]>
Commit | Line | Data |
---|---|---|
5bb7910a AL |
1 | /* |
2 | * QEMU live migration | |
3 | * | |
4 | * Copyright IBM, Corp. 2008 | |
5 | * | |
6 | * Authors: | |
7 | * Anthony Liguori <aliguori@us.ibm.com> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
10 | * the COPYING file in the top-level directory. | |
11 | * | |
6b620ca3 PB |
12 | * Contributions after 2012-01-13 are licensed under the terms of the |
13 | * GNU GPL, version 2 or (at your option) any later version. | |
5bb7910a AL |
14 | */ |
15 | ||
1393a485 | 16 | #include "qemu/osdep.h" |
f348b6d1 | 17 | #include "qemu/cutils.h" |
d49b6836 | 18 | #include "qemu/error-report.h" |
795c40b8 | 19 | #include "migration/blocker.h" |
f4dbe1bf | 20 | #include "exec.h" |
7fcac4a2 | 21 | #include "fd.h" |
61e8b148 | 22 | #include "socket.h" |
e1a3ecee | 23 | #include "rdma.h" |
7b1e1a22 | 24 | #include "ram.h" |
84a899de | 25 | #include "migration/global_state.h" |
c4b63b7c | 26 | #include "migration/misc.h" |
6666c96a | 27 | #include "migration.h" |
20a519a0 | 28 | #include "savevm.h" |
40014d81 | 29 | #include "qemu-file-channel.h" |
08a0aee1 | 30 | #include "qemu-file.h" |
987772d9 | 31 | #include "migration/vmstate.h" |
737e150e | 32 | #include "block/block.h" |
cc7a8ea7 | 33 | #include "qapi/qmp/qerror.h" |
6c595cde | 34 | #include "qapi/util.h" |
ab28bd23 | 35 | #include "qemu/rcu.h" |
2c9e6fec | 36 | #include "block.h" |
be07b0ac | 37 | #include "postcopy-ram.h" |
766bd176 | 38 | #include "qemu/thread.h" |
791e7c82 | 39 | #include "qmp-commands.h" |
c09e5bb1 | 40 | #include "trace.h" |
598cd2bd | 41 | #include "qapi-event.h" |
51180423 | 42 | #include "exec/target_page.h" |
61b67d47 | 43 | #include "io/channel-buffer.h" |
35a6ed4f | 44 | #include "migration/colo.h" |
4ffdb337 | 45 | #include "hw/boards.h" |
9d18af93 | 46 | #include "monitor/monitor.h" |
065e2813 | 47 | |
dc325627 | 48 | #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */ |
5bb7910a | 49 | |
5b4e1eb7 JQ |
50 | /* Amount of time to allocate to each "chunk" of bandwidth-throttled |
51 | * data. */ | |
52 | #define BUFFER_DELAY 100 | |
53 | #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) | |
54 | ||
2ff30257 AA |
55 | /* Time in milliseconds we are allowed to stop the source, |
56 | * for sending the last part */ | |
57 | #define DEFAULT_MIGRATE_SET_DOWNTIME 300 | |
58 | ||
87c9cc1c DHB |
59 | /* Maximum migrate downtime set to 2000 seconds */ |
60 | #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 | |
61 | #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) | |
62 | ||
8706d2d5 LL |
63 | /* Default compression thread count */ |
64 | #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 | |
3fcb38c2 LL |
65 | /* Default decompression thread count, usually decompression is at |
66 | * least 4 times as fast as compression.*/ | |
67 | #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 | |
8706d2d5 LL |
68 | /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
69 | #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 | |
1626fee3 | 70 | /* Define default autoconverge cpu throttle migration parameters */ |
d85a31d1 JH |
71 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 |
72 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 | |
8706d2d5 | 73 | |
17ad9b35 OW |
74 | /* Migration XBZRLE default cache size */ |
75 | #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024) | |
76 | ||
68b53591 HZ |
77 | /* The delay time (in ms) between two COLO checkpoints |
78 | * Note: Please change this default value to 10000 when we support hybrid mode. | |
79 | */ | |
80 | #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200 | |
81 | ||
99a0db9b GH |
82 | static NotifierList migration_state_notifiers = |
83 | NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); | |
84 | ||
adde220a DDAG |
85 | static bool deferred_incoming; |
86 | ||
da6f1790 JQ |
87 | /* Messages sent on the return path from destination to source */ |
88 | enum mig_rp_message_type { | |
89 | MIG_RP_MSG_INVALID = 0, /* Must be 0 */ | |
90 | MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ | |
91 | MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ | |
92 | ||
93 | MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ | |
94 | MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ | |
95 | ||
96 | MIG_RP_MSG_MAX | |
97 | }; | |
98 | ||
17549e84 JQ |
99 | /* When we add fault tolerance, we could have several |
100 | migrations at once. For now we don't need to add | |
101 | dynamic creation of migration */ | |
102 | ||
e5cb7e76 PX |
103 | static MigrationState *current_migration; |
104 | ||
8b0b29dc PX |
105 | static bool migration_object_check(MigrationState *ms, Error **errp); |
106 | ||
e5cb7e76 PX |
107 | void migration_object_init(void) |
108 | { | |
4ffdb337 | 109 | MachineState *ms = MACHINE(qdev_get_machine()); |
8b0b29dc | 110 | Error *err = NULL; |
4ffdb337 | 111 | |
e5cb7e76 PX |
112 | /* This can only be called once. */ |
113 | assert(!current_migration); | |
114 | current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); | |
4ffdb337 | 115 | |
8b0b29dc PX |
116 | if (!migration_object_check(current_migration, &err)) { |
117 | error_report_err(err); | |
118 | exit(1); | |
119 | } | |
120 | ||
4ffdb337 PX |
121 | /* |
122 | * We cannot really do this in migration_instance_init() since at | |
123 | * that time global properties are not yet applied, then this | |
124 | * value will be definitely replaced by something else. | |
125 | */ | |
126 | if (ms->enforce_config_section) { | |
127 | current_migration->send_configuration = true; | |
128 | } | |
e5cb7e76 PX |
129 | } |
130 | ||
bca7856a | 131 | /* For outgoing */ |
859bc756 | 132 | MigrationState *migrate_get_current(void) |
17549e84 | 133 | { |
e5cb7e76 PX |
134 | /* This can only be called after the object created. */ |
135 | assert(current_migration); | |
136 | return current_migration; | |
17549e84 JQ |
137 | } |
138 | ||
bca7856a DDAG |
139 | MigrationIncomingState *migration_incoming_get_current(void) |
140 | { | |
b4b076da JQ |
141 | static bool once; |
142 | static MigrationIncomingState mis_current; | |
bca7856a | 143 | |
b4b076da JQ |
144 | if (!once) { |
145 | mis_current.state = MIGRATION_STATUS_NONE; | |
146 | memset(&mis_current, 0, sizeof(MigrationIncomingState)); | |
b4b076da JQ |
147 | qemu_mutex_init(&mis_current.rp_mutex); |
148 | qemu_event_init(&mis_current.main_thread_load_event, false); | |
149 | once = true; | |
150 | } | |
151 | return &mis_current; | |
bca7856a DDAG |
152 | } |
153 | ||
154 | void migration_incoming_state_destroy(void) | |
155 | { | |
b4b076da JQ |
156 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
157 | ||
3482655b | 158 | if (mis->to_src_file) { |
660819b1 PX |
159 | /* Tell source that we are done */ |
160 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
161 | qemu_fclose(mis->to_src_file); |
162 | mis->to_src_file = NULL; | |
163 | } | |
164 | ||
660819b1 PX |
165 | if (mis->from_src_file) { |
166 | qemu_fclose(mis->from_src_file); | |
167 | mis->from_src_file = NULL; | |
168 | } | |
169 | ||
b4b076da | 170 | qemu_event_destroy(&mis->main_thread_load_event); |
bca7856a DDAG |
171 | } |
172 | ||
b05dc723 JQ |
173 | static void migrate_generate_event(int new_state) |
174 | { | |
175 | if (migrate_use_events()) { | |
176 | qapi_event_send_migration(new_state, &error_abort); | |
b05dc723 JQ |
177 | } |
178 | } | |
179 | ||
adde220a DDAG |
180 | /* |
181 | * Called on -incoming with a defer: uri. | |
182 | * The migration can be started later after any parameters have been | |
183 | * changed. | |
184 | */ | |
185 | static void deferred_incoming_migration(Error **errp) | |
186 | { | |
187 | if (deferred_incoming) { | |
188 | error_setg(errp, "Incoming migration already deferred"); | |
189 | } | |
190 | deferred_incoming = true; | |
191 | } | |
192 | ||
da6f1790 JQ |
193 | /* |
194 | * Send a message on the return channel back to the source | |
195 | * of the migration. | |
196 | */ | |
197 | static void migrate_send_rp_message(MigrationIncomingState *mis, | |
198 | enum mig_rp_message_type message_type, | |
199 | uint16_t len, void *data) | |
200 | { | |
201 | trace_migrate_send_rp_message((int)message_type, len); | |
202 | qemu_mutex_lock(&mis->rp_mutex); | |
203 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); | |
204 | qemu_put_be16(mis->to_src_file, len); | |
205 | qemu_put_buffer(mis->to_src_file, data, len); | |
206 | qemu_fflush(mis->to_src_file); | |
207 | qemu_mutex_unlock(&mis->rp_mutex); | |
208 | } | |
209 | ||
1e2d90eb DDAG |
210 | /* Request a range of pages from the source VM at the given |
211 | * start address. | |
212 | * rbname: Name of the RAMBlock to request the page in, if NULL it's the same | |
213 | * as the last request (a name must have been given previously) | |
214 | * Start: Address offset within the RB | |
215 | * Len: Length in bytes required - must be a multiple of pagesize | |
216 | */ | |
217 | void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname, | |
218 | ram_addr_t start, size_t len) | |
219 | { | |
cb8d4c8f | 220 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb DDAG |
221 | size_t msglen = 12; /* start + len */ |
222 | ||
223 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
224 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
225 | ||
226 | if (rbname) { | |
