]>
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" |
6a1751b7 | 19 | #include "qemu/main-loop.h" |
caf71f86 | 20 | #include "migration/migration.h" |
0d82d0e8 | 21 | #include "migration/qemu-file.h" |
9c17d615 | 22 | #include "sysemu/sysemu.h" |
737e150e | 23 | #include "block/block.h" |
cc7a8ea7 | 24 | #include "qapi/qmp/qerror.h" |
6c595cde | 25 | #include "qapi/util.h" |
1de7afc9 | 26 | #include "qemu/sockets.h" |
ab28bd23 | 27 | #include "qemu/rcu.h" |
caf71f86 | 28 | #include "migration/block.h" |
e0b266f0 | 29 | #include "migration/postcopy-ram.h" |
766bd176 | 30 | #include "qemu/thread.h" |
791e7c82 | 31 | #include "qmp-commands.h" |
c09e5bb1 | 32 | #include "trace.h" |
598cd2bd | 33 | #include "qapi-event.h" |
070afca2 | 34 | #include "qom/cpu.h" |
6c595cde DDAG |
35 | #include "exec/memory.h" |
36 | #include "exec/address-spaces.h" | |
61b67d47 | 37 | #include "io/channel-buffer.h" |
e1226365 | 38 | #include "io/channel-tls.h" |
35a6ed4f | 39 | #include "migration/colo.h" |
065e2813 | 40 | |
dc325627 | 41 | #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */ |
5bb7910a | 42 | |
5b4e1eb7 JQ |
43 | /* Amount of time to allocate to each "chunk" of bandwidth-throttled |
44 | * data. */ | |
45 | #define BUFFER_DELAY 100 | |
46 | #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) | |
47 | ||
2ff30257 AA |
48 | /* Time in milliseconds we are allowed to stop the source, |
49 | * for sending the last part */ | |
50 | #define DEFAULT_MIGRATE_SET_DOWNTIME 300 | |
51 | ||
87c9cc1c DHB |
52 | /* Maximum migrate downtime set to 2000 seconds */ |
53 | #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 | |
54 | #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) | |
55 | ||
8706d2d5 LL |
56 | /* Default compression thread count */ |
57 | #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 | |
3fcb38c2 LL |
58 | /* Default decompression thread count, usually decompression is at |
59 | * least 4 times as fast as compression.*/ | |
60 | #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 | |
8706d2d5 LL |
61 | /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
62 | #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 | |
1626fee3 | 63 | /* Define default autoconverge cpu throttle migration parameters */ |
d85a31d1 JH |
64 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 |
65 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 | |
8706d2d5 | 66 | |
17ad9b35 OW |
67 | /* Migration XBZRLE default cache size */ |
68 | #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024) | |
69 | ||
68b53591 HZ |
70 | /* The delay time (in ms) between two COLO checkpoints |
71 | * Note: Please change this default value to 10000 when we support hybrid mode. | |
72 | */ | |
73 | #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200 | |
74 | ||
99a0db9b GH |
75 | static NotifierList migration_state_notifiers = |
76 | NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); | |
77 | ||
adde220a DDAG |
78 | static bool deferred_incoming; |
79 | ||
093e3c42 DDAG |
80 | /* |
81 | * Current state of incoming postcopy; note this is not part of | |
82 | * MigrationIncomingState since it's state is used during cleanup | |
83 | * at the end as MIS is being freed. | |
84 | */ | |
85 | static PostcopyState incoming_postcopy_state; | |
86 | ||
17549e84 JQ |
87 | /* When we add fault tolerance, we could have several |
88 | migrations at once. For now we don't need to add | |
89 | dynamic creation of migration */ | |
90 | ||
bca7856a | 91 | /* For outgoing */ |
859bc756 | 92 | MigrationState *migrate_get_current(void) |
17549e84 | 93 | { |
6c595cde | 94 | static bool once; |
17549e84 | 95 | static MigrationState current_migration = { |
31194731 | 96 | .state = MIGRATION_STATUS_NONE, |
17ad9b35 | 97 | .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE, |
7e114f8c | 98 | .mbps = -1, |
2594f56d DB |
99 | .parameters = { |
100 | .compress_level = DEFAULT_MIGRATE_COMPRESS_LEVEL, | |
101 | .compress_threads = DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT, | |
102 | .decompress_threads = DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT, | |
103 | .cpu_throttle_initial = DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL, | |
104 | .cpu_throttle_increment = DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT, | |
2ff30257 AA |
105 | .max_bandwidth = MAX_THROTTLE, |
106 | .downtime_limit = DEFAULT_MIGRATE_SET_DOWNTIME, | |
68b53591 | 107 | .x_checkpoint_delay = DEFAULT_MIGRATE_X_CHECKPOINT_DELAY, |
2594f56d | 108 | }, |
17549e84 JQ |
109 | }; |
110 | ||
6c595cde DDAG |
111 | if (!once) { |
112 | qemu_mutex_init(¤t_migration.src_page_req_mutex); | |
4af245dc DB |
113 | current_migration.parameters.tls_creds = g_strdup(""); |
114 | current_migration.parameters.tls_hostname = g_strdup(""); | |
6c595cde DDAG |
115 | once = true; |
116 | } | |
17549e84 JQ |
117 | return ¤t_migration; |
118 | } | |
119 | ||
bca7856a DDAG |
120 | MigrationIncomingState *migration_incoming_get_current(void) |
121 | { | |
b4b076da JQ |
122 | static bool once; |
123 | static MigrationIncomingState mis_current; | |
bca7856a | 124 | |
b4b076da JQ |
125 | if (!once) { |
126 | mis_current.state = MIGRATION_STATUS_NONE; | |
127 | memset(&mis_current, 0, sizeof(MigrationIncomingState)); | |
128 | QLIST_INIT(&mis_current.loadvm_handlers); | |
129 | qemu_mutex_init(&mis_current.rp_mutex); | |
130 | qemu_event_init(&mis_current.main_thread_load_event, false); | |
131 | once = true; | |
132 | } | |
133 | return &mis_current; | |
bca7856a DDAG |
134 | } |
135 | ||
136 | void migration_incoming_state_destroy(void) | |
137 | { | |
b4b076da JQ |
138 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
139 | ||
140 | qemu_event_destroy(&mis->main_thread_load_event); | |
141 | loadvm_free_handlers(mis); | |
bca7856a DDAG |
142 | } |
143 | ||
df4b1024 JQ |
144 | |
145 | typedef struct { | |
13d16814 | 146 | bool optional; |
df4b1024 JQ |
147 | uint32_t size; |
148 | uint8_t runstate[100]; | |
172c4356 JQ |
149 | RunState state; |
150 | bool received; | |
df4b1024 JQ |
151 | } GlobalState; |
152 | ||
153 | static GlobalState global_state; | |
154 | ||
560d027b | 155 | int global_state_store(void) |
df4b1024 JQ |
156 | { |
157 | if (!runstate_store((char *)global_state.runstate, | |
158 | sizeof(global_state.runstate))) { | |
159 | error_report("runstate name too big: %s", global_state.runstate); | |
160 | trace_migrate_state_too_big(); | |
161 | return -EINVAL; | |
162 | } | |
163 | return 0; | |
164 | } | |
165 | ||
c69adea4 AP |
166 | void global_state_store_running(void) |
167 | { | |
168 | const char *state = RunState_lookup[RUN_STATE_RUNNING]; | |
169 | strncpy((char *)global_state.runstate, | |
170 | state, sizeof(global_state.runstate)); | |
171 | } | |
172 | ||
172c4356 | 173 | static bool global_state_received(void) |
df4b1024 | 174 | { |
172c4356 JQ |
175 | return global_state.received; |
176 | } | |
177 | ||
178 | static RunState global_state_get_runstate(void) | |
179 | { | |
180 | return global_state.state; | |
df4b1024 JQ |
181 | } |
182 | ||
13d16814 JQ |
183 | void global_state_set_optional(void) |
184 | { | |
185 | global_state.optional = true; | |
186 | } | |
187 | ||
188 | static bool global_state_needed(void *opaque) | |
189 | { | |
190 | GlobalState *s = opaque; | |
191 | char *runstate = (char *)s->runstate; | |
192 | ||
193 | /* If it is not optional, it is mandatory */ | |
194 | ||
195 | if (s->optional == false) { | |
196 | return true; | |
197 | } | |
198 | ||
199 | /* If state is running or paused, it is not needed */ | |
200 | ||
201 | if (strcmp(runstate, "running") == 0 || | |
202 | strcmp(runstate, "paused") == 0) { | |
203 | return false; | |
204 | } | |
205 | ||
206 | /* for any other state it is needed */ | |
207 | return true; | |
208 | } | |
209 | ||
df4b1024 JQ |
210 | static int global_state_post_load(void *opaque, int version_id) |
211 | { | |
212 | GlobalState *s = opaque; | |
172c4356 JQ |
213 | Error *local_err = NULL; |
214 | int r; | |
df4b1024 JQ |
215 | char *runstate = (char *)s->runstate; |
216 | ||
172c4356 | 217 | s->received = true; |
df4b1024 JQ |
218 | trace_migrate_global_state_post_load(runstate); |
219 | ||
7fb1cf16 | 220 | r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE__MAX, |
df4b1024 JQ |
221 | -1, &local_err); |
222 | ||
172c4356 JQ |
223 | if (r == -1) { |
224 | if (local_err) { | |
225 | error_report_err(local_err); | |
df4b1024 | 226 | } |
172c4356 | 227 | return -EINVAL; |
df4b1024 | 228 | } |
172c4356 | 229 | s->state = r; |
df4b1024 | 230 | |
172c4356 | 231 | return 0; |
df4b1024 JQ |
232 | } |
233 | ||
234 | static void global_state_pre_save(void *opaque) | |
235 | { | |
236 | GlobalState *s = opaque; | |
237 | ||
238 | trace_migrate_global_state_pre_save((char *)s->runstate); | |
239 | s->size = strlen((char *)s->runstate) + 1; | |
240 | } | |
241 | ||
242 | static const VMStateDescription vmstate_globalstate = { | |
243 | .name = "globalstate", | |
244 | .version_id = 1, | |
245 | .minimum_version_id = 1, | |
246 | .post_load = global_state_post_load, | |
247 | .pre_save = global_state_pre_save, | |
13d16814 | 248 | .needed = global_state_needed, |
df4b1024 JQ |
249 | .fields = (VMStateField[]) { |
250 | VMSTATE_UINT32(size, GlobalState), | |
251 | VMSTATE_BUFFER(runstate, GlobalState), | |
252 | VMSTATE_END_OF_LIST() | |
253 | }, | |
254 | }; | |
255 | ||
256 | void register_global_state(void) | |
257 | { | |
258 | /* We would use it independently that we receive it */ | |
259 | strcpy((char *)&global_state.runstate, ""); | |
172c4356 | 260 | global_state.received = false; |
df4b1024 JQ |
261 | vmstate_register(NULL, 0, &vmstate_globalstate, &global_state); |
262 | } | |
263 | ||
b05dc723 JQ |
264 | static void migrate_generate_event(int new_state) |
265 | { | |
266 | if (migrate_use_events()) { | |
267 | qapi_event_send_migration(new_state, &error_abort); | |
b05dc723 JQ |
268 | } |
269 | } | |
270 | ||
adde220a DDAG |
271 | /* |
272 | * Called on -incoming with a defer: uri. | |
273 | * The migration can be started later after any parameters have been | |
274 | * changed. | |
275 | */ | |
276 | static void deferred_incoming_migration(Error **errp) | |
277 | { | |
278 | if (deferred_incoming) { | |
279 | error_setg(errp, "Incoming migration already deferred"); | |
280 | } | |
281 | deferred_incoming = true; | |
282 | } | |
