]>
Commit | Line | Data |
---|---|---|
5bb7910a AL |
1 | /* |
2 | * QEMU live migration | |
3 | * | |
4 | * Copyright IBM, Corp. 2008 | |
5 | * | |
6 | * Authors: | |
7 | * Anthony Liguori <aliguori@us.ibm.com> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
10 | * the COPYING file in the top-level directory. | |
11 | * | |
6b620ca3 PB |
12 | * Contributions after 2012-01-13 are licensed under the terms of the |
13 | * GNU GPL, version 2 or (at your option) any later version. | |
5bb7910a AL |
14 | */ |
15 | ||
1393a485 | 16 | #include "qemu/osdep.h" |
f348b6d1 | 17 | #include "qemu/cutils.h" |
d49b6836 | 18 | #include "qemu/error-report.h" |
795c40b8 | 19 | #include "migration/blocker.h" |
f4dbe1bf | 20 | #include "exec.h" |
7fcac4a2 | 21 | #include "fd.h" |
61e8b148 | 22 | #include "socket.h" |
e1a3ecee | 23 | #include "rdma.h" |
7b1e1a22 | 24 | #include "ram.h" |
84a899de | 25 | #include "migration/global_state.h" |
c4b63b7c | 26 | #include "migration/misc.h" |
6666c96a | 27 | #include "migration.h" |
20a519a0 | 28 | #include "savevm.h" |
40014d81 | 29 | #include "qemu-file-channel.h" |
08a0aee1 | 30 | #include "qemu-file.h" |
987772d9 | 31 | #include "migration/vmstate.h" |
737e150e | 32 | #include "block/block.h" |
e688df6b | 33 | #include "qapi/error.h" |
9af23989 MA |
34 | #include "qapi/qapi-commands-migration.h" |
35 | #include "qapi/qapi-events-migration.h" | |
cc7a8ea7 | 36 | #include "qapi/qmp/qerror.h" |
15280c36 | 37 | #include "qapi/qmp/qnull.h" |
ab28bd23 | 38 | #include "qemu/rcu.h" |
2c9e6fec | 39 | #include "block.h" |
be07b0ac | 40 | #include "postcopy-ram.h" |
766bd176 | 41 | #include "qemu/thread.h" |
c09e5bb1 | 42 | #include "trace.h" |
51180423 | 43 | #include "exec/target_page.h" |
61b67d47 | 44 | #include "io/channel-buffer.h" |
35a6ed4f | 45 | #include "migration/colo.h" |
4ffdb337 | 46 | #include "hw/boards.h" |
9d18af93 | 47 | #include "monitor/monitor.h" |
065e2813 | 48 | |
dc325627 | 49 | #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */ |
5bb7910a | 50 | |
5b4e1eb7 JQ |
51 | /* Amount of time to allocate to each "chunk" of bandwidth-throttled |
52 | * data. */ | |
53 | #define BUFFER_DELAY 100 | |
54 | #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) | |
55 | ||
2ff30257 AA |
56 | /* Time in milliseconds we are allowed to stop the source, |
57 | * for sending the last part */ | |
58 | #define DEFAULT_MIGRATE_SET_DOWNTIME 300 | |
59 | ||
87c9cc1c DHB |
60 | /* Maximum migrate downtime set to 2000 seconds */ |
61 | #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 | |
62 | #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) | |
63 | ||
8706d2d5 LL |
64 | /* Default compression thread count */ |
65 | #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 | |
3fcb38c2 LL |
66 | /* Default decompression thread count, usually decompression is at |
67 | * least 4 times as fast as compression.*/ | |
68 | #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 | |
8706d2d5 LL |
69 | /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ |
70 | #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 | |
1626fee3 | 71 | /* Define default autoconverge cpu throttle migration parameters */ |
d85a31d1 JH |
72 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 |
73 | #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 | |
8706d2d5 | 74 | |
17ad9b35 | 75 | /* Migration XBZRLE default cache size */ |
73af8dd8 | 76 | #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024) |
17ad9b35 | 77 | |
68b53591 HZ |
78 | /* The delay time (in ms) between two COLO checkpoints |
79 | * Note: Please change this default value to 10000 when we support hybrid mode. | |
80 | */ | |
81 | #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200 | |
4075fb1c | 82 | #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2 |
0fb86605 | 83 | #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16 |
68b53591 | 84 | |
99a0db9b GH |
85 | static NotifierList migration_state_notifiers = |
86 | NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); | |
87 | ||
adde220a DDAG |
88 | static bool deferred_incoming; |
89 | ||
da6f1790 JQ |
90 | /* Messages sent on the return path from destination to source */ |
91 | enum mig_rp_message_type { | |
92 | MIG_RP_MSG_INVALID = 0, /* Must be 0 */ | |
93 | MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ | |
94 | MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ | |
95 | ||
96 | MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ | |
97 | MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ | |
98 | ||
99 | MIG_RP_MSG_MAX | |
100 | }; | |
101 | ||
17549e84 JQ |
102 | /* When we add fault tolerance, we could have several |
103 | migrations at once. For now we don't need to add | |
104 | dynamic creation of migration */ | |
105 | ||
e5cb7e76 PX |
106 | static MigrationState *current_migration; |
107 | ||
8b0b29dc | 108 | static bool migration_object_check(MigrationState *ms, Error **errp); |
0331c8ca DDAG |
109 | static int migration_maybe_pause(MigrationState *s, |
110 | int *current_active_state, | |
111 | int new_state); | |
8b0b29dc | 112 | |
e5cb7e76 PX |
113 | void migration_object_init(void) |
114 | { | |
4ffdb337 | 115 | MachineState *ms = MACHINE(qdev_get_machine()); |
8b0b29dc | 116 | Error *err = NULL; |
4ffdb337 | 117 | |
e5cb7e76 PX |
118 | /* This can only be called once. */ |
119 | assert(!current_migration); | |
120 | current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); | |
4ffdb337 | 121 | |
8b0b29dc PX |
122 | if (!migration_object_check(current_migration, &err)) { |
123 | error_report_err(err); | |
124 | exit(1); | |
125 | } | |
126 | ||
4ffdb337 PX |
127 | /* |
128 | * We cannot really do this in migration_instance_init() since at | |
129 | * that time global properties are not yet applied, then this | |
130 | * value will be definitely replaced by something else. | |
131 | */ | |
132 | if (ms->enforce_config_section) { | |
133 | current_migration->send_configuration = true; | |
134 | } | |
e5cb7e76 PX |
135 | } |
136 | ||
1f895604 VSO |
137 | void migration_object_finalize(void) |
138 | { | |
139 | object_unref(OBJECT(current_migration)); | |
140 | } | |
141 | ||
bca7856a | 142 | /* For outgoing */ |
859bc756 | 143 | MigrationState *migrate_get_current(void) |
17549e84 | 144 | { |
e5cb7e76 PX |
145 | /* This can only be called after the object created. */ |
146 | assert(current_migration); | |
147 | return current_migration; | |
17549e84 JQ |
148 | } |
149 | ||
bca7856a DDAG |
150 | MigrationIncomingState *migration_incoming_get_current(void) |
151 | { | |
b4b076da JQ |
152 | static bool once; |
153 | static MigrationIncomingState mis_current; | |
bca7856a | 154 | |
b4b076da JQ |
155 | if (!once) { |
156 | mis_current.state = MIGRATION_STATUS_NONE; | |
157 | memset(&mis_current, 0, sizeof(MigrationIncomingState)); | |
00fa4fc8 DDAG |
158 | mis_current.postcopy_remote_fds = g_array_new(FALSE, TRUE, |
159 | sizeof(struct PostCopyFD)); | |
b4b076da JQ |
160 | qemu_mutex_init(&mis_current.rp_mutex); |
161 | qemu_event_init(&mis_current.main_thread_load_event, false); | |
b411b844 | 162 | qemu_sem_init(&mis_current.postcopy_pause_sem_dst, 0); |
3a7804c3 | 163 | qemu_sem_init(&mis_current.postcopy_pause_sem_fault, 0); |
b35ebdf0 VSO |
164 | |
165 | init_dirty_bitmap_incoming_migration(); | |
166 | ||
b4b076da JQ |
167 | once = true; |
168 | } | |
169 | return &mis_current; | |
bca7856a DDAG |
170 | } |
171 | ||
172 | void migration_incoming_state_destroy(void) | |
173 | { | |
b4b076da JQ |
174 | struct MigrationIncomingState *mis = migration_incoming_get_current(); |
175 | ||
3482655b | 176 | if (mis->to_src_file) { |
660819b1 PX |
177 | /* Tell source that we are done */ |
178 | migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); | |
3482655b PX |
179 | qemu_fclose(mis->to_src_file); |
180 | mis->to_src_file = NULL; | |
181 | } | |
182 | ||
660819b1 PX |
183 | if (mis->from_src_file) { |
184 | qemu_fclose(mis->from_src_file); | |
185 | mis->from_src_file = NULL; | |
186 | } | |
00fa4fc8 DDAG |
187 | if (mis->postcopy_remote_fds) { |
188 | g_array_free(mis->postcopy_remote_fds, TRUE); | |
189 | mis->postcopy_remote_fds = NULL; | |
190 | } | |
660819b1 | 191 | |
5089e186 | 192 | qemu_event_reset(&mis->main_thread_load_event); |
bca7856a DDAG |
193 | } |
194 | ||
b05dc723 JQ |
195 | static void migrate_generate_event(int new_state) |
196 | { | |
197 | if (migrate_use_events()) { | |
198 | qapi_event_send_migration(new_state, &error_abort); | |
b05dc723 JQ |
199 | } |
200 | } | |
201 | ||
adde220a DDAG |
202 | /* |
203 | * Called on -incoming with a defer: uri. | |
204 | * The migration can be started later after any parameters have been | |
205 | * changed. | |
206 | */ | |
207 | static void deferred_incoming_migration(Error **errp) | |
208 | { | |
209 | if (deferred_incoming) { | |
210 | error_setg(errp, "Incoming migration already deferred"); | |
211 | } | |
212 | deferred_incoming = true; | |
213 | } | |
214 | ||
da6f1790 JQ |
215 | /* |
216 | * Send a message on the return channel back to the source | |
217 | * of the migration. | |
218 | */ | |
d6208e35 PX |
219 | static int migrate_send_rp_message(MigrationIncomingState *mis, |
220 | enum mig_rp_message_type message_type, | |
221 | uint16_t len, void *data) | |
da6f1790 | 222 | { |
d6208e35 PX |
223 | int ret = 0; |
224 | ||
da6f1790 JQ |
225 | trace_migrate_send_rp_message((int)message_type, len); |
226 | qemu_mutex_lock(&mis->rp_mutex); | |
d6208e35 PX |
227 | |
228 | /* | |
229 | * It's possible that the file handle got lost due to network | |
230 | * failures. | |
231 | */ | |
232 | if (!mis->to_src_file) { | |
233 | ret = -EIO; | |
234 | goto error; | |
235 | } | |
236 | ||
da6f1790 JQ |
237 | qemu_put_be16(mis->to_src_file, (unsigned int)message_type); |
238 | qemu_put_be16(mis->to_src_file, len); | |
239 | qemu_put_buffer(mis->to_src_file, data, len); | |
240 | qemu_fflush(mis->to_src_file); | |
d6208e35 PX |
241 | |
242 | /* It's possible that qemu file got error during sending */ | |
243 | ret = qemu_file_get_error(mis->to_src_file); | |
244 | ||
245 | error: | |
da6f1790 | 246 | qemu_mutex_unlock(&mis->rp_mutex); |
d6208e35 | 247 | return ret; |
da6f1790 JQ |
248 | } |
249 | ||
1e2d90eb DDAG |
250 | /* Request a range of pages from the source VM at the given |
251 | * start address. | |
252 | * rbname: Name of the RAMBlock to request the page in, if NULL it's the same | |
253 | * as the last request (a name must have been given previously) | |
254 | * Start: Address offset within the RB | |
255 | * Len: Length in bytes required - must be a multiple of pagesize | |
256 | */ | |
d6208e35 PX |
257 | int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname, |
258 | ram_addr_t start, size_t len) | |
1e2d90eb | 259 | { |
cb8d4c8f | 260 | uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ |
1e2d90eb | 261 | size_t msglen = 12; /* start + len */ |
d6208e35 | 262 | enum mig_rp_message_type msg_type; |
1e2d90eb DDAG |
263 | |
264 | *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); | |
265 | *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); | |
266 | ||
267 | if (rbname) { | |
268 | int rbname_len = strlen(rbname); | |
269 | assert(rbname_len < 256); | |
270 | ||
271 | bufc[msglen++] = rbname_len; | |
272 | memcpy(bufc + msglen, rbname, rbname_len); | |
273 | msglen += rbname_len; | |
d6208e35 | 274 | msg_type = MIG_RP_MSG_REQ_PAGES_ID; |
1e2d90eb | 275 | } else { |
d6208e35 | 276 | msg_type = MIG_RP_MSG_REQ_PAGES; |
1e2d90eb | 277 | } |
d6208e35 PX |
278 | |
279 | return migrate_send_rp_message(mis, msg_type, msglen, bufc); | |
1e2d90eb DDAG |
280 | } |
281 | ||
43eaae28 | 282 | void qemu_start_incoming_migration(const char *uri, Error **errp) |
5bb7910a | 283 | { |
34c9dd8e AL |
284 | const char *p; |
285 | ||
7cf1fe6d | 286 | qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort); |
adde220a DDAG |
