]>
Commit | Line | Data |
---|---|---|
35a6ed4f HZ |
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 | * This work is licensed under the terms of the GNU GPL, version 2 or | |
10 | * later. See the COPYING file in the top-level directory. | |
11 | */ | |
12 | ||
13 | #include "qemu/osdep.h" | |
0b827d5e | 14 | #include "sysemu/sysemu.h" |
e688df6b | 15 | #include "qapi/error.h" |
9af23989 | 16 | #include "qapi/qapi-commands-migration.h" |
40014d81 | 17 | #include "qemu-file-channel.h" |
6666c96a | 18 | #include "migration.h" |
08a0aee1 | 19 | #include "qemu-file.h" |
20a519a0 | 20 | #include "savevm.h" |
35a6ed4f | 21 | #include "migration/colo.h" |
2c9e6fec | 22 | #include "block.h" |
a91246c9 | 23 | #include "io/channel-buffer.h" |
0b827d5e | 24 | #include "trace.h" |
56ba83d2 | 25 | #include "qemu/error-report.h" |
d89e666e | 26 | #include "migration/failover.h" |
335d10cd | 27 | #ifdef CONFIG_REPLICATION |
2c9639ec | 28 | #include "replication.h" |
335d10cd | 29 | #endif |
131b2153 ZC |
30 | #include "net/colo-compare.h" |
31 | #include "net/colo.h" | |
8e48ac95 | 32 | #include "block/block.h" |
9ecff6d6 | 33 | #include "qapi/qapi-events-migration.h" |
f56c0065 | 34 | #include "qapi/qmp/qerror.h" |
3f6df99d | 35 | #include "sysemu/cpus.h" |
7b343530 | 36 | #include "net/filter.h" |
35a6ed4f | 37 | |
a8664ba5 | 38 | static bool vmstate_loading; |
131b2153 | 39 | static Notifier packets_compare_notifier; |
a8664ba5 | 40 | |
5ed0deca ZC |
41 | /* User need to know colo mode after COLO failover */ |
42 | static COLOMode last_colo_mode; | |
43 | ||
a91246c9 HZ |
44 | #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024) |
45 | ||
0b827d5e HZ |
46 | bool migration_in_colo_state(void) |
47 | { | |
48 | MigrationState *s = migrate_get_current(); | |
49 | ||
50 | return (s->state == MIGRATION_STATUS_COLO); | |
51 | } | |
52 | ||
25d0c16f HZ |
53 | bool migration_incoming_in_colo_state(void) |
54 | { | |
55 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
56 | ||
57 | return mis && (mis->state == MIGRATION_STATUS_COLO); | |
58 | } | |
59 | ||
b3f7f0c5 HZ |
60 | static bool colo_runstate_is_stopped(void) |
61 | { | |
62 | return runstate_check(RUN_STATE_COLO) || !runstate_is_running(); | |
63 | } | |
64 | ||
9d2db376 HZ |
65 | static void secondary_vm_do_failover(void) |
66 | { | |
3ebb9c4f ZC |
67 | /* COLO needs enable block-replication */ |
68 | #ifdef CONFIG_REPLICATION | |
9d2db376 HZ |
69 | int old_state; |
70 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
8e48ac95 | 71 | Error *local_err = NULL; |
9d2db376 | 72 | |
a8664ba5 HZ |
73 | /* Can not do failover during the process of VM's loading VMstate, Or |
74 | * it will break the secondary VM. | |
75 | */ | |
76 | if (vmstate_loading) { | |
77 | old_state = failover_set_state(FAILOVER_STATUS_ACTIVE, | |
78 | FAILOVER_STATUS_RELAUNCH); | |
79 | if (old_state != FAILOVER_STATUS_ACTIVE) { | |
80 | error_report("Unknown error while do failover for secondary VM," | |
977c736f | 81 | "old_state: %s", FailoverStatus_str(old_state)); |
a8664ba5 HZ |
82 | } |
83 | return; | |
84 | } | |
85 | ||
9d2db376 HZ |
86 | migrate_set_state(&mis->state, MIGRATION_STATUS_COLO, |
87 | MIGRATION_STATUS_COMPLETED); | |
88 | ||
8e48ac95 ZC |
89 | replication_stop_all(true, &local_err); |
90 | if (local_err) { | |
91 | error_report_err(local_err); | |
92 | } | |
93 | ||
7b343530 HZ |
94 | /* Notify all filters of all NIC to do checkpoint */ |
95 | colo_notify_filters_event(COLO_EVENT_FAILOVER, &local_err); | |
96 | if (local_err) { | |
97 | error_report_err(local_err); | |
98 | } | |
99 | ||
9d2db376 HZ |
100 | if (!autostart) { |
101 | error_report("\"-S\" qemu option will be ignored in secondary side"); | |
102 | /* recover runstate to normal migration finish state */ | |
103 | autostart = true; | |
104 | } | |
c937b9a6 HZ |
105 | /* |
106 | * Make sure COLO incoming thread not block in recv or send, | |
