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 3000
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 uint32_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
);
481 if (mark
== COLO_COMPARE_FREE_SECONDARY
) {
482 conn
->compare_seq
= spkt
->seq_end
;
483 packet_destroy(spkt
, NULL
);
486 if (mark
== (COLO_COMPARE_FREE_PRIMARY
| COLO_COMPARE_FREE_SECONDARY
)) {
487 conn
->compare_seq
= ppkt
->seq_end
;
488 colo_release_primary_pkt(s
, ppkt
);
489 packet_destroy(spkt
, NULL
);
493 g_queue_push_head(&conn
->primary_list
, ppkt
);
494 g_queue_push_head(&conn
->secondary_list
, spkt
);
496 #ifdef DEBUG_COLO_PACKETS
497 qemu_hexdump(stderr
, "colo-compare ppkt", ppkt
->data
, ppkt
->size
);
498 qemu_hexdump(stderr
, "colo-compare spkt", spkt
->data
, spkt
->size
);
501 colo_compare_inconsistency_notify(s
);
507 * Called from the compare thread on the primary
508 * for compare udp packet
510 static int colo_packet_compare_udp(Packet
*spkt
, Packet
*ppkt
)
512 uint16_t network_header_length
= ppkt
->ip
->ip_hl
<< 2;
513 uint16_t offset
= network_header_length
+ ETH_HLEN
+ ppkt
->vnet_hdr_len
;
515 trace_colo_compare_main("compare udp");
518 * Because of ppkt and spkt are both in the same connection,
519 * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are
520 * same with spkt. In addition, IP header's Identification is a random
521 * field, we can handle it in IP fragmentation function later.
522 * COLO just concern the response net packet payload from primary guest
523 * and secondary guest are same or not, So we ignored all IP header include
524 * other field like TOS,TTL,IP Checksum. we only need to compare
525 * the ip payload here.
527 if (ppkt
->size
!= spkt
->size
) {
528 trace_colo_compare_main("UDP: payload size of packets are different");
531 if (colo_compare_packet_payload(ppkt
, spkt
, offset
, offset
,
532 ppkt
->size
- offset
)) {
533 trace_colo_compare_udp_miscompare("primary pkt size", ppkt
->size
);
534 trace_colo_compare_udp_miscompare("Secondary pkt size", spkt
->size
);
535 #ifdef DEBUG_COLO_PACKETS
536 qemu_hexdump(stderr
, "colo-compare pri pkt", ppkt
->data
, ppkt
->size
);
537 qemu_hexdump(stderr
, "colo-compare sec pkt", spkt
->data
, spkt
->size
);
546 * Called from the compare thread on the primary
547 * for compare icmp packet
549 static int colo_packet_compare_icmp(Packet
*spkt
, Packet
*ppkt
)
551 uint16_t network_header_length
= ppkt
->ip
->ip_hl
<< 2;
552 uint16_t offset
= network_header_length
+ ETH_HLEN
+ ppkt
->vnet_hdr_len
;
554 trace_colo_compare_main("compare icmp");
557 * Because of ppkt and spkt are both in the same connection,
558 * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are
559 * same with spkt. In addition, IP header's Identification is a random
560 * field, we can handle it in IP fragmentation function later.
561 * COLO just concern the response net packet payload from primary guest
562 * and secondary guest are same or not, So we ignored all IP header include
563 * other field like TOS,TTL,IP Checksum. we only need to compare
564 * the ip payload here.