283 | ||
1e2d90eb DDAG |
284 | /* Request a range of pages from the source VM at the given |
285 | * start address. | |
286 | * rbname: Name of the RAMBlock to request the page in, if NULL it's the same | |
287 | * as the last request (a name must have been given previously) | |
288 | * Start: Address offset within the RB | |
289 | * Len: Length in bytes required - must be a multiple of pagesize | |
290 | */ | |
291 | void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname, | |
292 | ram_addr_t start, size_t len) | |
293 | { | |
cb8d4c8f | 294 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb DDAG |
295 | size_t msglen = 12; /* start + len */ |
296 | ||
297 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
298 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
299 | ||
300 | if (rbname) { | |
301 | int rbname_len = strlen(rbname); | |
302 | assert(rbname_len < 256); | |
303 | ||
304 | bufc[msglen++] = rbname_len; | |
305 | memcpy(bufc + msglen, rbname, rbname_len); | |
306 | msglen += rbname_len; | |
307 | migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc); | |
308 | } else { | |
309 | migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc); | |
310 | } | |
311 | } | |
312 | ||
43eaae28 | 313 | void qemu_start_incoming_migration(const char *uri, Error **errp) |
5bb7910a | 314 | { |
34c9dd8e AL |
315 | const char *p; |
316 | ||
7cf1fe6d | 317 | qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort); |
adde220a DDAG |
318 | if (!strcmp(uri, "defer")) { |
319 | deferred_incoming_migration(errp); | |
320 | } else if (strstart(uri, "tcp:", &p)) { | |
43eaae28 | 321 | tcp_start_incoming_migration(p, errp); |
2da776db | 322 | #ifdef CONFIG_RDMA |
adde220a | 323 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
324 | rdma_start_incoming_migration(p, errp); |
325 | #endif | |
adde220a | 326 | } else if (strstart(uri, "exec:", &p)) { |
43eaae28 | 327 | exec_start_incoming_migration(p, errp); |
adde220a | 328 | } else if (strstart(uri, "unix:", &p)) { |
43eaae28 | 329 | unix_start_incoming_migration(p, errp); |
adde220a | 330 | } else if (strstart(uri, "fd:", &p)) { |
43eaae28 | 331 | fd_start_incoming_migration(p, errp); |
adde220a | 332 | } else { |
312fd5f2 | 333 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 334 | } |
5bb7910a AL |
335 | } |
336 | ||
0aa6aefc DL |
337 | static void process_incoming_migration_bh(void *opaque) |
338 | { | |
339 | Error *local_err = NULL; | |
340 | MigrationIncomingState *mis = opaque; | |
341 | ||
342 | /* Make sure all file formats flush their mutable metadata */ | |
343 | bdrv_invalidate_cache_all(&local_err); | |
344 | if (local_err) { | |
345 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
346 | MIGRATION_STATUS_FAILED); | |
347 | error_report_err(local_err); | |
348 | migrate_decompress_threads_join(); | |
349 | exit(EXIT_FAILURE); | |
350 | } | |
351 | ||
d35ff5e6 KW |
352 | /* If we get an error here, just don't restart the VM yet. */ |
353 | blk_resume_after_migration(&local_err); | |
354 | if (local_err) { | |
355 | error_free(local_err); | |
356 | local_err = NULL; | |
357 | autostart = false; | |
358 | } | |
359 | ||
0aa6aefc DL |
360 | /* |
361 | * This must happen after all error conditions are dealt with and | |
362 | * we're sure the VM is going to be running on this host. | |
363 | */ | |
364 | qemu_announce_self(); | |
365 | ||
366 | /* If global state section was not received or we are in running | |
367 | state, we need to obey autostart. Any other state is set with | |
368 | runstate_set. */ | |
369 | ||
370 | if (!global_state_received() || | |
371 | global_state_get_runstate() == RUN_STATE_RUNNING) { | |
372 | if (autostart) { | |
373 | vm_start(); | |
374 | } else { | |
375 | runstate_set(RUN_STATE_PAUSED); | |
376 | } | |
377 | } else { | |
378 | runstate_set(global_state_get_runstate()); | |
379 | } | |
380 | migrate_decompress_threads_join(); | |
381 | /* | |
382 | * This must happen after any state changes since as soon as an external | |
383 | * observer sees this event they might start to prod at the VM assuming | |
384 | * it's ready to use. | |
385 | */ | |
386 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
387 | MIGRATION_STATUS_COMPLETED); | |
388 | qemu_bh_delete(mis->bh); | |
389 | migration_incoming_state_destroy(); | |
390 | } | |
391 | ||
82a4da79 | 392 | static void process_incoming_migration_co(void *opaque) |
511c0231 | 393 | { |
82a4da79 | 394 | QEMUFile *f = opaque; |
b4b076da | 395 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 396 | PostcopyState ps; |
1c12e1f5 PB |
397 | int ret; |
398 | ||
b4b076da | 399 | mis->from_src_file = f; |
67f11b5c | 400 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 401 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
93d7af6f HZ |
402 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
403 | MIGRATION_STATUS_ACTIVE); | |
1c12e1f5 | 404 | ret = qemu_loadvm_state(f); |
bca7856a | 405 | |
e9bef235 DDAG |
406 | ps = postcopy_state_get(); |
407 | trace_process_incoming_migration_co_end(ret, ps); | |
408 | if (ps != POSTCOPY_INCOMING_NONE) { | |
409 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
410 | /* | |
411 | * Where a migration had postcopy enabled (and thus went to advise) | |
412 | * but managed to complete within the precopy period, we can use | |
413 | * the normal exit. | |
414 | */ | |
415 | postcopy_ram_incoming_cleanup(mis); | |
416 | } else if (ret >= 0) { | |
417 | /* | |
418 | * Postcopy was started, cleanup should happen at the end of the | |
419 | * postcopy thread. | |
420 | */ | |
421 | trace_process_incoming_migration_co_postcopy_end_main(); | |
422 | return; | |
423 | } | |
424 | /* Else if something went wrong then just fall out of the normal exit */ | |
425 | } | |
426 | ||
25d0c16f HZ |
427 | /* we get COLO info, and know if we are in COLO mode */ |
428 | if (!ret && migration_incoming_enable_colo()) { | |
429 | mis->migration_incoming_co = qemu_coroutine_self(); | |
430 | qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", | |
431 | colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); | |
432 | mis->have_colo_incoming_thread = true; | |
433 | qemu_coroutine_yield(); | |
434 | ||
435 | /* Wait checkpoint incoming thread exit before free resource */ | |
436 | qemu_thread_join(&mis->colo_incoming_thread); | |
437 | } | |
438 | ||
1c12e1f5 | 439 | qemu_fclose(f); |
905f26f2 | 440 | free_xbzrle_decoded_buf(); |
bca7856a | 441 | |
1c12e1f5 | 442 | if (ret < 0) { |
93d7af6f HZ |
443 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, |
444 | MIGRATION_STATUS_FAILED); | |
db80face | 445 | error_report("load of migration failed: %s", strerror(-ret)); |
3fcb38c2 | 446 | migrate_decompress_threads_join(); |
4aead692 | 447 | exit(EXIT_FAILURE); |
511c0231 | 448 | } |
511c0231 | 449 | |
0aa6aefc DL |
450 | mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); |
451 | qemu_bh_schedule(mis->bh); | |
511c0231 JQ |
452 | } |
453 | ||
22724f49 | 454 | void migration_fd_process_incoming(QEMUFile *f) |
82a4da79 | 455 | { |
0b8b8753 | 456 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f); |
82a4da79 | 457 | |
3fcb38c2 | 458 | migrate_decompress_threads_create(); |
06ad5135 | 459 | qemu_file_set_blocking(f, false); |
0b8b8753 | 460 | qemu_coroutine_enter(co); |
82a4da79 PB |
461 | } |
462 | ||
48f07489 | 463 | |
22724f49 DB |
464 | void migration_channel_process_incoming(MigrationState *s, |
465 | QIOChannel *ioc) | |
48f07489 | 466 | { |
e1226365 DB |
467 | trace_migration_set_incoming_channel( |
468 | ioc, object_get_typename(OBJECT(ioc))); | |
469 | ||
470 | if (s->parameters.tls_creds && | |
4af245dc | 471 | *s->parameters.tls_creds && |
e1226365 DB |
472 | !object_dynamic_cast(OBJECT(ioc), |
473 | TYPE_QIO_CHANNEL_TLS)) { | |
474 | Error *local_err = NULL; | |
22724f49 | 475 | migration_tls_channel_process_incoming(s, ioc, &local_err); |
e1226365 DB |
476 | if (local_err) { |
477 | error_report_err(local_err); | |
478 | } | |
479 | } else { | |
480 | QEMUFile *f = qemu_fopen_channel_input(ioc); | |
22724f49 | 481 | migration_fd_process_incoming(f); |
e1226365 | 482 | } |
48f07489 DB |
483 | } |
484 | ||
485 | ||
22724f49 DB |
486 | void migration_channel_connect(MigrationState *s, |
487 | QIOChannel *ioc, | |
488 | const char *hostname) | |
48f07489 | 489 | { |
e1226365 DB |
490 | trace_migration_set_outgoing_channel( |
491 | ioc, object_get_typename(OBJECT(ioc)), hostname); | |
492 | ||
493 | if (s->parameters.tls_creds && | |
4af245dc | 494 | *s->parameters.tls_creds && |
e1226365 DB |
495 | !object_dynamic_cast(OBJECT(ioc), |
496 | TYPE_QIO_CHANNEL_TLS)) { | |
497 | Error *local_err = NULL; | |
22724f49 | 498 | migration_tls_channel_connect(s, ioc, hostname, &local_err); |
e1226365 DB |
499 | if (local_err) { |
500 | migrate_fd_error(s, local_err); | |
501 | error_free(local_err); | |
502 | } | |
503 | } else { | |
504 | QEMUFile *f = qemu_fopen_channel_output(ioc); | |
48f07489 | 505 | |
e1226365 | 506 | s->to_dst_file = f; |
48f07489 | 507 | |
e1226365 DB |
508 | migrate_fd_connect(s); |
509 | } | |
48f07489 DB |
510 | } |
511 | ||
512 | ||
6decec93 DDAG |
513 | /* |
514 | * Send a message on the return channel back to the source | |
515 | * of the migration. | |
516 | */ | |
517 | void migrate_send_rp_message(MigrationIncomingState *mis, | |
518 | enum mig_rp_message_type message_type, | |
519 | uint16_t len, void *data) | |
520 | { | |
521 | trace_migrate_send_rp_message((int)message_type, len); | |
522 | qemu_mutex_lock(&mis->rp_mutex); | |
523 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); | |
524 | qemu_put_be16(mis->to_src_file, len); | |
525 | qemu_put_buffer(mis->to_src_file, data, len); | |
526 | qemu_fflush(mis->to_src_file); | |
527 | qemu_mutex_unlock(&mis->rp_mutex); | |
528 | } | |
529 | ||
530 | /* | |
531 | * Send a 'SHUT' message on the return channel with the given value | |
532 | * to indicate that we've finished with the RP. Non-0 value indicates | |
533 | * error. | |
534 | */ | |
535 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
536 | uint32_t value) | |
537 | { | |
538 | uint32_t buf; | |
539 | ||
540 | buf = cpu_to_be32(value); | |
541 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
542 | } | |
543 | ||
544 | /* | |
545 | * Send a 'PONG' message on the return channel with the given value | |
546 | * (normally in response to a 'PING') | |
