]>
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 <zhangchen.fnst@cn.fujitsu.com> | |
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" | |
4d43a603 | 28 | #include "chardev/char-fe.h" |
7dce4e6f ZC |
29 | #include "qemu/sockets.h" |
30 | #include "qapi-visit.h" | |
59509ec1 | 31 | #include "net/colo.h" |
dd321ecf | 32 | #include "sysemu/iothread.h" |
7dce4e6f ZC |
33 | |
34 | #define TYPE_COLO_COMPARE "colo-compare" | |
35 | #define COLO_COMPARE(obj) \ | |
36 | OBJECT_CHECK(CompareState, (obj), TYPE_COLO_COMPARE) | |
37 | ||
0682e15b | 38 | #define COMPARE_READ_LEN_MAX NET_BUFSIZE |
b6540d40 ZC |
39 | #define MAX_QUEUE_SIZE 1024 |
40 | ||
0682e15b ZC |
41 | /* TODO: Should be configurable */ |
42 | #define REGULAR_PACKET_CHECK_MS 3000 | |
43 | ||
59509ec1 | 44 | /* |
61c5f469 ZC |
45 | * + CompareState ++ |
46 | * | | | |
47 | * +---------------+ +---------------+ +---------------+ | |
48 | * | conn list + - > conn + ------- > conn + -- > ...... | |
49 | * +---------------+ +---------------+ +---------------+ | |
50 | * | | | | | | | |
51 | * +---------------+ +---v----+ +---v----+ +---v----+ +---v----+ | |
52 | * |primary | |secondary |primary | |secondary | |
53 | * |packet | |packet + |packet | |packet + | |
54 | * +--------+ +--------+ +--------+ +--------+ | |
55 | * | | | | | |
56 | * +---v----+ +---v----+ +---v----+ +---v----+ | |
57 | * |primary | |secondary |primary | |secondary | |
58 | * |packet | |packet + |packet | |packet + | |
59 | * +--------+ +--------+ +--------+ +--------+ | |
60 | * | | | | | |
61 | * +---v----+ +---v----+ +---v----+ +---v----+ | |
62 | * |primary | |secondary |primary | |secondary | |
63 | * |packet | |packet + |packet | |packet + | |
64 | * +--------+ +--------+ +--------+ +--------+ | |
65 | */ | |
7dce4e6f ZC |
66 | typedef struct CompareState { |
67 | Object parent; | |
68 | ||
69 | char *pri_indev; | |
70 | char *sec_indev; | |
71 | char *outdev; | |
32a6ebec MAL |
72 | CharBackend chr_pri_in; |
73 | CharBackend chr_sec_in; | |
74 | CharBackend chr_out; | |
7dce4e6f ZC |
75 | SocketReadState pri_rs; |
76 | SocketReadState sec_rs; | |
aa3a7032 | 77 | bool vnet_hdr; |
59509ec1 | 78 | |
61c5f469 ZC |
79 | /* |
80 | * Record the connection that through the NIC | |
81 | * Element type: Connection | |
b6540d40 ZC |
82 | */ |
83 | GQueue conn_list; | |
61c5f469 | 84 | /* Record the connection without repetition */ |
59509ec1 | 85 | GHashTable *connection_track_table; |
dfd917a9 | 86 | |
dd321ecf | 87 | IOThread *iothread; |
b43decb0 | 88 | GMainContext *worker_context; |
dd321ecf | 89 | QEMUTimer *packet_check_timer; |
7dce4e6f ZC |
90 | } CompareState; |
91 | ||
92 | typedef struct CompareClass { | |
93 | ObjectClass parent_class; | |
94 | } CompareClass; | |
95 | ||
59509ec1 ZC |
96 | enum { |
97 | PRIMARY_IN = 0, | |
98 | SECONDARY_IN, | |
99 | }; | |
100 | ||
3037e7a5 | 101 | static int compare_chr_send(CompareState *s, |
59509ec1 | 102 | const uint8_t *buf, |
aa3a7032 ZC |
103 | uint32_t size, |
104 | uint32_t vnet_hdr_len); | |
59509ec1 | 105 | |
a935cc31 ZC |
106 | static gint seq_sorter(Packet *a, Packet *b, gpointer data) |
107 | { | |
108 | struct tcphdr *atcp, *btcp; | |
109 | ||
110 | atcp = (struct tcphdr *)(a->transport_header); | |
111 | btcp = (struct tcphdr *)(b->transport_header); | |
112 | return ntohl(atcp->th_seq) - ntohl(btcp->th_seq); | |
113 | } | |
114 | ||
8850d4ca MZ |
115 | /* |
116 | * Return 1 on success, if return 0 means the | |
117 | * packet will be dropped | |
118 | */ | |
119 | static int colo_insert_packet(GQueue *queue, Packet *pkt) | |
120 | { | |
121 | if (g_queue_get_length(queue) <= MAX_QUEUE_SIZE) { | |
122 | if (pkt->ip->ip_p == IPPROTO_TCP) { | |
123 | g_queue_insert_sorted(queue, | |
124 | pkt, | |
125 | (GCompareDataFunc)seq_sorter, | |
126 | NULL); | |
127 | } else { | |
128 | g_queue_push_tail(queue, pkt); | |
129 | } | |
130 | return 1; | |
131 | } | |
132 | return 0; | |
133 | } | |
134 | ||
59509ec1 ZC |
135 | /* |
136 | * Return 0 on success, if return -1 means the pkt | |
137 | * is unsupported(arp and ipv6) and will be sent later | |
138 | */ | |
8ec14402 | 139 | static int packet_enqueue(CompareState *s, int mode, Connection **con) |
