]>
Commit | Line | Data |
---|---|---|
7dce4e6f ZC |
1 | /* |
2 | * COarse-grain LOck-stepping Virtual Machines for Non-stop Service (COLO) | |
3 | * (a.k.a. Fault Tolerance or Continuous Replication) | |
4 | * | |
5 | * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD. | |
6 | * Copyright (c) 2016 FUJITSU LIMITED | |
7 | * Copyright (c) 2016 Intel Corporation | |
8 | * | |
9 | * Author: Zhang Chen <[email protected]> | |
10 | * | |
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. | |
13 | */ | |
14 | ||
15 | #include "qemu/osdep.h" | |
16 | #include "qemu/error-report.h" | |
59509ec1 | 17 | #include "trace.h" |
7dce4e6f ZC |
18 | #include "qemu-common.h" |
19 | #include "qapi/qmp/qerror.h" | |
20 | #include "qapi/error.h" | |
21 | #include "net/net.h" | |
f4b61836 | 22 | #include "net/eth.h" |
7dce4e6f ZC |
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" | |
59509ec1 | 31 | #include "net/colo.h" |
7dce4e6f ZC |
32 | |
33 | #define TYPE_COLO_COMPARE "colo-compare" | |
34 | #define COLO_COMPARE(obj) \ | |
35 | OBJECT_CHECK(CompareState, (obj), TYPE_COLO_COMPARE) | |
36 | ||
0682e15b | 37 | #define COMPARE_READ_LEN_MAX NET_BUFSIZE |
b6540d40 ZC |
38 | #define MAX_QUEUE_SIZE 1024 |
39 | ||
0682e15b ZC |
40 | /* TODO: Should be configurable */ |
41 | #define REGULAR_PACKET_CHECK_MS 3000 | |
42 | ||
59509ec1 ZC |
43 | /* |
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 | +--------+ +--------+ +--------+ +--------+ | |
64 | */ | |
7dce4e6f ZC |
65 | typedef struct CompareState { |
66 | Object parent; | |
67 | ||
68 | char *pri_indev; | |
69 | char *sec_indev; | |
70 | char *outdev; | |
32a6ebec MAL |
71 | CharBackend chr_pri_in; |
72 | CharBackend chr_sec_in; | |
73 | CharBackend chr_out; | |
7dce4e6f ZC |
74 | SocketReadState pri_rs; |
75 | SocketReadState sec_rs; | |
59509ec1 | 76 | |
b6540d40 ZC |
77 | /* connection list: the connections belonged to this NIC could be found |
78 | * in this list. | |
79 | * element type: Connection | |
80 | */ | |
81 | GQueue conn_list; | |
59509ec1 ZC |
82 | /* hashtable to save connection */ |
83 | GHashTable *connection_track_table; | |
0682e15b ZC |
84 | /* compare thread, a thread for each NIC */ |
85 | QemuThread thread; | |
dfd917a9 | 86 | |
b43decb0 | 87 | GMainContext *worker_context; |
dfd917a9 | 88 | GMainLoop *compare_loop; |
7dce4e6f ZC |
89 | } CompareState; |
90 | ||
91 | typedef struct CompareClass { | |
92 | ObjectClass parent_class; | |
93 | } CompareClass; | |
94 | ||
59509ec1 ZC |
95 | enum { |
96 | PRIMARY_IN = 0, | |
97 | SECONDARY_IN, | |
98 | }; | |
99 | ||
5345fdb4 | 100 | static int compare_chr_send(CharBackend *out, |
59509ec1 ZC |
101 | const uint8_t *buf, |
102 | uint32_t size); | |
103 | ||
a935cc31 ZC |
104 | static gint seq_sorter(Packet *a, Packet *b, gpointer data) |
105 | { | |
106 | struct tcphdr *atcp, *btcp; | |
107 | ||
108 | atcp = (struct tcphdr *)(a->transport_header); | |
109 | btcp = (struct tcphdr *)(b->transport_header); | |
110 | return ntohl(atcp->th_seq) - ntohl(btcp->th_seq); | |
111 | } | |
112 | ||
59509ec1 ZC |
113 | /* |
114 | * Return 0 on success, if return -1 means the pkt | |
115 | * is unsupported(arp and ipv6) and will be sent later | |
116 | */ | |
117 | static int packet_enqueue(CompareState *s, int mode) | |
118 | { | |
b6540d40 | 119 | ConnectionKey key; |
59509ec1 | 120 | Packet *pkt = NULL; |
b6540d40 | 121 | Connection *conn; |
59509ec1 ZC |
122 | |
123 | if (mode == PRIMARY_IN) { | |
124 | pkt = packet_new(s->pri_rs.buf, s->pri_rs.packet_len); | |
125 | } else { | |
126 | pkt = packet_new(s->sec_rs.buf, s->sec_rs.packet_len); | |
127 | } | |
128 | ||
129 | if (parse_packet_early(pkt)) { | |
130 | packet_destroy(pkt, NULL); | |
131 | pkt = NULL; | |
