]>
Commit | Line | Data |
---|---|---|
5bb7910a AL |
1 | /* |
2 | * QEMU live migration | |
3 | * | |
4 | * Copyright IBM, Corp. 2008 | |
5 | * | |
6 | * Authors: | |
7 | * Anthony Liguori <aliguori@us.ibm.com> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
10 | * the COPYING file in the top-level directory. | |
11 | * | |
6b620ca3 PB |
12 | * Contributions after 2012-01-13 are licensed under the terms of the |
13 | * GNU GPL, version 2 or (at your option) any later version. | |
5bb7910a AL |
14 | */ |
15 | ||
1393a485 | 16 | #include "qemu/osdep.h" |
f348b6d1 | 17 | #include "qemu/cutils.h" |
d49b6836 | 18 | #include "qemu/error-report.h" |
795c40b8 | 19 | #include "migration/blocker.h" |
f4dbe1bf | 20 | #include "exec.h" |
7fcac4a2 | 21 | #include "fd.h" |
61e8b148 | 22 | #include "socket.h" |
e1a3ecee | 23 | #include "rdma.h" |
7b1e1a22 | 24 | #include "ram.h" |
84a899de | 25 | #include "migration/global_state.h" |
c4b63b7c | 26 | #include "migration/misc.h" |
6666c96a | 27 | #include "migration.h" |
20a519a0 | 28 | #include "savevm.h" |
40014d81 | 29 | #include "qemu-file-channel.h" |
08a0aee1 | 30 | #include "qemu-file.h" |
987772d9 | 31 | #include "migration/vmstate.h" |
737e150e | 32 | #include "block/block.h" |
cc7a8ea7 | 33 | #include "qapi/qmp/qerror.h" |
6c595cde | 34 | #include "qapi/util.h" |
ab28bd23 | 35 | #include "qemu/rcu.h" |
2c9e6fec | 36 | #include "block.h" |
be07b0ac | 37 | #include "postcopy-ram.h" |
766bd176 | 38 | #include "qemu/thread.h" |
791e7c82 | 39 | #include "qmp-commands.h" |
c09e5bb1 | 40 | #include "trace.h" |
598cd2bd | 41 | #include "qapi-event.h" |
51180423 | 42 | #include "exec/target_page.h" |
61b67d47 | 43 | #include "io/channel-buffer.h" |
35a6ed4f | 44 | #include "migration/colo.h" |
4ffdb337 | 45 | #include "hw/boards.h" |
9d18af93 | 46 | #include "monitor/monitor.h" |
065e2813 | 47 | |
dc325627 | 48 | #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */ |
5bb7910a | 49 | |
5b4e1eb7 JQ |
50 | /* Amount of time to allocate to each "chunk" of bandwidth-throttled |
51 | * data. */ | |
52 | #define BUFFER_DELAY 100 | |
53 | #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) | |
54 | ||
2ff30257 AA |
55 | /* Time in milliseconds we are allowed to stop the source, |
56 | * for sending the last part */ | |
57 | #define DEFAULT_MIGRATE_SET_DOWNTIME 300 | |
58 | ||
87c9cc1c DHB |
59 | /* Maximum migrate downtime set to 2000 seconds */ |
60 | #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 | |
61 | #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) | |
62 | ||
8706d2d5 LL |
63 | /* Default compression thread count */ |
64 | #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 | |
3fcb38c2 LL |
65 | /* Default decompression thread count, usually decompression is at |
66 | * least 4 times as fast as compression.*/ | |
67 | #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 | |
8706d2d5 LL |
68 | /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
69 | #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 | |
1626fee3 | 70 | /* Define default autoconverge cpu throttle migration parameters */ |
d85a31d1 JH |
71 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 |
72 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 | |
8706d2d5 | 73 | |
17ad9b35 OW |
74 | /* Migration XBZRLE default cache size */ |
75 | #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024) | |
76 | ||
68b53591 HZ |
77 | /* The delay time (in ms) between two COLO checkpoints |
78 | * Note: Please change this default value to 10000 when we support hybrid mode. | |
79 | */ | |
80 | #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200 | |
81 | ||
99a0db9b GH |
82 | static NotifierList migration_state_notifiers = |
83 | NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); | |
84 | ||
adde220a DDAG |
85 | static bool deferred_incoming; |
86 | ||
da6f1790 JQ |
87 | /* Messages sent on the return path from destination to source */ |
88 | enum mig_rp_message_type { | |
89 | MIG_RP_MSG_INVALID = 0, /* Must be 0 */ | |
90 | MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ | |
91 | MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ | |
92 | ||
93 | MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ | |
94 | MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ | |
95 | ||
96 | MIG_RP_MSG_MAX | |
97 | }; | |
98 | ||
17549e84 JQ |
99 | /* When we add fault tolerance, we could have several |
100 | migrations at once. For now we don't need to add | |
101 | dynamic creation of migration */ | |
102 | ||
e5cb7e76 PX |
103 | static MigrationState *current_migration; |
104 | ||
105 | void migration_object_init(void) | |
106 | { | |
4ffdb337 PX |
107 | MachineState *ms = MACHINE(qdev_get_machine()); |
108 | ||
e5cb7e76 PX |
109 | /* This can only be called once. */ |
110 | assert(!current_migration); | |
111 | current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); | |
4ffdb337 PX |
112 | |
113 | /* | |
114 | * We cannot really do this in migration_instance_init() since at | |
115 | * that time global properties are not yet applied, then this | |
116 | * value will be definitely replaced by something else. | |
117 | */ | |
118 | if (ms->enforce_config_section) { | |
119 | current_migration->send_configuration = true; | |
120 | } | |
e5cb7e76 PX |
121 | } |
122 | ||
bca7856a | 123 | /* For outgoing */ |
859bc756 | 124 | MigrationState *migrate_get_current(void) |
17549e84 | 125 | { |
e5cb7e76 PX |
126 | /* This can only be called after the object created. */ |
127 | assert(current_migration); | |
128 | return current_migration; | |
17549e84 JQ |
129 | } |
130 | ||
3df663e5 PX |
131 | void migration_only_migratable_set(void) |
132 | { | |
133 | migrate_get_current()->only_migratable = true; | |
134 | } | |
135 | ||
bca7856a DDAG |
136 | MigrationIncomingState *migration_incoming_get_current(void) |
137 | { | |
b4b076da JQ |
138 | static bool once; |
139 | static MigrationIncomingState mis_current; | |
bca7856a | 140 | |
b4b076da JQ |
141 | if (!once) { |
142 | mis_current.state = MIGRATION_STATUS_NONE; | |
143 | memset(&mis_current, 0, sizeof(MigrationIncomingState)); | |
b4b076da JQ |
144 | qemu_mutex_init(&mis_current.rp_mutex); |
145 | qemu_event_init(&mis_current.main_thread_load_event, false); | |
146 | once = true; | |
147 | } | |
148 | return &mis_current; | |
bca7856a DDAG |
149 | } |
150 | ||
151 | void migration_incoming_state_destroy(void) | |
152 | { | |
b4b076da JQ |
153 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
154 | ||
3482655b | 155 | if (mis->to_src_file) { |
660819b1 PX |
156 | /* Tell source that we are done */ |
157 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
158 | qemu_fclose(mis->to_src_file); |
159 | mis->to_src_file = NULL; | |
160 | } | |
161 | ||
660819b1 PX |
162 | if (mis->from_src_file) { |
163 | qemu_fclose(mis->from_src_file); | |
164 | mis->from_src_file = NULL; | |
165 | } | |
166 | ||
b4b076da | 167 | qemu_event_destroy(&mis->main_thread_load_event); |
bca7856a DDAG |
168 | } |
169 | ||
b05dc723 JQ |
170 | static void migrate_generate_event(int new_state) |
171 | { | |
172 | if (migrate_use_events()) { | |
173 | qapi_event_send_migration(new_state, &error_abort); | |
b05dc723 JQ |
174 | } |
175 | } | |
176 | ||
adde220a DDAG |
177 | /* |
178 | * Called on -incoming with a defer: uri. | |
179 | * The migration can be started later after any parameters have been | |
180 | * changed. | |
181 | */ | |
182 | static void deferred_incoming_migration(Error **errp) | |
183 | { | |
184 | if (deferred_incoming) { | |
185 | error_setg(errp, "Incoming migration already deferred"); | |
186 | } | |
187 | deferred_incoming = true; | |
188 | } | |
189 | ||
da6f1790 JQ |
190 | /* |
191 | * Send a message on the return channel back to the source | |
192 | * of the migration. | |
193 | */ | |
194 | static void migrate_send_rp_message(MigrationIncomingState *mis, | |
195 | enum mig_rp_message_type message_type, | |
196 | uint16_t len, void *data) | |
197 | { | |
198 | trace_migrate_send_rp_message((int)message_type, len); | |
199 | qemu_mutex_lock(&mis->rp_mutex); | |
200 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); | |
201 | qemu_put_be16(mis->to_src_file, len); | |
202 | qemu_put_buffer(mis->to_src_file, data, len); | |
203 | qemu_fflush(mis->to_src_file); | |
204 | qemu_mutex_unlock(&mis->rp_mutex); | |
205 | } | |
206 | ||
1e2d90eb DDAG |
207 | /* Request a range of pages from the source VM at the given |
208 | * start address. | |
209 | * rbname: Name of the RAMBlock to request the page in, if NULL it's the same | |
210 | * as the last request (a name must have been given previously) | |
211 | * Start: Address offset within the RB | |
212 | * Len: Length in bytes required - must be a multiple of pagesize | |
213 | */ | |
214 | void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname, | |
215 | ram_addr_t start, size_t len) | |
216 | { | |
cb8d4c8f | 217 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb DDAG |
218 | size_t msglen = 12; /* start + len */ |
219 | ||
220 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
221 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
222 | ||
223 | if (rbname) { | |
224 | int rbname_len = strlen(rbname); | |
225 | assert(rbname_len < 256); | |
226 | ||
227 | bufc[msglen++] = rbname_len; | |
228 | memcpy(bufc + msglen, rbname, rbname_len); | |
229 | msglen += rbname_len; | |
230 | migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc); | |
231 | } else { | |
232 | migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc); | |
233 | } | |
234 | } | |
235 | ||
43eaae28 | 236 | void qemu_start_incoming_migration(const char *uri, Error **errp) |
5bb7910a | 237 | { |
34c9dd8e AL |
238 | const char *p; |
239 | ||
7cf1fe6d | 240 | qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort); |
adde220a DDAG |
241 | if (!strcmp(uri, "defer")) { |
242 | deferred_incoming_migration(errp); | |
243 | } else if (strstart(uri, "tcp:", &p)) { | |
43eaae28 | 244 | tcp_start_incoming_migration(p, errp); |
2da776db | 245 | #ifdef CONFIG_RDMA |
adde220a | 246 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
247 | rdma_start_incoming_migration(p, errp); |
248 | #endif | |
adde220a | 249 | } else if (strstart(uri, "exec:", &p)) { |
43eaae28 | 250 | exec_start_incoming_migration(p, errp); |
adde220a | 251 | } else if (strstart(uri, "unix:", &p)) { |