59509ec1 | 140 | { |
b6540d40 | 141 | ConnectionKey key; |
59509ec1 | 142 | Packet *pkt = NULL; |
b6540d40 | 143 | Connection *conn; |
59509ec1 ZC |
144 | |
145 | if (mode == PRIMARY_IN) { | |
ada1a33f ZC |
146 | pkt = packet_new(s->pri_rs.buf, |
147 | s->pri_rs.packet_len, | |
148 | s->pri_rs.vnet_hdr_len); | |
59509ec1 | 149 | } else { |
ada1a33f ZC |
150 | pkt = packet_new(s->sec_rs.buf, |
151 | s->sec_rs.packet_len, | |
152 | s->sec_rs.vnet_hdr_len); | |
59509ec1 ZC |
153 | } |
154 | ||
155 | if (parse_packet_early(pkt)) { | |
156 | packet_destroy(pkt, NULL); | |
157 | pkt = NULL; | |
158 | return -1; | |
159 | } | |
b6540d40 | 160 | fill_connection_key(pkt, &key); |
59509ec1 | 161 | |
b6540d40 ZC |
162 | conn = connection_get(s->connection_track_table, |
163 | &key, | |
164 | &s->conn_list); | |
59509ec1 | 165 | |
b6540d40 ZC |
166 | if (!conn->processing) { |
167 | g_queue_push_tail(&s->conn_list, conn); | |
168 | conn->processing = true; | |
169 | } | |
170 | ||
171 | if (mode == PRIMARY_IN) { | |
8850d4ca | 172 | if (!colo_insert_packet(&conn->primary_list, pkt)) { |
b6540d40 ZC |
173 | error_report("colo compare primary queue size too big," |
174 | "drop packet"); | |
175 | } | |
176 | } else { | |
8850d4ca | 177 | if (!colo_insert_packet(&conn->secondary_list, pkt)) { |
b6540d40 ZC |
178 | error_report("colo compare secondary queue size too big," |
179 | "drop packet"); | |
180 | } | |
181 | } | |
8ec14402 | 182 | con = &conn; |
59509ec1 ZC |
183 | |
184 | return 0; | |
185 | } | |
186 | ||
0682e15b ZC |
187 | /* |
188 | * The IP packets sent by primary and secondary | |
189 | * will be compared in here | |
190 | * TODO support ip fragment, Out-Of-Order | |
191 | * return: 0 means packet same | |
192 | * > 0 || < 0 means packet different | |
193 | */ | |
6f5009c3 ZC |
194 | static int colo_packet_compare_common(Packet *ppkt, |
195 | Packet *spkt, | |
196 | int poffset, | |
197 | int soffset) | |
0682e15b | 198 | { |
d87aa138 | 199 | if (trace_event_get_state_backends(TRACE_COLO_COMPARE_MISCOMPARE)) { |
e630b2bf ZC |
200 | char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20]; |
201 | ||
202 | strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src)); | |
203 | strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst)); | |
204 | strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src)); | |
205 | strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst)); | |
206 | ||
207 | trace_colo_compare_ip_info(ppkt->size, pri_ip_src, | |
208 | pri_ip_dst, spkt->size, | |
209 | sec_ip_src, sec_ip_dst); | |
210 | } | |
0682e15b | 211 | |
6f5009c3 ZC |
212 | poffset = ppkt->vnet_hdr_len + poffset; |
213 | soffset = ppkt->vnet_hdr_len + soffset; | |
d63b366a | 214 | |
6f5009c3 ZC |
215 | if (ppkt->size - poffset == spkt->size - soffset) { |
216 | return memcmp(ppkt->data + poffset, | |
217 | spkt->data + soffset, | |
218 | spkt->size - soffset); | |
0682e15b | 219 | } else { |
2ad7ca4c | 220 | trace_colo_compare_main("Net packet size are not the same"); |
0682e15b ZC |
221 | return -1; |
222 | } | |
223 | } | |
224 | ||
f4b61836 ZC |
225 | /* |
226 | * Called from the compare thread on the primary | |
227 | * for compare tcp packet | |
228 | * compare_tcp copied from Dr. David Alan Gilbert's branch | |
229 | */ | |
230 | static int colo_packet_compare_tcp(Packet *spkt, Packet *ppkt) | |
0682e15b | 231 | { |
f4b61836 ZC |
232 | struct tcphdr *ptcp, *stcp; |
233 | int res; | |
f4b61836 ZC |
234 | |
235 | trace_colo_compare_main("compare tcp"); | |
2ad7ca4c | 236 | |
f4b61836 ZC |
237 | ptcp = (struct tcphdr *)ppkt->transport_header; |
238 | stcp = (struct tcphdr *)spkt->transport_header; | |
239 | ||
240 | /* | |
241 | * The 'identification' field in the IP header is *very* random | |
242 | * it almost never matches. Fudge this by ignoring differences in | |
243 | * unfragmented packets; they'll normally sort themselves out if different | |
244 | * anyway, and it should recover at the TCP level. | |
245 | * An alternative would be to get both the primary and secondary to rewrite | |
246 | * somehow; but that would need some sync traffic to sync the state | |
247 | */ | |
248 | if (ntohs(ppkt->ip->ip_off) & IP_DF) { | |