227 | int rbname_len = strlen(rbname); | |
228 | assert(rbname_len < 256); | |
229 | ||
230 | bufc[msglen++] = rbname_len; | |
231 | memcpy(bufc + msglen, rbname, rbname_len); | |
232 | msglen += rbname_len; | |
233 | migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc); | |
234 | } else { | |
235 | migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc); | |
236 | } | |
237 | } | |
238 | ||
43eaae28 | 239 | void qemu_start_incoming_migration(const char *uri, Error **errp) |
5bb7910a | 240 | { |
34c9dd8e AL |
241 | const char *p; |
242 | ||
7cf1fe6d | 243 | qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort); |
adde220a DDAG |
244 | if (!strcmp(uri, "defer")) { |
245 | deferred_incoming_migration(errp); | |
246 | } else if (strstart(uri, "tcp:", &p)) { | |
43eaae28 | 247 | tcp_start_incoming_migration(p, errp); |
2da776db | 248 | #ifdef CONFIG_RDMA |
adde220a | 249 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
250 | rdma_start_incoming_migration(p, errp); |
251 | #endif | |
adde220a | 252 | } else if (strstart(uri, "exec:", &p)) { |
43eaae28 | 253 | exec_start_incoming_migration(p, errp); |
adde220a | 254 | } else if (strstart(uri, "unix:", &p)) { |
43eaae28 | 255 | unix_start_incoming_migration(p, errp); |
adde220a | 256 | } else if (strstart(uri, "fd:", &p)) { |
43eaae28 | 257 | fd_start_incoming_migration(p, errp); |
adde220a | 258 | } else { |
312fd5f2 | 259 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 260 | } |
5bb7910a AL |
261 | } |
262 | ||
0aa6aefc DL |
263 | static void process_incoming_migration_bh(void *opaque) |
264 | { | |
265 | Error *local_err = NULL; | |
266 | MigrationIncomingState *mis = opaque; | |
267 | ||
ace21a58 KW |
268 | /* Make sure all file formats flush their mutable metadata. |
269 | * If we get an error here, just don't restart the VM yet. */ | |
0aa6aefc | 270 | bdrv_invalidate_cache_all(&local_err); |
d35ff5e6 | 271 | if (local_err) { |
ace21a58 | 272 | error_report_err(local_err); |
d35ff5e6 KW |
273 | local_err = NULL; |
274 | autostart = false; | |
275 | } | |
276 | ||
0aa6aefc DL |
277 | /* |
278 | * This must happen after all error conditions are dealt with and | |
279 | * we're sure the VM is going to be running on this host. | |
280 | */ | |
281 | qemu_announce_self(); | |
282 | ||
283 | /* If global state section was not received or we are in running | |
284 | state, we need to obey autostart. Any other state is set with | |
285 | runstate_set. */ | |
286 | ||
287 | if (!global_state_received() || | |
288 | global_state_get_runstate() == RUN_STATE_RUNNING) { | |
289 | if (autostart) { | |
290 | vm_start(); | |
291 | } else { | |
292 | runstate_set(RUN_STATE_PAUSED); | |
293 | } | |
294 | } else { | |
295 | runstate_set(global_state_get_runstate()); | |
296 | } | |
0aa6aefc DL |
297 | /* |
298 | * This must happen after any state changes since as soon as an external | |
299 | * observer sees this event they might start to prod at the VM assuming | |
300 | * it's ready to use. | |
301 | */ | |
302 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
303 | MIGRATION_STATUS_COMPLETED); | |
304 | qemu_bh_delete(mis->bh); | |
305 | migration_incoming_state_destroy(); | |
306 | } | |
307 | ||
82a4da79 | 308 | static void process_incoming_migration_co(void *opaque) |
511c0231 | 309 | { |
82a4da79 | 310 | QEMUFile *f = opaque; |
b4b076da | 311 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 312 | PostcopyState ps; |
1c12e1f5 PB |
313 | int ret; |
314 | ||
b4b076da | 315 | mis->from_src_file = f; |
67f11b5c | 316 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 317 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
93d7af6f HZ |
318 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
319 | MIGRATION_STATUS_ACTIVE); | |
1c12e1f5 | 320 | ret = qemu_loadvm_state(f); |
bca7856a | 321 | |
e9bef235 DDAG |
322 | ps = postcopy_state_get(); |
323 | trace_process_incoming_migration_co_end(ret, ps); | |
324 | if (ps != POSTCOPY_INCOMING_NONE) { | |
325 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
326 | /* | |
327 | * Where a migration had postcopy enabled (and thus went to advise) | |
328 | * but managed to complete within the precopy period, we can use | |
329 | * the normal exit. | |
330 | */ | |
331 | postcopy_ram_incoming_cleanup(mis); | |
332 | } else if (ret >= 0) { | |
333 | /* | |
334 | * Postcopy was started, cleanup should happen at the end of the | |
335 | * postcopy thread. | |
336 | */ | |
337 | trace_process_incoming_migration_co_postcopy_end_main(); | |
338 | return; | |
339 | } | |
340 | /* Else if something went wrong then just fall out of the normal exit */ | |
341 | } | |
342 | ||
25d0c16f HZ |
343 | /* we get COLO info, and know if we are in COLO mode */ |
344 | if (!ret && migration_incoming_enable_colo()) { | |
345 | mis->migration_incoming_co = qemu_coroutine_self(); | |
346 | qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", | |
347 | colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); | |
348 | mis->have_colo_incoming_thread = true; | |
349 | qemu_coroutine_yield(); | |
350 | ||
351 | /* Wait checkpoint incoming thread exit before free resource */ | |
352 | qemu_thread_join(&mis->colo_incoming_thread); | |
353 | } | |
354 | ||
1c12e1f5 | 355 | if (ret < 0) { |
93d7af6f HZ |
356 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, |
357 | MIGRATION_STATUS_FAILED); | |
db80face | 358 | error_report("load of migration failed: %s", strerror(-ret)); |
3a0f2cea | 359 | qemu_fclose(mis->from_src_file); |
4aead692 | 360 | exit(EXIT_FAILURE); |
511c0231 | 361 | } |
0aa6aefc DL |
362 | mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); |
363 | qemu_bh_schedule(mis->bh); | |
511c0231 JQ |
364 | } |
365 | ||
22724f49 | 366 | void migration_fd_process_incoming(QEMUFile *f) |
82a4da79 | 367 | { |
0b8b8753 | 368 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f); |
82a4da79 | 369 | |
06ad5135 | 370 | qemu_file_set_blocking(f, false); |
0b8b8753 | 371 | qemu_coroutine_enter(co); |
82a4da79 PB |
372 | } |
373 | ||
6decec93 DDAG |
374 | /* |
375 | * Send a 'SHUT' message on the return channel with the given value | |
376 | * to indicate that we've finished with the RP. Non-0 value indicates | |
377 | * error. | |
378 | */ | |
379 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
380 | uint32_t value) | |
381 | { | |
382 | uint32_t buf; | |
383 | ||
384 | buf = cpu_to_be32(value); | |
385 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
386 | } | |
387 | ||
388 | /* | |
389 | * Send a 'PONG' message on the return channel with the given value | |
390 | * (normally in response to a 'PING') | |
391 | */ | |
392 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
393 | uint32_t value) | |
394 | { | |
395 | uint32_t buf; | |
396 | ||
397 | buf = cpu_to_be32(value); | |
398 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
399 | } | |
400 | ||
bbf6da32 OW |
401 | MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) |
402 | { | |
403 | MigrationCapabilityStatusList *head = NULL; | |
404 | MigrationCapabilityStatusList *caps; | |
405 | MigrationState *s = migrate_get_current(); | |
406 | int i; | |
407 | ||
387eedeb | 408 | caps = NULL; /* silence compiler warning */ |
7fb1cf16 | 409 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
ed1701c6 DDAG |
410 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
411 | if (i == MIGRATION_CAPABILITY_BLOCK) { | |
412 | continue; | |
413 | } | |
414 | #endif | |
bbf6da32 OW |
415 | if (head == NULL) { |
416 | head = g_malloc0(sizeof(*caps)); | |
417 | caps = head; | |
418 | } else { | |
419 | caps->next = g_malloc0(sizeof(*caps)); | |
420 | caps = caps->next; | |
421 | } | |
422 | caps->value = | |
423 | g_malloc(sizeof(*caps->value)); | |
424 | caps->value->capability = i; | |
425 | caps->value->state = s->enabled_capabilities[i]; | |
426 | } | |
427 | ||
428 | return head; | |
429 | } | |
430 | ||
85de8323 LL |
431 | MigrationParameters *qmp_query_migrate_parameters(Error **errp) |
432 | { | |
433 | MigrationParameters *params; | |
434 | MigrationState *s = migrate_get_current(); | |
435 | ||
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; |
de63ab61 | 447 | params->has_tls_creds = !!s->parameters.tls_creds; |
69ef1f36 | 448 | params->tls_creds = g_strdup(s->parameters.tls_creds); |
de63ab61 | 449 | params->has_tls_hostname = !!s->parameters.tls_hostname; |
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 | ||
476c72aa | 744 | static void migrate_params_apply(MigrationParameters *params) |
16d063bc PX |
745 | { |
746 | MigrationState *s = migrate_get_current(); | |
747 | ||
7f375e04 EB |
748 | if (params->has_compress_level) { |
749 | s->parameters.compress_level = params->compress_level; | |
85de8323 | 750 | } |
476c72aa | 751 | |
7f375e04 EB |
752 | if (params->has_compress_threads) { |
753 | s->parameters.compress_threads = params->compress_threads; | |
85de8323 | 754 | } |
476c72aa | 755 | |
7f375e04 EB |
756 | if (params->has_decompress_threads) { |
757 | s->parameters.decompress_threads = params->decompress_threads; | |
85de8323 | 758 | } |
476c72aa | 759 | |
7f375e04 EB |
760 | if (params->has_cpu_throttle_initial) { |
761 | s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; | |
1626fee3 | 762 | } |
476c72aa | 763 | |
7f375e04 EB |
764 | if (params->has_cpu_throttle_increment) { |
