timer: a9gtimer: remove loop to auto-increment comparator
[qemu/ar7.git] / net / colo-compare.c
blob109990fd8ec85083493c2a8c0524ab97c8dec0fd
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 "qemu-common.h"
19 #include "qapi/qmp/qerror.h"
20 #include "qapi/error.h"
21 #include "net/net.h"
22 #include "net/eth.h"
23 #include "qom/object_interfaces.h"
24 #include "qemu/iov.h"
25 #include "qom/object.h"
26 #include "qemu/typedefs.h"
27 #include "net/queue.h"
28 #include "sysemu/char.h"
29 #include "qemu/sockets.h"
30 #include "qapi-visit.h"
31 #include "net/colo.h"
33 #define TYPE_COLO_COMPARE "colo-compare"
34 #define COLO_COMPARE(obj) \
35 OBJECT_CHECK(CompareState, (obj), TYPE_COLO_COMPARE)
37 #define COMPARE_READ_LEN_MAX NET_BUFSIZE
38 #define MAX_QUEUE_SIZE 1024
40 /* TODO: Should be configurable */
41 #define REGULAR_PACKET_CHECK_MS 3000
44 + CompareState ++
45 | |
46 +---------------+ +---------------+ +---------------+
47 |conn list +--->conn +--------->conn |
48 +---------------+ +---------------+ +---------------+
49 | | | | | |
50 +---------------+ +---v----+ +---v----+ +---v----+ +---v----+
51 |primary | |secondary |primary | |secondary
52 |packet | |packet + |packet | |packet +
53 +--------+ +--------+ +--------+ +--------+
54 | | | |
55 +---v----+ +---v----+ +---v----+ +---v----+
56 |primary | |secondary |primary | |secondary
57 |packet | |packet + |packet | |packet +
58 +--------+ +--------+ +--------+ +--------+
59 | | | |
60 +---v----+ +---v----+ +---v----+ +---v----+
61 |primary | |secondary |primary | |secondary
62 |packet | |packet + |packet | |packet +
63 +--------+ +--------+ +--------+ +--------+
65 typedef struct CompareState {
66 Object parent;
68 char *pri_indev;
69 char *sec_indev;
70 char *outdev;
71 CharBackend chr_pri_in;
72 CharBackend chr_sec_in;
73 CharBackend chr_out;
74 SocketReadState pri_rs;
75 SocketReadState sec_rs;
77 /* connection list: the connections belonged to this NIC could be found
78 * in this list.
79 * element type: Connection
81 GQueue conn_list;
82 /* hashtable to save connection */
83 GHashTable *connection_track_table;
84 /* compare thread, a thread for each NIC */
85 QemuThread thread;
86 /* Timer used on the primary to find packets that are never matched */
87 QEMUTimer *timer;
88 QemuMutex timer_check_lock;
89 } CompareState;
91 typedef struct CompareClass {
92 ObjectClass parent_class;
93 } CompareClass;
95 typedef struct CompareChardevProps {
96 bool is_socket;
97 } CompareChardevProps;
99 enum {
100 PRIMARY_IN = 0,
101 SECONDARY_IN,
104 static int compare_chr_send(CharBackend *out,
105 const uint8_t *buf,
106 uint32_t size);
109 * Return 0 on success, if return -1 means the pkt
110 * is unsupported(arp and ipv6) and will be sent later
112 static int packet_enqueue(CompareState *s, int mode)
114 ConnectionKey key;
115 Packet *pkt = NULL;
116 Connection *conn;
118 if (mode == PRIMARY_IN) {
119 pkt = packet_new(s->pri_rs.buf, s->pri_rs.packet_len);
120 } else {
121 pkt = packet_new(s->sec_rs.buf, s->sec_rs.packet_len);
