]>
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" |
ab28bd23 | 34 | #include "qemu/rcu.h" |
2c9e6fec | 35 | #include "block.h" |
be07b0ac | 36 | #include "postcopy-ram.h" |
766bd176 | 37 | #include "qemu/thread.h" |
791e7c82 | 38 | #include "qmp-commands.h" |
c09e5bb1 | 39 | #include "trace.h" |
598cd2bd | 40 | #include "qapi-event.h" |
51180423 | 41 | #include "exec/target_page.h" |
61b67d47 | 42 | #include "io/channel-buffer.h" |
35a6ed4f | 43 | #include "migration/colo.h" |
4ffdb337 | 44 | #include "hw/boards.h" |
9d18af93 | 45 | #include "monitor/monitor.h" |
065e2813 | 46 | |
dc325627 | 47 | #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */ |
5bb7910a | 48 | |
5b4e1eb7 JQ |
49 | /* Amount of time to allocate to each "chunk" of bandwidth-throttled |
50 | * data. */ | |
51 | #define BUFFER_DELAY 100 | |
52 | #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) | |
53 | ||
2ff30257 AA |
54 | /* Time in milliseconds we are allowed to stop the source, |
55 | * for sending the last part */ | |
56 | #define DEFAULT_MIGRATE_SET_DOWNTIME 300 | |
57 | ||
87c9cc1c DHB |
58 | /* Maximum migrate downtime set to 2000 seconds */ |
59 | #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 | |
60 | #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) | |
61 | ||
8706d2d5 LL |
62 | /* Default compression thread count */ |
63 | #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 | |
3fcb38c2 LL |
64 | /* Default decompression thread count, usually decompression is at |
65 | * least 4 times as fast as compression.*/ | |
66 | #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 | |
8706d2d5 LL |
67 | /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
68 | #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 | |
1626fee3 | 69 | /* Define default autoconverge cpu throttle migration parameters */ |
d85a31d1 JH |
70 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 |
71 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 | |
8706d2d5 | 72 | |
17ad9b35 OW |
73 | /* Migration XBZRLE default cache size */ |
74 | #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024) | |
75 | ||
68b53591 HZ |
76 | /* The delay time (in ms) between two COLO checkpoints |
77 | * Note: Please change this default value to 10000 when we support hybrid mode. | |
78 | */ | |
79 | #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200 | |
80 | ||
99a0db9b GH |
81 | static NotifierList migration_state_notifiers = |
82 | NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); | |
83 | ||
adde220a DDAG |
84 | static bool deferred_incoming; |
85 | ||
da6f1790 JQ |
86 | /* Messages sent on the return path from destination to source */ |
87 | enum mig_rp_message_type { | |
88 | MIG_RP_MSG_INVALID = 0, /* Must be 0 */ | |
89 | MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ | |
90 | MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ | |
91 | ||
92 | MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ | |
93 | MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ | |
94 | ||
95 | MIG_RP_MSG_MAX | |
96 | }; | |
97 | ||
17549e84 JQ |
98 | /* When we add fault tolerance, we could have several |
99 | migrations at once. For now we don't need to add | |
100 | dynamic creation of migration */ | |
101 | ||
e5cb7e76 PX |
102 | static MigrationState *current_migration; |
103 | ||
8b0b29dc PX |
104 | static bool migration_object_check(MigrationState *ms, Error **errp); |
105 | ||
e5cb7e76 PX |
106 | void migration_object_init(void) |
107 | { | |
4ffdb337 | 108 | MachineState *ms = MACHINE(qdev_get_machine()); |
8b0b29dc | 109 | Error *err = NULL; |
4ffdb337 | 110 | |
e5cb7e76 PX |
111 | /* This can only be called once. */ |
112 | assert(!current_migration); | |
113 | current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); | |
4ffdb337 | 114 | |
8b0b29dc PX |
115 | if (!migration_object_check(current_migration, &err)) { |
116 | error_report_err(err); | |
117 | exit(1); | |
118 | } | |
119 | ||
4ffdb337 PX |
120 | /* |
121 | * We cannot really do this in migration_instance_init() since at | |
122 | * that time global properties are not yet applied, then this | |
123 | * value will be definitely replaced by something else. | |
124 | */ | |
125 | if (ms->enforce_config_section) { | |
126 | current_migration->send_configuration = true; | |
127 | } | |
e5cb7e76 PX |
128 | } |
129 | ||
bca7856a | 130 | /* For outgoing */ |
859bc756 | 131 | MigrationState *migrate_get_current(void) |
17549e84 | 132 | { |
e5cb7e76 PX |
133 | /* This can only be called after the object created. */ |
134 | assert(current_migration); | |
135 | return current_migration; | |
17549e84 JQ |
136 | } |
137 | ||
bca7856a DDAG |
138 | MigrationIncomingState *migration_incoming_get_current(void) |
139 | { | |
b4b076da JQ |
140 | static bool once; |
141 | static MigrationIncomingState mis_current; | |
bca7856a | 142 | |
b4b076da JQ |
143 | if (!once) { |
144 | mis_current.state = MIGRATION_STATUS_NONE; | |
145 | memset(&mis_current, 0, sizeof(MigrationIncomingState)); | |
b4b076da JQ |
146 | qemu_mutex_init(&mis_current.rp_mutex); |
147 | qemu_event_init(&mis_current.main_thread_load_event, false); | |
148 | once = true; | |
149 | } | |
150 | return &mis_current; | |
bca7856a DDAG |
151 | } |
152 | ||
153 | void migration_incoming_state_destroy(void) | |
154 | { | |
b4b076da JQ |
155 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
156 | ||
3482655b | 157 | if (mis->to_src_file) { |
660819b1 PX |
158 | /* Tell source that we are done */ |
159 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
160 | qemu_fclose(mis->to_src_file); |
161 | mis->to_src_file = NULL; | |
162 | } | |
163 | ||
660819b1 PX |
164 | if (mis->from_src_file) { |
165 | qemu_fclose(mis->from_src_file); | |
166 | mis->from_src_file = NULL; | |
167 | } | |
168 | ||
5089e186 | 169 | qemu_event_reset(&mis->main_thread_load_event); |
bca7856a DDAG |
170 | } |
171 | ||
b05dc723 JQ |
172 | static void migrate_generate_event(int new_state) |
173 | { | |
174 | if (migrate_use_events()) { | |
175 | qapi_event_send_migration(new_state, &error_abort); | |
b05dc723 JQ |
176 | } |
177 | } | |
178 | ||
adde220a DDAG |
179 | /* |
180 | * Called on -incoming with a defer: uri. | |
181 | * The migration can be started later after any parameters have been | |
182 | * changed. | |
183 | */ | |
184 | static void deferred_incoming_migration(Error **errp) | |
185 | { | |
186 | if (deferred_incoming) { | |
187 | error_setg(errp, "Incoming migration already deferred"); | |
188 | } | |
189 | deferred_incoming = true; | |
190 | } | |
191 | ||
da6f1790 JQ |
192 | /* |
193 | * Send a message on the return channel back to the source | |
194 | * of the migration. | |
195 | */ | |
196 | static void migrate_send_rp_message(MigrationIncomingState *mis, | |
197 | enum mig_rp_message_type message_type, | |
198 | uint16_t len, void *data) | |
199 | { | |
200 | trace_migrate_send_rp_message((int)message_type, len); | |
201 | qemu_mutex_lock(&mis->rp_mutex); | |
202 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); | |
203 | qemu_put_be16(mis->to_src_file, len); | |
204 | qemu_put_buffer(mis->to_src_file, data, len); | |
205 | qemu_fflush(mis->to_src_file); | |
206 | qemu_mutex_unlock(&mis->rp_mutex); | |
207 | } | |
208 | ||
1e2d90eb DDAG |
209 | /* Request a range of pages from the source VM at the given |
210 | * start address. | |
211 | * rbname: Name of the RAMBlock to request the page in, if NULL it's the same | |
212 | * as the last request (a name must have been given previously) | |
213 | * Start: Address offset within the RB | |
214 | * Len: Length in bytes required - must be a multiple of pagesize | |
215 | */ | |
216 | void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname, | |
217 | ram_addr_t start, size_t len) | |
218 | { | |
cb8d4c8f | 219 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb DDAG |
220 | size_t msglen = 12; /* start + len */ |
221 | ||
222 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
223 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
224 | ||
225 | if (rbname) { | |
226 | int rbname_len = strlen(rbname); | |
227 | assert(rbname_len < 256); | |
228 | ||
229 | bufc[msglen++] = rbname_len; | |
230 | memcpy(bufc + msglen, rbname, rbname_len); | |
231 | msglen += rbname_len; | |
232 | migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc); | |
233 | } else { | |
234 | migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc); | |
235 | } | |
236 | } | |
237 | ||
43eaae28 | 238 | void qemu_start_incoming_migration(const char *uri, Error **errp) |
5bb7910a | 239 | { |
34c9dd8e AL |
240 | const char *p; |
241 | ||
7cf1fe6d | 242 | qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort); |
adde220a DDAG |
243 | if (!strcmp(uri, "defer")) { |
244 | deferred_incoming_migration(errp); | |
245 | } else if (strstart(uri, "tcp:", &p)) { | |
43eaae28 | 246 | tcp_start_incoming_migration(p, errp); |
2da776db | 247 | #ifdef CONFIG_RDMA |
adde220a | 248 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
249 | rdma_start_incoming_migration(p, errp); |
250 | #endif | |
adde220a | 251 | } else if (strstart(uri, "exec:", &p)) { |
43eaae28 | 252 | exec_start_incoming_migration(p, errp); |
adde220a | 253 | } else if (strstart(uri, "unix:", &p)) { |
43eaae28 | 254 | unix_start_incoming_migration(p, errp); |
adde220a | 255 | } else if (strstart(uri, "fd:", &p)) { |
43eaae28 | 256 | fd_start_incoming_migration(p, errp); |
adde220a | 257 | } else { |
312fd5f2 | 258 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 259 | } |
5bb7910a AL |
260 | } |
261 | ||
0aa6aefc DL |
262 | static void process_incoming_migration_bh(void *opaque) |
263 | { | |
264 | Error *local_err = NULL; | |
265 | MigrationIncomingState *mis = opaque; | |
266 | ||
ace21a58 KW |
267 | /* Make sure all file formats flush their mutable metadata. |
268 | * If we get an error here, just don't restart the VM yet. */ | |
0aa6aefc | 269 | bdrv_invalidate_cache_all(&local_err); |
d35ff5e6 | 270 | if (local_err) { |
ace21a58 | 271 | error_report_err(local_err); |
d35ff5e6 KW |
272 | local_err = NULL; |
273 | autostart = false; | |
274 | } | |
275 | ||
0aa6aefc DL |
276 | /* |
277 | * This must happen after all error conditions are dealt with and | |
278 | * we're sure the VM is going to be running on this host. | |
279 | */ | |
280 | qemu_announce_self(); | |
281 | ||
282 | /* If global state section was not received or we are in running | |
283 | state, we need to obey autostart. Any other state is set with | |
284 | runstate_set. */ | |
285 | ||
286 | if (!global_state_received() || | |
287 | global_state_get_runstate() == RUN_STATE_RUNNING) { | |
288 | if (autostart) { | |
289 | vm_start(); | |
290 | } else { | |
291 | runstate_set(RUN_STATE_PAUSED); | |
292 | } | |
293 | } else { | |
294 | runstate_set(global_state_get_runstate()); | |
295 | } | |
0aa6aefc DL |
296 | /* |
297 | * This must happen after any state changes since as soon as an external | |
298 | * observer sees this event they might start to prod at the VM assuming | |
299 | * it's ready to use. | |
300 | */ | |
301 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
302 | MIGRATION_STATUS_COMPLETED); | |
303 | qemu_bh_delete(mis->bh); | |
304 | migration_incoming_state_destroy(); | |
305 | } | |
306 | ||
82a4da79 | 307 | static void process_incoming_migration_co(void *opaque) |