547 | */ | |
548 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
549 | uint32_t value) | |
550 | { | |
551 | uint32_t buf; | |
552 | ||
553 | buf = cpu_to_be32(value); | |
554 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
555 | } | |
556 | ||
bbf6da32 OW |
557 | MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) |
558 | { | |
559 | MigrationCapabilityStatusList *head = NULL; | |
560 | MigrationCapabilityStatusList *caps; | |
561 | MigrationState *s = migrate_get_current(); | |
562 | int i; | |
563 | ||
387eedeb | 564 | caps = NULL; /* silence compiler warning */ |
7fb1cf16 | 565 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
35a6ed4f HZ |
566 | if (i == MIGRATION_CAPABILITY_X_COLO && !colo_supported()) { |
567 | continue; | |
568 | } | |
bbf6da32 OW |
569 | if (head == NULL) { |
570 | head = g_malloc0(sizeof(*caps)); | |
571 | caps = head; | |
572 | } else { | |
573 | caps->next = g_malloc0(sizeof(*caps)); | |
574 | caps = caps->next; | |
575 | } | |
576 | caps->value = | |
577 | g_malloc(sizeof(*caps->value)); | |
578 | caps->value->capability = i; | |
579 | caps->value->state = s->enabled_capabilities[i]; | |
580 | } | |
581 | ||
582 | return head; | |
583 | } | |
584 | ||
85de8323 LL |
585 | MigrationParameters *qmp_query_migrate_parameters(Error **errp) |
586 | { | |
587 | MigrationParameters *params; | |
588 | MigrationState *s = migrate_get_current(); | |
589 | ||
590 | params = g_malloc0(sizeof(*params)); | |
de63ab61 | 591 | params->has_compress_level = true; |
2594f56d | 592 | params->compress_level = s->parameters.compress_level; |
de63ab61 | 593 | params->has_compress_threads = true; |
2594f56d | 594 | params->compress_threads = s->parameters.compress_threads; |
de63ab61 | 595 | params->has_decompress_threads = true; |
2594f56d | 596 | params->decompress_threads = s->parameters.decompress_threads; |
de63ab61 | 597 | params->has_cpu_throttle_initial = true; |
2594f56d | 598 | params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; |
de63ab61 | 599 | params->has_cpu_throttle_increment = true; |
2594f56d | 600 | params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; |
de63ab61 | 601 | params->has_tls_creds = !!s->parameters.tls_creds; |
69ef1f36 | 602 | params->tls_creds = g_strdup(s->parameters.tls_creds); |
de63ab61 | 603 | params->has_tls_hostname = !!s->parameters.tls_hostname; |
69ef1f36 | 604 | params->tls_hostname = g_strdup(s->parameters.tls_hostname); |
2ff30257 AA |
605 | params->has_max_bandwidth = true; |
606 | params->max_bandwidth = s->parameters.max_bandwidth; | |
607 | params->has_downtime_limit = true; | |
608 | params->downtime_limit = s->parameters.downtime_limit; | |
fe39a4d4 | 609 | params->has_x_checkpoint_delay = true; |
68b53591 | 610 | params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; |
85de8323 LL |
611 | |
612 | return params; | |
613 | } | |
614 | ||
f6844b99 DDAG |
615 | /* |
616 | * Return true if we're already in the middle of a migration | |
617 | * (i.e. any of the active or setup states) | |
618 | */ | |
619 | static bool migration_is_setup_or_active(int state) | |
620 | { | |
621 | switch (state) { | |
622 | case MIGRATION_STATUS_ACTIVE: | |
9ec055ae | 623 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
f6844b99 DDAG |
624 | case MIGRATION_STATUS_SETUP: |
625 | return true; | |
626 | ||
627 | default: | |
628 | return false; | |
629 | ||
630 | } | |
631 | } | |
632 | ||
f36d55af OW |
633 | static void get_xbzrle_cache_stats(MigrationInfo *info) |
634 | { | |
635 | if (migrate_use_xbzrle()) { | |
636 | info->has_xbzrle_cache = true; | |
637 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); | |
638 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
639 | info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred(); | |
640 | info->xbzrle_cache->pages = xbzrle_mig_pages_transferred(); | |
641 | info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss(); | |
8bc39233 | 642 | info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate(); |
f36d55af OW |
643 | info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow(); |
644 | } | |
645 | } | |
646 | ||
a22463a5 DDAG |
647 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
648 | { | |
649 | info->has_ram = true; | |
650 | info->ram = g_malloc0(sizeof(*info->ram)); | |
651 | info->ram->transferred = ram_bytes_transferred(); | |
652 | info->ram->total = ram_bytes_total(); | |
653 | info->ram->duplicate = dup_mig_pages_transferred(); | |
654 | info->ram->skipped = skipped_mig_pages_transferred(); | |
655 | info->ram->normal = norm_mig_pages_transferred(); | |
656 | info->ram->normal_bytes = norm_mig_bytes_transferred(); | |
657 | info->ram->mbps = s->mbps; | |
658 | info->ram->dirty_sync_count = s->dirty_sync_count; | |
d3bf5418 | 659 | info->ram->postcopy_requests = s->postcopy_requests; |
a22463a5 DDAG |
660 | |
661 | if (s->state != MIGRATION_STATUS_COMPLETED) { | |
662 | info->ram->remaining = ram_bytes_remaining(); | |
663 | info->ram->dirty_pages_rate = s->dirty_pages_rate; | |
664 | } | |
665 | } | |
666 | ||
791e7c82 | 667 | MigrationInfo *qmp_query_migrate(Error **errp) |
5bb7910a | 668 | { |
791e7c82 | 669 | MigrationInfo *info = g_malloc0(sizeof(*info)); |
17549e84 JQ |
670 | MigrationState *s = migrate_get_current(); |
671 | ||
672 | switch (s->state) { | |
31194731 | 673 | case MIGRATION_STATUS_NONE: |
17549e84 JQ |
674 | /* no migration has happened ever */ |
675 | break; | |
31194731 | 676 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 677 | info->has_status = true; |
ed4fbd10 | 678 | info->has_total_time = false; |
29ae8a41 | 679 | break; |
31194731 HZ |
680 | case MIGRATION_STATUS_ACTIVE: |
681 | case MIGRATION_STATUS_CANCELLING: | |
791e7c82 | 682 | info->has_status = true; |
7aa939af | 683 | info->has_total_time = true; |
bc72ad67 | 684 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) |
7aa939af | 685 | - s->total_time; |
2c52ddf1 JQ |
686 | info->has_expected_downtime = true; |
687 | info->expected_downtime = s->expected_downtime; | |
ed4fbd10 MH |
688 | info->has_setup_time = true; |
689 | info->setup_time = s->setup_time; | |
17549e84 | 690 | |
a22463a5 | 691 | populate_ram_info(info, s); |
8d017193 | 692 | |
17549e84 | 693 | if (blk_mig_active()) { |
791e7c82 LC |
694 | info->has_disk = true; |
695 | info->disk = g_malloc0(sizeof(*info->disk)); | |
696 | info->disk->transferred = blk_mig_bytes_transferred(); | |
697 | info->disk->remaining = blk_mig_bytes_remaining(); | |
698 | info->disk->total = blk_mig_bytes_total(); | |
ff8d81d8 | 699 | } |
f36d55af | 700 | |
4782893e | 701 | if (cpu_throttle_active()) { |
d85a31d1 JH |
702 | info->has_cpu_throttle_percentage = true; |
703 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
4782893e JH |
704 | } |
705 | ||
9ec055ae DDAG |
706 | get_xbzrle_cache_stats(info); |
707 | break; | |
708 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
709 | /* Mostly the same as active; TODO add some postcopy stats */ | |
710 | info->has_status = true; | |
711 | info->has_total_time = true; | |
712 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) | |
713 | - s->total_time; | |
714 | info->has_expected_downtime = true; | |
715 | info->expected_downtime = s->expected_downtime; | |
716 | info->has_setup_time = true; | |
717 | info->setup_time = s->setup_time; | |
718 | ||
a22463a5 | 719 | populate_ram_info(info, s); |
9ec055ae DDAG |
720 | |
721 | if (blk_mig_active()) { | |
722 | info->has_disk = true; | |
723 | info->disk = g_malloc0(sizeof(*info->disk)); | |
724 | info->disk->transferred = blk_mig_bytes_transferred(); | |
725 | info->disk->remaining = blk_mig_bytes_remaining(); | |
726 | info->disk->total = blk_mig_bytes_total(); | |
727 | } | |
728 | ||
f36d55af | 729 | get_xbzrle_cache_stats(info); |
17549e84 | 730 | break; |
0b827d5e HZ |
731 | case MIGRATION_STATUS_COLO: |
732 | info->has_status = true; | |
733 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
734 | break; | |
31194731 | 735 | case MIGRATION_STATUS_COMPLETED: |
f36d55af OW |
736 | get_xbzrle_cache_stats(info); |
737 | ||
791e7c82 | 738 | info->has_status = true; |
00c14997 | 739 | info->has_total_time = true; |
7aa939af | 740 | info->total_time = s->total_time; |
9c5a9fcf JQ |
741 | info->has_downtime = true; |
742 | info->downtime = s->downtime; | |
ed4fbd10 MH |
743 | info->has_setup_time = true; |
744 | info->setup_time = s->setup_time; | |
d5f8a570 | 745 | |
a22463a5 | 746 | populate_ram_info(info, s); |
17549e84 | 747 | break; |
31194731 | 748 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 749 | info->has_status = true; |
d59ce6f3 DB |
750 | if (s->error) { |
751 | info->has_error_desc = true; | |
752 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
753 | } | |
17549e84 | 754 | break; |
31194731 | 755 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 756 | info->has_status = true; |
17549e84 | 757 | break; |
5bb7910a | 758 | } |
cde63fbe | 759 | info->status = s->state; |
791e7c82 LC |
760 | |
761 | return info; | |
5bb7910a AL |
762 | } |
763 | ||
00458433 OW |
764 | void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params, |
765 | Error **errp) | |
766 | { | |
767 | MigrationState *s = migrate_get_current(); | |
768 | MigrationCapabilityStatusList *cap; | |
096631bd | 769 | bool old_postcopy_cap = migrate_postcopy_ram(); |
00458433 | 770 | |
f6844b99 | 771 | if (migration_is_setup_or_active(s->state)) { |
c6bd8c70 | 772 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
00458433 OW |
773 | return; |
774 | } | |
775 | ||
776 | for (cap = params; cap; cap = cap->next) { | |
35a6ed4f HZ |
777 | if (cap->value->capability == MIGRATION_CAPABILITY_X_COLO) { |
778 | if (!colo_supported()) { | |
779 | error_setg(errp, "COLO is not currently supported, please" | |
780 | " configure with --enable-colo option in order to" | |
781 | " support COLO feature"); | |
782 | continue; | |
783 | } | |
784 | } | |
00458433 OW |
785 | s->enabled_capabilities[cap->value->capability] = cap->value->state; |
786 | } | |
53dd370c DDAG |
787 | |
788 | if (migrate_postcopy_ram()) { | |
789 | if (migrate_use_compression()) { | |
790 | /* The decompression threads asynchronously write into RAM | |
791 | * rather than use the atomic copies needed to avoid | |
792 | * userfaulting. It should be possible to fix the decompression | |
793 | * threads for compatibility in future. | |