765 | s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; | |
1626fee3 | 766 | } |
476c72aa | 767 | |
7f375e04 | 768 | if (params->has_tls_creds) { |
69ef1f36 | 769 | g_free(s->parameters.tls_creds); |
7f375e04 | 770 | s->parameters.tls_creds = g_strdup(params->tls_creds); |
69ef1f36 | 771 | } |
476c72aa | 772 | |
7f375e04 | 773 | if (params->has_tls_hostname) { |
69ef1f36 | 774 | g_free(s->parameters.tls_hostname); |
7f375e04 | 775 | s->parameters.tls_hostname = g_strdup(params->tls_hostname); |
69ef1f36 | 776 | } |
476c72aa | 777 | |
2ff30257 AA |
778 | if (params->has_max_bandwidth) { |
779 | s->parameters.max_bandwidth = params->max_bandwidth; | |
780 | if (s->to_dst_file) { | |
781 | qemu_file_set_rate_limit(s->to_dst_file, | |
782 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); | |
783 | } | |
784 | } | |
476c72aa | 785 | |
2ff30257 AA |
786 | if (params->has_downtime_limit) { |
787 | s->parameters.downtime_limit = params->downtime_limit; | |
788 | } | |
68b53591 HZ |
789 | |
790 | if (params->has_x_checkpoint_delay) { | |
791 | s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; | |
479125d5 HZ |
792 | if (migration_in_colo_state()) { |
793 | colo_checkpoint_notify(s); | |
794 | } | |
68b53591 | 795 | } |
476c72aa | 796 | |
2833c59b JQ |
797 | if (params->has_block_incremental) { |
798 | s->parameters.block_incremental = params->block_incremental; | |
799 | } | |
85de8323 LL |
800 | } |
801 | ||
476c72aa PX |
802 | void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp) |
803 | { | |
804 | if (!migrate_params_check(params, errp)) { | |
805 | /* Invalid parameter */ | |
806 | return; | |
807 | } | |
808 | ||
809 | migrate_params_apply(params); | |
810 | } | |
811 | ||
2594f56d | 812 | |
4886a1bc DDAG |
813 | void qmp_migrate_start_postcopy(Error **errp) |
814 | { | |
815 | MigrationState *s = migrate_get_current(); | |
816 | ||
817 | if (!migrate_postcopy_ram()) { | |
a54d340b | 818 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
819 | " the start of migration"); |
820 | return; | |
821 | } | |
822 | ||
823 | if (s->state == MIGRATION_STATUS_NONE) { | |
824 | error_setg(errp, "Postcopy must be started after migration has been" | |
825 | " started"); | |
826 | return; | |
827 | } | |
828 | /* | |
829 | * we don't error if migration has finished since that would be racy | |
830 | * with issuing this command. | |
831 | */ | |
832 | atomic_set(&s->start_postcopy, true); | |
833 | } | |
834 | ||
065e2813 AL |
835 | /* shared migration helpers */ |
836 | ||
48781e5b | 837 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 838 | { |
48781e5b | 839 | if (atomic_cmpxchg(state, old_state, new_state) == old_state) { |
4ba4bc5e | 840 | trace_migrate_set_state(new_state); |
b05dc723 | 841 | migrate_generate_event(new_state); |
51cf4c1a Z |
842 | } |
843 | } | |
844 | ||
4e4a3d3a PX |
845 | static MigrationCapabilityStatusList *migrate_cap_add( |
846 | MigrationCapabilityStatusList *list, | |
847 | MigrationCapability index, | |
848 | bool state) | |
2833c59b JQ |
849 | { |
850 | MigrationCapabilityStatusList *cap; | |
851 | ||
852 | cap = g_new0(MigrationCapabilityStatusList, 1); | |
853 | cap->value = g_new0(MigrationCapabilityStatus, 1); | |
4e4a3d3a PX |
854 | cap->value->capability = index; |
855 | cap->value->state = state; | |
856 | cap->next = list; | |
857 | ||
858 | return cap; | |
859 | } | |
860 | ||
861 | void migrate_set_block_enabled(bool value, Error **errp) | |
862 | { | |
863 | MigrationCapabilityStatusList *cap; | |
864 | ||
865 | cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value); | |
2833c59b JQ |
866 | qmp_migrate_set_capabilities(cap, errp); |
867 | qapi_free_MigrationCapabilityStatusList(cap); | |
868 | } | |
869 | ||
870 | static void migrate_set_block_incremental(MigrationState *s, bool value) | |
871 | { | |
872 | s->parameters.block_incremental = value; | |
873 | } | |
874 | ||
875 | static void block_cleanup_parameters(MigrationState *s) | |
876 | { | |
877 | if (s->must_remove_block_options) { | |
878 | /* setting to false can never fail */ | |
879 | migrate_set_block_enabled(false, &error_abort); | |
880 | migrate_set_block_incremental(s, false); | |
881 | s->must_remove_block_options = false; | |
882 | } | |
883 | } | |
884 | ||
bb1fadc4 | 885 | static void migrate_fd_cleanup(void *opaque) |
065e2813 | 886 | { |
bb1fadc4 PB |
887 | MigrationState *s = opaque; |
888 | ||
889 | qemu_bh_delete(s->cleanup_bh); | |
890 | s->cleanup_bh = NULL; | |
891 | ||
89a02a9f | 892 | if (s->to_dst_file) { |
9013dca5 | 893 | trace_migrate_fd_cleanup(); |
404a7c05 | 894 | qemu_mutex_unlock_iothread(); |
1d34e4bf DDAG |
895 | if (s->migration_thread_running) { |
896 | qemu_thread_join(&s->thread); | |
897 | s->migration_thread_running = false; | |
898 | } | |
404a7c05 PB |
899 | qemu_mutex_lock_iothread(); |
900 | ||
89a02a9f HZ |
901 | qemu_fclose(s->to_dst_file); |
902 | s->to_dst_file = NULL; | |
065e2813 AL |
903 | } |
904 | ||
9ec055ae DDAG |
905 | assert((s->state != MIGRATION_STATUS_ACTIVE) && |
906 | (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE)); | |
7a2c1721 | 907 | |
94f5a437 | 908 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 909 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 910 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 911 | } |
a3fa1d78 PB |
912 | |
913 | notifier_list_notify(&migration_state_notifiers, s); | |
2833c59b | 914 | block_cleanup_parameters(s); |
065e2813 AL |
915 | } |
916 | ||
d59ce6f3 | 917 | void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 918 | { |
25174055 | 919 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 920 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
921 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
922 | MIGRATION_STATUS_FAILED); | |
d59ce6f3 DB |
923 | if (!s->error) { |
924 | s->error = error_copy(error); | |
925 | } | |
bb1fadc4 | 926 | notifier_list_notify(&migration_state_notifiers, s); |
2833c59b | 927 | block_cleanup_parameters(s); |
458cf28e JQ |
928 | } |
929 | ||
0edda1c4 | 930 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 931 | { |
6f2b811a | 932 | int old_state ; |
89a02a9f | 933 | QEMUFile *f = migrate_get_current()->to_dst_file; |
9013dca5 | 934 | trace_migrate_fd_cancel(); |
065e2813 | 935 | |
70b20477 DDAG |
936 | if (s->rp_state.from_dst_file) { |
937 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
938 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
939 | } | |
940 | ||
6f2b811a Z |
941 | do { |
942 | old_state = s->state; | |
f6844b99 | 943 | if (!migration_is_setup_or_active(old_state)) { |
6f2b811a Z |
944 | break; |
945 | } | |
48781e5b | 946 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 947 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
948 | |
949 | /* | |
950 | * If we're unlucky the migration code might be stuck somewhere in a | |
951 | * send/write while the network has failed and is waiting to timeout; | |
952 | * if we've got shutdown(2) available then we can force it to quit. | |
953 | * The outgoing qemu file gets closed in migrate_fd_cleanup that is | |
954 | * called in a bh, so there is no race against this cancel. | |
955 | */ | |
31194731 | 956 | if (s->state == MIGRATION_STATUS_CANCELLING && f) { |
a26ba26e DDAG |
957 | qemu_file_shutdown(f); |
958 | } | |
1d2acc31 HZ |
959 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
960 | Error *local_err = NULL; | |
961 | ||
962 | bdrv_invalidate_cache_all(&local_err); | |
963 | if (local_err) { | |
964 | error_report_err(local_err); | |
965 | } else { | |
966 | s->block_inactive = false; | |
967 | } | |
968 | } | |
2833c59b | 969 | block_cleanup_parameters(s); |
065e2813 AL |
970 | } |
971 | ||
99a0db9b GH |
972 | void add_migration_state_change_notifier(Notifier *notify) |
973 | { | |
974 | notifier_list_add(&migration_state_notifiers, notify); | |
975 | } | |
976 | ||
977 | void remove_migration_state_change_notifier(Notifier *notify) | |
978 | { | |
31552529 | 979 | notifier_remove(notify); |
99a0db9b GH |
980 | } |
981 | ||
02edd2e7 | 982 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 983 | { |
31194731 | 984 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
985 | } |
986 | ||
7073693b | 987 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 988 | { |
31194731 | 989 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 990 | } |
0edda1c4 | 991 | |
afe2df69 GH |
992 | bool migration_has_failed(MigrationState *s) |
993 | { | |
31194731 HZ |
994 | return (s->state == MIGRATION_STATUS_CANCELLED || |
995 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
996 | } |
997 | ||
5727309d | 998 | bool migration_in_postcopy(void) |
9ec055ae | 999 | { |
5727309d JQ |
1000 | MigrationState *s = migrate_get_current(); |
1001 | ||
9ec055ae DDAG |
1002 | return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); |
1003 | } | |
1004 | ||
b82fc321 DDAG |
1005 | bool migration_in_postcopy_after_devices(MigrationState *s) |
1006 | { | |
5727309d | 1007 | return migration_in_postcopy() && s->postcopy_after_devices; |
b82fc321 DDAG |
1008 | } |
1009 | ||
fab35005 | 1010 | bool migration_is_idle(void) |
fe44dc91 | 1011 | { |
fab35005 | 1012 | MigrationState *s = migrate_get_current(); |
fe44dc91 AA |
1013 | |
1014 | switch (s->state) { | |
1015 | case MIGRATION_STATUS_NONE: | |
1016 | case MIGRATION_STATUS_CANCELLED: | |
1017 | case MIGRATION_STATUS_COMPLETED: | |