511c0231 | 308 | { |
82a4da79 | 309 | QEMUFile *f = opaque; |
b4b076da | 310 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 311 | PostcopyState ps; |
1c12e1f5 PB |
312 | int ret; |
313 | ||
b4b076da | 314 | mis->from_src_file = f; |
67f11b5c | 315 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 316 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
93d7af6f HZ |
317 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
318 | MIGRATION_STATUS_ACTIVE); | |
1c12e1f5 | 319 | ret = qemu_loadvm_state(f); |
bca7856a | 320 | |
e9bef235 DDAG |
321 | ps = postcopy_state_get(); |
322 | trace_process_incoming_migration_co_end(ret, ps); | |
323 | if (ps != POSTCOPY_INCOMING_NONE) { | |
324 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
325 | /* | |
326 | * Where a migration had postcopy enabled (and thus went to advise) | |
327 | * but managed to complete within the precopy period, we can use | |
328 | * the normal exit. | |
329 | */ | |
330 | postcopy_ram_incoming_cleanup(mis); | |
331 | } else if (ret >= 0) { | |
332 | /* | |
333 | * Postcopy was started, cleanup should happen at the end of the | |
334 | * postcopy thread. | |
335 | */ | |
336 | trace_process_incoming_migration_co_postcopy_end_main(); | |
337 | return; | |
338 | } | |
339 | /* Else if something went wrong then just fall out of the normal exit */ | |
340 | } | |
341 | ||
25d0c16f HZ |
342 | /* we get COLO info, and know if we are in COLO mode */ |
343 | if (!ret && migration_incoming_enable_colo()) { | |
344 | mis->migration_incoming_co = qemu_coroutine_self(); | |
345 | qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", | |
346 | colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); | |
347 | mis->have_colo_incoming_thread = true; | |
348 | qemu_coroutine_yield(); | |
349 | ||
350 | /* Wait checkpoint incoming thread exit before free resource */ | |
351 | qemu_thread_join(&mis->colo_incoming_thread); | |
352 | } | |
353 | ||
1c12e1f5 | 354 | if (ret < 0) { |
93d7af6f HZ |
355 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, |
356 | MIGRATION_STATUS_FAILED); | |
db80face | 357 | error_report("load of migration failed: %s", strerror(-ret)); |
3a0f2cea | 358 | qemu_fclose(mis->from_src_file); |
4aead692 | 359 | exit(EXIT_FAILURE); |
511c0231 | 360 | } |
0aa6aefc DL |
361 | mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); |
362 | qemu_bh_schedule(mis->bh); | |
511c0231 JQ |
363 | } |
364 | ||
22724f49 | 365 | void migration_fd_process_incoming(QEMUFile *f) |
82a4da79 | 366 | { |
0b8b8753 | 367 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f); |
82a4da79 | 368 | |
06ad5135 | 369 | qemu_file_set_blocking(f, false); |
0b8b8753 | 370 | qemu_coroutine_enter(co); |
82a4da79 PB |
371 | } |
372 | ||
6decec93 DDAG |
373 | /* |
374 | * Send a 'SHUT' message on the return channel with the given value | |
375 | * to indicate that we've finished with the RP. Non-0 value indicates | |
376 | * error. | |
377 | */ | |
378 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
379 | uint32_t value) | |
380 | { | |
381 | uint32_t buf; | |
382 | ||
383 | buf = cpu_to_be32(value); | |
384 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
385 | } | |
386 | ||
387 | /* | |
388 | * Send a 'PONG' message on the return channel with the given value | |
389 | * (normally in response to a 'PING') | |
390 | */ | |
391 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
392 | uint32_t value) | |
393 | { | |
394 | uint32_t buf; | |
395 | ||
396 | buf = cpu_to_be32(value); | |
397 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
398 | } | |
399 | ||
bbf6da32 OW |
400 | MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) |
401 | { | |
402 | MigrationCapabilityStatusList *head = NULL; | |
403 | MigrationCapabilityStatusList *caps; | |
404 | MigrationState *s = migrate_get_current(); | |
405 | int i; | |
406 | ||
387eedeb | 407 | caps = NULL; /* silence compiler warning */ |
7fb1cf16 | 408 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
ed1701c6 DDAG |
409 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
410 | if (i == MIGRATION_CAPABILITY_BLOCK) { | |
411 | continue; | |
412 | } | |
413 | #endif | |
bbf6da32 OW |
414 | if (head == NULL) { |
415 | head = g_malloc0(sizeof(*caps)); | |
416 | caps = head; | |
417 | } else { | |
418 | caps->next = g_malloc0(sizeof(*caps)); | |
419 | caps = caps->next; | |
420 | } | |
421 | caps->value = | |
422 | g_malloc(sizeof(*caps->value)); | |
423 | caps->value->capability = i; | |
424 | caps->value->state = s->enabled_capabilities[i]; | |
425 | } | |
426 | ||
427 | return head; | |
428 | } | |
429 | ||
85de8323 LL |
430 | MigrationParameters *qmp_query_migrate_parameters(Error **errp) |
431 | { | |
432 | MigrationParameters *params; | |
433 | MigrationState *s = migrate_get_current(); | |
434 | ||
e87fae4c | 435 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
85de8323 | 436 | params = g_malloc0(sizeof(*params)); |
de63ab61 | 437 | params->has_compress_level = true; |
2594f56d | 438 | params->compress_level = s->parameters.compress_level; |
de63ab61 | 439 | params->has_compress_threads = true; |
2594f56d | 440 | params->compress_threads = s->parameters.compress_threads; |
de63ab61 | 441 | params->has_decompress_threads = true; |
2594f56d | 442 | params->decompress_threads = s->parameters.decompress_threads; |
de63ab61 | 443 | params->has_cpu_throttle_initial = true; |
2594f56d | 444 | params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; |
de63ab61 | 445 | params->has_cpu_throttle_increment = true; |
2594f56d | 446 | params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; |
8cc99dcd | 447 | params->has_tls_creds = true; |
69ef1f36 | 448 | params->tls_creds = g_strdup(s->parameters.tls_creds); |
8cc99dcd | 449 | params->has_tls_hostname = true; |
69ef1f36 | 450 | params->tls_hostname = g_strdup(s->parameters.tls_hostname); |
2ff30257 AA |
451 | params->has_max_bandwidth = true; |
452 | params->max_bandwidth = s->parameters.max_bandwidth; | |
453 | params->has_downtime_limit = true; | |
454 | params->downtime_limit = s->parameters.downtime_limit; | |
fe39a4d4 | 455 | params->has_x_checkpoint_delay = true; |
68b53591 | 456 | params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; |
2833c59b JQ |
457 | params->has_block_incremental = true; |
458 | params->block_incremental = s->parameters.block_incremental; | |
85de8323 LL |
459 | |
460 | return params; | |
461 | } | |
462 | ||
f6844b99 DDAG |
463 | /* |
464 | * Return true if we're already in the middle of a migration | |
465 | * (i.e. any of the active or setup states) | |
466 | */ | |
467 | static bool migration_is_setup_or_active(int state) | |
468 | { | |
469 | switch (state) { | |
470 | case MIGRATION_STATUS_ACTIVE: | |
9ec055ae | 471 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
f6844b99 DDAG |
472 | case MIGRATION_STATUS_SETUP: |
473 | return true; | |
474 | ||
475 | default: | |
476 | return false; | |
477 | ||
478 | } | |
479 | } | |
480 | ||
a22463a5 DDAG |
481 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
482 | { | |
483 | info->has_ram = true; | |
484 | info->ram = g_malloc0(sizeof(*info->ram)); | |
9360447d | 485 | info->ram->transferred = ram_counters.transferred; |
a22463a5 | 486 | info->ram->total = ram_bytes_total(); |
9360447d | 487 | info->ram->duplicate = ram_counters.duplicate; |
bedf53c1 JQ |
488 | /* legacy value. It is not used anymore */ |
489 | info->ram->skipped = 0; | |
9360447d JQ |
490 | info->ram->normal = ram_counters.normal; |
491 | info->ram->normal_bytes = ram_counters.normal * | |
20afaed9 | 492 | qemu_target_page_size(); |
a22463a5 | 493 | info->ram->mbps = s->mbps; |
9360447d JQ |
494 | info->ram->dirty_sync_count = ram_counters.dirty_sync_count; |
495 | info->ram->postcopy_requests = ram_counters.postcopy_requests; | |
030ce1f8 | 496 | info->ram->page_size = qemu_target_page_size(); |
a22463a5 | 497 | |
114f5aee JQ |
498 | if (migrate_use_xbzrle()) { |
499 | info->has_xbzrle_cache = true; | |
500 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); | |
501 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
9360447d JQ |
502 | info->xbzrle_cache->bytes = xbzrle_counters.bytes; |
503 | info->xbzrle_cache->pages = xbzrle_counters.pages; | |
504 | info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss; | |
505 | info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate; | |
506 | info->xbzrle_cache->overflow = xbzrle_counters.overflow; | |
114f5aee JQ |
507 | } |
508 | ||
338182c8 JQ |
509 | if (cpu_throttle_active()) { |
510 | info->has_cpu_throttle_percentage = true; | |
511 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
512 | } | |
513 | ||
a22463a5 DDAG |
514 | if (s->state != MIGRATION_STATUS_COMPLETED) { |
515 | info->ram->remaining = ram_bytes_remaining(); | |
9360447d | 516 | info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate; |
a22463a5 DDAG |
517 | } |
518 | } | |
519 | ||
930ac04c JQ |
520 | static void populate_disk_info(MigrationInfo *info) |
521 | { | |
522 | if (blk_mig_active()) { | |
523 | info->has_disk = true; | |
524 | info->disk = g_malloc0(sizeof(*info->disk)); | |
525 | info->disk->transferred = blk_mig_bytes_transferred(); | |
526 | info->disk->remaining = blk_mig_bytes_remaining(); | |
527 | info->disk->total = blk_mig_bytes_total(); | |
528 | } | |
529 | } | |
530 | ||
791e7c82 | 531 | MigrationInfo *qmp_query_migrate(Error **errp) |
5bb7910a | 532 | { |
791e7c82 | 533 | MigrationInfo *info = g_malloc0(sizeof(*info)); |
17549e84 JQ |
534 | MigrationState *s = migrate_get_current(); |
535 | ||
536 | switch (s->state) { | |
31194731 | 537 | case MIGRATION_STATUS_NONE: |
17549e84 JQ |
538 | /* no migration has happened ever */ |
539 | break; | |
31194731 | 540 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 541 | info->has_status = true; |
ed4fbd10 | 542 | info->has_total_time = false; |
29ae8a41 | 543 | break; |
31194731 HZ |
544 | case MIGRATION_STATUS_ACTIVE: |
545 | case MIGRATION_STATUS_CANCELLING: | |
9ec055ae | 546 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
c8f9f4f4 | 547 | /* TODO add some postcopy stats */ |
9ec055ae DDAG |
548 | info->has_status = true; |
549 | info->has_total_time = true; | |
550 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) | |
551 | - s->total_time; | |
552 | info->has_expected_downtime = true; | |
553 | info->expected_downtime = s->expected_downtime; | |
554 | info->has_setup_time = true; | |
555 | info->setup_time = s->setup_time; | |
556 | ||
a22463a5 | 557 | populate_ram_info(info, s); |
930ac04c | 558 | populate_disk_info(info); |
17549e84 | 559 | break; |
0b827d5e HZ |
560 | case MIGRATION_STATUS_COLO: |
561 | info->has_status = true; | |
562 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
563 | break; | |
31194731 | 564 | case MIGRATION_STATUS_COMPLETED: |
791e7c82 | 565 | info->has_status = true; |
00c14997 | 566 | info->has_total_time = true; |
7aa939af | 567 | info->total_time = s->total_time; |
9c5a9fcf JQ |
568 | info->has_downtime = true; |
569 | info->downtime = s->downtime; | |
ed4fbd10 MH |
570 | info->has_setup_time = true; |
571 | info->setup_time = s->setup_time; | |
d5f8a570 | 572 | |
a22463a5 | 573 | populate_ram_info(info, s); |
17549e84 | 574 | break; |