43eaae28 | 252 | unix_start_incoming_migration(p, errp); |
adde220a | 253 | } else if (strstart(uri, "fd:", &p)) { |
43eaae28 | 254 | fd_start_incoming_migration(p, errp); |
adde220a | 255 | } else { |
312fd5f2 | 256 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 257 | } |
5bb7910a AL |
258 | } |
259 | ||
0aa6aefc DL |
260 | static void process_incoming_migration_bh(void *opaque) |
261 | { | |
262 | Error *local_err = NULL; | |
263 | MigrationIncomingState *mis = opaque; | |
264 | ||
ace21a58 KW |
265 | /* Make sure all file formats flush their mutable metadata. |
266 | * If we get an error here, just don't restart the VM yet. */ | |
0aa6aefc | 267 | bdrv_invalidate_cache_all(&local_err); |
d35ff5e6 | 268 | if (local_err) { |
ace21a58 | 269 | error_report_err(local_err); |
d35ff5e6 KW |
270 | local_err = NULL; |
271 | autostart = false; | |
272 | } | |
273 | ||
0aa6aefc DL |
274 | /* |
275 | * This must happen after all error conditions are dealt with and | |
276 | * we're sure the VM is going to be running on this host. | |
277 | */ | |
278 | qemu_announce_self(); | |
279 | ||
280 | /* If global state section was not received or we are in running | |
281 | state, we need to obey autostart. Any other state is set with | |
282 | runstate_set. */ | |
283 | ||
284 | if (!global_state_received() || | |
285 | global_state_get_runstate() == RUN_STATE_RUNNING) { | |
286 | if (autostart) { | |
287 | vm_start(); | |
288 | } else { | |
289 | runstate_set(RUN_STATE_PAUSED); | |
290 | } | |
291 | } else { | |
292 | runstate_set(global_state_get_runstate()); | |
293 | } | |
294 | migrate_decompress_threads_join(); | |
295 | /* | |
296 | * This must happen after any state changes since as soon as an external | |
297 | * observer sees this event they might start to prod at the VM assuming | |
298 | * it's ready to use. | |
299 | */ | |
300 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
301 | MIGRATION_STATUS_COMPLETED); | |
302 | qemu_bh_delete(mis->bh); | |
303 | migration_incoming_state_destroy(); | |
304 | } | |
305 | ||
82a4da79 | 306 | static void process_incoming_migration_co(void *opaque) |
511c0231 | 307 | { |
82a4da79 | 308 | QEMUFile *f = opaque; |
b4b076da | 309 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 310 | PostcopyState ps; |
1c12e1f5 PB |
311 | int ret; |
312 | ||
b4b076da | 313 | mis->from_src_file = f; |
67f11b5c | 314 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 315 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
93d7af6f HZ |
316 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
317 | MIGRATION_STATUS_ACTIVE); | |
1c12e1f5 | 318 | ret = qemu_loadvm_state(f); |
bca7856a | 319 | |
e9bef235 DDAG |
320 | ps = postcopy_state_get(); |
321 | trace_process_incoming_migration_co_end(ret, ps); | |
322 | if (ps != POSTCOPY_INCOMING_NONE) { | |
323 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
324 | /* | |
325 | * Where a migration had postcopy enabled (and thus went to advise) | |
326 | * but managed to complete within the precopy period, we can use | |
327 | * the normal exit. | |
328 | */ | |
329 | postcopy_ram_incoming_cleanup(mis); | |
330 | } else if (ret >= 0) { | |
331 | /* | |
332 | * Postcopy was started, cleanup should happen at the end of the | |
333 | * postcopy thread. | |
334 | */ | |
335 | trace_process_incoming_migration_co_postcopy_end_main(); | |
336 | return; | |
337 | } | |
338 | /* Else if something went wrong then just fall out of the normal exit */ | |
339 | } | |
340 | ||
25d0c16f HZ |
341 | /* we get COLO info, and know if we are in COLO mode */ |
342 | if (!ret && migration_incoming_enable_colo()) { | |
343 | mis->migration_incoming_co = qemu_coroutine_self(); | |
344 | qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", | |
345 | colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); | |
346 | mis->have_colo_incoming_thread = true; | |
347 | qemu_coroutine_yield(); | |
348 | ||
349 | /* Wait checkpoint incoming thread exit before free resource */ | |
350 | qemu_thread_join(&mis->colo_incoming_thread); | |
351 | } | |
352 | ||
1c12e1f5 | 353 | if (ret < 0) { |
93d7af6f HZ |
354 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, |
355 | MIGRATION_STATUS_FAILED); | |
db80face | 356 | error_report("load of migration failed: %s", strerror(-ret)); |
3fcb38c2 | 357 | migrate_decompress_threads_join(); |
4aead692 | 358 | exit(EXIT_FAILURE); |
511c0231 | 359 | } |
511c0231 | 360 | |
e8199e48 LV |
361 | free_xbzrle_decoded_buf(); |
362 | ||
0aa6aefc DL |
363 | mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); |
364 | qemu_bh_schedule(mis->bh); | |
511c0231 JQ |
365 | } |
366 | ||
22724f49 | 367 | void migration_fd_process_incoming(QEMUFile *f) |
82a4da79 | 368 | { |
0b8b8753 | 369 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f); |
82a4da79 | 370 | |
3fcb38c2 | 371 | migrate_decompress_threads_create(); |
06ad5135 | 372 | qemu_file_set_blocking(f, false); |
0b8b8753 | 373 | qemu_coroutine_enter(co); |
82a4da79 PB |
374 | } |
375 | ||
6decec93 DDAG |
376 | /* |
377 | * Send a 'SHUT' message on the return channel with the given value | |
378 | * to indicate that we've finished with the RP. Non-0 value indicates | |
379 | * error. | |
380 | */ | |
381 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
382 | uint32_t value) | |
383 | { | |
384 | uint32_t buf; | |
385 | ||
386 | buf = cpu_to_be32(value); | |
387 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
388 | } | |
389 | ||
390 | /* | |
391 | * Send a 'PONG' message on the return channel with the given value | |
392 | * (normally in response to a 'PING') | |
393 | */ | |
394 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
395 | uint32_t value) | |
396 | { | |
397 | uint32_t buf; | |
398 | ||
399 | buf = cpu_to_be32(value); | |
400 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
401 | } | |
402 | ||
bbf6da32 OW |
403 | MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) |
404 | { | |
405 | MigrationCapabilityStatusList *head = NULL; | |
406 | MigrationCapabilityStatusList *caps; | |
407 | MigrationState *s = migrate_get_current(); | |
408 | int i; | |
409 | ||
387eedeb | 410 | caps = NULL; /* silence compiler warning */ |
7fb1cf16 | 411 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
ed1701c6 DDAG |
412 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
413 | if (i == MIGRATION_CAPABILITY_BLOCK) { | |
414 | continue; | |
415 | } | |
416 | #endif | |
35a6ed4f HZ |
417 | if (i == MIGRATION_CAPABILITY_X_COLO && !colo_supported()) { |
418 | continue; | |
419 | } | |
bbf6da32 OW |
420 | if (head == NULL) { |
421 | head = g_malloc0(sizeof(*caps)); | |
422 | caps = head; | |
423 | } else { | |
424 | caps->next = g_malloc0(sizeof(*caps)); | |
425 | caps = caps->next; | |
426 | } | |
427 | caps->value = | |
428 | g_malloc(sizeof(*caps->value)); | |
429 | caps->value->capability = i; | |
430 | caps->value->state = s->enabled_capabilities[i]; | |
431 | } | |
432 | ||
433 | return head; | |
434 | } | |
435 | ||
85de8323 LL |
436 | MigrationParameters *qmp_query_migrate_parameters(Error **errp) |
437 | { | |
438 | MigrationParameters *params; | |
439 | MigrationState *s = migrate_get_current(); | |
440 | ||
441 | params = g_malloc0(sizeof(*params)); | |
de63ab61 | 442 | params->has_compress_level = true; |
2594f56d | 443 | params->compress_level = s->parameters.compress_level; |
de63ab61 | 444 | params->has_compress_threads = true; |
2594f56d | 445 | params->compress_threads = s->parameters.compress_threads; |
de63ab61 | 446 | params->has_decompress_threads = true; |
2594f56d | 447 | params->decompress_threads = s->parameters.decompress_threads; |
de63ab61 | 448 | params->has_cpu_throttle_initial = true; |
2594f56d | 449 | params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; |
de63ab61 | 450 | params->has_cpu_throttle_increment = true; |
2594f56d | 451 | params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; |
de63ab61 | 452 | params->has_tls_creds = !!s->parameters.tls_creds; |
69ef1f36 | 453 | params->tls_creds = g_strdup(s->parameters.tls_creds); |
de63ab61 | 454 | params->has_tls_hostname = !!s->parameters.tls_hostname; |
69ef1f36 | 455 | params->tls_hostname = g_strdup(s->parameters.tls_hostname); |
2ff30257 AA |
456 | params->has_max_bandwidth = true; |
457 | params->max_bandwidth = s->parameters.max_bandwidth; | |
458 | params->has_downtime_limit = true; | |
459 | params->downtime_limit = s->parameters.downtime_limit; | |
fe39a4d4 | 460 | params->has_x_checkpoint_delay = true; |
68b53591 | 461 | params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; |
2833c59b JQ |
462 | params->has_block_incremental = true; |
463 | params->block_incremental = s->parameters.block_incremental; | |
85de8323 LL |
464 | |
465 | return params; | |
466 | } | |
467 | ||
f6844b99 DDAG |
468 | /* |
469 | * Return true if we're already in the middle of a migration | |
470 | * (i.e. any of the active or setup states) | |
471 | */ | |
472 | static bool migration_is_setup_or_active(int state) | |
473 | { | |
474 | switch (state) { | |
475 | case MIGRATION_STATUS_ACTIVE: | |
9ec055ae | 476 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
f6844b99 DDAG |
477 | case MIGRATION_STATUS_SETUP: |
478 | return true; | |
479 | ||
480 | default: | |
481 | return false; | |
482 | ||
483 | } | |
484 | } | |
485 | ||
a22463a5 DDAG |
486 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
487 | { | |
488 | info->has_ram = true; | |
489 | info->ram = g_malloc0(sizeof(*info->ram)); | |
9360447d | 490 | info->ram->transferred = ram_counters.transferred; |
a22463a5 | 491 | info->ram->total = ram_bytes_total(); |
9360447d | 492 | info->ram->duplicate = ram_counters.duplicate; |
bedf53c1 JQ |
493 | /* legacy value. It is not used anymore */ |
494 | info->ram->skipped = 0; | |
9360447d JQ |
495 | info->ram->normal = ram_counters.normal; |
496 | info->ram->normal_bytes = ram_counters.normal * | |
20afaed9 | 497 | qemu_target_page_size(); |
a22463a5 | 498 | info->ram->mbps = s->mbps; |
9360447d JQ |
499 | info->ram->dirty_sync_count = ram_counters.dirty_sync_count; |
500 | info->ram->postcopy_requests = ram_counters.postcopy_requests; | |
030ce1f8 | 501 | info->ram->page_size = qemu_target_page_size(); |
a22463a5 | 502 | |
114f5aee JQ |
503 | if (migrate_use_xbzrle()) { |
504 | info->has_xbzrle_cache = true; | |