249 | spkt->ip->ip_id = ppkt->ip->ip_id; | |
250 | /* and the sum will be different if the IDs were different */ | |
251 | spkt->ip->ip_sum = ppkt->ip->ip_sum; | |
252 | } | |
253 | ||
184d4d42 ZC |
254 | /* |
255 | * Check tcp header length for tcp option field. | |
256 | * th_off > 5 means this tcp packet have options field. | |
257 | * The tcp options maybe always different. | |
258 | * for example: | |
259 | * From RFC 7323. | |
260 | * TCP Timestamps option (TSopt): | |
261 | * Kind: 8 | |
262 | * | |
263 | * Length: 10 bytes | |
264 | * | |
265 | * +-------+-------+---------------------+---------------------+ | |
266 | * |Kind=8 | 10 | TS Value (TSval) |TS Echo Reply (TSecr)| | |
267 | * +-------+-------+---------------------+---------------------+ | |
268 | * 1 1 4 4 | |
269 | * | |
270 | * In this case the primary guest's timestamp always different with | |
271 | * the secondary guest's timestamp. COLO just focus on payload, | |
272 | * so we just need skip this field. | |
273 | */ | |
274 | if (ptcp->th_off > 5) { | |
6f5009c3 ZC |
275 | ptrdiff_t ptcp_offset, stcp_offset; |
276 | ||
277 | ptcp_offset = ppkt->transport_header - (uint8_t *)ppkt->data | |
278 | + (ptcp->th_off * 4) - ppkt->vnet_hdr_len; | |
279 | stcp_offset = spkt->transport_header - (uint8_t *)spkt->data | |
280 | + (stcp->th_off * 4) - spkt->vnet_hdr_len; | |
281 | ||
282 | /* | |
283 | * When network is busy, some tcp options(like sack) will unpredictable | |
284 | * occur in primary side or secondary side. it will make packet size | |
285 | * not same, but the two packet's payload is identical. colo just | |
286 | * care about packet payload, so we skip the option field. | |
287 | */ | |
288 | res = colo_packet_compare_common(ppkt, spkt, ptcp_offset, stcp_offset); | |
184d4d42 | 289 | } else if (ptcp->th_sum == stcp->th_sum) { |
6f5009c3 | 290 | res = colo_packet_compare_common(ppkt, spkt, ETH_HLEN, ETH_HLEN); |
6efeb328 ZC |
291 | } else { |
292 | res = -1; | |
293 | } | |
f4b61836 | 294 | |
d87aa138 SH |
295 | if (res != 0 && |
296 | trace_event_get_state_backends(TRACE_COLO_COMPARE_MISCOMPARE)) { | |
f583dca9 ZC |
297 | char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20]; |
298 | ||
299 | strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src)); | |
300 | strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst)); | |
301 | strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src)); | |
302 | strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst)); | |
303 | ||
304 | trace_colo_compare_ip_info(ppkt->size, pri_ip_src, | |
305 | pri_ip_dst, spkt->size, | |
306 | sec_ip_src, sec_ip_dst); | |
307 | ||
308 | trace_colo_compare_tcp_info("pri tcp packet", | |
309 | ntohl(ptcp->th_seq), | |
310 | ntohl(ptcp->th_ack), | |
311 | res, ptcp->th_flags, | |
312 | ppkt->size); | |
313 | ||
314 | trace_colo_compare_tcp_info("sec tcp packet", | |
315 | ntohl(stcp->th_seq), | |
316 | ntohl(stcp->th_ack), | |
317 | res, stcp->th_flags, | |
318 | spkt->size); | |
2061c14c ZC |
319 | |
320 | qemu_hexdump((char *)ppkt->data, stderr, | |
321 | "colo-compare ppkt", ppkt->size); | |
322 | qemu_hexdump((char *)spkt->data, stderr, | |
323 | "colo-compare spkt", spkt->size); | |
f4b61836 ZC |
324 | } |
325 | ||
326 | return res; | |
327 | } | |
328 | ||
329 | /* | |
330 | * Called from the compare thread on the primary | |
331 | * for compare udp packet | |
332 | */ | |
333 | static int colo_packet_compare_udp(Packet *spkt, Packet *ppkt) | |
334 | { | |
335 | int ret; | |
6efeb328 | 336 | int network_header_length = ppkt->ip->ip_hl * 4; |
f4b61836 ZC |
337 | |
338 | trace_colo_compare_main("compare udp"); | |
2ad7ca4c | 339 | |
6efeb328 ZC |
340 | /* |
341 | * Because of ppkt and spkt are both in the same connection, | |
342 | * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are | |
343 | * same with spkt. In addition, IP header's Identification is a random | |
344 | * field, we can handle it in IP fragmentation function later. | |
345 | * COLO just concern the response net packet payload from primary guest | |
346 | * and secondary guest are same or not, So we ignored all IP header include | |
347 | * other field like TOS,TTL,IP Checksum. we only need to compare | |