132 | return -1; | |
133 | } | |
b6540d40 | 134 | fill_connection_key(pkt, &key); |
59509ec1 | 135 | |
b6540d40 ZC |
136 | conn = connection_get(s->connection_track_table, |
137 | &key, | |
138 | &s->conn_list); | |
59509ec1 | 139 | |
b6540d40 ZC |
140 | if (!conn->processing) { |
141 | g_queue_push_tail(&s->conn_list, conn); | |
142 | conn->processing = true; | |
143 | } | |
144 | ||
145 | if (mode == PRIMARY_IN) { | |
146 | if (g_queue_get_length(&conn->primary_list) <= | |
147 | MAX_QUEUE_SIZE) { | |
148 | g_queue_push_tail(&conn->primary_list, pkt); | |
a935cc31 ZC |
149 | if (conn->ip_proto == IPPROTO_TCP) { |
150 | g_queue_sort(&conn->primary_list, | |
151 | (GCompareDataFunc)seq_sorter, | |
152 | NULL); | |
153 | } | |
b6540d40 ZC |
154 | } else { |
155 | error_report("colo compare primary queue size too big," | |
156 | "drop packet"); | |
157 | } | |
158 | } else { | |
159 | if (g_queue_get_length(&conn->secondary_list) <= | |
160 | MAX_QUEUE_SIZE) { | |
161 | g_queue_push_tail(&conn->secondary_list, pkt); | |
a935cc31 ZC |
162 | if (conn->ip_proto == IPPROTO_TCP) { |
163 | g_queue_sort(&conn->secondary_list, | |
164 | (GCompareDataFunc)seq_sorter, | |
165 | NULL); | |
166 | } | |
b6540d40 ZC |
167 | } else { |
168 | error_report("colo compare secondary queue size too big," | |
169 | "drop packet"); | |
170 | } | |
171 | } | |
59509ec1 ZC |
172 | |
173 | return 0; | |
174 | } | |
175 | ||
0682e15b ZC |
176 | /* |
177 | * The IP packets sent by primary and secondary | |
178 | * will be compared in here | |
179 | * TODO support ip fragment, Out-Of-Order | |
180 | * return: 0 means packet same | |
181 | * > 0 || < 0 means packet different | |
182 | */ | |
6efeb328 | 183 | static int colo_packet_compare_common(Packet *ppkt, Packet *spkt, int offset) |
0682e15b | 184 | { |
e630b2bf ZC |
185 | if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) { |
186 | char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20]; | |
187 | ||
188 | strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src)); | |
189 | strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst)); | |
190 | strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src)); | |
191 | strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst)); | |
192 | ||
193 | trace_colo_compare_ip_info(ppkt->size, pri_ip_src, | |
194 | pri_ip_dst, spkt->size, | |
195 | sec_ip_src, sec_ip_dst); | |
196 | } | |
0682e15b ZC |
197 | |
198 | if (ppkt->size == spkt->size) { | |
6efeb328 ZC |
199 | return memcmp(ppkt->data + offset, spkt->data + offset, |
200 | spkt->size - offset); | |
0682e15b | 201 | } else { |
2ad7ca4c | 202 | trace_colo_compare_main("Net packet size are not the same"); |
0682e15b ZC |
203 | return -1; |
204 | } | |
205 | } | |
206 | ||
f4b61836 ZC |
207 | /* |
208 | * Called from the compare thread on the primary | |
209 | * for compare tcp packet | |
210 | * compare_tcp copied from Dr. David Alan Gilbert's branch | |
211 | */ | |
212 | static int colo_packet_compare_tcp(Packet *spkt, Packet *ppkt) | |
0682e15b | 213 | { |
f4b61836 ZC |
214 | struct tcphdr *ptcp, *stcp; |
215 | int res; | |
f4b61836 ZC |
216 | |
217 | trace_colo_compare_main("compare tcp"); | |
2ad7ca4c | 218 | |
f4b61836 ZC |
219 | ptcp = (struct tcphdr *)ppkt->transport_header; |
220 | stcp = (struct tcphdr *)spkt->transport_header; | |
221 | ||
222 | /* | |
223 | * The 'identification' field in the IP header is *very* random | |
224 | * it almost never matches. Fudge this by ignoring differences in | |
225 | * unfragmented packets; they'll normally sort themselves out if different | |
226 | * anyway, and it should recover at the TCP level. | |
227 | * An alternative would be to get both the primary and secondary to rewrite | |
