Merge tag 'pull-loongarch-20241016' of https://gitlab.com/gaosong/qemu into staging
[qemu/armbru.git] / net / colo-compare.c
blobc4ad0ab71fa25c3c677fdd804c0719e0a7b8f483
1 /*
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/error-report.h"
17 #include "trace.h"
18 #include "qapi/error.h"
19 #include "net/net.h"
20 #include "net/eth.h"
21 #include "qom/object_interfaces.h"
22 #include "qemu/iov.h"
23 #include "qom/object.h"
24 #include "net/queue.h"
25 #include "chardev/char-fe.h"
26 #include "qemu/sockets.h"
27 #include "colo.h"
28 #include "sysemu/iothread.h"
29 #include "net/colo-compare.h"
30 #include "migration/colo.h"
31 #include "util.h"
33 #include "block/aio-wait.h"
34 #include "qemu/coroutine.h"
36 #define TYPE_COLO_COMPARE "colo-compare"
37 typedef struct CompareState CompareState;
38 DECLARE_INSTANCE_CHECKER(CompareState, COLO_COMPARE,
39 TYPE_COLO_COMPARE)
41 static QTAILQ_HEAD(, CompareState) net_compares =
42 QTAILQ_HEAD_INITIALIZER(net_compares);
44 static NotifierList colo_compare_notifiers =
45 NOTIFIER_LIST_INITIALIZER(colo_compare_notifiers);
47 #define COMPARE_READ_LEN_MAX NET_BUFSIZE
48 #define MAX_QUEUE_SIZE 1024
50 #define COLO_COMPARE_FREE_PRIMARY 0x01
51 #define COLO_COMPARE_FREE_SECONDARY 0x02
53 #define REGULAR_PACKET_CHECK_MS 1000
54 #define DEFAULT_TIME_OUT_MS 3000
56 /* #define DEBUG_COLO_PACKETS */
58 static QemuMutex colo_compare_mutex;
59 static bool colo_compare_active;
60 static QemuMutex event_mtx;
61 static QemuCond event_complete_cond;
62 static int event_unhandled_count;
63 static uint32_t max_queue_size;
66 * + CompareState ++
67 * | |
68 * +---------------+ +---------------+ +---------------+
69 * | conn list + - > conn + ------- > conn + -- > ......
70 * +---------------+ +---------------+ +---------------+
71 * | | | | | |
72 * +---------------+ +---v----+ +---v----+ +---v----+ +---v----+
73 * |primary | |secondary |primary | |secondary
74 * |packet | |packet + |packet | |packet +
75 * +--------+ +--------+ +--------+ +--------+
76 * | | | |
77 * +---v----+ +---v----+ +---v----+ +---v----+
78 * |primary | |secondary |primary | |secondary
79 * |packet | |packet + |packet | |packet +
80 * +--------+ +--------+ +--------+ +--------+
81 * | | | |
82 * +---v----+ +---v----+ +---v----+ +---v----+
83 * |primary | |secondary |primary | |secondary
84 * |packet | |packet + |packet | |packet +
85 * +--------+ +--------+ +--------+ +--------+
88 typedef struct SendCo {
89 Coroutine *co;
90 struct CompareState *s;
91 CharBackend *chr;
92 GQueue send_list;
93 bool notify_remote_frame;
94 bool done;
95 int ret;
96 } SendCo;
98 typedef struct SendEntry {
99 uint32_t size;
100 uint32_t vnet_hdr_len;
101 uint8_t *buf;
102 } SendEntry;
104 struct CompareState {
105 Object parent;
107 char *pri_indev;
108 char *sec_indev;
109 char *outdev;
110 char *notify_dev;
111 CharBackend chr_pri_in;
112 CharBackend chr_sec_in;
113 CharBackend chr_out;
114 CharBackend chr_notify_dev;
115 SocketReadState pri_rs;
116 SocketReadState sec_rs;
117 SocketReadState notify_rs;
118 SendCo out_sendco;
119 SendCo notify_sendco;
120 bool vnet_hdr;
121 uint64_t compare_timeout;
122 uint32_t expired_scan_cycle;
125 * Record the connection that through the NIC
126 * Element type: Connection
128 GQueue conn_list;
129 /* Record the connection without repetition */
130 GHashTable *connection_track_table;
132 IOThread *iothread;
133 GMainContext *worker_context;
134 QEMUTimer *packet_check_timer;
136 QEMUBH *event_bh;
137 enum colo_event event;
139 QTAILQ_ENTRY(CompareState) next;
142 typedef struct CompareClass {
143 ObjectClass parent_class;
144 } CompareClass;
146 enum {
147 PRIMARY_IN = 0,
148 SECONDARY_IN,
151 static const char *colo_mode[] = {
152 [PRIMARY_IN] = "primary",
153 [SECONDARY_IN] = "secondary",
156 static int compare_chr_send(CompareState *s,
157 uint8_t *buf,
158 uint32_t size,
159 uint32_t vnet_hdr_len,
160 bool notify_remote_frame,
161 bool zero_copy);
163 static bool packet_matches_str(const char *str,
164 const uint8_t *buf,
165 uint32_t packet_len)
167 if (packet_len != strlen(str)) {
168 return false;
171 return !memcmp(str, buf, packet_len);
174 static void notify_remote_frame(CompareState *s)
176 char msg[] = "DO_CHECKPOINT";
177 int ret = 0;
179 ret = compare_chr_send(s, (uint8_t *)msg, strlen(msg), 0, true, false);
180 if (ret < 0) {
181 error_report("Notify Xen COLO-frame failed");
185 static void colo_compare_inconsistency_notify(CompareState *s)
187 if (s->notify_dev) {
188 notify_remote_frame(s);
189 } else {
190 notifier_list_notify(&colo_compare_notifiers,
191 NULL);
195 /* Use restricted to colo_insert_packet() */