566 if (ppkt
->size
!= spkt
->size
) {
567 trace_colo_compare_main("ICMP: payload size of packets are different");
570 if (colo_compare_packet_payload(ppkt
, spkt
, offset
, offset
,
571 ppkt
->size
- offset
)) {
572 trace_colo_compare_icmp_miscompare("primary pkt size",
574 trace_colo_compare_icmp_miscompare("Secondary pkt size",
576 #ifdef DEBUG_COLO_PACKETS
577 qemu_hexdump(stderr
, "colo-compare pri pkt", ppkt
->data
, ppkt
->size
);
578 qemu_hexdump(stderr
, "colo-compare sec pkt", spkt
->data
, spkt
->size
);
587 * Called from the compare thread on the primary
588 * for compare other packet
590 static int colo_packet_compare_other(Packet
*spkt
, Packet
*ppkt
)
592 uint16_t offset
= ppkt
->vnet_hdr_len
;
594 trace_colo_compare_main("compare other");
595 if (trace_event_get_state_backends(TRACE_COLO_COMPARE_IP_INFO
)) {
596 char pri_ip_src
[20], pri_ip_dst
[20], sec_ip_src
[20], sec_ip_dst
[20];
598 strcpy(pri_ip_src
, inet_ntoa(ppkt
->ip
->ip_src
));
599 strcpy(pri_ip_dst
, inet_ntoa(ppkt
->ip
->ip_dst
));
600 strcpy(sec_ip_src
, inet_ntoa(spkt
->ip
->ip_src
));
601 strcpy(sec_ip_dst
, inet_ntoa(spkt
->ip
->ip_dst
));
603 trace_colo_compare_ip_info(ppkt
->size
, pri_ip_src
,
604 pri_ip_dst
, spkt
->size
,
605 sec_ip_src
, sec_ip_dst
);
608 if (ppkt
->size
!= spkt
->size
) {
609 trace_colo_compare_main("Other: payload size of packets are different");
612 return colo_compare_packet_payload(ppkt
, spkt
, offset
, offset
,
613 ppkt
->size
- offset
);
616 static int colo_old_packet_check_one(Packet
*pkt
, int64_t *check_time
)
618 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
620 if ((now
- pkt
->creation_ms
) > (*check_time
)) {
621 trace_colo_old_packet_check_found(pkt
->creation_ms
);
628 void colo_compare_register_notifier(Notifier
*notify
)
630 notifier_list_add(&colo_compare_notifiers
, notify
);
633 void colo_compare_unregister_notifier(Notifier
*notify
)
635 notifier_remove(notify
);
638 static int colo_old_packet_check_one_conn(Connection
*conn
,
641 GList
*result
= NULL
;
643 result
= g_queue_find_custom(&conn
->primary_list
,
645 (GCompareFunc
)colo_old_packet_check_one
);
648 /* Do checkpoint will flush old packet */
649 colo_compare_inconsistency_notify(s
);
657 * Look for old packets that the secondary hasn't matched,
658 * if we have some then we have to checkpoint to wake
661 static void colo_old_packet_check(void *opaque
)
663 CompareState
*s
= opaque
;
666 * If we find one old packet, stop finding job and notify
667 * COLO frame do checkpoint.
669 g_queue_find_custom(&s
->conn_list
, s
,
670 (GCompareFunc
)colo_old_packet_check_one_conn
);
673 static void colo_compare_packet(CompareState
*s
, Connection
*conn
,
674 int (*HandlePacket
)(Packet
*spkt
,
678 GList
*result
= NULL
;
680 while (!g_queue_is_empty(&conn
->primary_list
) &&
681 !g_queue_is_empty(&conn
->secondary_list
)) {
682 pkt
= g_queue_pop_head(&conn
->primary_list
);
683 result
= g_queue_find_custom(&conn
->secondary_list
,
684 pkt
, (GCompareFunc
)HandlePacket
);
687 colo_release_primary_pkt(s
, pkt
);
688 g_queue_remove(&conn
->secondary_list
, result
->data
);
691 * If one packet arrive late, the secondary_list or
692 * primary_list will be empty, so we can't compare it
693 * until next comparison. If the packets in the list are
694 * timeout, it will trigger a checkpoint request.