107 | * If mis->from_src_file and mis->to_src_file use the same fd, | |
108 | * The second shutdown() will return -1, we ignore this value, | |
109 | * It is harmless. | |
110 | */ | |
111 | if (mis->from_src_file) { | |
112 | qemu_file_shutdown(mis->from_src_file); | |
113 | } | |
114 | if (mis->to_src_file) { | |
115 | qemu_file_shutdown(mis->to_src_file); | |
116 | } | |
9d2db376 HZ |
117 | |
118 | old_state = failover_set_state(FAILOVER_STATUS_ACTIVE, | |
119 | FAILOVER_STATUS_COMPLETED); | |
120 | if (old_state != FAILOVER_STATUS_ACTIVE) { | |
121 | error_report("Incorrect state (%s) while doing failover for " | |
977c736f | 122 | "secondary VM", FailoverStatus_str(old_state)); |
9d2db376 HZ |
123 | return; |
124 | } | |
c937b9a6 HZ |
125 | /* Notify COLO incoming thread that failover work is finished */ |
126 | qemu_sem_post(&mis->colo_incoming_sem); | |
1fe6ab26 | 127 | |
9d2db376 HZ |
128 | /* For Secondary VM, jump to incoming co */ |
129 | if (mis->migration_incoming_co) { | |
130 | qemu_coroutine_enter(mis->migration_incoming_co); | |
131 | } | |
3ebb9c4f ZC |
132 | #else |
133 | abort(); | |
134 | #endif | |
9d2db376 HZ |
135 | } |
136 | ||
b3f7f0c5 HZ |
137 | static void primary_vm_do_failover(void) |
138 | { | |
3ebb9c4f | 139 | #ifdef CONFIG_REPLICATION |
b3f7f0c5 HZ |
140 | MigrationState *s = migrate_get_current(); |
141 | int old_state; | |
8e48ac95 | 142 | Error *local_err = NULL; |
b3f7f0c5 HZ |
143 | |
144 | migrate_set_state(&s->state, MIGRATION_STATUS_COLO, | |
145 | MIGRATION_STATUS_COMPLETED); | |
2518aec1 HZ |
146 | /* |
147 | * kick COLO thread which might wait at | |
148 | * qemu_sem_wait(&s->colo_checkpoint_sem). | |
149 | */ | |
150 | colo_checkpoint_notify(migrate_get_current()); | |
b3f7f0c5 | 151 | |
c937b9a6 HZ |
152 | /* |
153 | * Wake up COLO thread which may blocked in recv() or send(), | |
154 | * The s->rp_state.from_dst_file and s->to_dst_file may use the | |
155 | * same fd, but we still shutdown the fd for twice, it is harmless. | |
156 | */ | |
157 | if (s->to_dst_file) { | |
158 | qemu_file_shutdown(s->to_dst_file); | |
159 | } | |
160 | if (s->rp_state.from_dst_file) { | |
161 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
162 | } | |
163 | ||
b3f7f0c5 HZ |
164 | old_state = failover_set_state(FAILOVER_STATUS_ACTIVE, |
165 | FAILOVER_STATUS_COMPLETED); | |
166 | if (old_state != FAILOVER_STATUS_ACTIVE) { | |
167 | error_report("Incorrect state (%s) while doing failover for Primary VM", | |
977c736f | 168 | FailoverStatus_str(old_state)); |
b3f7f0c5 HZ |
169 | return; |
170 | } | |
8e48ac95 ZC |
171 | |
172 | replication_stop_all(true, &local_err); | |
173 | if (local_err) { | |
174 | error_report_err(local_err); | |
175 | local_err = NULL; | |
176 | } | |
177 | ||
c937b9a6 HZ |
178 | /* Notify COLO thread that failover work is finished */ |
179 | qemu_sem_post(&s->colo_exit_sem); | |
3ebb9c4f ZC |
180 | #else |
181 | abort(); | |
182 | #endif | |
b3f7f0c5 HZ |
183 | } |
184 | ||
aad555c2 ZC |
185 | COLOMode get_colo_mode(void) |
186 | { | |
187 | if (migration_in_colo_state()) { | |
188 | return COLO_MODE_PRIMARY; | |
189 | } else if (migration_incoming_in_colo_state()) { | |
190 | return COLO_MODE_SECONDARY; | |
191 | } else { | |
41b6b779 | 192 | return COLO_MODE_NONE; |
aad555c2 ZC |
193 | } |
194 | } | |
195 | ||
b3f7f0c5 HZ |
196 | void colo_do_failover(MigrationState *s) |
197 | { | |
198 | /* Make sure VM stopped while failover happened. */ | |
199 | if (!colo_runstate_is_stopped()) { | |
200 | vm_stop_force_state(RUN_STATE_COLO); | |
201 | } | |
202 | ||
82cd368c ZC |
203 | switch (get_colo_mode()) { |
204 | case COLO_MODE_PRIMARY: | |
b3f7f0c5 | 205 | primary_vm_do_failover(); |
82cd368c ZC |
206 | break; |
207 | case COLO_MODE_SECONDARY: | |
9d2db376 | 208 | secondary_vm_do_failover(); |
82cd368c ZC |
209 | break; |
210 | default: | |
211 | error_report("colo_do_failover failed because the colo mode" | |
212 | " could not be obtained"); | |