196 static gint seq_sorter(Packet *a, Packet *b, gpointer data)
198 return b->tcp_seq - a->tcp_seq;
201 static void fill_pkt_tcp_info(void *data, uint32_t *max_ack)
203 Packet *pkt = data;
204 struct tcp_hdr *tcphd;
206 tcphd = (struct tcp_hdr *)pkt->transport_header;
208 pkt->tcp_seq = ntohl(tcphd->th_seq);
209 pkt->tcp_ack = ntohl(tcphd->th_ack);
210 /* Need to consider ACK will bigger than uint32_t MAX */
211 *max_ack = pkt->tcp_ack - *max_ack > 0 ? pkt->tcp_ack : *max_ack;
212 pkt->header_size = pkt->transport_header - (uint8_t *)pkt->data
213 + (tcphd->th_off << 2);
214 pkt->payload_size = pkt->size - pkt->header_size;
215 pkt->seq_end = pkt->tcp_seq + pkt->payload_size;
216 pkt->flags = tcphd->th_flags;
220 * Return 1 on success, if return 0 means the
221 * packet will be dropped
223 static int colo_insert_packet(GQueue *queue, Packet *pkt, uint32_t *max_ack)
225 if (g_queue_get_length(queue) <= max_queue_size) {
226 if (pkt->ip->ip_p == IPPROTO_TCP) {
227 fill_pkt_tcp_info(pkt, max_ack);
228 g_queue_insert_sorted(queue,
229 pkt,
230 (GCompareDataFunc)seq_sorter,
231 NULL);
232 } else {
233 g_queue_push_tail(queue, pkt);
235 return 1;
237 return 0;
241 * Return 0 on success, if return -1 means the pkt
242 * is unsupported(arp and ipv6) and will be sent later
244 static int packet_enqueue(CompareState *s, int mode, Connection **con)
246 ConnectionKey key;
247 Packet *pkt = NULL;
248 Connection *conn;
249 int ret;
251 if (mode == PRIMARY_IN) {
252 pkt = packet_new(s->pri_rs.buf,
253 s->pri_rs.packet_len,
254 s->pri_rs.vnet_hdr_len);
255 } else {
256 pkt = packet_new(s->sec_rs.buf,
257 s->sec_rs.packet_len,
258 s->sec_rs.vnet_hdr_len);
261 if (parse_packet_early(pkt)) {
262 packet_destroy(pkt, NULL);
263 pkt = NULL;
264 return -1;
266 fill_connection_key(pkt, &key, false);
268 conn = connection_get(s->connection_track_table,
269 &key,
270 &s->conn_list);
272 if (!conn->processing) {
273 g_queue_push_tail(&s->conn_list, conn);
274 conn->processing = true;
277 if (mode == PRIMARY_IN) {
278 ret = colo_insert_packet(&conn->primary_list, pkt, &conn->pack);
279 } else {
280 ret = colo_insert_packet(&conn->secondary_list, pkt, &conn->sack);
283 if (!ret) {
284 trace_colo_compare_drop_packet(colo_mode[mode],
285 "queue size too big, drop packet");
286 packet_destroy(pkt, NULL);
287 pkt = NULL;
290 *con = conn;
292 return 0;
295 static inline bool after(uint32_t seq1, uint32_t seq2)
297 return (int32_t)(seq1 - seq2) > 0;
300 static void colo_release_primary_pkt(CompareState *s, Packet *pkt)
302 int ret;
303 ret = compare_chr_send(s,
304 pkt->data,
305 pkt->size,
306 pkt->vnet_hdr_len,
307 false,
308 true);
309 if (ret < 0) {
310 error_report("colo send primary packet failed");
312 trace_colo_compare_main("packet same and release packet");
313 packet_destroy_partial(pkt, NULL);
317 * The IP packets sent by primary and secondary
318 * will be compared in here
319 * TODO support ip fragment, Out-Of-Order
320 * return: 0 means packet same
321 * > 0 || < 0 means packet different
323 static int colo_compare_packet_payload(Packet *ppkt,
324 Packet *spkt,
325 uint16_t poffset,
326 uint16_t soffset,
327 uint16_t len)
330 if (trace_event_get_state_backends(TRACE_COLO_COMPARE_IP_INFO)) {
331 char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20];
333 strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src));
334 strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst));
335 strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src));
336 strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst));
338 trace_colo_compare_ip_info(ppkt->size, pri_ip_src,
339 pri_ip_dst, spkt->size,
340 sec_ip_src, sec_ip_dst);
343 return memcmp(ppkt->data + poffset, spkt->data + soffset, len);
347 * return true means that the payload is consist and
348 * need to make the next comparison, false means do
349 * the checkpoint
351 static bool colo_mark_tcp_pkt(Packet *ppkt, Packet *spkt,
352 int8_t *mark, uint32_t max_ack)
354 *mark = 0;
356 if (ppkt->tcp_seq == spkt->tcp_seq && ppkt->seq_end == spkt->seq_end) {
357 if (!colo_compare_packet_payload(ppkt, spkt,
358 ppkt->header_size, spkt->header_size,
359 ppkt->payload_size)) {
360 *mark = COLO_COMPARE_FREE_SECONDARY | COLO_COMPARE_FREE_PRIMARY;
361 return true;
365 /* one part of secondary packet payload still need to be compared */
366 if (!after(ppkt->seq_end, spkt->seq_end)) {
367 if (!colo_compare_packet_payload(ppkt, spkt,
368 ppkt->header_size + ppkt->offset,
369 spkt->header_size + spkt->offset,
370 ppkt->payload_size - ppkt->offset)) {
371 if (!after(ppkt->tcp_ack, max_ack)) {
372 *mark = COLO_COMPARE_FREE_PRIMARY;