228 | * somehow; but that would need some sync traffic to sync the state | |
229 | */ | |
230 | if (ntohs(ppkt->ip->ip_off) & IP_DF) { | |
231 | spkt->ip->ip_id = ppkt->ip->ip_id; | |
232 | /* and the sum will be different if the IDs were different */ | |
233 | spkt->ip->ip_sum = ppkt->ip->ip_sum; | |
234 | } | |
235 | ||
184d4d42 ZC |
236 | /* |
237 | * Check tcp header length for tcp option field. | |
238 | * th_off > 5 means this tcp packet have options field. | |
239 | * The tcp options maybe always different. | |
240 | * for example: | |
241 | * From RFC 7323. | |
242 | * TCP Timestamps option (TSopt): | |
243 | * Kind: 8 | |
244 | * | |
245 | * Length: 10 bytes | |
246 | * | |
247 | * +-------+-------+---------------------+---------------------+ | |
248 | * |Kind=8 | 10 | TS Value (TSval) |TS Echo Reply (TSecr)| | |
249 | * +-------+-------+---------------------+---------------------+ | |
250 | * 1 1 4 4 | |
251 | * | |
252 | * In this case the primary guest's timestamp always different with | |
253 | * the secondary guest's timestamp. COLO just focus on payload, | |
254 | * so we just need skip this field. | |
255 | */ | |
256 | if (ptcp->th_off > 5) { | |
257 | ptrdiff_t tcp_offset; | |
258 | tcp_offset = ppkt->transport_header - (uint8_t *)ppkt->data | |
259 | + (ptcp->th_off * 4); | |
260 | res = colo_packet_compare_common(ppkt, spkt, tcp_offset); | |
261 | } else if (ptcp->th_sum == stcp->th_sum) { | |
6efeb328 ZC |
262 | res = colo_packet_compare_common(ppkt, spkt, ETH_HLEN); |
263 | } else { | |
264 | res = -1; | |
265 | } | |
f4b61836 | 266 | |
51b9d495 PM |
267 | if (res != 0 && trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) { |
268 | trace_colo_compare_pkt_info_src(inet_ntoa(ppkt->ip->ip_src), | |
269 | ntohl(stcp->th_seq), | |
270 | ntohl(stcp->th_ack), | |
271 | res, stcp->th_flags, | |
272 | spkt->size); | |
273 | ||
274 | trace_colo_compare_pkt_info_dst(inet_ntoa(ppkt->ip->ip_dst), | |
275 | ntohl(ptcp->th_seq), | |
276 | ntohl(ptcp->th_ack), | |
277 | res, ptcp->th_flags, | |
278 | ppkt->size); | |
2061c14c ZC |
279 | |
280 | qemu_hexdump((char *)ppkt->data, stderr, | |
281 | "colo-compare ppkt", ppkt->size); | |
282 | qemu_hexdump((char *)spkt->data, stderr, | |
283 | "colo-compare spkt", spkt->size); | |
f4b61836 ZC |
284 | } |
285 | ||
286 | return res; | |
287 | } | |
288 | ||
289 | /* | |
290 | * Called from the compare thread on the primary | |
291 | * for compare udp packet | |
292 | */ | |
293 | static int colo_packet_compare_udp(Packet *spkt, Packet *ppkt) | |
294 | { | |
295 | int ret; | |
6efeb328 | 296 | int network_header_length = ppkt->ip->ip_hl * 4; |
f4b61836 ZC |
297 | |
298 | trace_colo_compare_main("compare udp"); | |
2ad7ca4c | 299 | |
6efeb328 ZC |
300 | /* |
301 | * Because of ppkt and spkt are both in the same connection, | |
302 | * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are | |
303 | * same with spkt. In addition, IP header's Identification is a random | |
304 | * field, we can handle it in IP fragmentation function later. | |
305 | * COLO just concern the response net packet payload from primary guest | |
306 | * and secondary guest are same or not, So we ignored all IP header include | |
307 | * other field like TOS,TTL,IP Checksum. we only need to compare | |
308 | * the ip payload here. | |
309 | */ | |
310 | ret = colo_packet_compare_common(ppkt, spkt, | |
311 | network_header_length + ETH_HLEN); | |
f4b61836 ZC |
312 | |
313 | if (ret) { | |
314 | trace_colo_compare_udp_miscompare("primary pkt size", ppkt->size); | |
f4b61836 | 315 | trace_colo_compare_udp_miscompare("Secondary pkt size", spkt->size); |
1723a7f7 ZC |
316 | if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) { |