124 if (parse_packet_early(pkt)) {
125 packet_destroy(pkt, NULL);
126 pkt = NULL;
127 return -1;
129 fill_connection_key(pkt, &key);
131 conn = connection_get(s->connection_track_table,
132 &key,
133 &s->conn_list);
135 if (!conn->processing) {
136 g_queue_push_tail(&s->conn_list, conn);
137 conn->processing = true;
140 if (mode == PRIMARY_IN) {
141 if (g_queue_get_length(&conn->primary_list) <=
142 MAX_QUEUE_SIZE) {
143 g_queue_push_tail(&conn->primary_list, pkt);
144 } else {
145 error_report("colo compare primary queue size too big,"
146 "drop packet");
148 } else {
149 if (g_queue_get_length(&conn->secondary_list) <=
150 MAX_QUEUE_SIZE) {
151 g_queue_push_tail(&conn->secondary_list, pkt);
152 } else {
153 error_report("colo compare secondary queue size too big,"
154 "drop packet");
158 return 0;
162 * The IP packets sent by primary and secondary
163 * will be compared in here
164 * TODO support ip fragment, Out-Of-Order
165 * return: 0 means packet same
166 * > 0 || < 0 means packet different
168 static int colo_packet_compare(Packet *ppkt, Packet *spkt)
170 trace_colo_compare_ip_info(ppkt->size, inet_ntoa(ppkt->ip->ip_src),
171 inet_ntoa(ppkt->ip->ip_dst), spkt->size,
172 inet_ntoa(spkt->ip->ip_src),
173 inet_ntoa(spkt->ip->ip_dst));
175 if (ppkt->size == spkt->size) {
176 return memcmp(ppkt->data, spkt->data, spkt->size);
177 } else {
178 return -1;
183 * Called from the compare thread on the primary
184 * for compare tcp packet
185 * compare_tcp copied from Dr. David Alan Gilbert's branch
187 static int colo_packet_compare_tcp(Packet *spkt, Packet *ppkt)
189 struct tcphdr *ptcp, *stcp;
190 int res;
191 char *sdebug, *ddebug;
193 trace_colo_compare_main("compare tcp");
194 if (ppkt->size != spkt->size) {
195 if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) {
196 trace_colo_compare_main("pkt size not same");
198 return -1;
201 ptcp = (struct tcphdr *)ppkt->transport_header;
202 stcp = (struct tcphdr *)spkt->transport_header;
205 * The 'identification' field in the IP header is *very* random
206 * it almost never matches. Fudge this by ignoring differences in
207 * unfragmented packets; they'll normally sort themselves out if different
208 * anyway, and it should recover at the TCP level.
209 * An alternative would be to get both the primary and secondary to rewrite
210 * somehow; but that would need some sync traffic to sync the state
212 if (ntohs(ppkt->ip->ip_off) & IP_DF) {
213 spkt->ip->ip_id = ppkt->ip->ip_id;
214 /* and the sum will be different if the IDs were different */
215 spkt->ip->ip_sum = ppkt->ip->ip_sum;
218 res = memcmp(ppkt->data + ETH_HLEN, spkt->data + ETH_HLEN,
219 (spkt->size - ETH_HLEN));
221 if (res != 0 && trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) {
222 sdebug = strdup(inet_ntoa(ppkt->ip->ip_src));
223 ddebug = strdup(inet_ntoa(ppkt->ip->ip_dst));
224 fprintf(stderr, "%s: src/dst: %s/%s p: seq/ack=%u/%u"
225 " s: seq/ack=%u/%u res=%d flags=%x/%x\n",
226 __func__, sdebug, ddebug,
227 (unsigned int)ntohl(ptcp->th_seq),
228 (unsigned int)ntohl(ptcp->th_ack),