31194731 | 575 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 576 | info->has_status = true; |
d59ce6f3 DB |
577 | if (s->error) { |
578 | info->has_error_desc = true; | |
579 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
580 | } | |
17549e84 | 581 | break; |
31194731 | 582 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 583 | info->has_status = true; |
17549e84 | 584 | break; |
5bb7910a | 585 | } |
cde63fbe | 586 | info->status = s->state; |
791e7c82 LC |
587 | |
588 | return info; | |
5bb7910a AL |
589 | } |
590 | ||
4a84214e PX |
591 | /** |
592 | * @migration_caps_check - check capability validity | |
593 | * | |
594 | * @cap_list: old capability list, array of bool | |
595 | * @params: new capabilities to be applied soon | |
596 | * @errp: set *errp if the check failed, with reason | |
597 | * | |
598 | * Returns true if check passed, otherwise false. | |
599 | */ | |
600 | static bool migrate_caps_check(bool *cap_list, | |
601 | MigrationCapabilityStatusList *params, | |
602 | Error **errp) | |
00458433 | 603 | { |
00458433 | 604 | MigrationCapabilityStatusList *cap; |
4a84214e | 605 | bool old_postcopy_cap; |
00458433 | 606 | |
4a84214e | 607 | old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
00458433 OW |
608 | |
609 | for (cap = params; cap; cap = cap->next) { | |
4a84214e PX |
610 | cap_list[cap->value->capability] = cap->value->state; |
611 | } | |
612 | ||
ed1701c6 | 613 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
4a84214e PX |
614 | if (cap_list[MIGRATION_CAPABILITY_BLOCK]) { |
615 | error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) " | |
616 | "block migration"); | |
617 | error_append_hint(errp, "Use drive_mirror+NBD instead.\n"); | |
618 | return false; | |
00458433 | 619 | } |
4a84214e | 620 | #endif |
53dd370c | 621 | |
4a84214e PX |
622 | if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) { |
623 | if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) { | |
53dd370c DDAG |
624 | /* The decompression threads asynchronously write into RAM |
625 | * rather than use the atomic copies needed to avoid | |
626 | * userfaulting. It should be possible to fix the decompression | |
627 | * threads for compatibility in future. | |
628 | */ | |
4a84214e PX |
629 | error_setg(errp, "Postcopy is not currently compatible " |
630 | "with compression"); | |
631 | return false; | |
53dd370c | 632 | } |
4a84214e | 633 | |
096631bd DDAG |
634 | /* This check is reasonably expensive, so only when it's being |
635 | * set the first time, also it's only the destination that needs | |
636 | * special support. | |
637 | */ | |
638 | if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) && | |
639 | !postcopy_ram_supported_by_host()) { | |
640 | /* postcopy_ram_supported_by_host will have emitted a more | |
641 | * detailed message | |
642 | */ | |
4a84214e PX |
643 | error_setg(errp, "Postcopy is not supported"); |
644 | return false; | |
096631bd | 645 | } |
53dd370c | 646 | } |
4a84214e PX |
647 | |
648 | return true; | |
649 | } | |
650 | ||
651 | void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params, | |
652 | Error **errp) | |
653 | { | |
654 | MigrationState *s = migrate_get_current(); | |
655 | MigrationCapabilityStatusList *cap; | |
656 | ||
657 | if (migration_is_setup_or_active(s->state)) { | |
658 | error_setg(errp, QERR_MIGRATION_ACTIVE); | |
659 | return; | |
660 | } | |
661 | ||
662 | if (!migrate_caps_check(s->enabled_capabilities, params, errp)) { | |
663 | return; | |
664 | } | |
665 | ||
666 | for (cap = params; cap; cap = cap->next) { | |
667 | s->enabled_capabilities[cap->value->capability] = cap->value->state; | |
668 | } | |
00458433 OW |
669 | } |
670 | ||
16d063bc PX |
671 | /* |
672 | * Check whether the parameters are valid. Error will be put into errp | |
673 | * (if provided). Return true if valid, otherwise false. | |
674 | */ | |
675 | static bool migrate_params_check(MigrationParameters *params, Error **errp) | |
85de8323 | 676 | { |
7f375e04 EB |
677 | if (params->has_compress_level && |
678 | (params->compress_level < 0 || params->compress_level > 9)) { | |
c6bd8c70 MA |
679 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", |
680 | "is invalid, it should be in the range of 0 to 9"); | |
16d063bc | 681 | return false; |
85de8323 | 682 | } |
16d063bc | 683 | |
7f375e04 EB |
684 | if (params->has_compress_threads && |
685 | (params->compress_threads < 1 || params->compress_threads > 255)) { | |
c6bd8c70 MA |
686 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
687 | "compress_threads", | |
688 | "is invalid, it should be in the range of 1 to 255"); | |
16d063bc | 689 | return false; |
85de8323 | 690 | } |
16d063bc | 691 | |
7f375e04 EB |
692 | if (params->has_decompress_threads && |
693 | (params->decompress_threads < 1 || params->decompress_threads > 255)) { | |
c6bd8c70 MA |
694 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
695 | "decompress_threads", | |
696 | "is invalid, it should be in the range of 1 to 255"); | |
16d063bc | 697 | return false; |
85de8323 | 698 | } |
16d063bc | 699 | |
7f375e04 EB |
700 | if (params->has_cpu_throttle_initial && |
701 | (params->cpu_throttle_initial < 1 || | |
702 | params->cpu_throttle_initial > 99)) { | |
1626fee3 | 703 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 704 | "cpu_throttle_initial", |
1626fee3 | 705 | "an integer in the range of 1 to 99"); |
16d063bc | 706 | return false; |
1626fee3 | 707 | } |
16d063bc | 708 | |
7f375e04 EB |
709 | if (params->has_cpu_throttle_increment && |
710 | (params->cpu_throttle_increment < 1 || | |
711 | params->cpu_throttle_increment > 99)) { | |
1626fee3 | 712 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 713 | "cpu_throttle_increment", |
1626fee3 | 714 | "an integer in the range of 1 to 99"); |
16d063bc | 715 | return false; |
1626fee3 | 716 | } |
16d063bc | 717 | |
2ff30257 AA |
718 | if (params->has_max_bandwidth && |
719 | (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) { | |
720 | error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the" | |
721 | " range of 0 to %zu bytes/second", SIZE_MAX); | |
16d063bc | 722 | return false; |
2ff30257 | 723 | } |
16d063bc | 724 | |
2ff30257 | 725 | if (params->has_downtime_limit && |
87c9cc1c DHB |
726 | (params->downtime_limit < 0 || |
727 | params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { | |
728 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " | |
729 | "the range of 0 to %d milliseconds", | |
730 | MAX_MIGRATE_DOWNTIME); | |
16d063bc | 731 | return false; |
2ff30257 | 732 | } |
16d063bc | 733 | |
68b53591 HZ |
734 | if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) { |
735 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
736 | "x_checkpoint_delay", | |
737 | "is invalid, it should be positive"); | |
16d063bc PX |
738 | return false; |
739 | } | |
740 | ||
741 | return true; | |
742 | } | |
743 | ||
1bda8b3c MA |
744 | static void migrate_params_test_apply(MigrateSetParameters *params, |
745 | MigrationParameters *dest) | |
746 | { | |
747 | *dest = migrate_get_current()->parameters; | |
748 | ||
749 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ | |
750 | ||
751 | if (params->has_compress_level) { | |
752 | dest->compress_level = params->compress_level; | |
753 | } | |
754 | ||
755 | if (params->has_compress_threads) { | |
756 | dest->compress_threads = params->compress_threads; | |
757 | } | |
758 | ||
759 | if (params->has_decompress_threads) { | |
760 | dest->decompress_threads = params->decompress_threads; | |
761 | } | |
762 | ||
763 | if (params->has_cpu_throttle_initial) { | |
764 | dest->cpu_throttle_initial = params->cpu_throttle_initial; | |
765 | } | |
766 | ||
767 | if (params->has_cpu_throttle_increment) { | |
768 | dest->cpu_throttle_increment = params->cpu_throttle_increment; | |
769 | } | |
770 | ||
771 | if (params->has_tls_creds) { | |
01fa5598 MA |
772 | assert(params->tls_creds->type == QTYPE_QSTRING); |
773 | dest->tls_creds = g_strdup(params->tls_creds->u.s); | |
1bda8b3c MA |
774 | } |
775 | ||
776 | if (params->has_tls_hostname) { | |
01fa5598 MA |
777 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
778 | dest->tls_hostname = g_strdup(params->tls_hostname->u.s); | |
1bda8b3c MA |
779 | } |
780 | ||
781 | if (params->has_max_bandwidth) { | |
782 | dest->max_bandwidth = params->max_bandwidth; | |
783 | } | |
784 | ||
785 | if (params->has_downtime_limit) { | |
786 | dest->downtime_limit = params->downtime_limit; | |
787 | } | |
788 | ||
789 | if (params->has_x_checkpoint_delay) { | |
790 | dest->x_checkpoint_delay = params->x_checkpoint_delay; | |
791 | } | |
792 | ||
793 | if (params->has_block_incremental) { | |
794 | dest->block_incremental = params->block_incremental; | |
795 | } | |
796 | } | |
797 | ||
798 | static void migrate_params_apply(MigrateSetParameters *params) | |
16d063bc PX |
799 | { |
800 | MigrationState *s = migrate_get_current(); | |
801 | ||
e87fae4c MA |
802 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
803 | ||
7f375e04 EB |
804 | if (params->has_compress_level) { |
805 | s->parameters.compress_level = params->compress_level; | |
85de8323 | 806 | } |
476c72aa | 807 | |
7f375e04 EB |
808 | if (params->has_compress_threads) { |
809 | s->parameters.compress_threads = params->compress_threads; | |
85de8323 | 810 | } |
476c72aa | 811 | |
7f375e04 EB |
812 | if (params->has_decompress_threads) { |
813 | s->parameters.decompress_threads = params->decompress_threads; | |
85de8323 | 814 | } |
476c72aa | 815 | |
7f375e04 EB |
816 | if (params->has_cpu_throttle_initial) { |
817 | s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; | |
1626fee3 | 818 | } |
476c72aa | 819 | |
7f375e04 EB |
820 | if (params->has_cpu_throttle_increment) { |
821 | s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; | |
1626fee3 | 822 | } |
476c72aa | 823 | |
7f375e04 | 824 | if (params->has_tls_creds) { |
69ef1f36 | 825 | g_free(s->parameters.tls_creds); |
01fa5598 MA |
826 | assert(params->tls_creds->type == QTYPE_QSTRING); |
827 | s->parameters.tls_creds = g_strdup(params->tls_creds->u.s); | |
69ef1f36 | 828 | } |
476c72aa | 829 | |
7f375e04 | 830 | if (params->has_tls_hostname) { |
69ef1f36 | 831 | g_free(s->parameters.tls_hostname); |
01fa5598 MA |
832 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
833 | s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s); | |
69ef1f36 | 834 | } |
476c72aa | 835 | |
2ff30257 AA |
836 | if (params->has_max_bandwidth) { |
837 | s->parameters.max_bandwidth = params->max_bandwidth; | |
838 | if (s->to_dst_file) { | |
839 | qemu_file_set_rate_limit(s->to_dst_file, | |
840 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); | |
841 | } | |
842 | } | |
476c72aa | 843 | |
2ff30257 AA |
844 | if (params->has_downtime_limit) { |
845 | s->parameters.downtime_limit = params->downtime_limit; | |
846 | } | |
68b53591 HZ |
847 | |
848 | if (params->has_x_checkpoint_delay) { | |
849 | s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; | |
479125d5 HZ |
850 | if (migration_in_colo_state()) { |
851 | colo_checkpoint_notify(s); | |
852 | } | |
68b53591 | 853 | } |
476c72aa | 854 | |
2833c59b JQ |
855 | if (params->has_block_incremental) { |
856 | s->parameters.block_incremental = params->block_incremental; | |
857 | } | |
85de8323 LL |
858 | } |
859 | ||
1bda8b3c | 860 | void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp) |