317 | qemu_hexdump((char *)ppkt->data, stderr, "colo-compare pri pkt", | |
318 | ppkt->size); | |
319 | qemu_hexdump((char *)spkt->data, stderr, "colo-compare sec pkt", | |
320 | spkt->size); | |
321 | } | |
f4b61836 ZC |
322 | } |
323 | ||
324 | return ret; | |
325 | } | |
326 | ||
327 | /* | |
328 | * Called from the compare thread on the primary | |
329 | * for compare icmp packet | |
330 | */ | |
331 | static int colo_packet_compare_icmp(Packet *spkt, Packet *ppkt) | |
332 | { | |
6efeb328 ZC |
333 | int network_header_length = ppkt->ip->ip_hl * 4; |
334 | ||
f4b61836 | 335 | trace_colo_compare_main("compare icmp"); |
f4b61836 | 336 | |
6efeb328 ZC |
337 | /* |
338 | * Because of ppkt and spkt are both in the same connection, | |
339 | * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are | |
340 | * same with spkt. In addition, IP header's Identification is a random | |
341 | * field, we can handle it in IP fragmentation function later. | |
342 | * COLO just concern the response net packet payload from primary guest | |
343 | * and secondary guest are same or not, So we ignored all IP header include | |
344 | * other field like TOS,TTL,IP Checksum. we only need to compare | |
345 | * the ip payload here. | |
346 | */ | |
347 | if (colo_packet_compare_common(ppkt, spkt, | |
348 | network_header_length + ETH_HLEN)) { | |
f4b61836 ZC |
349 | trace_colo_compare_icmp_miscompare("primary pkt size", |
350 | ppkt->size); | |
f4b61836 ZC |
351 | trace_colo_compare_icmp_miscompare("Secondary pkt size", |
352 | spkt->size); | |
1723a7f7 ZC |
353 | if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) { |
354 | qemu_hexdump((char *)ppkt->data, stderr, "colo-compare pri pkt", | |
355 | ppkt->size); | |
356 | qemu_hexdump((char *)spkt->data, stderr, "colo-compare sec pkt", | |
357 | spkt->size); | |
358 | } | |
f4b61836 ZC |
359 | return -1; |
360 | } else { | |
361 | return 0; | |
362 | } | |
363 | } | |
364 | ||
365 | /* | |
366 | * Called from the compare thread on the primary | |
367 | * for compare other packet | |
368 | */ | |
369 | static int colo_packet_compare_other(Packet *spkt, Packet *ppkt) | |
370 | { | |
371 | trace_colo_compare_main("compare other"); | |
e630b2bf ZC |
372 | if (trace_event_get_state(TRACE_COLO_COMPARE_MISCOMPARE)) { |
373 | char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20]; | |
374 | ||
375 | strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src)); | |
376 | strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst)); | |
377 | strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src)); | |
378 | strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst)); | |
379 | ||
380 | trace_colo_compare_ip_info(ppkt->size, pri_ip_src, | |
381 | pri_ip_dst, spkt->size, | |
382 | sec_ip_src, sec_ip_dst); | |
383 | } | |
384 | ||
6efeb328 | 385 | return colo_packet_compare_common(ppkt, spkt, 0); |
0682e15b ZC |
386 | } |
387 | ||
388 | static int colo_old_packet_check_one(Packet *pkt, int64_t *check_time) | |
389 | { | |
390 | int64_t now = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
391 | ||
392 | if ((now - pkt->creation_ms) > (*check_time)) { | |
393 | trace_colo_old_packet_check_found(pkt->creation_ms); | |
394 | return 0; | |
395 | } else { | |
396 | return 1; | |
397 | } | |
398 | } | |
399 | ||
d25a7dab ZC |
400 | static int colo_old_packet_check_one_conn(Connection *conn, |
401 | void *user_data) | |
0682e15b | 402 | { |
0682e15b ZC |
403 | GList *result = NULL; |
404 | int64_t check_time = REGULAR_PACKET_CHECK_MS; | |
405 | ||
406 | result = g_queue_find_custom(&conn->primary_list, | |
407 | &check_time, | |
408 | (GCompareFunc)colo_old_packet_check_one); | |
409 | ||
410 | if (result) { | |
