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
9 * Author: Zhang Chen <zhangchen.fnst@cn.fujitsu.com>
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-common.h"
17 #include "qemu/error-report.h"
19 #include "qapi/error.h"
22 #include "qom/object_interfaces.h"
24 #include "qom/object.h"
25 #include "net/queue.h"
26 #include "chardev/char-fe.h"
27 #include "qemu/sockets.h"
29 #include "sysemu/iothread.h"
30 #include "net/colo-compare.h"
31 #include "migration/colo.h"
32 #include "migration/migration.h"
35 #include "block/aio-wait.h"
36 #include "qemu/coroutine.h"
38 #define TYPE_COLO_COMPARE "colo-compare"
39 typedef struct CompareState CompareState
;
40 DECLARE_INSTANCE_CHECKER(CompareState
, COLO_COMPARE
,
43 static QTAILQ_HEAD(, CompareState
) net_compares
=
44 QTAILQ_HEAD_INITIALIZER(net_compares
);
46 static NotifierList colo_compare_notifiers
=
47 NOTIFIER_LIST_INITIALIZER(colo_compare_notifiers
);
49 #define COMPARE_READ_LEN_MAX NET_BUFSIZE
50 #define MAX_QUEUE_SIZE 1024
52 #define COLO_COMPARE_FREE_PRIMARY 0x01
53 #define COLO_COMPARE_FREE_SECONDARY 0x02
55 #define REGULAR_PACKET_CHECK_MS 1000
56 #define DEFAULT_TIME_OUT_MS 3000
58 /* #define DEBUG_COLO_PACKETS */
60 static QemuMutex colo_compare_mutex
;
61 static bool colo_compare_active
;
62 static QemuMutex event_mtx
;
63 static QemuCond event_complete_cond
;
64 static int event_unhandled_count
;
65 static uint32_t max_queue_size
;
70 * +---------------+ +---------------+ +---------------+
71 * | conn list + - > conn + ------- > conn + -- > ......
72 * +---------------+ +---------------+ +---------------+
74 * +---------------+ +---v----+ +---v----+ +---v----+ +---v----+
75 * |primary | |secondary |primary | |secondary
76 * |packet | |packet + |packet | |packet +
77 * +--------+ +--------+ +--------+ +--------+
79 * +---v----+ +---v----+ +---v----+ +---v----+
80 * |primary | |secondary |primary | |secondary
81 * |packet | |packet + |packet | |packet +
82 * +--------+ +--------+ +--------+ +--------+
84 * +---v----+ +---v----+ +---v----+ +---v----+
85 * |primary | |secondary |primary | |secondary
86 * |packet | |packet + |packet | |packet +
87 * +--------+ +--------+ +--------+ +--------+
90 typedef struct SendCo
{
92 struct CompareState
*s
;
95 bool notify_remote_frame
;
100 typedef struct SendEntry
{
102 uint32_t vnet_hdr_len
;
106 struct CompareState
{
113 CharBackend chr_pri_in
;
114 CharBackend chr_sec_in
;
116 CharBackend chr_notify_dev
;
117 SocketReadState pri_rs
;
118 SocketReadState sec_rs
;
119 SocketReadState notify_rs
;
121 SendCo notify_sendco
;
123 uint64_t compare_timeout
;
124 uint32_t expired_scan_cycle
;
127 * Record the connection that through the NIC
128 * Element type: Connection
131 /* Record the connection without repetition */
132 GHashTable
*connection_track_table
;
135 GMainContext
*worker_context
;
136 QEMUTimer
*packet_check_timer
;
139 enum colo_event event
;
141 QTAILQ_ENTRY(CompareState
) next
;
144 typedef struct CompareClass
{
145 ObjectClass parent_class
;
153 static const char *colo_mode
[] = {
154 [PRIMARY_IN
] = "primary",
155 [SECONDARY_IN
] = "secondary",
158 static int compare_chr_send(CompareState
*s
,
161 uint32_t vnet_hdr_len
,
162 bool notify_remote_frame
,
165 static bool packet_matches_str(const char *str
,
169 if (packet_len
!= strlen(str
)) {
173 return !memcmp(str
, buf
, strlen(str
));
176 static void notify_remote_frame(CompareState
*s
)
178 char msg
[] = "DO_CHECKPOINT";
181 ret
= compare_chr_send(s
, (uint8_t *)msg
, strlen(msg
), 0, true, false);
183 error_report("Notify Xen COLO-frame failed");
187 static void colo_compare_inconsistency_notify(CompareState
*s
)
190 notify_remote_frame(s
);
192 notifier_list_notify(&colo_compare_notifiers
,
193 migrate_get_current());
197 /* Use restricted to colo_insert_packet() */
198 static gint
seq_sorter(Packet
*a
, Packet
*b
, gpointer data
)
200 return a
->tcp_seq
- b
->tcp_seq
;
203 static void fill_pkt_tcp_info(void *data
, uint32_t *max_ack
)
206 struct tcp_hdr
*tcphd
;
208 tcphd
= (struct tcp_hdr
*)pkt
->transport_header
;
210 pkt
->tcp_seq
= ntohl(tcphd
->th_seq
);
211 pkt
->tcp_ack
= ntohl(tcphd
->th_ack
);
212 *max_ack
= *max_ack
> pkt
->tcp_ack
? *max_ack
: pkt
->tcp_ack
;
213 pkt
->header_size
= pkt
->transport_header
- (uint8_t *)pkt
->data
214 + (tcphd
->th_off
<< 2) - pkt
->vnet_hdr_len
;
215 pkt
->payload_size
= pkt
->size
- pkt
->header_size
;
216 pkt
->seq_end
= pkt
->tcp_seq
+ pkt
->payload_size
;
217 pkt
->flags
= tcphd
->th_flags
;
221 * Return 1 on success, if return 0 means the