476c72aa | 861 | { |
1bda8b3c MA |
862 | MigrationParameters tmp; |
863 | ||
01fa5598 MA |
864 | /* TODO Rewrite "" to null instead */ |
865 | if (params->has_tls_creds | |
866 | && params->tls_creds->type == QTYPE_QNULL) { | |
867 | QDECREF(params->tls_creds->u.n); | |
868 | params->tls_creds->type = QTYPE_QSTRING; | |
869 | params->tls_creds->u.s = strdup(""); | |
870 | } | |
871 | /* TODO Rewrite "" to null instead */ | |
872 | if (params->has_tls_hostname | |
873 | && params->tls_hostname->type == QTYPE_QNULL) { | |
874 | QDECREF(params->tls_hostname->u.n); | |
875 | params->tls_hostname->type = QTYPE_QSTRING; | |
876 | params->tls_hostname->u.s = strdup(""); | |
877 | } | |
878 | ||
1bda8b3c MA |
879 | migrate_params_test_apply(params, &tmp); |
880 | ||
881 | if (!migrate_params_check(&tmp, errp)) { | |
476c72aa PX |
882 | /* Invalid parameter */ |
883 | return; | |
884 | } | |
885 | ||
886 | migrate_params_apply(params); | |
887 | } | |
888 | ||
2594f56d | 889 | |
4886a1bc DDAG |
890 | void qmp_migrate_start_postcopy(Error **errp) |
891 | { | |
892 | MigrationState *s = migrate_get_current(); | |
893 | ||
894 | if (!migrate_postcopy_ram()) { | |
a54d340b | 895 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
896 | " the start of migration"); |
897 | return; | |
898 | } | |
899 | ||
900 | if (s->state == MIGRATION_STATUS_NONE) { | |
901 | error_setg(errp, "Postcopy must be started after migration has been" | |
902 | " started"); | |
903 | return; | |
904 | } | |
905 | /* | |
906 | * we don't error if migration has finished since that would be racy | |
907 | * with issuing this command. | |
908 | */ | |
909 | atomic_set(&s->start_postcopy, true); | |
910 | } | |
911 | ||
065e2813 AL |
912 | /* shared migration helpers */ |
913 | ||
48781e5b | 914 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 915 | { |
a31fedee | 916 | assert(new_state < MIGRATION_STATUS__MAX); |
48781e5b | 917 | if (atomic_cmpxchg(state, old_state, new_state) == old_state) { |
a31fedee | 918 | trace_migrate_set_state(MigrationStatus_str(new_state)); |
b05dc723 | 919 | migrate_generate_event(new_state); |
51cf4c1a Z |
920 | } |
921 | } | |
922 | ||
4e4a3d3a PX |
923 | static MigrationCapabilityStatusList *migrate_cap_add( |
924 | MigrationCapabilityStatusList *list, | |
925 | MigrationCapability index, | |
926 | bool state) | |
2833c59b JQ |
927 | { |
928 | MigrationCapabilityStatusList *cap; | |
929 | ||
930 | cap = g_new0(MigrationCapabilityStatusList, 1); | |
931 | cap->value = g_new0(MigrationCapabilityStatus, 1); | |
4e4a3d3a PX |
932 | cap->value->capability = index; |
933 | cap->value->state = state; | |
934 | cap->next = list; | |
935 | ||
936 | return cap; | |
937 | } | |
938 | ||
939 | void migrate_set_block_enabled(bool value, Error **errp) | |
940 | { | |
941 | MigrationCapabilityStatusList *cap; | |
942 | ||
943 | cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value); | |
2833c59b JQ |
944 | qmp_migrate_set_capabilities(cap, errp); |
945 | qapi_free_MigrationCapabilityStatusList(cap); | |
946 | } | |
947 | ||
948 | static void migrate_set_block_incremental(MigrationState *s, bool value) | |
949 | { | |
950 | s->parameters.block_incremental = value; | |
951 | } | |
952 | ||
953 | static void block_cleanup_parameters(MigrationState *s) | |
954 | { | |
955 | if (s->must_remove_block_options) { | |
956 | /* setting to false can never fail */ | |
957 | migrate_set_block_enabled(false, &error_abort); | |
958 | migrate_set_block_incremental(s, false); | |
959 | s->must_remove_block_options = false; | |
960 | } | |
961 | } | |
962 | ||
bb1fadc4 | 963 | static void migrate_fd_cleanup(void *opaque) |
065e2813 | 964 | { |
bb1fadc4 PB |
965 | MigrationState *s = opaque; |
966 | ||
967 | qemu_bh_delete(s->cleanup_bh); | |
968 | s->cleanup_bh = NULL; | |
969 | ||
89a02a9f | 970 | if (s->to_dst_file) { |
9013dca5 | 971 | trace_migrate_fd_cleanup(); |
404a7c05 | 972 | qemu_mutex_unlock_iothread(); |
1d34e4bf DDAG |
973 | if (s->migration_thread_running) { |
974 | qemu_thread_join(&s->thread); | |
975 | s->migration_thread_running = false; | |
976 | } | |
404a7c05 PB |
977 | qemu_mutex_lock_iothread(); |
978 | ||
89a02a9f HZ |
979 | qemu_fclose(s->to_dst_file); |
980 | s->to_dst_file = NULL; | |
065e2813 AL |
981 | } |
982 | ||
9ec055ae DDAG |
983 | assert((s->state != MIGRATION_STATUS_ACTIVE) && |
984 | (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE)); | |
7a2c1721 | 985 | |
94f5a437 | 986 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 987 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 988 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 989 | } |
a3fa1d78 PB |
990 | |
991 | notifier_list_notify(&migration_state_notifiers, s); | |
2833c59b | 992 | block_cleanup_parameters(s); |
065e2813 AL |
993 | } |
994 | ||
d59ce6f3 | 995 | void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 996 | { |
25174055 | 997 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 998 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
999 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1000 | MIGRATION_STATUS_FAILED); | |
d59ce6f3 DB |
1001 | if (!s->error) { |
1002 | s->error = error_copy(error); | |
1003 | } | |
bb1fadc4 | 1004 | notifier_list_notify(&migration_state_notifiers, s); |
2833c59b | 1005 | block_cleanup_parameters(s); |
458cf28e JQ |
1006 | } |
1007 | ||
0edda1c4 | 1008 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 1009 | { |
6f2b811a | 1010 | int old_state ; |
89a02a9f | 1011 | QEMUFile *f = migrate_get_current()->to_dst_file; |
9013dca5 | 1012 | trace_migrate_fd_cancel(); |
065e2813 | 1013 | |
70b20477 DDAG |
1014 | if (s->rp_state.from_dst_file) { |
1015 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
1016 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
1017 | } | |
1018 | ||
6f2b811a Z |
1019 | do { |
1020 | old_state = s->state; | |
f6844b99 | 1021 | if (!migration_is_setup_or_active(old_state)) { |
6f2b811a Z |
1022 | break; |
1023 | } | |
48781e5b | 1024 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 1025 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
1026 | |
1027 | /* | |
1028 | * If we're unlucky the migration code might be stuck somewhere in a | |
1029 | * send/write while the network has failed and is waiting to timeout; | |
1030 | * if we've got shutdown(2) available then we can force it to quit. | |
1031 | * The outgoing qemu file gets closed in migrate_fd_cleanup that is | |
1032 | * called in a bh, so there is no race against this cancel. | |
1033 | */ | |
31194731 | 1034 | if (s->state == MIGRATION_STATUS_CANCELLING && f) { |
a26ba26e DDAG |
1035 | qemu_file_shutdown(f); |
1036 | } | |
1d2acc31 HZ |
1037 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
1038 | Error *local_err = NULL; | |
1039 | ||
1040 | bdrv_invalidate_cache_all(&local_err); | |
1041 | if (local_err) { | |
1042 | error_report_err(local_err); | |
1043 | } else { | |
1044 | s->block_inactive = false; | |
1045 | } | |
1046 | } | |
2833c59b | 1047 | block_cleanup_parameters(s); |
065e2813 AL |
1048 | } |
1049 | ||
99a0db9b GH |
1050 | void add_migration_state_change_notifier(Notifier *notify) |
1051 | { | |
1052 | notifier_list_add(&migration_state_notifiers, notify); | |
1053 | } | |
1054 | ||
1055 | void remove_migration_state_change_notifier(Notifier *notify) | |
1056 | { | |
31552529 | 1057 | notifier_remove(notify); |
99a0db9b GH |
1058 | } |
1059 | ||
02edd2e7 | 1060 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 1061 | { |
31194731 | 1062 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
1063 | } |
1064 | ||
7073693b | 1065 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 1066 | { |
31194731 | 1067 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 1068 | } |
0edda1c4 | 1069 | |
afe2df69 GH |
1070 | bool migration_has_failed(MigrationState *s) |
1071 | { | |
31194731 HZ |
1072 | return (s->state == MIGRATION_STATUS_CANCELLED || |
1073 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
1074 | } |
1075 | ||
5727309d | 1076 | bool migration_in_postcopy(void) |
9ec055ae | 1077 | { |
5727309d JQ |
1078 | MigrationState *s = migrate_get_current(); |
1079 | ||
9ec055ae DDAG |
1080 | return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); |
1081 | } | |
1082 | ||
b82fc321 DDAG |
1083 | bool migration_in_postcopy_after_devices(MigrationState *s) |
1084 | { | |
5727309d | 1085 | return migration_in_postcopy() && s->postcopy_after_devices; |
b82fc321 DDAG |
1086 | } |
1087 | ||
fab35005 | 1088 | bool migration_is_idle(void) |
fe44dc91 | 1089 | { |
fab35005 | 1090 | MigrationState *s = migrate_get_current(); |
fe44dc91 AA |
1091 | |
1092 | switch (s->state) { | |
1093 | case MIGRATION_STATUS_NONE: | |
1094 | case MIGRATION_STATUS_CANCELLED: | |
1095 | case MIGRATION_STATUS_COMPLETED: | |
1096 | case MIGRATION_STATUS_FAILED: | |
1097 | return true; | |
1098 | case MIGRATION_STATUS_SETUP: | |
1099 | case MIGRATION_STATUS_CANCELLING: | |
1100 | case MIGRATION_STATUS_ACTIVE: | |
1101 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1102 | case MIGRATION_STATUS_COLO: | |
1103 | return false; | |
1104 | case MIGRATION_STATUS__MAX: | |
1105 | g_assert_not_reached(); | |
1106 | } | |
1107 | ||
1108 | return false; | |
1109 | } | |
1110 | ||
a0762d9e | 1111 | MigrationState *migrate_init(void) |
0edda1c4 | 1112 | { |
17549e84 | 1113 | MigrationState *s = migrate_get_current(); |
bbf6da32 | 1114 | |
389775d1 DDAG |
1115 | /* |
1116 | * Reinitialise all migration state, except | |
1117 | * parameters/capabilities that the user set, and | |
1118 | * locks. | |
1119 | */ | |
1120 | s->bytes_xfer = 0; | |
1121 | s->xfer_limit = 0; | |
1122 | s->cleanup_bh = 0; | |
89a02a9f | 1123 | s->to_dst_file = NULL; |
389775d1 | 1124 | s->state = MIGRATION_STATUS_NONE; |
389775d1 DDAG |
1125 | s->rp_state.from_dst_file = NULL; |
1126 | s->rp_state.error = false; | |
1127 | s->mbps = 0.0; | |
1128 | s->downtime = 0; | |
1129 | s->expected_downtime = 0; | |
389775d1 | 1130 | s->setup_time = 0; |
389775d1 | 1131 | s->start_postcopy = false; |
b82fc321 | 1132 | s->postcopy_after_devices = false; |
389775d1 | 1133 | s->migration_thread_running = false; |
d59ce6f3 DB |
1134 | error_free(s->error); |
1135 | s->error = NULL; | |
389775d1 | 1136 | |
48781e5b | 1137 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 1138 | |
bc72ad67 | 1139 | s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0edda1c4 JQ |
1140 | return s; |
1141 | } | |
cab30143 | 1142 | |
fa2756b7 AL |
1143 | static GSList *migration_blockers; |
1144 | ||
fe44dc91 | 1145 | int migrate_add_blocker(Error *reason, Error **errp) |
fa2756b7 | 1146 | { |
3df663e5 | 1147 | if (migrate_get_current()->only_migratable) { |
b67b8c3a AA |
1148 | error_propagate(errp, error_copy(reason)); |
1149 | error_prepend(errp, "disallowing migration blocker " | |
1150 | "(--only_migratable) for: "); | |
1151 | return -EACCES; | |
1152 | } | |
1153 | ||
fab35005 | 1154 | if (migration_is_idle()) { |
fe44dc91 AA |
1155 | migration_blockers = g_slist_prepend(migration_blockers, reason); |
1156 | return 0; | |
1157 | } | |
1158 | ||
1159 | error_propagate(errp, error_copy(reason)); | |
1160 | error_prepend(errp, "disallowing migration blocker (migration in " | |
1161 | "progress) for: "); | |
1162 | return -EBUSY; | |