b3f7f0c5 HZ |
213 | } |
214 | } | |
215 | ||
335d10cd | 216 | #ifdef CONFIG_REPLICATION |
2c9639ec ZC |
217 | void qmp_xen_set_replication(bool enable, bool primary, |
218 | bool has_failover, bool failover, | |
219 | Error **errp) | |
220 | { | |
221 | ReplicationMode mode = primary ? | |
222 | REPLICATION_MODE_PRIMARY : | |
223 | REPLICATION_MODE_SECONDARY; | |
224 | ||
225 | if (has_failover && enable) { | |
226 | error_setg(errp, "Parameter 'failover' is only for" | |
227 | " stopping replication"); | |
228 | return; | |
229 | } | |
230 | ||
231 | if (enable) { | |
232 | replication_start_all(mode, errp); | |
233 | } else { | |
234 | if (!has_failover) { | |
235 | failover = NULL; | |
236 | } | |
237 | replication_stop_all(failover, failover ? NULL : errp); | |
238 | } | |
239 | } | |
240 | ||
daa33c52 ZC |
241 | ReplicationStatus *qmp_query_xen_replication_status(Error **errp) |
242 | { | |
243 | Error *err = NULL; | |
244 | ReplicationStatus *s = g_new0(ReplicationStatus, 1); | |
245 | ||
246 | replication_get_error_all(&err); | |
247 | if (err) { | |
248 | s->error = true; | |
249 | s->has_desc = true; | |
250 | s->desc = g_strdup(error_get_pretty(err)); | |
251 | } else { | |
252 | s->error = false; | |
253 | } | |
254 | ||
255 | error_free(err); | |
256 | return s; | |
257 | } | |
258 | ||
259 | void qmp_xen_colo_do_checkpoint(Error **errp) | |
260 | { | |
261 | replication_do_checkpoint_all(errp); | |
262 | } | |
335d10cd | 263 | #endif |
daa33c52 | 264 | |
f56c0065 ZC |
265 | COLOStatus *qmp_query_colo_status(Error **errp) |
266 | { | |
267 | COLOStatus *s = g_new0(COLOStatus, 1); | |
268 | ||
269 | s->mode = get_colo_mode(); | |
5ed0deca | 270 | s->last_mode = last_colo_mode; |
f56c0065 ZC |
271 | |
272 | switch (failover_get_state()) { | |
273 | case FAILOVER_STATUS_NONE: | |
274 | s->reason = COLO_EXIT_REASON_NONE; | |
275 | break; | |
1fe6ab26 | 276 | case FAILOVER_STATUS_COMPLETED: |
f56c0065 ZC |
277 | s->reason = COLO_EXIT_REASON_REQUEST; |
278 | break; | |
279 | default: | |
3a43ac47 ZC |
280 | if (migration_in_colo_state()) { |
281 | s->reason = COLO_EXIT_REASON_PROCESSING; | |
282 | } else { | |
283 | s->reason = COLO_EXIT_REASON_ERROR; | |
284 | } | |
f56c0065 ZC |
285 | } |
286 | ||
287 | return s; | |
288 | } | |
289 | ||
4f97558e HZ |
290 | static void colo_send_message(QEMUFile *f, COLOMessage msg, |
291 | Error **errp) | |
292 | { | |
293 | int ret; | |
294 | ||
295 | if (msg >= COLO_MESSAGE__MAX) { | |
296 | error_setg(errp, "%s: Invalid message", __func__); | |
297 | return; | |
298 | } | |
299 | qemu_put_be32(f, msg); | |
300 | qemu_fflush(f); | |
301 | ||
302 | ret = qemu_file_get_error(f); | |
303 | if (ret < 0) { | |
304 | error_setg_errno(errp, -ret, "Can't send COLO message"); | |
305 | } | |
977c736f | 306 | trace_colo_send_message(COLOMessage_str(msg)); |
4f97558e HZ |
307 | } |
308 | ||
a91246c9 HZ |
309 | static void colo_send_message_value(QEMUFile *f, COLOMessage msg, |
310 | uint64_t value, Error **errp) | |
311 | { | |
312 | Error *local_err = NULL; | |
313 | int ret; | |
314 | ||
315 | colo_send_message(f, msg, &local_err); | |
316 | if (local_err) { | |
317 | error_propagate(errp, local_err); | |
318 | return; | |
319 | } | |
320 | qemu_put_be64(f, value); | |
321 | qemu_fflush(f); | |
322 | ||
323 | ret = qemu_file_get_error(f); | |
324 | if (ret < 0) { | |
325 | error_setg_errno(errp, -ret, "Failed to send value for message:%s", | |
977c736f | 326 | COLOMessage_str(msg)); |
a91246c9 HZ |
327 | } |
328 | } | |
329 | ||
4f97558e HZ |
330 | static COLOMessage colo_receive_message(QEMUFile *f, Error **errp) |
331 | { | |
332 | COLOMessage msg; | |
333 | int ret; | |
334 | ||
335 | msg = qemu_get_be32(f); | |
336 | ret = qemu_file_get_error(f); | |
337 | if (ret < 0) { | |
338 | error_setg_errno(errp, -ret, "Can't receive COLO message"); | |
339 | return msg; | |
340 | } | |
341 | if (msg >= COLO_MESSAGE__MAX) { | |