505 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); | |
506 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
9360447d JQ |
507 | info->xbzrle_cache->bytes = xbzrle_counters.bytes; |
508 | info->xbzrle_cache->pages = xbzrle_counters.pages; | |
509 | info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss; | |
510 | info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate; | |
511 | info->xbzrle_cache->overflow = xbzrle_counters.overflow; | |
114f5aee JQ |
512 | } |
513 | ||
338182c8 JQ |
514 | if (cpu_throttle_active()) { |
515 | info->has_cpu_throttle_percentage = true; | |
516 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
517 | } | |
518 | ||
a22463a5 DDAG |
519 | if (s->state != MIGRATION_STATUS_COMPLETED) { |
520 | info->ram->remaining = ram_bytes_remaining(); | |
9360447d | 521 | info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate; |
a22463a5 DDAG |
522 | } |
523 | } | |
524 | ||
930ac04c JQ |
525 | static void populate_disk_info(MigrationInfo *info) |
526 | { | |
527 | if (blk_mig_active()) { | |
528 | info->has_disk = true; | |
529 | info->disk = g_malloc0(sizeof(*info->disk)); | |
530 | info->disk->transferred = blk_mig_bytes_transferred(); | |
531 | info->disk->remaining = blk_mig_bytes_remaining(); | |
532 | info->disk->total = blk_mig_bytes_total(); | |
533 | } | |
534 | } | |
535 | ||
791e7c82 | 536 | MigrationInfo *qmp_query_migrate(Error **errp) |
5bb7910a | 537 | { |
791e7c82 | 538 | MigrationInfo *info = g_malloc0(sizeof(*info)); |
17549e84 JQ |
539 | MigrationState *s = migrate_get_current(); |
540 | ||
541 | switch (s->state) { | |
31194731 | 542 | case MIGRATION_STATUS_NONE: |
17549e84 JQ |
543 | /* no migration has happened ever */ |
544 | break; | |
31194731 | 545 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 546 | info->has_status = true; |
ed4fbd10 | 547 | info->has_total_time = false; |
29ae8a41 | 548 | break; |
31194731 HZ |
549 | case MIGRATION_STATUS_ACTIVE: |
550 | case MIGRATION_STATUS_CANCELLING: | |
9ec055ae | 551 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
c8f9f4f4 | 552 | /* TODO add some postcopy stats */ |
9ec055ae DDAG |
553 | info->has_status = true; |
554 | info->has_total_time = true; | |
555 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) | |
556 | - s->total_time; | |
557 | info->has_expected_downtime = true; | |
558 | info->expected_downtime = s->expected_downtime; | |
559 | info->has_setup_time = true; | |
560 | info->setup_time = s->setup_time; | |
561 | ||
a22463a5 | 562 | populate_ram_info(info, s); |
930ac04c | 563 | populate_disk_info(info); |
17549e84 | 564 | break; |
0b827d5e HZ |
565 | case MIGRATION_STATUS_COLO: |
566 | info->has_status = true; | |
567 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
568 | break; | |
31194731 | 569 | case MIGRATION_STATUS_COMPLETED: |
791e7c82 | 570 | info->has_status = true; |
00c14997 | 571 | info->has_total_time = true; |
7aa939af | 572 | info->total_time = s->total_time; |
9c5a9fcf JQ |
573 | info->has_downtime = true; |
574 | info->downtime = s->downtime; | |
ed4fbd10 MH |
575 | info->has_setup_time = true; |
576 | info->setup_time = s->setup_time; | |
d5f8a570 | 577 | |
a22463a5 | 578 | populate_ram_info(info, s); |
17549e84 | 579 | break; |
31194731 | 580 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 581 | info->has_status = true; |
d59ce6f3 DB |
582 | if (s->error) { |
583 | info->has_error_desc = true; | |
584 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
585 | } | |
17549e84 | 586 | break; |
31194731 | 587 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 588 | info->has_status = true; |
17549e84 | 589 | break; |
5bb7910a | 590 | } |
cde63fbe | 591 | info->status = s->state; |
791e7c82 LC |
592 | |
593 | return info; | |
5bb7910a AL |
594 | } |
595 | ||
00458433 OW |
596 | void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params, |
597 | Error **errp) | |
598 | { | |
599 | MigrationState *s = migrate_get_current(); | |
600 | MigrationCapabilityStatusList *cap; | |
096631bd | 601 | bool old_postcopy_cap = migrate_postcopy_ram(); |
00458433 | 602 | |
f6844b99 | 603 | if (migration_is_setup_or_active(s->state)) { |
c6bd8c70 | 604 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
00458433 OW |
605 | return; |
606 | } | |
607 | ||
608 | for (cap = params; cap; cap = cap->next) { | |
ed1701c6 DDAG |
609 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
610 | if (cap->value->capability == MIGRATION_CAPABILITY_BLOCK | |
611 | && cap->value->state) { | |
612 | error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) " | |
613 | "block migration"); | |
614 | error_append_hint(errp, "Use drive_mirror+NBD instead.\n"); | |
615 | continue; | |
616 | } | |
617 | #endif | |
35a6ed4f HZ |
618 | if (cap->value->capability == MIGRATION_CAPABILITY_X_COLO) { |
619 | if (!colo_supported()) { | |
620 | error_setg(errp, "COLO is not currently supported, please" | |
621 | " configure with --enable-colo option in order to" | |
622 | " support COLO feature"); | |
623 | continue; | |
624 | } | |
625 | } | |
00458433 OW |
626 | s->enabled_capabilities[cap->value->capability] = cap->value->state; |
627 | } | |
53dd370c DDAG |
628 | |
629 | if (migrate_postcopy_ram()) { | |
630 | if (migrate_use_compression()) { | |
631 | /* The decompression threads asynchronously write into RAM | |
632 | * rather than use the atomic copies needed to avoid | |
633 | * userfaulting. It should be possible to fix the decompression | |
634 | * threads for compatibility in future. | |
635 | */ | |
636 | error_report("Postcopy is not currently compatible with " | |
637 | "compression"); | |
32c3db5b | 638 | s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] = |
53dd370c DDAG |
639 | false; |
640 | } | |
096631bd DDAG |
641 | /* This check is reasonably expensive, so only when it's being |
642 | * set the first time, also it's only the destination that needs | |
643 | * special support. | |
644 | */ | |
645 | if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) && | |
646 | !postcopy_ram_supported_by_host()) { | |
647 | /* postcopy_ram_supported_by_host will have emitted a more | |
648 | * detailed message | |
649 | */ | |
650 | error_report("Postcopy is not supported"); | |
651 | s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] = | |
652 | false; | |
653 | } | |
53dd370c | 654 | } |
00458433 OW |
655 | } |
656 | ||
7f375e04 | 657 | void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp) |
85de8323 LL |
658 | { |
659 | MigrationState *s = migrate_get_current(); | |
660 | ||
7f375e04 EB |
661 | if (params->has_compress_level && |
662 | (params->compress_level < 0 || params->compress_level > 9)) { | |
c6bd8c70 MA |
663 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", |
664 | "is invalid, it should be in the range of 0 to 9"); | |
85de8323 LL |
665 | return; |
666 | } | |
7f375e04 EB |
667 | if (params->has_compress_threads && |
668 | (params->compress_threads < 1 || params->compress_threads > 255)) { | |
c6bd8c70 MA |
669 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
670 | "compress_threads", | |
671 | "is invalid, it should be in the range of 1 to 255"); | |
85de8323 LL |
672 | return; |
673 | } | |
7f375e04 EB |
674 | if (params->has_decompress_threads && |
675 | (params->decompress_threads < 1 || params->decompress_threads > 255)) { | |
c6bd8c70 MA |
676 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
677 | "decompress_threads", | |
678 | "is invalid, it should be in the range of 1 to 255"); | |
85de8323 LL |
679 | return; |
680 | } | |
7f375e04 EB |
681 | if (params->has_cpu_throttle_initial && |
682 | (params->cpu_throttle_initial < 1 || | |
683 | params->cpu_throttle_initial > 99)) { | |
1626fee3 | 684 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 685 | "cpu_throttle_initial", |
1626fee3 | 686 | "an integer in the range of 1 to 99"); |
091ecc8b | 687 | return; |
1626fee3 | 688 | } |
7f375e04 EB |
689 | if (params->has_cpu_throttle_increment && |
690 | (params->cpu_throttle_increment < 1 || | |
691 | params->cpu_throttle_increment > 99)) { | |
1626fee3 | 692 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 693 | "cpu_throttle_increment", |
1626fee3 | 694 | "an integer in the range of 1 to 99"); |
091ecc8b | 695 | return; |
1626fee3 | 696 | } |
2ff30257 AA |
697 | if (params->has_max_bandwidth && |
698 | (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) { | |
699 | error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the" | |
700 | " range of 0 to %zu bytes/second", SIZE_MAX); | |
701 | return; | |
702 | } | |
703 | if (params->has_downtime_limit && | |
87c9cc1c DHB |
704 | (params->downtime_limit < 0 || |
705 | params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { | |
706 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " | |
707 | "the range of 0 to %d milliseconds", | |
708 | MAX_MIGRATE_DOWNTIME); | |
2ff30257 AA |
709 | return; |
710 | } | |
68b53591 HZ |
711 | if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) { |
712 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
713 | "x_checkpoint_delay", | |
714 | "is invalid, it should be positive"); | |
715 | } | |
85de8323 | 716 | |
7f375e04 EB |
717 | if (params->has_compress_level) { |
718 | s->parameters.compress_level = params->compress_level; | |
85de8323 | 719 | } |
7f375e04 EB |
720 | if (params->has_compress_threads) { |
721 | s->parameters.compress_threads = params->compress_threads; | |
85de8323 | 722 | } |
7f375e04 EB |
723 | if (params->has_decompress_threads) { |
724 | s->parameters.decompress_threads = params->decompress_threads; | |
85de8323 | 725 | } |
7f375e04 EB |
726 | if (params->has_cpu_throttle_initial) { |
727 | s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; | |
1626fee3 | 728 | } |
7f375e04 EB |
729 | if (params->has_cpu_throttle_increment) { |
730 | s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; | |
1626fee3 | 731 | } |
7f375e04 | 732 | if (params->has_tls_creds) { |
69ef1f36 | 733 | g_free(s->parameters.tls_creds); |
7f375e04 | 734 | s->parameters.tls_creds = g_strdup(params->tls_creds); |
69ef1f36 | 735 | } |
7f375e04 | 736 | if (params->has_tls_hostname) { |
69ef1f36 | 737 | g_free(s->parameters.tls_hostname); |
7f375e04 | 738 | s->parameters.tls_hostname = g_strdup(params->tls_hostname); |
69ef1f36 | 739 | } |
2ff30257 AA |
740 | if (params->has_max_bandwidth) { |