373 spkt->offset += ppkt->payload_size - ppkt->offset;
374 return true;
375 } else {
376 /* secondary guest hasn't ack the data, don't send
377 * out this packet
379 return false;
382 } else {
383 /* primary packet is longer than secondary packet, compare
384 * the same part and mark the primary packet offset
386 if (!colo_compare_packet_payload(ppkt, spkt,
387 ppkt->header_size + ppkt->offset,
388 spkt->header_size + spkt->offset,
389 spkt->payload_size - spkt->offset)) {
390 *mark = COLO_COMPARE_FREE_SECONDARY;
391 ppkt->offset += spkt->payload_size - spkt->offset;
392 return true;
396 return false;
399 static void colo_compare_tcp(CompareState *s, Connection *conn)
401 Packet *ppkt = NULL, *spkt = NULL;
402 int8_t mark;
405 * If ppkt and spkt have the same payload, but ppkt's ACK
406 * is greater than spkt's ACK, in this case we can not
407 * send the ppkt because it will cause the secondary guest
408 * to miss sending some data in the next. Therefore, we
409 * record the maximum ACK in the current queue at both
410 * primary side and secondary side. Only when the ack is
411 * less than the smaller of the two maximum ack, then we
412 * can ensure that the packet's payload is acknowledged by
413 * primary and secondary.
415 uint32_t min_ack = conn->pack - conn->sack > 0 ?
416 conn->sack : conn->pack;
418 pri:
419 if (g_queue_is_empty(&conn->primary_list)) {
420 return;
422 ppkt = g_queue_pop_tail(&conn->primary_list);
423 sec:
424 if (g_queue_is_empty(&conn->secondary_list)) {
425 g_queue_push_tail(&conn->primary_list, ppkt);
426 return;
428 spkt = g_queue_pop_tail(&conn->secondary_list);
430 if (ppkt->tcp_seq == ppkt->seq_end) {
431 colo_release_primary_pkt(s, ppkt);
432 ppkt = NULL;
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);
438 ppkt = NULL;
441 if (spkt->tcp_seq == spkt->seq_end) {
442 packet_destroy(spkt, NULL);
443 if (!ppkt) {
444 goto pri;
445 } else {
446 goto sec;
448 } else {
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);
452 if (!ppkt) {
453 goto pri;
454 } else {
455 goto sec;
458 if (!ppkt) {
459 g_queue_push_tail(&conn->secondary_list, spkt);
460 goto pri;
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_tail(&conn->secondary_list, spkt);
479 goto pri;
480 } else if (mark == COLO_COMPARE_FREE_SECONDARY) {
481 conn->compare_seq = spkt->seq_end;
482 packet_destroy(spkt, NULL);
483 goto sec;
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);
488 goto pri;
490 } else {
491 g_queue_push_tail(&conn->primary_list, ppkt);
492 g_queue_push_tail(&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);
497 #endif
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");
527 return -1;
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);
536 #endif
537 return -1;
538 } else {
539 return 0;
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");
566 return -1;
568 if (colo_compare_packet_payload(ppkt, spkt, offset, offset,
569 ppkt->size - offset)) {
570 trace_colo_compare_icmp_miscompare("primary pkt size",
571 ppkt->size);
572 trace_colo_compare_icmp_miscompare("Secondary pkt size",
573 spkt->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);
577 #endif
578 return -1;
579 } else {
580 return 0;
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 (ppkt->size != spkt->size) {
594 trace_colo_compare_main("Other: payload size of packets are different");
595 return -1;
597 return colo_compare_packet_payload(ppkt, spkt, offset, offset,
598 ppkt->size - offset);
601 static int colo_old_packet_check_one(Packet *pkt, int64_t *check_time)
603 int64_t now = qemu_clock_get_ms(QEMU_CLOCK_HOST);
605 if ((now - pkt->creation_ms) > (*check_time)) {
606 trace_colo_old_packet_check_found(pkt->creation_ms);
607 return 0;
608 } else {
609 return 1;
613 void colo_compare_register_notifier(Notifier *notify)
615 notifier_list_add(&colo_compare_notifiers, notify);
618 void colo_compare_unregister_notifier(Notifier *notify)
620 notifier_remove(notify);
623 static int colo_old_packet_check_one_conn(Connection *conn,
624 CompareState *s)
626 if (!g_queue_is_empty(&conn->primary_list)) {
627 if (g_queue_find_custom(&conn->primary_list,
628 &s->compare_timeout,
629 (GCompareFunc)colo_old_packet_check_one))
630 goto out;
633 if (!g_queue_is_empty(&conn->secondary_list)) {
634 if (g_queue_find_custom(&conn->secondary_list,
635 &s->compare_timeout,
636 (GCompareFunc)colo_old_packet_check_one))
637 goto out;
640 return 1;
642 out:
643 /* Do checkpoint will flush old packet */
644 colo_compare_inconsistency_notify(s);
645 return 0;
649 * Look for old packets that the secondary hasn't matched,
650 * if we have some then we have to checkpoint to wake
651 * the secondary up.