342 | error_setg(errp, "%s: Invalid message", __func__); | |
343 | return msg; | |
344 | } | |
977c736f | 345 | trace_colo_receive_message(COLOMessage_str(msg)); |
4f97558e HZ |
346 | return msg; |
347 | } | |
348 | ||
349 | static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg, | |
350 | Error **errp) | |
351 | { | |
352 | COLOMessage msg; | |
353 | Error *local_err = NULL; | |
354 | ||
355 | msg = colo_receive_message(f, &local_err); | |
356 | if (local_err) { | |
357 | error_propagate(errp, local_err); | |
358 | return; | |
359 | } | |
360 | if (msg != expect_msg) { | |
361 | error_setg(errp, "Unexpected COLO message %d, expected %d", | |
362 | msg, expect_msg); | |
363 | } | |
364 | } | |
365 | ||
4291d372 HZ |
366 | static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg, |
367 | Error **errp) | |
368 | { | |
369 | Error *local_err = NULL; | |
370 | uint64_t value; | |
371 | int ret; | |
372 | ||
373 | colo_receive_check_message(f, expect_msg, &local_err); | |
374 | if (local_err) { | |
375 | error_propagate(errp, local_err); | |
376 | return 0; | |
377 | } | |
378 | ||
379 | value = qemu_get_be64(f); | |
380 | ret = qemu_file_get_error(f); | |
381 | if (ret < 0) { | |
382 | error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s", | |
977c736f | 383 | COLOMessage_str(expect_msg)); |
4291d372 HZ |
384 | } |
385 | return value; | |
386 | } | |
387 | ||
a91246c9 HZ |
388 | static int colo_do_checkpoint_transaction(MigrationState *s, |
389 | QIOChannelBuffer *bioc, | |
390 | QEMUFile *fb) | |
4f97558e HZ |
391 | { |
392 | Error *local_err = NULL; | |
a91246c9 | 393 | int ret = -1; |
4f97558e HZ |
394 | |
395 | colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST, | |
396 | &local_err); | |
397 | if (local_err) { | |
398 | goto out; | |
399 | } | |
400 | ||
401 | colo_receive_check_message(s->rp_state.from_dst_file, | |
402 | COLO_MESSAGE_CHECKPOINT_REPLY, &local_err); | |
403 | if (local_err) { | |
404 | goto out; | |
405 | } | |
a91246c9 HZ |
406 | /* Reset channel-buffer directly */ |
407 | qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL); | |
408 | bioc->usage = 0; | |
4f97558e | 409 | |
a91246c9 | 410 | qemu_mutex_lock_iothread(); |
b3f7f0c5 HZ |
411 | if (failover_get_state() != FAILOVER_STATUS_NONE) { |
412 | qemu_mutex_unlock_iothread(); | |
413 | goto out; | |
414 | } | |
a91246c9 HZ |
415 | vm_stop_force_state(RUN_STATE_COLO); |
416 | qemu_mutex_unlock_iothread(); | |
417 | trace_colo_vm_state_change("run", "stop"); | |
b3f7f0c5 HZ |
418 | /* |
419 | * Failover request bh could be called after vm_stop_force_state(), | |
420 | * So we need check failover_request_is_active() again. | |
421 | */ | |
422 | if (failover_get_state() != FAILOVER_STATUS_NONE) { | |
423 | goto out; | |
424 | } | |
a91246c9 | 425 | |
131b2153 ZC |
426 | colo_notify_compares_event(NULL, COLO_EVENT_CHECKPOINT, &local_err); |
427 | if (local_err) { | |
428 | goto out; | |
429 | } | |
430 | ||
a91246c9 | 431 | /* Disable block migration */ |
ce7c817c | 432 | migrate_set_block_enabled(false, &local_err); |
a91246c9 | 433 | qemu_mutex_lock_iothread(); |
3ebb9c4f ZC |
434 | |
435 | #ifdef CONFIG_REPLICATION | |
8e48ac95 ZC |
436 | replication_do_checkpoint_all(&local_err); |
437 | if (local_err) { | |
438 | qemu_mutex_unlock_iothread(); | |
439 | goto out; | |
440 | } | |
3ebb9c4f ZC |
441 | #else |
442 | abort(); | |
443 | #endif | |
4f97558e HZ |
444 | |
445 | colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err); | |
446 | if (local_err) { | |
3f6df99d ZC |
447 | qemu_mutex_unlock_iothread(); |
448 | goto out; | |
449 | } | |
450 | /* Note: device state is saved into buffer */ | |
451 | ret = qemu_save_device_state(fb); | |
452 | ||
453 | qemu_mutex_unlock_iothread(); | |
454 | if (ret < 0) { | |
4f97558e HZ |
455 | goto out; |
456 | } | |
3f6df99d ZC |
457 | /* |
458 | * Only save VM's live state, which not including device state. | |
459 | * TODO: We may need a timeout mechanism to prevent COLO process | |