287 | if (!strcmp(uri, "defer")) { |
288 | deferred_incoming_migration(errp); | |
289 | } else if (strstart(uri, "tcp:", &p)) { | |
43eaae28 | 290 | tcp_start_incoming_migration(p, errp); |
2da776db | 291 | #ifdef CONFIG_RDMA |
adde220a | 292 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
293 | rdma_start_incoming_migration(p, errp); |
294 | #endif | |
adde220a | 295 | } else if (strstart(uri, "exec:", &p)) { |
43eaae28 | 296 | exec_start_incoming_migration(p, errp); |
adde220a | 297 | } else if (strstart(uri, "unix:", &p)) { |
43eaae28 | 298 | unix_start_incoming_migration(p, errp); |
adde220a | 299 | } else if (strstart(uri, "fd:", &p)) { |
43eaae28 | 300 | fd_start_incoming_migration(p, errp); |
adde220a | 301 | } else { |
312fd5f2 | 302 | error_setg(errp, "unknown migration protocol: %s", uri); |
8ca5e801 | 303 | } |
5bb7910a AL |
304 | } |
305 | ||
0aa6aefc DL |
306 | static void process_incoming_migration_bh(void *opaque) |
307 | { | |
308 | Error *local_err = NULL; | |
309 | MigrationIncomingState *mis = opaque; | |
310 | ||
a18a73d7 DDAG |
311 | /* Make sure all file formats flush their mutable metadata. |
312 | * If we get an error here, just don't restart the VM yet. */ | |
313 | bdrv_invalidate_cache_all(&local_err); | |
314 | if (local_err) { | |
315 | error_report_err(local_err); | |
316 | local_err = NULL; | |
317 | autostart = false; | |
d35ff5e6 KW |
318 | } |
319 | ||
0aa6aefc DL |
320 | /* |
321 | * This must happen after all error conditions are dealt with and | |
322 | * we're sure the VM is going to be running on this host. | |
323 | */ | |
324 | qemu_announce_self(); | |
325 | ||
f986c3d2 JQ |
326 | if (multifd_load_cleanup(&local_err) != 0) { |
327 | error_report_err(local_err); | |
328 | autostart = false; | |
329 | } | |
0aa6aefc DL |
330 | /* If global state section was not received or we are in running |
331 | state, we need to obey autostart. Any other state is set with | |
332 | runstate_set. */ | |
333 | ||
b35ebdf0 VSO |
334 | dirty_bitmap_mig_before_vm_start(); |
335 | ||
0aa6aefc DL |
336 | if (!global_state_received() || |
337 | global_state_get_runstate() == RUN_STATE_RUNNING) { | |
338 | if (autostart) { | |
339 | vm_start(); | |
340 | } else { | |
341 | runstate_set(RUN_STATE_PAUSED); | |
342 | } | |
343 | } else { | |
344 | runstate_set(global_state_get_runstate()); | |
345 | } | |
0aa6aefc DL |
346 | /* |
347 | * This must happen after any state changes since as soon as an external | |
348 | * observer sees this event they might start to prod at the VM assuming | |
349 | * it's ready to use. | |
350 | */ | |
351 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, | |
352 | MIGRATION_STATUS_COMPLETED); | |
353 | qemu_bh_delete(mis->bh); | |
354 | migration_incoming_state_destroy(); | |
355 | } | |
356 | ||
82a4da79 | 357 | static void process_incoming_migration_co(void *opaque) |
511c0231 | 358 | { |
b4b076da | 359 | MigrationIncomingState *mis = migration_incoming_get_current(); |
e9bef235 | 360 | PostcopyState ps; |
1c12e1f5 PB |
361 | int ret; |
362 | ||
4f0fae7f | 363 | assert(mis->from_src_file); |
67f11b5c | 364 | mis->largest_page_size = qemu_ram_pagesize_largest(); |
093e3c42 | 365 | postcopy_state_set(POSTCOPY_INCOMING_NONE); |
93d7af6f HZ |
366 | migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, |
367 | MIGRATION_STATUS_ACTIVE); | |
4f0fae7f | 368 | ret = qemu_loadvm_state(mis->from_src_file); |
bca7856a | 369 | |
e9bef235 DDAG |
370 | ps = postcopy_state_get(); |
371 | trace_process_incoming_migration_co_end(ret, ps); | |
372 | if (ps != POSTCOPY_INCOMING_NONE) { | |
373 | if (ps == POSTCOPY_INCOMING_ADVISE) { | |
374 | /* | |
375 | * Where a migration had postcopy enabled (and thus went to advise) | |
376 | * but managed to complete within the precopy period, we can use | |
377 | * the normal exit. | |
378 | */ | |
379 | postcopy_ram_incoming_cleanup(mis); | |
380 | } else if (ret >= 0) { | |
381 | /* | |
382 | * Postcopy was started, cleanup should happen at the end of the | |
383 | * postcopy thread. | |
384 | */ | |
385 | trace_process_incoming_migration_co_postcopy_end_main(); | |
386 | return; | |
387 | } | |
388 | /* Else if something went wrong then just fall out of the normal exit */ | |
389 | } | |
390 | ||
25d0c16f HZ |
391 | /* we get COLO info, and know if we are in COLO mode */ |
392 | if (!ret && migration_incoming_enable_colo()) { | |
393 | mis->migration_incoming_co = qemu_coroutine_self(); | |
394 | qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", | |
395 | colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); | |
396 | mis->have_colo_incoming_thread = true; | |
397 | qemu_coroutine_yield(); | |
398 | ||
399 | /* Wait checkpoint incoming thread exit before free resource */ | |
400 | qemu_thread_join(&mis->colo_incoming_thread); | |
401 | } | |
402 | ||
1c12e1f5 | 403 | if (ret < 0) { |
f986c3d2 JQ |
404 | Error *local_err = NULL; |
405 | ||
93d7af6f HZ |
406 | migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, |
407 | MIGRATION_STATUS_FAILED); | |
db80face | 408 | error_report("load of migration failed: %s", strerror(-ret)); |
3a0f2cea | 409 | qemu_fclose(mis->from_src_file); |
f986c3d2 JQ |
410 | if (multifd_load_cleanup(&local_err) != 0) { |
411 | error_report_err(local_err); | |
412 | } | |
4aead692 | 413 | exit(EXIT_FAILURE); |
511c0231 | 414 | } |
0aa6aefc DL |
415 | mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); |
416 | qemu_bh_schedule(mis->bh); | |
511c0231 JQ |
417 | } |
418 | ||
e595a01a | 419 | static void migration_incoming_setup(QEMUFile *f) |
82a4da79 | 420 | { |
4f0fae7f | 421 | MigrationIncomingState *mis = migration_incoming_get_current(); |
82a4da79 | 422 | |
f986c3d2 JQ |
423 | if (multifd_load_setup() != 0) { |
424 | /* We haven't been able to create multifd threads | |
425 | nothing better to do */ | |
426 | exit(EXIT_FAILURE); | |
427 | } | |
428 | ||
4f0fae7f JQ |
429 | if (!mis->from_src_file) { |
430 | mis->from_src_file = f; | |
431 | } | |
06ad5135 | 432 | qemu_file_set_blocking(f, false); |
e595a01a JQ |
433 | } |
434 | ||
36c2f8be | 435 | void migration_incoming_process(void) |
e595a01a JQ |
436 | { |
437 | Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL); | |
0b8b8753 | 438 | qemu_coroutine_enter(co); |
82a4da79 PB |
439 | } |
440 | ||
e595a01a JQ |
441 | void migration_fd_process_incoming(QEMUFile *f) |
442 | { | |
443 | migration_incoming_setup(f); | |
444 | migration_incoming_process(); | |
445 | } | |
446 | ||
4f0fae7f JQ |
447 | void migration_ioc_process_incoming(QIOChannel *ioc) |
448 | { | |
449 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
450 | ||
451 | if (!mis->from_src_file) { | |
452 | QEMUFile *f = qemu_fopen_channel_input(ioc); | |
36c2f8be | 453 | migration_incoming_setup(f); |
71bb07db | 454 | return; |
4f0fae7f | 455 | } |
71bb07db | 456 | multifd_recv_new_channel(ioc); |
4f0fae7f JQ |
457 | } |
458 | ||
428d8908 JQ |
459 | /** |
460 | * @migration_has_all_channels: We have received all channels that we need | |
461 | * | |
462 | * Returns true when we have got connections to all the channels that | |
463 | * we need for migration. | |
464 | */ | |
465 | bool migration_has_all_channels(void) | |
466 | { | |
62c1e0ca JQ |
467 | bool all_channels; |
468 | ||
469 | all_channels = multifd_recv_all_channels_created(); | |
470 | ||
471 | return all_channels; | |
428d8908 JQ |
472 | } |
473 | ||
6decec93 DDAG |
474 | /* |
475 | * Send a 'SHUT' message on the return channel with the given value | |
476 | * to indicate that we've finished with the RP. Non-0 value indicates | |
477 | * error. | |
478 | */ | |
479 | void migrate_send_rp_shut(MigrationIncomingState *mis, | |
480 | uint32_t value) | |
481 | { | |
482 | uint32_t buf; | |
483 | ||
484 | buf = cpu_to_be32(value); | |
485 | migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); | |
486 | } | |
487 | ||
488 | /* | |
489 | * Send a 'PONG' message on the return channel with the given value | |
490 | * (normally in response to a 'PING') | |
491 | */ | |
492 | void migrate_send_rp_pong(MigrationIncomingState *mis, | |
493 | uint32_t value) | |
494 | { | |
495 | uint32_t buf; | |
496 | ||
497 | buf = cpu_to_be32(value); | |
498 | migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); | |
499 | } | |
500 | ||
bbf6da32 OW |
501 | MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) |
502 | { | |
503 | MigrationCapabilityStatusList *head = NULL; | |
504 | MigrationCapabilityStatusList *caps; | |
505 | MigrationState *s = migrate_get_current(); | |
506 | int i; | |
507 | ||
387eedeb | 508 | caps = NULL; /* silence compiler warning */ |
7fb1cf16 | 509 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
ed1701c6 DDAG |
510 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
511 | if (i == MIGRATION_CAPABILITY_BLOCK) { | |
512 | continue; | |
513 | } | |
514 | #endif | |
bbf6da32 OW |
515 | if (head == NULL) { |
516 | head = g_malloc0(sizeof(*caps)); | |
517 | caps = head; | |
518 | } else { | |
519 | caps->next = g_malloc0(sizeof(*caps)); | |
520 | caps = caps->next; | |
521 | } | |
522 | caps->value = | |
523 | g_malloc(sizeof(*caps->value)); | |
524 | caps->value->capability = i; | |
525 | caps->value->state = s->enabled_capabilities[i]; | |
526 | } | |
527 | ||
528 | return head; | |
529 | } | |
530 | ||
85de8323 LL |
531 | MigrationParameters *qmp_query_migrate_parameters(Error **errp) |
532 | { | |
533 | MigrationParameters *params; | |
534 | MigrationState *s = migrate_get_current(); | |
535 | ||
e87fae4c | 536 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
85de8323 | 537 | params = g_malloc0(sizeof(*params)); |
de63ab61 | 538 | params->has_compress_level = true; |
2594f56d | 539 | params->compress_level = s->parameters.compress_level; |
de63ab61 | 540 | params->has_compress_threads = true; |
2594f56d | 541 | params->compress_threads = s->parameters.compress_threads; |
de63ab61 | 542 | params->has_decompress_threads = true; |
2594f56d | 543 | params->decompress_threads = s->parameters.decompress_threads; |
de63ab61 | 544 | params->has_cpu_throttle_initial = true; |
2594f56d | 545 | params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; |
de63ab61 | 546 | params->has_cpu_throttle_increment = true; |
2594f56d | 547 | params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; |
8cc99dcd | 548 | params->has_tls_creds = true; |
69ef1f36 | 549 | params->tls_creds = g_strdup(s->parameters.tls_creds); |
8cc99dcd | 550 | params->has_tls_hostname = true; |
69ef1f36 | 551 | params->tls_hostname = g_strdup(s->parameters.tls_hostname); |
2ff30257 AA |
552 | params->has_max_bandwidth = true; |
553 | params->max_bandwidth = s->parameters.max_bandwidth; | |
554 | params->has_downtime_limit = true; | |
555 | params->downtime_limit = s->parameters.downtime_limit; | |
fe39a4d4 | 556 | params->has_x_checkpoint_delay = true; |
68b53591 | 557 | params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; |
2833c59b JQ |
558 | params->has_block_incremental = true; |
559 | params->block_incremental = s->parameters.block_incremental; | |
4075fb1c JQ |
560 | params->has_x_multifd_channels = true; |
561 | params->x_multifd_channels = s->parameters.x_multifd_channels; | |
0fb86605 JQ |
562 | params->has_x_multifd_page_count = true; |
563 | params->x_multifd_page_count = s->parameters.x_multifd_page_count; | |
73af8dd8 JQ |
564 | params->has_xbzrle_cache_size = true; |
565 | params->xbzrle_cache_size = s->parameters.xbzrle_cache_size; | |
85de8323 LL |
566 | |
567 | return params; | |
568 | } | |
569 | ||
f6844b99 DDAG |
570 | /* |
571 | * Return true if we're already in the middle of a migration | |
572 | * (i.e. any of the active or setup states) | |
573 | */ | |
574 | static bool migration_is_setup_or_active(int state) | |
575 | { | |
576 | switch (state) { | |
577 | case MIGRATION_STATUS_ACTIVE: | |
9ec055ae | 578 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
