECK_MS);
605 g_source_set_callback(timeout_source,
606 (GSourceFunc)check_old_packet_regular, s, NULL);
607 g_source_attach(timeout_source, s->worker_context);
609 g_main_loop_run(s->compare_loop);
611 g_source_unref(timeout_source);
612 g_main_loop_unref(s->compare_loop);
613 g_main_context_unref(s->worker_context);
617 static char *compare_get_pri_indev(Object *obj, Error **errp)
619 CompareState *s = COLO_COMPARE(obj);
621 return g_strdup(s->pri_indev);
624 static void compare_set_pri_indev(Object *obj, const char *value, Error **errp)
626 CompareState *s = COLO_COMPARE(obj);
628 g_free(s->pri_indev);
629 s->pri_indev = g_strdup(value);
632 static char *compare_get_sec_indev(Object *obj, Error **errp)
634 CompareState *s = COLO_COMPARE(obj);
636 return g_strdup(s->sec_indev);
639 static void compare_set_sec_indev(Object *obj, const char *value, Error **errp)
641 CompareState *s = COLO_COMPARE(obj);
643 g_free(s->sec_indev);
644 s->sec_indev = g_strdup(value);
647 static char *compare_get_outdev(Object *obj, Error **errp)
649 CompareState *s = COLO_COMPARE(obj);
651 return g_strdup(s->outdev);
654 static void compare_set_outdev(Object *obj, const char *value, Error **errp)
656 CompareState *s = COLO_COMPARE(obj);
659 s->outdev = g_strdup(value);
662 static void compare_pri_rs_finalize(SocketReadState *pri_rs)
664 CompareState *s = container_of(pri_rs, CompareState, pri_rs);
666 if (packet_enqueue(s, PRIMARY_IN)) {
667 trace_colo_compare_main("primary: unsupported packet in");
668 compare_chr_send(&s->chr_out, pri_rs->buf, pri_rs->packet_len);
670 /* compare connection */
671 g_queue_foreach(&s->conn_list, colo_compare_connection, s);
675 static void compare_sec_rs_finalize(SocketReadState *sec_rs)
677 CompareState *s = container_of(sec_rs, CompareState, sec_rs);
679 if (packet_enqueue(s, SECONDARY_IN)) {
680 trace_colo_compare_main("secondary: unsupported packet in");
682 /* compare connection */
683 g_queue_foreach(&s->conn_list, colo_compare_connection, s);
689 * Return 0 is success.
690 * Return 1 is failed.
692 static int find_and_check_chardev(Chardev **chr,
696 *chr = qemu_chr_find(chr_name);
698 error_setg(errp, "Device '%s' not found",
703 if (!qemu_chr_has_feature(*chr, QEMU_CHAR_FEATURE_RECONNECTABLE)) {
704 error_setg(errp, "chardev \"%s\" is not reconnectable",
713 * Called from the main thread on the primary
714 * to setup colo-compare.
716 static void colo_compare_complete(UserCreatable *uc, Error **errp)
718 CompareState *s = COLO_COMPARE(uc);
720 char thread_name[64];
721 static int compare_id;
723 if (!s->pri_indev || !s->sec_indev || !s->outdev) {
724 error_setg(errp, "colo compare needs 'primary_in' ,"
725 "'secondary_in','outdev' property set");
727 } else if (!strcmp(s->pri_indev, s->outdev) ||
728 !strcmp(s->sec_indev, s->outdev) ||
729 !strcmp(s->pri_indev, s->sec_indev)) {
730 error_setg(errp, "'indev' and 'outdev' could not be same "
731 "for compare module");
735 if (find_and_check_chardev(&chr, s->pri_indev, errp) ||
736 !qemu_chr_fe_init(&s->chr_pri_in, chr, errp)) {
740 if (find_and_check_chardev(&chr, s->sec_indev, errp) ||
741 !qemu_chr_fe_init(&s->chr_sec_in, chr, errp)) {
745 if (find_and_check_chardev(&chr, s->outdev, errp) ||
746 !qemu_chr_fe_init(&s->chr_out, chr, errp)) {
750 net_socket_rs_init(&s->pri_rs, compare_pri_rs_finalize, false);
751 net_socket_rs_init(&s->sec_rs, compare_sec_rs_finalize, false);
753 g_queue_init(&s->conn_list);
755 s->connection_track_table = g_hash_table_new_full(connection_key_hash,
756 connection_key_equal,
760 sprintf(thread_name, "colo-compare %d", compare_id);
761 qemu_thread_create(&s->thread, thread_name,
762 colo_compare_thread, s,
763 QEMU_THREAD_JOINABLE);
769 static void colo_flush_packets(void *opaque, void *user_data)
771 CompareState *s = user_data;
772 Connection *conn = opaque;
775 while (!g_queue_is_empty(&conn->primary_list)) {
776 pkt = g_queue_pop_head(&conn->primary_list);
777 compare_chr_send(&s->chr_out, pkt->data, pkt->size);
778 packet_destroy(pkt, NULL);
780 while (!g_queue_is_empty(&conn->secondary_list)) {
781 pkt = g_queue_pop_head(&conn->secondary_list);
782 packet_destroy(pkt, NULL);
786 static void colo_compare_class_init(ObjectClass *oc, void *data)
788 UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
790 ucc->complete = colo_compare_complete;
793 static void colo_compare_init(Object *obj)
795 object_property_add_str(obj, "primary_in",
796 compare_get_pri_indev, compare_set_pri_indev,
798 object_property_add_str(obj, "secondary_in",
799 compare_get_sec_indev, compare_set_sec_indev,
801 object_property_add_str(obj, "outdev",
802 compare_get_outdev, compare_set_outdev,
806 static void colo_compare_finalize(Object *obj)
808 CompareState *s = COLO_COMPARE(obj);
810 qemu_chr_fe_deinit(&s->chr_pri_in, false);
811 qemu_chr_fe_deinit(&s->chr_sec_in, false);
812 qemu_chr_fe_deinit(&s->chr_out, false);
814 g_main_loop_quit(s->compare_loop);
815 qemu_thread_join(&s->thread);
817 /* Release all unhandled packets after compare thead exited */
818 g_queue_foreach(&s->conn_list, colo_flush_packets, s);
820 g_queue_clear(&s->conn_list);
822 g_hash_table_destroy(s->connection_track_table);
823 g_free(s->pri_indev);
824 g_free(s->sec_indev);
828 static const TypeInfo colo_compare_info = {
829 .name = TYPE_COLO_COMPARE,
830 .parent = TYPE_OBJECT,
831 .instance_size = sizeof(CompareState),
832 .instance_init = colo_compare_init,
833 .instance_finalize = colo_compare_finalize,
834 .class_size = sizeof(CompareClass),
835 .class_init = colo_compare_class_init,
836 .interfaces = (InterfaceInfo[]) {
837 { TYPE_USER_CREATABLE },
842 static void register_types(void)
844 type_register_static(&colo_compare_info);
847 type_init(register_types);