348 | * the ip payload here. | |
349 | */ | |
350 | ret = colo_packet_compare_common(ppkt, spkt, | |
6f5009c3 | 351 | network_header_length + ETH_HLEN, |
6efeb328 | 352 | network_header_length + ETH_HLEN); |
f4b61836 ZC |
353 | |
354 | if (ret) { | |
355 | trace_colo_compare_udp_miscompare("primary pkt size", ppkt->size); | |
f4b61836 | 356 | trace_colo_compare_udp_miscompare("Secondary pkt size", spkt->size); |
d87aa138 | 357 | if (trace_event_get_state_backends(TRACE_COLO_COMPARE_MISCOMPARE)) { |
1723a7f7 ZC |
358 | qemu_hexdump((char *)ppkt->data, stderr, "colo-compare pri pkt", |
359 | ppkt->size); | |
360 | qemu_hexdump((char *)spkt->data, stderr, "colo-compare sec pkt", | |
361 | spkt->size); | |
362 | } | |
f4b61836 ZC |
363 | } |
364 | ||
365 | return ret; | |
366 | } | |
367 | ||
368 | /* | |
369 | * Called from the compare thread on the primary | |
370 | * for compare icmp packet | |
371 | */ | |
372 | static int colo_packet_compare_icmp(Packet *spkt, Packet *ppkt) | |
373 | { | |
6efeb328 ZC |
374 | int network_header_length = ppkt->ip->ip_hl * 4; |
375 | ||
f4b61836 | 376 | trace_colo_compare_main("compare icmp"); |
f4b61836 | 377 | |
6efeb328 ZC |
378 | /* |
379 | * Because of ppkt and spkt are both in the same connection, | |
380 | * The ppkt's src ip, dst ip, src port, dst port, ip_proto all are | |
381 | * same with spkt. In addition, IP header's Identification is a random | |
382 | * field, we can handle it in IP fragmentation function later. | |
383 | * COLO just concern the response net packet payload from primary guest | |
384 | * and secondary guest are same or not, So we ignored all IP header include | |
385 | * other field like TOS,TTL,IP Checksum. we only need to compare | |
386 | * the ip payload here. | |
387 | */ | |
388 | if (colo_packet_compare_common(ppkt, spkt, | |
6f5009c3 | 389 | network_header_length + ETH_HLEN, |
6efeb328 | 390 | network_header_length + ETH_HLEN)) { |
f4b61836 ZC |
391 | trace_colo_compare_icmp_miscompare("primary pkt size", |
392 | ppkt->size); | |
f4b61836 ZC |
393 | trace_colo_compare_icmp_miscompare("Secondary pkt size", |
394 | spkt->size); | |
d87aa138 | 395 | if (trace_event_get_state_backends(TRACE_COLO_COMPARE_MISCOMPARE)) { |
1723a7f7 ZC |
396 | qemu_hexdump((char *)ppkt->data, stderr, "colo-compare pri pkt", |
397 | ppkt->size); | |
398 | qemu_hexdump((char *)spkt->data, stderr, "colo-compare sec pkt", | |
399 | spkt->size); | |
400 | } | |
f4b61836 ZC |
401 | return -1; |
402 | } else { | |
403 | return 0; | |
404 | } | |
405 | } | |
406 | ||
407 | /* | |
408 | * Called from the compare thread on the primary | |
409 | * for compare other packet | |
410 | */ | |
411 | static int colo_packet_compare_other(Packet *spkt, Packet *ppkt) | |
412 | { | |
413 | trace_colo_compare_main("compare other"); | |
d87aa138 | 414 | if (trace_event_get_state_backends(TRACE_COLO_COMPARE_MISCOMPARE)) { |
e630b2bf ZC |
415 | char pri_ip_src[20], pri_ip_dst[20], sec_ip_src[20], sec_ip_dst[20]; |
416 | ||
417 | strcpy(pri_ip_src, inet_ntoa(ppkt->ip->ip_src)); | |
418 | strcpy(pri_ip_dst, inet_ntoa(ppkt->ip->ip_dst)); | |
419 | strcpy(sec_ip_src, inet_ntoa(spkt->ip->ip_src)); | |
420 | strcpy(sec_ip_dst, inet_ntoa(spkt->ip->ip_dst)); | |
421 | ||
422 | trace_colo_compare_ip_info(ppkt->size, pri_ip_src, | |
423 | pri_ip_dst, spkt->size, | |
424 | sec_ip_src, sec_ip_dst); | |
425 | } | |
426 | ||
6f5009c3 | 427 | return colo_packet_compare_common(ppkt, spkt, 0, 0); |
0682e15b ZC |
428 | } |
429 | ||
430 | static int colo_old_packet_check_one(Packet *pkt, int64_t *check_time) | |
431 | { | |
432 | int64_t now = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
433 | ||
434 | if ((now - pkt->creation_ms) > (*check_time)) { | |
435 | trace_colo_old_packet_check_found(pkt->creation_ms); | |
436 | return 0; | |
437 | } else { | |
438 | return 1; | |
439 | } | |
440 | } | |
441 | ||
d25a7dab ZC |
442 | static int colo_old_packet_check_one_conn(Connection *conn, |
443 | void *user_data) | |
0682e15b | 444 | { |
0682e15b ZC |
445 | GList *result = NULL; |
446 | int64_t check_time = REGULAR_PACKET_CHECK_MS; | |
447 | ||
448 | result = g_queue_find_custom(&conn->primary_list, | |
449 | &check_time, | |