a688d2c1 | 579 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
f6844b99 | 580 | case MIGRATION_STATUS_SETUP: |
31e06077 DDAG |
581 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
582 | case MIGRATION_STATUS_DEVICE: | |
f6844b99 DDAG |
583 | return true; |
584 | ||
585 | default: | |
586 | return false; | |
587 | ||
588 | } | |
589 | } | |
590 | ||
a22463a5 DDAG |
591 | static void populate_ram_info(MigrationInfo *info, MigrationState *s) |
592 | { | |
593 | info->has_ram = true; | |
594 | info->ram = g_malloc0(sizeof(*info->ram)); | |
9360447d | 595 | info->ram->transferred = ram_counters.transferred; |
a22463a5 | 596 | info->ram->total = ram_bytes_total(); |
9360447d | 597 | info->ram->duplicate = ram_counters.duplicate; |
bedf53c1 JQ |
598 | /* legacy value. It is not used anymore */ |
599 | info->ram->skipped = 0; | |
9360447d JQ |
600 | info->ram->normal = ram_counters.normal; |
601 | info->ram->normal_bytes = ram_counters.normal * | |
20afaed9 | 602 | qemu_target_page_size(); |
a22463a5 | 603 | info->ram->mbps = s->mbps; |
9360447d JQ |
604 | info->ram->dirty_sync_count = ram_counters.dirty_sync_count; |
605 | info->ram->postcopy_requests = ram_counters.postcopy_requests; | |
030ce1f8 | 606 | info->ram->page_size = qemu_target_page_size(); |
a22463a5 | 607 | |
114f5aee JQ |
608 | if (migrate_use_xbzrle()) { |
609 | info->has_xbzrle_cache = true; | |
610 | info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); | |
611 | info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); | |
9360447d JQ |
612 | info->xbzrle_cache->bytes = xbzrle_counters.bytes; |
613 | info->xbzrle_cache->pages = xbzrle_counters.pages; | |
614 | info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss; | |
615 | info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate; | |
616 | info->xbzrle_cache->overflow = xbzrle_counters.overflow; | |
114f5aee JQ |
617 | } |
618 | ||
338182c8 JQ |
619 | if (cpu_throttle_active()) { |
620 | info->has_cpu_throttle_percentage = true; | |
621 | info->cpu_throttle_percentage = cpu_throttle_get_percentage(); | |
622 | } | |
623 | ||
a22463a5 DDAG |
624 | if (s->state != MIGRATION_STATUS_COMPLETED) { |
625 | info->ram->remaining = ram_bytes_remaining(); | |
9360447d | 626 | info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate; |
a22463a5 DDAG |
627 | } |
628 | } | |
629 | ||
930ac04c JQ |
630 | static void populate_disk_info(MigrationInfo *info) |
631 | { | |
632 | if (blk_mig_active()) { | |
633 | info->has_disk = true; | |
634 | info->disk = g_malloc0(sizeof(*info->disk)); | |
635 | info->disk->transferred = blk_mig_bytes_transferred(); | |
636 | info->disk->remaining = blk_mig_bytes_remaining(); | |
637 | info->disk->total = blk_mig_bytes_total(); | |
638 | } | |
639 | } | |
640 | ||
65ace060 | 641 | static void fill_source_migration_info(MigrationInfo *info) |
5bb7910a | 642 | { |
17549e84 JQ |
643 | MigrationState *s = migrate_get_current(); |
644 | ||
645 | switch (s->state) { | |
31194731 | 646 | case MIGRATION_STATUS_NONE: |
17549e84 | 647 | /* no migration has happened ever */ |
65ace060 AP |
648 | /* do not overwrite destination migration status */ |
649 | return; | |
17549e84 | 650 | break; |
31194731 | 651 | case MIGRATION_STATUS_SETUP: |
29ae8a41 | 652 | info->has_status = true; |
ed4fbd10 | 653 | info->has_total_time = false; |
29ae8a41 | 654 | break; |
31194731 HZ |
655 | case MIGRATION_STATUS_ACTIVE: |
656 | case MIGRATION_STATUS_CANCELLING: | |
9ec055ae | 657 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: |
31e06077 DDAG |
658 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
659 | case MIGRATION_STATUS_DEVICE: | |
a688d2c1 | 660 | case MIGRATION_STATUS_POSTCOPY_PAUSED: |
c8f9f4f4 | 661 | /* TODO add some postcopy stats */ |
9ec055ae DDAG |
662 | info->has_status = true; |
663 | info->has_total_time = true; | |
664 | info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) | |
4af246a3 | 665 | - s->start_time; |
9ec055ae DDAG |
666 | info->has_expected_downtime = true; |
667 | info->expected_downtime = s->expected_downtime; | |
668 | info->has_setup_time = true; | |
669 | info->setup_time = s->setup_time; | |
670 | ||
a22463a5 | 671 | populate_ram_info(info, s); |
930ac04c | 672 | populate_disk_info(info); |
17549e84 | 673 | break; |
0b827d5e HZ |
674 | case MIGRATION_STATUS_COLO: |
675 | info->has_status = true; | |
676 | /* TODO: display COLO specific information (checkpoint info etc.) */ | |
677 | break; | |
31194731 | 678 | case MIGRATION_STATUS_COMPLETED: |
791e7c82 | 679 | info->has_status = true; |
00c14997 | 680 | info->has_total_time = true; |
7aa939af | 681 | info->total_time = s->total_time; |
9c5a9fcf JQ |
682 | info->has_downtime = true; |
683 | info->downtime = s->downtime; | |
ed4fbd10 MH |
684 | info->has_setup_time = true; |
685 | info->setup_time = s->setup_time; | |
d5f8a570 | 686 | |
a22463a5 | 687 | populate_ram_info(info, s); |
17549e84 | 688 | break; |
31194731 | 689 | case MIGRATION_STATUS_FAILED: |
791e7c82 | 690 | info->has_status = true; |
d59ce6f3 DB |
691 | if (s->error) { |
692 | info->has_error_desc = true; | |
693 | info->error_desc = g_strdup(error_get_pretty(s->error)); | |
694 | } | |
17549e84 | 695 | break; |
31194731 | 696 | case MIGRATION_STATUS_CANCELLED: |
791e7c82 | 697 | info->has_status = true; |
17549e84 | 698 | break; |
5bb7910a | 699 | } |
cde63fbe | 700 | info->status = s->state; |
5bb7910a AL |
701 | } |
702 | ||
4a84214e PX |
703 | /** |
704 | * @migration_caps_check - check capability validity | |
705 | * | |
706 | * @cap_list: old capability list, array of bool | |
707 | * @params: new capabilities to be applied soon | |
708 | * @errp: set *errp if the check failed, with reason | |
709 | * | |
710 | * Returns true if check passed, otherwise false. | |
711 | */ | |
712 | static bool migrate_caps_check(bool *cap_list, | |
713 | MigrationCapabilityStatusList *params, | |
714 | Error **errp) | |
00458433 | 715 | { |
00458433 | 716 | MigrationCapabilityStatusList *cap; |
4a84214e | 717 | bool old_postcopy_cap; |
d7651f15 | 718 | MigrationIncomingState *mis = migration_incoming_get_current(); |
00458433 | 719 | |
4a84214e | 720 | old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
00458433 OW |
721 | |
722 | for (cap = params; cap; cap = cap->next) { | |
4a84214e PX |
723 | cap_list[cap->value->capability] = cap->value->state; |
724 | } | |
725 | ||
ed1701c6 | 726 | #ifndef CONFIG_LIVE_BLOCK_MIGRATION |
4a84214e PX |
727 | if (cap_list[MIGRATION_CAPABILITY_BLOCK]) { |
728 | error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) " | |
729 | "block migration"); | |
730 | error_append_hint(errp, "Use drive_mirror+NBD instead.\n"); | |
731 | return false; | |
00458433 | 732 | } |
4a84214e | 733 | #endif |
53dd370c | 734 | |
4a84214e PX |
735 | if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) { |
736 | if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) { | |
53dd370c DDAG |
737 | /* The decompression threads asynchronously write into RAM |
738 | * rather than use the atomic copies needed to avoid | |
739 | * userfaulting. It should be possible to fix the decompression | |
740 | * threads for compatibility in future. | |
741 | */ | |
4a84214e PX |
742 | error_setg(errp, "Postcopy is not currently compatible " |
743 | "with compression"); | |
744 | return false; | |
53dd370c | 745 | } |
4a84214e | 746 | |
096631bd DDAG |
747 | /* This check is reasonably expensive, so only when it's being |
748 | * set the first time, also it's only the destination that needs | |
749 | * special support. | |
750 | */ | |
751 | if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) && | |
d7651f15 | 752 | !postcopy_ram_supported_by_host(mis)) { |
096631bd DDAG |
753 | /* postcopy_ram_supported_by_host will have emitted a more |
754 | * detailed message | |
755 | */ | |
4a84214e PX |
756 | error_setg(errp, "Postcopy is not supported"); |
757 | return false; | |
096631bd | 758 | } |
53dd370c | 759 | } |
4a84214e PX |
760 | |
761 | return true; | |
762 | } | |
763 | ||
65ace060 AP |
764 | static void fill_destination_migration_info(MigrationInfo *info) |
765 | { | |
766 | MigrationIncomingState *mis = migration_incoming_get_current(); | |
767 | ||
768 | switch (mis->state) { | |
769 | case MIGRATION_STATUS_NONE: | |
770 | return; | |
771 | break; | |
772 | case MIGRATION_STATUS_SETUP: | |
773 | case MIGRATION_STATUS_CANCELLING: | |
774 | case MIGRATION_STATUS_CANCELLED: | |
775 | case MIGRATION_STATUS_ACTIVE: | |
776 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
777 | case MIGRATION_STATUS_FAILED: | |
778 | case MIGRATION_STATUS_COLO: | |
779 | info->has_status = true; | |
780 | break; | |
781 | case MIGRATION_STATUS_COMPLETED: | |
782 | info->has_status = true; | |
783 | fill_destination_postcopy_migration_info(info); | |
784 | break; | |
785 | } | |
786 | info->status = mis->state; | |
787 | } | |
788 | ||
789 | MigrationInfo *qmp_query_migrate(Error **errp) | |
790 | { | |
791 | MigrationInfo *info = g_malloc0(sizeof(*info)); | |
792 | ||
793 | fill_destination_migration_info(info); | |
794 | fill_source_migration_info(info); | |
795 | ||
796 | return info; | |
797 | } | |
798 | ||
4a84214e PX |
799 | void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params, |
800 | Error **errp) | |
801 | { | |
802 | MigrationState *s = migrate_get_current(); | |
803 | MigrationCapabilityStatusList *cap; | |
dd0ee30c | 804 | bool cap_list[MIGRATION_CAPABILITY__MAX]; |
4a84214e PX |
805 | |
806 | if (migration_is_setup_or_active(s->state)) { | |
807 | error_setg(errp, QERR_MIGRATION_ACTIVE); | |
808 | return; | |
809 | } | |
810 | ||
dd0ee30c PX |
811 | memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list)); |
812 | if (!migrate_caps_check(cap_list, params, errp)) { | |
4a84214e PX |
813 | return; |
814 | } | |
815 | ||
816 | for (cap = params; cap; cap = cap->next) { | |
817 | s->enabled_capabilities[cap->value->capability] = cap->value->state; | |
818 | } | |
00458433 OW |
819 | } |
820 | ||
16d063bc PX |
821 | /* |
822 | * Check whether the parameters are valid. Error will be put into errp | |
823 | * (if provided). Return true if valid, otherwise false. | |
824 | */ | |
825 | static bool migrate_params_check(MigrationParameters *params, Error **errp) | |
85de8323 | 826 | { |
7f375e04 | 827 | if (params->has_compress_level && |
741d4086 | 828 | (params->compress_level > 9)) { |
c6bd8c70 MA |
829 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", |
830 | "is invalid, it should be in the range of 0 to 9"); | |
16d063bc | 831 | return false; |
85de8323 | 832 | } |
16d063bc | 833 | |
741d4086 | 834 | if (params->has_compress_threads && (params->compress_threads < 1)) { |
c6bd8c70 MA |
835 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
836 | "compress_threads", | |
837 | "is invalid, it should be in the range of 1 to 255"); | |
16d063bc | 838 | return false; |
85de8323 | 839 | } |
16d063bc | 840 | |
741d4086 | 841 | if (params->has_decompress_threads && (params->decompress_threads < 1)) { |
c6bd8c70 MA |
842 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
843 | "decompress_threads", | |
844 | "is invalid, it should be in the range of 1 to 255"); | |
16d063bc | 845 | return false; |
85de8323 | 846 | } |
16d063bc | 847 | |
7f375e04 EB |
848 | if (params->has_cpu_throttle_initial && |
849 | (params->cpu_throttle_initial < 1 || | |
850 | params->cpu_throttle_initial > 99)) { | |
1626fee3 | 851 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 852 | "cpu_throttle_initial", |
1626fee3 | 853 | "an integer in the range of 1 to 99"); |
16d063bc | 854 | return false; |
1626fee3 | 855 | } |
16d063bc | 856 | |
7f375e04 EB |
857 | if (params->has_cpu_throttle_increment && |
858 | (params->cpu_throttle_increment < 1 || | |
859 | params->cpu_throttle_increment > 99)) { | |
1626fee3 | 860 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