653 static void colo_old_packet_check(void *opaque)
655 CompareState *s = opaque;
658 * If we find one old packet, stop finding job and notify
659 * COLO frame do checkpoint.
661 g_queue_find_custom(&s->conn_list, s,
662 (GCompareFunc)colo_old_packet_check_one_conn);
665 static void colo_compare_packet(CompareState *s, Connection *conn,
666 int (*HandlePacket)(Packet *spkt,
667 Packet *ppkt))
669 Packet *pkt = NULL;
670 GList *result = NULL;
672 while (!g_queue_is_empty(&conn->primary_list) &&
673 !g_queue_is_empty(&conn->secondary_list)) {
674 pkt = g_queue_pop_tail(&conn->primary_list);
675 result = g_queue_find_custom(&conn->secondary_list,
676 pkt, (GCompareFunc)HandlePacket);
678 if (result) {
679 colo_release_primary_pkt(s, pkt);
680 packet_destroy(result->data, NULL);
681 g_queue_delete_link(&conn->secondary_list, result);
682 } else {
684 * If one packet arrive late, the secondary_list or
685 * primary_list will be empty, so we can't compare it
686 * until next comparison. If the packets in the list are
687 * timeout, it will trigger a checkpoint request.
689 trace_colo_compare_main("packet different");
690 g_queue_push_tail(&conn->primary_list, pkt);
692 colo_compare_inconsistency_notify(s);
693 break;
699 * Called from the compare thread on the primary
700 * for compare packet with secondary list of the
701 * specified connection when a new packet was
702 * queued to it.
704 static void colo_compare_connection(void *opaque, void *user_data)
706 CompareState *s = user_data;
707 Connection *conn = opaque;
709 switch (conn->ip_proto) {
710 case IPPROTO_TCP:
711 colo_compare_tcp(s, conn);
712 break;
713 case IPPROTO_UDP:
714 colo_compare_packet(s, conn, colo_packet_compare_udp);
715 break;
716 case IPPROTO_ICMP:
717 colo_compare_packet(s, conn, colo_packet_compare_icmp);
718 break;
719 default:
720 colo_compare_packet(s, conn, colo_packet_compare_other);
721 break;
725 static void coroutine_fn _compare_chr_send(void *opaque)
727 SendCo *sendco = opaque;
728 CompareState *s = sendco->s;
729 int ret = 0;
731 while (!g_queue_is_empty(&sendco->send_list)) {
732 SendEntry *entry = g_queue_pop_tail(&sendco->send_list);
733 uint32_t len = htonl(entry->size);
735 ret = qemu_chr_fe_write_all(sendco->chr, (uint8_t *)&len, sizeof(len));
737 if (ret != sizeof(len)) {
738 g_free(entry->buf);
739 g_slice_free(SendEntry, entry);
740 goto err;
743 if (!sendco->notify_remote_frame && s->vnet_hdr) {
745 * We send vnet header len make other module(like filter-redirector)
746 * know how to parse net packet correctly.
748 len = htonl(entry->vnet_hdr_len);
750 ret = qemu_chr_fe_write_all(sendco->chr,
751 (uint8_t *)&len,
752 sizeof(len));
754 if (ret != sizeof(len)) {
755 g_free(entry->buf);
756 g_slice_free(SendEntry, entry);
757 goto err;
761 ret = qemu_chr_fe_write_all(sendco->chr,
762 (uint8_t *)entry->buf,
763 entry->size);
765 if (ret != entry->size) {
766 g_free(entry->buf);
767 g_slice_free(SendEntry, entry);
768 goto err;
771 g_free(entry->buf);
772 g_slice_free(SendEntry, entry);
775 sendco->ret = 0;
776 goto out;
778 err:
779 while (!g_queue_is_empty(&sendco->send_list)) {
780 SendEntry *entry = g_queue_pop_tail(&sendco->send_list);
781 g_free(entry->buf);
782 g_slice_free(SendEntry, entry);
784 sendco->ret = ret < 0 ? ret : -EIO;
785 out:
786 sendco->co = NULL;
787 sendco->done = true;
788 aio_wait_kick();
791 static int compare_chr_send(CompareState *s,
792 uint8_t *buf,
793 uint32_t size,
794 uint32_t vnet_hdr_len,
795 bool notify_remote_frame,
796 bool zero_copy)
798 SendCo *sendco;
799 SendEntry *entry;
801 if (notify_remote_frame) {
802 sendco = &s->notify_sendco;
803 } else {
804 sendco = &s->out_sendco;
807 if (!size) {
808 return -1;
811 entry = g_slice_new(SendEntry);
812 entry->size = size;
813 entry->vnet_hdr_len = vnet_hdr_len;
814 if (zero_copy) {
815 entry->buf = buf;
816 } else {
817 entry->buf = g_malloc(size);
818 memcpy(entry->buf, buf, size);
820 g_queue_push_tail(&sendco->send_list, entry);
822 if (sendco->done) {
823 sendco->co = qemu_coroutine_create(_compare_chr_send, sendco);
824 sendco->done = false;
825 qemu_coroutine_enter(sendco->co);
826 if (sendco->done) {
827 /* report early errors */
828 return sendco->ret;
832 /* assume success */
833 return 0;
836 static int compare_chr_can_read(void *opaque)
838 return COMPARE_READ_LEN_MAX;
842 * Called from the main thread on the primary for packets
843 * arriving over the socket from the primary.