222 * packet will be dropped
224 static int colo_insert_packet(GQueue
*queue
, Packet
*pkt
, uint32_t *max_ack
)
226 if (g_queue_get_length(queue
) <= max_queue_size
) {
227 if (pkt
->ip
->ip_p
== IPPROTO_TCP
) {
228 fill_pkt_tcp_info(pkt
, max_ack
);
229 g_queue_insert_sorted(queue
,
231 (GCompareDataFunc
)seq_sorter
,
234 g_queue_push_tail(queue
, pkt
);
242 * Return 0 on success, if return -1 means the pkt
243 * is unsupported(arp and ipv6) and will be sent later
245 static int packet_enqueue(CompareState
*s
, int mode
, Connection
**con
)
252 if (mode
== PRIMARY_IN
) {
253 pkt
= packet_new(s
->pri_rs
.buf
,
254 s
->pri_rs
.packet_len
,
255 s
->pri_rs
.vnet_hdr_len
);
257 pkt
= packet_new(s
->sec_rs
.buf
,
258 s
->sec_rs
.packet_len
,
259 s
->sec_rs
.vnet_hdr_len
);
262 if (parse_packet_early(pkt
)) {
263 packet_destroy(pkt
, NULL
);
267 fill_connection_key(pkt
, &key
);
269 conn
= connection_get(s
->connection_track_table
,
273 if (!conn
->processing
) {
274 g_queue_push_tail(&s
->conn_list
, conn
);
275 conn
->processing
= true;
278 if (mode
== PRIMARY_IN
) {
279 ret
= colo_insert_packet(&conn
->primary_list
, pkt
, &conn
->pack
);
281 ret
= colo_insert_packet(&conn
->secondary_list
, pkt
, &conn
->sack
);
285 trace_colo_compare_drop_packet(colo_mode
[mode
],
286 "queue size too big, drop packet");
287 packet_destroy(pkt
, NULL
);
296 static inline bool after(uint32_t seq1
, uint32_t seq2
)
298 return (int32_t)(seq1
- seq2
) > 0;
301 static void colo_release_primary_pkt(CompareState
*s
, Packet
*pkt
)
304 ret
= compare_chr_send(s
,
311 error_report("colo send primary packet failed");
313 trace_colo_compare_main("packet same and release packet");
314 packet_destroy_partial(pkt
, NULL
);
318 * The IP packets sent by primary and secondary
319 * will be compared in here
320 * TODO support ip fragment, Out-Of-Order
321 * return: 0 means packet same
322 * > 0 || < 0 means packet different
324 static int colo_compare_packet_payload(Packet
*ppkt
,
331 if (trace_event_get_state_backends(TRACE_COLO_COMPARE_IP_INFO
)) {
332 char pri_ip_src
[20], pri_ip_dst
[20], sec_ip_src
[20], sec_ip_dst
[20];
334 strcpy(pri_ip_src
, inet_ntoa(ppkt
->ip
->ip_src
));
335 strcpy(pri_ip_dst
, inet_ntoa(ppkt
->ip
->ip_dst
));
336 strcpy(sec_ip_src
, inet_ntoa(spkt
->ip
->ip_src
));
337 strcpy(sec_ip_dst
, inet_ntoa(spkt
->ip
->ip_dst
));
339 trace_colo_compare_ip_info(ppkt
->size
, pri_ip_src
,
340 pri_ip_dst
, spkt
->size
,
341 sec_ip_src
, sec_ip_dst
);
344 return memcmp(ppkt
->data
+ poffset
, spkt
->data
+ soffset
, len
);
348 * return true means that the payload is consist and
349 * need to make the next comparison, false means do
352 static bool colo_mark_tcp_pkt(Packet
*ppkt
, Packet
*spkt
,
353 int8_t *mark
, uint32_t max_ack
)
357 if (ppkt
->tcp_seq
== spkt
->tcp_seq
&& ppkt
->seq_end
== spkt
->seq_end
) {
358 if (!colo_compare_packet_payload(ppkt
, spkt
,
359 ppkt
->header_size
, spkt
->header_size
,
360 ppkt
->payload_size
)) {
361 *mark
= COLO_COMPARE_FREE_SECONDARY
| COLO_COMPARE_FREE_PRIMARY
;
366 /* one part of secondary packet payload still need to be compared */
367 if (!after(ppkt
->seq_end
, spkt
->seq_end
)) {
368 if (!colo_compare_packet_payload(ppkt
, spkt
,
369 ppkt
->header_size
+ ppkt
->offset
,
370 spkt
->header_size
+ spkt
->offset
,
371 ppkt
->payload_size
- ppkt
->offset
)) {
372 if (!after(ppkt
->tcp_ack
, max_ack
)) {
373 *mark
= COLO_COMPARE_FREE_PRIMARY
;
374 spkt
->offset
+= ppkt
->payload_size
- ppkt
->offset
;
377 /* secondary guest hasn't ack the data, don't send
384 /* primary packet is longer than secondary packet, compare
385 * the same part and mark the primary packet offset
387 if (!colo_compare_packet_payload(ppkt
, spkt
,
388 ppkt
->header_size
+ ppkt
->offset
,
389 spkt
->header_size
+ spkt
->offset
,
390 spkt
->payload_size
- spkt
->offset
)) {
391 *mark
= COLO_COMPARE_FREE_SECONDARY
;
392 ppkt
->offset
+= spkt
->payload_size
- spkt
->offset
;
400 static void colo_compare_tcp(CompareState
*s
, Connection
*conn
)
402 Packet
*ppkt
= NULL
, *spkt
= NULL
;
406 * If ppkt and spkt have the same payload, but ppkt's ACK
407 * is greater than spkt's ACK, in this case we can not
408 * send the ppkt because it will cause the secondary guest
409 * to miss sending some data in the next. Therefore, we
410 * record the maximum ACK in the current queue at both
411 * primary side and secondary side. Only when the ack is
412 * less than the smaller of the two maximum ack, then we
413 * can ensure that the packet's payload is acknowledged by
414 * primary and secondary.