d85a31d1 | 861 | "cpu_throttle_increment", |
1626fee3 | 862 | "an integer in the range of 1 to 99"); |
16d063bc | 863 | return false; |
1626fee3 | 864 | } |
16d063bc | 865 | |
741d4086 | 866 | if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) { |
2ff30257 AA |
867 | error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the" |
868 | " range of 0 to %zu bytes/second", SIZE_MAX); | |
16d063bc | 869 | return false; |
2ff30257 | 870 | } |
16d063bc | 871 | |
2ff30257 | 872 | if (params->has_downtime_limit && |
741d4086 | 873 | (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { |
87c9cc1c DHB |
874 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " |
875 | "the range of 0 to %d milliseconds", | |
876 | MAX_MIGRATE_DOWNTIME); | |
16d063bc | 877 | return false; |
2ff30257 | 878 | } |
16d063bc | 879 | |
741d4086 JQ |
880 | /* x_checkpoint_delay is now always positive */ |
881 | ||
882 | if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) { | |
4075fb1c JQ |
883 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
884 | "multifd_channels", | |
885 | "is invalid, it should be in the range of 1 to 255"); | |
886 | return false; | |
887 | } | |
0fb86605 | 888 | if (params->has_x_multifd_page_count && |
741d4086 JQ |
889 | (params->x_multifd_page_count < 1 || |
890 | params->x_multifd_page_count > 10000)) { | |
0fb86605 JQ |
891 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, |
892 | "multifd_page_count", | |
893 | "is invalid, it should be in the range of 1 to 10000"); | |
894 | return false; | |
895 | } | |
16d063bc | 896 | |
73af8dd8 JQ |
897 | if (params->has_xbzrle_cache_size && |
898 | (params->xbzrle_cache_size < qemu_target_page_size() || | |
899 | !is_power_of_2(params->xbzrle_cache_size))) { | |
900 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, | |
901 | "xbzrle_cache_size", | |
902 | "is invalid, it should be bigger than target page size" | |
903 | " and a power of two"); | |
904 | return false; | |
905 | } | |
906 | ||
16d063bc PX |
907 | return true; |
908 | } | |
909 | ||
1bda8b3c MA |
910 | static void migrate_params_test_apply(MigrateSetParameters *params, |
911 | MigrationParameters *dest) | |
912 | { | |
913 | *dest = migrate_get_current()->parameters; | |
914 | ||
915 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ | |
916 | ||
917 | if (params->has_compress_level) { | |
918 | dest->compress_level = params->compress_level; | |
919 | } | |
920 | ||
921 | if (params->has_compress_threads) { | |
922 | dest->compress_threads = params->compress_threads; | |
923 | } | |
924 | ||
925 | if (params->has_decompress_threads) { | |
926 | dest->decompress_threads = params->decompress_threads; | |
927 | } | |
928 | ||
929 | if (params->has_cpu_throttle_initial) { | |
930 | dest->cpu_throttle_initial = params->cpu_throttle_initial; | |
931 | } | |
932 | ||
933 | if (params->has_cpu_throttle_increment) { | |
934 | dest->cpu_throttle_increment = params->cpu_throttle_increment; | |
935 | } | |
936 | ||
937 | if (params->has_tls_creds) { | |
01fa5598 MA |
938 | assert(params->tls_creds->type == QTYPE_QSTRING); |
939 | dest->tls_creds = g_strdup(params->tls_creds->u.s); | |
1bda8b3c MA |
940 | } |
941 | ||
942 | if (params->has_tls_hostname) { | |
01fa5598 MA |
943 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
944 | dest->tls_hostname = g_strdup(params->tls_hostname->u.s); | |
1bda8b3c MA |
945 | } |
946 | ||
947 | if (params->has_max_bandwidth) { | |
948 | dest->max_bandwidth = params->max_bandwidth; | |
949 | } | |
950 | ||
951 | if (params->has_downtime_limit) { | |
952 | dest->downtime_limit = params->downtime_limit; | |
953 | } | |
954 | ||
955 | if (params->has_x_checkpoint_delay) { | |
956 | dest->x_checkpoint_delay = params->x_checkpoint_delay; | |
957 | } | |
958 | ||
959 | if (params->has_block_incremental) { | |
960 | dest->block_incremental = params->block_incremental; | |
961 | } | |
5e7577a1 JQ |
962 | if (params->has_x_multifd_channels) { |
963 | dest->x_multifd_channels = params->x_multifd_channels; | |
964 | } | |
965 | if (params->has_x_multifd_page_count) { | |
966 | dest->x_multifd_page_count = params->x_multifd_page_count; | |
967 | } | |
73af8dd8 JQ |
968 | if (params->has_xbzrle_cache_size) { |
969 | dest->xbzrle_cache_size = params->xbzrle_cache_size; | |
970 | } | |
1bda8b3c MA |
971 | } |
972 | ||
73af8dd8 | 973 | static void migrate_params_apply(MigrateSetParameters *params, Error **errp) |
16d063bc PX |
974 | { |
975 | MigrationState *s = migrate_get_current(); | |
976 | ||
e87fae4c MA |
977 | /* TODO use QAPI_CLONE() instead of duplicating it inline */ |
978 | ||
7f375e04 EB |
979 | if (params->has_compress_level) { |
980 | s->parameters.compress_level = params->compress_level; | |
85de8323 | 981 | } |
476c72aa | 982 | |
7f375e04 EB |
983 | if (params->has_compress_threads) { |
984 | s->parameters.compress_threads = params->compress_threads; | |
85de8323 | 985 | } |
476c72aa | 986 | |
7f375e04 EB |
987 | if (params->has_decompress_threads) { |
988 | s->parameters.decompress_threads = params->decompress_threads; | |
85de8323 | 989 | } |
476c72aa | 990 | |
7f375e04 EB |
991 | if (params->has_cpu_throttle_initial) { |
992 | s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; | |
1626fee3 | 993 | } |
476c72aa | 994 | |
7f375e04 EB |
995 | if (params->has_cpu_throttle_increment) { |
996 | s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; | |
1626fee3 | 997 | } |
476c72aa | 998 | |
7f375e04 | 999 | if (params->has_tls_creds) { |
69ef1f36 | 1000 | g_free(s->parameters.tls_creds); |
01fa5598 MA |
1001 | assert(params->tls_creds->type == QTYPE_QSTRING); |
1002 | s->parameters.tls_creds = g_strdup(params->tls_creds->u.s); | |
69ef1f36 | 1003 | } |
476c72aa | 1004 | |
7f375e04 | 1005 | if (params->has_tls_hostname) { |
69ef1f36 | 1006 | g_free(s->parameters.tls_hostname); |
01fa5598 MA |
1007 | assert(params->tls_hostname->type == QTYPE_QSTRING); |
1008 | s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s); | |
69ef1f36 | 1009 | } |
476c72aa | 1010 | |
2ff30257 AA |
1011 | if (params->has_max_bandwidth) { |
1012 | s->parameters.max_bandwidth = params->max_bandwidth; | |
1013 | if (s->to_dst_file) { | |
1014 | qemu_file_set_rate_limit(s->to_dst_file, | |
1015 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); | |
1016 | } | |
1017 | } | |
476c72aa | 1018 | |
2ff30257 AA |
1019 | if (params->has_downtime_limit) { |
1020 | s->parameters.downtime_limit = params->downtime_limit; | |
1021 | } | |
68b53591 HZ |
1022 | |
1023 | if (params->has_x_checkpoint_delay) { | |
1024 | s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; | |
479125d5 HZ |
1025 | if (migration_in_colo_state()) { |
1026 | colo_checkpoint_notify(s); | |
1027 | } | |
68b53591 | 1028 | } |
476c72aa | 1029 | |
2833c59b JQ |
1030 | if (params->has_block_incremental) { |
1031 | s->parameters.block_incremental = params->block_incremental; | |
1032 | } | |
4075fb1c JQ |
1033 | if (params->has_x_multifd_channels) { |
1034 | s->parameters.x_multifd_channels = params->x_multifd_channels; | |
1035 | } | |
0fb86605 JQ |
1036 | if (params->has_x_multifd_page_count) { |
1037 | s->parameters.x_multifd_page_count = params->x_multifd_page_count; | |
1038 | } | |
73af8dd8 JQ |
1039 | if (params->has_xbzrle_cache_size) { |
1040 | s->parameters.xbzrle_cache_size = params->xbzrle_cache_size; | |
1041 | xbzrle_cache_resize(params->xbzrle_cache_size, errp); | |
1042 | } | |
85de8323 LL |
1043 | } |
1044 | ||
1bda8b3c | 1045 | void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp) |
476c72aa | 1046 | { |
1bda8b3c MA |
1047 | MigrationParameters tmp; |
1048 | ||
01fa5598 MA |
1049 | /* TODO Rewrite "" to null instead */ |
1050 | if (params->has_tls_creds | |
1051 | && params->tls_creds->type == QTYPE_QNULL) { | |
cb3e7f08 | 1052 | qobject_unref(params->tls_creds->u.n); |
01fa5598 MA |
1053 | params->tls_creds->type = QTYPE_QSTRING; |
1054 | params->tls_creds->u.s = strdup(""); | |
1055 | } | |
1056 | /* TODO Rewrite "" to null instead */ | |
1057 | if (params->has_tls_hostname | |
1058 | && params->tls_hostname->type == QTYPE_QNULL) { | |
cb3e7f08 | 1059 | qobject_unref(params->tls_hostname->u.n); |
01fa5598 MA |
1060 | params->tls_hostname->type = QTYPE_QSTRING; |
1061 | params->tls_hostname->u.s = strdup(""); | |
1062 | } | |
1063 | ||
1bda8b3c MA |
1064 | migrate_params_test_apply(params, &tmp); |
1065 | ||
1066 | if (!migrate_params_check(&tmp, errp)) { | |
476c72aa PX |
1067 | /* Invalid parameter */ |
1068 | return; | |
1069 | } | |
1070 | ||
73af8dd8 | 1071 | migrate_params_apply(params, errp); |
476c72aa PX |
1072 | } |
1073 | ||
2594f56d | 1074 | |
4886a1bc DDAG |
1075 | void qmp_migrate_start_postcopy(Error **errp) |
1076 | { | |
1077 | MigrationState *s = migrate_get_current(); | |
1078 | ||
16b0fd32 | 1079 | if (!migrate_postcopy()) { |
a54d340b | 1080 | error_setg(errp, "Enable postcopy with migrate_set_capability before" |
4886a1bc DDAG |
1081 | " the start of migration"); |
1082 | return; | |
1083 | } | |
1084 | ||
1085 | if (s->state == MIGRATION_STATUS_NONE) { | |
1086 | error_setg(errp, "Postcopy must be started after migration has been" | |
1087 | " started"); | |
1088 | return; | |
1089 | } | |
1090 | /* | |
1091 | * we don't error if migration has finished since that would be racy | |
1092 | * with issuing this command. | |
1093 | */ | |
1094 | atomic_set(&s->start_postcopy, true); | |
1095 | } | |
1096 | ||
065e2813 AL |
1097 | /* shared migration helpers */ |
1098 | ||
48781e5b | 1099 | void migrate_set_state(int *state, int old_state, int new_state) |
51cf4c1a | 1100 | { |
a31fedee | 1101 | assert(new_state < MIGRATION_STATUS__MAX); |
48781e5b | 1102 | if (atomic_cmpxchg(state, old_state, new_state) == old_state) { |
a31fedee | 1103 | trace_migrate_set_state(MigrationStatus_str(new_state)); |
b05dc723 | 1104 | migrate_generate_event(new_state); |
51cf4c1a Z |
1105 | } |
1106 | } | |
1107 | ||
4e4a3d3a PX |
1108 | static MigrationCapabilityStatusList *migrate_cap_add( |
1109 | MigrationCapabilityStatusList *list, | |
1110 | MigrationCapability index, | |
1111 | bool state) | |
2833c59b JQ |
1112 | { |
1113 | MigrationCapabilityStatusList *cap; | |
1114 | ||
1115 | cap = g_new0(MigrationCapabilityStatusList, 1); | |
1116 | cap->value = g_new0(MigrationCapabilityStatus, 1); | |
4e4a3d3a PX |
1117 | cap->value->capability = index; |
1118 | cap->value->state = state; | |
1119 | cap->next = list; | |
1120 | ||
1121 | return cap; | |
1122 | } | |
1123 | ||
1124 | void migrate_set_block_enabled(bool value, Error **errp) | |
1125 | { | |
1126 | MigrationCapabilityStatusList *cap; | |
1127 | ||
1128 | cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value); | |
2833c59b JQ |
1129 | qmp_migrate_set_capabilities(cap, errp); |
1130 | qapi_free_MigrationCapabilityStatusList(cap); | |
1131 | } | |
1132 | ||
1133 | static void migrate_set_block_incremental(MigrationState *s, bool value) | |
1134 | { | |
1135 | s->parameters.block_incremental = value; | |
1136 | } | |
1137 | ||
1138 | static void block_cleanup_parameters(MigrationState *s) | |
1139 | { | |
1140 | if (s->must_remove_block_options) { | |
1141 | /* setting to false can never fail */ | |
1142 | migrate_set_block_enabled(false, &error_abort); | |
1143 | migrate_set_block_incremental(s, false); | |
1144 | s->must_remove_block_options = false; | |
1145 | } | |
1146 | } | |
1147 | ||
bb1fadc4 | 1148 | static void migrate_fd_cleanup(void *opaque) |
065e2813 | 1149 | { |
bb1fadc4 PB |
1150 | MigrationState *s = opaque; |
1151 | ||
1152 | qemu_bh_delete(s->cleanup_bh); | |
1153 | s->cleanup_bh = NULL; | |
1154 | ||
0ceccd85 PX |
1155 | qemu_savevm_state_cleanup(); |
1156 | ||
89a02a9f | 1157 | if (s->to_dst_file) { |
f986c3d2 JQ |
1158 | Error *local_err = NULL; |
1159 | ||
9013dca5 | 1160 | trace_migrate_fd_cleanup(); |
404a7c05 | 1161 | qemu_mutex_unlock_iothread(); |
1d34e4bf DDAG |
1162 | if (s->migration_thread_running) { |