411 | /* do checkpoint will flush old packet */ | |
412 | /* TODO: colo_notify_checkpoint();*/ | |
d25a7dab | 413 | return 0; |
0682e15b | 414 | } |
d25a7dab ZC |
415 | |
416 | return 1; | |
0682e15b ZC |
417 | } |
418 | ||
419 | /* | |
420 | * Look for old packets that the secondary hasn't matched, | |
421 | * if we have some then we have to checkpoint to wake | |
422 | * the secondary up. | |
423 | */ | |
424 | static void colo_old_packet_check(void *opaque) | |
425 | { | |
426 | CompareState *s = opaque; | |
427 | ||
d25a7dab ZC |
428 | /* |
429 | * If we find one old packet, stop finding job and notify | |
430 | * COLO frame do checkpoint. | |
431 | */ | |
432 | g_queue_find_custom(&s->conn_list, NULL, | |
433 | (GCompareFunc)colo_old_packet_check_one_conn); | |
0682e15b ZC |
434 | } |
435 | ||
436 | /* | |
437 | * Called from the compare thread on the primary | |
438 | * for compare connection | |
439 | */ | |
440 | static void colo_compare_connection(void *opaque, void *user_data) | |
441 | { | |
442 | CompareState *s = user_data; | |
443 | Connection *conn = opaque; | |
444 | Packet *pkt = NULL; | |
445 | GList *result = NULL; | |
446 | int ret; | |
447 | ||
448 | while (!g_queue_is_empty(&conn->primary_list) && | |
449 | !g_queue_is_empty(&conn->secondary_list)) { | |
0682e15b | 450 | pkt = g_queue_pop_tail(&conn->primary_list); |
f4b61836 ZC |
451 | switch (conn->ip_proto) { |
452 | case IPPROTO_TCP: | |
453 | result = g_queue_find_custom(&conn->secondary_list, | |
454 | pkt, (GCompareFunc)colo_packet_compare_tcp); | |
455 | break; | |
456 | case IPPROTO_UDP: | |
457 | result = g_queue_find_custom(&conn->secondary_list, | |
458 | pkt, (GCompareFunc)colo_packet_compare_udp); | |
459 | break; | |
460 | case IPPROTO_ICMP: | |
461 | result = g_queue_find_custom(&conn->secondary_list, | |
462 | pkt, (GCompareFunc)colo_packet_compare_icmp); | |
463 | break; | |
464 | default: | |
465 | result = g_queue_find_custom(&conn->secondary_list, | |
466 | pkt, (GCompareFunc)colo_packet_compare_other); | |
467 | break; | |
468 | } | |
0682e15b ZC |
469 | |
470 | if (result) { | |
5345fdb4 | 471 | ret = compare_chr_send(&s->chr_out, pkt->data, pkt->size); |
0682e15b ZC |
472 | if (ret < 0) { |
473 | error_report("colo_send_primary_packet failed"); | |
474 | } | |
475 | trace_colo_compare_main("packet same and release packet"); | |
476 | g_queue_remove(&conn->secondary_list, result->data); | |
477 | packet_destroy(pkt, NULL); | |
478 | } else { | |
479 | /* | |
480 | * If one packet arrive late, the secondary_list or | |
481 | * primary_list will be empty, so we can't compare it | |
482 | * until next comparison. | |
483 | */ | |
484 | trace_colo_compare_main("packet different"); | |
0682e15b | 485 | g_queue_push_tail(&conn->primary_list, pkt); |
0682e15b ZC |
486 | /* TODO: colo_notify_checkpoint();*/ |
487 | break; | |
488 | } | |
489 | } | |
490 | } | |
491 | ||
5345fdb4 | 492 | static int compare_chr_send(CharBackend *out, |
59509ec1 ZC |
493 | const uint8_t *buf, |
494 | uint32_t size) | |
495 | { | |
496 | int ret = 0; | |
497 | uint32_t len = htonl(size); | |
498 | ||
499 | if (!size) { | |
500 | return 0; | |
501 | } | |
502 | ||
503 | ret = qemu_chr_fe_write_all(out, (uint8_t *)&len, sizeof(len)); | |
504 | if (ret != sizeof(len)) { | |
505 | goto err; | |
506 | } | |
507 | ||
508 | ret = qemu_chr_fe_write_all(out, (uint8_t *)buf, size); | |
509 | if (ret != size) { | |
510 | goto err; | |
511 | } | |
512 | ||
513 | return 0; | |
514 | ||
515 | err: | |
516 | return ret < 0 ? ret : -EIO; | |
517 | } | |
518 | ||
0682e15b ZC |
519 | static int compare_chr_can_read(void *opaque) |
520 | { | |
521 | return COMPARE_READ_LEN_MAX; | |
522 | } | |
523 | ||
524 | /* | |