450 | (GCompareFunc)colo_old_packet_check_one); | |
451 | ||
452 | if (result) { | |
61c5f469 ZC |
453 | /* Do checkpoint will flush old packet */ |
454 | /* | |
455 | * TODO: Notify colo frame to do checkpoint. | |
456 | * colo_compare_inconsistent_notify(); | |
457 | */ | |
d25a7dab | 458 | return 0; |
0682e15b | 459 | } |
d25a7dab ZC |
460 | |
461 | return 1; | |
0682e15b ZC |
462 | } |
463 | ||
464 | /* | |
465 | * Look for old packets that the secondary hasn't matched, | |
466 | * if we have some then we have to checkpoint to wake | |
467 | * the secondary up. | |
468 | */ | |
469 | static void colo_old_packet_check(void *opaque) | |
470 | { | |
471 | CompareState *s = opaque; | |
472 | ||
d25a7dab ZC |
473 | /* |
474 | * If we find one old packet, stop finding job and notify | |
475 | * COLO frame do checkpoint. | |
476 | */ | |
477 | g_queue_find_custom(&s->conn_list, NULL, | |
478 | (GCompareFunc)colo_old_packet_check_one_conn); | |
0682e15b ZC |
479 | } |
480 | ||
481 | /* | |
482 | * Called from the compare thread on the primary | |
483 | * for compare connection | |
484 | */ | |
485 | static void colo_compare_connection(void *opaque, void *user_data) | |
486 | { | |
487 | CompareState *s = user_data; | |
488 | Connection *conn = opaque; | |
489 | Packet *pkt = NULL; | |
490 | GList *result = NULL; | |
491 | int ret; | |
492 | ||
493 | while (!g_queue_is_empty(&conn->primary_list) && | |
494 | !g_queue_is_empty(&conn->secondary_list)) { | |
626bba98 | 495 | pkt = g_queue_pop_head(&conn->primary_list); |
f4b61836 ZC |
496 | switch (conn->ip_proto) { |
497 | case IPPROTO_TCP: | |
498 | result = g_queue_find_custom(&conn->secondary_list, | |
499 | pkt, (GCompareFunc)colo_packet_compare_tcp); | |
500 | break; | |
501 | case IPPROTO_UDP: | |
502 | result = g_queue_find_custom(&conn->secondary_list, | |
503 | pkt, (GCompareFunc)colo_packet_compare_udp); | |
504 | break; | |
505 | case IPPROTO_ICMP: | |
506 | result = g_queue_find_custom(&conn->secondary_list, | |
507 | pkt, (GCompareFunc)colo_packet_compare_icmp); | |
508 | break; | |
509 | default: | |
510 | result = g_queue_find_custom(&conn->secondary_list, | |
511 | pkt, (GCompareFunc)colo_packet_compare_other); | |
512 | break; | |
513 | } | |
0682e15b ZC |
514 | |
515 | if (result) { | |
aa3a7032 ZC |
516 | ret = compare_chr_send(s, |
517 | pkt->data, | |
518 | pkt->size, | |
519 | pkt->vnet_hdr_len); | |
0682e15b ZC |
520 | if (ret < 0) { |
521 | error_report("colo_send_primary_packet failed"); | |
522 | } | |
523 | trace_colo_compare_main("packet same and release packet"); | |
524 | g_queue_remove(&conn->secondary_list, result->data); | |
525 | packet_destroy(pkt, NULL); | |
526 | } else { | |
527 | /* | |
528 | * If one packet arrive late, the secondary_list or | |
529 | * primary_list will be empty, so we can't compare it | |
530 | * until next comparison. | |
531 | */ | |
532 | trace_colo_compare_main("packet different"); | |
626bba98 | 533 | g_queue_push_head(&conn->primary_list, pkt); |
0682e15b ZC |
534 | /* TODO: colo_notify_checkpoint();*/ |
535 | break; | |
536 | } | |
537 | } | |
538 | } | |
539 | ||
3037e7a5 | 540 | static int compare_chr_send(CompareState *s, |
59509ec1 | 541 | const uint8_t *buf, |
aa3a7032 ZC |
542 | uint32_t size, |
543 | uint32_t vnet_hdr_len) | |
59509ec1 ZC |
544 | { |
545 | int ret = 0; | |
546 | uint32_t len = htonl(size); | |
547 | ||
548 | if (!size) { | |
549 | return 0; | |
550 | } | |
551 | ||
3037e7a5 | 552 | ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)&len, sizeof(len)); |
59509ec1 ZC |
553 | if (ret != sizeof(len)) { |
554 | goto err; | |
555 | } | |
556 | ||
aa3a7032 ZC |
557 | if (s->vnet_hdr) { |
558 | /* | |
559 | * We send vnet header len make other module(like filter-redirector) | |
560 | * know how to parse net packet correctly. | |
561 | */ | |
562 | len = htonl(vnet_hdr_len); | |
563 | ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)&len, sizeof(len)); | |
564 | if (ret != sizeof(len)) { | |
565 | goto err; | |
566 | } | |
567 | } | |
568 | ||
3037e7a5 | 569 | ret = qemu_chr_fe_write_all(&s->chr_out, (uint8_t *)buf, size); |
59509ec1 ZC |
570 | if (ret != size) { |
571 | goto err; | |
572 | } | |
573 | ||
574 | return 0; | |
575 | ||