696 trace_colo_compare_main("packet different");
697 g_queue_push_head(&conn
->primary_list
, pkt
);
699 colo_compare_inconsistency_notify(s
);
706 * Called from the compare thread on the primary
707 * for compare packet with secondary list of the
708 * specified connection when a new packet was
711 static void colo_compare_connection(void *opaque
, void *user_data
)
713 CompareState
*s
= user_data
;
714 Connection
*conn
= opaque
;
716 switch (conn
->ip_proto
) {
718 colo_compare_tcp(s
, conn
);
721 colo_compare_packet(s
, conn
, colo_packet_compare_udp
);
724 colo_compare_packet(s
, conn
, colo_packet_compare_icmp
);
727 colo_compare_packet(s
, conn
, colo_packet_compare_other
);
732 static void coroutine_fn
_compare_chr_send(void *opaque
)
734 SendCo
*sendco
= opaque
;
735 CompareState
*s
= sendco
->s
;
738 while (!g_queue_is_empty(&sendco
->send_list
)) {
739 SendEntry
*entry
= g_queue_pop_tail(&sendco
->send_list
);
740 uint32_t len
= htonl(entry
->size
);
742 ret
= qemu_chr_fe_write_all(sendco
->chr
, (uint8_t *)&len
, sizeof(len
));
744 if (ret
!= sizeof(len
)) {
746 g_slice_free(SendEntry
, entry
);
750 if (!sendco
->notify_remote_frame
&& s
->vnet_hdr
) {
752 * We send vnet header len make other module(like filter-redirector)
753 * know how to parse net packet correctly.
755 len
= htonl(entry
->vnet_hdr_len
);
757 ret
= qemu_chr_fe_write_all(sendco
->chr
,
761 if (ret
!= sizeof(len
)) {
763 g_slice_free(SendEntry
, entry
);
768 ret
= qemu_chr_fe_write_all(sendco
->chr
,
769 (uint8_t *)entry
->buf
,
772 if (ret
!= entry
->size
) {
774 g_slice_free(SendEntry
, entry
);
779 g_slice_free(SendEntry
, entry
);
786 while (!g_queue_is_empty(&sendco
->send_list
)) {
787 SendEntry
*entry
= g_queue_pop_tail(&sendco
->send_list
);
789 g_slice_free(SendEntry
, entry
);
791 sendco
->ret
= ret
< 0 ? ret
: -EIO
;
798 static int compare_chr_send(CompareState
*s
,
801 uint32_t vnet_hdr_len
,
802 bool notify_remote_frame
,
808 if (notify_remote_frame
) {
809 sendco
= &s
->notify_sendco
;
811 sendco
= &s
->out_sendco
;
818 entry
= g_slice_new(SendEntry
);
820 entry
->vnet_hdr_len
= vnet_hdr_len
;
824 entry
->buf
= g_malloc(size
);
825 memcpy(entry
->buf
, buf
, size
);
827 g_queue_push_head(&sendco
->send_list
, entry
);
830 sendco
->co
= qemu_coroutine_create(_compare_chr_send
, sendco
);
831 sendco
->done
= false;
832 qemu_coroutine_enter(sendco
->co
);
834 /* report early errors */
843 static int compare_chr_can_read(void *opaque
)
845 return COMPARE_READ_LEN_MAX
;
849 * Called from the main thread on the primary for packets
850 * arriving over the socket from the primary.
852 static void compare_pri_chr_in(void *opaque
, const uint8_t *buf
, int size
)
854 CompareState
*s
= COLO_COMPARE(opaque
);
857 ret
= net_fill_rstate(&s
->pri_rs
, buf
, size
);
859 qemu_chr_fe_set_handlers(&s
->chr_pri_in
, NULL
, NULL
, NULL
, NULL
,
861 error_report("colo-compare primary_in error");
866 * Called from the main thread on the primary for packets
867 * arriving over the socket from the secondary.
869 static void compare_sec_chr_in(void *opaque
, const uint8_t *buf
, int size
)
871 CompareState
*s
= COLO_COMPARE(opaque
);
874 ret
= net_fill_rstate(&s
->sec_rs
, buf
, size
);
876 qemu_chr_fe_set_handlers(&s
->chr_sec_in
, NULL
, NULL
, NULL
, NULL
,
878 error_report("colo-compare secondary_in error");
882 static void compare_notify_chr(void *opaque
, const uint8_t *buf
, int size
)
884 CompareState
*s
= COLO_COMPARE(opaque
);
887 ret
= net_fill_rstate(&s
->notify_rs
, buf
, size
);
889 qemu_chr_fe_set_handlers(&s
->chr_notify_dev
, NULL
, NULL
, NULL
, NULL
,
891 error_report("colo-compare notify_dev error");
896 * Check old packet regularly so it can watch for any packets
897 * that the secondary hasn't produced equivalents of.
