]>
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" |
795c40b8 | 20 | #include "migration/blocker.h" |
f4dbe1bf | 21 | #include "exec.h" |
7fcac4a2 | 22 | #include "fd.h" |
caf71f86 | 23 | #include "migration/migration.h" |
20a519a0 | 24 | #include "savevm.h" |
40014d81 | 25 | #include "qemu-file-channel.h" |
08a0aee1 | 26 | #include "qemu-file.h" |
987772d9 | 27 | #include "migration/vmstate.h" |
9c17d615 | 28 | #include "sysemu/sysemu.h" |
737e150e | 29 | #include "block/block.h" |
cc7a8ea7 | 30 | #include "qapi/qmp/qerror.h" |
6c595cde | 31 | #include "qapi/util.h" |
1de7afc9 | 32 | #include "qemu/sockets.h" |
ab28bd23 | 33 | #include "qemu/rcu.h" |
caf71f86 | 34 | #include "migration/block.h" |
be07b0ac | 35 | #include "postcopy-ram.h" |
766bd176 | 36 | #include "qemu/thread.h" |
791e7c82 | 37 | #include "qmp-commands.h" |
c09e5bb1 | 38 | #include "trace.h" |
598cd2bd | 39 | #include "qapi-event.h" |
070afca2 | 40 | #include "qom/cpu.h" |
6c595cde DDAG |
41 | #include "exec/memory.h" |
42 | #include "exec/address-spaces.h" | |
51180423 | 43 | #include "exec/target_page.h" |
61b67d47 | 44 | #include "io/channel-buffer.h" |
e1226365 | 45 | #include "io/channel-tls.h" |
35a6ed4f | 46 | #include "migration/colo.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 | ||
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 | 111 | if (!once) { |
4af245dc DB |
112 | current_migration.parameters.tls_creds = g_strdup(""); |
113 | current_migration.parameters.tls_hostname = g_strdup(""); | |
6c595cde DDAG |
114 | once = true; |
115 | } | |
17549e84 JQ |
116 | return ¤t_migration; |
117 | } | |
118 | ||
bca7856a DDAG |
119 | MigrationIncomingState *migration_incoming_get_current(void) |
120 | { | |
b4b076da JQ |
121 | static bool once; |
122 | static MigrationIncomingState mis_current; | |
bca7856a | 123 | |
b4b076da JQ |
124 | if (!once) { |
125 | mis_current.state = MIGRATION_STATUS_NONE; | |
126 | memset(&mis_current, 0, sizeof(MigrationIncomingState)); | |
b4b076da JQ |
127 | qemu_mutex_init(&mis_current.rp_mutex); |
128 | qemu_event_init(&mis_current.main_thread_load_event, false); | |
129 | once = true; | |
130 | } | |
131 | return &mis_current; | |
bca7856a DDAG |
132 | } |
133 | ||
134 | void migration_incoming_state_destroy(void) | |
135 | { | |
b4b076da JQ |
136 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
137 | ||
3482655b | 138 | if (mis->to_src_file) { |
660819b1 PX |
139 | /* Tell source that we are done */ |
140 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
141 | qemu_fclose(mis->to_src_file); |
142 | mis->to_src_file = NULL; | |
143 | } | |
144 | ||
660819b1 PX |
145 | if (mis->from_src_file) { |
146 | qemu_fclose(mis->from_src_file); | |
147 | mis->from_src_file = NULL; | |
148 | } | |
149 | ||
b4b076da | 150 | qemu_event_destroy(&mis->main_thread_load_event); |
bca7856a DDAG |
151 | } |
152 | ||
df4b1024 JQ |
153 | |
154 | typedef struct { | |
13d16814 | 155 | bool optional; |
df4b1024 JQ |
156 | uint32_t size; |
157 | uint8_t runstate[100]; | |
172c4356 JQ |
158 | RunState state; |
159 | bool received; | |
df4b1024 JQ |
160 | } GlobalState; |
161 | ||
162 | static GlobalState global_state; | |
163 | ||
560d027b | 164 | int global_state_store(void) |
df4b1024 JQ |
165 | { |
166 | if (!runstate_store((char *)global_state.runstate, | |
167 | sizeof(global_state.runstate))) { | |
168 | error_report("runstate name too big: %s", global_state.runstate); | |
169 | trace_migrate_state_too_big(); | |
170 | return -EINVAL; | |
171 | } | |
172 | return 0; | |
173 | } | |
174 | ||
c69adea4 AP |
175 | void global_state_store_running(void) |
176 | { | |
177 | const char *state = RunState_lookup[RUN_STATE_RUNNING]; | |
178 | strncpy((char *)global_state.runstate, | |
179 | state, sizeof(global_state.runstate)); | |
180 | } | |
181 | ||
172c4356 | 182 | static bool global_state_received(void) |
df4b1024 | 183 | { |
172c4356 JQ |
184 | return global_state.received; |
185 | } | |
186 | ||
187 | static RunState global_state_get_runstate(void) | |
188 | { | |
189 | return global_state.state; | |
df4b1024 JQ |
190 | } |
191 | ||
13d16814 JQ |
192 | void global_state_set_optional(void) |
193 | { | |
194 | global_state.optional = true; | |
195 | } | |
196 | ||
197 | static bool global_state_needed(void *opaque) | |
198 | { | |
199 | GlobalState *s = opaque; | |
200 | char *runstate = (char *)s->runstate; | |
201 | ||
202 | /* If it is not optional, it is mandatory */ | |
203 | ||
204 | if (s->optional == false) { | |
205 | return true; | |
206 | } | |
207 | ||
208 | /* If state is running or paused, it is not needed */ | |
209 | ||
210 | if (strcmp(runstate, "running") == 0 || | |
211 | strcmp(runstate, "paused") == 0) { | |
212 | return false; | |
213 | } | |
214 | ||
215 | /* for any other state it is needed */ | |
216 | return true; | |
217 | } | |
218 | ||
df4b1024 JQ |
219 | static int global_state_post_load(void *opaque, int version_id) |
220 | { | |
221 | GlobalState *s = opaque; | |
172c4356 JQ |
222 | Error *local_err = NULL; |
223 | int r; | |
df4b1024 JQ |
224 | char *runstate = (char *)s->runstate; |
225 | ||
172c4356 | 226 | s->received = true; |
df4b1024 JQ |
227 | trace_migrate_global_state_post_load(runstate); |
228 | ||
7fb1cf16 | 229 | r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE__MAX, |
df4b1024 JQ |
230 | -1, &local_err); |
231 | ||
172c4356 JQ |
232 | if (r == -1) { |
233 | if (local_err) { | |
234 | error_report_err(local_err); | |
df4b1024 | 235 | } |
172c4356 | 236 | return -EINVAL; |
df4b1024 | 237 | } |
172c4356 | 238 | s->state = r; |
df4b1024 | 239 | |
172c4356 | 240 | return 0; |
df4b1024 JQ |
241 | } |
242 | ||
243 | static void global_state_pre_save(void *opaque) | |
244 | { | |
245 | GlobalState *s = opaque; | |
246 | ||
247 | trace_migrate_global_state_pre_save((char *)s->runstate); | |
248 | s->size = strlen((char *)s->runstate) + 1; | |
249 | } | |
250 | ||
251 | static const VMStateDescription vmstate_globalstate = { | |
252 | .name = "globalstate", | |
253 | .version_id = 1, | |
254 | .minimum_version_id = 1, | |
255 | .post_load = global_state_post_load, | |
256 | .pre_save = global_state_pre_save, | |
13d16814 | 257 | .needed = global_state_needed, |
df4b1024 JQ |
258 | .fields = (VMStateField[]) { |
259 | VMSTATE_UINT32(size, GlobalState), | |
260 | VMSTATE_BUFFER(runstate, GlobalState), | |
261 | VMSTATE_END_OF_LIST() | |
262 | }, | |
263 | }; | |
264 | ||
265 | void register_global_state(void) | |
266 | { | |
267 | /* We would use it independently that we receive it */ | |
268 | strcpy((char *)&global_state.runstate, ""); | |
172c4356 | 269 | global_state.received = false; |
df4b1024 JQ |
270 | vmstate_register(NULL, 0, &vmstate_globalstate, &global_state); |
271 | } | |
272 | ||
b05dc723 JQ |
273 | static void migrate_generate_event(int new_state) |
274 | { | |
275 | if (migrate_use_events()) { | |
276 | qapi_event_send_migration(new_state, &error_abort); | |
b05dc723 JQ |
277 | } |
278 | } | |
279 | ||
adde220a DDAG |
280 | /* |
281 | * Called on -incoming with a defer: uri. | |
282 | * The migration can be started later after any parameters have been | |
283 | * changed. | |
284 | */ | |
285 | static void deferred_incoming_migration(Error **errp) | |
286 | { | |
287 | if (deferred_incoming) { | |
288 | error_setg(errp, "Incoming migration already deferred"); | |
289 | } | |
290 | deferred_incoming = true; | |
291 | } | |
292 | ||
1e2d90eb DDAG |
293 | /* Request a range of pages from the source VM at the given |
294 | * start address. | |
295 | * rbname: Name of the RAMBlock to request the page in, if NULL it's the same | |
296 | * as the last request (a name must have been given previously) | |
297 | * Start: Address offset within the RB | |
298 | * Len: Length in bytes required - must be a multiple of pagesize | |
299 | */ | |
300 | void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname, | |
301 | ram_addr_t start, size_t len) | |
302 | { | |
cb8d4c8f | 303 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb DDAG |
304 | size_t msglen = 12; /* start + len */ |
305 | ||
306 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
307 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
308 | ||
309 | if (rbname) { | |
310 | int rbname_len = strlen(rbname); | |
311 | assert(rbname_len < 256); | |
312 | ||
313 | bufc[msglen++] = rbname_len; | |
314 | memcpy(bufc + msglen, rbname, rbname_len); | |
315 | msglen += rbname_len; | |
316 | migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc); | |
317 | } else { | |
318 | migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc); | |
319 | } | |
320 | } | |
321 | ||
43eaae28 | 322 | void qemu_start_incoming_migration(const char *uri, Error **errp) |
5bb7910a | 323 | { |
34c9dd8e AL |
324 | const char *p; |
325 | ||
7cf1fe6d | 326 | qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort); |
adde220a DDAG |
327 | if (!strcmp(uri, "defer")) { |
328 | deferred_incoming_migration(errp); | |
329 | } else if (strstart(uri, "tcp:", &p)) { | |
43eaae28 | 330 | tcp_start_incoming_migration(p, errp); |
2da776db | 331 | #ifdef CONFIG_RDMA |
adde220a | 332 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
333 | rdma_start_incoming_migration(p, errp); |
334 | #endif | |
adde220a | 335 | } else if (strstart(uri, "exec:", &p)) { |
43eaae28 | 336 | exec_start_incoming_migration(p, errp); |
adde220a | 337 | } else if (strstart(uri, "unix:", &p)) { |