1163 | qemu_thread_join(&s->thread); | |
1164 | s->migration_thread_running = false; | |
1165 | } | |
404a7c05 PB |
1166 | qemu_mutex_lock_iothread(); |
1167 | ||
f986c3d2 JQ |
1168 | if (multifd_save_cleanup(&local_err) != 0) { |
1169 | error_report_err(local_err); | |
1170 | } | |
89a02a9f HZ |
1171 | qemu_fclose(s->to_dst_file); |
1172 | s->to_dst_file = NULL; | |
065e2813 AL |
1173 | } |
1174 | ||
9ec055ae DDAG |
1175 | assert((s->state != MIGRATION_STATUS_ACTIVE) && |
1176 | (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE)); | |
7a2c1721 | 1177 | |
94f5a437 | 1178 | if (s->state == MIGRATION_STATUS_CANCELLING) { |
48781e5b | 1179 | migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, |
94f5a437 | 1180 | MIGRATION_STATUS_CANCELLED); |
7a2c1721 | 1181 | } |
a3fa1d78 | 1182 | |
87db1a7d JQ |
1183 | if (s->error) { |
1184 | /* It is used on info migrate. We can't free it */ | |
1185 | error_report_err(error_copy(s->error)); | |
1186 | } | |
a3fa1d78 | 1187 | notifier_list_notify(&migration_state_notifiers, s); |
2833c59b | 1188 | block_cleanup_parameters(s); |
065e2813 AL |
1189 | } |
1190 | ||
87db1a7d JQ |
1191 | void migrate_set_error(MigrationState *s, const Error *error) |
1192 | { | |
1193 | qemu_mutex_lock(&s->error_mutex); | |
1194 | if (!s->error) { | |
1195 | s->error = error_copy(error); | |
1196 | } | |
1197 | qemu_mutex_unlock(&s->error_mutex); | |
1198 | } | |
1199 | ||
d59ce6f3 | 1200 | void migrate_fd_error(MigrationState *s, const Error *error) |
065e2813 | 1201 | { |
25174055 | 1202 | trace_migrate_fd_error(error_get_pretty(error)); |
89a02a9f | 1203 | assert(s->to_dst_file == NULL); |
48781e5b HZ |
1204 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1205 | MIGRATION_STATUS_FAILED); | |
87db1a7d | 1206 | migrate_set_error(s, error); |
458cf28e JQ |
1207 | } |
1208 | ||
0edda1c4 | 1209 | static void migrate_fd_cancel(MigrationState *s) |
065e2813 | 1210 | { |
6f2b811a | 1211 | int old_state ; |
89a02a9f | 1212 | QEMUFile *f = migrate_get_current()->to_dst_file; |
9013dca5 | 1213 | trace_migrate_fd_cancel(); |
065e2813 | 1214 | |
70b20477 DDAG |
1215 | if (s->rp_state.from_dst_file) { |
1216 | /* shutdown the rp socket, so causing the rp thread to shutdown */ | |
1217 | qemu_file_shutdown(s->rp_state.from_dst_file); | |
1218 | } | |
1219 | ||
6f2b811a Z |
1220 | do { |
1221 | old_state = s->state; | |
f6844b99 | 1222 | if (!migration_is_setup_or_active(old_state)) { |
6f2b811a Z |
1223 | break; |
1224 | } | |
a7b36b48 DDAG |
1225 | /* If the migration is paused, kick it out of the pause */ |
1226 | if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) { | |
1227 | qemu_sem_post(&s->pause_sem); | |
1228 | } | |
48781e5b | 1229 | migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); |
31194731 | 1230 | } while (s->state != MIGRATION_STATUS_CANCELLING); |
a26ba26e DDAG |
1231 | |
1232 | /* | |
1233 | * If we're unlucky the migration code might be stuck somewhere in a | |
1234 | * send/write while the network has failed and is waiting to timeout; | |
1235 | * if we've got shutdown(2) available then we can force it to quit. | |
1236 | * The outgoing qemu file gets closed in migrate_fd_cleanup that is | |
1237 | * called in a bh, so there is no race against this cancel. | |
1238 | */ | |
31194731 | 1239 | if (s->state == MIGRATION_STATUS_CANCELLING && f) { |
a26ba26e DDAG |
1240 | qemu_file_shutdown(f); |
1241 | } | |
1d2acc31 HZ |
1242 | if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { |
1243 | Error *local_err = NULL; | |
1244 | ||
1245 | bdrv_invalidate_cache_all(&local_err); | |
1246 | if (local_err) { | |
1247 | error_report_err(local_err); | |
1248 | } else { | |
1249 | s->block_inactive = false; | |
1250 | } | |
1251 | } | |
065e2813 AL |
1252 | } |
1253 | ||
99a0db9b GH |
1254 | void add_migration_state_change_notifier(Notifier *notify) |
1255 | { | |
1256 | notifier_list_add(&migration_state_notifiers, notify); | |
1257 | } | |
1258 | ||
1259 | void remove_migration_state_change_notifier(Notifier *notify) | |
1260 | { | |
31552529 | 1261 | notifier_remove(notify); |
99a0db9b GH |
1262 | } |
1263 | ||
02edd2e7 | 1264 | bool migration_in_setup(MigrationState *s) |
afe2df69 | 1265 | { |
31194731 | 1266 | return s->state == MIGRATION_STATUS_SETUP; |
afe2df69 GH |
1267 | } |
1268 | ||
7073693b | 1269 | bool migration_has_finished(MigrationState *s) |
99a0db9b | 1270 | { |
31194731 | 1271 | return s->state == MIGRATION_STATUS_COMPLETED; |
99a0db9b | 1272 | } |
0edda1c4 | 1273 | |
afe2df69 GH |
1274 | bool migration_has_failed(MigrationState *s) |
1275 | { | |
31194731 HZ |
1276 | return (s->state == MIGRATION_STATUS_CANCELLED || |
1277 | s->state == MIGRATION_STATUS_FAILED); | |
afe2df69 GH |
1278 | } |
1279 | ||
5727309d | 1280 | bool migration_in_postcopy(void) |
9ec055ae | 1281 | { |
5727309d JQ |
1282 | MigrationState *s = migrate_get_current(); |
1283 | ||
9ec055ae DDAG |
1284 | return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); |
1285 | } | |
1286 | ||
b82fc321 DDAG |
1287 | bool migration_in_postcopy_after_devices(MigrationState *s) |
1288 | { | |
5727309d | 1289 | return migration_in_postcopy() && s->postcopy_after_devices; |
b82fc321 DDAG |
1290 | } |
1291 | ||
fab35005 | 1292 | bool migration_is_idle(void) |
fe44dc91 | 1293 | { |
fab35005 | 1294 | MigrationState *s = migrate_get_current(); |
fe44dc91 AA |
1295 | |
1296 | switch (s->state) { | |
1297 | case MIGRATION_STATUS_NONE: | |
1298 | case MIGRATION_STATUS_CANCELLED: | |
1299 | case MIGRATION_STATUS_COMPLETED: | |
1300 | case MIGRATION_STATUS_FAILED: | |
1301 | return true; | |
1302 | case MIGRATION_STATUS_SETUP: | |
1303 | case MIGRATION_STATUS_CANCELLING: | |
1304 | case MIGRATION_STATUS_ACTIVE: | |
1305 | case MIGRATION_STATUS_POSTCOPY_ACTIVE: | |
1306 | case MIGRATION_STATUS_COLO: | |
31e06077 DDAG |
1307 | case MIGRATION_STATUS_PRE_SWITCHOVER: |
1308 | case MIGRATION_STATUS_DEVICE: | |
fe44dc91 AA |
1309 | return false; |
1310 | case MIGRATION_STATUS__MAX: | |
1311 | g_assert_not_reached(); | |
1312 | } | |
1313 | ||
1314 | return false; | |
1315 | } | |
1316 | ||
3e0c8050 | 1317 | void migrate_init(MigrationState *s) |
0edda1c4 | 1318 | { |
389775d1 DDAG |
1319 | /* |
1320 | * Reinitialise all migration state, except | |
1321 | * parameters/capabilities that the user set, and | |
1322 | * locks. | |
1323 | */ | |
1324 | s->bytes_xfer = 0; | |
1325 | s->xfer_limit = 0; | |
1326 | s->cleanup_bh = 0; | |
89a02a9f | 1327 | s->to_dst_file = NULL; |
389775d1 | 1328 | s->state = MIGRATION_STATUS_NONE; |
389775d1 DDAG |
1329 | s->rp_state.from_dst_file = NULL; |
1330 | s->rp_state.error = false; | |
1331 | s->mbps = 0.0; | |
1332 | s->downtime = 0; | |
1333 | s->expected_downtime = 0; | |
389775d1 | 1334 | s->setup_time = 0; |
389775d1 | 1335 | s->start_postcopy = false; |
b82fc321 | 1336 | s->postcopy_after_devices = false; |
389775d1 | 1337 | s->migration_thread_running = false; |
d59ce6f3 DB |
1338 | error_free(s->error); |
1339 | s->error = NULL; | |
389775d1 | 1340 | |
48781e5b | 1341 | migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); |
0edda1c4 | 1342 | |
4af246a3 PX |
1343 | s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
1344 | s->total_time = 0; | |
7287cbd4 | 1345 | s->vm_was_running = false; |
b15df1ae PX |
1346 | s->iteration_initial_bytes = 0; |
1347 | s->threshold_size = 0; | |
0edda1c4 | 1348 | } |
cab30143 | 1349 | |
fa2756b7 AL |
1350 | static GSList *migration_blockers; |
1351 | ||
fe44dc91 | 1352 | int migrate_add_blocker(Error *reason, Error **errp) |
fa2756b7 | 1353 | { |
3df663e5 | 1354 | if (migrate_get_current()->only_migratable) { |
b67b8c3a AA |
1355 | error_propagate(errp, error_copy(reason)); |
1356 | error_prepend(errp, "disallowing migration blocker " | |
1357 | "(--only_migratable) for: "); | |
1358 | return -EACCES; | |
1359 | } | |
1360 | ||
fab35005 | 1361 | if (migration_is_idle()) { |
fe44dc91 AA |
1362 | migration_blockers = g_slist_prepend(migration_blockers, reason); |
1363 | return 0; | |
1364 | } | |
1365 | ||
1366 | error_propagate(errp, error_copy(reason)); | |
1367 | error_prepend(errp, "disallowing migration blocker (migration in " | |
1368 | "progress) for: "); | |
1369 | return -EBUSY; | |
fa2756b7 AL |
1370 | } |
1371 | ||
1372 | void migrate_del_blocker(Error *reason) | |
1373 | { | |
1374 | migration_blockers = g_slist_remove(migration_blockers, reason); | |
1375 | } | |
1376 | ||
bf1ae1f4 DDAG |
1377 | void qmp_migrate_incoming(const char *uri, Error **errp) |
1378 | { | |
1379 | Error *local_err = NULL; | |
4debb5f5 | 1380 | static bool once = true; |
bf1ae1f4 DDAG |
1381 | |
1382 | if (!deferred_incoming) { | |
4debb5f5 | 1383 | error_setg(errp, "For use with '-incoming defer'"); |
bf1ae1f4 DDAG |
1384 | return; |
1385 | } | |
4debb5f5 DDAG |
1386 | if (!once) { |
1387 | error_setg(errp, "The incoming migration has already been started"); | |
1388 | } | |
bf1ae1f4 DDAG |
1389 | |
1390 | qemu_start_incoming_migration(uri, &local_err); | |
1391 | ||
1392 | if (local_err) { | |
1393 | error_propagate(errp, local_err); | |
1394 | return; | |
1395 | } | |
1396 | ||
4debb5f5 | 1397 | once = false; |
bf1ae1f4 DDAG |
1398 | } |
1399 | ||
24f3902b GK |
1400 | bool migration_is_blocked(Error **errp) |
1401 | { | |
1402 | if (qemu_savevm_state_blocked(errp)) { | |
1403 | return true; | |
1404 | } | |
1405 | ||
1406 | if (migration_blockers) { | |
250561e1 | 1407 | error_propagate(errp, error_copy(migration_blockers->data)); |
24f3902b GK |
1408 | return true; |
1409 | } | |
1410 | ||
1411 | return false; | |
1412 | } | |
1413 | ||
e1c37d0e LC |
1414 | void qmp_migrate(const char *uri, bool has_blk, bool blk, |
1415 | bool has_inc, bool inc, bool has_detach, bool detach, | |
7a4da28b | 1416 | bool has_resume, bool resume, Error **errp) |
cab30143 | 1417 | { |
be7059cd | 1418 | Error *local_err = NULL; |
17549e84 | 1419 | MigrationState *s = migrate_get_current(); |
cab30143 | 1420 | const char *p; |
cab30143 | 1421 | |
f6844b99 | 1422 | if (migration_is_setup_or_active(s->state) || |
0b827d5e HZ |
1423 | s->state == MIGRATION_STATUS_CANCELLING || |
1424 | s->state == MIGRATION_STATUS_COLO) { | |
c6bd8c70 | 1425 | error_setg(errp, QERR_MIGRATION_ACTIVE); |
e1c37d0e | 1426 | return; |
cab30143 | 1427 | } |
ca99993a DDAG |
1428 | if (runstate_check(RUN_STATE_INMIGRATE)) { |
1429 | error_setg(errp, "Guest is waiting for an incoming migration"); | |
1430 | return; | |
1431 | } | |
1432 | ||
24f3902b | 1433 | if (migration_is_blocked(errp)) { |
e1c37d0e | 1434 | return; |
fa2756b7 AL |
1435 | } |
1436 | ||
2833c59b JQ |
1437 | if ((has_blk && blk) || (has_inc && inc)) { |
1438 | if (migrate_use_block() || migrate_use_block_incremental()) { | |
1439 | error_setg(errp, "Command options are incompatible with " | |
1440 | "current migration capabilities"); | |
1441 | return; | |
1442 | } | |
1443 | migrate_set_block_enabled(true, &local_err); | |
1444 | if (local_err) { | |
1445 | error_propagate(errp, local_err); | |
1446 | return; | |
1447 | } | |
1448 | s->must_remove_block_options = true; | |
1449 | } | |
1450 | ||
1451 | if (has_inc && inc) { | |
1452 | migrate_set_block_incremental(s, true); | |
1453 | } | |
1454 | ||
3e0c8050 | 1455 | migrate_init(s); |
cab30143 JQ |
1456 | |
1457 | if (strstart(uri, "tcp:", &p)) { | |
f37afb5a | 1458 | tcp_start_outgoing_migration(s, p, &local_err); |
2da776db | 1459 | #ifdef CONFIG_RDMA |
41310c68 | 1460 | } else if (strstart(uri, "rdma:", &p)) { |
2da776db MH |
1461 | rdma_start_outgoing_migration(s, p, &local_err); |
1462 | #endif | |
cab30143 | 1463 | } else if (strstart(uri, "exec:", &p)) { |
f37afb5a | 1464 | exec_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1465 | } else if (strstart(uri, "unix:", &p)) { |
f37afb5a | 1466 | unix_start_outgoing_migration(s, p, &local_err); |
cab30143 | 1467 | } else if (strstart(uri, "fd:", &p)) { |