741 | s->parameters.max_bandwidth = params->max_bandwidth; | |
742 | if (s->to_dst_file) { | |
743 | qemu_file_set_rate_limit(s->to_dst_file, | |
744 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); | |
745 | } | |
746 | } | |
747 | if (params->has_downtime_limit) { | |
748 | s->parameters.downtime_limit = params->downtime_limit; | |
749 | } | |
68b53591 HZ |
750 | |
751 | if (params->has_x_checkpoint_delay) { | |
752 | s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; | |
479125d5 HZ |
753 | if (migration_in_colo_state()) { |
754 | colo_checkpoint_notify(s); | |
755 | } | |
68b53591 | 756 | } |
2833c59b JQ |
757 | if (params->has_block_incremental) { |
758 | s->parameters.block_incremental = params->block_incremental; | |
759 | } | |
85de8323 LL |
760 | } |
761 | ||
2594f56d | 762 | |
4886a1bc DDAG |
763 | void qmp_migrate_start_postcopy(Error **errp) |
764 | { | |
765 | MigrationState *s = migrate_get_current(); | |
766 | ||
767 | if (!migrate_postcopy_ram()) { | |
a54d340b | 768 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
769 | " the start of migration"); |
770 | return; | |
771 | } | |
772 | ||
773 | if (s->state == MIGRATION_STATUS_NONE) { | |
774 | error_setg(errp, "Postcopy must be started after migration has been" | |
775 | " started"); | |
776 | return; | |
777 | } | |
778 | /* | |
779 | * we don't error if migration has finished since that would be racy | |
780 | * with issuing this command. | |
781 | */ | |
782 | atomic_set(&s->start_postcopy, true); | |
783 | } | |
784 | ||
065e2813 AL |
785 | /* shared migration helpers */ |
786 | ||
48781e5b | 787 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 788 | { |
48781e5b | 789 | if (atomic_cmpxchg(state, old_state, new_state) == old_state) { |
4ba4bc5e | 790 | trace_migrate_set_state(new_state); |
b05dc723 | 791 | migrate_generate_event(new_state); |
51cf4c1a Z |
792 | } |
793 | } | |
794 | ||
2833c59b JQ |
795 | void migrate_set_block_enabled(bool value, Error **errp) |
796 | { | |
797 | MigrationCapabilityStatusList *cap; | |
798 | ||
799 | cap = g_new0(MigrationCapabilityStatusList, 1); | |
800 | cap->value = g_new0(MigrationCapabilityStatus, 1); | |
801 | cap->value->capability = MIGRATION_CAPABILITY_BLOCK; | |
802 | cap->value->state = value; | |
803 | qmp_migrate_set_capabilities(cap, errp); | |
804 | qapi_free_MigrationCapabilityStatusList(cap); | |
805 | } | |
806 | ||
807 | static void migrate_set_block_incremental(MigrationState *s, bool value) | |
808 | { | |
809 | s->parameters.block_incremental = value; | |
810 | } | |
811 | ||
812 | static void block_cleanup_parameters(MigrationState *s) | |
813 | { | |
814 | if (s->must_remove_block_options) { | |
815 | /* setting to false can never fail */ | |
816 | migrate_set_block_enabled(false, &error_abort); | |
817 | migrate_set_block_incremental(s, false); | |
818 | s->must_remove_block_options = false; | |
819 | } | |
820 | } | |
821 | ||
bb1fadc4 | 822 | static void migrate_fd_cleanup(void *opaque) |
065e2813 | 823 | { |
bb1fadc4 PB |
824 | MigrationState *s = opaque; |
825 | ||
826 | qemu_bh_delete(s->cleanup_bh); | |
827 | s->cleanup_bh = NULL; | |
828 | ||
89a02a9f | 829 | if (s->to_dst_file) { |
9013dca5 | 830 | trace_migrate_fd_cleanup(); |
404a7c05 | 831 | qemu_mutex_unlock_iothread(); |
1d34e4bf DDAG |
832 | if (s->migration_thread_running) { |
833 | qemu_thread_join(&s->thread); | |
834 | s->migration_thread_running = false; | |
835 | } | |
404a7c05 PB |
836 | qemu_mutex_lock_iothread(); |
837 | ||
8706d2d5 | 838 | migrate_compress_threads_join(); |
89a02a9f HZ |
839 | qemu_fclose(s->to_dst_file); |
840 | s->to_dst_file = NULL; | |
065e2813 AL |
841 | } |
842 | ||
9ec055ae DDAG |
843 | assert((s->state != MIGRATION_STATUS_ACTIVE) && |
844 | (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE)); | |
7a2c1721 | 845 | |
94f5a437 | 846 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 847 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 848 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 849 | } |
a3fa1d78 PB |
850 | |
851 | notifier_list_notify(&migration_state_notifiers, s); | |
2833c59b | 852 | block_cleanup_parameters(s); |
065e2813 AL |
853 | } |
854 | ||
d59ce6f3 | 855 | void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 856 | { |
25174055 | 857 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 858 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
859 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
860 | MIGRATION_STATUS_FAILED); | |
d59ce6f3 DB |
861 | if (!s->error) { |
862 | s->error = error_copy(error); | |
863 | } | |
bb1fadc4 | 864 | notifier_list_notify(&migration_state_notifiers, s); |
2833c59b | 865 | block_cleanup_parameters(s); |
458cf28e JQ |
866 | } |
867 | ||
0edda1c4 | 868 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 869 | { |
6f2b811a | 870 | int old_state ; |
89a02a9f | 871 | QEMUFile *f = migrate_get_current()->to_dst_file; |
9013dca5 | 872 | trace_migrate_fd_cancel(); |
065e2813 | 873 | |
70b20477 DDAG |
874 | if (s->rp_state.from_dst_file) { |
875 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
876 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
877 | } | |
878 | ||
6f2b811a Z |
879 | do { |
880 | old_state = s->state; | |
f6844b99 | 881 | if (!migration_is_setup_or_active(old_state)) { |
6f2b811a Z |
882 | break; |
883 | } | |
48781e5b | 884 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 885 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
886 | |
887 | /* | |
888 | * If we're unlucky the migration code might be stuck somewhere in a | |
889 | * send/write while the network has failed and is waiting to timeout; | |
890 | * if we've got shutdown(2) available then we can force it to quit. | |
891 | * The outgoing qemu file gets closed in migrate_fd_cleanup that is | |
892 | * called in a bh, so there is no race against this cancel. | |
893 | */ | |
31194731 | 894 | if (s->state == MIGRATION_STATUS_CANCELLING && f) { |
a26ba26e DDAG |
895 | qemu_file_shutdown(f); |
896 | } | |
1d2acc31 HZ |
897 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
898 | Error *local_err = NULL; | |
899 | ||
900 | bdrv_invalidate_cache_all(&local_err); | |
901 | if (local_err) { | |
902 | error_report_err(local_err); | |
903 | } else { | |
904 | s->block_inactive = false; | |
905 | } | |
906 | } | |
2833c59b | 907 | block_cleanup_parameters(s); |
065e2813 AL |
908 | } |
909 | ||
99a0db9b GH |
910 | void add_migration_state_change_notifier(Notifier *notify) |
911 | { | |
912 | notifier_list_add(&migration_state_notifiers, notify); | |
913 | } | |
914 | ||
915 | void remove_migration_state_change_notifier(Notifier *notify) | |
916 | { | |
31552529 | 917 | notifier_remove(notify); |
99a0db9b GH |
918 | } |
919 | ||
02edd2e7 | 920 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 921 | { |
31194731 | 922 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
923 | } |
924 | ||
7073693b | 925 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 926 | { |
31194731 | 927 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 928 | } |
0edda1c4 | 929 | |
afe2df69 GH |
930 | bool migration_has_failed(MigrationState *s) |
931 | { | |
31194731 HZ |
932 | return (s->state == MIGRATION_STATUS_CANCELLED || |
933 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
934 | } |
935 | ||
5727309d | 936 | bool migration_in_postcopy(void) |
9ec055ae | 937 | { |
5727309d JQ |
938 | MigrationState *s = migrate_get_current(); |
939 | ||
9ec055ae DDAG |
940 | return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); |
941 | } | |
942 | ||
b82fc321 DDAG |
943 | bool migration_in_postcopy_after_devices(MigrationState *s) |
944 | { | |
5727309d | 945 | return migration_in_postcopy() && s->postcopy_after_devices; |
b82fc321 DDAG |
946 | } |
947 | ||
fab35005 | 948 | bool migration_is_idle(void) |
fe44dc91 | 949 | { |
fab35005 | 950 | MigrationState *s = migrate_get_current(); |
fe44dc91 AA |
951 | |
952 | switch (s->state) { | |
953 | case MIGRATION_STATUS_NONE: | |
954 | case MIGRATION_STATUS_CANCELLED: | |
955 | case MIGRATION_STATUS_COMPLETED: | |
956 | case MIGRATION_STATUS_FAILED: | |
957 | return true; | |
958 | case MIGRATION_STATUS_SETUP: | |
959 | case MIGRATION_STATUS_CANCELLING: | |
960 | case MIGRATION_STATUS_ACTIVE: | |
961 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
962 | case MIGRATION_STATUS_COLO: | |
963 | return false; | |
964 | case MIGRATION_STATUS__MAX: | |
965 | g_assert_not_reached(); | |
966 | } | |
967 | ||
968 | return false; | |
969 | } | |
970 | ||
a0762d9e | 971 | MigrationState *migrate_init(void) |
0edda1c4 | 972 | { |
17549e84 | 973 | MigrationState *s = migrate_get_current(); |
bbf6da32 | 974 | |
389775d1 DDAG |
975 | /* |
976 | * Reinitialise all migration state, except | |
977 | * parameters/capabilities that the user set, and | |
978 | * locks. | |
979 | */ | |
980 | s->bytes_xfer = 0; | |
981 | s->xfer_limit = 0; | |
982 | s->cleanup_bh = 0; | |
89a02a9f | 983 | s->to_dst_file = NULL; |
389775d1 | 984 | s->state = MIGRATION_STATUS_NONE; |
389775d1 DDAG |
985 | s->rp_state.from_dst_file = NULL; |
986 | s->rp_state.error = false; | |
987 | s->mbps = 0.0; | |
988 | s->downtime = 0; | |
989 | s->expected_downtime = 0; | |
389775d1 | 990 | s->setup_time = 0; |
389775d1 | 991 | s->start_postcopy = false; |
b82fc321 | 992 | s->postcopy_after_devices = false; |
389775d1 | 993 | s->migration_thread_running = false; |
d59ce6f3 DB |
994 | error_free(s->error); |
995 | s->error = NULL; | |
389775d1 | 996 | |
48781e5b | 997 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 998 | |
bc72ad67 | 999 | s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0edda1c4 JQ |
1000 | return s; |
1001 | } | |
cab30143 | 1002 | |
fa2756b7 AL |
1003 | static GSList *migration_blockers; |
1004 | ||
fe44dc91 | 1005 | int migrate_add_blocker(Error *reason, Error **errp) |
fa2756b7 | 1006 | { |
3df663e5 | 1007 | if (migrate_get_current()->only_migratable) { |
b67b8c3a AA |
1008 | error_propagate(errp, error_copy(reason)); |
1009 | error_prepend(errp, "disallowing migration blocker " | |
1010 | "(--only_migratable) for: "); | |
1011 | return -EACCES; | |
1012 | } | |
1013 | ||
fab35005 | 1014 | if (migration_is_idle()) { |
fe44dc91 AA |
1015 | migration_blockers = g_slist_prepend(migration_blockers, reason); |
1016 | return 0; | |
1017 | } | |
1018 | ||
1019 | error_propagate(errp, error_copy(reason)); | |