416 uint32_t min_ack
= conn
->pack
> conn
->sack
? conn
->sack
: conn
->pack
;
419 if (g_queue_is_empty(&conn
->primary_list
)) {
422 ppkt
= g_queue_pop_head(&conn
->primary_list
);
424 if (g_queue_is_empty(&conn
->secondary_list
)) {
425 g_queue_push_head(&conn
->primary_list
, ppkt
);
428 spkt
= g_queue_pop_head(&conn
->secondary_list
);
430 if (ppkt
->tcp_seq
== ppkt
->seq_end
) {
431 colo_release_primary_pkt(s
, ppkt
);
435 if (ppkt
&& conn
->compare_seq
&& !after(ppkt
->seq_end
, conn
->compare_seq
)) {
436 trace_colo_compare_main("pri: this packet has compared");
437 colo_release_primary_pkt(s
, ppkt
);
441 if (spkt
->tcp_seq
== spkt
->seq_end
) {
442 packet_destroy(spkt
, NULL
);
449 if (conn
->compare_seq
&& !after(spkt
->seq_end
, conn
->compare_seq
)) {
450 trace_colo_compare_main("sec: this packet has compared");
451 packet_destroy(spkt
, NULL
);
459 g_queue_push_head(&conn
->secondary_list
, spkt
);
464 if (colo_mark_tcp_pkt(ppkt
, spkt
, &mark
, min_ack
)) {
465 trace_colo_compare_tcp_info("pri",
466 ppkt
->tcp_seq
, ppkt
->tcp_ack
,
467 ppkt
->header_size
, ppkt
->payload_size
,
468 ppkt
->offset
, ppkt
->flags
);
470 trace_colo_compare_tcp_info("sec",
471 spkt
->tcp_seq
, spkt
->tcp_ack
,
472 spkt
->header_size
, spkt
->payload_size
,
473 spkt
->offset
, spkt
->flags
);
475 if (mark
== COLO_COMPARE_FREE_PRIMARY
) {
476 conn
->compare_seq
= ppkt
->seq_end
;
477 colo_release_primary_pkt(s
, ppkt
);
478 g_queue_push_head(&conn
->secondary_list
, spkt
);
480 } else if (mark
== COLO_COMPARE_FREE_SECONDARY
) {
481 conn
->compare_seq
= spkt
->seq_end
;
482 packet_destroy(spkt
, NULL
);
484 } else if (mark
== (COLO_COMPARE_FREE_PRIMARY
| COLO_COMPARE_FREE_SECONDARY
)) {
485 conn
->compare_seq
= ppkt
->seq_end
;
486 colo_release_primary_pkt(s
, ppkt
);
487 packet_destroy(spkt
, NULL
);
491 g_queue_push_head(&conn
->primary_list
, ppkt
);
492 g_queue_push_head(&conn
->secondary_list
, spkt
);
494 #ifdef DEBUG_COLO_PACKETS
495 qemu_hexdump(stderr
, "colo-compare ppkt", ppkt
->data
, ppkt
->size
);
496 qemu_hexdump(stderr
, "colo-compare spkt", spkt
->data
, spkt
->size
);
499 colo_compare_inconsistency_notify(s
);
505 * Called from the compare thread on the primary
506 * for compare udp packet
508 static int colo_packet_compare_udp(Packet
*spkt
, Packet
*ppkt
)
510 uint16_t network_header_length
= ppkt
->ip
->ip_hl
<< 2;
511 uint16_t offset
= network_header_length
+ ETH_HLEN
+ ppkt
->vnet_hdr_len
;
513 trace_colo_compare_main("compare udp");
516 * Because of ppkt and spkt are both in the same connection,
517 * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are
518 * same with spkt. In addition, IP header's Identification is a random
519 * field, we can handle it in IP fragmentation function later.
520 * COLO just concern the response net packet payload from primary guest
521 * and secondary guest are same or not, So we ignored all IP header include
522 * other field like TOS,TTL,IP Checksum. we only need to compare
523 * the ip payload here.
525 if (ppkt
->size
!= spkt
->size
) {
526 trace_colo_compare_main("UDP: payload size of packets are different");
529 if (colo_compare_packet_payload(ppkt
, spkt
, offset
, offset
,
530 ppkt
->size
- offset
)) {
531 trace_colo_compare_udp_miscompare("primary pkt size", ppkt
->size
);
532 trace_colo_compare_udp_miscompare("Secondary pkt size", spkt
->size
);
533 #ifdef DEBUG_COLO_PACKETS
534 qemu_hexdump(stderr
, "colo-compare pri pkt", ppkt
->data
, ppkt
->size
);
535 qemu_hexdump(stderr
, "colo-compare sec pkt", spkt
->data
, spkt
->size
);
544 * Called from the compare thread on the primary
545 * for compare icmp packet
547 static int colo_packet_compare_icmp(Packet
*spkt
, Packet
*ppkt
)
549 uint16_t network_header_length
= ppkt
->ip
->ip_hl
<< 2;
550 uint16_t offset
= network_header_length
+ ETH_HLEN
+ ppkt
->vnet_hdr_len
;
552 trace_colo_compare_main("compare icmp");
555 * Because of ppkt and spkt are both in the same connection,
556 * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are
557 * same with spkt. In addition, IP header's Identification is a random
558 * field, we can handle it in IP fragmentation function later.
559 * COLO just concern the response net packet payload from primary guest
560 * and secondary guest are same or not, So we ignored all IP header include
561 * other field like TOS,TTL,IP Checksum. we only need to compare
562 * the ip payload here.