fa2756b7 AL |
1163 | } |
1164 | ||
1165 | void migrate_del_blocker(Error *reason) | |
1166 | { | |
1167 | migration_blockers = g_slist_remove(migration_blockers, reason); | |
1168 | } | |
1169 | ||
bf1ae1f4 DDAG |
1170 | void qmp_migrate_incoming(const char *uri, Error **errp) |
1171 | { | |
1172 | Error *local_err = NULL; | |
4debb5f5 | 1173 | static bool once = true; |
bf1ae1f4 DDAG |
1174 | |
1175 | if (!deferred_incoming) { | |
4debb5f5 | 1176 | error_setg(errp, "For use with '-incoming defer'"); |
bf1ae1f4 DDAG |
1177 | return; |
1178 | } | |
4debb5f5 DDAG |
1179 | if (!once) { |
1180 | error_setg(errp, "The incoming migration has already been started"); | |
1181 | } | |
bf1ae1f4 DDAG |
1182 | |
1183 | qemu_start_incoming_migration(uri, &local_err); | |
1184 | ||
1185 | if (local_err) { | |
1186 | error_propagate(errp, local_err); | |
1187 | return; | |
1188 | } | |
1189 | ||
4debb5f5 | 1190 | once = false; |
bf1ae1f4 DDAG |
1191 | } |
1192 | ||
24f3902b GK |
1193 | bool migration_is_blocked(Error **errp) |
1194 | { | |
1195 | if (qemu_savevm_state_blocked(errp)) { | |
1196 | return true; | |
1197 | } | |
1198 | ||
1199 | if (migration_blockers) { | |
250561e1 | 1200 | error_propagate(errp, error_copy(migration_blockers->data)); |
24f3902b GK |
1201 | return true; |
1202 | } | |
1203 | ||
1204 | return false; | |
1205 | } | |
1206 | ||
e1c37d0e LC |
1207 | void qmp_migrate(const char *uri, bool has_blk, bool blk, |
1208 | bool has_inc, bool inc, bool has_detach, bool detach, | |
1209 | Error **errp) | |
cab30143 | 1210 | { |
be7059cd | 1211 | Error *local_err = NULL; |
17549e84 | 1212 | MigrationState *s = migrate_get_current(); |
cab30143 | 1213 | const char *p; |
cab30143 | 1214 | |
f6844b99 | 1215 | if (migration_is_setup_or_active(s->state) || |
0b827d5e HZ |
1216 | s->state == MIGRATION_STATUS_CANCELLING || |
1217 | s->state == MIGRATION_STATUS_COLO) { | |
c6bd8c70 | 1218 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
e1c37d0e | 1219 | return; |
cab30143 | 1220 | } |
ca99993a DDAG |
1221 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
1222 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
1223 | return; | |
1224 | } | |
1225 | ||
24f3902b | 1226 | if (migration_is_blocked(errp)) { |
e1c37d0e | 1227 | return; |
fa2756b7 AL |
1228 | } |
1229 | ||
2833c59b JQ |
1230 | if ((has_blk && blk) || (has_inc && inc)) { |
1231 | if (migrate_use_block() || migrate_use_block_incremental()) { | |
1232 | error_setg(errp, "Command options are incompatible with " | |
1233 | "current migration capabilities"); | |
1234 | return; | |
1235 | } | |
1236 | migrate_set_block_enabled(true, &local_err); | |
1237 | if (local_err) { | |
1238 | error_propagate(errp, local_err); | |
1239 | return; | |
1240 | } | |
1241 | s->must_remove_block_options = true; | |
1242 | } | |
1243 | ||
1244 | if (has_inc && inc) { | |
1245 | migrate_set_block_incremental(s, true); | |
1246 | } | |
1247 | ||
a0762d9e | 1248 | s = migrate_init(); |
cab30143 JQ |
1249 | |
1250 | if (strstart(uri, "tcp:", &p)) { | |
f37afb5a | 1251 | tcp_start_outgoing_migration(s, p, &local_err); |
2da776db | 1252 | #ifdef CONFIG_RDMA |
41310c68 | 1253 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
1254 | rdma_start_outgoing_migration(s, p, &local_err); |
1255 | #endif | |
cab30143 | 1256 | } else if (strstart(uri, "exec:", &p)) { |
f37afb5a | 1257 | exec_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1258 | } else if (strstart(uri, "unix:", &p)) { |
f37afb5a | 1259 | unix_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1260 | } else if (strstart(uri, "fd:", &p)) { |
f37afb5a | 1261 | fd_start_outgoing_migration(s, p, &local_err); |
99a0db9b | 1262 | } else { |
c6bd8c70 MA |
1263 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", |
1264 | "a valid migration protocol"); | |
48781e5b HZ |
1265 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1266 | MIGRATION_STATUS_FAILED); | |
e1c37d0e | 1267 | return; |
cab30143 JQ |
1268 | } |
1269 | ||
f37afb5a | 1270 | if (local_err) { |
d59ce6f3 | 1271 | migrate_fd_error(s, local_err); |
f37afb5a | 1272 | error_propagate(errp, local_err); |
e1c37d0e | 1273 | return; |
1299c631 | 1274 | } |
cab30143 JQ |
1275 | } |
1276 | ||
6cdedb07 | 1277 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 1278 | { |
17549e84 | 1279 | migrate_fd_cancel(migrate_get_current()); |
cab30143 JQ |
1280 | } |
1281 | ||
9e1ba4cc OW |
1282 | void qmp_migrate_set_cache_size(int64_t value, Error **errp) |
1283 | { | |
1284 | MigrationState *s = migrate_get_current(); | |
c91e681a | 1285 | int64_t new_size; |
9e1ba4cc OW |
1286 | |
1287 | /* Check for truncation */ | |
1288 | if (value != (size_t)value) { | |
c6bd8c70 MA |
1289 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1290 | "exceeding address space"); | |
9e1ba4cc OW |
1291 | return; |
1292 | } | |
1293 | ||
a5615b14 OW |
1294 | /* Cache should not be larger than guest ram size */ |
1295 | if (value > ram_bytes_total()) { | |
c6bd8c70 MA |
1296 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1297 | "exceeds guest ram size "); | |
a5615b14 OW |
1298 | return; |
1299 | } | |
1300 | ||
c91e681a OW |
1301 | new_size = xbzrle_cache_resize(value); |
1302 | if (new_size < 0) { | |
c6bd8c70 MA |
1303 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1304 | "is smaller than page size"); | |
c91e681a OW |
1305 | return; |
1306 | } | |
1307 | ||
1308 | s->xbzrle_cache_size = new_size; | |
9e1ba4cc OW |
1309 | } |
1310 | ||
1311 | int64_t qmp_query_migrate_cache_size(Error **errp) | |
1312 | { | |
1313 | return migrate_xbzrle_cache_size(); | |
1314 | } | |
1315 | ||
3dc85383 | 1316 | void qmp_migrate_set_speed(int64_t value, Error **errp) |
cab30143 | 1317 | { |
1bda8b3c | 1318 | MigrateSetParameters p = { |
2ff30257 AA |
1319 | .has_max_bandwidth = true, |
1320 | .max_bandwidth = value, | |
1321 | }; | |
cab30143 | 1322 | |
2ff30257 | 1323 | qmp_migrate_set_parameters(&p, errp); |
cab30143 JQ |
1324 | } |
1325 | ||
4f0a993b | 1326 | void qmp_migrate_set_downtime(double value, Error **errp) |
cab30143 | 1327 | { |
87c9cc1c DHB |
1328 | if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) { |
1329 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " | |
1330 | "the range of 0 to %d seconds", | |
1331 | MAX_MIGRATE_DOWNTIME_SECONDS); | |
1332 | return; | |
1333 | } | |
1334 | ||
2ff30257 AA |
1335 | value *= 1000; /* Convert to milliseconds */ |
1336 | value = MAX(0, MIN(INT64_MAX, value)); | |
1337 | ||
1bda8b3c | 1338 | MigrateSetParameters p = { |
2ff30257 AA |
1339 | .has_downtime_limit = true, |
1340 | .downtime_limit = value, | |
1341 | }; | |
1342 | ||
1343 | qmp_migrate_set_parameters(&p, errp); | |
99a0db9b | 1344 | } |
17ad9b35 | 1345 | |
53f09a10 PB |
1346 | bool migrate_release_ram(void) |
1347 | { | |
1348 | MigrationState *s; | |
1349 | ||
1350 | s = migrate_get_current(); | |
1351 | ||
1352 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; | |
1353 | } | |
1354 | ||
53dd370c DDAG |
1355 | bool migrate_postcopy_ram(void) |
1356 | { | |
1357 | MigrationState *s; | |
1358 | ||
1359 | s = migrate_get_current(); | |
1360 | ||
32c3db5b | 1361 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
53dd370c DDAG |
1362 | } |
1363 | ||
bde1e2ec CV |
1364 | bool migrate_auto_converge(void) |
1365 | { | |
1366 | MigrationState *s; | |
1367 | ||
1368 | s = migrate_get_current(); | |
1369 | ||
1370 | return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; | |
1371 | } | |
1372 | ||
323004a3 PL |
1373 | bool migrate_zero_blocks(void) |
1374 | { | |
1375 | MigrationState *s; | |
1376 | ||
1377 | s = migrate_get_current(); | |
1378 | ||
1379 | return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; | |
1380 | } | |
1381 | ||
8706d2d5 LL |
1382 | bool migrate_use_compression(void) |
1383 | { | |
dde4e694 LL |
1384 | MigrationState *s; |
1385 | ||
1386 | s = migrate_get_current(); | |
1387 | ||
1388 | return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; | |
8706d2d5 LL |
1389 | } |
1390 | ||
1391 | int migrate_compress_level(void) | |
1392 | { | |
1393 | MigrationState *s; | |
1394 | ||
1395 | s = migrate_get_current(); | |
1396 | ||
2594f56d | 1397 | return s->parameters.compress_level; |
8706d2d5 LL |
1398 | } |
1399 | ||
1400 | int migrate_compress_threads(void) | |
1401 | { | |
1402 | MigrationState *s; | |
1403 | ||
1404 | s = migrate_get_current(); | |
1405 | ||
2594f56d | 1406 | return s->parameters.compress_threads; |
8706d2d5 LL |
1407 | } |
1408 | ||
3fcb38c2 LL |
1409 | int migrate_decompress_threads(void) |
1410 | { | |
1411 | MigrationState *s; | |
1412 | ||
1413 | s = migrate_get_current(); | |
1414 | ||
2594f56d | 1415 | return s->parameters.decompress_threads; |
3fcb38c2 LL |
1416 | } |
1417 | ||
b05dc723 JQ |
1418 | bool migrate_use_events(void) |
1419 | { | |
1420 | MigrationState *s; | |
1421 | ||
1422 | s = migrate_get_current(); | |
1423 | ||
1424 | return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; | |
1425 | } | |
1426 | ||
17ad9b35 OW |
1427 | int migrate_use_xbzrle(void) |
1428 | { | |
1429 | MigrationState *s; | |
1430 | ||
1431 | s = migrate_get_current(); | |
1432 | ||
1433 | return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; | |
1434 | } | |
1435 | ||
1436 | int64_t migrate_xbzrle_cache_size(void) | |
1437 | { | |
1438 | MigrationState *s; | |
1439 | ||
1440 | s = migrate_get_current(); | |
1441 | ||
1442 | return s->xbzrle_cache_size; | |
1443 | } | |
0d82d0e8 | 1444 | |
2833c59b JQ |
1445 | bool migrate_use_block(void) |
1446 | { | |
1447 | MigrationState *s; | |
1448 | ||
1449 | s = migrate_get_current(); | |
1450 | ||
1451 | return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK]; | |
1452 | } | |
1453 | ||
c788ada8 PX |
1454 | bool migrate_use_return_path(void) |
1455 | { | |
1456 | MigrationState *s; | |
1457 | ||
1458 | s = migrate_get_current(); | |
1459 | ||
1460 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH]; | |
1461 | } | |
1462 | ||
2833c59b JQ |
1463 | bool migrate_use_block_incremental(void) |
1464 | { | |
1465 | MigrationState *s; | |
1466 | ||
1467 | s = migrate_get_current(); | |
1468 | ||
1469 | return s->parameters.block_incremental; | |
1470 | } | |
1471 | ||
70b20477 DDAG |
1472 | /* migration thread support */ |
1473 | /* | |
1474 | * Something bad happened to the RP stream, mark an error | |
1475 | * The caller shall print or trace something to indicate why | |
1476 | */ | |
1477 | static void mark_source_rp_bad(MigrationState *s) | |
1478 | { | |
1479 | s->rp_state.error = true; | |
1480 | } | |
1481 | ||
1482 | static struct rp_cmd_args { | |
1483 | ssize_t len; /* -1 = variable */ | |
1484 | const char *name; | |
1485 | } rp_cmd_args[] = { | |
1486 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
1487 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
1488 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
1489 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
1490 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
70b20477 DDAG |
1491 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
1492 | }; | |
1493 | ||
1e2d90eb DDAG |
1494 | /* |
1495 | * Process a request for pages received on the return path, | |
1496 | * We're allowed to send more than requested (e.g. to round to our page size) | |
1497 | * and we don't need to send pages that have already been sent. | |