1020 | error_prepend(errp, "disallowing migration blocker (migration in " | |
1021 | "progress) for: "); | |
1022 | return -EBUSY; | |
fa2756b7 AL |
1023 | } |
1024 | ||
1025 | void migrate_del_blocker(Error *reason) | |
1026 | { | |
1027 | migration_blockers = g_slist_remove(migration_blockers, reason); | |
1028 | } | |
1029 | ||
bf1ae1f4 DDAG |
1030 | void qmp_migrate_incoming(const char *uri, Error **errp) |
1031 | { | |
1032 | Error *local_err = NULL; | |
4debb5f5 | 1033 | static bool once = true; |
bf1ae1f4 DDAG |
1034 | |
1035 | if (!deferred_incoming) { | |
4debb5f5 | 1036 | error_setg(errp, "For use with '-incoming defer'"); |
bf1ae1f4 DDAG |
1037 | return; |
1038 | } | |
4debb5f5 DDAG |
1039 | if (!once) { |
1040 | error_setg(errp, "The incoming migration has already been started"); | |
1041 | } | |
bf1ae1f4 DDAG |
1042 | |
1043 | qemu_start_incoming_migration(uri, &local_err); | |
1044 | ||
1045 | if (local_err) { | |
1046 | error_propagate(errp, local_err); | |
1047 | return; | |
1048 | } | |
1049 | ||
4debb5f5 | 1050 | once = false; |
bf1ae1f4 DDAG |
1051 | } |
1052 | ||
24f3902b GK |
1053 | bool migration_is_blocked(Error **errp) |
1054 | { | |
1055 | if (qemu_savevm_state_blocked(errp)) { | |
1056 | return true; | |
1057 | } | |
1058 | ||
1059 | if (migration_blockers) { | |
250561e1 | 1060 | error_propagate(errp, error_copy(migration_blockers->data)); |
24f3902b GK |
1061 | return true; |
1062 | } | |
1063 | ||
1064 | return false; | |
1065 | } | |
1066 | ||
e1c37d0e LC |
1067 | void qmp_migrate(const char *uri, bool has_blk, bool blk, |
1068 | bool has_inc, bool inc, bool has_detach, bool detach, | |
1069 | Error **errp) | |
cab30143 | 1070 | { |
be7059cd | 1071 | Error *local_err = NULL; |
17549e84 | 1072 | MigrationState *s = migrate_get_current(); |
cab30143 | 1073 | const char *p; |
cab30143 | 1074 | |
f6844b99 | 1075 | if (migration_is_setup_or_active(s->state) || |
0b827d5e HZ |
1076 | s->state == MIGRATION_STATUS_CANCELLING || |
1077 | s->state == MIGRATION_STATUS_COLO) { | |
c6bd8c70 | 1078 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
e1c37d0e | 1079 | return; |
cab30143 | 1080 | } |
ca99993a DDAG |
1081 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
1082 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
1083 | return; | |
1084 | } | |
1085 | ||
24f3902b | 1086 | if (migration_is_blocked(errp)) { |
e1c37d0e | 1087 | return; |
fa2756b7 AL |
1088 | } |
1089 | ||
2833c59b JQ |
1090 | if ((has_blk && blk) || (has_inc && inc)) { |
1091 | if (migrate_use_block() || migrate_use_block_incremental()) { | |
1092 | error_setg(errp, "Command options are incompatible with " | |
1093 | "current migration capabilities"); | |
1094 | return; | |
1095 | } | |
1096 | migrate_set_block_enabled(true, &local_err); | |
1097 | if (local_err) { | |
1098 | error_propagate(errp, local_err); | |
1099 | return; | |
1100 | } | |
1101 | s->must_remove_block_options = true; | |
1102 | } | |
1103 | ||
1104 | if (has_inc && inc) { | |
1105 | migrate_set_block_incremental(s, true); | |
1106 | } | |
1107 | ||
a0762d9e | 1108 | s = migrate_init(); |
cab30143 JQ |
1109 | |
1110 | if (strstart(uri, "tcp:", &p)) { | |
f37afb5a | 1111 | tcp_start_outgoing_migration(s, p, &local_err); |
2da776db | 1112 | #ifdef CONFIG_RDMA |
41310c68 | 1113 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
1114 | rdma_start_outgoing_migration(s, p, &local_err); |
1115 | #endif | |
cab30143 | 1116 | } else if (strstart(uri, "exec:", &p)) { |
f37afb5a | 1117 | exec_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1118 | } else if (strstart(uri, "unix:", &p)) { |
f37afb5a | 1119 | unix_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1120 | } else if (strstart(uri, "fd:", &p)) { |
f37afb5a | 1121 | fd_start_outgoing_migration(s, p, &local_err); |
99a0db9b | 1122 | } else { |
c6bd8c70 MA |
1123 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", |
1124 | "a valid migration protocol"); | |
48781e5b HZ |
1125 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1126 | MIGRATION_STATUS_FAILED); | |
e1c37d0e | 1127 | return; |
cab30143 JQ |
1128 | } |
1129 | ||
f37afb5a | 1130 | if (local_err) { |
d59ce6f3 | 1131 | migrate_fd_error(s, local_err); |
f37afb5a | 1132 | error_propagate(errp, local_err); |
e1c37d0e | 1133 | return; |
1299c631 | 1134 | } |
cab30143 JQ |
1135 | } |
1136 | ||
6cdedb07 | 1137 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 1138 | { |
17549e84 | 1139 | migrate_fd_cancel(migrate_get_current()); |
cab30143 JQ |
1140 | } |
1141 | ||
9e1ba4cc OW |
1142 | void qmp_migrate_set_cache_size(int64_t value, Error **errp) |
1143 | { | |
1144 | MigrationState *s = migrate_get_current(); | |
c91e681a | 1145 | int64_t new_size; |
9e1ba4cc OW |
1146 | |
1147 | /* Check for truncation */ | |
1148 | if (value != (size_t)value) { | |
c6bd8c70 MA |
1149 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1150 | "exceeding address space"); | |
9e1ba4cc OW |
1151 | return; |
1152 | } | |
1153 | ||
a5615b14 OW |
1154 | /* Cache should not be larger than guest ram size */ |
1155 | if (value > ram_bytes_total()) { | |
c6bd8c70 MA |
1156 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1157 | "exceeds guest ram size "); | |
a5615b14 OW |
1158 | return; |
1159 | } | |
1160 | ||
c91e681a OW |
1161 | new_size = xbzrle_cache_resize(value); |
1162 | if (new_size < 0) { | |
c6bd8c70 MA |
1163 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1164 | "is smaller than page size"); | |
c91e681a OW |
1165 | return; |
1166 | } | |
1167 | ||
1168 | s->xbzrle_cache_size = new_size; | |
9e1ba4cc OW |
1169 | } |
1170 | ||
1171 | int64_t qmp_query_migrate_cache_size(Error **errp) | |
1172 | { | |
1173 | return migrate_xbzrle_cache_size(); | |
1174 | } | |
1175 | ||
3dc85383 | 1176 | void qmp_migrate_set_speed(int64_t value, Error **errp) |
cab30143 | 1177 | { |
2ff30257 AA |
1178 | MigrationParameters p = { |
1179 | .has_max_bandwidth = true, | |
1180 | .max_bandwidth = value, | |
1181 | }; | |
cab30143 | 1182 | |
2ff30257 | 1183 | qmp_migrate_set_parameters(&p, errp); |
cab30143 JQ |
1184 | } |
1185 | ||
4f0a993b | 1186 | void qmp_migrate_set_downtime(double value, Error **errp) |
cab30143 | 1187 | { |
87c9cc1c DHB |
1188 | if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) { |
1189 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " | |
1190 | "the range of 0 to %d seconds", | |
1191 | MAX_MIGRATE_DOWNTIME_SECONDS); | |
1192 | return; | |
1193 | } | |
1194 | ||
2ff30257 AA |
1195 | value *= 1000; /* Convert to milliseconds */ |
1196 | value = MAX(0, MIN(INT64_MAX, value)); | |
1197 | ||
1198 | MigrationParameters p = { | |
1199 | .has_downtime_limit = true, | |
1200 | .downtime_limit = value, | |
1201 | }; | |
1202 | ||
1203 | qmp_migrate_set_parameters(&p, errp); | |
99a0db9b | 1204 | } |
17ad9b35 | 1205 | |
53f09a10 PB |
1206 | bool migrate_release_ram(void) |
1207 | { | |
1208 | MigrationState *s; | |
1209 | ||
1210 | s = migrate_get_current(); | |
1211 | ||
1212 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; | |
1213 | } | |
1214 | ||
53dd370c DDAG |
1215 | bool migrate_postcopy_ram(void) |
1216 | { | |
1217 | MigrationState *s; | |
1218 | ||
1219 | s = migrate_get_current(); | |
1220 | ||
32c3db5b | 1221 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
53dd370c DDAG |
1222 | } |
1223 | ||
bde1e2ec CV |
1224 | bool migrate_auto_converge(void) |
1225 | { | |
1226 | MigrationState *s; | |
1227 | ||
1228 | s = migrate_get_current(); | |
1229 | ||
1230 | return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; | |
1231 | } | |
1232 | ||
323004a3 PL |
1233 | bool migrate_zero_blocks(void) |
1234 | { | |
1235 | MigrationState *s; | |
1236 | ||
1237 | s = migrate_get_current(); | |
1238 | ||
1239 | return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; | |
1240 | } | |
1241 | ||
8706d2d5 LL |
1242 | bool migrate_use_compression(void) |
1243 | { | |
dde4e694 LL |
1244 | MigrationState *s; |
1245 | ||
1246 | s = migrate_get_current(); | |
1247 | ||
1248 | return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; | |
8706d2d5 LL |
1249 | } |
1250 | ||
1251 | int migrate_compress_level(void) | |
1252 | { | |
1253 | MigrationState *s; | |
1254 | ||
1255 | s = migrate_get_current(); | |
1256 | ||
2594f56d | 1257 | return s->parameters.compress_level; |
8706d2d5 LL |
1258 | } |
1259 | ||
1260 | int migrate_compress_threads(void) | |
1261 | { | |
1262 | MigrationState *s; | |
1263 | ||
1264 | s = migrate_get_current(); | |
1265 | ||
2594f56d | 1266 | return s->parameters.compress_threads; |
8706d2d5 LL |
1267 | } |
1268 | ||
3fcb38c2 LL |
1269 | int migrate_decompress_threads(void) |
1270 | { | |
1271 | MigrationState *s; | |
1272 | ||
1273 | s = migrate_get_current(); | |
1274 | ||
2594f56d | 1275 | return s->parameters.decompress_threads; |
3fcb38c2 LL |
1276 | } |
1277 | ||
b05dc723 JQ |
1278 | bool migrate_use_events(void) |
1279 | { | |
1280 | MigrationState *s; | |
1281 | ||
1282 | s = migrate_get_current(); | |
1283 | ||
1284 | return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; | |
1285 | } | |
1286 | ||
17ad9b35 OW |
1287 | int migrate_use_xbzrle(void) |
1288 | { | |
1289 | MigrationState *s; | |
1290 | ||
1291 | s = migrate_get_current(); | |
1292 | ||
1293 | return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; | |
1294 | } | |
1295 | ||
1296 | int64_t migrate_xbzrle_cache_size(void) | |
1297 | { | |
1298 | MigrationState *s; | |
1299 | ||
1300 | s = migrate_get_current(); | |
1301 | ||
1302 | return s->xbzrle_cache_size; | |
1303 | } | |
0d82d0e8 | 1304 | |
2833c59b JQ |
1305 | bool migrate_use_block(void) |
1306 | { | |
1307 | MigrationState *s; | |
1308 | ||
1309 | s = migrate_get_current(); | |
1310 | ||
1311 | return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK]; | |
1312 | } | |
1313 | ||
1314 | bool migrate_use_block_incremental(void) | |
1315 | { | |
1316 | MigrationState *s; | |
1317 | ||
1318 | s = migrate_get_current(); | |
1319 | ||
1320 | return s->parameters.block_incremental; | |
1321 | } | |
1322 | ||
70b20477 DDAG |
1323 | /* migration thread support */ |
1324 | /* | |
1325 | * Something bad happened to the RP stream, mark an error | |
1326 | * The caller shall print or trace something to indicate why | |
1327 | */ | |
1328 | static void mark_source_rp_bad(MigrationState *s) | |
1329 | { | |