845 static void compare_pri_chr_in(void *opaque, const uint8_t *buf, int size)
847 CompareState *s = COLO_COMPARE(opaque);
848 int ret;
850 ret = net_fill_rstate(&s->pri_rs, buf, size);
851 if (ret == -1) {
852 qemu_chr_fe_set_handlers(&s->chr_pri_in, NULL, NULL, NULL, NULL,
853 NULL, NULL, true);
854 error_report("colo-compare primary_in error");
859 * Called from the main thread on the primary for packets
860 * arriving over the socket from the secondary.
862 static void compare_sec_chr_in(void *opaque, const uint8_t *buf, int size)
864 CompareState *s = COLO_COMPARE(opaque);
865 int ret;
867 ret = net_fill_rstate(&s->sec_rs, buf, size);
868 if (ret == -1) {
869 qemu_chr_fe_set_handlers(&s->chr_sec_in, NULL, NULL, NULL, NULL,
870 NULL, NULL, true);
871 error_report("colo-compare secondary_in error");
875 static void compare_notify_chr(void *opaque, const uint8_t *buf, int size)
877 CompareState *s = COLO_COMPARE(opaque);
878 int ret;
880 ret = net_fill_rstate(&s->notify_rs, buf, size);
881 if (ret == -1) {
882 qemu_chr_fe_set_handlers(&s->chr_notify_dev, NULL, NULL, NULL, NULL,
883 NULL, NULL, true);
884 error_report("colo-compare notify_dev error");
889 * Check old packet regularly so it can watch for any packets
890 * that the secondary hasn't produced equivalents of.
892 static void check_old_packet_regular(void *opaque)
894 CompareState *s = opaque;
896 /* if have old packet we will notify checkpoint */
897 colo_old_packet_check(s);
898 timer_mod(s->packet_check_timer, qemu_clock_get_ms(QEMU_CLOCK_HOST) +
899 s->expired_scan_cycle);
902 /* Public API, Used for COLO frame to notify compare event */
903 void colo_notify_compares_event(void *opaque, int event, Error **errp)
905 CompareState *s;
906 qemu_mutex_lock(&colo_compare_mutex);
908 if (!colo_compare_active) {
909 qemu_mutex_unlock(&colo_compare_mutex);
910 return;
913 qemu_mutex_lock(&event_mtx);
914 QTAILQ_FOREACH(s, &net_compares, next) {
915 s->event = event;
916 qemu_bh_schedule(s->event_bh);
917 event_unhandled_count++;
919 /* Wait all compare threads to finish handling this event */
920 while (event_unhandled_count > 0) {
921 qemu_cond_wait(&event_complete_cond, &event_mtx);
924 qemu_mutex_unlock(&event_mtx);
925 qemu_mutex_unlock(&colo_compare_mutex);
928 static void colo_compare_timer_init(CompareState *s)
930 AioContext *ctx = iothread_get_aio_context(s->iothread);
932 s->packet_check_timer = aio_timer_new(ctx, QEMU_CLOCK_HOST,
933 SCALE_MS, check_old_packet_regular,
935 timer_mod(s->packet_check_timer, qemu_clock_get_ms(QEMU_CLOCK_HOST) +
936 s->expired_scan_cycle);
939 static void colo_compare_timer_del(CompareState *s)
941 if (s->packet_check_timer) {
942 timer_free(s->packet_check_timer);
943 s->packet_check_timer = NULL;
947 static void colo_flush_packets(void *opaque, void *user_data);
949 static void colo_compare_handle_event(void *opaque)
951 CompareState *s = opaque;
953 switch (s->event) {
954 case COLO_EVENT_CHECKPOINT:
955 g_queue_foreach(&s->conn_list, colo_flush_packets, s);
956 break;
957 case COLO_EVENT_FAILOVER:
958 break;
959 default:
960 break;
963 qemu_mutex_lock(&event_mtx);
964 assert(event_unhandled_count > 0);
965 event_unhandled_count--;
966 qemu_cond_broadcast(&event_complete_cond);
967 qemu_mutex_unlock(&event_mtx);
970 static void colo_compare_iothread(CompareState *s)
972 AioContext *ctx = iothread_get_aio_context(s->iothread);
973 object_ref(OBJECT(s->iothread));
974 s->worker_context = iothread_get_g_main_context(s->iothread);
976 qemu_chr_fe_set_handlers(&s->chr_pri_in, compare_chr_can_read,
977 compare_pri_chr_in, NULL, NULL,
978 s, s->worker_context, true);
979 qemu_chr_fe_set_handlers(&s->chr_sec_in, compare_chr_can_read,
980 compare_sec_chr_in, NULL, NULL,
981 s, s->worker_context, true);
982 if (s->notify_dev) {
983 qemu_chr_fe_set_handlers(&s->chr_notify_dev, compare_chr_can_read,
984 compare_notify_chr, NULL, NULL,
985 s, s->worker_context, true);
988 colo_compare_timer_init(s);
989 s->event_bh = aio_bh_new(ctx, colo_compare_handle_event, s);
992 static char *compare_get_pri_indev(Object *obj, Error **errp)
994 CompareState *s = COLO_COMPARE(obj);
996 return g_strdup(s->pri_indev);
999 static void compare_set_pri_indev(Object *obj, const char *value, Error **errp)
1001 CompareState *s = COLO_COMPARE(obj);
1003 g_free(s->pri_indev);
1004 s->pri_indev = g_strdup(value);
1007 static char *compare_get_sec_indev(Object *obj, Error **errp)
1009 CompareState *s = COLO_COMPARE(obj);
1011 return g_strdup(s->sec_indev);
1014 static void compare_set_sec_indev(Object *obj, const char *value, Error **errp)
1016 CompareState *s = COLO_COMPARE(obj);
1018 g_free(s->sec_indev);