564 if (ppkt
->size
!= spkt
->size
) {
565 trace_colo_compare_main("ICMP: payload size of packets are different");
568 if (colo_compare_packet_payload(ppkt
, spkt
, offset
, offset
,
569 ppkt
->size
- offset
)) {
570 trace_colo_compare_icmp_miscompare("primary pkt size",
572 trace_colo_compare_icmp_miscompare("Secondary pkt size",
574 #ifdef DEBUG_COLO_PACKETS
575 qemu_hexdump(stderr
, "colo-compare pri pkt", ppkt
->data
, ppkt
->size
);
576 qemu_hexdump(stderr
, "colo-compare sec pkt", spkt
->data
, spkt
->size
);
585 * Called from the compare thread on the primary
586 * for compare other packet
588 static int colo_packet_compare_other(Packet
*spkt
, Packet
*ppkt
)
590 uint16_t offset
= ppkt
->vnet_hdr_len
;
592 trace_colo_compare_main("compare other");
593 if (trace_event_get_state_backends(TRACE_COLO_COMPARE_IP_INFO
)) {
594 char pri_ip_src
[20], pri_ip_dst
[20], sec_ip_src
[20], sec_ip_dst
[20];
596 strcpy(pri_ip_src
, inet_ntoa(ppkt
->ip
->ip_src
));
597 strcpy(pri_ip_dst
, inet_ntoa(ppkt
->ip
->ip_dst
));
598 strcpy(sec_ip_src
, inet_ntoa(spkt
->ip
->ip_src
));
599 strcpy(sec_ip_dst
, inet_ntoa(spkt
->ip
->ip_dst
));
601 trace_colo_compare_ip_info(ppkt
->size
, pri_ip_src
,
602 pri_ip_dst
, spkt
->size
,
603 sec_ip_src
, sec_ip_dst
);
606 if (ppkt
->size
!= spkt
->size
) {
607 trace_colo_compare_main("Other: payload size of packets are different");
610 return colo_compare_packet_payload(ppkt
, spkt
, offset
, offset
,
611 ppkt
->size
- offset
);
614 static int colo_old_packet_check_one(Packet
*pkt
, int64_t *check_time
)
616 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
618 if ((now
- pkt
->creation_ms
) > (*check_time
)) {
619 trace_colo_old_packet_check_found(pkt
->creation_ms
);
626 void colo_compare_register_notifier(Notifier
*notify
)
628 notifier_list_add(&colo_compare_notifiers
, notify
);
631 void colo_compare_unregister_notifier(Notifier
*notify
)
633 notifier_remove(notify
);
636 static int colo_old_packet_check_one_conn(Connection
*conn
,
639 if (!g_queue_is_empty(&conn
->primary_list
)) {
640 if (g_queue_find_custom(&conn
->primary_list
,
642 (GCompareFunc
)colo_old_packet_check_one
))
646 if (!g_queue_is_empty(&conn
->secondary_list
)) {
647 if (g_queue_find_custom(&conn
->secondary_list
,
649 (GCompareFunc
)colo_old_packet_check_one
))
656 /* Do checkpoint will flush old packet */
657 colo_compare_inconsistency_notify(s
);
662 * Look for old packets that the secondary hasn't matched,
663 * if we have some then we have to checkpoint to wake
666 static void colo_old_packet_check(void *opaque
)
668 CompareState
*s
= opaque
;
671 * If we find one old packet, stop finding job and notify
672 * COLO frame do checkpoint.
674 g_queue_find_custom(&s
->conn_list
, s
,
675 (GCompareFunc
)colo_old_packet_check_one_conn
);
678 static void colo_compare_packet(CompareState
*s
, Connection
*conn
,
679 int (*HandlePacket
)(Packet
*spkt
,
683 GList
*result
= NULL
;
685 while (!g_queue_is_empty(&conn
->primary_list
) &&
686 !g_queue_is_empty(&conn
->secondary_list
)) {
687 pkt
= g_queue_pop_head(&conn
->primary_list
);
688 result
= g_queue_find_custom(&conn
->secondary_list
,
689 pkt
, (GCompareFunc
)HandlePacket
);
692 colo_release_primary_pkt(s
, pkt
);
693 g_queue_remove(&conn
->secondary_list
, result
->data
);
696 * If one packet arrive late, the secondary_list or
697 * primary_list will be empty, so we can't compare it
698 * until next comparison. If the packets in the list are
699 * timeout, it will trigger a checkpoint request.
701 trace_colo_compare_main("packet different");
702 g_queue_push_head(&conn
->primary_list
, pkt
);
704 colo_compare_inconsistency_notify(s
);
711 * Called from the compare thread on the primary
712 * for compare packet with secondary list of the
713 * specified connection when a new packet was
716 static void colo_compare_connection(void *opaque
, void *user_data
)
718 CompareState
*s
= user_data
;
719 Connection
*conn
= opaque
;
721 switch (conn
->ip_proto
) {
723 colo_compare_tcp(s
, conn
);
726 colo_compare_packet(s
, conn
, colo_packet_compare_udp
);
729 colo_compare_packet(s
, conn
, colo_packet_compare_icmp
);
732 colo_compare_packet(s
, conn
, colo_packet_compare_other
);
737 static void coroutine_fn
_compare_chr_send(void *opaque
)
739 SendCo
*sendco
= opaque
;
740 CompareState
*s
= sendco
->s
;
743 while (!g_queue_is_empty(&sendco
->send_list
)) {
744 SendEntry
*entry
= g_queue_pop_tail(&sendco
->send_list
);
745 uint32_t len
= htonl(entry
->size
);
747 ret
= qemu_chr_fe_write_all(sendco
->chr
, (uint8_t *)&len
, sizeof(len
));
749 if (ret
!= sizeof(len
)) {
751 g_slice_free(SendEntry
, entry
);
755 if (!sendco
->notify_remote_frame
&& s
->vnet_hdr
) {
757 * We send vnet header len make other module(like filter-redirector)
758 * know how to parse net packet correctly.