1018 | case MIGRATION_STATUS_FAILED: | |
1019 | return true; | |
1020 | case MIGRATION_STATUS_SETUP: | |
1021 | case MIGRATION_STATUS_CANCELLING: | |
1022 | case MIGRATION_STATUS_ACTIVE: | |
1023 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1024 | case MIGRATION_STATUS_COLO: | |
1025 | return false; | |
1026 | case MIGRATION_STATUS__MAX: | |
1027 | g_assert_not_reached(); | |
1028 | } | |
1029 | ||
1030 | return false; | |
1031 | } | |
1032 | ||
a0762d9e | 1033 | MigrationState *migrate_init(void) |
0edda1c4 | 1034 | { |
17549e84 | 1035 | MigrationState *s = migrate_get_current(); |
bbf6da32 | 1036 | |
389775d1 DDAG |
1037 | /* |
1038 | * Reinitialise all migration state, except | |
1039 | * parameters/capabilities that the user set, and | |
1040 | * locks. | |
1041 | */ | |
1042 | s->bytes_xfer = 0; | |
1043 | s->xfer_limit = 0; | |
1044 | s->cleanup_bh = 0; | |
89a02a9f | 1045 | s->to_dst_file = NULL; |
389775d1 | 1046 | s->state = MIGRATION_STATUS_NONE; |
389775d1 DDAG |
1047 | s->rp_state.from_dst_file = NULL; |
1048 | s->rp_state.error = false; | |
1049 | s->mbps = 0.0; | |
1050 | s->downtime = 0; | |
1051 | s->expected_downtime = 0; | |
389775d1 | 1052 | s->setup_time = 0; |
389775d1 | 1053 | s->start_postcopy = false; |
b82fc321 | 1054 | s->postcopy_after_devices = false; |
389775d1 | 1055 | s->migration_thread_running = false; |
d59ce6f3 DB |
1056 | error_free(s->error); |
1057 | s->error = NULL; | |
389775d1 | 1058 | |
48781e5b | 1059 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 1060 | |
bc72ad67 | 1061 | s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0edda1c4 JQ |
1062 | return s; |
1063 | } | |
cab30143 | 1064 | |
fa2756b7 AL |
1065 | static GSList *migration_blockers; |
1066 | ||
fe44dc91 | 1067 | int migrate_add_blocker(Error *reason, Error **errp) |
fa2756b7 | 1068 | { |
3df663e5 | 1069 | if (migrate_get_current()->only_migratable) { |
b67b8c3a AA |
1070 | error_propagate(errp, error_copy(reason)); |
1071 | error_prepend(errp, "disallowing migration blocker " | |
1072 | "(--only_migratable) for: "); | |
1073 | return -EACCES; | |
1074 | } | |
1075 | ||
fab35005 | 1076 | if (migration_is_idle()) { |
fe44dc91 AA |
1077 | migration_blockers = g_slist_prepend(migration_blockers, reason); |
1078 | return 0; | |
1079 | } | |
1080 | ||
1081 | error_propagate(errp, error_copy(reason)); | |
1082 | error_prepend(errp, "disallowing migration blocker (migration in " | |
1083 | "progress) for: "); | |
1084 | return -EBUSY; | |
fa2756b7 AL |
1085 | } |
1086 | ||
1087 | void migrate_del_blocker(Error *reason) | |
1088 | { | |
1089 | migration_blockers = g_slist_remove(migration_blockers, reason); | |
1090 | } | |
1091 | ||
bf1ae1f4 DDAG |
1092 | void qmp_migrate_incoming(const char *uri, Error **errp) |
1093 | { | |
1094 | Error *local_err = NULL; | |
4debb5f5 | 1095 | static bool once = true; |
bf1ae1f4 DDAG |
1096 | |
1097 | if (!deferred_incoming) { | |
4debb5f5 | 1098 | error_setg(errp, "For use with '-incoming defer'"); |
bf1ae1f4 DDAG |
1099 | return; |
1100 | } | |
4debb5f5 DDAG |
1101 | if (!once) { |
1102 | error_setg(errp, "The incoming migration has already been started"); | |
1103 | } | |
bf1ae1f4 DDAG |
1104 | |
1105 | qemu_start_incoming_migration(uri, &local_err); | |
1106 | ||
1107 | if (local_err) { | |
1108 | error_propagate(errp, local_err); | |
1109 | return; | |
1110 | } | |
1111 | ||
4debb5f5 | 1112 | once = false; |
bf1ae1f4 DDAG |
1113 | } |
1114 | ||
24f3902b GK |
1115 | bool migration_is_blocked(Error **errp) |
1116 | { | |
1117 | if (qemu_savevm_state_blocked(errp)) { | |
1118 | return true; | |
1119 | } | |
1120 | ||
1121 | if (migration_blockers) { | |
250561e1 | 1122 | error_propagate(errp, error_copy(migration_blockers->data)); |
24f3902b GK |
1123 | return true; |
1124 | } | |
1125 | ||
1126 | return false; | |
1127 | } | |
1128 | ||
e1c37d0e LC |
1129 | void qmp_migrate(const char *uri, bool has_blk, bool blk, |
1130 | bool has_inc, bool inc, bool has_detach, bool detach, | |
1131 | Error **errp) | |
cab30143 | 1132 | { |
be7059cd | 1133 | Error *local_err = NULL; |
17549e84 | 1134 | MigrationState *s = migrate_get_current(); |
cab30143 | 1135 | const char *p; |
cab30143 | 1136 | |
f6844b99 | 1137 | if (migration_is_setup_or_active(s->state) || |
0b827d5e HZ |
1138 | s->state == MIGRATION_STATUS_CANCELLING || |
1139 | s->state == MIGRATION_STATUS_COLO) { | |
c6bd8c70 | 1140 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
e1c37d0e | 1141 | return; |
cab30143 | 1142 | } |
ca99993a DDAG |
1143 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
1144 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
1145 | return; | |
1146 | } | |
1147 | ||
24f3902b | 1148 | if (migration_is_blocked(errp)) { |
e1c37d0e | 1149 | return; |
fa2756b7 AL |
1150 | } |
1151 | ||
2833c59b JQ |
1152 | if ((has_blk && blk) || (has_inc && inc)) { |
1153 | if (migrate_use_block() || migrate_use_block_incremental()) { | |
1154 | error_setg(errp, "Command options are incompatible with " | |
1155 | "current migration capabilities"); | |
1156 | return; | |
1157 | } | |
1158 | migrate_set_block_enabled(true, &local_err); | |
1159 | if (local_err) { | |
1160 | error_propagate(errp, local_err); | |
1161 | return; | |
1162 | } | |
1163 | s->must_remove_block_options = true; | |
1164 | } | |
1165 | ||
1166 | if (has_inc && inc) { | |
1167 | migrate_set_block_incremental(s, true); | |
1168 | } | |
1169 | ||
a0762d9e | 1170 | s = migrate_init(); |
cab30143 JQ |
1171 | |
1172 | if (strstart(uri, "tcp:", &p)) { | |
f37afb5a | 1173 | tcp_start_outgoing_migration(s, p, &local_err); |
2da776db | 1174 | #ifdef CONFIG_RDMA |
41310c68 | 1175 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
1176 | rdma_start_outgoing_migration(s, p, &local_err); |
1177 | #endif | |
cab30143 | 1178 | } else if (strstart(uri, "exec:", &p)) { |
f37afb5a | 1179 | exec_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1180 | } else if (strstart(uri, "unix:", &p)) { |
f37afb5a | 1181 | unix_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1182 | } else if (strstart(uri, "fd:", &p)) { |
f37afb5a | 1183 | fd_start_outgoing_migration(s, p, &local_err); |
99a0db9b | 1184 | } else { |
c6bd8c70 MA |
1185 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", |
1186 | "a valid migration protocol"); | |
48781e5b HZ |
1187 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1188 | MIGRATION_STATUS_FAILED); | |
e1c37d0e | 1189 | return; |
cab30143 JQ |
1190 | } |
1191 | ||
f37afb5a | 1192 | if (local_err) { |
d59ce6f3 | 1193 | migrate_fd_error(s, local_err); |
f37afb5a | 1194 | error_propagate(errp, local_err); |
e1c37d0e | 1195 | return; |
1299c631 | 1196 | } |
cab30143 JQ |
1197 | } |
1198 | ||
6cdedb07 | 1199 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 1200 | { |
17549e84 | 1201 | migrate_fd_cancel(migrate_get_current()); |
cab30143 JQ |
1202 | } |
1203 | ||
9e1ba4cc OW |
1204 | void qmp_migrate_set_cache_size(int64_t value, Error **errp) |
1205 | { | |
1206 | MigrationState *s = migrate_get_current(); | |
c91e681a | 1207 | int64_t new_size; |
9e1ba4cc OW |
1208 | |
1209 | /* Check for truncation */ | |
1210 | if (value != (size_t)value) { | |
c6bd8c70 MA |
1211 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1212 | "exceeding address space"); | |
9e1ba4cc OW |
1213 | return; |
1214 | } | |
1215 | ||
a5615b14 OW |
1216 | /* Cache should not be larger than guest ram size */ |
1217 | if (value > ram_bytes_total()) { | |
c6bd8c70 MA |
1218 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1219 | "exceeds guest ram size "); | |
a5615b14 OW |
1220 | return; |
1221 | } | |
1222 | ||
c91e681a OW |
1223 | new_size = xbzrle_cache_resize(value); |
1224 | if (new_size < 0) { | |
c6bd8c70 MA |
1225 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1226 | "is smaller than page size"); | |
c91e681a OW |
1227 | return; |
1228 | } | |
1229 | ||
1230 | s->xbzrle_cache_size = new_size; | |
9e1ba4cc OW |
1231 | } |
1232 | ||
1233 | int64_t qmp_query_migrate_cache_size(Error **errp) | |
1234 | { | |
1235 | return migrate_xbzrle_cache_size(); | |
1236 | } | |
1237 | ||
3dc85383 | 1238 | void qmp_migrate_set_speed(int64_t value, Error **errp) |
cab30143 | 1239 | { |
2ff30257 AA |
1240 | MigrationParameters p = { |
1241 | .has_max_bandwidth = true, | |
1242 | .max_bandwidth = value, | |
1243 | }; | |
cab30143 | 1244 | |
2ff30257 | 1245 | qmp_migrate_set_parameters(&p, errp); |
cab30143 JQ |
1246 | } |
1247 | ||
4f0a993b | 1248 | void qmp_migrate_set_downtime(double value, Error **errp) |
cab30143 | 1249 | { |
87c9cc1c DHB |
1250 | if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) { |
1251 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " | |
1252 | "the range of 0 to %d seconds", | |
1253 | MAX_MIGRATE_DOWNTIME_SECONDS); | |
1254 | return; | |
1255 | } | |
1256 | ||
2ff30257 AA |
1257 | value *= 1000; /* Convert to milliseconds */ |
1258 | value = MAX(0, MIN(INT64_MAX, value)); | |
1259 | ||
1260 | MigrationParameters p = { | |
1261 | .has_downtime_limit = true, | |
1262 | .downtime_limit = value, | |
1263 | }; | |
1264 | ||
1265 | qmp_migrate_set_parameters(&p, errp); | |
99a0db9b | 1266 | } |
17ad9b35 | 1267 | |
53f09a10 PB |
1268 | bool migrate_release_ram(void) |
1269 | { | |
1270 | MigrationState *s; | |
1271 | ||
1272 | s = migrate_get_current(); | |
1273 | ||