460 | * to be blocked here. | |
461 | */ | |
462 | qemu_savevm_live_state(s->to_dst_file); | |
463 | ||
464 | qemu_fflush(fb); | |
465 | ||
a91246c9 HZ |
466 | /* |
467 | * We need the size of the VMstate data in Secondary side, | |
468 | * With which we can decide how much data should be read. | |
469 | */ | |
470 | colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE, | |
471 | bioc->usage, &local_err); | |
472 | if (local_err) { | |
473 | goto out; | |
474 | } | |
4f97558e | 475 | |
a91246c9 HZ |
476 | qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage); |
477 | qemu_fflush(s->to_dst_file); | |
478 | ret = qemu_file_get_error(s->to_dst_file); | |
479 | if (ret < 0) { | |
480 | goto out; | |
481 | } | |
4f97558e HZ |
482 | |
483 | colo_receive_check_message(s->rp_state.from_dst_file, | |
484 | COLO_MESSAGE_VMSTATE_RECEIVED, &local_err); | |
485 | if (local_err) { | |
486 | goto out; | |
487 | } | |
488 | ||
489 | colo_receive_check_message(s->rp_state.from_dst_file, | |
490 | COLO_MESSAGE_VMSTATE_LOADED, &local_err); | |
491 | if (local_err) { | |
492 | goto out; | |
493 | } | |
494 | ||
a91246c9 HZ |
495 | ret = 0; |
496 | ||
497 | qemu_mutex_lock_iothread(); | |
498 | vm_start(); | |
499 | qemu_mutex_unlock_iothread(); | |
500 | trace_colo_vm_state_change("stop", "run"); | |
4f97558e | 501 | |
4f97558e HZ |
502 | out: |
503 | if (local_err) { | |
504 | error_report_err(local_err); | |
505 | } | |
a91246c9 | 506 | return ret; |
4f97558e HZ |
507 | } |
508 | ||
131b2153 ZC |
509 | static void colo_compare_notify_checkpoint(Notifier *notifier, void *data) |
510 | { | |
511 | colo_checkpoint_notify(data); | |
512 | } | |
513 | ||
0b827d5e HZ |
514 | static void colo_process_checkpoint(MigrationState *s) |
515 | { | |
a91246c9 HZ |
516 | QIOChannelBuffer *bioc; |
517 | QEMUFile *fb = NULL; | |
479125d5 | 518 | int64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_HOST); |
4f97558e HZ |
519 | Error *local_err = NULL; |
520 | int ret; | |
521 | ||
5ed0deca ZC |
522 | last_colo_mode = get_colo_mode(); |
523 | if (last_colo_mode != COLO_MODE_PRIMARY) { | |
524 | error_report("COLO mode must be COLO_MODE_PRIMARY"); | |
525 | return; | |
526 | } | |
527 | ||
aef06085 HZ |
528 | failover_init_state(); |
529 | ||
56ba83d2 HZ |
530 | s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file); |
531 | if (!s->rp_state.from_dst_file) { | |
532 | error_report("Open QEMUFile from_dst_file failed"); | |
533 | goto out; | |
534 | } | |
535 | ||
131b2153 ZC |
536 | packets_compare_notifier.notify = colo_compare_notify_checkpoint; |
537 | colo_compare_register_notifier(&packets_compare_notifier); | |
538 | ||
4f97558e HZ |
539 | /* |
540 | * Wait for Secondary finish loading VM states and enter COLO | |
541 | * restore. | |
542 | */ | |
543 | colo_receive_check_message(s->rp_state.from_dst_file, | |
544 | COLO_MESSAGE_CHECKPOINT_READY, &local_err); | |
545 | if (local_err) { | |
546 | goto out; | |
547 | } | |
a91246c9 HZ |
548 | bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE); |
549 | fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc)); | |
550 | object_unref(OBJECT(bioc)); | |
4f97558e | 551 | |
0b827d5e | 552 | qemu_mutex_lock_iothread(); |
3ebb9c4f | 553 | #ifdef CONFIG_REPLICATION |
8e48ac95 ZC |
554 | replication_start_all(REPLICATION_MODE_PRIMARY, &local_err); |
555 | if (local_err) { | |
556 | qemu_mutex_unlock_iothread(); | |
557 | goto out; | |
558 | } | |
3ebb9c4f ZC |
559 | #else |
560 | abort(); | |
561 | #endif | |
8e48ac95 | 562 | |
0b827d5e HZ |
563 | vm_start(); |
564 | qemu_mutex_unlock_iothread(); | |
565 | trace_colo_vm_state_change("stop", "run"); | |
566 | ||
479125d5 HZ |
567 | timer_mod(s->colo_delay_timer, |
568 | current_time + s->parameters.x_checkpoint_delay); | |
569 | ||
4f97558e | 570 | while (s->state == MIGRATION_STATUS_COLO) { |
b3f7f0c5 HZ |
571 | if (failover_get_state() != FAILOVER_STATUS_NONE) { |