1330 | s->rp_state.error = true; | |
1331 | } | |
1332 | ||
1333 | static struct rp_cmd_args { | |
1334 | ssize_t len; /* -1 = variable */ | |
1335 | const char *name; | |
1336 | } rp_cmd_args[] = { | |
1337 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
1338 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
1339 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
1340 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
1341 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
70b20477 DDAG |
1342 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
1343 | }; | |
1344 | ||
1e2d90eb DDAG |
1345 | /* |
1346 | * Process a request for pages received on the return path, | |
1347 | * We're allowed to send more than requested (e.g. to round to our page size) | |
1348 | * and we don't need to send pages that have already been sent. | |
1349 | */ | |
1350 | static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
1351 | ram_addr_t start, size_t len) | |
1352 | { | |
6c595cde DDAG |
1353 | long our_host_ps = getpagesize(); |
1354 | ||
1e2d90eb | 1355 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
1356 | |
1357 | /* | |
1358 | * Since we currently insist on matching page sizes, just sanity check | |
1359 | * we're being asked for whole host pages. | |
1360 | */ | |
1361 | if (start & (our_host_ps-1) || | |
1362 | (len & (our_host_ps-1))) { | |
1363 | error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT | |
1364 | " len: %zd", __func__, start, len); | |
1365 | mark_source_rp_bad(ms); | |
1366 | return; | |
1367 | } | |
1368 | ||
96506894 | 1369 | if (ram_save_queue_pages(rbname, start, len)) { |
6c595cde DDAG |
1370 | mark_source_rp_bad(ms); |
1371 | } | |
1e2d90eb DDAG |
1372 | } |
1373 | ||
70b20477 DDAG |
1374 | /* |
1375 | * Handles messages sent on the return path towards the source VM | |
1376 | * | |
1377 | */ | |
1378 | static void *source_return_path_thread(void *opaque) | |
1379 | { | |
1380 | MigrationState *ms = opaque; | |
1381 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
1382 | uint16_t header_len, header_type; | |
568b01ca | 1383 | uint8_t buf[512]; |
70b20477 | 1384 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
1385 | ram_addr_t start = 0; /* =0 to silence warning */ |
1386 | size_t len = 0, expected_len; | |
70b20477 DDAG |
1387 | int res; |
1388 | ||
1389 | trace_source_return_path_thread_entry(); | |
1390 | while (!ms->rp_state.error && !qemu_file_get_error(rp) && | |
1391 | migration_is_setup_or_active(ms->state)) { | |
1392 | trace_source_return_path_thread_loop_top(); | |
1393 | header_type = qemu_get_be16(rp); | |
1394 | header_len = qemu_get_be16(rp); | |
1395 | ||
1396 | if (header_type >= MIG_RP_MSG_MAX || | |
1397 | header_type == MIG_RP_MSG_INVALID) { | |
1398 | error_report("RP: Received invalid message 0x%04x length 0x%04x", | |
1399 | header_type, header_len); | |
1400 | mark_source_rp_bad(ms); | |
1401 | goto out; | |
1402 | } | |
1403 | ||
1404 | if ((rp_cmd_args[header_type].len != -1 && | |
1405 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 1406 | header_len > sizeof(buf)) { |
70b20477 DDAG |
1407 | error_report("RP: Received '%s' message (0x%04x) with" |
1408 | "incorrect length %d expecting %zu", | |
1409 | rp_cmd_args[header_type].name, header_type, header_len, | |
1410 | (size_t)rp_cmd_args[header_type].len); | |
1411 | mark_source_rp_bad(ms); | |
1412 | goto out; | |
1413 | } | |
1414 | ||
1415 | /* We know we've got a valid header by this point */ | |
1416 | res = qemu_get_buffer(rp, buf, header_len); | |
1417 | if (res != header_len) { | |
1418 | error_report("RP: Failed reading data for message 0x%04x" | |
1419 | " read %d expected %d", | |
1420 | header_type, res, header_len); | |
1421 | mark_source_rp_bad(ms); | |
1422 | goto out; | |
1423 | } | |
1424 | ||
1425 | /* OK, we have the message and the data */ | |
1426 | switch (header_type) { | |
1427 | case MIG_RP_MSG_SHUT: | |
4d885131 | 1428 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
1429 | trace_source_return_path_thread_shut(sibling_error); |
1430 | if (sibling_error) { | |
1431 | error_report("RP: Sibling indicated error %d", sibling_error); | |
1432 | mark_source_rp_bad(ms); | |
1433 | } | |
1434 | /* | |
1435 | * We'll let the main thread deal with closing the RP | |
1436 | * we could do a shutdown(2) on it, but we're the only user | |
1437 | * anyway, so there's nothing gained. | |
1438 | */ | |
1439 | goto out; | |
1440 | ||
1441 | case MIG_RP_MSG_PONG: | |
4d885131 | 1442 | tmp32 = ldl_be_p(buf); |
70b20477 DDAG |
1443 | trace_source_return_path_thread_pong(tmp32); |
1444 | break; | |
1445 | ||
1e2d90eb | 1446 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
1447 | start = ldq_be_p(buf); |
1448 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1449 | migrate_handle_rp_req_pages(ms, NULL, start, len); |
1450 | break; | |
1451 | ||
1452 | case MIG_RP_MSG_REQ_PAGES_ID: | |
1453 | expected_len = 12 + 1; /* header + termination */ | |
1454 | ||
1455 | if (header_len >= expected_len) { | |
4d885131 PM |
1456 | start = ldq_be_p(buf); |
1457 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1458 | /* Now we expect an idstr */ |
1459 | tmp32 = buf[12]; /* Length of the following idstr */ | |
1460 | buf[13 + tmp32] = '\0'; | |
1461 | expected_len += tmp32; | |
1462 | } | |
1463 | if (header_len != expected_len) { | |
1464 | error_report("RP: Req_Page_id with length %d expecting %zd", | |
1465 | header_len, expected_len); | |
1466 | mark_source_rp_bad(ms); | |
1467 | goto out; | |
1468 | } | |
1469 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); | |
1470 | break; | |
1471 | ||
70b20477 DDAG |
1472 | default: |
1473 | break; | |
1474 | } | |
1475 | } | |
5df5416e | 1476 | if (qemu_file_get_error(rp)) { |
70b20477 DDAG |
1477 | trace_source_return_path_thread_bad_end(); |
1478 | mark_source_rp_bad(ms); | |
1479 | } | |
1480 | ||
1481 | trace_source_return_path_thread_end(); | |
1482 | out: | |
1483 | ms->rp_state.from_dst_file = NULL; | |
1484 | qemu_fclose(rp); | |
1485 | return NULL; | |
1486 | } | |
1487 | ||
70b20477 DDAG |
1488 | static int open_return_path_on_source(MigrationState *ms) |
1489 | { | |
1490 | ||
89a02a9f | 1491 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
1492 | if (!ms->rp_state.from_dst_file) { |
1493 | return -1; | |
1494 | } | |
1495 | ||
1496 | trace_open_return_path_on_source(); | |
1497 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", | |
1498 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
1499 | ||
1500 | trace_open_return_path_on_source_continue(); | |
1501 | ||
1502 | return 0; | |
1503 | } | |
1504 | ||
70b20477 DDAG |
1505 | /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ |
1506 | static int await_return_path_close_on_source(MigrationState *ms) | |
1507 | { | |
1508 | /* | |
1509 | * If this is a normal exit then the destination will send a SHUT and the | |
1510 | * rp_thread will exit, however if there's an error we need to cause | |
1511 | * it to exit. | |
1512 | */ | |
89a02a9f | 1513 | if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { |
70b20477 DDAG |
1514 | /* |
1515 | * shutdown(2), if we have it, will cause it to unblock if it's stuck | |
1516 | * waiting for the destination. | |
1517 | */ | |
1518 | qemu_file_shutdown(ms->rp_state.from_dst_file); | |
1519 | mark_source_rp_bad(ms); | |
1520 | } | |
1521 | trace_await_return_path_close_on_source_joining(); | |
1522 | qemu_thread_join(&ms->rp_state.rp_thread); | |
1523 | trace_await_return_path_close_on_source_close(); | |
1524 | return ms->rp_state.error; | |
1525 | } | |
1526 | ||
1d34e4bf DDAG |
1527 | /* |
1528 | * Switch from normal iteration to postcopy | |
1529 | * Returns non-0 on error | |
1530 | */ | |
1531 | static int postcopy_start(MigrationState *ms, bool *old_vm_running) | |
1532 | { | |
1533 | int ret; | |
61b67d47 DB |
1534 | QIOChannelBuffer *bioc; |
1535 | QEMUFile *fb; | |
1d34e4bf | 1536 | int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
ef8d6488 | 1537 | bool restart_block = false; |
48781e5b | 1538 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, |
1d34e4bf DDAG |
1539 | MIGRATION_STATUS_POSTCOPY_ACTIVE); |
1540 | ||
1541 | trace_postcopy_start(); | |
1542 | qemu_mutex_lock_iothread(); | |
1543 | trace_postcopy_start_set_run(); | |
1544 | ||
1545 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); | |
1546 | *old_vm_running = runstate_is_running(); | |
1547 | global_state_store(); | |
1548 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
76b1c7fe KW |
1549 | if (ret < 0) { |
1550 | goto fail; | |
1551 | } | |
1d34e4bf | 1552 | |
76b1c7fe | 1553 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
1554 | if (ret < 0) { |
1555 | goto fail; | |
1556 | } | |
ef8d6488 | 1557 | restart_block = true; |
1d34e4bf | 1558 | |
1c0d249d DDAG |
1559 | /* |
1560 | * Cause any non-postcopiable, but iterative devices to | |
1561 | * send out their final data. | |
1562 | */ | |
a1fbe750 | 1563 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); |
1c0d249d | 1564 | |
1d34e4bf DDAG |
1565 | /* |
1566 | * in Finish migrate and with the io-lock held everything should | |
1567 | * be quiet, but we've potentially still got dirty pages and we | |
1568 | * need to tell the destination to throw any pages it's already received | |
1569 | * that are dirty | |
1570 | */ | |
1571 | if (ram_postcopy_send_discard_bitmap(ms)) { | |
1572 | error_report("postcopy send discard bitmap failed"); | |
1573 | goto fail; | |
1574 | } | |
1575 | ||
1576 | /* | |
1577 | * send rest of state - note things that are doing postcopy | |
1578 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
1579 | * wrap their state up here | |
1580 | */ | |
89a02a9f | 1581 | qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); |
1d34e4bf | 1582 | /* Ping just for debugging, helps line traces up */ |
89a02a9f | 1583 | qemu_savevm_send_ping(ms->to_dst_file, 2); |
1d34e4bf DDAG |
1584 | |
1585 | /* | |
1586 | * While loading the device state we may trigger page transfer | |
1587 | * requests and the fd must be free to process those, and thus | |
1588 | * the destination must read the whole device state off the fd before | |
1589 | * it starts processing it. Unfortunately the ad-hoc migration format | |
1590 | * doesn't allow the destination to know the size to read without fully | |