1274 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; | |
1275 | } | |
1276 | ||
53dd370c DDAG |
1277 | bool migrate_postcopy_ram(void) |
1278 | { | |
1279 | MigrationState *s; | |
1280 | ||
1281 | s = migrate_get_current(); | |
1282 | ||
32c3db5b | 1283 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
53dd370c DDAG |
1284 | } |
1285 | ||
bde1e2ec CV |
1286 | bool migrate_auto_converge(void) |
1287 | { | |
1288 | MigrationState *s; | |
1289 | ||
1290 | s = migrate_get_current(); | |
1291 | ||
1292 | return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; | |
1293 | } | |
1294 | ||
323004a3 PL |
1295 | bool migrate_zero_blocks(void) |
1296 | { | |
1297 | MigrationState *s; | |
1298 | ||
1299 | s = migrate_get_current(); | |
1300 | ||
1301 | return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; | |
1302 | } | |
1303 | ||
8706d2d5 LL |
1304 | bool migrate_use_compression(void) |
1305 | { | |
dde4e694 LL |
1306 | MigrationState *s; |
1307 | ||
1308 | s = migrate_get_current(); | |
1309 | ||
1310 | return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; | |
8706d2d5 LL |
1311 | } |
1312 | ||
1313 | int migrate_compress_level(void) | |
1314 | { | |
1315 | MigrationState *s; | |
1316 | ||
1317 | s = migrate_get_current(); | |
1318 | ||
2594f56d | 1319 | return s->parameters.compress_level; |
8706d2d5 LL |
1320 | } |
1321 | ||
1322 | int migrate_compress_threads(void) | |
1323 | { | |
1324 | MigrationState *s; | |
1325 | ||
1326 | s = migrate_get_current(); | |
1327 | ||
2594f56d | 1328 | return s->parameters.compress_threads; |
8706d2d5 LL |
1329 | } |
1330 | ||
3fcb38c2 LL |
1331 | int migrate_decompress_threads(void) |
1332 | { | |
1333 | MigrationState *s; | |
1334 | ||
1335 | s = migrate_get_current(); | |
1336 | ||
2594f56d | 1337 | return s->parameters.decompress_threads; |
3fcb38c2 LL |
1338 | } |
1339 | ||
b05dc723 JQ |
1340 | bool migrate_use_events(void) |
1341 | { | |
1342 | MigrationState *s; | |
1343 | ||
1344 | s = migrate_get_current(); | |
1345 | ||
1346 | return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; | |
1347 | } | |
1348 | ||
17ad9b35 OW |
1349 | int migrate_use_xbzrle(void) |
1350 | { | |
1351 | MigrationState *s; | |
1352 | ||
1353 | s = migrate_get_current(); | |
1354 | ||
1355 | return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; | |
1356 | } | |
1357 | ||
1358 | int64_t migrate_xbzrle_cache_size(void) | |
1359 | { | |
1360 | MigrationState *s; | |
1361 | ||
1362 | s = migrate_get_current(); | |
1363 | ||
1364 | return s->xbzrle_cache_size; | |
1365 | } | |
0d82d0e8 | 1366 | |
2833c59b JQ |
1367 | bool migrate_use_block(void) |
1368 | { | |
1369 | MigrationState *s; | |
1370 | ||
1371 | s = migrate_get_current(); | |
1372 | ||
1373 | return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK]; | |
1374 | } | |
1375 | ||
c788ada8 PX |
1376 | bool migrate_use_return_path(void) |
1377 | { | |
1378 | MigrationState *s; | |
1379 | ||
1380 | s = migrate_get_current(); | |
1381 | ||
1382 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH]; | |
1383 | } | |
1384 | ||
2833c59b JQ |
1385 | bool migrate_use_block_incremental(void) |
1386 | { | |
1387 | MigrationState *s; | |
1388 | ||
1389 | s = migrate_get_current(); | |
1390 | ||
1391 | return s->parameters.block_incremental; | |
1392 | } | |
1393 | ||
70b20477 DDAG |
1394 | /* migration thread support */ |
1395 | /* | |
1396 | * Something bad happened to the RP stream, mark an error | |
1397 | * The caller shall print or trace something to indicate why | |
1398 | */ | |
1399 | static void mark_source_rp_bad(MigrationState *s) | |
1400 | { | |
1401 | s->rp_state.error = true; | |
1402 | } | |
1403 | ||
1404 | static struct rp_cmd_args { | |
1405 | ssize_t len; /* -1 = variable */ | |
1406 | const char *name; | |
1407 | } rp_cmd_args[] = { | |
1408 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
1409 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
1410 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
1411 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
1412 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
70b20477 DDAG |
1413 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
1414 | }; | |
1415 | ||
1e2d90eb DDAG |
1416 | /* |
1417 | * Process a request for pages received on the return path, | |
1418 | * We're allowed to send more than requested (e.g. to round to our page size) | |
1419 | * and we don't need to send pages that have already been sent. | |
1420 | */ | |
1421 | static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
1422 | ram_addr_t start, size_t len) | |
1423 | { | |
6c595cde DDAG |
1424 | long our_host_ps = getpagesize(); |
1425 | ||
1e2d90eb | 1426 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
1427 | |
1428 | /* | |
1429 | * Since we currently insist on matching page sizes, just sanity check | |
1430 | * we're being asked for whole host pages. | |
1431 | */ | |
1432 | if (start & (our_host_ps-1) || | |
1433 | (len & (our_host_ps-1))) { | |
1434 | error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT | |
1435 | " len: %zd", __func__, start, len); | |
1436 | mark_source_rp_bad(ms); | |
1437 | return; | |
1438 | } | |
1439 | ||
96506894 | 1440 | if (ram_save_queue_pages(rbname, start, len)) { |
6c595cde DDAG |
1441 | mark_source_rp_bad(ms); |
1442 | } | |
1e2d90eb DDAG |
1443 | } |
1444 | ||
70b20477 DDAG |
1445 | /* |
1446 | * Handles messages sent on the return path towards the source VM | |
1447 | * | |
1448 | */ | |
1449 | static void *source_return_path_thread(void *opaque) | |
1450 | { | |
1451 | MigrationState *ms = opaque; | |
1452 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
1453 | uint16_t header_len, header_type; | |
568b01ca | 1454 | uint8_t buf[512]; |
70b20477 | 1455 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
1456 | ram_addr_t start = 0; /* =0 to silence warning */ |
1457 | size_t len = 0, expected_len; | |
70b20477 DDAG |
1458 | int res; |
1459 | ||
1460 | trace_source_return_path_thread_entry(); | |
1461 | while (!ms->rp_state.error && !qemu_file_get_error(rp) && | |
1462 | migration_is_setup_or_active(ms->state)) { | |
1463 | trace_source_return_path_thread_loop_top(); | |
1464 | header_type = qemu_get_be16(rp); | |
1465 | header_len = qemu_get_be16(rp); | |
1466 | ||
1467 | if (header_type >= MIG_RP_MSG_MAX || | |
1468 | header_type == MIG_RP_MSG_INVALID) { | |
1469 | error_report("RP: Received invalid message 0x%04x length 0x%04x", | |
1470 | header_type, header_len); | |
1471 | mark_source_rp_bad(ms); | |
1472 | goto out; | |
1473 | } | |
1474 | ||
1475 | if ((rp_cmd_args[header_type].len != -1 && | |
1476 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 1477 | header_len > sizeof(buf)) { |
70b20477 DDAG |
1478 | error_report("RP: Received '%s' message (0x%04x) with" |
1479 | "incorrect length %d expecting %zu", | |
1480 | rp_cmd_args[header_type].name, header_type, header_len, | |
1481 | (size_t)rp_cmd_args[header_type].len); | |
1482 | mark_source_rp_bad(ms); | |
1483 | goto out; | |
1484 | } | |
1485 | ||
1486 | /* We know we've got a valid header by this point */ | |
1487 | res = qemu_get_buffer(rp, buf, header_len); | |
1488 | if (res != header_len) { | |
1489 | error_report("RP: Failed reading data for message 0x%04x" | |
1490 | " read %d expected %d", | |
1491 | header_type, res, header_len); | |
1492 | mark_source_rp_bad(ms); | |
1493 | goto out; | |
1494 | } | |
1495 | ||
1496 | /* OK, we have the message and the data */ | |
1497 | switch (header_type) { | |
1498 | case MIG_RP_MSG_SHUT: | |
4d885131 | 1499 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
1500 | trace_source_return_path_thread_shut(sibling_error); |
1501 | if (sibling_error) { | |
1502 | error_report("RP: Sibling indicated error %d", sibling_error); | |
1503 | mark_source_rp_bad(ms); | |
1504 | } | |
1505 | /* | |
1506 | * We'll let the main thread deal with closing the RP | |
1507 | * we could do a shutdown(2) on it, but we're the only user | |
1508 | * anyway, so there's nothing gained. | |
1509 | */ | |
1510 | goto out; | |
1511 | ||
1512 | case MIG_RP_MSG_PONG: | |
4d885131 | 1513 | tmp32 = ldl_be_p(buf); |
70b20477 DDAG |
1514 | trace_source_return_path_thread_pong(tmp32); |
1515 | break; | |
1516 | ||
1e2d90eb | 1517 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
1518 | start = ldq_be_p(buf); |
1519 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1520 | migrate_handle_rp_req_pages(ms, NULL, start, len); |
1521 | break; | |
1522 | ||
1523 | case MIG_RP_MSG_REQ_PAGES_ID: | |
1524 | expected_len = 12 + 1; /* header + termination */ | |
1525 | ||
1526 | if (header_len >= expected_len) { | |
4d885131 PM |
1527 | start = ldq_be_p(buf); |
1528 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1529 | /* Now we expect an idstr */ |
1530 | tmp32 = buf[12]; /* Length of the following idstr */ | |
1531 | buf[13 + tmp32] = '\0'; | |
1532 | expected_len += tmp32; | |
1533 | } | |
1534 | if (header_len != expected_len) { | |
1535 | error_report("RP: Req_Page_id with length %d expecting %zd", | |
1536 | header_len, expected_len); | |
1537 | mark_source_rp_bad(ms); | |
1538 | goto out; | |
1539 | } | |
1540 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); | |
1541 | break; | |
1542 | ||
70b20477 DDAG |
1543 | default: |
1544 | break; | |
1545 | } | |
1546 | } | |