1498 | */ | |
1499 | static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
1500 | ram_addr_t start, size_t len) | |
1501 | { | |
6c595cde DDAG |
1502 | long our_host_ps = getpagesize(); |
1503 | ||
1e2d90eb | 1504 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
1505 | |
1506 | /* | |
1507 | * Since we currently insist on matching page sizes, just sanity check | |
1508 | * we're being asked for whole host pages. | |
1509 | */ | |
1510 | if (start & (our_host_ps-1) || | |
1511 | (len & (our_host_ps-1))) { | |
1512 | error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT | |
1513 | " len: %zd", __func__, start, len); | |
1514 | mark_source_rp_bad(ms); | |
1515 | return; | |
1516 | } | |
1517 | ||
96506894 | 1518 | if (ram_save_queue_pages(rbname, start, len)) { |
6c595cde DDAG |
1519 | mark_source_rp_bad(ms); |
1520 | } | |
1e2d90eb DDAG |
1521 | } |
1522 | ||
70b20477 DDAG |
1523 | /* |
1524 | * Handles messages sent on the return path towards the source VM | |
1525 | * | |
1526 | */ | |
1527 | static void *source_return_path_thread(void *opaque) | |
1528 | { | |
1529 | MigrationState *ms = opaque; | |
1530 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
1531 | uint16_t header_len, header_type; | |
568b01ca | 1532 | uint8_t buf[512]; |
70b20477 | 1533 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
1534 | ram_addr_t start = 0; /* =0 to silence warning */ |
1535 | size_t len = 0, expected_len; | |
70b20477 DDAG |
1536 | int res; |
1537 | ||
1538 | trace_source_return_path_thread_entry(); | |
1539 | while (!ms->rp_state.error && !qemu_file_get_error(rp) && | |
1540 | migration_is_setup_or_active(ms->state)) { | |
1541 | trace_source_return_path_thread_loop_top(); | |
1542 | header_type = qemu_get_be16(rp); | |
1543 | header_len = qemu_get_be16(rp); | |
1544 | ||
1545 | if (header_type >= MIG_RP_MSG_MAX || | |
1546 | header_type == MIG_RP_MSG_INVALID) { | |
1547 | error_report("RP: Received invalid message 0x%04x length 0x%04x", | |
1548 | header_type, header_len); | |
1549 | mark_source_rp_bad(ms); | |
1550 | goto out; | |
1551 | } | |
1552 | ||
1553 | if ((rp_cmd_args[header_type].len != -1 && | |
1554 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 1555 | header_len > sizeof(buf)) { |
70b20477 DDAG |
1556 | error_report("RP: Received '%s' message (0x%04x) with" |
1557 | "incorrect length %d expecting %zu", | |
1558 | rp_cmd_args[header_type].name, header_type, header_len, | |
1559 | (size_t)rp_cmd_args[header_type].len); | |
1560 | mark_source_rp_bad(ms); | |
1561 | goto out; | |
1562 | } | |
1563 | ||
1564 | /* We know we've got a valid header by this point */ | |
1565 | res = qemu_get_buffer(rp, buf, header_len); | |
1566 | if (res != header_len) { | |
1567 | error_report("RP: Failed reading data for message 0x%04x" | |
1568 | " read %d expected %d", | |
1569 | header_type, res, header_len); | |
1570 | mark_source_rp_bad(ms); | |
1571 | goto out; | |
1572 | } | |
1573 | ||
1574 | /* OK, we have the message and the data */ | |
1575 | switch (header_type) { | |
1576 | case MIG_RP_MSG_SHUT: | |
4d885131 | 1577 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
1578 | trace_source_return_path_thread_shut(sibling_error); |
1579 | if (sibling_error) { | |
1580 | error_report("RP: Sibling indicated error %d", sibling_error); | |
1581 | mark_source_rp_bad(ms); | |
1582 | } | |
1583 | /* | |
1584 | * We'll let the main thread deal with closing the RP | |
1585 | * we could do a shutdown(2) on it, but we're the only user | |
1586 | * anyway, so there's nothing gained. | |
1587 | */ | |
1588 | goto out; | |
1589 | ||
1590 | case MIG_RP_MSG_PONG: | |
4d885131 | 1591 | tmp32 = ldl_be_p(buf); |
70b20477 DDAG |
1592 | trace_source_return_path_thread_pong(tmp32); |
1593 | break; | |
1594 | ||
1e2d90eb | 1595 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
1596 | start = ldq_be_p(buf); |
1597 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1598 | migrate_handle_rp_req_pages(ms, NULL, start, len); |
1599 | break; | |
1600 | ||
1601 | case MIG_RP_MSG_REQ_PAGES_ID: | |
1602 | expected_len = 12 + 1; /* header + termination */ | |
1603 | ||
1604 | if (header_len >= expected_len) { | |
4d885131 PM |
1605 | start = ldq_be_p(buf); |
1606 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1607 | /* Now we expect an idstr */ |
1608 | tmp32 = buf[12]; /* Length of the following idstr */ | |
1609 | buf[13 + tmp32] = '\0'; | |
1610 | expected_len += tmp32; | |
1611 | } | |
1612 | if (header_len != expected_len) { | |
1613 | error_report("RP: Req_Page_id with length %d expecting %zd", | |
1614 | header_len, expected_len); | |
1615 | mark_source_rp_bad(ms); | |
1616 | goto out; | |
1617 | } | |
1618 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); | |
1619 | break; | |
1620 | ||
70b20477 DDAG |
1621 | default: |
1622 | break; | |
1623 | } | |
1624 | } | |
5df5416e | 1625 | if (qemu_file_get_error(rp)) { |
70b20477 DDAG |
1626 | trace_source_return_path_thread_bad_end(); |
1627 | mark_source_rp_bad(ms); | |
1628 | } | |
1629 | ||
1630 | trace_source_return_path_thread_end(); | |
1631 | out: | |
1632 | ms->rp_state.from_dst_file = NULL; | |
1633 | qemu_fclose(rp); | |
1634 | return NULL; | |
1635 | } | |
1636 | ||
70b20477 DDAG |
1637 | static int open_return_path_on_source(MigrationState *ms) |
1638 | { | |
1639 | ||
89a02a9f | 1640 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
1641 | if (!ms->rp_state.from_dst_file) { |
1642 | return -1; | |
1643 | } | |
1644 | ||
1645 | trace_open_return_path_on_source(); | |
1646 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", | |
1647 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
1648 | ||
1649 | trace_open_return_path_on_source_continue(); | |
1650 | ||
1651 | return 0; | |
1652 | } | |
1653 | ||
70b20477 DDAG |
1654 | /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ |
1655 | static int await_return_path_close_on_source(MigrationState *ms) | |
1656 | { | |
1657 | /* | |
1658 | * If this is a normal exit then the destination will send a SHUT and the | |
1659 | * rp_thread will exit, however if there's an error we need to cause | |
1660 | * it to exit. | |
1661 | */ | |
89a02a9f | 1662 | if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { |
70b20477 DDAG |
1663 | /* |
1664 | * shutdown(2), if we have it, will cause it to unblock if it's stuck | |
1665 | * waiting for the destination. | |
1666 | */ | |
1667 | qemu_file_shutdown(ms->rp_state.from_dst_file); | |
1668 | mark_source_rp_bad(ms); | |
1669 | } | |
1670 | trace_await_return_path_close_on_source_joining(); | |
1671 | qemu_thread_join(&ms->rp_state.rp_thread); | |
1672 | trace_await_return_path_close_on_source_close(); | |
1673 | return ms->rp_state.error; | |
1674 | } | |
1675 | ||
1d34e4bf DDAG |
1676 | /* |
1677 | * Switch from normal iteration to postcopy | |
1678 | * Returns non-0 on error | |
1679 | */ | |
1680 | static int postcopy_start(MigrationState *ms, bool *old_vm_running) | |
1681 | { | |
1682 | int ret; | |
61b67d47 DB |
1683 | QIOChannelBuffer *bioc; |
1684 | QEMUFile *fb; | |
1d34e4bf | 1685 | int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
ef8d6488 | 1686 | bool restart_block = false; |
48781e5b | 1687 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, |
1d34e4bf DDAG |
1688 | MIGRATION_STATUS_POSTCOPY_ACTIVE); |
1689 | ||
1690 | trace_postcopy_start(); | |
1691 | qemu_mutex_lock_iothread(); | |
1692 | trace_postcopy_start_set_run(); | |
1693 | ||
1694 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); | |
1695 | *old_vm_running = runstate_is_running(); | |
1696 | global_state_store(); | |
1697 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
76b1c7fe KW |
1698 | if (ret < 0) { |
1699 | goto fail; | |
1700 | } | |
1d34e4bf | 1701 | |
76b1c7fe | 1702 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
1703 | if (ret < 0) { |
1704 | goto fail; | |
1705 | } | |
ef8d6488 | 1706 | restart_block = true; |
1d34e4bf | 1707 | |
1c0d249d DDAG |
1708 | /* |
1709 | * Cause any non-postcopiable, but iterative devices to | |
1710 | * send out their final data. | |
1711 | */ | |
a1fbe750 | 1712 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); |
1c0d249d | 1713 | |
1d34e4bf DDAG |
1714 | /* |
1715 | * in Finish migrate and with the io-lock held everything should | |
1716 | * be quiet, but we've potentially still got dirty pages and we | |
1717 | * need to tell the destination to throw any pages it's already received | |
1718 | * that are dirty | |
1719 | */ | |
1720 | if (ram_postcopy_send_discard_bitmap(ms)) { | |
1721 | error_report("postcopy send discard bitmap failed"); | |
1722 | goto fail; | |
1723 | } | |
1724 | ||
1725 | /* | |
1726 | * send rest of state - note things that are doing postcopy | |
1727 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
1728 | * wrap their state up here | |
1729 | */ | |
89a02a9f | 1730 | qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); |
1d34e4bf | 1731 | /* Ping just for debugging, helps line traces up */ |
89a02a9f | 1732 | qemu_savevm_send_ping(ms->to_dst_file, 2); |
1d34e4bf DDAG |
1733 | |
1734 | /* | |
1735 | * While loading the device state we may trigger page transfer | |
1736 | * requests and the fd must be free to process those, and thus | |
1737 | * the destination must read the whole device state off the fd before | |
1738 | * it starts processing it. Unfortunately the ad-hoc migration format | |
1739 | * doesn't allow the destination to know the size to read without fully | |
1740 | * parsing it through each devices load-state code (especially the open | |
1741 | * coded devices that use get/put). | |
1742 | * So we wrap the device state up in a package with a length at the start; | |
1743 | * to do this we use a qemu_buf to hold the whole of the device state. | |
1744 | */ | |
61b67d47 | 1745 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 1746 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
61b67d47 DB |
1747 | fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc)); |
1748 | object_unref(OBJECT(bioc)); | |
1d34e4bf | 1749 | |
c76201ab DDAG |
1750 | /* |
1751 | * Make sure the receiver can get incoming pages before we send the rest | |
1752 | * of the state | |
1753 | */ | |
1754 | qemu_savevm_send_postcopy_listen(fb); | |
1755 | ||
a1fbe750 | 1756 | qemu_savevm_state_complete_precopy(fb, false, false); |
1d34e4bf DDAG |
1757 | qemu_savevm_send_ping(fb, 3); |
1758 | ||
1759 | qemu_savevm_send_postcopy_run(fb); | |
1760 | ||
1761 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 1762 | |
ef8d6488 DDAG |
1763 | /* Last point of recovery; as soon as we send the package the destination |
1764 | * can open devices and potentially start running. | |
1765 | * Lets just check again we've not got any errors. | |
1766 | */ | |
1767 | ret = qemu_file_get_error(ms->to_dst_file); | |
1768 | if (ret) { | |
1769 | error_report("postcopy_start: Migration stream errored (pre package)"); | |
1770 | goto fail_closefb; | |
1771 | } | |
1772 | ||
1773 | restart_block = false; | |
1774 | ||
1d34e4bf | 1775 | /* Now send that blob */ |
61b67d47 | 1776 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
1777 | goto fail_closefb; |
1778 | } | |
1779 | qemu_fclose(fb); | |
b82fc321 DDAG |
1780 | |
1781 | /* Send a notify to give a chance for anything that needs to happen | |