525 | * Called from the main thread on the primary for packets | |
526 | * arriving over the socket from the primary. | |
527 | */ | |
528 | static void compare_pri_chr_in(void *opaque, const uint8_t *buf, int size) | |
529 | { | |
530 | CompareState *s = COLO_COMPARE(opaque); | |
531 | int ret; | |
532 | ||
533 | ret = net_fill_rstate(&s->pri_rs, buf, size); | |
534 | if (ret == -1) { | |
39ab61c6 MAL |
535 | qemu_chr_fe_set_handlers(&s->chr_pri_in, NULL, NULL, NULL, |
536 | NULL, NULL, true); | |
0682e15b ZC |
537 | error_report("colo-compare primary_in error"); |
538 | } | |
539 | } | |
540 | ||
541 | /* | |
542 | * Called from the main thread on the primary for packets | |
543 | * arriving over the socket from the secondary. | |
544 | */ | |
545 | static void compare_sec_chr_in(void *opaque, const uint8_t *buf, int size) | |
546 | { | |
547 | CompareState *s = COLO_COMPARE(opaque); | |
548 | int ret; | |
549 | ||
550 | ret = net_fill_rstate(&s->sec_rs, buf, size); | |
551 | if (ret == -1) { | |
39ab61c6 MAL |
552 | qemu_chr_fe_set_handlers(&s->chr_sec_in, NULL, NULL, NULL, |
553 | NULL, NULL, true); | |
0682e15b ZC |
554 | error_report("colo-compare secondary_in error"); |
555 | } | |
556 | } | |
557 | ||
66d2a242 HZ |
558 | /* |
559 | * Check old packet regularly so it can watch for any packets | |
560 | * that the secondary hasn't produced equivalents of. | |
561 | */ | |
562 | static gboolean check_old_packet_regular(void *opaque) | |
563 | { | |
564 | CompareState *s = opaque; | |
565 | ||
566 | /* if have old packet we will notify checkpoint */ | |
567 | colo_old_packet_check(s); | |
568 | ||
569 | return TRUE; | |
570 | } | |
571 | ||
0682e15b ZC |
572 | static void *colo_compare_thread(void *opaque) |
573 | { | |
0682e15b | 574 | CompareState *s = opaque; |
66d2a242 | 575 | GSource *timeout_source; |
0682e15b | 576 | |
b43decb0 | 577 | s->worker_context = g_main_context_new(); |
0682e15b | 578 | |
5345fdb4 | 579 | qemu_chr_fe_set_handlers(&s->chr_pri_in, compare_chr_can_read, |
b43decb0 | 580 | compare_pri_chr_in, NULL, s, s->worker_context, true); |
5345fdb4 | 581 | qemu_chr_fe_set_handlers(&s->chr_sec_in, compare_chr_can_read, |
b43decb0 | 582 | compare_sec_chr_in, NULL, s, s->worker_context, true); |
0682e15b | 583 | |
b43decb0 | 584 | s->compare_loop = g_main_loop_new(s->worker_context, FALSE); |
0682e15b | 585 | |
66d2a242 HZ |
586 | /* To kick any packets that the secondary doesn't match */ |
587 | timeout_source = g_timeout_source_new(REGULAR_PACKET_CHECK_MS); | |
588 | g_source_set_callback(timeout_source, | |
589 | (GSourceFunc)check_old_packet_regular, s, NULL); | |
b43decb0 | 590 | g_source_attach(timeout_source, s->worker_context); |
66d2a242 | 591 | |
dfd917a9 | 592 | g_main_loop_run(s->compare_loop); |
0682e15b | 593 | |
66d2a242 | 594 | g_source_unref(timeout_source); |
dfd917a9 | 595 | g_main_loop_unref(s->compare_loop); |
b43decb0 | 596 | g_main_context_unref(s->worker_context); |
0682e15b ZC |
597 | return NULL; |
598 | } | |
599 | ||
7dce4e6f ZC |
600 | static char *compare_get_pri_indev(Object *obj, Error **errp) |
601 | { | |
602 | CompareState *s = COLO_COMPARE(obj); | |
603 | ||
604 | return g_strdup(s->pri_indev); | |
605 | } | |
606 | ||
607 | static void compare_set_pri_indev(Object *obj, const char *value, Error **errp) | |
608 | { | |
609 | CompareState *s = COLO_COMPARE(obj); | |
610 | ||
611 | g_free(s->pri_indev); | |
612 | s->pri_indev = g_strdup(value); | |
613 | } | |
614 | ||
615 | static char *compare_get_sec_indev(Object *obj, Error **errp) | |
616 | { | |
617 | CompareState *s = COLO_COMPARE(obj); | |
618 | ||
619 | return g_strdup(s->sec_indev); | |
620 | } | |
621 | ||
622 | static void compare_set_sec_indev(Object *obj, const char *value, Error **errp) | |