5df5416e | 1547 | if (qemu_file_get_error(rp)) { |
70b20477 DDAG |
1548 | trace_source_return_path_thread_bad_end(); |
1549 | mark_source_rp_bad(ms); | |
1550 | } | |
1551 | ||
1552 | trace_source_return_path_thread_end(); | |
1553 | out: | |
1554 | ms->rp_state.from_dst_file = NULL; | |
1555 | qemu_fclose(rp); | |
1556 | return NULL; | |
1557 | } | |
1558 | ||
70b20477 DDAG |
1559 | static int open_return_path_on_source(MigrationState *ms) |
1560 | { | |
1561 | ||
89a02a9f | 1562 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
1563 | if (!ms->rp_state.from_dst_file) { |
1564 | return -1; | |
1565 | } | |
1566 | ||
1567 | trace_open_return_path_on_source(); | |
1568 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", | |
1569 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
1570 | ||
1571 | trace_open_return_path_on_source_continue(); | |
1572 | ||
1573 | return 0; | |
1574 | } | |
1575 | ||
70b20477 DDAG |
1576 | /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ |
1577 | static int await_return_path_close_on_source(MigrationState *ms) | |
1578 | { | |
1579 | /* | |
1580 | * If this is a normal exit then the destination will send a SHUT and the | |
1581 | * rp_thread will exit, however if there's an error we need to cause | |
1582 | * it to exit. | |
1583 | */ | |
89a02a9f | 1584 | if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { |
70b20477 DDAG |
1585 | /* |
1586 | * shutdown(2), if we have it, will cause it to unblock if it's stuck | |
1587 | * waiting for the destination. | |
1588 | */ | |
1589 | qemu_file_shutdown(ms->rp_state.from_dst_file); | |
1590 | mark_source_rp_bad(ms); | |
1591 | } | |
1592 | trace_await_return_path_close_on_source_joining(); | |
1593 | qemu_thread_join(&ms->rp_state.rp_thread); | |
1594 | trace_await_return_path_close_on_source_close(); | |
1595 | return ms->rp_state.error; | |
1596 | } | |
1597 | ||
1d34e4bf DDAG |
1598 | /* |
1599 | * Switch from normal iteration to postcopy | |
1600 | * Returns non-0 on error | |
1601 | */ | |
1602 | static int postcopy_start(MigrationState *ms, bool *old_vm_running) | |
1603 | { | |
1604 | int ret; | |
61b67d47 DB |
1605 | QIOChannelBuffer *bioc; |
1606 | QEMUFile *fb; | |
1d34e4bf | 1607 | int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
ef8d6488 | 1608 | bool restart_block = false; |
48781e5b | 1609 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, |
1d34e4bf DDAG |
1610 | MIGRATION_STATUS_POSTCOPY_ACTIVE); |
1611 | ||
1612 | trace_postcopy_start(); | |
1613 | qemu_mutex_lock_iothread(); | |
1614 | trace_postcopy_start_set_run(); | |
1615 | ||
1616 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); | |
1617 | *old_vm_running = runstate_is_running(); | |
1618 | global_state_store(); | |
1619 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
76b1c7fe KW |
1620 | if (ret < 0) { |
1621 | goto fail; | |
1622 | } | |
1d34e4bf | 1623 | |
76b1c7fe | 1624 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
1625 | if (ret < 0) { |
1626 | goto fail; | |
1627 | } | |
ef8d6488 | 1628 | restart_block = true; |
1d34e4bf | 1629 | |
1c0d249d DDAG |
1630 | /* |
1631 | * Cause any non-postcopiable, but iterative devices to | |
1632 | * send out their final data. | |
1633 | */ | |
a1fbe750 | 1634 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); |
1c0d249d | 1635 | |
1d34e4bf DDAG |
1636 | /* |
1637 | * in Finish migrate and with the io-lock held everything should | |
1638 | * be quiet, but we've potentially still got dirty pages and we | |
1639 | * need to tell the destination to throw any pages it's already received | |
1640 | * that are dirty | |
1641 | */ | |
1642 | if (ram_postcopy_send_discard_bitmap(ms)) { | |
1643 | error_report("postcopy send discard bitmap failed"); | |
1644 | goto fail; | |
1645 | } | |
1646 | ||
1647 | /* | |
1648 | * send rest of state - note things that are doing postcopy | |
1649 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
1650 | * wrap their state up here | |
1651 | */ | |
89a02a9f | 1652 | qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); |
1d34e4bf | 1653 | /* Ping just for debugging, helps line traces up */ |
89a02a9f | 1654 | qemu_savevm_send_ping(ms->to_dst_file, 2); |
1d34e4bf DDAG |
1655 | |
1656 | /* | |
1657 | * While loading the device state we may trigger page transfer | |
1658 | * requests and the fd must be free to process those, and thus | |
1659 | * the destination must read the whole device state off the fd before | |
1660 | * it starts processing it. Unfortunately the ad-hoc migration format | |
1661 | * doesn't allow the destination to know the size to read without fully | |
1662 | * parsing it through each devices load-state code (especially the open | |
1663 | * coded devices that use get/put). | |
1664 | * So we wrap the device state up in a package with a length at the start; | |
1665 | * to do this we use a qemu_buf to hold the whole of the device state. | |
1666 | */ | |
61b67d47 | 1667 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 1668 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
61b67d47 DB |
1669 | fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc)); |
1670 | object_unref(OBJECT(bioc)); | |
1d34e4bf | 1671 | |
c76201ab DDAG |
1672 | /* |
1673 | * Make sure the receiver can get incoming pages before we send the rest | |
1674 | * of the state | |
1675 | */ | |
1676 | qemu_savevm_send_postcopy_listen(fb); | |
1677 | ||
a1fbe750 | 1678 | qemu_savevm_state_complete_precopy(fb, false, false); |
1d34e4bf DDAG |
1679 | qemu_savevm_send_ping(fb, 3); |
1680 | ||
1681 | qemu_savevm_send_postcopy_run(fb); | |
1682 | ||
1683 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 1684 | |
ef8d6488 DDAG |
1685 | /* Last point of recovery; as soon as we send the package the destination |
1686 | * can open devices and potentially start running. | |
1687 | * Lets just check again we've not got any errors. | |
1688 | */ | |
1689 | ret = qemu_file_get_error(ms->to_dst_file); | |
1690 | if (ret) { | |
1691 | error_report("postcopy_start: Migration stream errored (pre package)"); | |
1692 | goto fail_closefb; | |
1693 | } | |
1694 | ||
1695 | restart_block = false; | |
1696 | ||
1d34e4bf | 1697 | /* Now send that blob */ |
61b67d47 | 1698 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
1699 | goto fail_closefb; |
1700 | } | |
1701 | qemu_fclose(fb); | |
b82fc321 DDAG |
1702 | |
1703 | /* Send a notify to give a chance for anything that needs to happen | |
1704 | * at the transition to postcopy and after the device state; in particular | |
1705 | * spice needs to trigger a transition now | |
1706 | */ | |
1707 | ms->postcopy_after_devices = true; | |
1708 | notifier_list_notify(&migration_state_notifiers, ms); | |
1709 | ||
1d34e4bf DDAG |
1710 | ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; |
1711 | ||
1712 | qemu_mutex_unlock_iothread(); | |
1713 | ||
1714 | /* | |
1715 | * Although this ping is just for debug, it could potentially be | |
1716 | * used for getting a better measurement of downtime at the source. | |
1717 | */ | |
89a02a9f | 1718 | qemu_savevm_send_ping(ms->to_dst_file, 4); |
1d34e4bf | 1719 | |
ced1c616 PB |
1720 | if (migrate_release_ram()) { |
1721 | ram_postcopy_migrated_memory_release(ms); | |
1722 | } | |
1723 | ||
89a02a9f | 1724 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf DDAG |
1725 | if (ret) { |
1726 | error_report("postcopy_start: Migration stream errored"); | |
48781e5b | 1727 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf DDAG |
1728 | MIGRATION_STATUS_FAILED); |
1729 | } | |
1730 | ||
1731 | return ret; | |
1732 | ||
1733 | fail_closefb: | |
1734 | qemu_fclose(fb); | |
1735 | fail: | |
48781e5b | 1736 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 1737 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
1738 | if (restart_block) { |
1739 | /* A failure happened early enough that we know the destination hasn't | |
1740 | * accessed block devices, so we're safe to recover. | |
1741 | */ | |
1742 | Error *local_err = NULL; | |
1743 | ||
1744 | bdrv_invalidate_cache_all(&local_err); | |
1745 | if (local_err) { | |
1746 | error_report_err(local_err); | |
1747 | } | |
1748 | } | |
1d34e4bf DDAG |
1749 | qemu_mutex_unlock_iothread(); |
1750 | return -1; | |
1751 | } | |
1752 | ||
09f6c85e DDAG |
1753 | /** |
1754 | * migration_completion: Used by migration_thread when there's not much left. | |
1755 | * The caller 'breaks' the loop when this returns. | |
1756 | * | |
1757 | * @s: Current migration state | |
36f48567 | 1758 | * @current_active_state: The migration state we expect to be in |
09f6c85e DDAG |
1759 | * @*old_vm_running: Pointer to old_vm_running flag |
1760 | * @*start_time: Pointer to time to update | |
1761 | */ | |
36f48567 DDAG |
1762 | static void migration_completion(MigrationState *s, int current_active_state, |
1763 | bool *old_vm_running, | |
09f6c85e DDAG |
1764 | int64_t *start_time) |
1765 | { | |
1766 | int ret; | |
1767 | ||
b10ac0c4 DDAG |
1768 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
1769 | qemu_mutex_lock_iothread(); | |
1770 | *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
1771 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); | |
1772 | *old_vm_running = runstate_is_running(); | |
1773 | ret = global_state_store(); | |
1774 | ||
1775 | if (!ret) { | |
a1fbe750 | 1776 | bool inactivate = !migrate_colo_enabled(); |