572 | error_report("failover request"); | |
573 | goto out; | |
574 | } | |
575 | ||
479125d5 | 576 | qemu_sem_wait(&s->colo_checkpoint_sem); |
18cc23d7 | 577 | |
2518aec1 HZ |
578 | if (s->state != MIGRATION_STATUS_COLO) { |
579 | goto out; | |
580 | } | |
a91246c9 | 581 | ret = colo_do_checkpoint_transaction(s, bioc, fb); |
4f97558e HZ |
582 | if (ret < 0) { |
583 | goto out; | |
584 | } | |
585 | } | |
0b827d5e | 586 | |
56ba83d2 | 587 | out: |
4f97558e HZ |
588 | /* Throw the unreported error message after exited from loop */ |
589 | if (local_err) { | |
590 | error_report_err(local_err); | |
591 | } | |
592 | ||
a91246c9 HZ |
593 | if (fb) { |
594 | qemu_fclose(fb); | |
595 | } | |
596 | ||
9ecff6d6 HZ |
597 | /* |
598 | * There are only two reasons we can get here, some error happened | |
599 | * or the user triggered failover. | |
600 | */ | |
601 | switch (failover_get_state()) { | |
1fe6ab26 | 602 | case FAILOVER_STATUS_COMPLETED: |
9ecff6d6 HZ |
603 | qapi_event_send_colo_exit(COLO_MODE_PRIMARY, |
604 | COLO_EXIT_REASON_REQUEST); | |
605 | break; | |
606 | default: | |
3a43ac47 ZC |
607 | qapi_event_send_colo_exit(COLO_MODE_PRIMARY, |
608 | COLO_EXIT_REASON_ERROR); | |
9ecff6d6 HZ |
609 | } |
610 | ||
c937b9a6 HZ |
611 | /* Hope this not to be too long to wait here */ |
612 | qemu_sem_wait(&s->colo_exit_sem); | |
613 | qemu_sem_destroy(&s->colo_exit_sem); | |
131b2153 ZC |
614 | |
615 | /* | |
616 | * It is safe to unregister notifier after failover finished. | |
617 | * Besides, colo_delay_timer and colo_checkpoint_sem can't be | |
618 | * released befor unregister notifier, or there will be use-after-free | |
619 | * error. | |
620 | */ | |
621 | colo_compare_unregister_notifier(&packets_compare_notifier); | |
622 | timer_del(s->colo_delay_timer); | |
623 | timer_free(s->colo_delay_timer); | |
624 | qemu_sem_destroy(&s->colo_checkpoint_sem); | |
625 | ||
c937b9a6 HZ |
626 | /* |
627 | * Must be called after failover BH is completed, | |
628 | * Or the failover BH may shutdown the wrong fd that | |
629 | * re-used by other threads after we release here. | |
630 | */ | |
56ba83d2 HZ |
631 | if (s->rp_state.from_dst_file) { |
632 | qemu_fclose(s->rp_state.from_dst_file); | |
633 | } | |
0b827d5e HZ |
634 | } |
635 | ||
479125d5 HZ |
636 | void colo_checkpoint_notify(void *opaque) |
637 | { | |
638 | MigrationState *s = opaque; | |
639 | int64_t next_notify_time; | |
640 | ||
641 | qemu_sem_post(&s->colo_checkpoint_sem); | |
642 | s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
643 | next_notify_time = s->colo_checkpoint_time + | |
644 | s->parameters.x_checkpoint_delay; | |
645 | timer_mod(s->colo_delay_timer, next_notify_time); | |
646 | } | |
647 | ||
0b827d5e HZ |
648 | void migrate_start_colo_process(MigrationState *s) |
649 | { | |
650 | qemu_mutex_unlock_iothread(); | |
479125d5 HZ |
651 | qemu_sem_init(&s->colo_checkpoint_sem, 0); |
652 | s->colo_delay_timer = timer_new_ms(QEMU_CLOCK_HOST, | |
653 | colo_checkpoint_notify, s); | |
654 | ||
c937b9a6 | 655 | qemu_sem_init(&s->colo_exit_sem, 0); |
0b827d5e HZ |
656 | migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE, |
657 | MIGRATION_STATUS_COLO); | |
658 | colo_process_checkpoint(s); | |
659 | qemu_mutex_lock_iothread(); | |
660 | } | |
25d0c16f | 661 | |
4f97558e HZ |
662 | static void colo_wait_handle_message(QEMUFile *f, int *checkpoint_request, |
663 | Error **errp) | |
664 | { | |
665 | COLOMessage msg; | |
666 | Error *local_err = NULL; | |
667 | ||
668 | msg = colo_receive_message(f, &local_err); | |
669 | if (local_err) { | |
670 | error_propagate(errp, local_err); | |
671 | return; | |
672 | } | |
673 | ||
674 | switch (msg) { | |
675 | case COLO_MESSAGE_CHECKPOINT_REQUEST: | |
676 | *checkpoint_request = 1; | |
677 | break; | |
678 | default: | |
679 | *checkpoint_request = 0; | |
680 | error_setg(errp, "Got unknown COLO message: %d", msg); | |
681 | break; | |
682 | } | |
683 | } | |
684 | ||
25d0c16f HZ |