1782 | * at the transition to postcopy and after the device state; in particular | |
1783 | * spice needs to trigger a transition now | |
1784 | */ | |
1785 | ms->postcopy_after_devices = true; | |
1786 | notifier_list_notify(&migration_state_notifiers, ms); | |
1787 | ||
1d34e4bf DDAG |
1788 | ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; |
1789 | ||
1790 | qemu_mutex_unlock_iothread(); | |
1791 | ||
1792 | /* | |
1793 | * Although this ping is just for debug, it could potentially be | |
1794 | * used for getting a better measurement of downtime at the source. | |
1795 | */ | |
89a02a9f | 1796 | qemu_savevm_send_ping(ms->to_dst_file, 4); |
1d34e4bf | 1797 | |
ced1c616 PB |
1798 | if (migrate_release_ram()) { |
1799 | ram_postcopy_migrated_memory_release(ms); | |
1800 | } | |
1801 | ||
89a02a9f | 1802 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf DDAG |
1803 | if (ret) { |
1804 | error_report("postcopy_start: Migration stream errored"); | |
48781e5b | 1805 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf DDAG |
1806 | MIGRATION_STATUS_FAILED); |
1807 | } | |
1808 | ||
1809 | return ret; | |
1810 | ||
1811 | fail_closefb: | |
1812 | qemu_fclose(fb); | |
1813 | fail: | |
48781e5b | 1814 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 1815 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
1816 | if (restart_block) { |
1817 | /* A failure happened early enough that we know the destination hasn't | |
1818 | * accessed block devices, so we're safe to recover. | |
1819 | */ | |
1820 | Error *local_err = NULL; | |
1821 | ||
1822 | bdrv_invalidate_cache_all(&local_err); | |
1823 | if (local_err) { | |
1824 | error_report_err(local_err); | |
1825 | } | |
1826 | } | |
1d34e4bf DDAG |
1827 | qemu_mutex_unlock_iothread(); |
1828 | return -1; | |
1829 | } | |
1830 | ||
09f6c85e DDAG |
1831 | /** |
1832 | * migration_completion: Used by migration_thread when there's not much left. | |
1833 | * The caller 'breaks' the loop when this returns. | |
1834 | * | |
1835 | * @s: Current migration state | |
36f48567 | 1836 | * @current_active_state: The migration state we expect to be in |
09f6c85e DDAG |
1837 | * @*old_vm_running: Pointer to old_vm_running flag |
1838 | * @*start_time: Pointer to time to update | |
1839 | */ | |
36f48567 DDAG |
1840 | static void migration_completion(MigrationState *s, int current_active_state, |
1841 | bool *old_vm_running, | |
09f6c85e DDAG |
1842 | int64_t *start_time) |
1843 | { | |
1844 | int ret; | |
1845 | ||
b10ac0c4 DDAG |
1846 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
1847 | qemu_mutex_lock_iothread(); | |
1848 | *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
1849 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); | |
1850 | *old_vm_running = runstate_is_running(); | |
1851 | ret = global_state_store(); | |
1852 | ||
1853 | if (!ret) { | |
a1fbe750 | 1854 | bool inactivate = !migrate_colo_enabled(); |
b10ac0c4 | 1855 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); |
f07fa4cb KW |
1856 | if (ret >= 0) { |
1857 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); | |
a1fbe750 FZ |
1858 | ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, |
1859 | inactivate); | |
f07fa4cb | 1860 | } |
a1fbe750 FZ |
1861 | if (inactivate && ret >= 0) { |
1862 | s->block_inactive = true; | |
b10ac0c4 DDAG |
1863 | } |
1864 | } | |
1865 | qemu_mutex_unlock_iothread(); | |
09f6c85e | 1866 | |
b10ac0c4 DDAG |
1867 | if (ret < 0) { |
1868 | goto fail; | |
09f6c85e | 1869 | } |
b10ac0c4 DDAG |
1870 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { |
1871 | trace_migration_completion_postcopy_end(); | |
1872 | ||
89a02a9f | 1873 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
b10ac0c4 | 1874 | trace_migration_completion_postcopy_end_after_complete(); |
09f6c85e | 1875 | } |
09f6c85e | 1876 | |
b10ac0c4 DDAG |
1877 | /* |
1878 | * If rp was opened we must clean up the thread before | |
1879 | * cleaning everything else up (since if there are no failures | |
1880 | * it will wait for the destination to send it's status in | |
1881 | * a SHUT command). | |
b10ac0c4 | 1882 | */ |
0425dc97 | 1883 | if (s->rp_state.from_dst_file) { |
b10ac0c4 | 1884 | int rp_error; |
0425dc97 | 1885 | trace_migration_return_path_end_before(); |
b10ac0c4 | 1886 | rp_error = await_return_path_close_on_source(s); |
0425dc97 | 1887 | trace_migration_return_path_end_after(rp_error); |
b10ac0c4 | 1888 | if (rp_error) { |
fe904ea8 | 1889 | goto fail_invalidate; |
b10ac0c4 | 1890 | } |
09f6c85e DDAG |
1891 | } |
1892 | ||
89a02a9f | 1893 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 1894 | trace_migration_completion_file_err(); |
fe904ea8 | 1895 | goto fail_invalidate; |
09f6c85e DDAG |
1896 | } |
1897 | ||
0b827d5e HZ |
1898 | if (!migrate_colo_enabled()) { |
1899 | migrate_set_state(&s->state, current_active_state, | |
1900 | MIGRATION_STATUS_COMPLETED); | |
1901 | } | |
1902 | ||
09f6c85e DDAG |
1903 | return; |
1904 | ||
fe904ea8 GK |
1905 | fail_invalidate: |
1906 | /* If not doing postcopy, vm_start() will be called: let's regain | |
1907 | * control on images. | |
1908 | */ | |
1909 | if (s->state == MIGRATION_STATUS_ACTIVE) { | |
1910 | Error *local_err = NULL; | |
1911 | ||
1d2acc31 | 1912 | qemu_mutex_lock_iothread(); |
fe904ea8 GK |
1913 | bdrv_invalidate_cache_all(&local_err); |
1914 | if (local_err) { | |
1915 | error_report_err(local_err); | |
1d2acc31 HZ |
1916 | } else { |
1917 | s->block_inactive = false; | |
fe904ea8 | 1918 | } |
1d2acc31 | 1919 | qemu_mutex_unlock_iothread(); |
fe904ea8 GK |
1920 | } |
1921 | ||
09f6c85e | 1922 | fail: |
48781e5b HZ |
1923 | migrate_set_state(&s->state, current_active_state, |
1924 | MIGRATION_STATUS_FAILED); | |
09f6c85e DDAG |
1925 | } |
1926 | ||
35a6ed4f HZ |
1927 | bool migrate_colo_enabled(void) |
1928 | { | |
1929 | MigrationState *s = migrate_get_current(); | |
1930 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; | |
1931 | } | |
1932 | ||
70b20477 DDAG |
1933 | /* |
1934 | * Master migration thread on the source VM. | |
1935 | * It drives the migration and pumps the data down the outgoing channel. | |
1936 | */ | |
5f496a1b | 1937 | static void *migration_thread(void *opaque) |
0d82d0e8 | 1938 | { |
9848a404 | 1939 | MigrationState *s = opaque; |
1d34e4bf | 1940 | /* Used by the bandwidth calcs, updated later */ |
bc72ad67 AB |
1941 | int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1942 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
be7172e2 | 1943 | int64_t initial_bytes = 0; |
faec066a PX |
1944 | /* |
1945 | * The final stage happens when the remaining data is smaller than | |
1946 | * this threshold; it's calculated from the requested downtime and | |
1947 | * measured bandwidth | |
1948 | */ | |
1949 | int64_t threshold_size = 0; | |
a3fa1d78 | 1950 | int64_t start_time = initial_time; |
94f5a437 | 1951 | int64_t end_time; |
a3fa1d78 | 1952 | bool old_vm_running = false; |
1d34e4bf DDAG |
1953 | bool entered_postcopy = false; |
1954 | /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */ | |
1955 | enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE; | |
0b827d5e | 1956 | bool enable_colo = migrate_colo_enabled(); |
76f5933a | 1957 | |
ab28bd23 PB |
1958 | rcu_register_thread(); |
1959 | ||
89a02a9f | 1960 | qemu_savevm_state_header(s->to_dst_file); |
1d34e4bf | 1961 | |
62a02658 PX |
1962 | /* |
1963 | * If we opened the return path, we need to make sure dst has it | |
1964 | * opened as well. | |
1965 | */ | |
1966 | if (s->rp_state.from_dst_file) { | |
1d34e4bf | 1967 | /* Now tell the dest that it should open its end so it can reply */ |
89a02a9f | 1968 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
1969 | |
1970 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 1971 | qemu_savevm_send_ping(s->to_dst_file, 1); |
0425dc97 | 1972 | } |
1d34e4bf | 1973 | |
0425dc97 | 1974 | if (migrate_postcopy_ram()) { |
1d34e4bf DDAG |
1975 | /* |
1976 | * Tell the destination that we *might* want to do postcopy later; | |
1977 | * if the other end can't do postcopy it should fail now, nice and | |
1978 | * early. | |
1979 | */ | |
89a02a9f | 1980 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
1981 | } |
1982 | ||
9907e842 | 1983 | qemu_savevm_state_setup(s->to_dst_file); |
0d82d0e8 | 1984 | |
bc72ad67 | 1985 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
48781e5b HZ |
1986 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1987 | MIGRATION_STATUS_ACTIVE); | |
29ae8a41 | 1988 | |
9ec055ae DDAG |
1989 | trace_migration_thread_setup_complete(); |
1990 | ||
1991 | while (s->state == MIGRATION_STATUS_ACTIVE || | |
1992 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
a3e879cd | 1993 | int64_t current_time; |
c369f40d | 1994 | uint64_t pending_size; |
0d82d0e8 | 1995 | |
89a02a9f | 1996 | if (!qemu_file_rate_limit(s->to_dst_file)) { |
c31b098f DDAG |
1997 | uint64_t pend_post, pend_nonpost; |
1998 | ||
faec066a PX |
1999 | qemu_savevm_state_pending(s->to_dst_file, threshold_size, |
2000 | &pend_nonpost, &pend_post); | |
c31b098f | 2001 | pending_size = pend_nonpost + pend_post; |
faec066a | 2002 | trace_migrate_pending(pending_size, threshold_size, |
c31b098f | 2003 | pend_post, pend_nonpost); |
faec066a | 2004 | if (pending_size && pending_size >= threshold_size) { |
1d34e4bf DDAG |
2005 | /* Still a significant amount to transfer */ |
2006 | ||
1d34e4bf DDAG |
2007 | if (migrate_postcopy_ram() && |
2008 | s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE && | |
faec066a | 2009 | pend_nonpost <= threshold_size && |
1d34e4bf DDAG |
2010 | atomic_read(&s->start_postcopy)) { |
2011 | ||
2012 | if (!postcopy_start(s, &old_vm_running)) { | |
2013 | current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE; | |
2014 | entered_postcopy = true; | |
2015 | } | |
2016 | ||
2017 | continue; | |
2018 | } | |
2019 | /* Just another iteration step */ | |
89a02a9f | 2020 | qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy); |
c369f40d | 2021 | } else { |
09f6c85e | 2022 | trace_migration_thread_low_pending(pending_size); |
1d34e4bf | 2023 | migration_completion(s, current_active_state, |
36f48567 | 2024 | &old_vm_running, &start_time); |
09f6c85e | 2025 | break; |
c369f40d JQ |
2026 | } |
2027 | } | |
f4410a5d | 2028 | |
89a02a9f | 2029 | if (qemu_file_get_error(s->to_dst_file)) { |
48781e5b HZ |
2030 | migrate_set_state(&s->state, current_active_state, |
2031 | MIGRATION_STATUS_FAILED); | |
1d34e4bf | 2032 | trace_migration_thread_file_err(); |
fd45ee2c PB |
2033 | break; |
2034 | } | |
bc72ad67 | 2035 | current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0d82d0e8 | 2036 | if (current_time >= initial_time + BUFFER_DELAY) { |
89a02a9f HZ |
2037 | uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) - |
2038 | initial_bytes; | |
77417f10 | 2039 | uint64_t time_spent = current_time - initial_time; |
a694ee34 | 2040 | double bandwidth = (double)transferred_bytes / time_spent; |
faec066a | 2041 | threshold_size = bandwidth * s->parameters.downtime_limit; |
0d82d0e8 | 2042 | |
5b648de0 WY |
2043 | s->mbps = (((double) transferred_bytes * 8.0) / |
2044 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
7e114f8c | 2045 | |
9013dca5 | 2046 | trace_migrate_transferred(transferred_bytes, time_spent, |
faec066a | 2047 | bandwidth, threshold_size); |