760 len
= htonl(entry
->vnet_hdr_len
);
762 ret
= qemu_chr_fe_write_all(sendco
->chr
,
766 if (ret
!= sizeof(len
)) {
768 g_slice_free(SendEntry
, entry
);
773 ret
= qemu_chr_fe_write_all(sendco
->chr
,
774 (uint8_t *)entry
->buf
,
777 if (ret
!= entry
->size
) {
779 g_slice_free(SendEntry
, entry
);
784 g_slice_free(SendEntry
, entry
);
791 while (!g_queue_is_empty(&sendco
->send_list
)) {
792 SendEntry
*entry
= g_queue_pop_tail(&sendco
->send_list
);
794 g_slice_free(SendEntry
, entry
);
796 sendco
->ret
= ret
< 0 ? ret
: -EIO
;
803 static int compare_chr_send(CompareState
*s
,
806 uint32_t vnet_hdr_len
,
807 bool notify_remote_frame
,
813 if (notify_remote_frame
) {
814 sendco
= &s
->notify_sendco
;
816 sendco
= &s
->out_sendco
;
823 entry
= g_slice_new(SendEntry
);
825 entry
->vnet_hdr_len
= vnet_hdr_len
;
829 entry
->buf
= g_malloc(size
);
830 memcpy(entry
->buf
, buf
, size
);
832 g_queue_push_head(&sendco
->send_list
, entry
);
835 sendco
->co
= qemu_coroutine_create(_compare_chr_send
, sendco
);
836 sendco
->done
= false;
837 qemu_coroutine_enter(sendco
->co
);
839 /* report early errors */
848 static int compare_chr_can_read(void *opaque
)
850 return COMPARE_READ_LEN_MAX
;
854 * Called from the main thread on the primary for packets
855 * arriving over the socket from the primary.
857 static void compare_pri_chr_in(void *opaque
, const uint8_t *buf
, int size
)
859 CompareState
*s
= COLO_COMPARE(opaque
);
862 ret
= net_fill_rstate(&s
->pri_rs
, buf
, size
);
864 qemu_chr_fe_set_handlers(&s
->chr_pri_in
, NULL
, NULL
, NULL
, NULL
,
866 error_report("colo-compare primary_in error");
871 * Called from the main thread on the primary for packets
872 * arriving over the socket from the secondary.
874 static void compare_sec_chr_in(void *opaque
, const uint8_t *buf
, int size
)
876 CompareState
*s
= COLO_COMPARE(opaque
);
879 ret
= net_fill_rstate(&s
->sec_rs
, buf
, size
);
881 qemu_chr_fe_set_handlers(&s
->chr_sec_in
, NULL
, NULL
, NULL
, NULL
,
883 error_report("colo-compare secondary_in error");
887 static void compare_notify_chr(void *opaque
, const uint8_t *buf
, int size
)
889 CompareState
*s
= COLO_COMPARE(opaque
);
892 ret
= net_fill_rstate(&s
->notify_rs
, buf
, size
);
894 qemu_chr_fe_set_handlers(&s
->chr_notify_dev
, NULL
, NULL
, NULL
, NULL
,
896 error_report("colo-compare notify_dev error");
901 * Check old packet regularly so it can watch for any packets
902 * that the secondary hasn't produced equivalents of.
904 static void check_old_packet_regular(void *opaque
)
906 CompareState
*s
= opaque
;
908 /* if have old packet we will notify checkpoint */
909 colo_old_packet_check(s
);
910 timer_mod(s
->packet_check_timer
, qemu_clock_get_ms(QEMU_CLOCK_HOST
) +
911 s
->expired_scan_cycle
);
914 /* Public API, Used for COLO frame to notify compare event */
915 void colo_notify_compares_event(void *opaque
, int event
, Error
**errp
)
918 qemu_mutex_lock(&colo_compare_mutex
);
920 if (!colo_compare_active
) {
921 qemu_mutex_unlock(&colo_compare_mutex
);
925 qemu_mutex_lock(&event_mtx
);
926 QTAILQ_FOREACH(s
, &net_compares
, next
) {
928 qemu_bh_schedule(s
->event_bh
);
929 event_unhandled_count
++;
931 /* Wait all compare threads to finish handling this event */
932 while (event_unhandled_count
> 0) {
933 qemu_cond_wait(&event_complete_cond
, &event_mtx
);
936 qemu_mutex_unlock(&event_mtx
);
937 qemu_mutex_unlock(&colo_compare_mutex
);
940 static void colo_compare_timer_init(CompareState
*s
)
942 AioContext
*ctx
= iothread_get_aio_context(s
->iothread
);
944 s
->packet_check_timer
= aio_timer_new(ctx
, QEMU_CLOCK_HOST
,
945 SCALE_MS
, check_old_packet_regular
,
947 timer_mod(s
->packet_check_timer
, qemu_clock_get_ms(QEMU_CLOCK_HOST
) +
948 s
->expired_scan_cycle
);
951 static void colo_compare_timer_del(CompareState
*s
)
953 if (s
->packet_check_timer
) {
954 timer_del(s
->packet_check_timer
);
955 timer_free(s
->packet_check_timer
);
956 s
->packet_check_timer
= NULL
;
960 static void colo_flush_packets(void *opaque
, void *user_data
);
962 static void colo_compare_handle_event(void *opaque
)
964 CompareState
*s
= opaque
;
967 case COLO_EVENT_CHECKPOINT
:
968 g_queue_foreach(&s
->conn_list
, colo_flush_packets
, s
);
970 case COLO_EVENT_FAILOVER
:
976 qemu_mutex_lock(&event_mtx
);
977 assert(event_unhandled_count
> 0);
978 event_unhandled_count
--;
979 qemu_cond_broadcast(&event_complete_cond
);
980 qemu_mutex_unlock(&event_mtx
);
983 static void colo_compare_iothread(CompareState
*s
)
985 AioContext
*ctx
= iothread_get_aio_context(s
->iothread
);
986 object_ref(OBJECT(s
->iothread
));
987 s
->worker_context
= iothread_get_g_main_context(s
->iothread
);
989 qemu_chr_fe_set_handlers(&s
->chr_pri_in
, compare_chr_can_read
,
990 compare_pri_chr_in
, NULL
, NULL
,
991 s
, s
->worker_context
, true);
992 qemu_chr_fe_set_handlers(&s
->chr_sec_in
, compare_chr_can_read
,
993 compare_sec_chr_in
, NULL
, NULL
,
994 s
, s
->worker_context
, true);
996 qemu_chr_fe_set_handlers(&s
->chr_notify_dev
, compare_chr_can_read
,
997 compare_notify_chr
, NULL