43eaae28 | 338 | unix_start_incoming_migration(p, errp); |
adde220a | 339 | } else if (strstart(uri, "fd:", &p)) { |
43eaae28 | 340 | fd_start_incoming_migration(p, errp); |
adde220a | 341 | } else { |
312fd5f2 | 342 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 343 | } |
5bb7910a AL |
344 | } |
345 | ||
0aa6aefc DL |
346 | static void process_incoming_migration_bh(void *opaque) |
347 | { | |
348 | Error *local_err = NULL; | |
349 | MigrationIncomingState *mis = opaque; | |
350 | ||
ace21a58 KW |
351 | /* Make sure all file formats flush their mutable metadata. |
352 | * If we get an error here, just don't restart the VM yet. */ | |
0aa6aefc | 353 | bdrv_invalidate_cache_all(&local_err); |
d35ff5e6 | 354 | if (local_err) { |
ace21a58 | 355 | error_report_err(local_err); |
d35ff5e6 KW |
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 | if (ret < 0) { |
93d7af6f HZ |
440 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, |
441 | MIGRATION_STATUS_FAILED); | |
db80face | 442 | error_report("load of migration failed: %s", strerror(-ret)); |
3fcb38c2 | 443 | migrate_decompress_threads_join(); |
4aead692 | 444 | exit(EXIT_FAILURE); |
511c0231 | 445 | } |
511c0231 | 446 | |
e8199e48 LV |
447 | free_xbzrle_decoded_buf(); |
448 | ||
0aa6aefc DL |
449 | mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); |
450 | qemu_bh_schedule(mis->bh); | |
511c0231 JQ |
451 | } |
452 | ||
22724f49 | 453 | void migration_fd_process_incoming(QEMUFile *f) |
82a4da79 | 454 | { |
0b8b8753 | 455 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f); |
82a4da79 | 456 | |
3fcb38c2 | 457 | migrate_decompress_threads_create(); |
06ad5135 | 458 | qemu_file_set_blocking(f, false); |
0b8b8753 | 459 | qemu_coroutine_enter(co); |
82a4da79 PB |
460 | } |
461 | ||
6decec93 DDAG |
462 | /* |
463 | * Send a message on the return channel back to the source | |
464 | * of the migration. | |
465 | */ | |
466 | void migrate_send_rp_message(MigrationIncomingState *mis, | |
467 | enum mig_rp_message_type message_type, | |
468 | uint16_t len, void *data) | |
469 | { | |
470 | trace_migrate_send_rp_message((int)message_type, len); | |
471 | qemu_mutex_lock(&mis->rp_mutex); | |
472 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); | |
473 | qemu_put_be16(mis->to_src_file, len); | |
474 | qemu_put_buffer(mis->to_src_file, data, len); | |
475 | qemu_fflush(mis->to_src_file); | |
476 | qemu_mutex_unlock(&mis->rp_mutex); | |
477 | } | |
478 | ||
479 | /* | |
480 | * Send a 'SHUT' message on the return channel with the given value | |
481 | * to indicate that we've finished with the RP. Non-0 value indicates | |
482 | * error. | |
483 | */ | |
484 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
485 | uint32_t value) | |
486 | { | |
487 | uint32_t buf; | |
488 | ||
489 | buf = cpu_to_be32(value); | |
490 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
491 | } | |
492 | ||
493 | /* | |
494 | * Send a 'PONG' message on the return channel with the given value | |
495 | * (normally in response to a 'PING') | |
496 | */ | |
497 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
498 | uint32_t value) | |
499 | { | |
500 | uint32_t buf; | |
501 | ||
502 | buf = cpu_to_be32(value); | |
503 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
504 | } | |
505 | ||
bbf6da32 OW |
506 | MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) |
507 | { | |
508 | MigrationCapabilityStatusList *head = NULL; | |
509 | MigrationCapabilityStatusList *caps; | |
510 | MigrationState *s = migrate_get_current(); | |
511 | int i; | |
512 | ||
387eedeb | 513 | caps = NULL; /* silence compiler warning */ |
7fb1cf16 | 514 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
ed1701c6 DDAG |
515 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
516 | if (i == MIGRATION_CAPABILITY_BLOCK) { | |
517 | continue; | |
518 | } | |
519 | #endif | |
35a6ed4f HZ |
520 | if (i == MIGRATION_CAPABILITY_X_COLO && !colo_supported()) { |
521 | continue; | |
522 | } | |
bbf6da32 OW |
523 | if (head == NULL) { |
524 | head = g_malloc0(sizeof(*caps)); | |
525 | caps = head; | |
526 | } else { | |
527 | caps->next = g_malloc0(sizeof(*caps)); | |
528 | caps = caps->next; | |
529 | } | |
530 | caps->value = | |
531 | g_malloc(sizeof(*caps->value)); | |
532 | caps->value->capability = i; | |
533 | caps->value->state = s->enabled_capabilities[i]; | |
534 | } | |
535 | ||
536 | return head; | |
537 | } | |
538 | ||
85de8323 LL |
539 | MigrationParameters *qmp_query_migrate_parameters(Error **errp) |
540 | { | |
541 | MigrationParameters *params; | |
542 | MigrationState *s = migrate_get_current(); | |
543 | ||
544 | params = g_malloc0(sizeof(*params)); | |
de63ab61 | 545 | params->has_compress_level = true; |
2594f56d | 546 | params->compress_level = s->parameters.compress_level; |
de63ab61 | 547 | params->has_compress_threads = true; |
2594f56d | 548 | params->compress_threads = s->parameters.compress_threads; |
de63ab61 | 549 | params->has_decompress_threads = true; |
2594f56d | 550 | params->decompress_threads = s->parameters.decompress_threads; |
de63ab61 | 551 | params->has_cpu_throttle_initial = true; |
2594f56d | 552 | params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; |
de63ab61 | 553 | params->has_cpu_throttle_increment = true; |
2594f56d | 554 | params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; |
de63ab61 | 555 | params->has_tls_creds = !!s->parameters.tls_creds; |
69ef1f36 | 556 | params->tls_creds = g_strdup(s->parameters.tls_creds); |
de63ab61 | 557 | params->has_tls_hostname = !!s->parameters.tls_hostname; |
69ef1f36 | 558 | params->tls_hostname = g_strdup(s->parameters.tls_hostname); |
2ff30257 AA |
559 | params->has_max_bandwidth = true; |
560 | params->max_bandwidth = s->parameters.max_bandwidth; | |
561 | params->has_downtime_limit = true; | |
562 | params->downtime_limit = s->parameters.downtime_limit; | |
fe39a4d4 | 563 | params->has_x_checkpoint_delay = true; |
68b53591 | 564 | params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; |
2833c59b JQ |
565 | params->has_block_incremental = true; |
566 | params->block_incremental = s->parameters.block_incremental; | |
85de8323 LL |
567 | |
568 | return params; | |
569 | } | |
570 | ||
f6844b99 DDAG |
571 | /* |
572 | * Return true if we're already in the middle of a migration | |
573 | * (i.e. any of the active or setup states) | |
574 | */ | |
575 | static bool migration_is_setup_or_active(int state) | |
576 | { | |
577 | switch (state) { | |
578 | case MIGRATION_STATUS_ACTIVE: | |
9ec055ae | 579 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
f6844b99 DDAG |
580 | case MIGRATION_STATUS_SETUP: |
581 | return true; | |
582 | ||
583 | default: | |
584 | return false; | |
585 | ||
586 | } | |
587 | } | |
588 | ||
f36d55af OW |
589 | static void get_xbzrle_cache_stats(MigrationInfo *info) |
590 | { | |
591 | if (migrate_use_xbzrle()) { | |
592 | info->has_xbzrle_cache = true; | |
593 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); | |
594 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
595 | info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred(); | |
596 | info->xbzrle_cache->pages = xbzrle_mig_pages_transferred(); | |
597 | info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss(); | |
8bc39233 | 598 | info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate(); |
f36d55af OW |
599 | info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow(); |
600 | } | |
601 | } | |
602 | ||
a22463a5 DDAG |
603 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
604 | { | |
605 | info->has_ram = true; | |
606 | info->ram = g_malloc0(sizeof(*info->ram)); | |
607 | info->ram->transferred = ram_bytes_transferred(); | |
608 | info->ram->total = ram_bytes_total(); | |
609 | info->ram->duplicate = dup_mig_pages_transferred(); | |
bedf53c1 JQ |
610 | /* legacy value. It is not used anymore */ |
611 | info->ram->skipped = 0; | |
a22463a5 | 612 | info->ram->normal = norm_mig_pages_transferred(); |
29cc3d8a | 613 | info->ram->normal_bytes = norm_mig_pages_transferred() * |
20afaed9 | 614 | qemu_target_page_size(); |
a22463a5 | 615 | info->ram->mbps = s->mbps; |
42d219d3 | 616 | info->ram->dirty_sync_count = ram_dirty_sync_count(); |
96506894 | 617 | info->ram->postcopy_requests = ram_postcopy_requests(); |
030ce1f8 | 618 | info->ram->page_size = qemu_target_page_size(); |
a22463a5 DDAG |
619 | |
620 | if (s->state != MIGRATION_STATUS_COMPLETED) { | |
621 | info->ram->remaining = ram_bytes_remaining(); | |
47ad8619 | 622 | info->ram->dirty_pages_rate = ram_dirty_pages_rate(); |
a22463a5 DDAG |
623 | } |
624 | } | |
625 | ||
791e7c82 | 626 | MigrationInfo *qmp_query_migrate(Error **errp) |
5bb7910a | 627 | { |
791e7c82 | 628 | MigrationInfo *info = g_malloc0(sizeof(*info)); |
17549e84 JQ |
629 | MigrationState *s = migrate_get_current(); |
630 | ||
631 | switch (s->state) { | |
31194731 | 632 | case MIGRATION_STATUS_NONE: |
17549e84 JQ |
633 | /* no migration has happened ever */ |
634 | break; | |
31194731 | 635 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 636 | info->has_status = true; |
ed4fbd10 | 637 | info->has_total_time = false; |
29ae8a41 | 638 | break; |
31194731 HZ |
639 | case MIGRATION_STATUS_ACTIVE: |
640 | case MIGRATION_STATUS_CANCELLING: | |
791e7c82 | 641 | info->has_status = true; |
7aa939af | 642 | info->has_total_time = true; |
bc72ad67 | 643 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) |