623 | { | |
624 | CompareState *s = COLO_COMPARE(obj); | |
625 | ||
626 | g_free(s->sec_indev); | |
627 | s->sec_indev = g_strdup(value); | |
628 | } | |
629 | ||
630 | static char *compare_get_outdev(Object *obj, Error **errp) | |
631 | { | |
632 | CompareState *s = COLO_COMPARE(obj); | |
633 | ||
634 | return g_strdup(s->outdev); | |
635 | } | |
636 | ||
637 | static void compare_set_outdev(Object *obj, const char *value, Error **errp) | |
638 | { | |
639 | CompareState *s = COLO_COMPARE(obj); | |
640 | ||
641 | g_free(s->outdev); | |
642 | s->outdev = g_strdup(value); | |
643 | } | |
644 | ||
645 | static void compare_pri_rs_finalize(SocketReadState *pri_rs) | |
646 | { | |
59509ec1 ZC |
647 | CompareState *s = container_of(pri_rs, CompareState, pri_rs); |
648 | ||
649 | if (packet_enqueue(s, PRIMARY_IN)) { | |
650 | trace_colo_compare_main("primary: unsupported packet in"); | |
5345fdb4 | 651 | compare_chr_send(&s->chr_out, pri_rs->buf, pri_rs->packet_len); |
0682e15b ZC |
652 | } else { |
653 | /* compare connection */ | |
654 | g_queue_foreach(&s->conn_list, colo_compare_connection, s); | |
59509ec1 | 655 | } |
7dce4e6f ZC |
656 | } |
657 | ||
658 | static void compare_sec_rs_finalize(SocketReadState *sec_rs) | |
659 | { | |
59509ec1 ZC |
660 | CompareState *s = container_of(sec_rs, CompareState, sec_rs); |
661 | ||
662 | if (packet_enqueue(s, SECONDARY_IN)) { | |
663 | trace_colo_compare_main("secondary: unsupported packet in"); | |
0682e15b ZC |
664 | } else { |
665 | /* compare connection */ | |
666 | g_queue_foreach(&s->conn_list, colo_compare_connection, s); | |
59509ec1 | 667 | } |
7dce4e6f ZC |
668 | } |
669 | ||
7dce4e6f ZC |
670 | |
671 | /* | |
672 | * Return 0 is success. | |
673 | * Return 1 is failed. | |
674 | */ | |
0ec7b3e7 | 675 | static int find_and_check_chardev(Chardev **chr, |
7dce4e6f ZC |
676 | char *chr_name, |
677 | Error **errp) | |
678 | { | |
7dce4e6f ZC |
679 | *chr = qemu_chr_find(chr_name); |
680 | if (*chr == NULL) { | |
681 | error_setg(errp, "Device '%s' not found", | |
682 | chr_name); | |
683 | return 1; | |
684 | } | |
685 | ||
0a73336d DB |
686 | if (!qemu_chr_has_feature(*chr, QEMU_CHAR_FEATURE_RECONNECTABLE)) { |
687 | error_setg(errp, "chardev \"%s\" is not reconnectable", | |
7dce4e6f ZC |
688 | chr_name); |
689 | return 1; | |
690 | } | |
fbf3cc3a | 691 | |
7dce4e6f ZC |
692 | return 0; |
693 | } | |
694 | ||
695 | /* | |
696 | * Called from the main thread on the primary | |
697 | * to setup colo-compare. | |
698 | */ | |
699 | static void colo_compare_complete(UserCreatable *uc, Error **errp) | |
700 | { | |
701 | CompareState *s = COLO_COMPARE(uc); | |
0ec7b3e7 | 702 | Chardev *chr; |
0682e15b ZC |
703 | char thread_name[64]; |
704 | static int compare_id; | |
7dce4e6f ZC |
705 | |
706 | if (!s->pri_indev || !s->sec_indev || !s->outdev) { | |
707 | error_setg(errp, "colo compare needs 'primary_in' ," | |
708 | "'secondary_in','outdev' property set"); | |
709 | return; | |
710 | } else if (!strcmp(s->pri_indev, s->outdev) || | |
711 | !strcmp(s->sec_indev, s->outdev) || | |
712 | !strcmp(s->pri_indev, s->sec_indev)) { | |
713 | error_setg(errp, "'indev' and 'outdev' could not be same " | |
714 | "for compare module"); | |
715 | return; | |
716 | } | |
717 | ||
5345fdb4 MAL |
718 | if (find_and_check_chardev(&chr, s->pri_indev, errp) || |
719 | !qemu_chr_fe_init(&s->chr_pri_in, chr, errp)) { | |
7dce4e6f ZC |
720 | return; |
721 | } | |
722 | ||
5345fdb4 MAL |
723 | if (find_and_check_chardev(&chr, s->sec_indev, errp) || |
724 | !qemu_chr_fe_init(&s->chr_sec_in, chr, errp)) { | |
7dce4e6f ZC |
725 | return; |
726 | } | |
727 | ||
5345fdb4 MAL |
728 | if (find_and_check_chardev(&chr, s->outdev, errp) || |