899 static void check_old_packet_regular(void *opaque
)
901 CompareState
*s
= opaque
;
903 /* if have old packet we will notify checkpoint */
904 colo_old_packet_check(s
);
905 timer_mod(s
->packet_check_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
906 s
->expired_scan_cycle
);
909 /* Public API, Used for COLO frame to notify compare event */
910 void colo_notify_compares_event(void *opaque
, int event
, Error
**errp
)
913 qemu_mutex_lock(&colo_compare_mutex
);
915 if (!colo_compare_active
) {
916 qemu_mutex_unlock(&colo_compare_mutex
);
920 qemu_mutex_lock(&event_mtx
);
921 QTAILQ_FOREACH(s
, &net_compares
, next
) {
923 qemu_bh_schedule(s
->event_bh
);
924 event_unhandled_count
++;
926 /* Wait all compare threads to finish handling this event */
927 while (event_unhandled_count
> 0) {
928 qemu_cond_wait(&event_complete_cond
, &event_mtx
);
931 qemu_mutex_unlock(&event_mtx
);
932 qemu_mutex_unlock(&colo_compare_mutex
);
935 static void colo_compare_timer_init(CompareState
*s
)
937 AioContext
*ctx
= iothread_get_aio_context(s
->iothread
);
939 s
->packet_check_timer
= aio_timer_new(ctx
, QEMU_CLOCK_VIRTUAL
,
940 SCALE_MS
, check_old_packet_regular
,
942 timer_mod(s
->packet_check_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
943 s
->expired_scan_cycle
);
946 static void colo_compare_timer_del(CompareState
*s
)
948 if (s
->packet_check_timer
) {
949 timer_del(s
->packet_check_timer
);
950 timer_free(s
->packet_check_timer
);
951 s
->packet_check_timer
= NULL
;
955 static void colo_flush_packets(void *opaque
, void *user_data
);
957 static void colo_compare_handle_event(void *opaque
)
959 CompareState
*s
= opaque
;
962 case COLO_EVENT_CHECKPOINT
:
963 g_queue_foreach(&s
->conn_list
, colo_flush_packets
, s
);
965 case COLO_EVENT_FAILOVER
:
971 qemu_mutex_lock(&event_mtx
);
972 assert(event_unhandled_count
> 0);
973 event_unhandled_count
--;
974 qemu_cond_broadcast(&event_complete_cond
);
975 qemu_mutex_unlock(&event_mtx
);
978 static void colo_compare_iothread(CompareState
*s
)
980 AioContext
*ctx
= iothread_get_aio_context(s
->iothread
);
981 object_ref(OBJECT(s
->iothread
));
982 s
->worker_context
= iothread_get_g_main_context(s
->iothread
);
984 qemu_chr_fe_set_handlers(&s
->chr_pri_in
, compare_chr_can_read
,
985 compare_pri_chr_in
, NULL
, NULL
,
986 s
, s
->worker_context
, true);
987 qemu_chr_fe_set_handlers(&s
->chr_sec_in
, compare_chr_can_read
,
988 compare_sec_chr_in
, NULL
, NULL
,
989 s
, s
->worker_context
, true);
991 qemu_chr_fe_set_handlers(&s
->chr_notify_dev
, compare_chr_can_read
,
992 compare_notify_chr
, NULL
, NULL
,
993 s
, s
->worker_context
, true);
996 colo_compare_timer_init(s
);
997 s
->event_bh
= aio_bh_new(ctx
, colo_compare_handle_event
, s
);
1000 static char *compare_get_pri_indev(Object
*obj
, Error
**errp
)
1002 CompareState
*s
= COLO_COMPARE(obj
);
1004 return g_strdup(s
->pri_indev
);
1007 static void compare_set_pri_indev(Object
*obj
, const char *value
, Error
**errp
)
1009 CompareState