f37afb5a | 1468 | fd_start_outgoing_migration(s, p, &local_err); |
99a0db9b | 1469 | } else { |
c6bd8c70 MA |
1470 | error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", |
1471 | "a valid migration protocol"); | |
48781e5b HZ |
1472 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1473 | MIGRATION_STATUS_FAILED); | |
09576e74 | 1474 | block_cleanup_parameters(s); |
e1c37d0e | 1475 | return; |
cab30143 JQ |
1476 | } |
1477 | ||
f37afb5a | 1478 | if (local_err) { |
d59ce6f3 | 1479 | migrate_fd_error(s, local_err); |
f37afb5a | 1480 | error_propagate(errp, local_err); |
e1c37d0e | 1481 | return; |
1299c631 | 1482 | } |
cab30143 JQ |
1483 | } |
1484 | ||
6cdedb07 | 1485 | void qmp_migrate_cancel(Error **errp) |
cab30143 | 1486 | { |
17549e84 | 1487 | migrate_fd_cancel(migrate_get_current()); |
cab30143 JQ |
1488 | } |
1489 | ||
89cfc02c DDAG |
1490 | void qmp_migrate_continue(MigrationStatus state, Error **errp) |
1491 | { | |
1492 | MigrationState *s = migrate_get_current(); | |
1493 | if (s->state != state) { | |
1494 | error_setg(errp, "Migration not in expected state: %s", | |
1495 | MigrationStatus_str(s->state)); | |
1496 | return; | |
1497 | } | |
1498 | qemu_sem_post(&s->pause_sem); | |
1499 | } | |
1500 | ||
9e1ba4cc OW |
1501 | void qmp_migrate_set_cache_size(int64_t value, Error **errp) |
1502 | { | |
73af8dd8 JQ |
1503 | MigrateSetParameters p = { |
1504 | .has_xbzrle_cache_size = true, | |
1505 | .xbzrle_cache_size = value, | |
1506 | }; | |
c91e681a | 1507 | |
73af8dd8 | 1508 | qmp_migrate_set_parameters(&p, errp); |
9e1ba4cc OW |
1509 | } |
1510 | ||
1511 | int64_t qmp_query_migrate_cache_size(Error **errp) | |
1512 | { | |
1513 | return migrate_xbzrle_cache_size(); | |
1514 | } | |
1515 | ||
3dc85383 | 1516 | void qmp_migrate_set_speed(int64_t value, Error **errp) |
cab30143 | 1517 | { |
1bda8b3c | 1518 | MigrateSetParameters p = { |
2ff30257 AA |
1519 | .has_max_bandwidth = true, |
1520 | .max_bandwidth = value, | |
1521 | }; | |
cab30143 | 1522 | |
2ff30257 | 1523 | qmp_migrate_set_parameters(&p, errp); |
cab30143 JQ |
1524 | } |
1525 | ||
4f0a993b | 1526 | void qmp_migrate_set_downtime(double value, Error **errp) |
cab30143 | 1527 | { |
87c9cc1c DHB |
1528 | if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) { |
1529 | error_setg(errp, "Parameter 'downtime_limit' expects an integer in " | |
1530 | "the range of 0 to %d seconds", | |
1531 | MAX_MIGRATE_DOWNTIME_SECONDS); | |
1532 | return; | |
1533 | } | |
1534 | ||
2ff30257 AA |
1535 | value *= 1000; /* Convert to milliseconds */ |
1536 | value = MAX(0, MIN(INT64_MAX, value)); | |
1537 | ||
1bda8b3c | 1538 | MigrateSetParameters p = { |
2ff30257 AA |
1539 | .has_downtime_limit = true, |
1540 | .downtime_limit = value, | |
1541 | }; | |
1542 | ||
1543 | qmp_migrate_set_parameters(&p, errp); | |
99a0db9b | 1544 | } |
17ad9b35 | 1545 | |
53f09a10 PB |
1546 | bool migrate_release_ram(void) |
1547 | { | |
1548 | MigrationState *s; | |
1549 | ||
1550 | s = migrate_get_current(); | |
1551 | ||
1552 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; | |
1553 | } | |
1554 | ||
53dd370c DDAG |
1555 | bool migrate_postcopy_ram(void) |
1556 | { | |
1557 | MigrationState *s; | |
1558 | ||
1559 | s = migrate_get_current(); | |
1560 | ||
32c3db5b | 1561 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; |
53dd370c DDAG |
1562 | } |
1563 | ||
58110f0a VSO |
1564 | bool migrate_postcopy(void) |
1565 | { | |
dd6bb914 | 1566 | return migrate_postcopy_ram() || migrate_dirty_bitmaps(); |
58110f0a VSO |
1567 | } |
1568 | ||
bde1e2ec CV |
1569 | bool migrate_auto_converge(void) |
1570 | { | |
1571 | MigrationState *s; | |
1572 | ||
1573 | s = migrate_get_current(); | |
1574 | ||
1575 | return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; | |
1576 | } | |
1577 | ||
323004a3 PL |
1578 | bool migrate_zero_blocks(void) |
1579 | { | |
1580 | MigrationState *s; | |
1581 | ||
1582 | s = migrate_get_current(); | |
1583 | ||
1584 | return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; | |
1585 | } | |
1586 | ||
f22f928e AP |
1587 | bool migrate_postcopy_blocktime(void) |
1588 | { | |
1589 | MigrationState *s; | |
1590 | ||
1591 | s = migrate_get_current(); | |
1592 | ||
1593 | return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME]; | |
1594 | } | |
1595 | ||
8706d2d5 LL |
1596 | bool migrate_use_compression(void) |
1597 | { | |
dde4e694 LL |
1598 | MigrationState *s; |
1599 | ||
1600 | s = migrate_get_current(); | |
1601 | ||
1602 | return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; | |
8706d2d5 LL |
1603 | } |
1604 | ||
1605 | int migrate_compress_level(void) | |
1606 | { | |
1607 | MigrationState *s; | |
1608 | ||
1609 | s = migrate_get_current(); | |
1610 | ||
2594f56d | 1611 | return s->parameters.compress_level; |
8706d2d5 LL |
1612 | } |
1613 | ||
1614 | int migrate_compress_threads(void) | |
1615 | { | |
1616 | MigrationState *s; | |
1617 | ||
1618 | s = migrate_get_current(); | |
1619 | ||
2594f56d | 1620 | return s->parameters.compress_threads; |
8706d2d5 LL |
1621 | } |
1622 | ||
3fcb38c2 LL |
1623 | int migrate_decompress_threads(void) |
1624 | { | |
1625 | MigrationState *s; | |
1626 | ||
1627 | s = migrate_get_current(); | |
1628 | ||
2594f56d | 1629 | return s->parameters.decompress_threads; |
3fcb38c2 LL |
1630 | } |
1631 | ||
55efc8c2 VSO |
1632 | bool migrate_dirty_bitmaps(void) |
1633 | { | |
1634 | MigrationState *s; | |
1635 | ||
1636 | s = migrate_get_current(); | |
1637 | ||
1638 | return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS]; | |
1639 | } | |
1640 | ||
b05dc723 JQ |
1641 | bool migrate_use_events(void) |
1642 | { | |
1643 | MigrationState *s; | |
1644 | ||
1645 | s = migrate_get_current(); | |
1646 | ||
1647 | return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; | |
1648 | } | |
1649 | ||
30126bbf JQ |
1650 | bool migrate_use_multifd(void) |
1651 | { | |
1652 | MigrationState *s; | |
1653 | ||
1654 | s = migrate_get_current(); | |
1655 | ||
1656 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD]; | |
1657 | } | |
1658 | ||
93fbd031 DDAG |
1659 | bool migrate_pause_before_switchover(void) |
1660 | { | |
1661 | MigrationState *s; | |
1662 | ||
1663 | s = migrate_get_current(); | |
1664 | ||
1665 | return s->enabled_capabilities[ | |
1666 | MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER]; | |
1667 | } | |
1668 | ||
4075fb1c JQ |
1669 | int migrate_multifd_channels(void) |
1670 | { | |
1671 | MigrationState *s; | |
1672 | ||
1673 | s = migrate_get_current(); | |
1674 | ||
1675 | return s->parameters.x_multifd_channels; | |
1676 | } | |
1677 | ||
0fb86605 JQ |
1678 | int migrate_multifd_page_count(void) |
1679 | { | |
1680 | MigrationState *s; | |
1681 | ||
1682 | s = migrate_get_current(); | |
1683 | ||
1684 | return s->parameters.x_multifd_page_count; | |
1685 | } | |
1686 | ||
17ad9b35 OW |
1687 | int migrate_use_xbzrle(void) |
1688 | { | |
1689 | MigrationState *s; | |
1690 | ||
1691 | s = migrate_get_current(); | |
1692 | ||
1693 | return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; | |
1694 | } | |
1695 | ||
1696 | int64_t migrate_xbzrle_cache_size(void) | |
1697 | { | |
1698 | MigrationState *s; | |
1699 | ||
1700 | s = migrate_get_current(); | |
1701 | ||
73af8dd8 | 1702 | return s->parameters.xbzrle_cache_size; |
17ad9b35 | 1703 | } |
0d82d0e8 | 1704 | |
2833c59b JQ |
1705 | bool migrate_use_block(void) |
1706 | { | |
1707 | MigrationState *s; | |
1708 | ||
1709 | s = migrate_get_current(); | |
1710 | ||
1711 | return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK]; | |
1712 | } | |
1713 | ||
c788ada8 PX |
1714 | bool migrate_use_return_path(void) |
1715 | { | |
1716 | MigrationState *s; | |
1717 | ||
1718 | s = migrate_get_current(); | |
1719 | ||
1720 | return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH]; | |
1721 | } | |
1722 | ||
2833c59b JQ |
1723 | bool migrate_use_block_incremental(void) |
1724 | { | |
1725 | MigrationState *s; | |
1726 | ||
1727 | s = migrate_get_current(); | |
1728 | ||
1729 | return s->parameters.block_incremental; | |
1730 | } | |
1731 | ||
70b20477 DDAG |
1732 | /* migration thread support */ |
1733 | /* | |
1734 | * Something bad happened to the RP stream, mark an error | |
1735 | * The caller shall print or trace something to indicate why | |
1736 | */ | |
1737 | static void mark_source_rp_bad(MigrationState *s) | |
1738 | { | |
1739 | s->rp_state.error = true; | |
1740 | } | |
1741 | ||
1742 | static struct rp_cmd_args { | |
1743 | ssize_t len; /* -1 = variable */ | |
1744 | const char *name; | |
1745 | } rp_cmd_args[] = { | |
1746 | [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, | |
1747 | [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, | |
1748 | [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, | |
1e2d90eb DDAG |
1749 | [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, |
1750 | [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, | |
70b20477 DDAG |
1751 | [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, |
1752 | }; | |
1753 | ||
1e2d90eb DDAG |
1754 | /* |
1755 | * Process a request for pages received on the return path, | |
1756 | * We're allowed to send more than requested (e.g. to round to our page size) | |
1757 | * and we don't need to send pages that have already been sent. | |
1758 | */ | |
1759 | static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, | |
1760 | ram_addr_t start, size_t len) | |
1761 | { | |
6c595cde DDAG |
1762 | long our_host_ps = getpagesize(); |
1763 | ||
1e2d90eb | 1764 | trace_migrate_handle_rp_req_pages(rbname, start, len); |
6c595cde DDAG |
1765 | |
1766 | /* | |
1767 | * Since we currently insist on matching page sizes, just sanity check | |
1768 | * we're being asked for whole host pages. | |
1769 | */ | |
1770 | if (start & (our_host_ps-1) || | |
1771 | (len & (our_host_ps-1))) { | |
1772 | error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT | |
1773 | " len: %zd", __func__, start, len); | |
1774 | mark_source_rp_bad(ms); | |
1775 | return; | |
1776 | } | |
1777 | ||
96506894 | 1778 | if (ram_save_queue_pages(rbname, start, len)) { |
6c595cde DDAG |
1779 | mark_source_rp_bad(ms); |
1780 | } | |
1e2d90eb DDAG |
1781 | } |
1782 | ||
14b1742e PX |
1783 | /* Return true to retry, false to quit */ |
1784 | static bool postcopy_pause_return_path_thread(MigrationState *s) | |
1785 | { | |
1786 | trace_postcopy_pause_return_path(); | |
1787 | ||
1788 | qemu_sem_wait(&s->postcopy_pause_rp_sem); | |
1789 | ||
1790 | trace_postcopy_pause_return_path_continued(); | |
1791 | ||
1792 | return true; | |
1793 | } | |
1794 | ||
70b20477 DDAG |
1795 | /* |
1796 | * Handles messages sent on the return path towards the source VM | |
1797 | * | |
1798 | */ | |
1799 | static void *source_return_path_thread(void *opaque) | |
1800 | { | |
1801 | MigrationState *ms = opaque; | |
1802 | QEMUFile *rp = ms->rp_state.from_dst_file; | |
1803 | uint16_t header_len, header_type; | |
568b01ca | 1804 | uint8_t buf[512]; |
70b20477 | 1805 | uint32_t tmp32, sibling_error; |
1e2d90eb DDAG |
1806 | ram_addr_t start = 0; /* =0 to silence warning */ |
1807 | size_t len = 0, expected_len; | |
70b20477 DDAG |
1808 | int res; |
1809 | ||
1810 | trace_source_return_path_thread_entry(); | |
14b1742e PX |
1811 | |
1812 | retry: | |
70b20477 DDAG |
1813 | while (!ms->rp_state.error && !qemu_file_get_error(rp) && |
1814 | migration_is_setup_or_active(ms->state)) { | |
1815 | trace_source_return_path_thread_loop_top(); | |
1816 | header_type = qemu_get_be16(rp); | |
1817 | header_len = qemu_get_be16(rp); | |
1818 | ||
7a9ddfbf PX |
1819 | if (qemu_file_get_error(rp)) { |
1820 | mark_source_rp_bad(ms); | |
1821 | goto out; | |
1822 | } | |
1823 | ||
70b20477 DDAG |
1824 | if (header_type >= MIG_RP_MSG_MAX || |
1825 | header_type == MIG_RP_MSG_INVALID) { | |
1826 | error_report("RP: Received invalid message 0x%04x length 0x%04x", | |