794 | */ | |
795 | error_report("Postcopy is not currently compatible with " | |
796 | "compression"); | |
32c3db5b | 797 | s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] = |
53dd370c DDAG |
798 | false; |
799 | } | |
096631bd DDAG |
800 | /* This check is reasonably expensive, so only when it's being |
801 | * set the first time, also it's only the destination that needs | |
802 | * special support. | |
803 | */ | |
804 | if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) && | |
805 | !postcopy_ram_supported_by_host()) { | |
806 | /* postcopy_ram_supported_by_host will have emitted a more | |
807 | * detailed message | |
808 | */ | |
809 | error_report("Postcopy is not supported"); | |
810 | s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] = | |
811 | false; | |
812 | } | |
53dd370c | 813 | } |
00458433 OW |
814 | } |
815 | ||
7f375e04 | 816 | void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp) |
85de8323 LL |
817 | { |
818 | MigrationState *s = migrate_get_current(); | |
819 | ||
7f375e04 EB |
820 | if (params->has_compress_level && |
821 | (params->compress_level < 0 || params->compress_level > 9)) { | |
c6bd8c70 MA |
822 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", |
823 | "is invalid, it should be in the range of 0 to 9"); | |
85de8323 LL |
824 | return; |
825 | } | |
7f375e04 EB |
826 | if (params->has_compress_threads && |
827 | (params->compress_threads < 1 || params->compress_threads > 255)) { | |
c6bd8c70 MA |
828 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
829 | "compress_threads", | |
830 | "is invalid, it should be in the range of 1 to 255"); | |
85de8323 LL |
831 | return; |
832 | } | |
7f375e04 EB |
833 | if (params->has_decompress_threads && |
834 | (params->decompress_threads < 1 || params->decompress_threads > 255)) { | |
c6bd8c70 MA |
835 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
836 | "decompress_threads", | |
837 | "is invalid, it should be in the range of 1 to 255"); | |
85de8323 LL |
838 | return; |
839 | } | |
7f375e04 EB |
840 | if (params->has_cpu_throttle_initial && |
841 | (params->cpu_throttle_initial < 1 || | |
842 | params->cpu_throttle_initial > 99)) { | |
1626fee3 | 843 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 844 | "cpu_throttle_initial", |
1626fee3 | 845 | "an integer in the range of 1 to 99"); |
091ecc8b | 846 | return; |
1626fee3 | 847 | } |
7f375e04 EB |
848 | if (params->has_cpu_throttle_increment && |
849 | (params->cpu_throttle_increment < 1 || | |
850 | params->cpu_throttle_increment > 99)) { | |
1626fee3 | 851 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 852 | "cpu_throttle_increment", |
1626fee3 | 853 | "an integer in the range of 1 to 99"); |
091ecc8b | 854 | return; |
1626fee3 | 855 | } |
2ff30257 AA |
856 | if (params->has_max_bandwidth && |
857 | (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) { | |
858 | error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the" | |
859 | " range of 0 to %zu bytes/second", SIZE_MAX); | |
860 | return; | |
861 | } | |
862 | if (params->has_downtime_limit && | |
87c9cc1c DHB |
863 | (params->downtime_limit < 0 || |
864 | params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { | |
865 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " | |
866 | "the range of 0 to %d milliseconds", | |
867 | MAX_MIGRATE_DOWNTIME); | |
2ff30257 AA |
868 | return; |
869 | } | |
68b53591 HZ |
870 | if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) { |
871 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
872 | "x_checkpoint_delay", | |
873 | "is invalid, it should be positive"); | |
874 | } | |
85de8323 | 875 | |
7f375e04 EB |
876 | if (params->has_compress_level) { |
877 | s->parameters.compress_level = params->compress_level; | |
85de8323 | 878 | } |
7f375e04 EB |
879 | if (params->has_compress_threads) { |
880 | s->parameters.compress_threads = params->compress_threads; | |
85de8323 | 881 | } |
7f375e04 EB |
882 | if (params->has_decompress_threads) { |
883 | s->parameters.decompress_threads = params->decompress_threads; | |
85de8323 | 884 | } |
7f375e04 EB |
885 | if (params->has_cpu_throttle_initial) { |
886 | s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; | |
1626fee3 | 887 | } |
7f375e04 EB |
888 | if (params->has_cpu_throttle_increment) { |
889 | s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; | |
1626fee3 | 890 | } |
7f375e04 | 891 | if (params->has_tls_creds) { |
69ef1f36 | 892 | g_free(s->parameters.tls_creds); |
7f375e04 | 893 | s->parameters.tls_creds = g_strdup(params->tls_creds); |
69ef1f36 | 894 | } |
7f375e04 | 895 | if (params->has_tls_hostname) { |
69ef1f36 | 896 | g_free(s->parameters.tls_hostname); |
7f375e04 | 897 | s->parameters.tls_hostname = g_strdup(params->tls_hostname); |
69ef1f36 | 898 | } |
2ff30257 AA |
899 | if (params->has_max_bandwidth) { |
900 | s->parameters.max_bandwidth = params->max_bandwidth; | |
901 | if (s->to_dst_file) { | |
902 | qemu_file_set_rate_limit(s->to_dst_file, | |
903 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); | |
904 | } | |
905 | } | |
906 | if (params->has_downtime_limit) { | |
907 | s->parameters.downtime_limit = params->downtime_limit; | |
908 | } | |
68b53591 HZ |
909 | |
910 | if (params->has_x_checkpoint_delay) { | |
911 | s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; | |
479125d5 HZ |
912 | if (migration_in_colo_state()) { |
913 | colo_checkpoint_notify(s); | |
914 | } | |
68b53591 | 915 | } |
85de8323 LL |
916 | } |
917 | ||
2594f56d | 918 | |
4886a1bc DDAG |
919 | void qmp_migrate_start_postcopy(Error **errp) |
920 | { | |
921 | MigrationState *s = migrate_get_current(); | |
922 | ||
923 | if (!migrate_postcopy_ram()) { | |
a54d340b | 924 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
925 | " the start of migration"); |
926 | return; | |
927 | } | |
928 | ||
929 | if (s->state == MIGRATION_STATUS_NONE) { | |
930 | error_setg(errp, "Postcopy must be started after migration has been" | |
931 | " started"); | |
932 | return; | |
933 | } | |
934 | /* | |
935 | * we don't error if migration has finished since that would be racy | |
936 | * with issuing this command. | |
937 | */ | |
938 | atomic_set(&s->start_postcopy, true); | |
939 | } | |
940 | ||
065e2813 AL |
941 | /* shared migration helpers */ |
942 | ||
48781e5b | 943 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 944 | { |
48781e5b | 945 | if (atomic_cmpxchg(state, old_state, new_state) == old_state) { |
4ba4bc5e | 946 | trace_migrate_set_state(new_state); |
b05dc723 | 947 | migrate_generate_event(new_state); |
51cf4c1a Z |
948 | } |
949 | } | |
950 | ||
bb1fadc4 | 951 | static void migrate_fd_cleanup(void *opaque) |
065e2813 | 952 | { |
bb1fadc4 PB |
953 | MigrationState *s = opaque; |
954 | ||
955 | qemu_bh_delete(s->cleanup_bh); | |
956 | s->cleanup_bh = NULL; | |
957 | ||
6c595cde DDAG |
958 | flush_page_queue(s); |
959 | ||
89a02a9f | 960 | if (s->to_dst_file) { |
9013dca5 | 961 | trace_migrate_fd_cleanup(); |
404a7c05 | 962 | qemu_mutex_unlock_iothread(); |
1d34e4bf DDAG |
963 | if (s->migration_thread_running) { |
964 | qemu_thread_join(&s->thread); | |
965 | s->migration_thread_running = false; | |
966 | } | |
404a7c05 PB |
967 | qemu_mutex_lock_iothread(); |
968 | ||
8706d2d5 | 969 | migrate_compress_threads_join(); |
89a02a9f HZ |
970 | qemu_fclose(s->to_dst_file); |
971 | s->to_dst_file = NULL; | |
065e2813 AL |
972 | } |
973 | ||
9ec055ae DDAG |
974 | assert((s->state != MIGRATION_STATUS_ACTIVE) && |
975 | (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE)); | |
7a2c1721 | 976 | |
94f5a437 | 977 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 978 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 979 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 980 | } |
a3fa1d78 PB |
981 | |
982 | notifier_list_notify(&migration_state_notifiers, s); | |
065e2813 AL |
983 | } |
984 | ||
d59ce6f3 | 985 | void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 986 | { |
25174055 | 987 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 988 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
989 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
990 | MIGRATION_STATUS_FAILED); | |
d59ce6f3 DB |
991 | if (!s->error) { |
992 | s->error = error_copy(error); | |
993 | } | |
bb1fadc4 | 994 | notifier_list_notify(&migration_state_notifiers, s); |
458cf28e JQ |
995 | } |
996 | ||
0edda1c4 | 997 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 998 | { |
6f2b811a | 999 | int old_state ; |
89a02a9f | 1000 | QEMUFile *f = migrate_get_current()->to_dst_file; |
9013dca5 | 1001 | trace_migrate_fd_cancel(); |
065e2813 | 1002 | |
70b20477 DDAG |
1003 | if (s->rp_state.from_dst_file) { |
1004 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
1005 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
1006 | } | |
1007 | ||
6f2b811a Z |
1008 | do { |
1009 | old_state = s->state; | |
f6844b99 | 1010 | if (!migration_is_setup_or_active(old_state)) { |
6f2b811a Z |
1011 | break; |
1012 | } | |
48781e5b | 1013 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 1014 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
1015 | |
1016 | /* | |
1017 | * If we're unlucky the migration code might be stuck somewhere in a | |
1018 | * send/write while the network has failed and is waiting to timeout; | |
1019 | * if we've got shutdown(2) available then we can force it to quit. | |
1020 | * The outgoing qemu file gets closed in migrate_fd_cleanup that is | |
1021 | * called in a bh, so there is no race against this cancel. | |
1022 | */ | |
31194731 | 1023 | if (s->state == MIGRATION_STATUS_CANCELLING && f) { |
a26ba26e DDAG |
1024 | qemu_file_shutdown(f); |
1025 | } | |
1d2acc31 HZ |
1026 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
1027 | Error *local_err = NULL; | |
1028 | ||
1029 | bdrv_invalidate_cache_all(&local_err); | |
1030 | if (local_err) { | |
1031 | error_report_err(local_err); | |
1032 | } else { | |
1033 | s->block_inactive = false; | |
1034 | } | |
1035 | } | |
065e2813 AL |
1036 | } |
1037 | ||
99a0db9b GH |
1038 | void add_migration_state_change_notifier(Notifier *notify) |
1039 | { | |
1040 | notifier_list_add(&migration_state_notifiers, notify); | |
1041 | } | |
1042 | ||
1043 | void remove_migration_state_change_notifier(Notifier *notify) | |
1044 | { | |
31552529 | 1045 | notifier_remove(notify); |