685 | void *colo_process_incoming_thread(void *opaque) |
686 | { | |
687 | MigrationIncomingState *mis = opaque; | |
4291d372 HZ |
688 | QEMUFile *fb = NULL; |
689 | QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */ | |
690 | uint64_t total_size; | |
691 | uint64_t value; | |
4f97558e | 692 | Error *local_err = NULL; |
3f6df99d | 693 | int ret; |
25d0c16f | 694 | |
74637e6f | 695 | rcu_register_thread(); |
c937b9a6 HZ |
696 | qemu_sem_init(&mis->colo_incoming_sem, 0); |
697 | ||
25d0c16f HZ |
698 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, |
699 | MIGRATION_STATUS_COLO); | |
700 | ||
5ed0deca ZC |
701 | last_colo_mode = get_colo_mode(); |
702 | if (last_colo_mode != COLO_MODE_SECONDARY) { | |
703 | error_report("COLO mode must be COLO_MODE_SECONDARY"); | |
704 | return NULL; | |
705 | } | |
706 | ||
aef06085 HZ |
707 | failover_init_state(); |
708 | ||
56ba83d2 HZ |
709 | mis->to_src_file = qemu_file_get_return_path(mis->from_src_file); |
710 | if (!mis->to_src_file) { | |
711 | error_report("COLO incoming thread: Open QEMUFile to_src_file failed"); | |
712 | goto out; | |
713 | } | |
714 | /* | |
715 | * Note: the communication between Primary side and Secondary side | |
716 | * should be sequential, we set the fd to unblocked in migration incoming | |
717 | * coroutine, and here we are in the COLO incoming thread, so it is ok to | |
718 | * set the fd back to blocked. | |
719 | */ | |
720 | qemu_file_set_blocking(mis->from_src_file, true); | |
721 | ||
4291d372 HZ |
722 | bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE); |
723 | fb = qemu_fopen_channel_input(QIO_CHANNEL(bioc)); | |
724 | object_unref(OBJECT(bioc)); | |
725 | ||
131b2153 | 726 | qemu_mutex_lock_iothread(); |
3ebb9c4f | 727 | #ifdef CONFIG_REPLICATION |
8e48ac95 ZC |
728 | replication_start_all(REPLICATION_MODE_SECONDARY, &local_err); |
729 | if (local_err) { | |
730 | qemu_mutex_unlock_iothread(); | |
731 | goto out; | |
732 | } | |
3ebb9c4f ZC |
733 | #else |
734 | abort(); | |
735 | #endif | |
131b2153 ZC |
736 | vm_start(); |
737 | trace_colo_vm_state_change("stop", "run"); | |
738 | qemu_mutex_unlock_iothread(); | |
739 | ||
4f97558e HZ |
740 | colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY, |
741 | &local_err); | |
742 | if (local_err) { | |
743 | goto out; | |
744 | } | |
745 | ||
746 | while (mis->state == MIGRATION_STATUS_COLO) { | |
02ba9265 | 747 | int request = 0; |
4f97558e HZ |
748 | |
749 | colo_wait_handle_message(mis->from_src_file, &request, &local_err); | |
750 | if (local_err) { | |
751 | goto out; | |
752 | } | |
753 | assert(request); | |
9d2db376 HZ |
754 | if (failover_get_state() != FAILOVER_STATUS_NONE) { |
755 | error_report("failover request"); | |
756 | goto out; | |
757 | } | |
758 | ||
131b2153 ZC |
759 | qemu_mutex_lock_iothread(); |
760 | vm_stop_force_state(RUN_STATE_COLO); | |
761 | trace_colo_vm_state_change("run", "stop"); | |
762 | qemu_mutex_unlock_iothread(); | |
763 | ||
4f97558e HZ |
764 | /* FIXME: This is unnecessary for periodic checkpoint mode */ |
765 | colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY, | |
766 | &local_err); | |
767 | if (local_err) { | |
768 | goto out; | |
769 | } | |
770 | ||
771 | colo_receive_check_message(mis->from_src_file, | |
772 | COLO_MESSAGE_VMSTATE_SEND, &local_err); | |
773 | if (local_err) { | |
774 | goto out; | |
775 | } | |
776 | ||
3f6df99d ZC |
777 | qemu_mutex_lock_iothread(); |
778 | cpu_synchronize_all_pre_loadvm(); | |
779 | ret = qemu_loadvm_state_main(mis->from_src_file, mis); | |
780 | qemu_mutex_unlock_iothread(); | |
781 | ||
782 | if (ret < 0) { | |
783 | error_report("Load VM's live state (ram) error"); | |
784 | goto out; | |
785 | } | |
786 | ||
4291d372 HZ |
787 | value = colo_receive_message_value(mis->from_src_file, |
788 | COLO_MESSAGE_VMSTATE_SIZE, &local_err); | |
789 | if (local_err) { | |
790 | goto out; | |
791 | } | |
792 | ||
793 | /* | |