729 | !qemu_chr_fe_init(&s->chr_out, chr, errp)) { | |
7dce4e6f ZC |
730 | return; |
731 | } | |
732 | ||
7dce4e6f ZC |
733 | net_socket_rs_init(&s->pri_rs, compare_pri_rs_finalize); |
734 | net_socket_rs_init(&s->sec_rs, compare_sec_rs_finalize); | |
735 | ||
b6540d40 ZC |
736 | g_queue_init(&s->conn_list); |
737 | ||
738 | s->connection_track_table = g_hash_table_new_full(connection_key_hash, | |
739 | connection_key_equal, | |
740 | g_free, | |
741 | connection_destroy); | |
59509ec1 | 742 | |
0682e15b ZC |
743 | sprintf(thread_name, "colo-compare %d", compare_id); |
744 | qemu_thread_create(&s->thread, thread_name, | |
745 | colo_compare_thread, s, | |
746 | QEMU_THREAD_JOINABLE); | |
747 | compare_id++; | |
748 | ||
7dce4e6f ZC |
749 | return; |
750 | } | |
751 | ||
dfd917a9 HZ |
752 | static void colo_flush_packets(void *opaque, void *user_data) |
753 | { | |
754 | CompareState *s = user_data; | |
755 | Connection *conn = opaque; | |
756 | Packet *pkt = NULL; | |
757 | ||
758 | while (!g_queue_is_empty(&conn->primary_list)) { | |
759 | pkt = g_queue_pop_head(&conn->primary_list); | |
760 | compare_chr_send(&s->chr_out, pkt->data, pkt->size); | |
761 | packet_destroy(pkt, NULL); | |
762 | } | |
763 | while (!g_queue_is_empty(&conn->secondary_list)) { | |
764 | pkt = g_queue_pop_head(&conn->secondary_list); | |
765 | packet_destroy(pkt, NULL); | |
766 | } | |
767 | } | |
768 | ||
7dce4e6f ZC |
769 | static void colo_compare_class_init(ObjectClass *oc, void *data) |
770 | { | |
771 | UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc); | |
772 | ||
773 | ucc->complete = colo_compare_complete; | |
774 | } | |
775 | ||
776 | static void colo_compare_init(Object *obj) | |
777 | { | |
778 | object_property_add_str(obj, "primary_in", | |
779 | compare_get_pri_indev, compare_set_pri_indev, | |
780 | NULL); | |
781 | object_property_add_str(obj, "secondary_in", | |
782 | compare_get_sec_indev, compare_set_sec_indev, | |
783 | NULL); | |
784 | object_property_add_str(obj, "outdev", | |
785 | compare_get_outdev, compare_set_outdev, | |
786 | NULL); | |
787 | } | |
788 | ||
789 | static void colo_compare_finalize(Object *obj) | |
790 | { | |
791 | CompareState *s = COLO_COMPARE(obj); | |
792 | ||
b43decb0 HZ |
793 | qemu_chr_fe_set_handlers(&s->chr_pri_in, NULL, NULL, NULL, NULL, |
794 | s->worker_context, true); | |
795 | qemu_chr_fe_set_handlers(&s->chr_sec_in, NULL, NULL, NULL, NULL, | |
796 | s->worker_context, true); | |
c39860e6 | 797 | qemu_chr_fe_deinit(&s->chr_out); |
7dce4e6f | 798 | |
dfd917a9 HZ |
799 | g_main_loop_quit(s->compare_loop); |
800 | qemu_thread_join(&s->thread); | |
b6540d40 | 801 | |
dfd917a9 HZ |
802 | /* Release all unhandled packets after compare thead exited */ |
803 | g_queue_foreach(&s->conn_list, colo_flush_packets, s); | |
804 | ||
727c2d76 | 805 | g_queue_clear(&s->conn_list); |
0682e15b | 806 | |
dfd917a9 | 807 | g_hash_table_destroy(s->connection_track_table); |
7dce4e6f ZC |
808 | g_free(s->pri_indev); |
809 | g_free(s->sec_indev); | |
810 | g_free(s->outdev); | |
811 | } | |
812 | ||
813 | static const TypeInfo colo_compare_info = { | |
814 | .name = TYPE_COLO_COMPARE, | |
815 | .parent = TYPE_OBJECT, | |
816 | .instance_size = sizeof(CompareState), | |
817 | .instance_init = colo_compare_init, | |
818 | .instance_finalize = colo_compare_finalize, | |
819 | .class_size = sizeof(CompareClass), | |
820 | .class_init = colo_compare_class_init, | |
821 | .interfaces = (InterfaceInfo[]) { | |
822 | { TYPE_USER_CREATABLE }, | |
823 | { } | |
824 | } | |
825 | }; | |
826 | ||
827 | static void register_types(void) | |
828 | { | |
829 | type_register_static(&colo_compare_info); | |
830 | } | |
831 | ||
832 | type_init(register_types); |