99a0db9b GH |
1046 | } |
1047 | ||
02edd2e7 | 1048 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 1049 | { |
31194731 | 1050 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
1051 | } |
1052 | ||
7073693b | 1053 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 1054 | { |
31194731 | 1055 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 1056 | } |
0edda1c4 | 1057 | |
afe2df69 GH |
1058 | bool migration_has_failed(MigrationState *s) |
1059 | { | |
31194731 HZ |
1060 | return (s->state == MIGRATION_STATUS_CANCELLED || |
1061 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
1062 | } |
1063 | ||
9ec055ae DDAG |
1064 | bool migration_in_postcopy(MigrationState *s) |
1065 | { | |
1066 | return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
1067 | } | |
1068 | ||
b82fc321 DDAG |
1069 | bool migration_in_postcopy_after_devices(MigrationState *s) |
1070 | { | |
1071 | return migration_in_postcopy(s) && s->postcopy_after_devices; | |
1072 | } | |
1073 | ||
fe44dc91 AA |
1074 | bool migration_is_idle(MigrationState *s) |
1075 | { | |
1076 | if (!s) { | |
1077 | s = migrate_get_current(); | |
1078 | } | |
1079 | ||
1080 | switch (s->state) { | |
1081 | case MIGRATION_STATUS_NONE: | |
1082 | case MIGRATION_STATUS_CANCELLED: | |
1083 | case MIGRATION_STATUS_COMPLETED: | |
1084 | case MIGRATION_STATUS_FAILED: | |
1085 | return true; | |
1086 | case MIGRATION_STATUS_SETUP: | |
1087 | case MIGRATION_STATUS_CANCELLING: | |
1088 | case MIGRATION_STATUS_ACTIVE: | |
1089 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1090 | case MIGRATION_STATUS_COLO: | |
1091 | return false; | |
1092 | case MIGRATION_STATUS__MAX: | |
1093 | g_assert_not_reached(); | |
1094 | } | |
1095 | ||
1096 | return false; | |
1097 | } | |
1098 | ||
aefeb18b | 1099 | MigrationState *migrate_init(const MigrationParams *params) |
0edda1c4 | 1100 | { |
17549e84 | 1101 | MigrationState *s = migrate_get_current(); |
bbf6da32 | 1102 | |
389775d1 DDAG |
1103 | /* |
1104 | * Reinitialise all migration state, except | |
1105 | * parameters/capabilities that the user set, and | |
1106 | * locks. | |
1107 | */ | |
1108 | s->bytes_xfer = 0; | |
1109 | s->xfer_limit = 0; | |
1110 | s->cleanup_bh = 0; | |
89a02a9f | 1111 | s->to_dst_file = NULL; |
389775d1 | 1112 | s->state = MIGRATION_STATUS_NONE; |
6607ae23 | 1113 | s->params = *params; |
389775d1 DDAG |
1114 | s->rp_state.from_dst_file = NULL; |
1115 | s->rp_state.error = false; | |
1116 | s->mbps = 0.0; | |
1117 | s->downtime = 0; | |
1118 | s->expected_downtime = 0; | |
1119 | s->dirty_pages_rate = 0; | |
1120 | s->dirty_bytes_rate = 0; | |
1121 | s->setup_time = 0; | |
1122 | s->dirty_sync_count = 0; | |
1123 | s->start_postcopy = false; | |
b82fc321 | 1124 | s->postcopy_after_devices = false; |
d3bf5418 | 1125 | s->postcopy_requests = 0; |
389775d1 DDAG |
1126 | s->migration_thread_running = false; |
1127 | s->last_req_rb = NULL; | |
d59ce6f3 DB |
1128 | error_free(s->error); |
1129 | s->error = NULL; | |
389775d1 | 1130 | |
48781e5b | 1131 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 1132 | |
6c595cde DDAG |
1133 | QSIMPLEQ_INIT(&s->src_page_requests); |
1134 | ||
bc72ad67 | 1135 | s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0edda1c4 JQ |
1136 | return s; |
1137 | } | |
cab30143 | 1138 | |
fa2756b7 AL |
1139 | static GSList *migration_blockers; |
1140 | ||
fe44dc91 | 1141 | int migrate_add_blocker(Error *reason, Error **errp) |
fa2756b7 | 1142 | { |
b67b8c3a AA |
1143 | if (only_migratable) { |
1144 | error_propagate(errp, error_copy(reason)); | |
1145 | error_prepend(errp, "disallowing migration blocker " | |
1146 | "(--only_migratable) for: "); | |
1147 | return -EACCES; | |
1148 | } | |
1149 | ||
fe44dc91 AA |
1150 | if (migration_is_idle(NULL)) { |
1151 | migration_blockers = g_slist_prepend(migration_blockers, reason); | |
1152 | return 0; | |
1153 | } | |
1154 | ||
1155 | error_propagate(errp, error_copy(reason)); | |
1156 | error_prepend(errp, "disallowing migration blocker (migration in " | |
1157 | "progress) for: "); | |
1158 | return -EBUSY; | |
fa2756b7 AL |
1159 | } |
1160 | ||
1161 | void migrate_del_blocker(Error *reason) | |
1162 | { | |
1163 | migration_blockers = g_slist_remove(migration_blockers, reason); | |
1164 | } | |
1165 | ||
7562f907 AA |
1166 | int check_migratable(Object *obj, Error **err) |
1167 | { | |
1168 | DeviceClass *dc = DEVICE_GET_CLASS(obj); | |
1169 | if (only_migratable && dc->vmsd) { | |
1170 | if (dc->vmsd->unmigratable) { | |
1171 | error_setg(err, "Device %s is not migratable, but " | |
1172 | "--only-migratable was specified", | |
1173 | object_get_typename(obj)); | |
1174 | return -1; | |
1175 | } | |
1176 | } | |
1177 | ||
1178 | return 0; | |
1179 | } | |
1180 | ||
bf1ae1f4 DDAG |
1181 | void qmp_migrate_incoming(const char *uri, Error **errp) |
1182 | { | |
1183 | Error *local_err = NULL; | |
4debb5f5 | 1184 | static bool once = true; |
bf1ae1f4 DDAG |
1185 | |
1186 | if (!deferred_incoming) { | |
4debb5f5 | 1187 | error_setg(errp, "For use with '-incoming defer'"); |
bf1ae1f4 DDAG |
1188 | return; |
1189 | } | |
4debb5f5 DDAG |
1190 | if (!once) { |
1191 | error_setg(errp, "The incoming migration has already been started"); | |
1192 | } | |
bf1ae1f4 DDAG |
1193 | |
1194 | qemu_start_incoming_migration(uri, &local_err); | |
1195 | ||
1196 | if (local_err) { | |
1197 | error_propagate(errp, local_err); | |
1198 | return; | |
1199 | } | |
1200 | ||
4debb5f5 | 1201 | once = false; |
bf1ae1f4 DDAG |
1202 | } |
1203 | ||
24f3902b GK |
1204 | bool migration_is_blocked(Error **errp) |
1205 | { | |
1206 | if (qemu_savevm_state_blocked(errp)) { | |
1207 | return true; | |
1208 | } | |
1209 | ||
1210 | if (migration_blockers) { | |
1211 | *errp = error_copy(migration_blockers->data); | |
1212 | return true; | |
1213 | } | |
1214 | ||
1215 | return false; | |
1216 | } | |
1217 | ||
e1c37d0e LC |
1218 | void qmp_migrate(const char *uri, bool has_blk, bool blk, |
1219 | bool has_inc, bool inc, bool has_detach, bool detach, | |
1220 | Error **errp) | |
cab30143 | 1221 | { |
be7059cd | 1222 | Error *local_err = NULL; |
17549e84 | 1223 | MigrationState *s = migrate_get_current(); |
6607ae23 | 1224 | MigrationParams params; |
cab30143 | 1225 | const char *p; |
cab30143 | 1226 | |
8c0426ae PP |
1227 | params.blk = has_blk && blk; |
1228 | params.shared = has_inc && inc; | |
6607ae23 | 1229 | |
f6844b99 | 1230 | if (migration_is_setup_or_active(s->state) || |
0b827d5e HZ |
1231 | s->state == MIGRATION_STATUS_CANCELLING || |
1232 | s->state == MIGRATION_STATUS_COLO) { | |
c6bd8c70 | 1233 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
e1c37d0e | 1234 | return; |
cab30143 | 1235 | } |
ca99993a DDAG |
1236 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
1237 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
1238 | return; | |
1239 | } | |
1240 | ||
24f3902b | 1241 | if (migration_is_blocked(errp)) { |
e1c37d0e | 1242 | return; |
fa2756b7 AL |
1243 | } |
1244 | ||
6607ae23 | 1245 | s = migrate_init(¶ms); |
cab30143 JQ |
1246 | |
1247 | if (strstart(uri, "tcp:", &p)) { | |
f37afb5a | 1248 | tcp_start_outgoing_migration(s, p, &local_err); |
2da776db | 1249 | #ifdef CONFIG_RDMA |
41310c68 | 1250 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
1251 | rdma_start_outgoing_migration(s, p, &local_err); |
1252 | #endif | |
cab30143 | 1253 | } else if (strstart(uri, "exec:", &p)) { |
f37afb5a | 1254 | exec_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1255 | } else if (strstart(uri, "unix:", &p)) { |
f37afb5a | 1256 | unix_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1257 | } else if (strstart(uri, "fd:", &p)) { |
f37afb5a | 1258 | fd_start_outgoing_migration(s, p, &local_err); |
99a0db9b | 1259 | } else { |
c6bd8c70 MA |
1260 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", |
1261 | "a valid migration protocol"); | |
48781e5b HZ |
1262 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1263 | MIGRATION_STATUS_FAILED); | |
e1c37d0e | 1264 | return; |
cab30143 JQ |
1265 | } |
1266 | ||
f37afb5a | 1267 | if (local_err) { |
d59ce6f3 | 1268 | migrate_fd_error(s, local_err); |
f37afb5a | 1269 | error_propagate(errp, local_err); |
e1c37d0e | 1270 | return; |
1299c631 | 1271 | } |
cab30143 JQ |
1272 | } |
1273 | ||
6cdedb07 | 1274 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 1275 | { |
17549e84 | 1276 | migrate_fd_cancel(migrate_get_current()); |
cab30143 JQ |
1277 | } |
1278 | ||
9e1ba4cc OW |
1279 | void qmp_migrate_set_cache_size(int64_t value, Error **errp) |
1280 | { | |
1281 | MigrationState *s = migrate_get_current(); | |
c91e681a | 1282 | int64_t new_size; |
9e1ba4cc OW |
1283 | |
1284 | /* Check for truncation */ | |
1285 | if (value != (size_t)value) { | |
c6bd8c70 MA |
1286 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1287 | "exceeding address space"); | |
9e1ba4cc OW |
1288 | return; |
1289 | } | |
1290 | ||
a5615b14 OW |
1291 | /* Cache should not be larger than guest ram size */ |
1292 | if (value > ram_bytes_total()) { | |
c6bd8c70 MA |
1293 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1294 | "exceeds guest ram size "); | |
a5615b14 OW |
1295 | return; |
1296 | } | |
1297 | ||
c91e681a OW |
1298 | new_size = xbzrle_cache_resize(value); |
1299 | if (new_size < 0) { | |
c6bd8c70 MA |
1300 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1301 | "is smaller than page size"); | |
c91e681a OW |
1302 | return; |
1303 | } | |
1304 | ||
1305 | s->xbzrle_cache_size = new_size; | |
9e1ba4cc OW |
1306 | } |
1307 | ||
1308 | int64_t qmp_query_migrate_cache_size(Error **errp) | |
1309 | { | |
1310 | return migrate_xbzrle_cache_size(); | |
1311 | } | |
1312 | ||
3dc85383 | 1313 | void qmp_migrate_set_speed(int64_t value, Error **errp) |
cab30143 | 1314 | { |
2ff30257 AA |
1315 | MigrationParameters p = { |
1316 | .has_max_bandwidth = true, | |
1317 | .max_bandwidth = value, | |
1318 | }; | |
cab30143 | 1319 | |
2ff30257 | 1320 | qmp_migrate_set_parameters(&p, errp); |
cab30143 JQ |
1321 | } |
1322 | ||
4f0a993b | 1323 | void qmp_migrate_set_downtime(double value, Error **errp) |
cab30143 | 1324 | { |
87c9cc1c DHB |
1325 | if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) { |
1326 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " | |
1327 | "the range of 0 to %d seconds", | |
1328 | MAX_MIGRATE_DOWNTIME_SECONDS); | |
1329 | return; | |
1330 | } | |