*s
= COLO_COMPARE(obj
);
1011 g_free(s
->pri_indev
);
1012 s
->pri_indev
= g_strdup(value
);
1015 static char *compare_get_sec_indev(Object
*obj
, Error
**errp
)
1017 CompareState
*s
= COLO_COMPARE(obj
);
1019 return g_strdup(s
->sec_indev
);
1022 static void compare_set_sec_indev(Object
*obj
, const char *value
, Error
**errp
)
1024 CompareState
*s
= COLO_COMPARE(obj
);
1026 g_free(s
->sec_indev
);
1027 s
->sec_indev
= g_strdup(value
);
1030 static char *compare_get_outdev(Object
*obj
, Error
**errp
)
1032 CompareState
*s
= COLO_COMPARE(obj
);
1034 return g_strdup(s
->outdev
);
1037 static void compare_set_outdev(Object
*obj
, const char *value
, Error
**errp
)
1039 CompareState
*s
= COLO_COMPARE(obj
);
1042 s
->outdev
= g_strdup(value
);
1045 static bool compare_get_vnet_hdr(Object
*obj
, Error
**errp
)
1047 CompareState
*s
= COLO_COMPARE(obj
);
1052 static void compare_set_vnet_hdr(Object
*obj
,
1056 CompareState
*s
= COLO_COMPARE(obj
);
1058 s
->vnet_hdr
= value
;
1061 static char *compare_get_notify_dev(Object
*obj
, Error
**errp
)
1063 CompareState
*s
= COLO_COMPARE(obj
);
1065 return g_strdup(s
->notify_dev
);
1068 static void compare_set_notify_dev(Object
*obj
, const char *value
, Error
**errp
)
1070 CompareState
*s
= COLO_COMPARE(obj
);
1072 g_free(s
->notify_dev
);
1073 s
->notify_dev
= g_strdup(value
);
1076 static void compare_get_timeout(Object
*obj
, Visitor
*v
,
1077 const char *name
, void *opaque
,
1080 CompareState
*s
= COLO_COMPARE(obj
);
1081 uint32_t value
= s
->compare_timeout
;
1083 visit_type_uint32(v
, name
, &value
, errp
);
1086 static void compare_set_timeout(Object
*obj
, Visitor
*v
,
1087 const char *name
, void *opaque
,
1090 CompareState
*s
= COLO_COMPARE(obj
);
1093 if (!visit_type_uint32(v
, name
, &value
, errp
)) {
1097 error_setg(errp
, "Property '%s.%s' requires a positive value",
1098 object_get_typename(obj
), name
);
1101 s
->compare_timeout
= value
;
1104 static void compare_get_expired_scan_cycle(Object
*obj
, Visitor
*v
,
1105 const char *name
, void *opaque
,
1108 CompareState
*s
= COLO_COMPARE(obj
);
1109 uint32_t value
= s
->expired_scan_cycle
;
1111 visit_type_uint32(v
, name
, &value
, errp
);
1114 static void compare_set_expired_scan_cycle(Object
*obj
, Visitor
*v
,
1115 const char *name
, void *opaque
,
1118 CompareState
*s
= COLO_COMPARE(obj
);
1121 if (!visit_type_uint32(v
, name
, &value
, errp
)) {
1125 error_setg(errp
, "Property '%s.%s' requires a positive value",
1126 object_get_typename(obj
), name
);
1129 s
->expired_scan_cycle
= value
;
1132 static void get_max_queue_size(Object
*obj
, Visitor
*v
,
1133 const char *name
, void *opaque
,
1136 uint32_t value
= max_queue_size
;
1138 visit_type_uint32(v
, name
, &value
, errp
);
1141 static void set_max_queue_size(Object
*obj
, Visitor
*v
,
1142 const char *name
, void *opaque
,
1145 Error
*local_err
= NULL
;
1148 visit_type_uint32(v
, name
, &value
, &local_err
);
1153 error_setg(&local_err
, "Property '%s.%s' requires a positive value",