7aa939af | 644 | - s->total_time; |
2c52ddf1 JQ |
645 | info->has_expected_downtime = true; |
646 | info->expected_downtime = s->expected_downtime; | |
ed4fbd10 MH |
647 | info->has_setup_time = true; |
648 | info->setup_time = s->setup_time; | |
17549e84 | 649 | |
a22463a5 | 650 | populate_ram_info(info, s); |
8d017193 | 651 | |
17549e84 | 652 | if (blk_mig_active()) { |
791e7c82 LC |
653 | info->has_disk = true; |
654 | info->disk = g_malloc0(sizeof(*info->disk)); | |
655 | info->disk->transferred = blk_mig_bytes_transferred(); | |
656 | info->disk->remaining = blk_mig_bytes_remaining(); | |
657 | info->disk->total = blk_mig_bytes_total(); | |
ff8d81d8 | 658 | } |
f36d55af | 659 | |
4782893e | 660 | if (cpu_throttle_active()) { |
d85a31d1 JH |
661 | info->has_cpu_throttle_percentage = true; |
662 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
4782893e JH |
663 | } |
664 | ||
9ec055ae DDAG |
665 | get_xbzrle_cache_stats(info); |
666 | break; | |
667 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
668 | /* Mostly the same as active; TODO add some postcopy stats */ | |
669 | info->has_status = true; | |
670 | info->has_total_time = true; | |
671 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) | |
672 | - s->total_time; | |
673 | info->has_expected_downtime = true; | |
674 | info->expected_downtime = s->expected_downtime; | |
675 | info->has_setup_time = true; | |
676 | info->setup_time = s->setup_time; | |
677 | ||
a22463a5 | 678 | populate_ram_info(info, s); |
9ec055ae DDAG |
679 | |
680 | if (blk_mig_active()) { | |
681 | info->has_disk = true; | |
682 | info->disk = g_malloc0(sizeof(*info->disk)); | |
683 | info->disk->transferred = blk_mig_bytes_transferred(); | |
684 | info->disk->remaining = blk_mig_bytes_remaining(); | |
685 | info->disk->total = blk_mig_bytes_total(); | |
686 | } | |
687 | ||
f36d55af | 688 | get_xbzrle_cache_stats(info); |
17549e84 | 689 | break; |
0b827d5e HZ |
690 | case MIGRATION_STATUS_COLO: |
691 | info->has_status = true; | |
692 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
693 | break; | |
31194731 | 694 | case MIGRATION_STATUS_COMPLETED: |
f36d55af OW |
695 | get_xbzrle_cache_stats(info); |
696 | ||
791e7c82 | 697 | info->has_status = true; |
00c14997 | 698 | info->has_total_time = true; |
7aa939af | 699 | info->total_time = s->total_time; |
9c5a9fcf JQ |
700 | info->has_downtime = true; |
701 | info->downtime = s->downtime; | |
ed4fbd10 MH |
702 | info->has_setup_time = true; |
703 | info->setup_time = s->setup_time; | |
d5f8a570 | 704 | |
a22463a5 | 705 | populate_ram_info(info, s); |
17549e84 | 706 | break; |
31194731 | 707 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 708 | info->has_status = true; |
d59ce6f3 DB |
709 | if (s->error) { |
710 | info->has_error_desc = true; | |
711 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
712 | } | |
17549e84 | 713 | break; |
31194731 | 714 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 715 | info->has_status = true; |
17549e84 | 716 | break; |
5bb7910a | 717 | } |
cde63fbe | 718 | info->status = s->state; |
791e7c82 LC |
719 | |
720 | return info; | |
5bb7910a AL |
721 | } |
722 | ||
00458433 OW |
723 | void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params, |
724 | Error **errp) | |
725 | { | |
726 | MigrationState *s = migrate_get_current(); | |
727 | MigrationCapabilityStatusList *cap; | |
096631bd | 728 | bool old_postcopy_cap = migrate_postcopy_ram(); |
00458433 | 729 | |
f6844b99 | 730 | if (migration_is_setup_or_active(s->state)) { |
c6bd8c70 | 731 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
00458433 OW |
732 | return; |
733 | } | |
734 | ||
735 | for (cap = params; cap; cap = cap->next) { | |
ed1701c6 DDAG |
736 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
737 | if (cap->value->capability == MIGRATION_CAPABILITY_BLOCK | |
738 | && cap->value->state) { | |
739 | error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) " | |
740 | "block migration"); | |
741 | error_append_hint(errp, "Use drive_mirror+NBD instead.\n"); | |
742 | continue; | |
743 | } | |
744 | #endif | |
35a6ed4f HZ |
745 | if (cap->value->capability == MIGRATION_CAPABILITY_X_COLO) { |
746 | if (!colo_supported()) { | |
747 | error_setg(errp, "COLO is not currently supported, please" | |
748 | " configure with --enable-colo option in order to" | |
749 | " support COLO feature"); | |
750 | continue; | |
751 | } | |
752 | } | |
00458433 OW |
753 | s->enabled_capabilities[cap->value->capability] = cap->value->state; |
754 | } | |
53dd370c DDAG |
755 | |
756 | if (migrate_postcopy_ram()) { | |
757 | if (migrate_use_compression()) { | |
758 | /* The decompression threads asynchronously write into RAM | |
759 | * rather than use the atomic copies needed to avoid | |
760 | * userfaulting. It should be possible to fix the decompression | |
761 | * threads for compatibility in future. | |
762 | */ | |
763 | error_report("Postcopy is not currently compatible with " | |
764 | "compression"); | |
32c3db5b | 765 | s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] = |
53dd370c DDAG |
766 | false; |
767 | } | |
096631bd DDAG |
768 | /* This check is reasonably expensive, so only when it's being |
769 | * set the first time, also it's only the destination that needs | |
770 | * special support. | |
771 | */ | |
772 | if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) && | |
773 | !postcopy_ram_supported_by_host()) { | |
774 | /* postcopy_ram_supported_by_host will have emitted a more | |
775 | * detailed message | |
776 | */ | |
777 | error_report("Postcopy is not supported"); | |
778 | s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] = | |
779 | false; | |
780 | } | |
53dd370c | 781 | } |
00458433 OW |
782 | } |
783 | ||
7f375e04 | 784 | void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp) |
85de8323 LL |
785 | { |
786 | MigrationState *s = migrate_get_current(); | |
787 | ||
7f375e04 EB |
788 | if (params->has_compress_level && |
789 | (params->compress_level < 0 || params->compress_level > 9)) { | |
c6bd8c70 MA |
790 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", |
791 | "is invalid, it should be in the range of 0 to 9"); | |
85de8323 LL |
792 | return; |
793 | } | |
7f375e04 EB |
794 | if (params->has_compress_threads && |
795 | (params->compress_threads < 1 || params->compress_threads > 255)) { | |
c6bd8c70 MA |
796 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
797 | "compress_threads", | |
798 | "is invalid, it should be in the range of 1 to 255"); | |
85de8323 LL |
799 | return; |
800 | } | |
7f375e04 EB |
801 | if (params->has_decompress_threads && |
802 | (params->decompress_threads < 1 || params->decompress_threads > 255)) { | |
c6bd8c70 MA |
803 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
804 | "decompress_threads", | |
805 | "is invalid, it should be in the range of 1 to 255"); | |
85de8323 LL |
806 | return; |
807 | } | |
7f375e04 EB |
808 | if (params->has_cpu_throttle_initial && |
809 | (params->cpu_throttle_initial < 1 || | |
810 | params->cpu_throttle_initial > 99)) { | |
1626fee3 | 811 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 812 | "cpu_throttle_initial", |
1626fee3 | 813 | "an integer in the range of 1 to 99"); |
091ecc8b | 814 | return; |
1626fee3 | 815 | } |
7f375e04 EB |
816 | if (params->has_cpu_throttle_increment && |
817 | (params->cpu_throttle_increment < 1 || | |
818 | params->cpu_throttle_increment > 99)) { | |
1626fee3 | 819 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 820 | "cpu_throttle_increment", |
1626fee3 | 821 | "an integer in the range of 1 to 99"); |
091ecc8b | 822 | return; |
1626fee3 | 823 | } |
2ff30257 AA |
824 | if (params->has_max_bandwidth && |
825 | (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) { | |
826 | error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the" | |
827 | " range of 0 to %zu bytes/second", SIZE_MAX); | |
828 | return; | |
829 | } | |
830 | if (params->has_downtime_limit && | |
87c9cc1c DHB |
831 | (params->downtime_limit < 0 || |
832 | params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { | |
833 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " | |
834 | "the range of 0 to %d milliseconds", | |
835 | MAX_MIGRATE_DOWNTIME); | |
2ff30257 AA |
836 | return; |
837 | } | |
68b53591 HZ |
838 | if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) { |
839 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
840 | "x_checkpoint_delay", | |
841 | "is invalid, it should be positive"); | |
842 | } | |
85de8323 | 843 | |
7f375e04 EB |
844 | if (params->has_compress_level) { |
845 | s->parameters.compress_level = params->compress_level; | |
85de8323 | 846 | } |
7f375e04 EB |
847 | if (params->has_compress_threads) { |
848 | s->parameters.compress_threads = params->compress_threads; | |
85de8323 | 849 | } |
7f375e04 EB |
850 | if (params->has_decompress_threads) { |
851 | s->parameters.decompress_threads = params->decompress_threads; | |
85de8323 | 852 | } |
7f375e04 EB |
853 | if (params->has_cpu_throttle_initial) { |
854 | s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; | |
1626fee3 | 855 | } |
7f375e04 EB |
856 | if (params->has_cpu_throttle_increment) { |
857 | s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; | |
1626fee3 | 858 | } |
7f375e04 | 859 | if (params->has_tls_creds) { |
69ef1f36 | 860 | g_free(s->parameters.tls_creds); |