1019 s->sec_indev = g_strdup(value);
1022 static char *compare_get_outdev(Object *obj, Error **errp)
1024 CompareState *s = COLO_COMPARE(obj);
1026 return g_strdup(s->outdev);
1029 static void compare_set_outdev(Object *obj, const char *value, Error **errp)
1031 CompareState *s = COLO_COMPARE(obj);
1033 g_free(s->outdev);
1034 s->outdev = g_strdup(value);
1037 static bool compare_get_vnet_hdr(Object *obj, Error **errp)
1039 CompareState *s = COLO_COMPARE(obj);
1041 return s->vnet_hdr;
1044 static void compare_set_vnet_hdr(Object *obj,
1045 bool value,
1046 Error **errp)
1048 CompareState *s = COLO_COMPARE(obj);
1050 s->vnet_hdr = value;
1053 static char *compare_get_notify_dev(Object *obj, Error **errp)
1055 CompareState *s = COLO_COMPARE(obj);
1057 return g_strdup(s->notify_dev);
1060 static void compare_set_notify_dev(Object *obj, const char *value, Error **errp)
1062 CompareState *s = COLO_COMPARE(obj);
1064 g_free(s->notify_dev);
1065 s->notify_dev = g_strdup(value);
1068 static void compare_get_timeout(Object *obj, Visitor *v,
1069 const char *name, void *opaque,
1070 Error **errp)
1072 CompareState *s = COLO_COMPARE(obj);
1073 uint64_t value = s->compare_timeout;
1075 visit_type_uint64(v, name, &value, errp);
1078 static void compare_set_timeout(Object *obj, Visitor *v,
1079 const char *name, void *opaque,
1080 Error **errp)
1082 CompareState *s = COLO_COMPARE(obj);
1083 uint32_t value;
1085 if (!visit_type_uint32(v, name, &value, errp)) {
1086 return;
1088 if (!value) {
1089 error_setg(errp, "Property '%s.%s' requires a positive value",
1090 object_get_typename(obj), name);
1091 return;
1093 s->compare_timeout = value;
1096 static void compare_get_expired_scan_cycle(Object *obj, Visitor *v,
1097 const char *name, void *opaque,
1098 Error **errp)
1100 CompareState *s = COLO_COMPARE(obj);
1101 uint32_t value = s->expired_scan_cycle;
1103 visit_type_uint32(v, name, &value, errp);
1106 static void compare_set_expired_scan_cycle(Object *obj, Visitor *v,
1107 const char *name, void *opaque,
1108 Error **errp)
1110 CompareState *s = COLO_COMPARE(obj);
1111 uint32_t value;
1113 if (!visit_type_uint32(v, name, &value, errp)) {
1114 return;
1116 if (!value) {
1117 error_setg(errp, "Property '%s.%s' requires a positive value",
1118 object_get_typename(obj), name);
1119 return;
1121 s->expired_scan_cycle = value;
1124 static void get_max_queue_size(Object *obj, Visitor *v,
1125 const char *name, void *opaque,
1126 Error **errp)
1128 uint32_t value = max_queue_size;
1130 visit_type_uint32(v, name, &value, errp);
1133 static void set_max_queue_size(Object *obj, Visitor *v,
1134 const char *name, void *opaque,
1135 Error **errp)
1137 uint64_t value;
1139 if (!visit_type_uint64(v, name, &value, errp)) {
1140 return;
1142 if (!value) {
1143 error_setg(errp, "Property '%s.%s' requires a positive value",
1144 object_get_typename(obj), name);
1145 return;
1147 max_queue_size = value;
1150 static void compare_pri_rs_finalize(SocketReadState *pri_rs)
1152 CompareState *s = container_of(pri_rs, CompareState, pri_rs);
1153 Connection *conn = NULL;
1155 if (packet_enqueue(s, PRIMARY_IN, &conn)) {
1156 trace_colo_compare_main("primary: unsupported packet in");
1157 compare_chr_send(s,
1158 pri_rs->buf,
1159 pri_rs->packet_len,
1160 pri_rs->vnet_hdr_len,
1161 false,
1162 false);
1163 } else {
1164 /* compare packet in the specified connection */
1165 colo_compare_connection(conn, s);
1169 static void compare_sec_rs_finalize(SocketReadState *sec_rs)
1171 CompareState *s = container_of(sec_rs, CompareState, sec_rs);
1172 Connection *conn = NULL;
1174 if (packet_enqueue(s, SECONDARY_IN, &conn)) {
1175 trace_colo_compare_main("secondary: unsupported packet in");
1176 } else {
1177 /* compare packet in the specified connection */
1178 colo_compare_connection(conn, s);
1182 static void compare_notify_rs_finalize(SocketReadState *notify_rs)
1184 CompareState *s = container_of(notify_rs, CompareState, notify_rs);
1186 const char msg[] = "COLO_COMPARE_GET_XEN_INIT";
1187 int ret;
1189 if (packet_matches_str("COLO_USERSPACE_PROXY_INIT",
1190 notify_rs->buf,
1191 notify_rs->packet_len)) {
1192 ret = compare_chr_send(s, (uint8_t *)msg, strlen(msg), 0, true, false);
1193 if (ret < 0) {
1194 error_report("Notify Xen COLO-frame INIT failed");
1196 } else if (packet_matches_str("COLO_CHECKPOINT",
1197 notify_rs->buf,
1198 notify_rs->packet_len)) {
1199 /* colo-compare do checkpoint, flush pri packet and remove sec packet */
1200 g_queue_foreach(&s->conn_list, colo_flush_packets, s);
1201 } else {
1202 error_report("COLO compare got unsupported instruction");
1207 * Return 0 is success.