1154 object_get_typename(obj
), name
);
1157 max_queue_size
= value
;
1160 error_propagate(errp
, local_err
);
1163 static void compare_pri_rs_finalize(SocketReadState
*pri_rs
)
1165 CompareState
*s
= container_of(pri_rs
, CompareState
, pri_rs
);
1166 Connection
*conn
= NULL
;
1168 if (packet_enqueue(s
, PRIMARY_IN
, &conn
)) {
1169 trace_colo_compare_main("primary: unsupported packet in");
1173 pri_rs
->vnet_hdr_len
,
1177 /* compare packet in the specified connection */
1178 colo_compare_connection(conn
, s
);
1182 static void compare_sec_rs_finalize(SocketReadState
*sec_rs
)
1184 CompareState
*s
= container_of(sec_rs
, CompareState
, sec_rs
);
1185 Connection
*conn
= NULL
;
1187 if (packet_enqueue(s
, SECONDARY_IN
, &conn
)) {
1188 trace_colo_compare_main("secondary: unsupported packet in");
1190 /* compare packet in the specified connection */
1191 colo_compare_connection(conn
, s
);
1195 static void compare_notify_rs_finalize(SocketReadState
*notify_rs
)
1197 CompareState
*s
= container_of(notify_rs
, CompareState
, notify_rs
);
1199 const char msg
[] = "COLO_COMPARE_GET_XEN_INIT";
1202 if (packet_matches_str("COLO_USERSPACE_PROXY_INIT",
1204 notify_rs
->packet_len
)) {
1205 ret
= compare_chr_send(s
, (uint8_t *)msg
, strlen(msg
), 0, true, false);
1207 error_report("Notify Xen COLO-frame INIT failed");
1209 } else if (packet_matches_str("COLO_CHECKPOINT",
1211 notify_rs
->packet_len
)) {
1212 /* colo-compare do checkpoint, flush pri packet and remove sec packet */
1213 g_queue_foreach(&s
->conn_list
, colo_flush_packets
, s
);
1215 error_report("COLO compare got unsupported instruction");
1220 * Return 0 is success.
1221 * Return 1 is failed.
1223 static int find_and_check_chardev(Chardev
**chr
,
1227 *chr
= qemu_chr_find(chr_name
);
1229 error_setg(errp
, "Device '%s' not found",
1234 if (!qemu_chr_has_feature(*chr
, QEMU_CHAR_FEATURE_RECONNECTABLE
)) {
1235 error_setg(errp
, "chardev \"%s\" is not reconnectable",
1240 if (!qemu_chr_has_feature(*chr
, QEMU_CHAR_FEATURE_GCONTEXT
)) {
1241 error_setg(errp
, "chardev \"%s\" cannot switch context",
1250 * Called from the main thread on the primary
1251 * to setup colo-compare.
1253 static void colo_compare_complete(UserCreatable
*uc
, Error
**errp
)
1255 CompareState
*s
= COLO_COMPARE(uc
);
1258 if (!s
->pri_indev
|| !s
->sec_indev
|| !s
->outdev
|| !s
->iothread
) {
1259 error_setg(errp
, "colo compare needs 'primary_in' ,"
1260 "'secondary_in','outdev','iothread' property set");
1262 } else if (!strcmp(s
->pri_indev
, s
->outdev
) ||
1263 !strcmp(s
->sec_indev
, s
->outdev
) ||
1264 !strcmp(s
->pri_indev
, s
->sec_indev
)) {
1265 error_setg(errp
, "'indev' and 'outdev' could not be same "
1266 "for compare module");
1270 if (!s
->compare_timeout
) {
1271 /* Set default value to 3000 MS */
1272 s
->compare_timeout
= DEFAULT_TIME_OUT_MS
;
1275 if (!s
->expired_scan_cycle
) {
1276 /* Set default value to 3000 MS */
1277 s
->expired_scan_cycle
= REGULAR_PACKET_CHECK_MS
;