1331 | ||
2ff30257 AA |
1332 | value *= 1000; /* Convert to milliseconds */ |
1333 | value = MAX(0, MIN(INT64_MAX, value)); | |
1334 | ||
1335 | MigrationParameters p = { | |
1336 | .has_downtime_limit = true, | |
1337 | .downtime_limit = value, | |
1338 | }; | |
1339 | ||
1340 | qmp_migrate_set_parameters(&p, errp); | |
99a0db9b | 1341 | } |
17ad9b35 | 1342 | |
53f09a10 PB |
1343 | bool migrate_release_ram(void) |
1344 | { | |
1345 | MigrationState *s; | |
1346 | ||
1347 | s = migrate_get_current(); | |
1348 | ||
1349 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; | |
1350 | } | |
1351 | ||
53dd370c DDAG |
1352 | bool migrate_postcopy_ram(void) |
1353 | { | |
1354 | MigrationState *s; | |
1355 | ||
1356 | s = migrate_get_current(); | |
1357 | ||
32c3db5b | 1358 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
53dd370c DDAG |
1359 | } |
1360 | ||
bde1e2ec CV |
1361 | bool migrate_auto_converge(void) |
1362 | { | |
1363 | MigrationState *s; | |
1364 | ||
1365 | s = migrate_get_current(); | |
1366 | ||
1367 | return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; | |
1368 | } | |
1369 | ||
323004a3 PL |
1370 | bool migrate_zero_blocks(void) |
1371 | { | |
1372 | MigrationState *s; | |
1373 | ||
1374 | s = migrate_get_current(); | |
1375 | ||
1376 | return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; | |
1377 | } | |
1378 | ||
8706d2d5 LL |
1379 | bool migrate_use_compression(void) |
1380 | { | |
dde4e694 LL |
1381 | MigrationState *s; |
1382 | ||
1383 | s = migrate_get_current(); | |
1384 | ||
1385 | return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; | |
8706d2d5 LL |
1386 | } |
1387 | ||
1388 | int migrate_compress_level(void) | |
1389 | { | |
1390 | MigrationState *s; | |
1391 | ||
1392 | s = migrate_get_current(); | |
1393 | ||
2594f56d | 1394 | return s->parameters.compress_level; |
8706d2d5 LL |
1395 | } |
1396 | ||
1397 | int migrate_compress_threads(void) | |
1398 | { | |
1399 | MigrationState *s; | |
1400 | ||
1401 | s = migrate_get_current(); | |
1402 | ||
2594f56d | 1403 | return s->parameters.compress_threads; |
8706d2d5 LL |
1404 | } |
1405 | ||
3fcb38c2 LL |
1406 | int migrate_decompress_threads(void) |
1407 | { | |
1408 | MigrationState *s; | |
1409 | ||
1410 | s = migrate_get_current(); | |
1411 | ||
2594f56d | 1412 | return s->parameters.decompress_threads; |
3fcb38c2 LL |
1413 | } |
1414 | ||
b05dc723 JQ |
1415 | bool migrate_use_events(void) |
1416 | { | |
1417 | MigrationState *s; | |
1418 | ||
1419 | s = migrate_get_current(); | |
1420 | ||
1421 | return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; | |
1422 | } | |
1423 | ||
17ad9b35 OW |
1424 | int migrate_use_xbzrle(void) |
1425 | { | |
1426 | MigrationState *s; | |
1427 | ||
1428 | s = migrate_get_current(); | |
1429 | ||
1430 | return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; | |
1431 | } | |
1432 | ||
1433 | int64_t migrate_xbzrle_cache_size(void) | |
1434 | { | |
1435 | MigrationState *s; | |
1436 | ||
1437 | s = migrate_get_current(); | |
1438 | ||
1439 | return s->xbzrle_cache_size; | |
1440 | } | |
0d82d0e8 | 1441 | |
70b20477 DDAG |
1442 | /* migration thread support */ |
1443 | /* | |
1444 | * Something bad happened to the RP stream, mark an error | |
1445 | * The caller shall print or trace something to indicate why | |
1446 | */ | |
1447 | static void mark_source_rp_bad(MigrationState *s) | |
1448 | { | |
1449 | s->rp_state.error = true; | |
1450 | } | |
1451 | ||
1452 | static struct rp_cmd_args { | |
1453 | ssize_t len; /* -1 = variable */ | |
1454 | const char *name; | |
1455 | } rp_cmd_args[] = { | |
1456 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
1457 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
1458 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
1459 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
1460 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
70b20477 DDAG |
1461 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
1462 | }; | |
1463 | ||
1e2d90eb DDAG |
1464 | /* |
1465 | * Process a request for pages received on the return path, | |
1466 | * We're allowed to send more than requested (e.g. to round to our page size) | |
1467 | * and we don't need to send pages that have already been sent. | |
1468 | */ | |
1469 | static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
1470 | ram_addr_t start, size_t len) | |
1471 | { | |
6c595cde DDAG |
1472 | long our_host_ps = getpagesize(); |
1473 | ||
1e2d90eb | 1474 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
1475 | |
1476 | /* | |
1477 | * Since we currently insist on matching page sizes, just sanity check | |
1478 | * we're being asked for whole host pages. | |
1479 | */ | |
1480 | if (start & (our_host_ps-1) || | |
1481 | (len & (our_host_ps-1))) { | |
1482 | error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT | |
1483 | " len: %zd", __func__, start, len); | |
1484 | mark_source_rp_bad(ms); | |
1485 | return; | |
1486 | } | |
1487 | ||
1488 | if (ram_save_queue_pages(ms, rbname, start, len)) { | |
1489 | mark_source_rp_bad(ms); | |
1490 | } | |
1e2d90eb DDAG |
1491 | } |
1492 | ||
70b20477 DDAG |
1493 | /* |
1494 | * Handles messages sent on the return path towards the source VM | |
1495 | * | |
1496 | */ | |
1497 | static void *source_return_path_thread(void *opaque) | |
1498 | { | |
1499 | MigrationState *ms = opaque; | |
1500 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
1501 | uint16_t header_len, header_type; | |
568b01ca | 1502 | uint8_t buf[512]; |
70b20477 | 1503 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
1504 | ram_addr_t start = 0; /* =0 to silence warning */ |
1505 | size_t len = 0, expected_len; | |
70b20477 DDAG |
1506 | int res; |
1507 | ||
1508 | trace_source_return_path_thread_entry(); | |
1509 | while (!ms->rp_state.error && !qemu_file_get_error(rp) && | |
1510 | migration_is_setup_or_active(ms->state)) { | |
1511 | trace_source_return_path_thread_loop_top(); | |
1512 | header_type = qemu_get_be16(rp); | |
1513 | header_len = qemu_get_be16(rp); | |
1514 | ||
1515 | if (header_type >= MIG_RP_MSG_MAX || | |
1516 | header_type == MIG_RP_MSG_INVALID) { | |
1517 | error_report("RP: Received invalid message 0x%04x length 0x%04x", | |
1518 | header_type, header_len); | |
1519 | mark_source_rp_bad(ms); | |
1520 | goto out; | |
1521 | } | |
1522 | ||
1523 | if ((rp_cmd_args[header_type].len != -1 && | |
1524 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 1525 | header_len > sizeof(buf)) { |
70b20477 DDAG |
1526 | error_report("RP: Received '%s' message (0x%04x) with" |
1527 | "incorrect length %d expecting %zu", | |
1528 | rp_cmd_args[header_type].name, header_type, header_len, | |
1529 | (size_t)rp_cmd_args[header_type].len); | |
1530 | mark_source_rp_bad(ms); | |
1531 | goto out; | |
1532 | } | |
1533 | ||
1534 | /* We know we've got a valid header by this point */ | |
1535 | res = qemu_get_buffer(rp, buf, header_len); | |
1536 | if (res != header_len) { | |
1537 | error_report("RP: Failed reading data for message 0x%04x" | |
1538 | " read %d expected %d", | |
1539 | header_type, res, header_len); | |
1540 | mark_source_rp_bad(ms); | |
1541 | goto out; | |
1542 | } | |
1543 | ||
1544 | /* OK, we have the message and the data */ | |
1545 | switch (header_type) { | |
1546 | case MIG_RP_MSG_SHUT: | |
4d885131 | 1547 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
1548 | trace_source_return_path_thread_shut(sibling_error); |
1549 | if (sibling_error) { | |
1550 | error_report("RP: Sibling indicated error %d", sibling_error); | |
1551 | mark_source_rp_bad(ms); | |
1552 | } | |
1553 | /* | |
1554 | * We'll let the main thread deal with closing the RP | |
1555 | * we could do a shutdown(2) on it, but we're the only user | |
1556 | * anyway, so there's nothing gained. | |
1557 | */ | |
1558 | goto out; | |
1559 | ||
1560 | case MIG_RP_MSG_PONG: | |
4d885131 | 1561 | tmp32 = ldl_be_p(buf); |
70b20477 DDAG |
1562 | trace_source_return_path_thread_pong(tmp32); |
1563 | break; | |
1564 | ||
1e2d90eb | 1565 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
1566 | start = ldq_be_p(buf); |
1567 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1568 | migrate_handle_rp_req_pages(ms, NULL, start, len); |
1569 | break; | |
1570 | ||
1571 | case MIG_RP_MSG_REQ_PAGES_ID: | |
1572 | expected_len = 12 + 1; /* header + termination */ | |
1573 | ||
1574 | if (header_len >= expected_len) { | |
4d885131 PM |
1575 | start = ldq_be_p(buf); |
1576 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1577 | /* Now we expect an idstr */ |
1578 | tmp32 = buf[12]; /* Length of the following idstr */ | |
1579 | buf[13 + tmp32] = '\0'; | |
1580 | expected_len += tmp32; | |
1581 | } | |
1582 | if (header_len != expected_len) { | |
1583 | error_report("RP: Req_Page_id with length %d expecting %zd", | |
1584 | header_len, expected_len); | |
1585 | mark_source_rp_bad(ms); | |
1586 | goto out; | |
1587 | } | |
1588 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); | |
1589 | break; | |
1590 | ||
70b20477 DDAG |
1591 | default: |
1592 | break; | |
1593 | } | |
1594 | } | |
5df5416e | 1595 | if (qemu_file_get_error(rp)) { |
70b20477 DDAG |
1596 | trace_source_return_path_thread_bad_end(); |
1597 | mark_source_rp_bad(ms); | |
1598 | } | |
1599 | ||
1600 | trace_source_return_path_thread_end(); | |
1601 | out: | |
1602 | ms->rp_state.from_dst_file = NULL; | |
1603 | qemu_fclose(rp); | |
1604 | return NULL; | |
1605 | } | |
1606 | ||
70b20477 DDAG |
1607 | static int open_return_path_on_source(MigrationState *ms) |
1608 | { | |
1609 | ||
89a02a9f | 1610 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
1611 | if (!ms->rp_state.from_dst_file) { |
1612 | return -1; | |
1613 | } | |
1614 | ||
1615 | trace_open_return_path_on_source(); | |
1616 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", | |
1617 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
1618 | ||
1619 | trace_open_return_path_on_source_continue(); | |
1620 | ||
1621 | return 0; | |
1622 | } | |
1623 | ||
70b20477 DDAG |
1624 | /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ |
1625 | static int await_return_path_close_on_source(MigrationState *ms) | |
1626 | { | |
1627 | /* | |
1628 | * If this is a normal exit then the destination will send a SHUT and the | |
1629 | * rp_thread will exit, however if there's an error we need to cause | |