229 (unsigned int)ntohl(stcp->th_seq),
230 (unsigned int)ntohl(stcp->th_ack),
231 res, ptcp->th_flags, stcp->th_flags);
233 fprintf(stderr, "Primary len = %d\n", ppkt->size);
234 qemu_hexdump((char *)ppkt->data, stderr, "colo-compare", ppkt->size);
235 fprintf(stderr, "Secondary len = %d\n", spkt->size);
236 qemu_hexdump((char *)spkt->data, stderr, "colo-compare", spkt->size);
238 g_free(sdebug);
239 g_free(ddebug);
242 return res;
246 * Called from the compare thread on the primary
247 * for compare udp packet
249 static int colo_packet_compare_udp(Packet *spkt, Packet *ppkt)
251 int ret;
253 trace_colo_compare_main("compare udp");
254 ret = colo_packet_compare(ppkt, spkt);
256 if (ret) {
257 trace_colo_compare_udp_miscompare("primary pkt size", ppkt->size);
258 qemu_hexdump((char *)ppkt->data, stderr, "colo-compare", ppkt->size);
259 trace_colo_compare_udp_miscompare("Secondary pkt size", spkt->size);
260 qemu_hexdump((char *)spkt->data, stderr, "colo-compare", spkt->size);
263 return ret;
267 * Called from the compare thread on the primary
268 * for compare icmp packet
270 static int colo_packet_compare_icmp(Packet *spkt, Packet *ppkt)
272 int network_length;
274 trace_colo_compare_main("compare icmp");
275 network_length = ppkt->ip->ip_hl * 4;
276 if (ppkt->size != spkt->size ||
277 ppkt->size < network_length + ETH_HLEN) {
278 return -1;
281 if (colo_packet_compare(ppkt, spkt)) {
282 trace_colo_compare_icmp_miscompare("primary pkt size",
283 ppkt->size);
284 qemu_hexdump((char *)ppkt->data, stderr, "colo-compare",
285 ppkt->size);
286 trace_colo_compare_icmp_miscompare("Secondary pkt size",
287 spkt->size);
288 qemu_hexdump((char *)spkt->data, stderr, "colo-compare",
289 spkt->size);
290 return -1;
291 } else {
292 return 0;
297 * Called from the compare thread on the primary
298 * for compare other packet
300 static int colo_packet_compare_other(Packet *spkt, Packet *ppkt)
302 trace_colo_compare_main("compare other");
303 trace_colo_compare_ip_info(ppkt->size, inet_ntoa(ppkt->ip->ip_src),
304 inet_ntoa(ppkt->ip->ip_dst), spkt->size,
305 inet_ntoa(spkt->ip->ip_src),
306 inet_ntoa(spkt->ip->ip_dst));
307 return colo_packet_compare(ppkt, spkt);
310 static int colo_old_packet_check_one(Packet *pkt, int64_t *check_time)
312 int64_t now = qemu_clock_get_ms(QEMU_CLOCK_HOST);
314 if ((now - pkt->creation_ms) > (*check_time)) {
315 trace_colo_old_packet_check_found(pkt->creation_ms);
316 return 0;
317 } else {
318 return 1;
322 static void colo_old_packet_check_one_conn(void *opaque,
323 void *user_data)
325 Connection *conn = opaque;
326 GList *result = NULL;
327 int64_t check_time = REGULAR_PACKET_CHECK_MS;
329 result = g_queue_find_custom(&conn->primary_list,
330 &check_time,
331 (GCompareFunc)colo_old_packet_check_one);
333 if (result) {
334 /* do checkpoint will flush old packet */
335 /* TODO: colo_notify_checkpoint();*/
340 * Look for old packets that the secondary hasn't matched,
341 * if we have some then we have to checkpoint to wake
342 * the secondary up.