90f8ae72 JQ |
2048 | /* if we haven't sent anything, we don't want to recalculate |
2049 | 10000 is a small enough number for our purposes */ | |
9360447d JQ |
2050 | if (ram_counters.dirty_pages_rate && transferred_bytes > 10000) { |
2051 | s->expected_downtime = ram_counters.dirty_pages_rate * | |
20afaed9 | 2052 | qemu_target_page_size() / bandwidth; |
90f8ae72 | 2053 | } |
0d82d0e8 | 2054 | |
89a02a9f | 2055 | qemu_file_reset_rate_limit(s->to_dst_file); |
0d82d0e8 | 2056 | initial_time = current_time; |
89a02a9f | 2057 | initial_bytes = qemu_ftell(s->to_dst_file); |
0d82d0e8 | 2058 | } |
89a02a9f | 2059 | if (qemu_file_rate_limit(s->to_dst_file)) { |
0d82d0e8 JQ |
2060 | /* usleep expects microseconds */ |
2061 | g_usleep((initial_time + BUFFER_DELAY - current_time)*1000); | |
2062 | } | |
a3fa1d78 PB |
2063 | } |
2064 | ||
1d34e4bf | 2065 | trace_migration_thread_after_loop(); |
070afca2 JH |
2066 | /* If we enabled cpu throttling for auto-converge, turn it off. */ |
2067 | cpu_throttle_stop(); | |
94f5a437 | 2068 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
070afca2 | 2069 | |
f4410a5d | 2070 | qemu_mutex_lock_iothread(); |
0b827d5e HZ |
2071 | /* |
2072 | * The resource has been allocated by migration will be reused in COLO | |
2073 | * process, so don't release them. | |
2074 | */ | |
2075 | if (!enable_colo) { | |
2076 | qemu_savevm_state_cleanup(); | |
2077 | } | |
31194731 | 2078 | if (s->state == MIGRATION_STATUS_COMPLETED) { |
89a02a9f | 2079 | uint64_t transferred_bytes = qemu_ftell(s->to_dst_file); |
a3fa1d78 | 2080 | s->total_time = end_time - s->total_time; |
1d34e4bf DDAG |
2081 | if (!entered_postcopy) { |
2082 | s->downtime = end_time - start_time; | |
2083 | } | |
d6ed7312 PL |
2084 | if (s->total_time) { |
2085 | s->mbps = (((double) transferred_bytes * 8.0) / | |
2086 | ((double) s->total_time)) / 1000; | |
2087 | } | |
a3fa1d78 PB |
2088 | runstate_set(RUN_STATE_POSTMIGRATE); |
2089 | } else { | |
0b827d5e HZ |
2090 | if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) { |
2091 | migrate_start_colo_process(s); | |
2092 | qemu_savevm_state_cleanup(); | |
2093 | /* | |
2094 | * Fixme: we will run VM in COLO no matter its old running state. | |
2095 | * After exited COLO, we will keep running. | |
2096 | */ | |
2097 | old_vm_running = true; | |
2098 | } | |
1d34e4bf | 2099 | if (old_vm_running && !entered_postcopy) { |
a3fa1d78 | 2100 | vm_start(); |
42da5550 DDAG |
2101 | } else { |
2102 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
2103 | runstate_set(RUN_STATE_POSTMIGRATE); | |
2104 | } | |
dba433c0 | 2105 | } |
0d82d0e8 | 2106 | } |
bb1fadc4 | 2107 | qemu_bh_schedule(s->cleanup_bh); |
dba433c0 | 2108 | qemu_mutex_unlock_iothread(); |
f4410a5d | 2109 | |
ab28bd23 | 2110 | rcu_unregister_thread(); |
0d82d0e8 JQ |
2111 | return NULL; |
2112 | } | |
2113 | ||
9848a404 | 2114 | void migrate_fd_connect(MigrationState *s) |
0d82d0e8 | 2115 | { |
2ff30257 | 2116 | s->expected_downtime = s->parameters.downtime_limit; |
bb1fadc4 | 2117 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s); |
0d82d0e8 | 2118 | |
9e4d2b98 | 2119 | qemu_file_set_blocking(s->to_dst_file, true); |
89a02a9f | 2120 | qemu_file_set_rate_limit(s->to_dst_file, |
2ff30257 | 2121 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); |
442773ce | 2122 | |
9287ac27 SH |
2123 | /* Notify before starting migration thread */ |
2124 | notifier_list_notify(&migration_state_notifiers, s); | |
2125 | ||
1d34e4bf | 2126 | /* |
c788ada8 PX |
2127 | * Open the return path. For postcopy, it is used exclusively. For |
2128 | * precopy, only if user specified "return-path" capability would | |
2129 | * QEMU uses the return path. | |
1d34e4bf | 2130 | */ |
c788ada8 | 2131 | if (migrate_postcopy_ram() || migrate_use_return_path()) { |
1d34e4bf DDAG |
2132 | if (open_return_path_on_source(s)) { |
2133 | error_report("Unable to open return-path for postcopy"); | |
48781e5b | 2134 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1d34e4bf DDAG |
2135 | MIGRATION_STATUS_FAILED); |
2136 | migrate_fd_cleanup(s); | |
2137 | return; | |
2138 | } | |
2139 | } | |
2140 | ||
009fad7f | 2141 | qemu_thread_create(&s->thread, "live_migration", migration_thread, s, |
bb1fadc4 | 2142 | QEMU_THREAD_JOINABLE); |
1d34e4bf | 2143 | s->migration_thread_running = true; |
0d82d0e8 | 2144 | } |
093e3c42 | 2145 | |
9d18af93 PX |
2146 | void migration_global_dump(Monitor *mon) |
2147 | { | |
2148 | MigrationState *ms = migrate_get_current(); | |
2149 | ||
2150 | monitor_printf(mon, "globals: store-global-state=%d, only_migratable=%d, " | |
2151 | "send-configuration=%d, send-section-footer=%d\n", | |
2152 | ms->store_global_state, ms->only_migratable, | |
2153 | ms->send_configuration, ms->send_section_footer); | |
2154 | } | |
2155 | ||
20814758 PX |
2156 | #define DEFINE_PROP_MIG_CAP(name, x) \ |
2157 | DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false) | |
2158 | ||
5272298c PX |
2159 | static Property migration_properties[] = { |
2160 | DEFINE_PROP_BOOL("store-global-state", MigrationState, | |
2161 | store_global_state, true), | |
3df663e5 | 2162 | DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false), |
71dd4c1a PX |
2163 | DEFINE_PROP_BOOL("send-configuration", MigrationState, |
2164 | send_configuration, true), | |
15c38503 PX |
2165 | DEFINE_PROP_BOOL("send-section-footer", MigrationState, |
2166 | send_section_footer, true), | |
89632faf PX |
2167 | |
2168 | /* Migration parameters */ | |
2169 | DEFINE_PROP_INT64("x-compress-level", MigrationState, | |
2170 | parameters.compress_level, | |
2171 | DEFAULT_MIGRATE_COMPRESS_LEVEL), | |
2172 | DEFINE_PROP_INT64("x-compress-threads", MigrationState, | |
2173 | parameters.compress_threads, | |
2174 | DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT), | |
2175 | DEFINE_PROP_INT64("x-decompress-threads", MigrationState, | |
2176 | parameters.decompress_threads, | |
2177 | DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT), | |
2178 | DEFINE_PROP_INT64("x-cpu-throttle-initial", MigrationState, | |
2179 | parameters.cpu_throttle_initial, | |
2180 | DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL), | |
2181 | DEFINE_PROP_INT64("x-cpu-throttle-increment", MigrationState, | |
2182 | parameters.cpu_throttle_increment, | |
2183 | DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT), | |
2184 | DEFINE_PROP_INT64("x-max-bandwidth", MigrationState, | |
2185 | parameters.max_bandwidth, MAX_THROTTLE), | |
2186 | DEFINE_PROP_INT64("x-downtime-limit", MigrationState, | |
2187 | parameters.downtime_limit, | |
2188 | DEFAULT_MIGRATE_SET_DOWNTIME), | |
2189 | DEFINE_PROP_INT64("x-checkpoint-delay", MigrationState, | |
2190 | parameters.x_checkpoint_delay, | |
2191 | DEFAULT_MIGRATE_X_CHECKPOINT_DELAY), | |
20814758 PX |
2192 | |
2193 | /* Migration capabilities */ | |
2194 | DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE), | |
2195 | DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL), | |
2196 | DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE), | |
2197 | DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS), | |
2198 | DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS), | |
2199 | DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS), | |
2200 | DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM), | |
2201 | DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO), | |
2202 | DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM), | |
2203 | DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK), | |
2204 | DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH), | |
2205 | ||
5272298c PX |
2206 | DEFINE_PROP_END_OF_LIST(), |
2207 | }; | |
2208 | ||
e5cb7e76 PX |
2209 | static void migration_class_init(ObjectClass *klass, void *data) |
2210 | { | |
2211 | DeviceClass *dc = DEVICE_CLASS(klass); | |
2212 | ||
2213 | dc->user_creatable = false; | |
5272298c | 2214 | dc->props = migration_properties; |
e5cb7e76 PX |
2215 | } |
2216 | ||
b91bf5e4 MAL |
2217 | static void migration_instance_finalize(Object *obj) |
2218 | { | |
2219 | MigrationState *ms = MIGRATION_OBJ(obj); | |
2220 | MigrationParameters *params = &ms->parameters; | |
2221 | ||
2222 | g_free(params->tls_hostname); | |
2223 | g_free(params->tls_creds); | |
2224 | } | |
2225 | ||
e5cb7e76 PX |
2226 | static void migration_instance_init(Object *obj) |
2227 | { | |
2228 | MigrationState *ms = MIGRATION_OBJ(obj); | |
8b0b29dc | 2229 | MigrationParameters *params = &ms->parameters; |
e5cb7e76 PX |
2230 | |
2231 | ms->state = MIGRATION_STATUS_NONE; | |
2232 | ms->xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE; | |
2233 | ms->mbps = -1; | |
8b0b29dc PX |
2234 | |
2235 | params->tls_hostname = g_strdup(""); | |
2236 | params->tls_creds = g_strdup(""); | |
2237 | ||
2238 | /* Set has_* up only for parameter checks */ | |
2239 | params->has_compress_level = true; | |
2240 | params->has_compress_threads = true; | |
2241 | params->has_decompress_threads = true; | |
2242 | params->has_cpu_throttle_initial = true; | |
2243 | params->has_cpu_throttle_increment = true; | |
2244 | params->has_max_bandwidth = true; | |
2245 | params->has_downtime_limit = true; | |
2246 | params->has_x_checkpoint_delay = true; | |
2247 | params->has_block_incremental = true; | |
2248 | } | |
2249 | ||
2250 | /* | |
2251 | * Return true if check pass, false otherwise. Error will be put | |
2252 | * inside errp if provided. | |
2253 | */ | |
2254 | static bool migration_object_check(MigrationState *ms, Error **errp) | |
2255 | { | |
6b19a7d9 PX |
2256 | MigrationCapabilityStatusList *head = NULL; |
2257 | /* Assuming all off */ | |
2258 | bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret; | |
2259 | int i; | |
2260 | ||
8b0b29dc PX |
2261 | if (!migrate_params_check(&ms->parameters, errp)) { |
2262 | return false; | |
2263 | } | |
2264 | ||
6b19a7d9 PX |
2265 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
2266 | if (ms->enabled_capabilities[i]) { | |
2267 | head = migrate_cap_add(head, i, true); | |
2268 | } | |
2269 | } | |
2270 | ||
2271 | ret = migrate_caps_check(cap_list, head, errp); | |
2272 | ||
2273 | /* It works with head == NULL */ | |
2274 | qapi_free_MigrationCapabilityStatusList(head); | |
2275 | ||
2276 | return ret; | |
e5cb7e76 PX |
2277 | } |
2278 | ||
2279 | static const TypeInfo migration_type = { | |
2280 | .name = TYPE_MIGRATION, | |
01f6e14c | 2281 | /* |
c8d3ff38 PX |
2282 | * NOTE: TYPE_MIGRATION is not really a device, as the object is |
2283 | * not created using qdev_create(), it is not attached to the qdev | |
2284 | * device tree, and it is never realized. | |
2285 | * | |
2286 | * TODO: Make this TYPE_OBJECT once QOM provides something like | |
2287 | * TYPE_DEVICE's "-global" properties. | |
01f6e14c | 2288 | */ |
e5cb7e76 PX |
2289 | .parent = TYPE_DEVICE, |
2290 | .class_init = migration_class_init, | |
2291 | .class_size = sizeof(MigrationClass), | |
2292 | .instance_size = sizeof(MigrationState), | |
2293 | .instance_init = migration_instance_init, | |
b91bf5e4 | 2294 | .instance_finalize = migration_instance_finalize, |
e5cb7e76 PX |
2295 | }; |
2296 | ||
2297 | static void register_migration_types(void) | |
2298 | { | |
2299 | type_register_static(&migration_type); | |
2300 | } | |
2301 | ||
2302 | type_init(register_migration_types); |