, NULL
,
998 s
, s
->worker_context
, true);
1001 colo_compare_timer_init(s
);
1002 s
->event_bh
= aio_bh_new(ctx
, colo_compare_handle_event
, s
);
1005 static char *compare_get_pri_indev(Object
*obj
, Error
**errp
)
1007 CompareState
*s
= COLO_COMPARE(obj
);
1009 return g_strdup(s
->pri_indev
);
1012 static void compare_set_pri_indev(Object
*obj
, const char *value
, Error
**errp
)
1014 CompareState
*s
= COLO_COMPARE(obj
);
1016 g_free(s
->pri_indev
);
1017 s
->pri_indev
= g_strdup(value
);
1020 static char *compare_get_sec_indev(Object
*obj
, Error
**errp
)
1022 CompareState
*s
= COLO_COMPARE(obj
);
1024 return g_strdup(s
->sec_indev
);
1027 static void compare_set_sec_indev(Object
*obj
, const char *value
, Error
**errp
)
1029 CompareState
*s
= COLO_COMPARE(obj
);
1031 g_free(s
->sec_indev
);
1032 s
->sec_indev
= g_strdup(value
);
1035 static char *compare_get_outdev(Object
*obj
, Error
**errp
)
1037 CompareState
*s
= COLO_COMPARE(obj
);
1039 return g_strdup(s
->outdev
);
1042 static void compare_set_outdev(Object
*obj
, const char *value
, Error
**errp
)
1044 CompareState
*s
= COLO_COMPARE(obj
);
1047 s
->outdev
= g_strdup(value
);
1050 static bool compare_get_vnet_hdr(Object
*obj
, Error
**errp
)
1052 CompareState
*s
= COLO_COMPARE(obj
);
1057 static void compare_set_vnet_hdr(Object
*obj
,
1061 CompareState
*s
= COLO_COMPARE(obj
);
1063 s
->vnet_hdr
= value
;
1066 static char *compare_get_notify_dev(Object
*obj
, Error
**errp
)
1068 CompareState
*s
= COLO_COMPARE(obj
);
1070 return g_strdup(s
->notify_dev
);
1073 static void compare_set_notify_dev(Object
*obj
, const char *value
, Error
**errp
)
1075 CompareState
*s
= COLO_COMPARE(obj
);
1077 g_free(s
->notify_dev
);
1078 s
->notify_dev
= g_strdup(value
);
1081 static void compare_get_timeout(Object
*obj
, Visitor
*v
,
1082 const char *name
, void *opaque
,
1085 CompareState
*s
= COLO_COMPARE(obj
);
1086 uint64_t value
= s
->compare_timeout
;
1088 visit_type_uint64(v
, name
, &value
, errp
);
1091 static void compare_set_timeout(Object
*obj
, Visitor
*v
,
1092 const char *name
, void *opaque
,
1095 CompareState
*s
= COLO_COMPARE(obj
);
1098 if (!visit_type_uint32(v
, name
, &value
, errp
)) {
1102 error_setg(errp
, "Property '%s.%s' requires a positive value",
1103 object_get_typename(obj
), name
);
1106 s
->compare_timeout
= value
;
1109 static void compare_get_expired_scan_cycle(Object
*obj
, Visitor
*v
,
1110 const char *name
, void *opaque
,
1113 CompareState
*s
= COLO_COMPARE(obj
);
1114 uint32_t value
= s
->expired_scan_cycle
;
1116 visit_type_uint32(v
, name
, &value
, errp
);
1119 static void compare_set_expired_scan_cycle(Object
*obj
, Visitor
*v
,
1120 const char *name
, void *opaque
,
1123 CompareState
*s
= COLO_COMPARE(obj
);
1126 if (!visit_type_uint32(v
, name
, &value
, errp
)) {
1130 error_setg(errp
, "Property '%s.%s' requires a positive value",
1131 object_get_typename(obj
), name
);
1134 s
->expired_scan_cycle
= value
;
1137 static void get_max_queue_size(Object
*obj
, Visitor
*v
,
1138 const char *name
, void *opaque
,
1141 uint32_t value
= max_queue_size
;
1143 visit_type_uint32(v
, name
, &value
, errp
);
1146 static void set_max_queue_size(Object
*obj
, Visitor
*v
,
1147 const char *name
, void *opaque
,
1150 Error
*local_err
= NULL
;
1153 visit_type_uint64(v
, name
, &value
, &local_err
);
1158 error_setg(&local_err
, "Property '%s.%s' requires a positive value",
1159 object_get_typename(obj
), name
);
1162 max_queue_size
= value
;
1165 error_propagate(errp
, local_err
);
1168 static void compare_pri_rs_finalize(SocketReadState
*pri_rs
)
1170 CompareState
*s
= container_of(pri_rs
, CompareState
, pri_rs
);
1171 Connection
*conn
= NULL
;
1173 if (packet_enqueue(s
, PRIMARY_IN
, &conn
)) {
1174 trace_colo_compare_main("primary: unsupported packet in");
1178 pri_rs
->vnet_hdr_len
,
1182 /* compare packet in the specified connection */
1183 colo_compare_connection(conn
, s
);
1187 static void compare_sec_rs_finalize(SocketReadState
*sec_rs
)
1189 CompareState
*s
= container_of(sec_rs
, CompareState
, sec_rs
);
1190 Connection
*conn
= NULL
;
1192 if (packet_enqueue(s
, SECONDARY_IN
, &conn
)) {
1193 trace_colo_compare_main("secondary: unsupported packet in");
1195 /* compare packet in the specified connection */
1196 colo_compare_connection(conn
, s
);
1200 static void compare_notify_rs_finalize(SocketReadState
*notify_rs
)
1202 CompareState
*s
= container_of(notify_rs
, CompareState
, notify_rs
);
1204 const char msg
[] = "COLO_COMPARE_GET_XEN_INIT";
1207 if (packet_matches_str("COLO_USERSPACE_PROXY_INIT",
1209 notify_rs
->packet_len
)) {
1210 ret
= compare_chr_send(s
, (uint8_t *)msg
, strlen(msg
), 0, true, false);
1212 error_report("Notify Xen COLO-frame INIT failed");
1214 } else if (packet_matches_str("COLO_CHECKPOINT",
1216 notify_rs
->packet_len
)) {
1217 /* colo-compare do checkpoint, flush pri packet and remove sec packet */
1218 g_queue_foreach(&s
->conn_list
, colo_flush_packets
, s
);
1220 error_report("COLO compare got unsupported instruction");
1225 * Return 0 is success.