576 | err: | |
577 | return ret < 0 ? ret : -EIO; | |
578 | } | |
579 | ||
0682e15b ZC |
580 | static int compare_chr_can_read(void *opaque) |
581 | { | |
582 | return COMPARE_READ_LEN_MAX; | |
583 | } | |
584 | ||
585 | /* | |
586 | * Called from the main thread on the primary for packets | |
587 | * arriving over the socket from the primary. | |
588 | */ | |
589 | static void compare_pri_chr_in(void *opaque, const uint8_t *buf, int size) | |
590 | { | |
591 | CompareState *s = COLO_COMPARE(opaque); | |
592 | int ret; | |
593 | ||
594 | ret = net_fill_rstate(&s->pri_rs, buf, size); | |
595 | if (ret == -1) { | |
81517ba3 | 596 | qemu_chr_fe_set_handlers(&s->chr_pri_in, NULL, NULL, NULL, NULL, |
39ab61c6 | 597 | NULL, NULL, true); |
0682e15b ZC |
598 | error_report("colo-compare primary_in error"); |
599 | } | |
600 | } | |
601 | ||
602 | /* | |
603 | * Called from the main thread on the primary for packets | |
604 | * arriving over the socket from the secondary. | |
605 | */ | |
606 | static void compare_sec_chr_in(void *opaque, const uint8_t *buf, int size) | |
607 | { | |
608 | CompareState *s = COLO_COMPARE(opaque); | |
609 | int ret; | |
610 | ||
611 | ret = net_fill_rstate(&s->sec_rs, buf, size); | |
612 | if (ret == -1) { | |
81517ba3 | 613 | qemu_chr_fe_set_handlers(&s->chr_sec_in, NULL, NULL, NULL, NULL, |
39ab61c6 | 614 | NULL, NULL, true); |
0682e15b ZC |
615 | error_report("colo-compare secondary_in error"); |
616 | } | |
617 | } | |
618 | ||
66d2a242 HZ |
619 | /* |
620 | * Check old packet regularly so it can watch for any packets | |
621 | * that the secondary hasn't produced equivalents of. | |
622 | */ | |
dd321ecf | 623 | static void check_old_packet_regular(void *opaque) |
66d2a242 HZ |
624 | { |
625 | CompareState *s = opaque; | |
626 | ||
627 | /* if have old packet we will notify checkpoint */ | |
628 | colo_old_packet_check(s); | |
dd321ecf WY |
629 | timer_mod(s->packet_check_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + |
630 | REGULAR_PACKET_CHECK_MS); | |
631 | } | |
632 | ||
633 | static void colo_compare_timer_init(CompareState *s) | |
634 | { | |
635 | AioContext *ctx = iothread_get_aio_context(s->iothread); | |
66d2a242 | 636 | |
dd321ecf WY |
637 | s->packet_check_timer = aio_timer_new(ctx, QEMU_CLOCK_VIRTUAL, |
638 | SCALE_MS, check_old_packet_regular, | |
639 | s); | |
640 | timer_mod(s->packet_check_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + | |
641 | REGULAR_PACKET_CHECK_MS); | |
66d2a242 HZ |
642 | } |
643 | ||
dd321ecf | 644 | static void colo_compare_timer_del(CompareState *s) |
0682e15b | 645 | { |
dd321ecf WY |
646 | if (s->packet_check_timer) { |
647 | timer_del(s->packet_check_timer); | |
648 | timer_free(s->packet_check_timer); | |
649 | s->packet_check_timer = NULL; | |
650 | } | |
651 | } | |
0682e15b | 652 | |
dd321ecf WY |
653 | static void colo_compare_iothread(CompareState *s) |
654 | { | |
655 | object_ref(OBJECT(s->iothread)); | |
656 | s->worker_context = iothread_get_g_main_context(s->iothread); | |
0682e15b | 657 | |
5345fdb4 | 658 | qemu_chr_fe_set_handlers(&s->chr_pri_in, compare_chr_can_read, |
81517ba3 AN |
659 | compare_pri_chr_in, NULL, NULL, |
660 | s, s->worker_context, true); | |
5345fdb4 | 661 | qemu_chr_fe_set_handlers(&s->chr_sec_in, compare_chr_can_read, |
81517ba3 AN |
662 | compare_sec_chr_in, NULL, NULL, |
663 | s, s->worker_context, true); | |
0682e15b | 664 | |
dd321ecf | 665 | colo_compare_timer_init(s); |
0682e15b ZC |
666 | } |
667 | ||
7dce4e6f ZC |
668 | static char *compare_get_pri_indev(Object *obj, Error **errp) |
669 | { | |
670 | CompareState *s = COLO_COMPARE(obj); | |
671 | ||
672 | return g_strdup(s->pri_indev); | |
673 | } | |
674 | ||
675 | static void compare_set_pri_indev(Object *obj, const char *value, Error **errp) | |
676 | { | |
677 | CompareState *s = COLO_COMPARE(obj); | |
678 | ||
679 | g_free(s->pri_indev); | |
680 | s->pri_indev = g_strdup(value); | |
681 | } | |
682 | ||
683 | static char *compare_get_sec_indev(Object *obj, Error **errp) | |
684 | { | |
685 | CompareState *s = COLO_COMPARE(obj); | |
686 | ||
687 | return g_strdup(s->sec_indev); | |
688 | } | |
689 | ||
690 | static void compare_set_sec_indev(Object *obj, const char *value, Error **errp) | |