1827 | header_type, header_len); | |
1828 | mark_source_rp_bad(ms); | |
1829 | goto out; | |
1830 | } | |
1831 | ||
1832 | if ((rp_cmd_args[header_type].len != -1 && | |
1833 | header_len != rp_cmd_args[header_type].len) || | |
568b01ca | 1834 | header_len > sizeof(buf)) { |
70b20477 DDAG |
1835 | error_report("RP: Received '%s' message (0x%04x) with" |
1836 | "incorrect length %d expecting %zu", | |
1837 | rp_cmd_args[header_type].name, header_type, header_len, | |
1838 | (size_t)rp_cmd_args[header_type].len); | |
1839 | mark_source_rp_bad(ms); | |
1840 | goto out; | |
1841 | } | |
1842 | ||
1843 | /* We know we've got a valid header by this point */ | |
1844 | res = qemu_get_buffer(rp, buf, header_len); | |
1845 | if (res != header_len) { | |
1846 | error_report("RP: Failed reading data for message 0x%04x" | |
1847 | " read %d expected %d", | |
1848 | header_type, res, header_len); | |
1849 | mark_source_rp_bad(ms); | |
1850 | goto out; | |
1851 | } | |
1852 | ||
1853 | /* OK, we have the message and the data */ | |
1854 | switch (header_type) { | |
1855 | case MIG_RP_MSG_SHUT: | |
4d885131 | 1856 | sibling_error = ldl_be_p(buf); |
70b20477 DDAG |
1857 | trace_source_return_path_thread_shut(sibling_error); |
1858 | if (sibling_error) { | |
1859 | error_report("RP: Sibling indicated error %d", sibling_error); | |
1860 | mark_source_rp_bad(ms); | |
1861 | } | |
1862 | /* | |
1863 | * We'll let the main thread deal with closing the RP | |
1864 | * we could do a shutdown(2) on it, but we're the only user | |
1865 | * anyway, so there's nothing gained. | |
1866 | */ | |
1867 | goto out; | |
1868 | ||
1869 | case MIG_RP_MSG_PONG: | |
4d885131 | 1870 | tmp32 = ldl_be_p(buf); |
70b20477 DDAG |
1871 | trace_source_return_path_thread_pong(tmp32); |
1872 | break; | |
1873 | ||
1e2d90eb | 1874 | case MIG_RP_MSG_REQ_PAGES: |
4d885131 PM |
1875 | start = ldq_be_p(buf); |
1876 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1877 | migrate_handle_rp_req_pages(ms, NULL, start, len); |
1878 | break; | |
1879 | ||
1880 | case MIG_RP_MSG_REQ_PAGES_ID: | |
1881 | expected_len = 12 + 1; /* header + termination */ | |
1882 | ||
1883 | if (header_len >= expected_len) { | |
4d885131 PM |
1884 | start = ldq_be_p(buf); |
1885 | len = ldl_be_p(buf + 8); | |
1e2d90eb DDAG |
1886 | /* Now we expect an idstr */ |
1887 | tmp32 = buf[12]; /* Length of the following idstr */ | |
1888 | buf[13 + tmp32] = '\0'; | |
1889 | expected_len += tmp32; | |
1890 | } | |
1891 | if (header_len != expected_len) { | |
1892 | error_report("RP: Req_Page_id with length %d expecting %zd", | |
1893 | header_len, expected_len); | |
1894 | mark_source_rp_bad(ms); | |
1895 | goto out; | |
1896 | } | |
1897 | migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); | |
1898 | break; | |
1899 | ||
70b20477 DDAG |
1900 | default: |
1901 | break; | |
1902 | } | |
1903 | } | |
14b1742e PX |
1904 | |
1905 | out: | |
1906 | res = qemu_file_get_error(rp); | |
1907 | if (res) { | |
1908 | if (res == -EIO) { | |
1909 | /* | |
1910 | * Maybe there is something we can do: it looks like a | |
1911 | * network down issue, and we pause for a recovery. | |
1912 | */ | |
1913 | if (postcopy_pause_return_path_thread(ms)) { | |
1914 | /* Reload rp, reset the rest */ | |
1915 | rp = ms->rp_state.from_dst_file; | |
1916 | ms->rp_state.error = false; | |
1917 | goto retry; | |
1918 | } | |
1919 | } | |
1920 | ||
70b20477 DDAG |
1921 | trace_source_return_path_thread_bad_end(); |
1922 | mark_source_rp_bad(ms); | |
1923 | } | |
1924 | ||
1925 | trace_source_return_path_thread_end(); | |
70b20477 DDAG |
1926 | ms->rp_state.from_dst_file = NULL; |
1927 | qemu_fclose(rp); | |
1928 | return NULL; | |
1929 | } | |
1930 | ||
70b20477 DDAG |
1931 | static int open_return_path_on_source(MigrationState *ms) |
1932 | { | |
1933 | ||
89a02a9f | 1934 | ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); |
70b20477 DDAG |
1935 | if (!ms->rp_state.from_dst_file) { |
1936 | return -1; | |
1937 | } | |
1938 | ||
1939 | trace_open_return_path_on_source(); | |
1940 | qemu_thread_create(&ms->rp_state.rp_thread, "return path", | |
1941 | source_return_path_thread, ms, QEMU_THREAD_JOINABLE); | |
1942 | ||
1943 | trace_open_return_path_on_source_continue(); | |
1944 | ||
1945 | return 0; | |
1946 | } | |
1947 | ||
70b20477 DDAG |
1948 | /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ |
1949 | static int await_return_path_close_on_source(MigrationState *ms) | |
1950 | { | |
1951 | /* | |
1952 | * If this is a normal exit then the destination will send a SHUT and the | |
1953 | * rp_thread will exit, however if there's an error we need to cause | |
1954 | * it to exit. | |
1955 | */ | |
89a02a9f | 1956 | if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { |
70b20477 DDAG |
1957 | /* |
1958 | * shutdown(2), if we have it, will cause it to unblock if it's stuck | |
1959 | * waiting for the destination. | |
1960 | */ | |
1961 | qemu_file_shutdown(ms->rp_state.from_dst_file); | |
1962 | mark_source_rp_bad(ms); | |
1963 | } | |
1964 | trace_await_return_path_close_on_source_joining(); | |
1965 | qemu_thread_join(&ms->rp_state.rp_thread); | |
1966 | trace_await_return_path_close_on_source_close(); | |
1967 | return ms->rp_state.error; | |
1968 | } | |
1969 | ||
1d34e4bf DDAG |
1970 | /* |
1971 | * Switch from normal iteration to postcopy | |
1972 | * Returns non-0 on error | |
1973 | */ | |
7287cbd4 | 1974 | static int postcopy_start(MigrationState *ms) |
1d34e4bf DDAG |
1975 | { |
1976 | int ret; | |
61b67d47 DB |
1977 | QIOChannelBuffer *bioc; |
1978 | QEMUFile *fb; | |
1d34e4bf | 1979 | int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
ef8d6488 | 1980 | bool restart_block = false; |
0331c8ca DDAG |
1981 | int cur_state = MIGRATION_STATUS_ACTIVE; |
1982 | if (!migrate_pause_before_switchover()) { | |
1983 | migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, | |
1984 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
1985 | } | |
1d34e4bf DDAG |
1986 | |
1987 | trace_postcopy_start(); | |
1988 | qemu_mutex_lock_iothread(); | |
1989 | trace_postcopy_start_set_run(); | |
1990 | ||
1991 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); | |
1d34e4bf DDAG |
1992 | global_state_store(); |
1993 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); | |
76b1c7fe KW |
1994 | if (ret < 0) { |
1995 | goto fail; | |
1996 | } | |
1d34e4bf | 1997 | |
0331c8ca DDAG |
1998 | ret = migration_maybe_pause(ms, &cur_state, |
1999 | MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2000 | if (ret < 0) { | |
2001 | goto fail; | |
2002 | } | |
2003 | ||
76b1c7fe | 2004 | ret = bdrv_inactivate_all(); |
1d34e4bf DDAG |
2005 | if (ret < 0) { |
2006 | goto fail; | |
2007 | } | |
ef8d6488 | 2008 | restart_block = true; |
1d34e4bf | 2009 | |
1c0d249d DDAG |
2010 | /* |
2011 | * Cause any non-postcopiable, but iterative devices to | |
2012 | * send out their final data. | |
2013 | */ | |
a1fbe750 | 2014 | qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); |
1c0d249d | 2015 | |
1d34e4bf DDAG |
2016 | /* |
2017 | * in Finish migrate and with the io-lock held everything should | |
2018 | * be quiet, but we've potentially still got dirty pages and we | |
2019 | * need to tell the destination to throw any pages it's already received | |
2020 | * that are dirty | |
2021 | */ | |
58110f0a VSO |
2022 | if (migrate_postcopy_ram()) { |
2023 | if (ram_postcopy_send_discard_bitmap(ms)) { | |
2024 | error_report("postcopy send discard bitmap failed"); | |
2025 | goto fail; | |
2026 | } | |
1d34e4bf DDAG |
2027 | } |
2028 | ||
2029 | /* | |
2030 | * send rest of state - note things that are doing postcopy | |
2031 | * will notice we're in POSTCOPY_ACTIVE and not actually | |
2032 | * wrap their state up here | |
2033 | */ | |
89a02a9f | 2034 | qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); |
58110f0a VSO |
2035 | if (migrate_postcopy_ram()) { |
2036 | /* Ping just for debugging, helps line traces up */ | |
2037 | qemu_savevm_send_ping(ms->to_dst_file, 2); | |
2038 | } | |
1d34e4bf DDAG |
2039 | |
2040 | /* | |
2041 | * While loading the device state we may trigger page transfer | |
2042 | * requests and the fd must be free to process those, and thus | |
2043 | * the destination must read the whole device state off the fd before | |
2044 | * it starts processing it. Unfortunately the ad-hoc migration format | |
2045 | * doesn't allow the destination to know the size to read without fully | |
2046 | * parsing it through each devices load-state code (especially the open | |
2047 | * coded devices that use get/put). | |
2048 | * So we wrap the device state up in a package with a length at the start; | |
2049 | * to do this we use a qemu_buf to hold the whole of the device state. | |
2050 | */ | |
61b67d47 | 2051 | bioc = qio_channel_buffer_new(4096); |
6f01f136 | 2052 | qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); |
61b67d47 DB |
2053 | fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc)); |
2054 | object_unref(OBJECT(bioc)); | |
1d34e4bf | 2055 | |
c76201ab DDAG |
2056 | /* |
2057 | * Make sure the receiver can get incoming pages before we send the rest | |
2058 | * of the state | |
2059 | */ | |
2060 | qemu_savevm_send_postcopy_listen(fb); | |
2061 | ||
a1fbe750 | 2062 | qemu_savevm_state_complete_precopy(fb, false, false); |
58110f0a VSO |
2063 | if (migrate_postcopy_ram()) { |
2064 | qemu_savevm_send_ping(fb, 3); | |
2065 | } | |
1d34e4bf DDAG |
2066 | |
2067 | qemu_savevm_send_postcopy_run(fb); | |
2068 | ||
2069 | /* <><> end of stuff going into the package */ | |
1d34e4bf | 2070 | |
ef8d6488 DDAG |
2071 | /* Last point of recovery; as soon as we send the package the destination |
2072 | * can open devices and potentially start running. | |
2073 | * Lets just check again we've not got any errors. | |
2074 | */ | |
2075 | ret = qemu_file_get_error(ms->to_dst_file); | |
2076 | if (ret) { | |
2077 | error_report("postcopy_start: Migration stream errored (pre package)"); | |
2078 | goto fail_closefb; | |
2079 | } | |
2080 | ||
2081 | restart_block = false; | |
2082 | ||
1d34e4bf | 2083 | /* Now send that blob */ |
61b67d47 | 2084 | if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { |
1d34e4bf DDAG |
2085 | goto fail_closefb; |
2086 | } | |
2087 | qemu_fclose(fb); | |
b82fc321 DDAG |
2088 | |
2089 | /* Send a notify to give a chance for anything that needs to happen | |
2090 | * at the transition to postcopy and after the device state; in particular | |
2091 | * spice needs to trigger a transition now | |
2092 | */ | |
2093 | ms->postcopy_after_devices = true; | |
2094 | notifier_list_notify(&migration_state_notifiers, ms); | |
2095 | ||
1d34e4bf DDAG |
2096 | ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; |
2097 | ||
2098 | qemu_mutex_unlock_iothread(); | |
2099 | ||
58110f0a VSO |
2100 | if (migrate_postcopy_ram()) { |
2101 | /* | |
2102 | * Although this ping is just for debug, it could potentially be | |
2103 | * used for getting a better measurement of downtime at the source. | |
2104 | */ | |
2105 | qemu_savevm_send_ping(ms->to_dst_file, 4); | |
2106 | } | |
1d34e4bf | 2107 | |
ced1c616 PB |
2108 | if (migrate_release_ram()) { |
2109 | ram_postcopy_migrated_memory_release(ms); | |
2110 | } | |
2111 | ||
89a02a9f | 2112 | ret = qemu_file_get_error(ms->to_dst_file); |
1d34e4bf DDAG |
2113 | if (ret) { |
2114 | error_report("postcopy_start: Migration stream errored"); | |
48781e5b | 2115 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf DDAG |
2116 | MIGRATION_STATUS_FAILED); |
2117 | } | |
2118 | ||
2119 | return ret; | |
2120 | ||
2121 | fail_closefb: | |
2122 | qemu_fclose(fb); | |
2123 | fail: | |
48781e5b | 2124 | migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, |
1d34e4bf | 2125 | MIGRATION_STATUS_FAILED); |
ef8d6488 DDAG |
2126 | if (restart_block) { |
2127 | /* A failure happened early enough that we know the destination hasn't | |