1630 | * it to exit. | |
1631 | */ | |
89a02a9f | 1632 | if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { |
70b20477 DDAG |
1633 | /* |
1634 | * shutdown(2), if we have it, will cause it to unblock if it's stuck | |
1635 | * waiting for the destination. | |
1636 | */ | |
1637 | qemu_file_shutdown(ms->rp_state.from_dst_file); | |
1638 | mark_source_rp_bad(ms); | |
1639 | } | |
1640 | trace_await_return_path_close_on_source_joining(); | |
1641 | qemu_thread_join(&ms->rp_state.rp_thread); | |
1642 | trace_await_return_path_close_on_source_close(); | |
1643 | return ms->rp_state.error; | |
1644 | } | |
1645 | ||
1d34e4bf DDAG |
1646 | /* |
1647 | * Switch from normal iteration to postcopy | |
1648 | * Returns non-0 on error | |
1649 | */ | |
1650 | static int postcopy_start(MigrationState *ms, bool *old_vm_running) | |
1651 | { | |
1652 | int ret; | |
61b67d47 DB |
1653 | QIOChannelBuffer *bioc; |
1654 | QEMUFile *fb; | |
1d34e4bf | 1655 | int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
ef8d6488 | 1656 | bool restart_block = false; |
48781e5b | 1657 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, |
1d34e4bf DDAG |
1658 | MIGRATION_STATUS_POSTCOPY_ACTIVE); |
1659 | ||
1660 | trace_postcopy_start(); | |
1661 | qemu_mutex_lock_iothread(); | |
1662 | trace_postcopy_start_set_run(); | |
1663 | ||
1664 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); | |
1665 | *old_vm_running = runstate_is_running(); | |
1666 | global_state_store(); | |
1667 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
76b1c7fe KW |
1668 | if (ret < 0) { |
1669 | goto fail; | |
1670 | } | |
1d34e4bf | 1671 | |
76b1c7fe | 1672 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
1673 | if (ret < 0) { |
1674 | goto fail; | |
1675 | } | |
ef8d6488 | 1676 | restart_block = true; |
1d34e4bf | 1677 | |
1c0d249d DDAG |
1678 | /* |
1679 | * Cause any non-postcopiable, but iterative devices to | |
1680 | * send out their final data. | |
1681 | */ | |
89a02a9f | 1682 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true); |
1c0d249d | 1683 | |
1d34e4bf DDAG |
1684 | /* |
1685 | * in Finish migrate and with the io-lock held everything should | |
1686 | * be quiet, but we've potentially still got dirty pages and we | |
1687 | * need to tell the destination to throw any pages it's already received | |
1688 | * that are dirty | |
1689 | */ | |
1690 | if (ram_postcopy_send_discard_bitmap(ms)) { | |
1691 | error_report("postcopy send discard bitmap failed"); | |
1692 | goto fail; | |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * send rest of state - note things that are doing postcopy | |
1697 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
1698 | * wrap their state up here | |
1699 | */ | |
89a02a9f | 1700 | qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); |
1d34e4bf | 1701 | /* Ping just for debugging, helps line traces up */ |
89a02a9f | 1702 | qemu_savevm_send_ping(ms->to_dst_file, 2); |
1d34e4bf DDAG |
1703 | |
1704 | /* | |
1705 | * While loading the device state we may trigger page transfer | |
1706 | * requests and the fd must be free to process those, and thus | |
1707 | * the destination must read the whole device state off the fd before | |
1708 | * it starts processing it. Unfortunately the ad-hoc migration format | |
1709 | * doesn't allow the destination to know the size to read without fully | |
1710 | * parsing it through each devices load-state code (especially the open | |
1711 | * coded devices that use get/put). | |
1712 | * So we wrap the device state up in a package with a length at the start; | |
1713 | * to do this we use a qemu_buf to hold the whole of the device state. | |
1714 | */ | |
61b67d47 | 1715 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 1716 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
61b67d47 DB |
1717 | fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc)); |
1718 | object_unref(OBJECT(bioc)); | |
1d34e4bf | 1719 | |
c76201ab DDAG |
1720 | /* |
1721 | * Make sure the receiver can get incoming pages before we send the rest | |
1722 | * of the state | |
1723 | */ | |
1724 | qemu_savevm_send_postcopy_listen(fb); | |
1725 | ||
1c0d249d | 1726 | qemu_savevm_state_complete_precopy(fb, false); |
1d34e4bf DDAG |
1727 | qemu_savevm_send_ping(fb, 3); |
1728 | ||
1729 | qemu_savevm_send_postcopy_run(fb); | |
1730 | ||
1731 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 1732 | |
ef8d6488 DDAG |
1733 | /* Last point of recovery; as soon as we send the package the destination |
1734 | * can open devices and potentially start running. | |
1735 | * Lets just check again we've not got any errors. | |
1736 | */ | |
1737 | ret = qemu_file_get_error(ms->to_dst_file); | |
1738 | if (ret) { | |
1739 | error_report("postcopy_start: Migration stream errored (pre package)"); | |
1740 | goto fail_closefb; | |
1741 | } | |
1742 | ||
1743 | restart_block = false; | |
1744 | ||
1d34e4bf | 1745 | /* Now send that blob */ |
61b67d47 | 1746 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
1747 | goto fail_closefb; |
1748 | } | |
1749 | qemu_fclose(fb); | |
b82fc321 DDAG |
1750 | |
1751 | /* Send a notify to give a chance for anything that needs to happen | |
1752 | * at the transition to postcopy and after the device state; in particular | |
1753 | * spice needs to trigger a transition now | |
1754 | */ | |
1755 | ms->postcopy_after_devices = true; | |
1756 | notifier_list_notify(&migration_state_notifiers, ms); | |
1757 | ||
1d34e4bf DDAG |
1758 | ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; |
1759 | ||
1760 | qemu_mutex_unlock_iothread(); | |
1761 | ||
1762 | /* | |
1763 | * Although this ping is just for debug, it could potentially be | |
1764 | * used for getting a better measurement of downtime at the source. | |
1765 | */ | |
89a02a9f | 1766 | qemu_savevm_send_ping(ms->to_dst_file, 4); |
1d34e4bf | 1767 | |
ced1c616 PB |
1768 | if (migrate_release_ram()) { |
1769 | ram_postcopy_migrated_memory_release(ms); | |
1770 | } | |
1771 | ||
89a02a9f | 1772 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf DDAG |
1773 | if (ret) { |
1774 | error_report("postcopy_start: Migration stream errored"); | |
48781e5b | 1775 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf DDAG |
1776 | MIGRATION_STATUS_FAILED); |
1777 | } | |
1778 | ||
1779 | return ret; | |
1780 | ||
1781 | fail_closefb: | |
1782 | qemu_fclose(fb); | |
1783 | fail: | |
48781e5b | 1784 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 1785 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
1786 | if (restart_block) { |
1787 | /* A failure happened early enough that we know the destination hasn't | |
1788 | * accessed block devices, so we're safe to recover. | |
1789 | */ | |
1790 | Error *local_err = NULL; | |
1791 | ||
1792 | bdrv_invalidate_cache_all(&local_err); | |
1793 | if (local_err) { | |
1794 | error_report_err(local_err); | |
1795 | } | |
1796 | } | |
1d34e4bf DDAG |
1797 | qemu_mutex_unlock_iothread(); |
1798 | return -1; | |
1799 | } | |
1800 | ||
09f6c85e DDAG |
1801 | /** |
1802 | * migration_completion: Used by migration_thread when there's not much left. | |
1803 | * The caller 'breaks' the loop when this returns. | |
1804 | * | |
1805 | * @s: Current migration state | |
36f48567 | 1806 | * @current_active_state: The migration state we expect to be in |
09f6c85e DDAG |
1807 | * @*old_vm_running: Pointer to old_vm_running flag |
1808 | * @*start_time: Pointer to time to update | |
1809 | */ | |
36f48567 DDAG |
1810 | static void migration_completion(MigrationState *s, int current_active_state, |
1811 | bool *old_vm_running, | |
09f6c85e DDAG |
1812 | int64_t *start_time) |
1813 | { | |
1814 | int ret; | |
1815 | ||
b10ac0c4 DDAG |
1816 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
1817 | qemu_mutex_lock_iothread(); | |
1818 | *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
1819 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); | |
1820 | *old_vm_running = runstate_is_running(); | |
1821 | ret = global_state_store(); | |
1822 | ||
1823 | if (!ret) { | |
1824 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
0b827d5e HZ |
1825 | /* |
1826 | * Don't mark the image with BDRV_O_INACTIVE flag if | |
1827 | * we will go into COLO stage later. | |
1828 | */ | |
1829 | if (ret >= 0 && !migrate_colo_enabled()) { | |
76b1c7fe KW |
1830 | ret = bdrv_inactivate_all(); |
1831 | } | |
b10ac0c4 | 1832 | if (ret >= 0) { |
89a02a9f HZ |
1833 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); |
1834 | qemu_savevm_state_complete_precopy(s->to_dst_file, false); | |
1d2acc31 | 1835 | s->block_inactive = true; |
b10ac0c4 DDAG |
1836 | } |
1837 | } | |
1838 | qemu_mutex_unlock_iothread(); | |
09f6c85e | 1839 | |
b10ac0c4 DDAG |
1840 | if (ret < 0) { |
1841 | goto fail; | |
09f6c85e | 1842 | } |
b10ac0c4 DDAG |
1843 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { |
1844 | trace_migration_completion_postcopy_end(); | |
1845 | ||
89a02a9f | 1846 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
b10ac0c4 | 1847 | trace_migration_completion_postcopy_end_after_complete(); |
09f6c85e | 1848 | } |
09f6c85e | 1849 | |
b10ac0c4 DDAG |
1850 | /* |
1851 | * If rp was opened we must clean up the thread before | |
1852 | * cleaning everything else up (since if there are no failures | |
1853 | * it will wait for the destination to send it's status in | |
1854 | * a SHUT command). | |
1855 | * Postcopy opens rp if enabled (even if it's not avtivated) | |
1856 | */ | |
1857 | if (migrate_postcopy_ram()) { | |
1858 | int rp_error; | |
1859 | trace_migration_completion_postcopy_end_before_rp(); | |
1860 | rp_error = await_return_path_close_on_source(s); | |
1861 | trace_migration_completion_postcopy_end_after_rp(rp_error); | |
1862 | if (rp_error) { | |
fe904ea8 | 1863 | goto fail_invalidate; |
b10ac0c4 | 1864 | } |
09f6c85e DDAG |
1865 | } |
1866 | ||
89a02a9f | 1867 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 1868 | trace_migration_completion_file_err(); |
fe904ea8 | 1869 | goto fail_invalidate; |
09f6c85e DDAG |
1870 | } |
1871 | ||
0b827d5e HZ |
1872 | if (!migrate_colo_enabled()) { |
1873 | migrate_set_state(&s->state, current_active_state, | |
1874 | MIGRATION_STATUS_COMPLETED); | |
1875 | } | |
1876 | ||
09f6c85e DDAG |
1877 | return; |
1878 | ||
fe904ea8 GK |
1879 | fail_invalidate: |
1880 | /* If not doing postcopy, vm_start() will be called: let's regain | |