691 | { | |
692 | CompareState *s = COLO_COMPARE(obj); | |
693 | ||
694 | g_free(s->sec_indev); | |
695 | s->sec_indev = g_strdup(value); | |
696 | } | |
697 | ||
698 | static char *compare_get_outdev(Object *obj, Error **errp) | |
699 | { | |
700 | CompareState *s = COLO_COMPARE(obj); | |
701 | ||
702 | return g_strdup(s->outdev); | |
703 | } | |
704 | ||
705 | static void compare_set_outdev(Object *obj, const char *value, Error **errp) | |
706 | { | |
707 | CompareState *s = COLO_COMPARE(obj); | |
708 | ||
709 | g_free(s->outdev); | |
710 | s->outdev = g_strdup(value); | |
711 | } | |
712 | ||
aa3a7032 ZC |
713 | static bool compare_get_vnet_hdr(Object *obj, Error **errp) |
714 | { | |
715 | CompareState *s = COLO_COMPARE(obj); | |
716 | ||
717 | return s->vnet_hdr; | |
718 | } | |
719 | ||
720 | static void compare_set_vnet_hdr(Object *obj, | |
721 | bool value, | |
722 | Error **errp) | |
723 | { | |
724 | CompareState *s = COLO_COMPARE(obj); | |
725 | ||
726 | s->vnet_hdr = value; | |
727 | } | |
728 | ||
7dce4e6f ZC |
729 | static void compare_pri_rs_finalize(SocketReadState *pri_rs) |
730 | { | |
59509ec1 | 731 | CompareState *s = container_of(pri_rs, CompareState, pri_rs); |
8ec14402 | 732 | Connection *conn = NULL; |
59509ec1 | 733 | |
8ec14402 | 734 | if (packet_enqueue(s, PRIMARY_IN, &conn)) { |
59509ec1 | 735 | trace_colo_compare_main("primary: unsupported packet in"); |
aa3a7032 ZC |
736 | compare_chr_send(s, |
737 | pri_rs->buf, | |
738 | pri_rs->packet_len, | |
739 | pri_rs->vnet_hdr_len); | |
0682e15b ZC |
740 | } else { |
741 | /* compare connection */ | |
8ec14402 | 742 | colo_compare_connection(conn, s); |
59509ec1 | 743 | } |
7dce4e6f ZC |
744 | } |
745 | ||
746 | static void compare_sec_rs_finalize(SocketReadState *sec_rs) | |
747 | { | |
59509ec1 | 748 | CompareState *s = container_of(sec_rs, CompareState, sec_rs); |
8ec14402 | 749 | Connection *conn = NULL; |
59509ec1 | 750 | |
8ec14402 | 751 | if (packet_enqueue(s, SECONDARY_IN, &conn)) { |
59509ec1 | 752 | trace_colo_compare_main("secondary: unsupported packet in"); |
0682e15b ZC |
753 | } else { |
754 | /* compare connection */ | |
8ec14402 | 755 | colo_compare_connection(conn, s); |
59509ec1 | 756 | } |
7dce4e6f ZC |
757 | } |
758 | ||
7dce4e6f ZC |
759 | |
760 | /* | |
761 | * Return 0 is success. | |
762 | * Return 1 is failed. | |
763 | */ | |
0ec7b3e7 | 764 | static int find_and_check_chardev(Chardev **chr, |
7dce4e6f ZC |
765 | char *chr_name, |
766 | Error **errp) | |
767 | { | |
7dce4e6f ZC |
768 | *chr = qemu_chr_find(chr_name); |
769 | if (*chr == NULL) { | |
770 | error_setg(errp, "Device '%s' not found", | |
771 | chr_name); | |
772 | return 1; | |
773 | } | |
774 | ||
0a73336d DB |
775 | if (!qemu_chr_has_feature(*chr, QEMU_CHAR_FEATURE_RECONNECTABLE)) { |
776 | error_setg(errp, "chardev \"%s\" is not reconnectable", | |
7dce4e6f ZC |
777 | chr_name); |
778 | return 1; | |
779 | } | |
fbf3cc3a | 780 | |
7dce4e6f ZC |
781 | return 0; |
782 | } | |
783 | ||
784 | /* | |
785 | * Called from the main thread on the primary | |
786 | * to setup colo-compare. | |
787 | */ | |
788 | static void colo_compare_complete(UserCreatable *uc, Error **errp) | |
789 | { | |
790 | CompareState *s = COLO_COMPARE(uc); | |
0ec7b3e7 | 791 | Chardev *chr; |
7dce4e6f | 792 | |
dd321ecf | 793 | if (!s->pri_indev || !s->sec_indev || !s->outdev || !s->iothread) { |
7dce4e6f | 794 | error_setg(errp, "colo compare needs 'primary_in' ," |
dd321ecf | 795 | "'secondary_in','outdev','iothread' property set"); |
7dce4e6f ZC |
796 | return; |
797 | } else if (!strcmp(s->pri_indev, s->outdev) || | |
798 | !strcmp(s->sec_indev, s->outdev) || | |
799 | !strcmp(s->pri_indev, s->sec_indev)) { | |
800 | error_setg(errp, "'indev' and 'outdev' could not be same " | |
801 | "for compare module"); | |
802 | return; | |
803 | } | |
804 | ||
5345fdb4 MAL |
805 | if (find_and_check_chardev(&chr, s->pri_indev, errp) || |
806 | !qemu_chr_fe_init(&s->chr_pri_in, chr, errp)) { | |
7dce4e6f ZC |
807 | return; |
808 | } | |
809 | ||
5345fdb4 MAL |
810 | if (find_and_check_chardev(&chr, s->sec_indev, errp) || |
811 | !qemu_chr_fe_init(&s->chr_sec_in, chr, errp)) { | |
7dce4e6f ZC |