1591 | * parsing it through each devices load-state code (especially the open | |
1592 | * coded devices that use get/put). | |
1593 | * So we wrap the device state up in a package with a length at the start; | |
1594 | * to do this we use a qemu_buf to hold the whole of the device state. | |
1595 | */ | |
61b67d47 | 1596 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 1597 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
61b67d47 DB |
1598 | fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc)); |
1599 | object_unref(OBJECT(bioc)); | |
1d34e4bf | 1600 | |
c76201ab DDAG |
1601 | /* |
1602 | * Make sure the receiver can get incoming pages before we send the rest | |
1603 | * of the state | |
1604 | */ | |
1605 | qemu_savevm_send_postcopy_listen(fb); | |
1606 | ||
a1fbe750 | 1607 | qemu_savevm_state_complete_precopy(fb, false, false); |
1d34e4bf DDAG |
1608 | qemu_savevm_send_ping(fb, 3); |
1609 | ||
1610 | qemu_savevm_send_postcopy_run(fb); | |
1611 | ||
1612 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 1613 | |
ef8d6488 DDAG |
1614 | /* Last point of recovery; as soon as we send the package the destination |
1615 | * can open devices and potentially start running. | |
1616 | * Lets just check again we've not got any errors. | |
1617 | */ | |
1618 | ret = qemu_file_get_error(ms->to_dst_file); | |
1619 | if (ret) { | |
1620 | error_report("postcopy_start: Migration stream errored (pre package)"); | |
1621 | goto fail_closefb; | |
1622 | } | |
1623 | ||
1624 | restart_block = false; | |
1625 | ||
1d34e4bf | 1626 | /* Now send that blob */ |
61b67d47 | 1627 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
1628 | goto fail_closefb; |
1629 | } | |
1630 | qemu_fclose(fb); | |
b82fc321 DDAG |
1631 | |
1632 | /* Send a notify to give a chance for anything that needs to happen | |
1633 | * at the transition to postcopy and after the device state; in particular | |
1634 | * spice needs to trigger a transition now | |
1635 | */ | |
1636 | ms->postcopy_after_devices = true; | |
1637 | notifier_list_notify(&migration_state_notifiers, ms); | |
1638 | ||
1d34e4bf DDAG |
1639 | ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; |
1640 | ||
1641 | qemu_mutex_unlock_iothread(); | |
1642 | ||
1643 | /* | |
1644 | * Although this ping is just for debug, it could potentially be | |
1645 | * used for getting a better measurement of downtime at the source. | |
1646 | */ | |
89a02a9f | 1647 | qemu_savevm_send_ping(ms->to_dst_file, 4); |
1d34e4bf | 1648 | |
ced1c616 PB |
1649 | if (migrate_release_ram()) { |
1650 | ram_postcopy_migrated_memory_release(ms); | |
1651 | } | |
1652 | ||
89a02a9f | 1653 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf DDAG |
1654 | if (ret) { |
1655 | error_report("postcopy_start: Migration stream errored"); | |
48781e5b | 1656 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf DDAG |
1657 | MIGRATION_STATUS_FAILED); |
1658 | } | |
1659 | ||
1660 | return ret; | |
1661 | ||
1662 | fail_closefb: | |
1663 | qemu_fclose(fb); | |
1664 | fail: | |
48781e5b | 1665 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 1666 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
1667 | if (restart_block) { |
1668 | /* A failure happened early enough that we know the destination hasn't | |
1669 | * accessed block devices, so we're safe to recover. | |
1670 | */ | |
1671 | Error *local_err = NULL; | |
1672 | ||
1673 | bdrv_invalidate_cache_all(&local_err); | |
1674 | if (local_err) { | |
1675 | error_report_err(local_err); | |
1676 | } | |
1677 | } | |
1d34e4bf DDAG |
1678 | qemu_mutex_unlock_iothread(); |
1679 | return -1; | |
1680 | } | |
1681 | ||
09f6c85e DDAG |
1682 | /** |
1683 | * migration_completion: Used by migration_thread when there's not much left. | |
1684 | * The caller 'breaks' the loop when this returns. | |
1685 | * | |
1686 | * @s: Current migration state | |
36f48567 | 1687 | * @current_active_state: The migration state we expect to be in |
09f6c85e DDAG |
1688 | * @*old_vm_running: Pointer to old_vm_running flag |
1689 | * @*start_time: Pointer to time to update | |
1690 | */ | |
36f48567 DDAG |
1691 | static void migration_completion(MigrationState *s, int current_active_state, |
1692 | bool *old_vm_running, | |
09f6c85e DDAG |
1693 | int64_t *start_time) |
1694 | { | |
1695 | int ret; | |
1696 | ||
b10ac0c4 DDAG |
1697 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
1698 | qemu_mutex_lock_iothread(); | |
1699 | *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
1700 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); | |
1701 | *old_vm_running = runstate_is_running(); | |
1702 | ret = global_state_store(); | |
1703 | ||
1704 | if (!ret) { | |
a1fbe750 | 1705 | bool inactivate = !migrate_colo_enabled(); |
b10ac0c4 | 1706 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); |
f07fa4cb KW |
1707 | if (ret >= 0) { |
1708 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); | |
a1fbe750 FZ |
1709 | ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, |
1710 | inactivate); | |
f07fa4cb | 1711 | } |
a1fbe750 FZ |
1712 | if (inactivate && ret >= 0) { |
1713 | s->block_inactive = true; | |
b10ac0c4 DDAG |
1714 | } |
1715 | } | |
1716 | qemu_mutex_unlock_iothread(); | |
09f6c85e | 1717 | |
b10ac0c4 DDAG |
1718 | if (ret < 0) { |
1719 | goto fail; | |
09f6c85e | 1720 | } |
b10ac0c4 DDAG |
1721 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { |
1722 | trace_migration_completion_postcopy_end(); | |
1723 | ||
89a02a9f | 1724 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
b10ac0c4 | 1725 | trace_migration_completion_postcopy_end_after_complete(); |
09f6c85e | 1726 | } |
09f6c85e | 1727 | |
b10ac0c4 DDAG |
1728 | /* |
1729 | * If rp was opened we must clean up the thread before | |
1730 | * cleaning everything else up (since if there are no failures | |
1731 | * it will wait for the destination to send it's status in | |
1732 | * a SHUT command). | |
b10ac0c4 | 1733 | */ |
0425dc97 | 1734 | if (s->rp_state.from_dst_file) { |
b10ac0c4 | 1735 | int rp_error; |
0425dc97 | 1736 | trace_migration_return_path_end_before(); |
b10ac0c4 | 1737 | rp_error = await_return_path_close_on_source(s); |
0425dc97 | 1738 | trace_migration_return_path_end_after(rp_error); |
b10ac0c4 | 1739 | if (rp_error) { |
fe904ea8 | 1740 | goto fail_invalidate; |
b10ac0c4 | 1741 | } |
09f6c85e DDAG |
1742 | } |
1743 | ||
89a02a9f | 1744 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 1745 | trace_migration_completion_file_err(); |
fe904ea8 | 1746 | goto fail_invalidate; |
09f6c85e DDAG |
1747 | } |
1748 | ||
0b827d5e HZ |
1749 | if (!migrate_colo_enabled()) { |
1750 | migrate_set_state(&s->state, current_active_state, | |
1751 | MIGRATION_STATUS_COMPLETED); | |
1752 | } | |
1753 | ||
09f6c85e DDAG |
1754 | return; |
1755 | ||
fe904ea8 GK |
1756 | fail_invalidate: |
1757 | /* If not doing postcopy, vm_start() will be called: let's regain | |
1758 | * control on images. | |
1759 | */ | |
1760 | if (s->state == MIGRATION_STATUS_ACTIVE) { | |
1761 | Error *local_err = NULL; | |
1762 | ||
1d2acc31 | 1763 | qemu_mutex_lock_iothread(); |
fe904ea8 GK |
1764 | bdrv_invalidate_cache_all(&local_err); |
1765 | if (local_err) { | |
1766 | error_report_err(local_err); | |
1d2acc31 HZ |
1767 | } else { |
1768 | s->block_inactive = false; | |
fe904ea8 | 1769 | } |
1d2acc31 | 1770 | qemu_mutex_unlock_iothread(); |
fe904ea8 GK |
1771 | } |
1772 | ||
09f6c85e | 1773 | fail: |
48781e5b HZ |
1774 | migrate_set_state(&s->state, current_active_state, |
1775 | MIGRATION_STATUS_FAILED); | |
09f6c85e DDAG |
1776 | } |
1777 | ||
35a6ed4f HZ |
1778 | bool migrate_colo_enabled(void) |
1779 | { | |
1780 | MigrationState *s = migrate_get_current(); | |
1781 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; | |
1782 | } | |
1783 | ||
70b20477 DDAG |
1784 | /* |
1785 | * Master migration thread on the source VM. | |
1786 | * It drives the migration and pumps the data down the outgoing channel. | |
1787 | */ | |
5f496a1b | 1788 | static void *migration_thread(void *opaque) |
0d82d0e8 | 1789 | { |
9848a404 | 1790 | MigrationState *s = opaque; |
1d34e4bf | 1791 | /* Used by the bandwidth calcs, updated later */ |
bc72ad67 AB |
1792 | int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1793 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
be7172e2 | 1794 | int64_t initial_bytes = 0; |
faec066a PX |
1795 | /* |
1796 | * The final stage happens when the remaining data is smaller than | |
1797 | * this threshold; it's calculated from the requested downtime and | |
1798 | * measured bandwidth | |
1799 | */ | |
1800 | int64_t threshold_size = 0; | |
a3fa1d78 | 1801 | int64_t start_time = initial_time; |
94f5a437 | 1802 | int64_t end_time; |
a3fa1d78 | 1803 | bool old_vm_running = false; |
1d34e4bf DDAG |
1804 | bool entered_postcopy = false; |
1805 | /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */ | |
1806 | enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE; | |
0b827d5e | 1807 | bool enable_colo = migrate_colo_enabled(); |
76f5933a | 1808 | |
ab28bd23 PB |
1809 | rcu_register_thread(); |
1810 | ||
89a02a9f | 1811 | qemu_savevm_state_header(s->to_dst_file); |
1d34e4bf | 1812 | |
62a02658 PX |
1813 | /* |
1814 | * If we opened the return path, we need to make sure dst has it | |
1815 | * opened as well. | |
1816 | */ | |
1817 | if (s->rp_state.from_dst_file) { | |
1d34e4bf | 1818 | /* Now tell the dest that it should open its end so it can reply */ |
89a02a9f | 1819 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
1820 | |
1821 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 1822 | qemu_savevm_send_ping(s->to_dst_file, 1); |
0425dc97 | 1823 | } |
1d34e4bf | 1824 | |
0425dc97 | 1825 | if (migrate_postcopy_ram()) { |
1d34e4bf DDAG |
1826 | /* |
1827 | * Tell the destination that we *might* want to do postcopy later; | |
1828 | * if the other end can't do postcopy it should fail now, nice and | |
1829 | * early. | |
1830 | */ | |
89a02a9f | 1831 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
1832 | } |
1833 | ||
a0762d9e | 1834 | qemu_savevm_state_begin(s->to_dst_file); |
0d82d0e8 | 1835 | |
bc72ad67 | 1836 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
48781e5b HZ |
1837 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1838 | MIGRATION_STATUS_ACTIVE); | |
29ae8a41 | 1839 | |
9ec055ae DDAG |
1840 | trace_migration_thread_setup_complete(); |
1841 | ||