1208 * Return 1 is failed.
1210 static int find_and_check_chardev(Chardev **chr,
1211 char *chr_name,
1212 Error **errp)
1214 *chr = qemu_chr_find(chr_name);
1215 if (*chr == NULL) {
1216 error_setg(errp, "Device '%s' not found",
1217 chr_name);
1218 return 1;
1221 if (!qemu_chr_has_feature(*chr, QEMU_CHAR_FEATURE_RECONNECTABLE)) {
1222 error_setg(errp, "chardev \"%s\" is not reconnectable",
1223 chr_name);
1224 return 1;
1227 if (!qemu_chr_has_feature(*chr, QEMU_CHAR_FEATURE_GCONTEXT)) {
1228 error_setg(errp, "chardev \"%s\" cannot switch context",
1229 chr_name);
1230 return 1;
1233 return 0;
1237 * Called from the main thread on the primary
1238 * to setup colo-compare.
1240 static void colo_compare_complete(UserCreatable *uc, Error **errp)
1242 CompareState *s = COLO_COMPARE(uc);
1243 Chardev *chr;
1245 if (!s->pri_indev || !s->sec_indev || !s->outdev || !s->iothread) {
1246 error_setg(errp, "colo compare needs 'primary_in' ,"
1247 "'secondary_in','outdev','iothread' property set");
1248 return;
1249 } else if (!strcmp(s->pri_indev, s->outdev) ||
1250 !strcmp(s->sec_indev, s->outdev) ||
1251 !strcmp(s->pri_indev, s->sec_indev)) {
1252 error_setg(errp, "'indev' and 'outdev' could not be same "
1253 "for compare module");
1254 return;
1257 if (!s->compare_timeout) {
1258 /* Set default value to 3000 MS */
1259 s->compare_timeout = DEFAULT_TIME_OUT_MS;
1262 if (!s->expired_scan_cycle) {
1263 /* Set default value to 1000 MS */
1264 s->expired_scan_cycle = REGULAR_PACKET_CHECK_MS;
1267 if (!max_queue_size) {
1268 /* Set default queue size to 1024 */
1269 max_queue_size = MAX_QUEUE_SIZE;
1272 if (find_and_check_chardev(&chr, s->pri_indev, errp) ||
1273 !qemu_chr_fe_init(&s->chr_pri_in, chr, errp)) {
1274 return;
1277 if (find_and_check_chardev(&chr, s->sec_indev, errp) ||
1278 !qemu_chr_fe_init(&s->chr_sec_in, chr, errp)) {
1279 return;
1282 if (find_and_check_chardev(&chr, s->outdev, errp) ||
1283 !qemu_chr_fe_init(&s->chr_out, chr, errp)) {
1284 return;
1287 net_socket_rs_init(&s->pri_rs, compare_pri_rs_finalize, s->vnet_hdr);
1288 net_socket_rs_init(&s->sec_rs, compare_sec_rs_finalize, s->vnet_hdr);
1290 /* Try to enable remote notify chardev, currently just for Xen COLO */
1291 if (s->notify_dev) {
1292 if (find_and_check_chardev(&chr, s->notify_dev, errp) ||
1293 !qemu_chr_fe_init(&s->chr_notify_dev, chr, errp)) {
1294 return;
1297 net_socket_rs_init(&s->notify_rs, compare_notify_rs_finalize,
1298 s->vnet_hdr);
1301 s->out_sendco.s = s;
1302 s->out_sendco.chr = &s->chr_out;
1303 s->out_sendco.notify_remote_frame = false;
1304 s->out_sendco.done = true;
1305 g_queue_init(&s->out_sendco.send_list);
1307 if (s->notify_dev) {
1308 s->notify_sendco.s = s;
1309 s->notify_sendco.chr = &s->chr_notify_dev;
1310 s->notify_sendco.notify_remote_frame = true;
1311 s->notify_sendco.done = true;
1312 g_queue_init(&s->notify_sendco.send_list);
1315 g_queue_init(&s->conn_list);
1317 s->connection_track_table = g_hash_table_new_full(connection_key_hash,
1318 connection_key_equal,
1319 g_free,
1320 NULL);
1322 colo_compare_iothread(s);
1324 qemu_mutex_lock(&colo_compare_mutex);
1325 if (!colo_compare_active) {
1326 qemu_mutex_init(&event_mtx);
1327 qemu_cond_init(&event_complete_cond);
1328 colo_compare_active = true;
1330 QTAILQ_INSERT_TAIL(&net_compares, s, next);
1331 qemu_mutex_unlock(&colo_compare_mutex);
1333 return;
1336 static void colo_flush_packets(void *opaque, void *user_data)
1338 CompareState *s = user_data;
1339 Connection *conn = opaque;
1340 Packet *pkt = NULL;
1342 while (!g_queue_is_empty(&conn->primary_list)) {
1343 pkt = g_queue_pop_tail(&conn->primary_list);
1344 compare_chr_send(s,
1345 pkt->data,
1346 pkt->size,
1347 pkt->vnet_hdr_len,
1348 false,
1349 true);
1350 packet_destroy_partial(pkt, NULL);
1352 while (!g_queue_is_empty(&conn->secondary_list)) {