1280 if (!max_queue_size
) {
1281 /* Set default queue size to 1024 */
1282 max_queue_size
= MAX_QUEUE_SIZE
;
1285 if (find_and_check_chardev(&chr
, s
->pri_indev
, errp
) ||
1286 !qemu_chr_fe_init(&s
->chr_pri_in
, chr
, errp
)) {
1290 if (find_and_check_chardev(&chr
, s
->sec_indev
, errp
) ||
1291 !qemu_chr_fe_init(&s
->chr_sec_in
, chr
, errp
)) {
1295 if (find_and_check_chardev(&chr
, s
->outdev
, errp
) ||
1296 !qemu_chr_fe_init(&s
->chr_out
, chr
, errp
)) {
1300 net_socket_rs_init(&s
->pri_rs
, compare_pri_rs_finalize
, s
->vnet_hdr
);
1301 net_socket_rs_init(&s
->sec_rs
, compare_sec_rs_finalize
, s
->vnet_hdr
);
1303 /* Try to enable remote notify chardev, currently just for Xen COLO */
1304 if (s
->notify_dev
) {
1305 if (find_and_check_chardev(&chr
, s
->notify_dev
, errp
) ||
1306 !qemu_chr_fe_init(&s
->chr_notify_dev
, chr
, errp
)) {
1310 net_socket_rs_init(&s
->notify_rs
, compare_notify_rs_finalize
,
1314 s
->out_sendco
.s
= s
;
1315 s
->out_sendco
.chr
= &s
->chr_out
;
1316 s
->out_sendco
.notify_remote_frame
= false;
1317 s
->out_sendco
.done
= true;
1318 g_queue_init(&s
->out_sendco
.send_list
);
1320 if (s
->notify_dev
) {
1321 s
->notify_sendco
.s
= s
;
1322 s
->notify_sendco
.chr
= &s
->chr_notify_dev
;
1323 s
->notify_sendco
.notify_remote_frame
= true;
1324 s
->notify_sendco
.done
= true;
1325 g_queue_init(&s
->notify_sendco
.send_list
);
1328 g_queue_init(&s
->conn_list
);
1330 s
->connection_track_table
= g_hash_table_new_full(connection_key_hash
,
1331 connection_key_equal
,
1333 connection_destroy
);
1335 colo_compare_iothread(s
);
1337 qemu_mutex_lock(&colo_compare_mutex
);
1338 if (!colo_compare_active
) {
1339 qemu_mutex_init(&event_mtx
);
1340 qemu_cond_init(&event_complete_cond
);
1341 colo_compare_active
= true;
1343 QTAILQ_INSERT_TAIL(&net_compares
, s
, next
);
1344 qemu_mutex_unlock(&colo_compare_mutex
);
1349 static void colo_flush_packets(void *opaque
, void *user_data
)
1351 CompareState
*s
= user_data
;
1352 Connection
*conn
= opaque
;
1355 while (!g_queue_is_empty(&conn
->primary_list
)) {
1356 pkt
= g_queue_pop_head(&conn
->primary_list
);
1363 packet_destroy_partial(pkt
, NULL
);
1365 while (!g_queue_is_empty(&conn
->secondary_list
)) {
1366 pkt
= g_queue_pop_head(&conn
->secondary_list
);
1367 packet_destroy(pkt
, NULL
);
1371 static void colo_compare_class_init(ObjectClass
*oc
, void *data
)
1373 UserCreatableClass
*ucc
= USER_CREATABLE_CLASS(oc
);
1375 ucc
->complete
= colo_compare_complete
;
1378 static void colo_compare_init(Object
*obj
)
1380 CompareState
*s
= COLO_COMPARE(obj
);
1382 object_property_add_str(obj
, "primary_in",
1383 compare_get_pri_indev
, compare_set_pri_indev
);
1384 object_property_add_str(obj
, "secondary_in",
1385 compare_get_sec_indev
, compare_set_sec_indev
);
1386 object_property_add_str(obj
, "outdev",
1387 compare_get_outdev
, compare_set_outdev
);
1388 object_property_add_link(obj
, "iothread", TYPE_IOTHREAD
,
1389 (Object
**)&s
->iothread
,
1390 object_property_allow_set_link