b10ac0c4 | 1777 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); |
f07fa4cb KW |
1778 | if (ret >= 0) { |
1779 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); | |
a1fbe750 FZ |
1780 | ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, |
1781 | inactivate); | |
f07fa4cb | 1782 | } |
a1fbe750 FZ |
1783 | if (inactivate && ret >= 0) { |
1784 | s->block_inactive = true; | |
b10ac0c4 DDAG |
1785 | } |
1786 | } | |
1787 | qemu_mutex_unlock_iothread(); | |
09f6c85e | 1788 | |
b10ac0c4 DDAG |
1789 | if (ret < 0) { |
1790 | goto fail; | |
09f6c85e | 1791 | } |
b10ac0c4 DDAG |
1792 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { |
1793 | trace_migration_completion_postcopy_end(); | |
1794 | ||
89a02a9f | 1795 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
b10ac0c4 | 1796 | trace_migration_completion_postcopy_end_after_complete(); |
09f6c85e | 1797 | } |
09f6c85e | 1798 | |
b10ac0c4 DDAG |
1799 | /* |
1800 | * If rp was opened we must clean up the thread before | |
1801 | * cleaning everything else up (since if there are no failures | |
1802 | * it will wait for the destination to send it's status in | |
1803 | * a SHUT command). | |
b10ac0c4 | 1804 | */ |
0425dc97 | 1805 | if (s->rp_state.from_dst_file) { |
b10ac0c4 | 1806 | int rp_error; |
0425dc97 | 1807 | trace_migration_return_path_end_before(); |
b10ac0c4 | 1808 | rp_error = await_return_path_close_on_source(s); |
0425dc97 | 1809 | trace_migration_return_path_end_after(rp_error); |
b10ac0c4 | 1810 | if (rp_error) { |
fe904ea8 | 1811 | goto fail_invalidate; |
b10ac0c4 | 1812 | } |
09f6c85e DDAG |
1813 | } |
1814 | ||
89a02a9f | 1815 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 1816 | trace_migration_completion_file_err(); |
fe904ea8 | 1817 | goto fail_invalidate; |
09f6c85e DDAG |
1818 | } |
1819 | ||
0b827d5e HZ |
1820 | if (!migrate_colo_enabled()) { |
1821 | migrate_set_state(&s->state, current_active_state, | |
1822 | MIGRATION_STATUS_COMPLETED); | |
1823 | } | |
1824 | ||
09f6c85e DDAG |
1825 | return; |
1826 | ||
fe904ea8 GK |
1827 | fail_invalidate: |
1828 | /* If not doing postcopy, vm_start() will be called: let's regain | |
1829 | * control on images. | |
1830 | */ | |
1831 | if (s->state == MIGRATION_STATUS_ACTIVE) { | |
1832 | Error *local_err = NULL; | |
1833 | ||
1d2acc31 | 1834 | qemu_mutex_lock_iothread(); |
fe904ea8 GK |
1835 | bdrv_invalidate_cache_all(&local_err); |
1836 | if (local_err) { | |
1837 | error_report_err(local_err); | |
1d2acc31 HZ |
1838 | } else { |
1839 | s->block_inactive = false; | |
fe904ea8 | 1840 | } |
1d2acc31 | 1841 | qemu_mutex_unlock_iothread(); |
fe904ea8 GK |
1842 | } |
1843 | ||
09f6c85e | 1844 | fail: |
48781e5b HZ |
1845 | migrate_set_state(&s->state, current_active_state, |
1846 | MIGRATION_STATUS_FAILED); | |
09f6c85e DDAG |
1847 | } |
1848 | ||
35a6ed4f HZ |
1849 | bool migrate_colo_enabled(void) |
1850 | { | |
1851 | MigrationState *s = migrate_get_current(); | |
1852 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; | |
1853 | } | |
1854 | ||
70b20477 DDAG |
1855 | /* |
1856 | * Master migration thread on the source VM. | |
1857 | * It drives the migration and pumps the data down the outgoing channel. | |
1858 | */ | |
5f496a1b | 1859 | static void *migration_thread(void *opaque) |
0d82d0e8 | 1860 | { |
9848a404 | 1861 | MigrationState *s = opaque; |
1d34e4bf | 1862 | /* Used by the bandwidth calcs, updated later */ |
bc72ad67 AB |
1863 | int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1864 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
be7172e2 | 1865 | int64_t initial_bytes = 0; |
faec066a PX |
1866 | /* |
1867 | * The final stage happens when the remaining data is smaller than | |
1868 | * this threshold; it's calculated from the requested downtime and | |
1869 | * measured bandwidth | |
1870 | */ | |
1871 | int64_t threshold_size = 0; | |
a3fa1d78 | 1872 | int64_t start_time = initial_time; |
94f5a437 | 1873 | int64_t end_time; |
a3fa1d78 | 1874 | bool old_vm_running = false; |
1d34e4bf DDAG |
1875 | bool entered_postcopy = false; |
1876 | /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */ | |
1877 | enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE; | |
0b827d5e | 1878 | bool enable_colo = migrate_colo_enabled(); |
76f5933a | 1879 | |
ab28bd23 PB |
1880 | rcu_register_thread(); |
1881 | ||
89a02a9f | 1882 | qemu_savevm_state_header(s->to_dst_file); |
1d34e4bf | 1883 | |
62a02658 PX |
1884 | /* |
1885 | * If we opened the return path, we need to make sure dst has it | |
1886 | * opened as well. | |
1887 | */ | |
1888 | if (s->rp_state.from_dst_file) { | |
1d34e4bf | 1889 | /* Now tell the dest that it should open its end so it can reply */ |
89a02a9f | 1890 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
1891 | |
1892 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 1893 | qemu_savevm_send_ping(s->to_dst_file, 1); |
0425dc97 | 1894 | } |
1d34e4bf | 1895 | |
0425dc97 | 1896 | if (migrate_postcopy_ram()) { |
1d34e4bf DDAG |
1897 | /* |
1898 | * Tell the destination that we *might* want to do postcopy later; | |
1899 | * if the other end can't do postcopy it should fail now, nice and | |
1900 | * early. | |
1901 | */ | |
89a02a9f | 1902 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
1903 | } |
1904 | ||
9907e842 | 1905 | qemu_savevm_state_setup(s->to_dst_file); |
0d82d0e8 | 1906 | |
bc72ad67 | 1907 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
48781e5b HZ |
1908 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1909 | MIGRATION_STATUS_ACTIVE); | |
29ae8a41 | 1910 | |
9ec055ae DDAG |
1911 | trace_migration_thread_setup_complete(); |
1912 | ||
1913 | while (s->state == MIGRATION_STATUS_ACTIVE || | |
1914 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
a3e879cd | 1915 | int64_t current_time; |
c369f40d | 1916 | uint64_t pending_size; |
0d82d0e8 | 1917 | |
89a02a9f | 1918 | if (!qemu_file_rate_limit(s->to_dst_file)) { |
c31b098f DDAG |
1919 | uint64_t pend_post, pend_nonpost; |
1920 | ||
faec066a PX |
1921 | qemu_savevm_state_pending(s->to_dst_file, threshold_size, |
1922 | &pend_nonpost, &pend_post); | |
c31b098f | 1923 | pending_size = pend_nonpost + pend_post; |
faec066a | 1924 | trace_migrate_pending(pending_size, threshold_size, |
c31b098f | 1925 | pend_post, pend_nonpost); |
faec066a | 1926 | if (pending_size && pending_size >= threshold_size) { |
1d34e4bf DDAG |
1927 | /* Still a significant amount to transfer */ |
1928 | ||
1d34e4bf DDAG |
1929 | if (migrate_postcopy_ram() && |
1930 | s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE && | |
faec066a | 1931 | pend_nonpost <= threshold_size && |
1d34e4bf DDAG |
1932 | atomic_read(&s->start_postcopy)) { |
1933 | ||
1934 | if (!postcopy_start(s, &old_vm_running)) { | |
1935 | current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE; | |
1936 | entered_postcopy = true; | |
1937 | } | |
1938 | ||
1939 | continue; | |
1940 | } | |
1941 | /* Just another iteration step */ | |
89a02a9f | 1942 | qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy); |
c369f40d | 1943 | } else { |
09f6c85e | 1944 | trace_migration_thread_low_pending(pending_size); |
1d34e4bf | 1945 | migration_completion(s, current_active_state, |
36f48567 | 1946 | &old_vm_running, &start_time); |
09f6c85e | 1947 | break; |
c369f40d JQ |
1948 | } |
1949 | } | |
f4410a5d | 1950 | |
89a02a9f | 1951 | if (qemu_file_get_error(s->to_dst_file)) { |
48781e5b HZ |
1952 | migrate_set_state(&s->state, current_active_state, |
1953 | MIGRATION_STATUS_FAILED); | |
1d34e4bf | 1954 | trace_migration_thread_file_err(); |
fd45ee2c PB |
1955 | break; |
1956 | } | |
bc72ad67 | 1957 | current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0d82d0e8 | 1958 | if (current_time >= initial_time + BUFFER_DELAY) { |
89a02a9f HZ |
1959 | uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) - |
1960 | initial_bytes; | |
77417f10 | 1961 | uint64_t time_spent = current_time - initial_time; |
a694ee34 | 1962 | double bandwidth = (double)transferred_bytes / time_spent; |
faec066a | 1963 | threshold_size = bandwidth * s->parameters.downtime_limit; |
0d82d0e8 | 1964 | |
5b648de0 WY |
1965 | s->mbps = (((double) transferred_bytes * 8.0) / |
1966 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
7e114f8c | 1967 | |
9013dca5 | 1968 | trace_migrate_transferred(transferred_bytes, time_spent, |
faec066a | 1969 | bandwidth, threshold_size); |
90f8ae72 JQ |
1970 | /* if we haven't sent anything, we don't want to recalculate |
1971 | 10000 is a small enough number for our purposes */ | |
9360447d JQ |
1972 | if (ram_counters.dirty_pages_rate && transferred_bytes > 10000) { |
1973 | s->expected_downtime = ram_counters.dirty_pages_rate * | |
20afaed9 | 1974 | qemu_target_page_size() / bandwidth; |
90f8ae72 | 1975 | } |
0d82d0e8 | 1976 | |
89a02a9f | 1977 | qemu_file_reset_rate_limit(s->to_dst_file); |
0d82d0e8 | 1978 | initial_time = current_time; |
89a02a9f | 1979 | initial_bytes = qemu_ftell(s->to_dst_file); |
0d82d0e8 | 1980 | } |
89a02a9f | 1981 | if (qemu_file_rate_limit(s->to_dst_file)) { |
0d82d0e8 JQ |
1982 | /* usleep expects microseconds */ |
1983 | g_usleep((initial_time + BUFFER_DELAY - current_time)*1000); | |
1984 | } | |
a3fa1d78 PB |
1985 | } |
1986 | ||
1d34e4bf | 1987 | trace_migration_thread_after_loop(); |
070afca2 JH |
1988 | /* If we enabled cpu throttling for auto-converge, turn it off. */ |
