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