794 | * Read VM device state data into channel buffer, | |
795 | * It's better to re-use the memory allocated. | |
796 | * Here we need to handle the channel buffer directly. | |
797 | */ | |
798 | if (value > bioc->capacity) { | |
799 | bioc->capacity = value; | |
800 | bioc->data = g_realloc(bioc->data, bioc->capacity); | |
801 | } | |
802 | total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value); | |
803 | if (total_size != value) { | |
804 | error_report("Got %" PRIu64 " VMState data, less than expected" | |
805 | " %" PRIu64, total_size, value); | |
806 | goto out; | |
807 | } | |
808 | bioc->usage = total_size; | |
809 | qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL); | |
4f97558e HZ |
810 | |
811 | colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED, | |
812 | &local_err); | |
813 | if (local_err) { | |
814 | goto out; | |
815 | } | |
816 | ||
4291d372 | 817 | qemu_mutex_lock_iothread(); |
a8664ba5 | 818 | vmstate_loading = true; |
3f6df99d ZC |
819 | ret = qemu_load_device_state(fb); |
820 | if (ret < 0) { | |
821 | error_report("COLO: load device state failed"); | |
4291d372 HZ |
822 | qemu_mutex_unlock_iothread(); |
823 | goto out; | |
824 | } | |
a8664ba5 | 825 | |
3ebb9c4f | 826 | #ifdef CONFIG_REPLICATION |
8e48ac95 ZC |
827 | replication_get_error_all(&local_err); |
828 | if (local_err) { | |
829 | qemu_mutex_unlock_iothread(); | |
830 | goto out; | |
831 | } | |
3ebb9c4f | 832 | |
8e48ac95 ZC |
833 | /* discard colo disk buffer */ |
834 | replication_do_checkpoint_all(&local_err); | |
835 | if (local_err) { | |
836 | qemu_mutex_unlock_iothread(); | |
837 | goto out; | |
838 | } | |
3ebb9c4f ZC |
839 | #else |
840 | abort(); | |
841 | #endif | |
7b343530 HZ |
842 | /* Notify all filters of all NIC to do checkpoint */ |
843 | colo_notify_filters_event(COLO_EVENT_CHECKPOINT, &local_err); | |
844 | ||
845 | if (local_err) { | |
846 | qemu_mutex_unlock_iothread(); | |
847 | goto out; | |
848 | } | |
849 | ||
a8664ba5 | 850 | vmstate_loading = false; |
131b2153 ZC |
851 | vm_start(); |
852 | trace_colo_vm_state_change("stop", "run"); | |
4291d372 | 853 | qemu_mutex_unlock_iothread(); |
4f97558e | 854 | |
a8664ba5 HZ |
855 | if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) { |
856 | failover_set_state(FAILOVER_STATUS_RELAUNCH, | |
857 | FAILOVER_STATUS_NONE); | |
858 | failover_request_active(NULL); | |
859 | goto out; | |
860 | } | |
861 | ||
4f97558e HZ |
862 | colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED, |
863 | &local_err); | |
864 | if (local_err) { | |
865 | goto out; | |
866 | } | |
867 | } | |
25d0c16f | 868 | |
56ba83d2 | 869 | out: |
a8664ba5 | 870 | vmstate_loading = false; |
4f97558e HZ |
871 | /* Throw the unreported error message after exited from loop */ |
872 | if (local_err) { | |
873 | error_report_err(local_err); | |
874 | } | |
875 | ||
3a43ac47 ZC |
876 | /* |
877 | * There are only two reasons we can get here, some error happened | |
878 | * or the user triggered failover. | |
879 | */ | |
9ecff6d6 | 880 | switch (failover_get_state()) { |
1fe6ab26 | 881 | case FAILOVER_STATUS_COMPLETED: |
9ecff6d6 HZ |
882 | qapi_event_send_colo_exit(COLO_MODE_SECONDARY, |
883 | COLO_EXIT_REASON_REQUEST); | |
884 | break; | |
885 | default: | |
3a43ac47 ZC |
886 | qapi_event_send_colo_exit(COLO_MODE_SECONDARY, |
887 | COLO_EXIT_REASON_ERROR); | |
9ecff6d6 HZ |
888 | } |
889 | ||
4291d372 HZ |
890 | if (fb) { |
891 | qemu_fclose(fb); | |
892 | } | |
893 | ||
c937b9a6 HZ |
894 | /* Hope this not to be too long to loop here */ |
895 | qemu_sem_wait(&mis->colo_incoming_sem); | |
896 | qemu_sem_destroy(&mis->colo_incoming_sem); | |
897 | /* Must be called after failover BH is completed */ | |
56ba83d2 HZ |
898 | if (mis->to_src_file) { |
899 | qemu_fclose(mis->to_src_file); | |
b8b5734b | 900 | mis->to_src_file = NULL; |
56ba83d2 | 901 | } |
25d0c16f | 902 | |
74637e6f | 903 | rcu_unregister_thread(); |
25d0c16f HZ |
904 | return NULL; |
905 | } |