7f375e04 | 861 | s->parameters.tls_creds = g_strdup(params->tls_creds); |
69ef1f36 | 862 | } |
7f375e04 | 863 | if (params->has_tls_hostname) { |
69ef1f36 | 864 | g_free(s->parameters.tls_hostname); |
7f375e04 | 865 | s->parameters.tls_hostname = g_strdup(params->tls_hostname); |
69ef1f36 | 866 | } |
2ff30257 AA |
867 | if (params->has_max_bandwidth) { |
868 | s->parameters.max_bandwidth = params->max_bandwidth; | |
869 | if (s->to_dst_file) { | |
870 | qemu_file_set_rate_limit(s->to_dst_file, | |
871 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); | |
872 | } | |
873 | } | |
874 | if (params->has_downtime_limit) { | |
875 | s->parameters.downtime_limit = params->downtime_limit; | |
876 | } | |
68b53591 HZ |
877 | |
878 | if (params->has_x_checkpoint_delay) { | |
879 | s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; | |
479125d5 HZ |
880 | if (migration_in_colo_state()) { |
881 | colo_checkpoint_notify(s); | |
882 | } | |
68b53591 | 883 | } |
2833c59b JQ |
884 | if (params->has_block_incremental) { |
885 | s->parameters.block_incremental = params->block_incremental; | |
886 | } | |
85de8323 LL |
887 | } |
888 | ||
2594f56d | 889 | |
4886a1bc DDAG |
890 | void qmp_migrate_start_postcopy(Error **errp) |
891 | { | |
892 | MigrationState *s = migrate_get_current(); | |
893 | ||
894 | if (!migrate_postcopy_ram()) { | |
a54d340b | 895 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
896 | " the start of migration"); |
897 | return; | |
898 | } | |
899 | ||
900 | if (s->state == MIGRATION_STATUS_NONE) { | |
901 | error_setg(errp, "Postcopy must be started after migration has been" | |
902 | " started"); | |
903 | return; | |
904 | } | |
905 | /* | |
906 | * we don't error if migration has finished since that would be racy | |
907 | * with issuing this command. | |
908 | */ | |
909 | atomic_set(&s->start_postcopy, true); | |
910 | } | |
911 | ||
065e2813 AL |
912 | /* shared migration helpers */ |
913 | ||
48781e5b | 914 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 915 | { |
48781e5b | 916 | if (atomic_cmpxchg(state, old_state, new_state) == old_state) { |
4ba4bc5e | 917 | trace_migrate_set_state(new_state); |
b05dc723 | 918 | migrate_generate_event(new_state); |
51cf4c1a Z |
919 | } |
920 | } | |
921 | ||
2833c59b JQ |
922 | void migrate_set_block_enabled(bool value, Error **errp) |
923 | { | |
924 | MigrationCapabilityStatusList *cap; | |
925 | ||
926 | cap = g_new0(MigrationCapabilityStatusList, 1); | |
927 | cap->value = g_new0(MigrationCapabilityStatus, 1); | |
928 | cap->value->capability = MIGRATION_CAPABILITY_BLOCK; | |
929 | cap->value->state = value; | |
930 | qmp_migrate_set_capabilities(cap, errp); | |
931 | qapi_free_MigrationCapabilityStatusList(cap); | |
932 | } | |
933 | ||
934 | static void migrate_set_block_incremental(MigrationState *s, bool value) | |
935 | { | |
936 | s->parameters.block_incremental = value; | |
937 | } | |
938 | ||
939 | static void block_cleanup_parameters(MigrationState *s) | |
940 | { | |
941 | if (s->must_remove_block_options) { | |
942 | /* setting to false can never fail */ | |
943 | migrate_set_block_enabled(false, &error_abort); | |
944 | migrate_set_block_incremental(s, false); | |
945 | s->must_remove_block_options = false; | |
946 | } | |
947 | } | |
948 | ||
bb1fadc4 | 949 | static void migrate_fd_cleanup(void *opaque) |
065e2813 | 950 | { |
bb1fadc4 PB |
951 | MigrationState *s = opaque; |
952 | ||
953 | qemu_bh_delete(s->cleanup_bh); | |
954 | s->cleanup_bh = NULL; | |
955 | ||
ec481c6c | 956 | migration_page_queue_free(); |
6c595cde | 957 | |
89a02a9f | 958 | if (s->to_dst_file) { |
9013dca5 | 959 | trace_migrate_fd_cleanup(); |
404a7c05 | 960 | qemu_mutex_unlock_iothread(); |
1d34e4bf DDAG |
961 | if (s->migration_thread_running) { |
962 | qemu_thread_join(&s->thread); | |
963 | s->migration_thread_running = false; | |
964 | } | |
404a7c05 PB |
965 | qemu_mutex_lock_iothread(); |
966 | ||
8706d2d5 | 967 | migrate_compress_threads_join(); |
89a02a9f HZ |
968 | qemu_fclose(s->to_dst_file); |
969 | s->to_dst_file = NULL; | |
065e2813 AL |
970 | } |
971 | ||
9ec055ae DDAG |
972 | assert((s->state != MIGRATION_STATUS_ACTIVE) && |
973 | (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE)); | |
7a2c1721 | 974 | |
94f5a437 | 975 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 976 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 977 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 978 | } |
a3fa1d78 PB |
979 | |
980 | notifier_list_notify(&migration_state_notifiers, s); | |
2833c59b | 981 | block_cleanup_parameters(s); |
065e2813 AL |
982 | } |
983 | ||
d59ce6f3 | 984 | void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 985 | { |
25174055 | 986 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 987 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
988 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
989 | MIGRATION_STATUS_FAILED); | |
d59ce6f3 DB |
990 | if (!s->error) { |
991 | s->error = error_copy(error); | |
992 | } | |
bb1fadc4 | 993 | notifier_list_notify(&migration_state_notifiers, s); |
2833c59b | 994 | block_cleanup_parameters(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 | } | |
2833c59b | 1036 | block_cleanup_parameters(s); |
065e2813 AL |
1037 | } |
1038 | ||
99a0db9b GH |
1039 | void add_migration_state_change_notifier(Notifier *notify) |
1040 | { | |
1041 | notifier_list_add(&migration_state_notifiers, notify); | |
1042 | } | |
1043 | ||
1044 | void remove_migration_state_change_notifier(Notifier *notify) | |
1045 | { | |
31552529 | 1046 | notifier_remove(notify); |
99a0db9b GH |
1047 | } |
1048 | ||
02edd2e7 | 1049 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 1050 | { |
31194731 | 1051 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
1052 | } |
1053 | ||
7073693b | 1054 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 1055 | { |
31194731 | 1056 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 1057 | } |
0edda1c4 | 1058 | |
afe2df69 GH |
1059 | bool migration_has_failed(MigrationState *s) |
1060 | { | |
31194731 HZ |
1061 | return (s->state == MIGRATION_STATUS_CANCELLED || |
1062 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
1063 | } |
1064 | ||
5727309d | 1065 | bool migration_in_postcopy(void) |
9ec055ae | 1066 | { |
5727309d JQ |
1067 | MigrationState *s = migrate_get_current(); |
1068 | ||
9ec055ae DDAG |
1069 | return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); |
1070 | } | |
1071 | ||
b82fc321 DDAG |
1072 | bool migration_in_postcopy_after_devices(MigrationState *s) |
1073 | { | |
5727309d | 1074 | return migration_in_postcopy() && s->postcopy_after_devices; |
b82fc321 DDAG |
1075 | } |
1076 | ||
fab35005 | 1077 | bool migration_is_idle(void) |
fe44dc91 | 1078 | { |
fab35005 | 1079 | MigrationState *s = migrate_get_current(); |
fe44dc91 AA |
1080 | |
1081 | switch (s->state) { | |
1082 | case MIGRATION_STATUS_NONE: | |
1083 | case MIGRATION_STATUS_CANCELLED: | |
1084 | case MIGRATION_STATUS_COMPLETED: | |
1085 | case MIGRATION_STATUS_FAILED: | |
1086 | return true; | |
1087 | case MIGRATION_STATUS_SETUP: | |
1088 | case MIGRATION_STATUS_CANCELLING: | |
1089 | case MIGRATION_STATUS_ACTIVE: | |
1090 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1091 | case MIGRATION_STATUS_COLO: | |
1092 | return false; | |
1093 | case MIGRATION_STATUS__MAX: | |
1094 | g_assert_not_reached(); | |
1095 | } | |
1096 | ||
1097 | return false; | |
1098 | } | |
1099 | ||
a0762d9e | 1100 | MigrationState *migrate_init(void) |
0edda1c4 | 1101 | { |
17549e84 | 1102 | MigrationState *s = migrate_get_current(); |
bbf6da32 | 1103 | |
389775d1 DDAG |
1104 | /* |
1105 | * Reinitialise all migration state, except | |
1106 | * parameters/capabilities that the user set, and | |
1107 | * locks. | |
1108 | */ | |
1109 | s->bytes_xfer = 0; | |
1110 | s->xfer_limit = 0; | |
1111 | s->cleanup_bh = 0; | |
89a02a9f | 1112 | s->to_dst_file = NULL; |
389775d1 | 1113 | s->state = MIGRATION_STATUS_NONE; |
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; | |
389775d1 | 1119 | s->setup_time = 0; |
389775d1 | 1120 | s->start_postcopy = false; |
b82fc321 | 1121 | s->postcopy_after_devices = false; |
389775d1 | 1122 | s->migration_thread_running = false; |
d59ce6f3 DB |
1123 | error_free(s->error); |
1124 | s->error = NULL; | |
389775d1 | 1125 | |
48781e5b | 1126 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 1127 | |
bc72ad67 | 1128 | s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0edda1c4 JQ |
1129 | return s; |
1130 | } | |
cab30143 | 1131 | |
fa2756b7 AL |
1132 | static GSList *migration_blockers; |
1133 | ||
fe44dc91 | 1134 | int migrate_add_blocker(Error *reason, Error **errp) |
fa2756b7 | 1135 | { |
b67b8c3a AA |
1136 | if (only_migratable) { |
1137 | error_propagate(errp, error_copy(reason)); | |
1138 | error_prepend(errp, "disallowing migration blocker " | |
1139 | "(--only_migratable) for: "); | |
1140 | return -EACCES; | |
1141 | } | |
1142 | ||
fab35005 | 1143 | if (migration_is_idle()) { |
fe44dc91 AA |
1144 | migration_blockers = g_slist_prepend(migration_blockers, reason); |
1145 | return 0; | |
1146 | } | |
1147 | ||
1148 | error_propagate(errp, error_copy(reason)); | |
1149 | error_prepend(errp, "disallowing migration blocker (migration in " | |
1150 | "progress) for: "); | |
1151 | return -EBUSY; | |
fa2756b7 AL |
1152 | } |
1153 | ||
1154 | void migrate_del_blocker(Error *reason) | |
1155 | { | |