1226 * Return 1 is failed.
1228 static int find_and_check_chardev(Chardev
**chr
,
1232 *chr
= qemu_chr_find(chr_name
);
1234 error_setg(errp
, "Device '%s' not found",
1239 if (!qemu_chr_has_feature(*chr
, QEMU_CHAR_FEATURE_RECONNECTABLE
)) {
1240 error_setg(errp
, "chardev \"%s\" is not reconnectable",
1245 if (!qemu_chr_has_feature(*chr
, QEMU_CHAR_FEATURE_GCONTEXT
)) {
1246 error_setg(errp
, "chardev \"%s\" cannot switch context",
1255 * Called from the main thread on the primary
1256 * to setup colo-compare.
1258 static void colo_compare_complete(UserCreatable
*uc
, Error
**errp
)
1260 CompareState
*s
= COLO_COMPARE(uc
);
1263 if (!s
->pri_indev
|| !s
->sec_indev
|| !s
->outdev
|| !s
->iothread
) {
1264 error_setg(errp
, "colo compare needs 'primary_in' ,"
1265 "'secondary_in','outdev','iothread' property set");
1267 } else if (!strcmp(s
->pri_indev
, s
->outdev
) ||
1268 !strcmp(s
->sec_indev
, s
->outdev
) ||
1269 !strcmp(s
->pri_indev
, s
->sec_indev
)) {
1270 error_setg(errp
, "'indev' and 'outdev' could not be same "
1271 "for compare module");
1275 if (!s
->compare_timeout
) {
1276 /* Set default value to 3000 MS */
1277 s
->compare_timeout
= DEFAULT_TIME_OUT_MS
;
1280 if (!s
->expired_scan_cycle
) {
1281 /* Set default value to 3000 MS */
1282 s
->expired_scan_cycle
= REGULAR_PACKET_CHECK_MS
;
1285 if (!max_queue_size
) {
1286 /* Set default queue size to 1024 */
1287 max_queue_size
= MAX_QUEUE_SIZE
;
1290 if (find_and_check_chardev(&chr
, s
->pri_indev
, errp
) ||
1291 !qemu_chr_fe_init(&s
->chr_pri_in
, chr
, errp
)) {
1295 if (find_and_check_chardev(&chr
, s
->sec_indev
, errp
) ||
1296 !qemu_chr_fe_init(&s
->chr_sec_in
, chr
, errp
)) {
1300 if (find_and_check_chardev(&chr
, s
->outdev
, errp
) ||
1301 !qemu_chr_fe_init(&s
->chr_out
, chr
, errp
)) {
1305 net_socket_rs_init(&s
->pri_rs
, compare_pri_rs_finalize
, s
->vnet_hdr
);
1306 net_socket_rs_init(&s
->sec_rs
, compare_sec_rs_finalize
, s
->vnet_hdr
);
1308 /* Try to enable remote notify chardev, currently just for Xen COLO */
1309 if (s
->notify_dev
) {
1310 if (find_and_check_chardev(&chr
, s
->notify_dev
, errp
) ||
1311 !qemu_chr_fe_init(&s
->chr_notify_dev
, chr
, errp
)) {
1315 net_socket_rs_init(&s
->notify_rs
, compare_notify_rs_finalize
,
1319 s
->out_sendco
.s
= s
;
1320 s
->out_sendco
.chr
= &s
->chr_out
;
1321 s
->out_sendco
.notify_remote_frame
= false;
1322 s
->out_sendco
.done
= true;
1323 g_queue_init(&s
->out_sendco
.send_list
);
1325 if (s
->notify_dev
) {
1326 s
->notify_sendco
.s
= s
;
1327 s
->notify_sendco
.chr
= &s
->chr_notify_dev
;
1328 s
->notify_sendco
.notify_remote_frame
= true;
1329 s
->notify_sendco
.done
= true;
1330 g_queue_init(&s
->notify_sendco
.send_list
);
1333 g_queue_init(&s
->conn_list
);
1335 s
->connection_track_table
= g_hash_table_new_full(connection_key_hash
,
1336 connection_key_equal
,
1338 connection_destroy
);
1340 colo_compare_iothread(s
);
1342 qemu_mutex_lock(&colo_compare_mutex
);
1343 if (!colo_compare_active
) {
1344 qemu_mutex_init(&event_mtx
);
1345 qemu_cond_init(&event_complete_cond
);
1346 colo_compare_active
= true;
1348 QTAILQ_INSERT_TAIL(&net_compares
, s
, next
);
1349 qemu_mutex_unlock(&colo_compare_mutex
);
1354 static void colo_flush_packets(void *opaque
, void *user_data
)
1356 CompareState
*s
= user_data
;
1357 Connection
*conn
= opaque
;
1360 while (!g_queue_is_empty(&conn
->primary_list
)) {
1361 pkt
= g_queue_pop_head(&conn
->primary_list
);
1368 packet_destroy_partial(pkt
, NULL
);
1370 while (!g_queue_is_empty(&conn
->secondary_list
)) {
1371 pkt
= g_queue_pop_head(&conn
->secondary_list
);
1372 packet_destroy(pkt
, NULL
);
1376 static void colo_compare_class_init(ObjectClass