1842 | while (s->state == MIGRATION_STATUS_ACTIVE || | |
1843 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
a3e879cd | 1844 | int64_t current_time; |
c369f40d | 1845 | uint64_t pending_size; |
0d82d0e8 | 1846 | |
89a02a9f | 1847 | if (!qemu_file_rate_limit(s->to_dst_file)) { |
c31b098f DDAG |
1848 | uint64_t pend_post, pend_nonpost; |
1849 | ||
faec066a PX |
1850 | qemu_savevm_state_pending(s->to_dst_file, threshold_size, |
1851 | &pend_nonpost, &pend_post); | |
c31b098f | 1852 | pending_size = pend_nonpost + pend_post; |
faec066a | 1853 | trace_migrate_pending(pending_size, threshold_size, |
c31b098f | 1854 | pend_post, pend_nonpost); |
faec066a | 1855 | if (pending_size && pending_size >= threshold_size) { |
1d34e4bf DDAG |
1856 | /* Still a significant amount to transfer */ |
1857 | ||
1d34e4bf DDAG |
1858 | if (migrate_postcopy_ram() && |
1859 | s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE && | |
faec066a | 1860 | pend_nonpost <= threshold_size && |
1d34e4bf DDAG |
1861 | atomic_read(&s->start_postcopy)) { |
1862 | ||
1863 | if (!postcopy_start(s, &old_vm_running)) { | |
1864 | current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE; | |
1865 | entered_postcopy = true; | |
1866 | } | |
1867 | ||
1868 | continue; | |
1869 | } | |
1870 | /* Just another iteration step */ | |
89a02a9f | 1871 | qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy); |
c369f40d | 1872 | } else { |
09f6c85e | 1873 | trace_migration_thread_low_pending(pending_size); |
1d34e4bf | 1874 | migration_completion(s, current_active_state, |
36f48567 | 1875 | &old_vm_running, &start_time); |
09f6c85e | 1876 | break; |
c369f40d JQ |
1877 | } |
1878 | } | |
f4410a5d | 1879 | |
89a02a9f | 1880 | if (qemu_file_get_error(s->to_dst_file)) { |
48781e5b HZ |
1881 | migrate_set_state(&s->state, current_active_state, |
1882 | MIGRATION_STATUS_FAILED); | |
1d34e4bf | 1883 | trace_migration_thread_file_err(); |
fd45ee2c PB |
1884 | break; |
1885 | } | |
bc72ad67 | 1886 | current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0d82d0e8 | 1887 | if (current_time >= initial_time + BUFFER_DELAY) { |
89a02a9f HZ |
1888 | uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) - |
1889 | initial_bytes; | |
77417f10 | 1890 | uint64_t time_spent = current_time - initial_time; |
a694ee34 | 1891 | double bandwidth = (double)transferred_bytes / time_spent; |
faec066a | 1892 | threshold_size = bandwidth * s->parameters.downtime_limit; |
0d82d0e8 | 1893 | |
5b648de0 WY |
1894 | s->mbps = (((double) transferred_bytes * 8.0) / |
1895 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
7e114f8c | 1896 | |
9013dca5 | 1897 | trace_migrate_transferred(transferred_bytes, time_spent, |
faec066a | 1898 | bandwidth, threshold_size); |
90f8ae72 JQ |
1899 | /* if we haven't sent anything, we don't want to recalculate |
1900 | 10000 is a small enough number for our purposes */ | |
9360447d JQ |
1901 | if (ram_counters.dirty_pages_rate && transferred_bytes > 10000) { |
1902 | s->expected_downtime = ram_counters.dirty_pages_rate * | |
20afaed9 | 1903 | qemu_target_page_size() / bandwidth; |
90f8ae72 | 1904 | } |
0d82d0e8 | 1905 | |
89a02a9f | 1906 | qemu_file_reset_rate_limit(s->to_dst_file); |
0d82d0e8 | 1907 | initial_time = current_time; |
89a02a9f | 1908 | initial_bytes = qemu_ftell(s->to_dst_file); |
0d82d0e8 | 1909 | } |
89a02a9f | 1910 | if (qemu_file_rate_limit(s->to_dst_file)) { |
0d82d0e8 JQ |
1911 | /* usleep expects microseconds */ |
1912 | g_usleep((initial_time + BUFFER_DELAY - current_time)*1000); | |
1913 | } | |
a3fa1d78 PB |
1914 | } |
1915 | ||
1d34e4bf | 1916 | trace_migration_thread_after_loop(); |
070afca2 JH |
1917 | /* If we enabled cpu throttling for auto-converge, turn it off. */ |
1918 | cpu_throttle_stop(); | |
94f5a437 | 1919 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
070afca2 | 1920 | |
f4410a5d | 1921 | qemu_mutex_lock_iothread(); |
0b827d5e HZ |
1922 | /* |
1923 | * The resource has been allocated by migration will be reused in COLO | |
1924 | * process, so don't release them. | |
1925 | */ | |
1926 | if (!enable_colo) { | |
1927 | qemu_savevm_state_cleanup(); | |
1928 | } | |
31194731 | 1929 | if (s->state == MIGRATION_STATUS_COMPLETED) { |
89a02a9f | 1930 | uint64_t transferred_bytes = qemu_ftell(s->to_dst_file); |
a3fa1d78 | 1931 | s->total_time = end_time - s->total_time; |
1d34e4bf DDAG |
1932 | if (!entered_postcopy) { |
1933 | s->downtime = end_time - start_time; | |
1934 | } | |
d6ed7312 PL |
1935 | if (s->total_time) { |
1936 | s->mbps = (((double) transferred_bytes * 8.0) / | |
1937 | ((double) s->total_time)) / 1000; | |
1938 | } | |
a3fa1d78 PB |
1939 | runstate_set(RUN_STATE_POSTMIGRATE); |
1940 | } else { | |
0b827d5e HZ |
1941 | if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) { |
1942 | migrate_start_colo_process(s); | |
1943 | qemu_savevm_state_cleanup(); | |
1944 | /* | |
1945 | * Fixme: we will run VM in COLO no matter its old running state. | |
1946 | * After exited COLO, we will keep running. | |
1947 | */ | |
1948 | old_vm_running = true; | |
1949 | } | |
1d34e4bf | 1950 | if (old_vm_running && !entered_postcopy) { |
a3fa1d78 | 1951 | vm_start(); |
42da5550 DDAG |
1952 | } else { |
1953 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
1954 | runstate_set(RUN_STATE_POSTMIGRATE); | |
1955 | } | |
dba433c0 | 1956 | } |
0d82d0e8 | 1957 | } |
bb1fadc4 | 1958 | qemu_bh_schedule(s->cleanup_bh); |
dba433c0 | 1959 | qemu_mutex_unlock_iothread(); |
f4410a5d | 1960 | |
ab28bd23 | 1961 | rcu_unregister_thread(); |
0d82d0e8 JQ |
1962 | return NULL; |
1963 | } | |
1964 | ||
9848a404 | 1965 | void migrate_fd_connect(MigrationState *s) |
0d82d0e8 | 1966 | { |
2ff30257 | 1967 | s->expected_downtime = s->parameters.downtime_limit; |
bb1fadc4 | 1968 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s); |
0d82d0e8 | 1969 | |
9e4d2b98 | 1970 | qemu_file_set_blocking(s->to_dst_file, true); |
89a02a9f | 1971 | qemu_file_set_rate_limit(s->to_dst_file, |
2ff30257 | 1972 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); |
442773ce | 1973 | |
9287ac27 SH |
1974 | /* Notify before starting migration thread */ |
1975 | notifier_list_notify(&migration_state_notifiers, s); | |
1976 | ||
1d34e4bf DDAG |
1977 | /* |
1978 | * Open the return path; currently for postcopy but other things might | |
1979 | * also want it. | |
1980 | */ | |
1981 | if (migrate_postcopy_ram()) { | |
1982 | if (open_return_path_on_source(s)) { | |
1983 | error_report("Unable to open return-path for postcopy"); | |
48781e5b | 1984 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1d34e4bf DDAG |
1985 | MIGRATION_STATUS_FAILED); |
1986 | migrate_fd_cleanup(s); | |
1987 | return; | |
1988 | } | |
1989 | } | |
1990 | ||
8706d2d5 | 1991 | migrate_compress_threads_create(); |
009fad7f | 1992 | qemu_thread_create(&s->thread, "live_migration", migration_thread, s, |
bb1fadc4 | 1993 | QEMU_THREAD_JOINABLE); |
1d34e4bf | 1994 | s->migration_thread_running = true; |
0d82d0e8 | 1995 | } |
093e3c42 | 1996 | |
9d18af93 PX |
1997 | void migration_global_dump(Monitor *mon) |
1998 | { | |
1999 | MigrationState *ms = migrate_get_current(); | |
2000 | ||
2001 | monitor_printf(mon, "globals: store-global-state=%d, only_migratable=%d, " | |
2002 | "send-configuration=%d, send-section-footer=%d\n", | |
2003 | ms->store_global_state, ms->only_migratable, | |
2004 | ms->send_configuration, ms->send_section_footer); | |
2005 | } | |
2006 | ||
5272298c PX |
2007 | static Property migration_properties[] = { |
2008 | DEFINE_PROP_BOOL("store-global-state", MigrationState, | |
2009 | store_global_state, true), | |
3df663e5 | 2010 | DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false), |
71dd4c1a PX |
2011 | DEFINE_PROP_BOOL("send-configuration", MigrationState, |
2012 | send_configuration, true), | |
15c38503 PX |
2013 | DEFINE_PROP_BOOL("send-section-footer", MigrationState, |
2014 | send_section_footer, true), | |
5272298c PX |
2015 | DEFINE_PROP_END_OF_LIST(), |
2016 | }; | |
2017 | ||
e5cb7e76 PX |
2018 | static void migration_class_init(ObjectClass *klass, void *data) |
2019 | { | |
2020 | DeviceClass *dc = DEVICE_CLASS(klass); | |
2021 | ||
2022 | dc->user_creatable = false; | |
5272298c | 2023 | dc->props = migration_properties; |
e5cb7e76 PX |
2024 | } |
2025 | ||
2026 | static void migration_instance_init(Object *obj) | |
2027 | { | |
2028 | MigrationState *ms = MIGRATION_OBJ(obj); | |
2029 | ||
2030 | ms->state = MIGRATION_STATUS_NONE; | |
2031 | ms->xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE; | |
2032 | ms->mbps = -1; | |
2033 | ms->parameters = (MigrationParameters) { | |
2034 | .compress_level = DEFAULT_MIGRATE_COMPRESS_LEVEL, | |
2035 | .compress_threads = DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT, | |
2036 | .decompress_threads = DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT, | |
2037 | .cpu_throttle_initial = DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL, | |
2038 | .cpu_throttle_increment = DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT, | |
2039 | .max_bandwidth = MAX_THROTTLE, | |
2040 | .downtime_limit = DEFAULT_MIGRATE_SET_DOWNTIME, | |
2041 | .x_checkpoint_delay = DEFAULT_MIGRATE_X_CHECKPOINT_DELAY, | |
2042 | }; | |
2043 | ms->parameters.tls_creds = g_strdup(""); | |
2044 | ms->parameters.tls_hostname = g_strdup(""); | |
2045 | } | |
2046 | ||
2047 | static const TypeInfo migration_type = { | |
2048 | .name = TYPE_MIGRATION, | |
01f6e14c PX |
2049 | /* |
2050 | * NOTE: "migration" itself is not really a device. We used | |
2051 | * TYPE_DEVICE here only to leverage some existing QDev features | |
2052 | * like "-global" properties, and HW_COMPAT_* fields (which are | |
2053 | * finally applied as global properties as well). If one day the | |
2054 | * global property feature can be migrated from QDev to QObject in | |
2055 | * general, then we can switch to QObject as well. | |
2056 | */ | |
e5cb7e76 PX |
2057 | .parent = TYPE_DEVICE, |
2058 | .class_init = migration_class_init, | |
2059 | .class_size = sizeof(MigrationClass), | |
2060 | .instance_size = sizeof(MigrationState), | |
2061 | .instance_init = migration_instance_init, | |
2062 | }; | |
2063 | ||
2064 | static void register_migration_types(void) | |
2065 | { | |
2066 | type_register_static(&migration_type); | |
2067 | } | |
2068 | ||
2069 | type_init(register_migration_types); |