344 static void colo_old_packet_check(void *opaque)
346 CompareState *s = opaque;
348 g_queue_foreach(&s->conn_list, colo_old_packet_check_one_conn, NULL);
352 * Called from the compare thread on the primary
353 * for compare connection
355 static void colo_compare_connection(void *opaque, void *user_data)
357 CompareState *s = user_data;
358 Connection *conn = opaque;
359 Packet *pkt = NULL;
360 GList *result = NULL;
361 int ret;
363 while (!g_queue_is_empty(&conn->primary_list) &&
364 !g_queue_is_empty(&conn->secondary_list)) {
365 qemu_mutex_lock(&s->timer_check_lock);
366 pkt = g_queue_pop_tail(&conn->primary_list);
367 qemu_mutex_unlock(&s->timer_check_lock);
368 switch (conn->ip_proto) {
369 case IPPROTO_TCP:
370 result = g_queue_find_custom(&conn->secondary_list,
371 pkt, (GCompareFunc)colo_packet_compare_tcp);
372 break;
373 case IPPROTO_UDP:
374 result = g_queue_find_custom(&conn->secondary_list,
375 pkt, (GCompareFunc)colo_packet_compare_udp);
376 break;
377 case IPPROTO_ICMP:
378 result = g_queue_find_custom(&conn->secondary_list,
379 pkt, (GCompareFunc)colo_packet_compare_icmp);
380 break;
381 default:
382 result = g_queue_find_custom(&conn->secondary_list,
383 pkt, (GCompareFunc)colo_packet_compare_other);
384 break;
387 if (result) {
388 ret = compare_chr_send(&s->chr_out, pkt->data, pkt->size);
389 if (ret < 0) {
390 error_report("colo_send_primary_packet failed");
392 trace_colo_compare_main("packet same and release packet");
393 g_queue_remove(&conn->secondary_list, result->data);
394 packet_destroy(pkt, NULL);
395 } else {
397 * If one packet arrive late, the secondary_list or
398 * primary_list will be empty, so we can't compare it
399 * until next comparison.
401 trace_colo_compare_main("packet different");
402 qemu_mutex_lock(&s->timer_check_lock);
403 g_queue_push_tail(&conn->primary_list, pkt);
404 qemu_mutex_unlock(&s->timer_check_lock);
405 /* TODO: colo_notify_checkpoint();*/
406 break;
411 static int compare_chr_send(CharBackend *out,
412 const uint8_t *buf,
413 uint32_t size)
415 int ret = 0;
416 uint32_t len = htonl(size);
418 if (!size) {
419 return 0;
422 ret = qemu_chr_fe_write_all(out, (uint8_t *)&len, sizeof(len));
423 if (ret != sizeof(len)) {
424 goto err;
427 ret = qemu_chr_fe_write_all(out, (uint8_t *)buf, size);
428 if (ret != size) {
429 goto err;
432 return 0;
434 err:
435 return ret < 0 ? ret : -EIO;
438 static int compare_chr_can_read(void *opaque)
440 return COMPARE_READ_LEN_MAX;
444 * Called from the main thread on the primary for packets
445 * arriving over the socket from the primary.
447 static void compare_pri_chr_in(void *opaque, const uint8_t *buf, int size)
449 CompareState *s = COLO_COMPARE(opaque);
450 int ret;
452 ret = net_fill_rstate(&s->pri_rs, buf, size);
453 if (ret == -1) {
454 qemu_chr_fe_set_handlers(&s->chr_pri_in, NULL, NULL, NULL,
455 NULL, NULL, true);
456 error_report("colo-compare primary_in error");
461 * Called from the main thread on the primary for packets
462 * arriving over the socket from the secondary.