812 | return; |
813 | } | |
814 | ||
5345fdb4 MAL |
815 | if (find_and_check_chardev(&chr, s->outdev, errp) || |
816 | !qemu_chr_fe_init(&s->chr_out, chr, errp)) { | |
7dce4e6f ZC |
817 | return; |
818 | } | |
819 | ||
aa3a7032 ZC |
820 | net_socket_rs_init(&s->pri_rs, compare_pri_rs_finalize, s->vnet_hdr); |
821 | net_socket_rs_init(&s->sec_rs, compare_sec_rs_finalize, s->vnet_hdr); | |
7dce4e6f | 822 | |
b6540d40 ZC |
823 | g_queue_init(&s->conn_list); |
824 | ||
825 | s->connection_track_table = g_hash_table_new_full(connection_key_hash, | |
826 | connection_key_equal, | |
827 | g_free, | |
828 | connection_destroy); | |
59509ec1 | 829 | |
dd321ecf | 830 | colo_compare_iothread(s); |
7dce4e6f ZC |
831 | return; |
832 | } | |
833 | ||
dfd917a9 HZ |
834 | static void colo_flush_packets(void *opaque, void *user_data) |
835 | { | |
836 | CompareState *s = user_data; | |
837 | Connection *conn = opaque; | |
838 | Packet *pkt = NULL; | |
839 | ||
840 | while (!g_queue_is_empty(&conn->primary_list)) { | |
841 | pkt = g_queue_pop_head(&conn->primary_list); | |
aa3a7032 ZC |
842 | compare_chr_send(s, |
843 | pkt->data, | |
844 | pkt->size, | |
845 | pkt->vnet_hdr_len); | |
dfd917a9 HZ |
846 | packet_destroy(pkt, NULL); |
847 | } | |
848 | while (!g_queue_is_empty(&conn->secondary_list)) { | |
849 | pkt = g_queue_pop_head(&conn->secondary_list); | |
850 | packet_destroy(pkt, NULL); | |
851 | } | |
852 | } | |
853 | ||
7dce4e6f ZC |
854 | static void colo_compare_class_init(ObjectClass *oc, void *data) |
855 | { | |
856 | UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc); | |
857 | ||
858 | ucc->complete = colo_compare_complete; | |
859 | } | |
860 | ||
861 | static void colo_compare_init(Object *obj) | |
862 | { | |
aa3a7032 ZC |
863 | CompareState *s = COLO_COMPARE(obj); |
864 | ||
7dce4e6f ZC |
865 | object_property_add_str(obj, "primary_in", |
866 | compare_get_pri_indev, compare_set_pri_indev, | |
867 | NULL); | |
868 | object_property_add_str(obj, "secondary_in", | |
869 | compare_get_sec_indev, compare_set_sec_indev, | |
870 | NULL); | |
871 | object_property_add_str(obj, "outdev", | |
872 | compare_get_outdev, compare_set_outdev, | |
873 | NULL); | |
dd321ecf WY |
874 | object_property_add_link(obj, "iothread", TYPE_IOTHREAD, |
875 | (Object **)&s->iothread, | |
876 | object_property_allow_set_link, | |
877 | OBJ_PROP_LINK_UNREF_ON_RELEASE, NULL); | |
aa3a7032 ZC |
878 | |
879 | s->vnet_hdr = false; | |
880 | object_property_add_bool(obj, "vnet_hdr_support", compare_get_vnet_hdr, | |
881 | compare_set_vnet_hdr, NULL); | |
7dce4e6f ZC |
882 | } |
883 | ||
884 | static void colo_compare_finalize(Object *obj) | |
885 | { | |
886 | CompareState *s = COLO_COMPARE(obj); | |
887 | ||
1ce2610c MAL |
888 | qemu_chr_fe_deinit(&s->chr_pri_in, false); |
889 | qemu_chr_fe_deinit(&s->chr_sec_in, false); | |
890 | qemu_chr_fe_deinit(&s->chr_out, false); | |
dd321ecf WY |
891 | if (s->iothread) { |
892 | colo_compare_timer_del(s); | |
893 | } | |
dfd917a9 HZ |
894 | /* Release all unhandled packets after compare thead exited */ |
895 | g_queue_foreach(&s->conn_list, colo_flush_packets, s); | |
896 | ||
727c2d76 | 897 | g_queue_clear(&s->conn_list); |
0682e15b | 898 | |
dd321ecf WY |
899 | if (s->connection_track_table) { |
900 | g_hash_table_destroy(s->connection_track_table); | |
901 | } | |
902 | ||
903 | if (s->iothread) { | |
904 | object_unref(OBJECT(s->iothread)); | |
905 | } | |
7dce4e6f ZC |
906 | g_free(s->pri_indev); |
907 | g_free(s->sec_indev); | |
908 | g_free(s->outdev); | |
909 | } | |
910 | ||
911 | static const TypeInfo colo_compare_info = { | |
912 | .name = TYPE_COLO_COMPARE, | |
913 | .parent = TYPE_OBJECT, | |
914 | .instance_size = sizeof(CompareState), | |
915 | .instance_init = colo_compare_init, | |
916 | .instance_finalize = colo_compare_finalize, | |
917 | .class_size = sizeof(CompareClass), | |
918 | .class_init = colo_compare_class_init, | |
919 | .interfaces = (InterfaceInfo[]) { | |
920 | { TYPE_USER_CREATABLE }, | |
921 | { } | |
922 | } | |
923 | }; | |
924 | ||
925 | static void register_types(void) | |
926 | { | |
927 | type_register_static(&colo_compare_info); | |
928 | } | |
929 | ||
930 | type_init(register_types); |