1353 pkt = g_queue_pop_tail(&conn->secondary_list);
1354 packet_destroy(pkt, NULL);
1358 static void colo_compare_class_init(ObjectClass *oc, void *data)
1360 UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
1362 ucc->complete = colo_compare_complete;
1365 static void colo_compare_init(Object *obj)
1367 CompareState *s = COLO_COMPARE(obj);
1369 object_property_add_str(obj, "primary_in",
1370 compare_get_pri_indev, compare_set_pri_indev);
1371 object_property_add_str(obj, "secondary_in",
1372 compare_get_sec_indev, compare_set_sec_indev);
1373 object_property_add_str(obj, "outdev",
1374 compare_get_outdev, compare_set_outdev);
1375 object_property_add_link(obj, "iothread", TYPE_IOTHREAD,
1376 (Object **)&s->iothread,
1377 object_property_allow_set_link,
1378 OBJ_PROP_LINK_STRONG);
1379 /* This parameter just for Xen COLO */
1380 object_property_add_str(obj, "notify_dev",
1381 compare_get_notify_dev, compare_set_notify_dev);
1383 object_property_add(obj, "compare_timeout", "uint64",
1384 compare_get_timeout,
1385 compare_set_timeout, NULL, NULL);
1387 object_property_add(obj, "expired_scan_cycle", "uint32",
1388 compare_get_expired_scan_cycle,
1389 compare_set_expired_scan_cycle, NULL, NULL);
1391 object_property_add(obj, "max_queue_size", "uint32",
1392 get_max_queue_size,
1393 set_max_queue_size, NULL, NULL);
1395 s->vnet_hdr = false;
1396 object_property_add_bool(obj, "vnet_hdr_support", compare_get_vnet_hdr,
1397 compare_set_vnet_hdr);
1400 void colo_compare_cleanup(void)
1402 CompareState *tmp = NULL;
1403 CompareState *n = NULL;
1405 QTAILQ_FOREACH_SAFE(tmp, &net_compares, next, n) {
1406 object_unparent(OBJECT(tmp));
1410 static void colo_compare_finalize(Object *obj)
1412 CompareState *s = COLO_COMPARE(obj);
1413 CompareState *tmp = NULL;
1415 qemu_mutex_lock(&colo_compare_mutex);
1416 QTAILQ_FOREACH(tmp, &net_compares, next) {
1417 if (tmp == s) {
1418 QTAILQ_REMOVE(&net_compares, s, next);
1419 break;
1422 if (QTAILQ_EMPTY(&net_compares)) {
1423 colo_compare_active = false;
1424 qemu_mutex_destroy(&event_mtx);
1425 qemu_cond_destroy(&event_complete_cond);
1427 qemu_mutex_unlock(&colo_compare_mutex);
1429 qemu_chr_fe_deinit(&s->chr_pri_in, false);
1430 qemu_chr_fe_deinit(&s->chr_sec_in, false);
1431 qemu_chr_fe_deinit(&s->chr_out, false);
1432 if (s->notify_dev) {
1433 qemu_chr_fe_deinit(&s->chr_notify_dev, false);
1436 colo_compare_timer_del(s);
1438 qemu_bh_delete(s->event_bh);
1440 AioContext *ctx = iothread_get_aio_context(s->iothread);
1441 AIO_WAIT_WHILE(ctx, !s->out_sendco.done);
1442 if (s->notify_dev) {
1443 AIO_WAIT_WHILE(ctx, !s->notify_sendco.done);
1446 /* Release all unhandled packets after compare thead exited */
1447 g_queue_foreach(&s->conn_list, colo_flush_packets, s);
1448 AIO_WAIT_WHILE(NULL, !s->out_sendco.done);
1450 g_queue_clear(&s->conn_list);
1451 g_queue_clear(&s->out_sendco.send_list);
1452 if (s->notify_dev) {
1453 g_queue_clear(&s->notify_sendco.send_list);
1456 if (s->connection_track_table) {
1457 g_hash_table_destroy(s->connection_track_table);
1460 object_unref(OBJECT(s->iothread));
1462 g_free(s->pri_indev);
1463 g_free(s->sec_indev);
1464 g_free(s->outdev);
1465 g_free(s->notify_dev);
1468 static void __attribute__((__constructor__)) colo_compare_init_globals(void)
1470 colo_compare_active = false;
1471 qemu_mutex_init(&colo_compare_mutex);
1474 static const TypeInfo colo_compare_info = {
1475 .name = TYPE_COLO_COMPARE,
1476 .parent = TYPE_OBJECT,
1477 .instance_size = sizeof(CompareState),
1478 .instance_init = colo_compare_init,
1479 .instance_finalize = colo_compare_finalize,
1480 .class_size = sizeof(CompareClass),
1481 .class_init = colo_compare_class_init,
1482 .interfaces = (InterfaceInfo[]) {
1483 { TYPE_USER_CREATABLE },
1488 static void register_types(void)
1490 type_register_static(&colo_compare_info);
1493 type_init(register_types);