464 static void compare_sec_chr_in(void *opaque, const uint8_t *buf, int size)
466 CompareState *s = COLO_COMPARE(opaque);
467 int ret;
469 ret = net_fill_rstate(&s->sec_rs, buf, size);
470 if (ret == -1) {
471 qemu_chr_fe_set_handlers(&s->chr_sec_in, NULL, NULL, NULL,
472 NULL, NULL, true);
473 error_report("colo-compare secondary_in error");
477 static void *colo_compare_thread(void *opaque)
479 GMainContext *worker_context;
480 GMainLoop *compare_loop;
481 CompareState *s = opaque;
483 worker_context = g_main_context_new();
485 qemu_chr_fe_set_handlers(&s->chr_pri_in, compare_chr_can_read,
486 compare_pri_chr_in, NULL, s, worker_context, true);
487 qemu_chr_fe_set_handlers(&s->chr_sec_in, compare_chr_can_read,
488 compare_sec_chr_in, NULL, s, worker_context, true);
490 compare_loop = g_main_loop_new(worker_context, FALSE);
492 g_main_loop_run(compare_loop);
494 g_main_loop_unref(compare_loop);
495 g_main_context_unref(worker_context);
496 return NULL;
499 static char *compare_get_pri_indev(Object *obj, Error **errp)
501 CompareState *s = COLO_COMPARE(obj);
503 return g_strdup(s->pri_indev);
506 static void compare_set_pri_indev(Object *obj, const char *value, Error **errp)
508 CompareState *s = COLO_COMPARE(obj);
510 g_free(s->pri_indev);
511 s->pri_indev = g_strdup(value);
514 static char *compare_get_sec_indev(Object *obj, Error **errp)
516 CompareState *s = COLO_COMPARE(obj);
518 return g_strdup(s->sec_indev);
521 static void compare_set_sec_indev(Object *obj, const char *value, Error **errp)
523 CompareState *s = COLO_COMPARE(obj);
525 g_free(s->sec_indev);
526 s->sec_indev = g_strdup(value);
529 static char *compare_get_outdev(Object *obj, Error **errp)
531 CompareState *s = COLO_COMPARE(obj);
533 return g_strdup(s->outdev);
536 static void compare_set_outdev(Object *obj, const char *value, Error **errp)
538 CompareState *s = COLO_COMPARE(obj);
540 g_free(s->outdev);
541 s->outdev = g_strdup(value);
544 static void compare_pri_rs_finalize(SocketReadState *pri_rs)
546 CompareState *s = container_of(pri_rs, CompareState, pri_rs);
548 if (packet_enqueue(s, PRIMARY_IN)) {
549 trace_colo_compare_main("primary: unsupported packet in");
550 compare_chr_send(&s->chr_out, pri_rs->buf, pri_rs->packet_len);
551 } else {
552 /* compare connection */
553 g_queue_foreach(&s->conn_list, colo_compare_connection, s);
557 static void compare_sec_rs_finalize(SocketReadState *sec_rs)
559 CompareState *s = container_of(sec_rs, CompareState, sec_rs);
561 if (packet_enqueue(s, SECONDARY_IN)) {
562 trace_colo_compare_main("secondary: unsupported packet in");
563 } else {
564 /* compare connection */
565 g_queue_foreach(&s->conn_list, colo_compare_connection, s);
571 * Return 0 is success.
572 * Return 1 is failed.
574 static int find_and_check_chardev(CharDriverState **chr,
575 char *chr_name,
576 Error **errp)
578 CompareChardevProps props;
580 *chr = qemu_chr_find(chr_name);
581 if (*chr == NULL) {
582 error_setg(errp, "Device '%s' not found",
583 chr_name);
584 return 1;
587 memset(&props, 0, sizeof(props));
589 if (!qemu_chr_has_feature(*chr, QEMU_CHAR_FEATURE_RECONNECTABLE)) {
590 error_setg(errp, "chardev \"%s\" is not reconnectable",
591 chr_name);
592 return 1;
595 return 0;
599 * Check old packet regularly so it can watch for any packets
600 * that the secondary hasn't produced equivalents of.
602 static void check_old_packet_regular(void *opaque)
604 CompareState *s = opaque;
606 timer_mod(s->timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
607 REGULAR_PACKET_CHECK_MS);
608 /* if have old packet we will notify checkpoint */
610 * TODO: Make timer handler run in compare thread
611 * like qemu_chr_add_handlers_full.
613 qemu_mutex_lock(&s->timer_check_lock);
614 colo_old_packet_check(s);
615 qemu_mutex_unlock(&s->timer_check_lock);
619 * Called from the main thread on the primary
620 * to setup colo-compare.