*oc
, void *data
)
1378 UserCreatableClass
*ucc
= USER_CREATABLE_CLASS(oc
);
1380 ucc
->complete
= colo_compare_complete
;
1383 static void colo_compare_init(Object
*obj
)
1385 CompareState
*s
= COLO_COMPARE(obj
);
1387 object_property_add_str(obj
, "primary_in",
1388 compare_get_pri_indev
, compare_set_pri_indev
);
1389 object_property_add_str(obj
, "secondary_in",
1390 compare_get_sec_indev
, compare_set_sec_indev
);
1391 object_property_add_str(obj
, "outdev",
1392 compare_get_outdev
, compare_set_outdev
);
1393 object_property_add_link(obj
, "iothread", TYPE_IOTHREAD
,
1394 (Object
**)&s
->iothread
,
1395 object_property_allow_set_link
,
1396 OBJ_PROP_LINK_STRONG
);
1397 /* This parameter just for Xen COLO */
1398 object_property_add_str(obj
, "notify_dev",
1399 compare_get_notify_dev
, compare_set_notify_dev
);
1401 object_property_add(obj
, "compare_timeout", "uint64",
1402 compare_get_timeout
,
1403 compare_set_timeout
, NULL
, NULL
);
1405 object_property_add(obj
, "expired_scan_cycle", "uint32",
1406 compare_get_expired_scan_cycle
,
1407 compare_set_expired_scan_cycle
, NULL
, NULL
);
1409 object_property_add(obj
, "max_queue_size", "uint32",
1411 set_max_queue_size
, NULL
, NULL
);
1413 s
->vnet_hdr
= false;
1414 object_property_add_bool(obj
, "vnet_hdr_support", compare_get_vnet_hdr
,
1415 compare_set_vnet_hdr
);
1418 static void colo_compare_finalize(Object
*obj
)
1420 CompareState
*s
= COLO_COMPARE(obj
);
1421 CompareState
*tmp
= NULL
;
1423 qemu_mutex_lock(&colo_compare_mutex
);
1424 QTAILQ_FOREACH(tmp
, &net_compares
, next
) {
1426 QTAILQ_REMOVE(&net_compares
, s
, next
);
1430 if (QTAILQ_EMPTY(&net_compares
)) {
1431 colo_compare_active
= false;
1432 qemu_mutex_destroy(&event_mtx
);
1433 qemu_cond_destroy(&event_complete_cond
);
1435 qemu_mutex_unlock(&colo_compare_mutex
);
1437 qemu_chr_fe_deinit(&s
->chr_pri_in
, false);
1438 qemu_chr_fe_deinit(&s
->chr_sec_in
, false);
1439 qemu_chr_fe_deinit(&s
->chr_out
, false);
1440 if (s
->notify_dev
) {
1441 qemu_chr_fe_deinit(&s
->chr_notify_dev
, false);
1444 colo_compare_timer_del(s
);
1446 qemu_bh_delete(s
->event_bh
);
1448 AioContext
*ctx
= iothread_get_aio_context(s
->iothread
);
1449 aio_context_acquire(ctx
);
1450 AIO_WAIT_WHILE(ctx
, !s
->out_sendco
.done
);
1451 if (s
->notify_dev
) {
1452 AIO_WAIT_WHILE(ctx
, !s
->notify_sendco
.done
);
1454 aio_context_release(ctx
);
1456 /* Release all unhandled packets after compare thead exited */
1457 g_queue_foreach(&s
->conn_list
, colo_flush_packets
, s
);
1458 AIO_WAIT_WHILE(NULL
, !s
->out_sendco
.done
);
1460 g_queue_clear(&s
->conn_list
);
1461 g_queue_clear(&s
->out_sendco
.send_list
);
1462 if (s
->notify_dev
) {
1463 g_queue_clear(&s
->notify_sendco
.send_list
);
1466 if (s
->connection_track_table
) {
1467 g_hash_table_destroy(s
->connection_track_table
);
1470 object_unref(OBJECT(s
->iothread
));
1472 g_free(s
->pri_indev
);
1473 g_free(s
->sec_indev
);
1475 g_free(s
->notify_dev
);
1478 static void __attribute__((__constructor__
)) colo_compare_init_globals(void)
1480 colo_compare_active
= false;
1481 qemu_mutex_init(&colo_compare_mutex
);
1484 static const TypeInfo colo_compare_info
= {
1485 .name
= TYPE_COLO_COMPARE
,
1486 .parent
= TYPE_OBJECT
,
1487 .instance_size
= sizeof(CompareState
),
1488 .instance_init
= colo_compare_init
,
1489 .instance_finalize
= colo_compare_finalize
,
1490 .class_size
= sizeof(CompareClass
),
1491 .class_init
= colo_compare_class_init
,
1492 .interfaces
= (InterfaceInfo
[]) {
1493 { TYPE_USER_CREATABLE
},
1498 static void register_types(void)
1500 type_register_static(&colo_compare_info
);
1503 type_init(register_types
);