1156 | migration_blockers = g_slist_remove(migration_blockers, reason); | |
1157 | } | |
1158 | ||
bf1ae1f4 DDAG |
1159 | void qmp_migrate_incoming(const char *uri, Error **errp) |
1160 | { | |
1161 | Error *local_err = NULL; | |
4debb5f5 | 1162 | static bool once = true; |
bf1ae1f4 DDAG |
1163 | |
1164 | if (!deferred_incoming) { | |
4debb5f5 | 1165 | error_setg(errp, "For use with '-incoming defer'"); |
bf1ae1f4 DDAG |
1166 | return; |
1167 | } | |
4debb5f5 DDAG |
1168 | if (!once) { |
1169 | error_setg(errp, "The incoming migration has already been started"); | |
1170 | } | |
bf1ae1f4 DDAG |
1171 | |
1172 | qemu_start_incoming_migration(uri, &local_err); | |
1173 | ||
1174 | if (local_err) { | |
1175 | error_propagate(errp, local_err); | |
1176 | return; | |
1177 | } | |
1178 | ||
4debb5f5 | 1179 | once = false; |
bf1ae1f4 DDAG |
1180 | } |
1181 | ||
24f3902b GK |
1182 | bool migration_is_blocked(Error **errp) |
1183 | { | |
1184 | if (qemu_savevm_state_blocked(errp)) { | |
1185 | return true; | |
1186 | } | |
1187 | ||
1188 | if (migration_blockers) { | |
1189 | *errp = error_copy(migration_blockers->data); | |
1190 | return true; | |
1191 | } | |
1192 | ||
1193 | return false; | |
1194 | } | |
1195 | ||
e1c37d0e LC |
1196 | void qmp_migrate(const char *uri, bool has_blk, bool blk, |
1197 | bool has_inc, bool inc, bool has_detach, bool detach, | |
1198 | Error **errp) | |
cab30143 | 1199 | { |
be7059cd | 1200 | Error *local_err = NULL; |
17549e84 | 1201 | MigrationState *s = migrate_get_current(); |
cab30143 | 1202 | const char *p; |
cab30143 | 1203 | |
f6844b99 | 1204 | if (migration_is_setup_or_active(s->state) || |
0b827d5e HZ |
1205 | s->state == MIGRATION_STATUS_CANCELLING || |
1206 | s->state == MIGRATION_STATUS_COLO) { | |
c6bd8c70 | 1207 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
e1c37d0e | 1208 | return; |
cab30143 | 1209 | } |
ca99993a DDAG |
1210 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
1211 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
1212 | return; | |
1213 | } | |
1214 | ||
24f3902b | 1215 | if (migration_is_blocked(errp)) { |
e1c37d0e | 1216 | return; |
fa2756b7 AL |
1217 | } |
1218 | ||
2833c59b JQ |
1219 | if ((has_blk && blk) || (has_inc && inc)) { |
1220 | if (migrate_use_block() || migrate_use_block_incremental()) { | |
1221 | error_setg(errp, "Command options are incompatible with " | |
1222 | "current migration capabilities"); | |
1223 | return; | |
1224 | } | |
1225 | migrate_set_block_enabled(true, &local_err); | |
1226 | if (local_err) { | |
1227 | error_propagate(errp, local_err); | |
1228 | return; | |
1229 | } | |
1230 | s->must_remove_block_options = true; | |
1231 | } | |
1232 | ||
1233 | if (has_inc && inc) { | |
1234 | migrate_set_block_incremental(s, true); | |
1235 | } | |
1236 | ||
a0762d9e | 1237 | s = migrate_init(); |
cab30143 JQ |
1238 | |
1239 | if (strstart(uri, "tcp:", &p)) { | |
f37afb5a | 1240 | tcp_start_outgoing_migration(s, p, &local_err); |
2da776db | 1241 | #ifdef CONFIG_RDMA |
41310c68 | 1242 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
1243 | rdma_start_outgoing_migration(s, p, &local_err); |
1244 | #endif | |
cab30143 | 1245 | } else if (strstart(uri, "exec:", &p)) { |
f37afb5a | 1246 | exec_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1247 | } else if (strstart(uri, "unix:", &p)) { |
f37afb5a | 1248 | unix_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1249 | } else if (strstart(uri, "fd:", &p)) { |
f37afb5a | 1250 | fd_start_outgoing_migration(s, p, &local_err); |
99a0db9b | 1251 | } else { |
c6bd8c70 MA |
1252 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", |
1253 | "a valid migration protocol"); | |
48781e5b HZ |
1254 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1255 | MIGRATION_STATUS_FAILED); | |
e1c37d0e | 1256 | return; |
cab30143 JQ |
1257 | } |
1258 | ||
f37afb5a | 1259 | if (local_err) { |
d59ce6f3 | 1260 | migrate_fd_error(s, local_err); |
f37afb5a | 1261 | error_propagate(errp, local_err); |
e1c37d0e | 1262 | return; |
1299c631 | 1263 | } |
cab30143 JQ |
1264 | } |
1265 | ||
6cdedb07 | 1266 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 1267 | { |
17549e84 | 1268 | migrate_fd_cancel(migrate_get_current()); |
cab30143 JQ |
1269 | } |
1270 | ||
9e1ba4cc OW |
1271 | void qmp_migrate_set_cache_size(int64_t value, Error **errp) |
1272 | { | |
1273 | MigrationState *s = migrate_get_current(); | |
c91e681a | 1274 | int64_t new_size; |
9e1ba4cc OW |
1275 | |
1276 | /* Check for truncation */ | |
1277 | if (value != (size_t)value) { | |
c6bd8c70 MA |
1278 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1279 | "exceeding address space"); | |
9e1ba4cc OW |
1280 | return; |
1281 | } | |
1282 | ||
a5615b14 OW |
1283 | /* Cache should not be larger than guest ram size */ |
1284 | if (value > ram_bytes_total()) { | |
c6bd8c70 MA |
1285 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1286 | "exceeds guest ram size "); | |
a5615b14 OW |
1287 | return; |
1288 | } | |
1289 | ||
c91e681a OW |
1290 | new_size = xbzrle_cache_resize(value); |
1291 | if (new_size < 0) { | |
c6bd8c70 MA |
1292 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", |
1293 | "is smaller than page size"); | |
c91e681a OW |
1294 | return; |
1295 | } | |
1296 | ||
1297 | s->xbzrle_cache_size = new_size; | |
9e1ba4cc OW |
1298 | } |
1299 | ||
1300 | int64_t qmp_query_migrate_cache_size(Error **errp) | |
1301 | { | |
1302 | return migrate_xbzrle_cache_size(); | |
1303 | } | |
1304 | ||
3dc85383 | 1305 | void qmp_migrate_set_speed(int64_t value, Error **errp) |
cab30143 | 1306 | { |
2ff30257 AA |
1307 | MigrationParameters p = { |
1308 | .has_max_bandwidth = true, | |
1309 | .max_bandwidth = value, | |
1310 | }; | |
cab30143 | 1311 | |
2ff30257 | 1312 | qmp_migrate_set_parameters(&p, errp); |
cab30143 JQ |
1313 | } |
1314 | ||
4f0a993b | 1315 | void qmp_migrate_set_downtime(double value, Error **errp) |
cab30143 | 1316 | { |
87c9cc1c DHB |
1317 | if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) { |
1318 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " | |
1319 | "the range of 0 to %d seconds", | |
1320 | MAX_MIGRATE_DOWNTIME_SECONDS); | |
1321 | return; | |
1322 | } | |
1323 | ||
2ff30257 AA |
1324 | value *= 1000; /* Convert to milliseconds */ |
1325 | value = MAX(0, MIN(INT64_MAX, value)); | |
1326 | ||
1327 | MigrationParameters p = { | |
1328 | .has_downtime_limit = true, | |
1329 | .downtime_limit = value, | |
1330 | }; | |
1331 | ||
1332 | qmp_migrate_set_parameters(&p, errp); | |
99a0db9b | 1333 | } |
17ad9b35 | 1334 | |
53f09a10 PB |
1335 | bool migrate_release_ram(void) |
1336 | { | |
1337 | MigrationState *s; | |
1338 | ||
1339 | s = migrate_get_current(); | |
1340 | ||
1341 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; | |
1342 | } | |
1343 | ||
53dd370c DDAG |
1344 | bool migrate_postcopy_ram(void) |
1345 | { | |
1346 | MigrationState *s; | |
1347 | ||
1348 | s = migrate_get_current(); | |
1349 | ||
32c3db5b | 1350 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
53dd370c DDAG |
1351 | } |
1352 | ||
bde1e2ec CV |
1353 | bool migrate_auto_converge(void) |
1354 | { | |
1355 | MigrationState *s; | |
1356 | ||
1357 | s = migrate_get_current(); | |
1358 | ||
1359 | return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; | |
1360 | } | |
1361 | ||
323004a3 PL |
1362 | bool migrate_zero_blocks(void) |
1363 | { | |
1364 | MigrationState *s; | |
1365 | ||
1366 | s = migrate_get_current(); | |
1367 | ||
1368 | return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; | |
1369 | } | |
1370 | ||
8706d2d5 LL |
1371 | bool migrate_use_compression(void) |
1372 | { | |
dde4e694 LL |
1373 | MigrationState *s; |
1374 | ||
1375 | s = migrate_get_current(); | |
1376 | ||
1377 | return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; | |
8706d2d5 LL |
1378 | } |
1379 | ||
1380 | int migrate_compress_level(void) | |
1381 | { | |
1382 | MigrationState *s; | |
1383 | ||
1384 | s = migrate_get_current(); | |
1385 | ||
2594f56d | 1386 | return s->parameters.compress_level; |
8706d2d5 LL |
1387 | } |
1388 | ||
1389 | int migrate_compress_threads(void) | |
1390 | { | |
1391 | MigrationState *s; | |
1392 | ||
1393 | s = migrate_get_current(); | |
1394 | ||
2594f56d | 1395 | return s->parameters.compress_threads; |
8706d2d5 LL |
1396 | } |
1397 | ||
3fcb38c2 LL |
1398 | int migrate_decompress_threads(void) |
1399 | { | |
1400 | MigrationState *s; | |
1401 | ||
1402 | s = migrate_get_current(); | |
1403 | ||
2594f56d | 1404 | return s->parameters.decompress_threads; |
3fcb38c2 LL |
1405 | } |
1406 | ||
b05dc723 JQ |
1407 | bool migrate_use_events(void) |
1408 | { | |
1409 | MigrationState *s; | |
1410 | ||
1411 | s = migrate_get_current(); | |
1412 | ||
1413 | return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; | |
1414 | } | |
1415 | ||
17ad9b35 OW |
1416 | int migrate_use_xbzrle(void) |
1417 | { | |
1418 | MigrationState *s; | |
1419 | ||
1420 | s = migrate_get_current(); | |
1421 | ||
1422 | return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; | |
1423 | } | |
1424 | ||
1425 | int64_t migrate_xbzrle_cache_size(void) | |
1426 | { | |
1427 | MigrationState *s; | |
1428 | ||
1429 | s = migrate_get_current(); | |
1430 | ||
1431 | return s->xbzrle_cache_size; | |
1432 | } | |
0d82d0e8 | 1433 | |
2833c59b JQ |
1434 | bool migrate_use_block(void) |
1435 | { | |
1436 | MigrationState *s; | |
1437 | ||
1438 | s = migrate_get_current(); | |