1989 | cpu_throttle_stop(); | |
94f5a437 | 1990 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
070afca2 | 1991 | |
f4410a5d | 1992 | qemu_mutex_lock_iothread(); |
0b827d5e HZ |
1993 | /* |
1994 | * The resource has been allocated by migration will be reused in COLO | |
1995 | * process, so don't release them. | |
1996 | */ | |
1997 | if (!enable_colo) { | |
1998 | qemu_savevm_state_cleanup(); | |
1999 | } | |
31194731 | 2000 | if (s->state == MIGRATION_STATUS_COMPLETED) { |
89a02a9f | 2001 | uint64_t transferred_bytes = qemu_ftell(s->to_dst_file); |
a3fa1d78 | 2002 | s->total_time = end_time - s->total_time; |
1d34e4bf DDAG |
2003 | if (!entered_postcopy) { |
2004 | s->downtime = end_time - start_time; | |
2005 | } | |
d6ed7312 PL |
2006 | if (s->total_time) { |
2007 | s->mbps = (((double) transferred_bytes * 8.0) / | |
2008 | ((double) s->total_time)) / 1000; | |
2009 | } | |
a3fa1d78 PB |
2010 | runstate_set(RUN_STATE_POSTMIGRATE); |
2011 | } else { | |
0b827d5e HZ |
2012 | if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) { |
2013 | migrate_start_colo_process(s); | |
2014 | qemu_savevm_state_cleanup(); | |
2015 | /* | |
2016 | * Fixme: we will run VM in COLO no matter its old running state. | |
2017 | * After exited COLO, we will keep running. | |
2018 | */ | |
2019 | old_vm_running = true; | |
2020 | } | |
1d34e4bf | 2021 | if (old_vm_running && !entered_postcopy) { |
a3fa1d78 | 2022 | vm_start(); |
42da5550 DDAG |
2023 | } else { |
2024 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
2025 | runstate_set(RUN_STATE_POSTMIGRATE); | |
2026 | } | |
dba433c0 | 2027 | } |
0d82d0e8 | 2028 | } |
bb1fadc4 | 2029 | qemu_bh_schedule(s->cleanup_bh); |
dba433c0 | 2030 | qemu_mutex_unlock_iothread(); |
f4410a5d | 2031 | |
ab28bd23 | 2032 | rcu_unregister_thread(); |
0d82d0e8 JQ |
2033 | return NULL; |
2034 | } | |
2035 | ||
9848a404 | 2036 | void migrate_fd_connect(MigrationState *s) |
0d82d0e8 | 2037 | { |
2ff30257 | 2038 | s->expected_downtime = s->parameters.downtime_limit; |
bb1fadc4 | 2039 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s); |
0d82d0e8 | 2040 | |
9e4d2b98 | 2041 | qemu_file_set_blocking(s->to_dst_file, true); |
89a02a9f | 2042 | qemu_file_set_rate_limit(s->to_dst_file, |
2ff30257 | 2043 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); |
442773ce | 2044 | |
9287ac27 SH |
2045 | /* Notify before starting migration thread */ |
2046 | notifier_list_notify(&migration_state_notifiers, s); | |
2047 | ||
1d34e4bf | 2048 | /* |
c788ada8 PX |
2049 | * Open the return path. For postcopy, it is used exclusively. For |
2050 | * precopy, only if user specified "return-path" capability would | |
2051 | * QEMU uses the return path. | |
1d34e4bf | 2052 | */ |
c788ada8 | 2053 | if (migrate_postcopy_ram() || migrate_use_return_path()) { |
1d34e4bf DDAG |
2054 | if (open_return_path_on_source(s)) { |
2055 | error_report("Unable to open return-path for postcopy"); | |
48781e5b | 2056 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1d34e4bf DDAG |
2057 | MIGRATION_STATUS_FAILED); |
2058 | migrate_fd_cleanup(s); | |
2059 | return; | |
2060 | } | |
2061 | } | |
2062 | ||
009fad7f | 2063 | qemu_thread_create(&s->thread, "live_migration", migration_thread, s, |
bb1fadc4 | 2064 | QEMU_THREAD_JOINABLE); |
1d34e4bf | 2065 | s->migration_thread_running = true; |
0d82d0e8 | 2066 | } |
093e3c42 | 2067 | |
9d18af93 PX |
2068 | void migration_global_dump(Monitor *mon) |
2069 | { | |
2070 | MigrationState *ms = migrate_get_current(); | |
2071 | ||
2072 | monitor_printf(mon, "globals: store-global-state=%d, only_migratable=%d, " | |
2073 | "send-configuration=%d, send-section-footer=%d\n", | |
2074 | ms->store_global_state, ms->only_migratable, | |
2075 | ms->send_configuration, ms->send_section_footer); | |
2076 | } | |
2077 | ||
20814758 PX |
2078 | #define DEFINE_PROP_MIG_CAP(name, x) \ |
2079 | DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false) | |
2080 | ||
5272298c PX |
2081 | static Property migration_properties[] = { |
2082 | DEFINE_PROP_BOOL("store-global-state", MigrationState, | |
2083 | store_global_state, true), | |
3df663e5 | 2084 | DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false), |
71dd4c1a PX |
2085 | DEFINE_PROP_BOOL("send-configuration", MigrationState, |
2086 | send_configuration, true), | |
15c38503 PX |
2087 | DEFINE_PROP_BOOL("send-section-footer", MigrationState, |
2088 | send_section_footer, true), | |
89632faf PX |
2089 | |
2090 | /* Migration parameters */ | |
2091 | DEFINE_PROP_INT64("x-compress-level", MigrationState, | |
2092 | parameters.compress_level, | |
2093 | DEFAULT_MIGRATE_COMPRESS_LEVEL), | |
2094 | DEFINE_PROP_INT64("x-compress-threads", MigrationState, | |
2095 | parameters.compress_threads, | |
2096 | DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT), | |
2097 | DEFINE_PROP_INT64("x-decompress-threads", MigrationState, | |
2098 | parameters.decompress_threads, | |
2099 | DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT), | |
2100 | DEFINE_PROP_INT64("x-cpu-throttle-initial", MigrationState, | |
2101 | parameters.cpu_throttle_initial, | |
2102 | DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL), | |
2103 | DEFINE_PROP_INT64("x-cpu-throttle-increment", MigrationState, | |
2104 | parameters.cpu_throttle_increment, | |
2105 | DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT), | |
2106 | DEFINE_PROP_INT64("x-max-bandwidth", MigrationState, | |
2107 | parameters.max_bandwidth, MAX_THROTTLE), | |
2108 | DEFINE_PROP_INT64("x-downtime-limit", MigrationState, | |
2109 | parameters.downtime_limit, | |
2110 | DEFAULT_MIGRATE_SET_DOWNTIME), | |
2111 | DEFINE_PROP_INT64("x-checkpoint-delay", MigrationState, | |
2112 | parameters.x_checkpoint_delay, | |
2113 | DEFAULT_MIGRATE_X_CHECKPOINT_DELAY), | |
20814758 PX |
2114 | |
2115 | /* Migration capabilities */ | |
2116 | DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE), | |
2117 | DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL), | |
2118 | DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE), | |
2119 | DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS), | |
2120 | DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS), | |
2121 | DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS), | |
2122 | DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM), | |
2123 | DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO), | |
2124 | DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM), | |
2125 | DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK), | |
2126 | DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH), | |
2127 | ||
5272298c PX |
2128 | DEFINE_PROP_END_OF_LIST(), |
2129 | }; | |
2130 | ||
e5cb7e76 PX |
2131 | static void migration_class_init(ObjectClass *klass, void *data) |
2132 | { | |
2133 | DeviceClass *dc = DEVICE_CLASS(klass); | |
2134 | ||
2135 | dc->user_creatable = false; | |
5272298c | 2136 | dc->props = migration_properties; |
e5cb7e76 PX |
2137 | } |
2138 | ||
2139 | static void migration_instance_init(Object *obj) | |
2140 | { | |
2141 | MigrationState *ms = MIGRATION_OBJ(obj); | |
8b0b29dc | 2142 | MigrationParameters *params = &ms->parameters; |
e5cb7e76 PX |
2143 | |
2144 | ms->state = MIGRATION_STATUS_NONE; | |
2145 | ms->xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE; | |
2146 | ms->mbps = -1; | |
8b0b29dc PX |
2147 | |
2148 | params->tls_hostname = g_strdup(""); | |
2149 | params->tls_creds = g_strdup(""); | |
2150 | ||
2151 | /* Set has_* up only for parameter checks */ | |
2152 | params->has_compress_level = true; | |
2153 | params->has_compress_threads = true; | |
2154 | params->has_decompress_threads = true; | |
2155 | params->has_cpu_throttle_initial = true; | |
2156 | params->has_cpu_throttle_increment = true; | |
2157 | params->has_max_bandwidth = true; | |
2158 | params->has_downtime_limit = true; | |
2159 | params->has_x_checkpoint_delay = true; | |
2160 | params->has_block_incremental = true; | |
2161 | } | |
2162 | ||
2163 | /* | |
2164 | * Return true if check pass, false otherwise. Error will be put | |
2165 | * inside errp if provided. | |
2166 | */ | |
2167 | static bool migration_object_check(MigrationState *ms, Error **errp) | |
2168 | { | |
2169 | if (!migrate_params_check(&ms->parameters, errp)) { | |
2170 | return false; | |
2171 | } | |
2172 | ||
2173 | return true; | |
e5cb7e76 PX |
2174 | } |
2175 | ||
2176 | static const TypeInfo migration_type = { | |
2177 | .name = TYPE_MIGRATION, | |
01f6e14c | 2178 | /* |
c8d3ff38 PX |
2179 | * NOTE: TYPE_MIGRATION is not really a device, as the object is |
2180 | * not created using qdev_create(), it is not attached to the qdev | |
2181 | * device tree, and it is never realized. | |
2182 | * | |
2183 | * TODO: Make this TYPE_OBJECT once QOM provides something like | |
2184 | * TYPE_DEVICE's "-global" properties. | |
01f6e14c | 2185 | */ |
e5cb7e76 PX |
2186 | .parent = TYPE_DEVICE, |
2187 | .class_init = migration_class_init, | |
2188 | .class_size = sizeof(MigrationClass), | |
2189 | .instance_size = sizeof(MigrationState), | |
2190 | .instance_init = migration_instance_init, | |
2191 | }; | |
2192 | ||
2193 | static void register_migration_types(void) | |
2194 | { | |
2195 | type_register_static(&migration_type); | |
2196 | } | |
2197 | ||
2198 | type_init(register_migration_types); |