622 static void colo_compare_complete(UserCreatable *uc, Error **errp)
624 CompareState *s = COLO_COMPARE(uc);
625 CharDriverState *chr;
626 char thread_name[64];
627 static int compare_id;
629 if (!s->pri_indev || !s->sec_indev || !s->outdev) {
630 error_setg(errp, "colo compare needs 'primary_in' ,"
631 "'secondary_in','outdev' property set");
632 return;
633 } else if (!strcmp(s->pri_indev, s->outdev) ||
634 !strcmp(s->sec_indev, s->outdev) ||
635 !strcmp(s->pri_indev, s->sec_indev)) {
636 error_setg(errp, "'indev' and 'outdev' could not be same "
637 "for compare module");
638 return;
641 if (find_and_check_chardev(&chr, s->pri_indev, errp) ||
642 !qemu_chr_fe_init(&s->chr_pri_in, chr, errp)) {
643 return;
646 if (find_and_check_chardev(&chr, s->sec_indev, errp) ||
647 !qemu_chr_fe_init(&s->chr_sec_in, chr, errp)) {
648 return;
651 if (find_and_check_chardev(&chr, s->outdev, errp) ||
652 !qemu_chr_fe_init(&s->chr_out, chr, errp)) {
653 return;
656 net_socket_rs_init(&s->pri_rs, compare_pri_rs_finalize);
657 net_socket_rs_init(&s->sec_rs, compare_sec_rs_finalize);
659 g_queue_init(&s->conn_list);
660 qemu_mutex_init(&s->timer_check_lock);
662 s->connection_track_table = g_hash_table_new_full(connection_key_hash,
663 connection_key_equal,
664 g_free,
665 connection_destroy);
667 sprintf(thread_name, "colo-compare %d", compare_id);
668 qemu_thread_create(&s->thread, thread_name,
669 colo_compare_thread, s,
670 QEMU_THREAD_JOINABLE);
671 compare_id++;
673 /* A regular timer to kick any packets that the secondary doesn't match */
674 s->timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, /* Only when guest runs */
675 check_old_packet_regular, s);
676 timer_mod(s->timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) +
677 REGULAR_PACKET_CHECK_MS);
679 return;
682 static void colo_compare_class_init(ObjectClass *oc, void *data)
684 UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc);
686 ucc->complete = colo_compare_complete;
689 static void colo_compare_init(Object *obj)
691 object_property_add_str(obj, "primary_in",
692 compare_get_pri_indev, compare_set_pri_indev,
693 NULL);
694 object_property_add_str(obj, "secondary_in",
695 compare_get_sec_indev, compare_set_sec_indev,
696 NULL);
697 object_property_add_str(obj, "outdev",
698 compare_get_outdev, compare_set_outdev,
699 NULL);
702 static void colo_compare_finalize(Object *obj)
704 CompareState *s = COLO_COMPARE(obj);
706 qemu_chr_fe_deinit(&s->chr_pri_in);
707 qemu_chr_fe_deinit(&s->chr_sec_in);
708 qemu_chr_fe_deinit(&s->chr_out);
710 g_queue_free(&s->conn_list);
712 if (qemu_thread_is_self(&s->thread)) {
713 /* compare connection */
714 g_queue_foreach(&s->conn_list, colo_compare_connection, s);
715 qemu_thread_join(&s->thread);
718 if (s->timer) {
719 timer_del(s->timer);
722 qemu_mutex_destroy(&s->timer_check_lock);
724 g_free(s->pri_indev);
725 g_free(s->sec_indev);
726 g_free(s->outdev);
729 static const TypeInfo colo_compare_info = {
730 .name = TYPE_COLO_COMPARE,
731 .parent = TYPE_OBJECT,
732 .instance_size = sizeof(CompareState),
733 .instance_init = colo_compare_init,
734 .instance_finalize = colo_compare_finalize,
735 .class_size = sizeof(CompareClass),
736 .class_init = colo_compare_class_init,
737 .interfaces = (InterfaceInfo[]) {
738 { TYPE_USER_CREATABLE },
743 static void register_types(void)
745 type_register_static(&colo_compare_info);
748 type_init(register_types);