1439 | ||
1440 | return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK]; | |
1441 | } | |
1442 | ||
1443 | bool migrate_use_block_incremental(void) | |
1444 | { | |
1445 | MigrationState *s; | |
1446 | ||
1447 | s = migrate_get_current(); | |
1448 | ||
1449 | return s->parameters.block_incremental; | |
1450 | } | |
1451 | ||
70b20477 DDAG |
1452 | /* migration thread support */ |
1453 | /* | |
1454 | * Something bad happened to the RP stream, mark an error | |
1455 | * The caller shall print or trace something to indicate why | |
1456 | */ | |
1457 | static void mark_source_rp_bad(MigrationState *s) | |
1458 | { | |
1459 | s->rp_state.error = true; | |
1460 | } | |
1461 | ||
1462 | static struct rp_cmd_args { | |
1463 | ssize_t len; /* -1 = variable */ | |
1464 | const char *name; | |
1465 | } rp_cmd_args[] = { | |
1466 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
1467 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
1468 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
1469 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
1470 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
70b20477 DDAG |
1471 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
1472 | }; | |
1473 | ||
1e2d90eb DDAG |
1474 | /* |
1475 | * Process a request for pages received on the return path, | |
1476 | * We're allowed to send more than requested (e.g. to round to our page size) | |
1477 | * and we don't need to send pages that have already been sent. | |
1478 | */ | |
1479 | static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
1480 | ram_addr_t start, size_t len) | |
1481 | { | |
6c595cde DDAG |
1482 | long our_host_ps = getpagesize(); |
1483 | ||
1e2d90eb | 1484 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
1485 | |
1486 | /* | |
1487 | * Since we currently insist on matching page sizes, just sanity check | |
1488 | * we're being asked for whole host pages. | |
1489 | */ | |
1490 | if (start & (our_host_ps-1) || | |
1491 | (len & (our_host_ps-1))) { | |
1492 | error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT | |
1493 | " len: %zd", __func__, start, len); | |
1494 | mark_source_rp_bad(ms); | |
1495 | return; | |
1496 | } | |
1497 | ||
96506894 | 1498 | if (ram_save_queue_pages(rbname, start, len)) { |
6c595cde DDAG |
1499 | mark_source_rp_bad(ms); |
1500 | } | |
1e2d90eb DDAG |
1501 | } |
1502 | ||
70b20477 DDAG |
1503 | /* |
1504 | * Handles messages sent on the return path towards the source VM | |
1505 | * | |
1506 | */ | |
1507 | static void *source_return_path_thread(void *opaque) | |
1508 | { | |
1509 | MigrationState *ms = opaque; | |
1510 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
1511 | uint16_t header_len, header_type; | |
568b01ca | 1512 | uint8_t buf[512]; |
70b20477 | 1513 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
1514 | ram_addr_t start = 0; /* =0 to silence warning */ |
1515 | size_t len = 0, expected_len; | |
70b20477 DDAG |
1516 | int res; |
1517 | ||
1518 | trace_source_return_path_thread_entry(); | |
1519 | while (!ms->rp_state.error && !qemu_file_get_error(rp) && | |
1520 | migration_is_setup_or_active(ms->state)) { | |
1521 | trace_source_return_path_thread_loop_top(); | |
1522 | header_type = qemu_get_be16(rp); | |
1523 | header_len = qemu_get_be16(rp); | |
1524 | ||
1525 | if (header_type >= MIG_RP_MSG_MAX || | |
1526 | header_type == MIG_RP_MSG_INVALID) { | |
1527 | error_report("RP: Received invalid message 0x%04x length 0x%04x", | |
1528 | header_type, header_len); | |
1529 | mark_source_rp_bad(ms); | |
1530 | goto out; | |
1531 | } | |
1532 | ||
1533 | if ((rp_cmd_args[header_type].len != -1 && | |
1534 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 1535 | header_len > sizeof(buf)) { |
70b20477 DDAG |
1536 | error_report("RP: Received '%s' message (0x%04x) with" |
1537 | "incorrect length %d expecting %zu", | |
1538 | rp_cmd_args[header_type].name, header_type, header_len, | |
1539 | (size_t)rp_cmd_args[header_type].len); | |
1540 | mark_source_rp_bad(ms); | |
1541 | goto out; | |
1542 | } | |
1543 | ||
1544 | /* We know we've got a valid header by this point */ | |
1545 | res = qemu_get_buffer(rp, buf, header_len); | |
1546 | if (res != header_len) { | |
1547 | error_report("RP: Failed reading data for message 0x%04x" | |
1548 | " read %d expected %d", | |
1549 | header_type, res, header_len); | |
1550 | mark_source_rp_bad(ms); | |
1551 | goto out; | |
1552 | } | |
1553 | ||
1554 | /* OK, we have the message and the data */ | |
1555 | switch (header_type) { | |
1556 | case MIG_RP_MSG_SHUT: | |
4d885131 | 1557 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
1558 | trace_source_return_path_thread_shut(sibling_error); |
1559 | if (sibling_error) { | |
1560 | error_report("RP: Sibling indicated error %d", sibling_error); | |
1561 | mark_source_rp_bad(ms); | |
1562 | } | |
1563 | /* | |
1564 | * We'll let the main thread deal with closing the RP | |
1565 | * we could do a shutdown(2) on it, but we're the only user | |
1566 | * anyway, so there's nothing gained. | |
1567 | */ | |
1568 | goto out; | |
1569 | ||
1570 | case MIG_RP_MSG_PONG: | |
4d885131 | 1571 | tmp32 = ldl_be_p(buf); |
70b20477 DDAG |
1572 | trace_source_return_path_thread_pong(tmp32); |
1573 | break; | |
1574 | ||
1e2d90eb | 1575 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
1576 | start = ldq_be_p(buf); |
1577 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1578 | migrate_handle_rp_req_pages(ms, NULL, start, len); |
1579 | break; | |
1580 | ||
1581 | case MIG_RP_MSG_REQ_PAGES_ID: | |
1582 | expected_len = 12 + 1; /* header + termination */ | |
1583 | ||
1584 | if (header_len >= expected_len) { | |
4d885131 PM |
1585 | start = ldq_be_p(buf); |
1586 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1587 | /* Now we expect an idstr */ |
1588 | tmp32 = buf[12]; /* Length of the following idstr */ | |
1589 | buf[13 + tmp32] = '\0'; | |
1590 | expected_len += tmp32; | |
1591 | } | |
1592 | if (header_len != expected_len) { | |
1593 | error_report("RP: Req_Page_id with length %d expecting %zd", | |
1594 | header_len, expected_len); | |
1595 | mark_source_rp_bad(ms); | |
1596 | goto out; | |
1597 | } | |
1598 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); | |
1599 | break; | |
1600 | ||
70b20477 DDAG |
1601 | default: |
1602 | break; | |
1603 | } | |
1604 | } | |
5df5416e | 1605 | if (qemu_file_get_error(rp)) { |
70b20477 DDAG |
1606 | trace_source_return_path_thread_bad_end(); |
1607 | mark_source_rp_bad(ms); | |
1608 | } | |
1609 | ||
1610 | trace_source_return_path_thread_end(); | |
1611 | out: | |
1612 | ms->rp_state.from_dst_file = NULL; | |
1613 | qemu_fclose(rp); | |
1614 | return NULL; | |
1615 | } | |
1616 | ||
70b20477 DDAG |
1617 | static int open_return_path_on_source(MigrationState *ms) |
1618 | { | |
1619 | ||
89a02a9f | 1620 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
1621 | if (!ms->rp_state.from_dst_file) { |
1622 | return -1; | |
1623 | } | |
1624 | ||
1625 | trace_open_return_path_on_source(); | |
1626 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", | |
1627 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
1628 | ||
1629 | trace_open_return_path_on_source_continue(); | |
1630 | ||
1631 | return 0; | |
1632 | } | |
1633 | ||
70b20477 DDAG |
1634 | /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ |
1635 | static int await_return_path_close_on_source(MigrationState *ms) | |
1636 | { | |
1637 | /* | |
1638 | * If this is a normal exit then the destination will send a SHUT and the | |
1639 | * rp_thread will exit, however if there's an error we need to cause | |
1640 | * it to exit. | |
1641 | */ | |
89a02a9f | 1642 | if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { |
70b20477 DDAG |
1643 | /* |
1644 | * shutdown(2), if we have it, will cause it to unblock if it's stuck | |
1645 | * waiting for the destination. | |
1646 | */ | |
1647 | qemu_file_shutdown(ms->rp_state.from_dst_file); | |
1648 | mark_source_rp_bad(ms); | |
1649 | } | |
1650 | trace_await_return_path_close_on_source_joining(); | |
1651 | qemu_thread_join(&ms->rp_state.rp_thread); | |
1652 | trace_await_return_path_close_on_source_close(); | |
1653 | return ms->rp_state.error; | |
1654 | } | |
1655 | ||
1d34e4bf DDAG |
1656 | /* |
1657 | * Switch from normal iteration to postcopy | |
1658 | * Returns non-0 on error | |
1659 | */ | |
1660 | static int postcopy_start(MigrationState *ms, bool *old_vm_running) | |
1661 | { | |
1662 | int ret; | |
61b67d47 DB |
1663 | QIOChannelBuffer *bioc; |
1664 | QEMUFile *fb; | |
1d34e4bf | 1665 | int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
ef8d6488 | 1666 | bool restart_block = false; |
48781e5b | 1667 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, |
1d34e4bf DDAG |
1668 | MIGRATION_STATUS_POSTCOPY_ACTIVE); |
1669 | ||
1670 | trace_postcopy_start(); | |
1671 | qemu_mutex_lock_iothread(); | |
1672 | trace_postcopy_start_set_run(); | |
1673 | ||
1674 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); | |
1675 | *old_vm_running = runstate_is_running(); | |
1676 | global_state_store(); | |
1677 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
76b1c7fe KW |
1678 | if (ret < 0) { |
1679 | goto fail; | |
1680 | } | |
1d34e4bf | 1681 | |
76b1c7fe | 1682 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
1683 | if (ret < 0) { |
1684 | goto fail; | |
1685 | } | |
ef8d6488 | 1686 | restart_block = true; |