1881 | * control on images. | |
1882 | */ | |
1883 | if (s->state == MIGRATION_STATUS_ACTIVE) { | |
1884 | Error *local_err = NULL; | |
1885 | ||
1d2acc31 | 1886 | qemu_mutex_lock_iothread(); |
fe904ea8 GK |
1887 | bdrv_invalidate_cache_all(&local_err); |
1888 | if (local_err) { | |
1889 | error_report_err(local_err); | |
1d2acc31 HZ |
1890 | } else { |
1891 | s->block_inactive = false; | |
fe904ea8 | 1892 | } |
1d2acc31 | 1893 | qemu_mutex_unlock_iothread(); |
fe904ea8 GK |
1894 | } |
1895 | ||
09f6c85e | 1896 | fail: |
48781e5b HZ |
1897 | migrate_set_state(&s->state, current_active_state, |
1898 | MIGRATION_STATUS_FAILED); | |
09f6c85e DDAG |
1899 | } |
1900 | ||
35a6ed4f HZ |
1901 | bool migrate_colo_enabled(void) |
1902 | { | |
1903 | MigrationState *s = migrate_get_current(); | |
1904 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; | |
1905 | } | |
1906 | ||
70b20477 DDAG |
1907 | /* |
1908 | * Master migration thread on the source VM. | |
1909 | * It drives the migration and pumps the data down the outgoing channel. | |
1910 | */ | |
5f496a1b | 1911 | static void *migration_thread(void *opaque) |
0d82d0e8 | 1912 | { |
9848a404 | 1913 | MigrationState *s = opaque; |
1d34e4bf | 1914 | /* Used by the bandwidth calcs, updated later */ |
bc72ad67 AB |
1915 | int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1916 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
be7172e2 | 1917 | int64_t initial_bytes = 0; |
0d82d0e8 | 1918 | int64_t max_size = 0; |
a3fa1d78 | 1919 | int64_t start_time = initial_time; |
94f5a437 | 1920 | int64_t end_time; |
a3fa1d78 | 1921 | bool old_vm_running = false; |
1d34e4bf DDAG |
1922 | bool entered_postcopy = false; |
1923 | /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */ | |
1924 | enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE; | |
0b827d5e | 1925 | bool enable_colo = migrate_colo_enabled(); |
76f5933a | 1926 | |
ab28bd23 PB |
1927 | rcu_register_thread(); |
1928 | ||
89a02a9f | 1929 | qemu_savevm_state_header(s->to_dst_file); |
1d34e4bf DDAG |
1930 | |
1931 | if (migrate_postcopy_ram()) { | |
1932 | /* Now tell the dest that it should open its end so it can reply */ | |
89a02a9f | 1933 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
1934 | |
1935 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 1936 | qemu_savevm_send_ping(s->to_dst_file, 1); |
1d34e4bf DDAG |
1937 | |
1938 | /* | |
1939 | * Tell the destination that we *might* want to do postcopy later; | |
1940 | * if the other end can't do postcopy it should fail now, nice and | |
1941 | * early. | |
1942 | */ | |
89a02a9f | 1943 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
1944 | } |
1945 | ||
89a02a9f | 1946 | qemu_savevm_state_begin(s->to_dst_file, &s->params); |
0d82d0e8 | 1947 | |
bc72ad67 | 1948 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
1d34e4bf | 1949 | current_active_state = MIGRATION_STATUS_ACTIVE; |
48781e5b HZ |
1950 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1951 | MIGRATION_STATUS_ACTIVE); | |
29ae8a41 | 1952 | |
9ec055ae DDAG |
1953 | trace_migration_thread_setup_complete(); |
1954 | ||
1955 | while (s->state == MIGRATION_STATUS_ACTIVE || | |
1956 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
a3e879cd | 1957 | int64_t current_time; |
c369f40d | 1958 | uint64_t pending_size; |
0d82d0e8 | 1959 | |
89a02a9f | 1960 | if (!qemu_file_rate_limit(s->to_dst_file)) { |
c31b098f DDAG |
1961 | uint64_t pend_post, pend_nonpost; |
1962 | ||
89a02a9f | 1963 | qemu_savevm_state_pending(s->to_dst_file, max_size, &pend_nonpost, |
c31b098f DDAG |
1964 | &pend_post); |
1965 | pending_size = pend_nonpost + pend_post; | |
1966 | trace_migrate_pending(pending_size, max_size, | |
1967 | pend_post, pend_nonpost); | |
b22ff1fb | 1968 | if (pending_size && pending_size >= max_size) { |
1d34e4bf DDAG |
1969 | /* Still a significant amount to transfer */ |
1970 | ||
1d34e4bf DDAG |
1971 | if (migrate_postcopy_ram() && |
1972 | s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE && | |
1973 | pend_nonpost <= max_size && | |
1974 | atomic_read(&s->start_postcopy)) { | |
1975 | ||
1976 | if (!postcopy_start(s, &old_vm_running)) { | |
1977 | current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE; | |
1978 | entered_postcopy = true; | |
1979 | } | |
1980 | ||
1981 | continue; | |
1982 | } | |
1983 | /* Just another iteration step */ | |
89a02a9f | 1984 | qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy); |
c369f40d | 1985 | } else { |
09f6c85e | 1986 | trace_migration_thread_low_pending(pending_size); |
1d34e4bf | 1987 | migration_completion(s, current_active_state, |
36f48567 | 1988 | &old_vm_running, &start_time); |
09f6c85e | 1989 | break; |
c369f40d JQ |
1990 | } |
1991 | } | |
f4410a5d | 1992 | |
89a02a9f | 1993 | if (qemu_file_get_error(s->to_dst_file)) { |
48781e5b HZ |
1994 | migrate_set_state(&s->state, current_active_state, |
1995 | MIGRATION_STATUS_FAILED); | |
1d34e4bf | 1996 | trace_migration_thread_file_err(); |
fd45ee2c PB |
1997 | break; |
1998 | } | |
bc72ad67 | 1999 | current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0d82d0e8 | 2000 | if (current_time >= initial_time + BUFFER_DELAY) { |
89a02a9f HZ |
2001 | uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) - |
2002 | initial_bytes; | |
77417f10 | 2003 | uint64_t time_spent = current_time - initial_time; |
a694ee34 | 2004 | double bandwidth = (double)transferred_bytes / time_spent; |
2ff30257 | 2005 | max_size = bandwidth * s->parameters.downtime_limit; |
0d82d0e8 | 2006 | |
5b648de0 WY |
2007 | s->mbps = (((double) transferred_bytes * 8.0) / |
2008 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
7e114f8c | 2009 | |
9013dca5 AK |
2010 | trace_migrate_transferred(transferred_bytes, time_spent, |
2011 | bandwidth, max_size); | |
90f8ae72 JQ |
2012 | /* if we haven't sent anything, we don't want to recalculate |
2013 | 10000 is a small enough number for our purposes */ | |
2014 | if (s->dirty_bytes_rate && transferred_bytes > 10000) { | |
2015 | s->expected_downtime = s->dirty_bytes_rate / bandwidth; | |
2016 | } | |
0d82d0e8 | 2017 | |
89a02a9f | 2018 | qemu_file_reset_rate_limit(s->to_dst_file); |
0d82d0e8 | 2019 | initial_time = current_time; |
89a02a9f | 2020 | initial_bytes = qemu_ftell(s->to_dst_file); |
0d82d0e8 | 2021 | } |
89a02a9f | 2022 | if (qemu_file_rate_limit(s->to_dst_file)) { |
0d82d0e8 JQ |
2023 | /* usleep expects microseconds */ |
2024 | g_usleep((initial_time + BUFFER_DELAY - current_time)*1000); | |
2025 | } | |
a3fa1d78 PB |
2026 | } |
2027 | ||
1d34e4bf | 2028 | trace_migration_thread_after_loop(); |
070afca2 JH |
2029 | /* If we enabled cpu throttling for auto-converge, turn it off. */ |
2030 | cpu_throttle_stop(); | |
94f5a437 | 2031 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
070afca2 | 2032 | |
f4410a5d | 2033 | qemu_mutex_lock_iothread(); |
0b827d5e HZ |
2034 | /* |
2035 | * The resource has been allocated by migration will be reused in COLO | |
2036 | * process, so don't release them. | |
2037 | */ | |
2038 | if (!enable_colo) { | |
2039 | qemu_savevm_state_cleanup(); | |
2040 | } | |
31194731 | 2041 | if (s->state == MIGRATION_STATUS_COMPLETED) { |
89a02a9f | 2042 | uint64_t transferred_bytes = qemu_ftell(s->to_dst_file); |
a3fa1d78 | 2043 | s->total_time = end_time - s->total_time; |
1d34e4bf DDAG |
2044 | if (!entered_postcopy) { |
2045 | s->downtime = end_time - start_time; | |
2046 | } | |
d6ed7312 PL |
2047 | if (s->total_time) { |
2048 | s->mbps = (((double) transferred_bytes * 8.0) / | |
2049 | ((double) s->total_time)) / 1000; | |
2050 | } | |
a3fa1d78 PB |
2051 | runstate_set(RUN_STATE_POSTMIGRATE); |
2052 | } else { | |
0b827d5e HZ |
2053 | if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) { |
2054 | migrate_start_colo_process(s); | |
2055 | qemu_savevm_state_cleanup(); | |
2056 | /* | |
2057 | * Fixme: we will run VM in COLO no matter its old running state. | |
2058 | * After exited COLO, we will keep running. | |
2059 | */ | |
2060 | old_vm_running = true; | |
2061 | } | |
1d34e4bf | 2062 | if (old_vm_running && !entered_postcopy) { |
a3fa1d78 | 2063 | vm_start(); |
42da5550 DDAG |
2064 | } else { |
2065 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
2066 | runstate_set(RUN_STATE_POSTMIGRATE); | |
2067 | } | |
dba433c0 | 2068 | } |
0d82d0e8 | 2069 | } |
bb1fadc4 | 2070 | qemu_bh_schedule(s->cleanup_bh); |
dba433c0 | 2071 | qemu_mutex_unlock_iothread(); |
f4410a5d | 2072 | |
ab28bd23 | 2073 | rcu_unregister_thread(); |
0d82d0e8 JQ |
2074 | return NULL; |
2075 | } | |
2076 | ||
9848a404 | 2077 | void migrate_fd_connect(MigrationState *s) |
0d82d0e8 | 2078 | { |
2ff30257 | 2079 | s->expected_downtime = s->parameters.downtime_limit; |
bb1fadc4 | 2080 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s); |
0d82d0e8 | 2081 | |
9e4d2b98 | 2082 | qemu_file_set_blocking(s->to_dst_file, true); |
89a02a9f | 2083 | qemu_file_set_rate_limit(s->to_dst_file, |
2ff30257 | 2084 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); |
442773ce | 2085 | |
9287ac27 SH |
2086 | /* Notify before starting migration thread */ |
2087 | notifier_list_notify(&migration_state_notifiers, s); | |
2088 | ||
1d34e4bf DDAG |
2089 | /* |
2090 | * Open the return path; currently for postcopy but other things might | |
2091 | * also want it. | |
2092 | */ | |
2093 | if (migrate_postcopy_ram()) { | |
2094 | if (open_return_path_on_source(s)) { | |
2095 | error_report("Unable to open return-path for postcopy"); | |
48781e5b | 2096 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1d34e4bf DDAG |
2097 | MIGRATION_STATUS_FAILED); |
2098 | migrate_fd_cleanup(s); | |
2099 | return; | |
2100 | } | |
2101 | } | |
2102 | ||
8706d2d5 | 2103 | migrate_compress_threads_create(); |
009fad7f | 2104 | qemu_thread_create(&s->thread, "live_migration", migration_thread, s, |
bb1fadc4 | 2105 | QEMU_THREAD_JOINABLE); |
1d34e4bf | 2106 | s->migration_thread_running = true; |
0d82d0e8 | 2107 | } |
093e3c42 DDAG |
2108 | |
2109 | PostcopyState postcopy_state_get(void) | |
2110 | { | |
2111 | return atomic_mb_read(&incoming_postcopy_state); | |
2112 | } | |
2113 | ||
2114 | /* Set the state and return the old state */ | |
2115 | PostcopyState postcopy_state_set(PostcopyState new_state) | |
2116 | { | |
2117 | return atomic_xchg(&incoming_postcopy_state, new_state); | |
2118 | } | |
2119 |