,
1391 OBJ_PROP_LINK_STRONG
);
1392 /* This parameter just for Xen COLO */
1393 object_property_add_str(obj
, "notify_dev",
1394 compare_get_notify_dev
, compare_set_notify_dev
);
1396 object_property_add(obj
, "compare_timeout", "uint32",
1397 compare_get_timeout
,
1398 compare_set_timeout
, NULL
, NULL
);
1400 object_property_add(obj
, "expired_scan_cycle", "uint32",
1401 compare_get_expired_scan_cycle
,
1402 compare_set_expired_scan_cycle
, NULL
, NULL
);
1404 object_property_add(obj
, "max_queue_size", "uint32",
1406 set_max_queue_size
, NULL
, NULL
);
1408 s
->vnet_hdr
= false;
1409 object_property_add_bool(obj
, "vnet_hdr_support", compare_get_vnet_hdr
,
1410 compare_set_vnet_hdr
);
1413 static void colo_compare_finalize(Object
*obj
)
1415 CompareState
*s
= COLO_COMPARE(obj
);
1416 CompareState
*tmp
= NULL
;
1418 qemu_mutex_lock(&colo_compare_mutex
);
1419 QTAILQ_FOREACH(tmp
, &net_compares
, next
) {
1421 QTAILQ_REMOVE(&net_compares
, s
, next
);
1425 if (QTAILQ_EMPTY(&net_compares
)) {
1426 colo_compare_active
= false;
1427 qemu_mutex_destroy(&event_mtx
);
1428 qemu_cond_destroy(&event_complete_cond
);
1430 qemu_mutex_unlock(&colo_compare_mutex
);
1432 qemu_chr_fe_deinit(&s
->chr_pri_in
, false);
1433 qemu_chr_fe_deinit(&s
->chr_sec_in
, false);
1434 qemu_chr_fe_deinit(&s
->chr_out
, false);
1435 if (s
->notify_dev
) {
1436 qemu_chr_fe_deinit(&s
->chr_notify_dev
, false);
1439 colo_compare_timer_del(s
);
1441 qemu_bh_delete(s
->event_bh
);
1443 AioContext
*ctx
= iothread_get_aio_context(s
->iothread
);
1444 aio_context_acquire(ctx
);
1445 AIO_WAIT_WHILE(ctx
, !s
->out_sendco
.done
);
1446 if (s
->notify_dev
) {
1447 AIO_WAIT_WHILE(ctx
, !s
->notify_sendco
.done
);
1449 aio_context_release(ctx
);
1451 /* Release all unhandled packets after compare thead exited */
1452 g_queue_foreach(&s
->conn_list
, colo_flush_packets
, s
);
1453 AIO_WAIT_WHILE(NULL
, !s
->out_sendco
.done
);
1455 g_queue_clear(&s
->conn_list
);
1456 g_queue_clear(&s
->out_sendco
.send_list
);
1457 if (s
->notify_dev
) {
1458 g_queue_clear(&s
->notify_sendco
.send_list
);
1461 if (s
->connection_track_table
) {
1462 g_hash_table_destroy(s
->connection_track_table
);
1465 object_unref(OBJECT(s
->iothread
));
1467 g_free(s
->pri_indev
);
1468 g_free(s
->sec_indev
);
1470 g_free(s
->notify_dev
);
1473 static void __attribute__((__constructor__
)) colo_compare_init_globals(void)
1475 colo_compare_active
= false;
1476 qemu_mutex_init(&colo_compare_mutex
);
1479 static const TypeInfo colo_compare_info
= {
1480 .name
= TYPE_COLO_COMPARE
,
1481 .parent
= TYPE_OBJECT
,
1482 .instance_size
= sizeof(CompareState
),
1483 .instance_init
= colo_compare_init
,
1484 .instance_finalize
= colo_compare_finalize
,
1485 .class_size
= sizeof(CompareClass
),
1486 .class_init
= colo_compare_class_init
,
1487 .interfaces
= (InterfaceInfo
[]) {
1488 { TYPE_USER_CREATABLE
},
1493 static void register_types(void)
1495 type_register_static(&colo_compare_info
);
1498 type_init(register_types
);