1d34e4bf | 1687 | |
1c0d249d DDAG |
1688 | /* |
1689 | * Cause any non-postcopiable, but iterative devices to | |
1690 | * send out their final data. | |
1691 | */ | |
89a02a9f | 1692 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true); |
1c0d249d | 1693 | |
1d34e4bf DDAG |
1694 | /* |
1695 | * in Finish migrate and with the io-lock held everything should | |
1696 | * be quiet, but we've potentially still got dirty pages and we | |
1697 | * need to tell the destination to throw any pages it's already received | |
1698 | * that are dirty | |
1699 | */ | |
1700 | if (ram_postcopy_send_discard_bitmap(ms)) { | |
1701 | error_report("postcopy send discard bitmap failed"); | |
1702 | goto fail; | |
1703 | } | |
1704 | ||
1705 | /* | |
1706 | * send rest of state - note things that are doing postcopy | |
1707 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
1708 | * wrap their state up here | |
1709 | */ | |
89a02a9f | 1710 | qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); |
1d34e4bf | 1711 | /* Ping just for debugging, helps line traces up */ |
89a02a9f | 1712 | qemu_savevm_send_ping(ms->to_dst_file, 2); |
1d34e4bf DDAG |
1713 | |
1714 | /* | |
1715 | * While loading the device state we may trigger page transfer | |
1716 | * requests and the fd must be free to process those, and thus | |
1717 | * the destination must read the whole device state off the fd before | |
1718 | * it starts processing it. Unfortunately the ad-hoc migration format | |
1719 | * doesn't allow the destination to know the size to read without fully | |
1720 | * parsing it through each devices load-state code (especially the open | |
1721 | * coded devices that use get/put). | |
1722 | * So we wrap the device state up in a package with a length at the start; | |
1723 | * to do this we use a qemu_buf to hold the whole of the device state. | |
1724 | */ | |
61b67d47 | 1725 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 1726 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
61b67d47 DB |
1727 | fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc)); |
1728 | object_unref(OBJECT(bioc)); | |
1d34e4bf | 1729 | |
c76201ab DDAG |
1730 | /* |
1731 | * Make sure the receiver can get incoming pages before we send the rest | |
1732 | * of the state | |
1733 | */ | |
1734 | qemu_savevm_send_postcopy_listen(fb); | |
1735 | ||
1c0d249d | 1736 | qemu_savevm_state_complete_precopy(fb, false); |
1d34e4bf DDAG |
1737 | qemu_savevm_send_ping(fb, 3); |
1738 | ||
1739 | qemu_savevm_send_postcopy_run(fb); | |
1740 | ||
1741 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 1742 | |
ef8d6488 DDAG |
1743 | /* Last point of recovery; as soon as we send the package the destination |
1744 | * can open devices and potentially start running. | |
1745 | * Lets just check again we've not got any errors. | |
1746 | */ | |
1747 | ret = qemu_file_get_error(ms->to_dst_file); | |
1748 | if (ret) { | |
1749 | error_report("postcopy_start: Migration stream errored (pre package)"); | |
1750 | goto fail_closefb; | |
1751 | } | |
1752 | ||
1753 | restart_block = false; | |
1754 | ||
1d34e4bf | 1755 | /* Now send that blob */ |
61b67d47 | 1756 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
1757 | goto fail_closefb; |
1758 | } | |
1759 | qemu_fclose(fb); | |
b82fc321 DDAG |
1760 | |
1761 | /* Send a notify to give a chance for anything that needs to happen | |
1762 | * at the transition to postcopy and after the device state; in particular | |
1763 | * spice needs to trigger a transition now | |
1764 | */ | |
1765 | ms->postcopy_after_devices = true; | |
1766 | notifier_list_notify(&migration_state_notifiers, ms); | |
1767 | ||
1d34e4bf DDAG |
1768 | ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; |
1769 | ||
1770 | qemu_mutex_unlock_iothread(); | |
1771 | ||
1772 | /* | |
1773 | * Although this ping is just for debug, it could potentially be | |
1774 | * used for getting a better measurement of downtime at the source. | |
1775 | */ | |
89a02a9f | 1776 | qemu_savevm_send_ping(ms->to_dst_file, 4); |
1d34e4bf | 1777 | |
ced1c616 PB |
1778 | if (migrate_release_ram()) { |
1779 | ram_postcopy_migrated_memory_release(ms); | |
1780 | } | |
1781 | ||
89a02a9f | 1782 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf DDAG |
1783 | if (ret) { |
1784 | error_report("postcopy_start: Migration stream errored"); | |
48781e5b | 1785 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf DDAG |
1786 | MIGRATION_STATUS_FAILED); |
1787 | } | |
1788 | ||
1789 | return ret; | |
1790 | ||
1791 | fail_closefb: | |
1792 | qemu_fclose(fb); | |
1793 | fail: | |
48781e5b | 1794 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 1795 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
1796 | if (restart_block) { |
1797 | /* A failure happened early enough that we know the destination hasn't | |
1798 | * accessed block devices, so we're safe to recover. | |
1799 | */ | |
1800 | Error *local_err = NULL; | |
1801 | ||
1802 | bdrv_invalidate_cache_all(&local_err); | |
1803 | if (local_err) { | |
1804 | error_report_err(local_err); | |
1805 | } | |
1806 | } | |
1d34e4bf DDAG |
1807 | qemu_mutex_unlock_iothread(); |
1808 | return -1; | |
1809 | } | |
1810 | ||
09f6c85e DDAG |
1811 | /** |
1812 | * migration_completion: Used by migration_thread when there's not much left. | |
1813 | * The caller 'breaks' the loop when this returns. | |
1814 | * | |
1815 | * @s: Current migration state | |
36f48567 | 1816 | * @current_active_state: The migration state we expect to be in |
09f6c85e DDAG |
1817 | * @*old_vm_running: Pointer to old_vm_running flag |
1818 | * @*start_time: Pointer to time to update | |
1819 | */ | |
36f48567 DDAG |
1820 | static void migration_completion(MigrationState *s, int current_active_state, |
1821 | bool *old_vm_running, | |
09f6c85e DDAG |
1822 | int64_t *start_time) |
1823 | { | |
1824 | int ret; | |
1825 | ||
b10ac0c4 DDAG |
1826 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
1827 | qemu_mutex_lock_iothread(); | |
1828 | *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
1829 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); | |
1830 | *old_vm_running = runstate_is_running(); | |
1831 | ret = global_state_store(); | |
1832 | ||
1833 | if (!ret) { | |
1834 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
0b827d5e HZ |
1835 | /* |
1836 | * Don't mark the image with BDRV_O_INACTIVE flag if | |
1837 | * we will go into COLO stage later. | |
1838 | */ | |
1839 | if (ret >= 0 && !migrate_colo_enabled()) { | |
76b1c7fe KW |
1840 | ret = bdrv_inactivate_all(); |
1841 | } | |
b10ac0c4 | 1842 | if (ret >= 0) { |
89a02a9f HZ |
1843 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); |
1844 | qemu_savevm_state_complete_precopy(s->to_dst_file, false); | |
1d2acc31 | 1845 | s->block_inactive = true; |
b10ac0c4 DDAG |
1846 | } |
1847 | } | |
1848 | qemu_mutex_unlock_iothread(); | |
09f6c85e | 1849 | |
b10ac0c4 DDAG |
1850 | if (ret < 0) { |
1851 | goto fail; | |
09f6c85e | 1852 | } |
b10ac0c4 DDAG |
1853 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { |
1854 | trace_migration_completion_postcopy_end(); | |
1855 | ||
89a02a9f | 1856 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
b10ac0c4 | 1857 | trace_migration_completion_postcopy_end_after_complete(); |
09f6c85e | 1858 | } |
09f6c85e | 1859 | |
b10ac0c4 DDAG |
1860 | /* |
1861 | * If rp was opened we must clean up the thread before | |
1862 | * cleaning everything else up (since if there are no failures | |
1863 | * it will wait for the destination to send it's status in | |
1864 | * a SHUT command). | |
1865 | * Postcopy opens rp if enabled (even if it's not avtivated) | |
1866 | */ | |
1867 | if (migrate_postcopy_ram()) { | |
1868 | int rp_error; | |
1869 | trace_migration_completion_postcopy_end_before_rp(); | |
1870 | rp_error = await_return_path_close_on_source(s); | |
1871 | trace_migration_completion_postcopy_end_after_rp(rp_error); | |
1872 | if (rp_error) { | |
fe904ea8 | 1873 | goto fail_invalidate; |
b10ac0c4 | 1874 | } |
09f6c85e DDAG |
1875 | } |
1876 | ||
89a02a9f | 1877 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 1878 | trace_migration_completion_file_err(); |
fe904ea8 | 1879 | goto fail_invalidate; |
09f6c85e DDAG |
1880 | } |
1881 | ||
0b827d5e HZ |
1882 | if (!migrate_colo_enabled()) { |
1883 | migrate_set_state(&s->state, current_active_state, | |
1884 | MIGRATION_STATUS_COMPLETED); | |
1885 | } | |
1886 | ||
09f6c85e DDAG |
1887 | return; |
1888 | ||
fe904ea8 GK |
1889 | fail_invalidate: |
1890 | /* If not doing postcopy, vm_start() will be called: let's regain | |
1891 | * control on images. | |
1892 | */ | |
1893 | if (s->state == MIGRATION_STATUS_ACTIVE) { | |
1894 | Error *local_err = NULL; | |
1895 | ||
1d2acc31 | 1896 | qemu_mutex_lock_iothread(); |
fe904ea8 GK |
1897 | bdrv_invalidate_cache_all(&local_err); |
1898 | if (local_err) { | |
1899 | error_report_err(local_err); | |
1d2acc31 HZ |
1900 | } else { |
1901 | s->block_inactive = false; | |
fe904ea8 | 1902 | } |
1d2acc31 | 1903 | qemu_mutex_unlock_iothread(); |
fe904ea8 GK |
1904 | } |
1905 | ||
09f6c85e | 1906 | fail: |
48781e5b HZ |
1907 | migrate_set_state(&s->state, current_active_state, |
1908 | MIGRATION_STATUS_FAILED); | |
09f6c85e DDAG |
1909 | } |
1910 | ||
35a6ed4f HZ |
1911 | bool migrate_colo_enabled(void) |
1912 | { | |
1913 | MigrationState *s = migrate_get_current(); | |
1914 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; | |
1915 | } | |
1916 | ||
70b20477 DDAG |
1917 | /* |
1918 | * Master migration thread on the source VM. | |
1919 | * It drives the migration and pumps the data down the outgoing channel. | |
1920 | */ | |
5f496a1b | 1921 | static void *migration_thread(void *opaque) |
0d82d0e8 | 1922 | { |
9848a404 | 1923 | MigrationState *s = opaque; |
1d34e4bf | 1924 | /* Used by the bandwidth calcs, updated later */ |
bc72ad67 AB |
1925 | int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1926 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); | |
be7172e2 | 1927 | int64_t initial_bytes = 0; |
faec066a PX |
1928 | /* |
1929 | * The final stage happens when the remaining data is smaller than | |
1930 | * this threshold; it's calculated from the requested downtime and | |
1931 | * measured bandwidth | |
1932 | */ | |
1933 | int64_t threshold_size = 0; | |
a3fa1d78 | 1934 | int64_t start_time = initial_time; |
94f5a437 | 1935 | int64_t end_time; |
a3fa1d78 | 1936 | bool old_vm_running = false; |
1d34e4bf DDAG |
1937 | bool entered_postcopy = false; |
1938 | /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */ | |
1939 | enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE; | |
0b827d5e | 1940 | bool enable_colo = migrate_colo_enabled(); |
76f5933a | 1941 | |
ab28bd23 PB |
1942 | rcu_register_thread(); |
1943 | ||
89a02a9f | 1944 | qemu_savevm_state_header(s->to_dst_file); |
1d34e4bf DDAG |
1945 | |
1946 | if (migrate_postcopy_ram()) { | |
1947 | /* Now tell the dest that it should open its end so it can reply */ | |
89a02a9f | 1948 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
1949 | |
1950 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 1951 | qemu_savevm_send_ping(s->to_dst_file, 1); |
1d34e4bf DDAG |
1952 | |
1953 | /* | |
1954 | * Tell the destination that we *might* want to do postcopy later; | |
1955 | * if the other end can't do postcopy it should fail now, nice and | |
1956 | * early. | |
1957 | */ | |
89a02a9f | 1958 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
1959 | } |
1960 | ||
a0762d9e | 1961 | qemu_savevm_state_begin(s->to_dst_file); |
0d82d0e8 | 1962 | |
bc72ad67 | 1963 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
48781e5b HZ |
1964 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1965 | MIGRATION_STATUS_ACTIVE); | |
29ae8a41 | 1966 | |
9ec055ae DDAG |
1967 | trace_migration_thread_setup_complete(); |
1968 | ||
1969 | while (s->state == MIGRATION_STATUS_ACTIVE || | |
1970 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
a3e879cd | 1971 | int64_t current_time; |
c369f40d | 1972 | uint64_t pending_size; |
0d82d0e8 | 1973 | |
89a02a9f | 1974 | if (!qemu_file_rate_limit(s->to_dst_file)) { |
c31b098f DDAG |
1975 | uint64_t pend_post, pend_nonpost; |
1976 | ||
faec066a PX |
1977 | qemu_savevm_state_pending(s->to_dst_file, threshold_size, |
1978 | &pend_nonpost, &pend_post); | |
c31b098f | 1979 | pending_size = pend_nonpost + pend_post; |
faec066a | 1980 | trace_migrate_pending(pending_size, threshold_size, |
c31b098f | 1981 | pend_post, pend_nonpost); |
faec066a | 1982 | if (pending_size && pending_size >= threshold_size) { |
1d34e4bf DDAG |
1983 | /* Still a significant amount to transfer */ |
1984 | ||
1d34e4bf DDAG |
1985 | if (migrate_postcopy_ram() && |
1986 | s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE && | |
faec066a | 1987 | pend_nonpost <= threshold_size && |
1d34e4bf DDAG |
1988 | atomic_read(&s->start_postcopy)) { |
1989 | ||
1990 | if (!postcopy_start(s, &old_vm_running)) { | |
1991 | current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE; | |
1992 | entered_postcopy = true; | |
1993 | } | |
1994 | ||
1995 | continue; | |
1996 | } | |
1997 | /* Just another iteration step */ | |
89a02a9f | 1998 | qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy); |
c369f40d | 1999 | } else { |
09f6c85e | 2000 | trace_migration_thread_low_pending(pending_size); |
1d34e4bf | 2001 | migration_completion(s, current_active_state, |
36f48567 | 2002 | &old_vm_running, &start_time); |
09f6c85e | 2003 | break; |
c369f40d JQ |
2004 | } |
2005 | } | |
f4410a5d | 2006 | |
89a02a9f | 2007 | if (qemu_file_get_error(s->to_dst_file)) { |
48781e5b HZ |
2008 | migrate_set_state(&s->state, current_active_state, |
2009 | MIGRATION_STATUS_FAILED); | |
1d34e4bf | 2010 | trace_migration_thread_file_err(); |
fd45ee2c PB |
2011 | break; |
2012 | } | |
bc72ad67 | 2013 | current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0d82d0e8 | 2014 | if (current_time >= initial_time + BUFFER_DELAY) { |
89a02a9f HZ |
2015 | uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) - |
2016 | initial_bytes; | |
77417f10 | 2017 | uint64_t time_spent = current_time - initial_time; |
a694ee34 | 2018 | double bandwidth = (double)transferred_bytes / time_spent; |
faec066a | 2019 | threshold_size = bandwidth * s->parameters.downtime_limit; |
0d82d0e8 | 2020 | |
5b648de0 WY |
2021 | s->mbps = (((double) transferred_bytes * 8.0) / |
2022 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
7e114f8c | 2023 | |
9013dca5 | 2024 | trace_migrate_transferred(transferred_bytes, time_spent, |
faec066a | 2025 | bandwidth, threshold_size); |
90f8ae72 JQ |
2026 | /* if we haven't sent anything, we don't want to recalculate |
2027 | 10000 is a small enough number for our purposes */ | |
47ad8619 JQ |
2028 | if (ram_dirty_pages_rate() && transferred_bytes > 10000) { |
2029 | s->expected_downtime = ram_dirty_pages_rate() * | |
20afaed9 | 2030 | qemu_target_page_size() / bandwidth; |
90f8ae72 | 2031 | } |
0d82d0e8 | 2032 | |
89a02a9f | 2033 | qemu_file_reset_rate_limit(s->to_dst_file); |
0d82d0e8 | 2034 | initial_time = current_time; |
89a02a9f | 2035 | initial_bytes = qemu_ftell(s->to_dst_file); |
0d82d0e8 | 2036 | } |
89a02a9f | 2037 | if (qemu_file_rate_limit(s->to_dst_file)) { |
0d82d0e8 JQ |
2038 | /* usleep expects microseconds */ |
2039 | g_usleep((initial_time + BUFFER_DELAY - current_time)*1000); | |
2040 | } | |
a3fa1d78 PB |
2041 | } |
2042 | ||
1d34e4bf | 2043 | trace_migration_thread_after_loop(); |
070afca2 JH |
2044 | /* If we enabled cpu throttling for auto-converge, turn it off. */ |
2045 | cpu_throttle_stop(); | |
94f5a437 | 2046 | end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
070afca2 | 2047 | |
f4410a5d | 2048 | qemu_mutex_lock_iothread(); |
0b827d5e HZ |
2049 | /* |
2050 | * The resource has been allocated by migration will be reused in COLO | |
2051 | * process, so don't release them. | |
2052 | */ | |
2053 | if (!enable_colo) { | |
2054 | qemu_savevm_state_cleanup(); | |
2055 | } | |
31194731 | 2056 | if (s->state == MIGRATION_STATUS_COMPLETED) { |
89a02a9f | 2057 | uint64_t transferred_bytes = qemu_ftell(s->to_dst_file); |
a3fa1d78 | 2058 | s->total_time = end_time - s->total_time; |
1d34e4bf DDAG |
2059 | if (!entered_postcopy) { |
2060 | s->downtime = end_time - start_time; | |
2061 | } | |
d6ed7312 PL |
2062 | if (s->total_time) { |
2063 | s->mbps = (((double) transferred_bytes * 8.0) / | |
2064 | ((double) s->total_time)) / 1000; | |
2065 | } | |
a3fa1d78 PB |
2066 | runstate_set(RUN_STATE_POSTMIGRATE); |
2067 | } else { | |
0b827d5e HZ |
2068 | if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) { |
2069 | migrate_start_colo_process(s); | |
2070 | qemu_savevm_state_cleanup(); | |
2071 | /* | |
2072 | * Fixme: we will run VM in COLO no matter its old running state. | |
2073 | * After exited COLO, we will keep running. | |
2074 | */ | |
2075 | old_vm_running = true; | |
2076 | } | |
1d34e4bf | 2077 | if (old_vm_running && !entered_postcopy) { |
a3fa1d78 | 2078 | vm_start(); |
42da5550 DDAG |
2079 | } else { |
2080 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
2081 | runstate_set(RUN_STATE_POSTMIGRATE); | |
2082 | } | |
dba433c0 | 2083 | } |
0d82d0e8 | 2084 | } |
bb1fadc4 | 2085 | qemu_bh_schedule(s->cleanup_bh); |
dba433c0 | 2086 | qemu_mutex_unlock_iothread(); |
f4410a5d | 2087 | |
ab28bd23 | 2088 | rcu_unregister_thread(); |
0d82d0e8 JQ |
2089 | return NULL; |
2090 | } | |
2091 | ||
9848a404 | 2092 | void migrate_fd_connect(MigrationState *s) |
0d82d0e8 | 2093 | { |
2ff30257 | 2094 | s->expected_downtime = s->parameters.downtime_limit; |
bb1fadc4 | 2095 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s); |
0d82d0e8 | 2096 | |
9e4d2b98 | 2097 | qemu_file_set_blocking(s->to_dst_file, true); |
89a02a9f | 2098 | qemu_file_set_rate_limit(s->to_dst_file, |
2ff30257 | 2099 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); |
442773ce | 2100 | |
9287ac27 SH |
2101 | /* Notify before starting migration thread */ |
2102 | notifier_list_notify(&migration_state_notifiers, s); | |
2103 | ||
1d34e4bf DDAG |
2104 | /* |
2105 | * Open the return path; currently for postcopy but other things might | |
2106 | * also want it. | |
2107 | */ | |
2108 | if (migrate_postcopy_ram()) { | |
2109 | if (open_return_path_on_source(s)) { | |
2110 | error_report("Unable to open return-path for postcopy"); | |
48781e5b | 2111 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1d34e4bf DDAG |
2112 | MIGRATION_STATUS_FAILED); |
2113 | migrate_fd_cleanup(s); | |
2114 | return; | |
2115 | } | |
2116 | } | |
2117 | ||
8706d2d5 | 2118 | migrate_compress_threads_create(); |
009fad7f | 2119 | qemu_thread_create(&s->thread, "live_migration", migration_thread, s, |
bb1fadc4 | 2120 | QEMU_THREAD_JOINABLE); |
1d34e4bf | 2121 | s->migration_thread_running = true; |
0d82d0e8 | 2122 | } |
093e3c42 | 2123 |