2128 | * accessed block devices, so we're safe to recover. | |
2129 | */ | |
2130 | Error *local_err = NULL; | |
2131 | ||
2132 | bdrv_invalidate_cache_all(&local_err); | |
2133 | if (local_err) { | |
2134 | error_report_err(local_err); | |
2135 | } | |
2136 | } | |
1d34e4bf DDAG |
2137 | qemu_mutex_unlock_iothread(); |
2138 | return -1; | |
2139 | } | |
2140 | ||
e91d8951 DDAG |
2141 | /** |
2142 | * migration_maybe_pause: Pause if required to by | |
2143 | * migrate_pause_before_switchover called with the iothread locked | |
2144 | * Returns: 0 on success | |
2145 | */ | |
0331c8ca DDAG |
2146 | static int migration_maybe_pause(MigrationState *s, |
2147 | int *current_active_state, | |
2148 | int new_state) | |
e91d8951 DDAG |
2149 | { |
2150 | if (!migrate_pause_before_switchover()) { | |
2151 | return 0; | |
2152 | } | |
2153 | ||
2154 | /* Since leaving this state is not atomic with posting the semaphore | |
2155 | * it's possible that someone could have issued multiple migrate_continue | |
2156 | * and the semaphore is incorrectly positive at this point; | |
2157 | * the docs say it's undefined to reinit a semaphore that's already | |
2158 | * init'd, so use timedwait to eat up any existing posts. | |
2159 | */ | |
2160 | while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) { | |
2161 | /* This block intentionally left blank */ | |
2162 | } | |
2163 | ||
2164 | qemu_mutex_unlock_iothread(); | |
2165 | migrate_set_state(&s->state, *current_active_state, | |
2166 | MIGRATION_STATUS_PRE_SWITCHOVER); | |
2167 | qemu_sem_wait(&s->pause_sem); | |
2168 | migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER, | |
0331c8ca DDAG |
2169 | new_state); |
2170 | *current_active_state = new_state; | |
e91d8951 DDAG |
2171 | qemu_mutex_lock_iothread(); |
2172 | ||
0331c8ca | 2173 | return s->state == new_state ? 0 : -EINVAL; |
e91d8951 DDAG |
2174 | } |
2175 | ||
09f6c85e DDAG |
2176 | /** |
2177 | * migration_completion: Used by migration_thread when there's not much left. | |
2178 | * The caller 'breaks' the loop when this returns. | |
2179 | * | |
2180 | * @s: Current migration state | |
09f6c85e | 2181 | */ |
2ad87305 | 2182 | static void migration_completion(MigrationState *s) |
09f6c85e DDAG |
2183 | { |
2184 | int ret; | |
2ad87305 | 2185 | int current_active_state = s->state; |
09f6c85e | 2186 | |
b10ac0c4 DDAG |
2187 | if (s->state == MIGRATION_STATUS_ACTIVE) { |
2188 | qemu_mutex_lock_iothread(); | |
64909f97 | 2189 | s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
b10ac0c4 | 2190 | qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); |
7287cbd4 | 2191 | s->vm_was_running = runstate_is_running(); |
b10ac0c4 DDAG |
2192 | ret = global_state_store(); |
2193 | ||
2194 | if (!ret) { | |
a1fbe750 | 2195 | bool inactivate = !migrate_colo_enabled(); |
b10ac0c4 | 2196 | ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); |
e91d8951 | 2197 | if (ret >= 0) { |
0331c8ca DDAG |
2198 | ret = migration_maybe_pause(s, ¤t_active_state, |
2199 | MIGRATION_STATUS_DEVICE); | |
e91d8951 | 2200 | } |
f07fa4cb KW |
2201 | if (ret >= 0) { |
2202 | qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); | |
a1fbe750 FZ |
2203 | ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, |
2204 | inactivate); | |
f07fa4cb | 2205 | } |
a1fbe750 FZ |
2206 | if (inactivate && ret >= 0) { |
2207 | s->block_inactive = true; | |
b10ac0c4 DDAG |
2208 | } |
2209 | } | |
2210 | qemu_mutex_unlock_iothread(); | |
09f6c85e | 2211 | |
b10ac0c4 DDAG |
2212 | if (ret < 0) { |
2213 | goto fail; | |
09f6c85e | 2214 | } |
b10ac0c4 DDAG |
2215 | } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { |
2216 | trace_migration_completion_postcopy_end(); | |
2217 | ||
89a02a9f | 2218 | qemu_savevm_state_complete_postcopy(s->to_dst_file); |
b10ac0c4 | 2219 | trace_migration_completion_postcopy_end_after_complete(); |
09f6c85e | 2220 | } |
09f6c85e | 2221 | |
b10ac0c4 DDAG |
2222 | /* |
2223 | * If rp was opened we must clean up the thread before | |
2224 | * cleaning everything else up (since if there are no failures | |
2225 | * it will wait for the destination to send it's status in | |
2226 | * a SHUT command). | |
b10ac0c4 | 2227 | */ |
0425dc97 | 2228 | if (s->rp_state.from_dst_file) { |
b10ac0c4 | 2229 | int rp_error; |
0425dc97 | 2230 | trace_migration_return_path_end_before(); |
b10ac0c4 | 2231 | rp_error = await_return_path_close_on_source(s); |
0425dc97 | 2232 | trace_migration_return_path_end_after(rp_error); |
b10ac0c4 | 2233 | if (rp_error) { |
fe904ea8 | 2234 | goto fail_invalidate; |
b10ac0c4 | 2235 | } |
09f6c85e DDAG |
2236 | } |
2237 | ||
89a02a9f | 2238 | if (qemu_file_get_error(s->to_dst_file)) { |
09f6c85e | 2239 | trace_migration_completion_file_err(); |
fe904ea8 | 2240 | goto fail_invalidate; |
09f6c85e DDAG |
2241 | } |
2242 | ||
0b827d5e HZ |
2243 | if (!migrate_colo_enabled()) { |
2244 | migrate_set_state(&s->state, current_active_state, | |
2245 | MIGRATION_STATUS_COMPLETED); | |
2246 | } | |
2247 | ||
09f6c85e DDAG |
2248 | return; |
2249 | ||
fe904ea8 GK |
2250 | fail_invalidate: |
2251 | /* If not doing postcopy, vm_start() will be called: let's regain | |
2252 | * control on images. | |
2253 | */ | |
6039dd5b DDAG |
2254 | if (s->state == MIGRATION_STATUS_ACTIVE || |
2255 | s->state == MIGRATION_STATUS_DEVICE) { | |
fe904ea8 GK |
2256 | Error *local_err = NULL; |
2257 | ||
1d2acc31 | 2258 | qemu_mutex_lock_iothread(); |
fe904ea8 GK |
2259 | bdrv_invalidate_cache_all(&local_err); |
2260 | if (local_err) { | |
2261 | error_report_err(local_err); | |
1d2acc31 HZ |
2262 | } else { |
2263 | s->block_inactive = false; | |
fe904ea8 | 2264 | } |
1d2acc31 | 2265 | qemu_mutex_unlock_iothread(); |
fe904ea8 GK |
2266 | } |
2267 | ||
09f6c85e | 2268 | fail: |
48781e5b HZ |
2269 | migrate_set_state(&s->state, current_active_state, |
2270 | MIGRATION_STATUS_FAILED); | |
09f6c85e DDAG |
2271 | } |
2272 | ||
35a6ed4f HZ |
2273 | bool migrate_colo_enabled(void) |
2274 | { | |
2275 | MigrationState *s = migrate_get_current(); | |
2276 | return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; | |
2277 | } | |
2278 | ||
b23c2ade PX |
2279 | typedef enum MigThrError { |
2280 | /* No error detected */ | |
2281 | MIG_THR_ERR_NONE = 0, | |
2282 | /* Detected error, but resumed successfully */ | |
2283 | MIG_THR_ERR_RECOVERED = 1, | |
2284 | /* Detected fatal error, need to exit */ | |
2285 | MIG_THR_ERR_FATAL = 2, | |
2286 | } MigThrError; | |
2287 | ||
2288 | /* | |
2289 | * We don't return until we are in a safe state to continue current | |
2290 | * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or | |
2291 | * MIG_THR_ERR_FATAL if unrecovery failure happened. | |
2292 | */ | |
2293 | static MigThrError postcopy_pause(MigrationState *s) | |
2294 | { | |
2295 | assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2296 | migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, | |
2297 | MIGRATION_STATUS_POSTCOPY_PAUSED); | |
2298 | ||
2299 | /* Current channel is possibly broken. Release it. */ | |
2300 | assert(s->to_dst_file); | |
2301 | qemu_file_shutdown(s->to_dst_file); | |
2302 | qemu_fclose(s->to_dst_file); | |
2303 | s->to_dst_file = NULL; | |
2304 | ||
2305 | error_report("Detected IO failure for postcopy. " | |
2306 | "Migration paused."); | |
2307 | ||
2308 | /* | |
2309 | * We wait until things fixed up. Then someone will setup the | |
2310 | * status back for us. | |
2311 | */ | |
2312 | while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { | |
2313 | qemu_sem_wait(&s->postcopy_pause_sem); | |
2314 | } | |
2315 | ||
2316 | trace_postcopy_pause_continued(); | |
2317 | ||
2318 | return MIG_THR_ERR_RECOVERED; | |
2319 | } | |
2320 | ||
2321 | static MigThrError migration_detect_error(MigrationState *s) | |
2322 | { | |
2323 | int ret; | |
2324 | ||
2325 | /* Try to detect any file errors */ | |
2326 | ret = qemu_file_get_error(s->to_dst_file); | |
2327 | ||
2328 | if (!ret) { | |
2329 | /* Everything is fine */ | |
2330 | return MIG_THR_ERR_NONE; | |
2331 | } | |
2332 | ||
2333 | if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) { | |
2334 | /* | |
2335 | * For postcopy, we allow the network to be down for a | |
2336 | * while. After that, it can be continued by a | |
2337 | * recovery phase. | |
2338 | */ | |
2339 | return postcopy_pause(s); | |
2340 | } else { | |
2341 | /* | |
2342 | * For precopy (or postcopy with error outside IO), we fail | |
2343 | * with no time. | |
2344 | */ | |
2345 | migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED); | |
2346 | trace_migration_thread_file_err(); | |
2347 | ||
2348 | /* Time to stop the migration, now. */ | |
2349 | return MIG_THR_ERR_FATAL; | |
2350 | } | |
2351 | } | |
2352 | ||
cf011f08 PX |
2353 | static void migration_calculate_complete(MigrationState *s) |
2354 | { | |
2355 | uint64_t bytes = qemu_ftell(s->to_dst_file); | |
2356 | int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
2357 | ||
2358 | s->total_time = end_time - s->start_time; | |
2359 | if (!s->downtime) { | |
2360 | /* | |
2361 | * It's still not set, so we are precopy migration. For | |
2362 | * postcopy, downtime is calculated during postcopy_start(). | |
2363 | */ | |
2364 | s->downtime = end_time - s->downtime_start; | |
2365 | } | |
2366 | ||
2367 | if (s->total_time) { | |
2368 | s->mbps = ((double) bytes * 8.0) / s->total_time / 1000; | |
2369 | } | |
2370 | } | |
2371 | ||
b15df1ae PX |
2372 | static void migration_update_counters(MigrationState *s, |
2373 | int64_t current_time) | |
2374 | { | |
2375 | uint64_t transferred, time_spent; | |
b15df1ae PX |
2376 | double bandwidth; |
2377 | ||
2378 | if (current_time < s->iteration_start_time + BUFFER_DELAY) { | |
2379 | return; | |
2380 | } | |
2381 | ||
2382 | transferred = qemu_ftell(s->to_dst_file) - s->iteration_initial_bytes; | |
2383 | time_spent = current_time - s->iteration_start_time; | |
2384 | bandwidth = (double)transferred / time_spent; | |
0781c1ed | 2385 | s->threshold_size = bandwidth * s->parameters.downtime_limit; |
b15df1ae PX |
2386 | |
2387 | s->mbps = (((double) transferred * 8.0) / | |
2388 | ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; | |
2389 | ||
2390 | /* | |
2391 | * if we haven't sent anything, we don't want to | |
2392 | * recalculate. 10000 is a small enough number for our purposes | |
2393 | */ | |
2394 | if (ram_counters.dirty_pages_rate && transferred > 10000) { | |
2395 | s->expected_downtime = ram_counters.dirty_pages_rate * | |
2396 | qemu_target_page_size() / bandwidth; | |
2397 | } | |
2398 | ||
2399 | qemu_file_reset_rate_limit(s->to_dst_file); | |
2400 | ||
2401 | s->iteration_start_time = current_time; | |
2402 | s->iteration_initial_bytes = qemu_ftell(s->to_dst_file); | |
2403 | ||
2404 | trace_migrate_transferred(transferred, time_spent, | |
0781c1ed | 2405 | bandwidth, s->threshold_size); |
b15df1ae PX |
2406 | } |
2407 | ||
2ad87305 PX |
2408 | /* Migration thread iteration status */ |
2409 | typedef enum { | |
2410 | MIG_ITERATE_RESUME, /* Resume current iteration */ | |
2411 | MIG_ITERATE_SKIP, /* Skip current iteration */ | |
2412 | MIG_ITERATE_BREAK, /* Break the loop */ | |
2413 | } MigIterateState; | |
2414 | ||
2415 | /* | |
2416 | * Return true if continue to the next iteration directly, false | |
2417 | * otherwise. | |
2418 | */ | |
2419 | static MigIterateState migration_iteration_run(MigrationState *s) | |
2420 | { | |
47995026 | 2421 | uint64_t pending_size, pend_pre, pend_compat, pend_post; |
2ad87305 PX |
2422 | bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE; |
2423 | ||
47995026 VSO |
2424 | qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre, |
2425 | &pend_compat, &pend_post); | |
2426 | pending_size = pend_pre + pend_compat + pend_post; | |
2ad87305 PX |
2427 | |
2428 | trace_migrate_pending(pending_size, s->threshold_size, | |
47995026 | 2429 | pend_pre, pend_compat, pend_post); |
2ad87305 PX |
2430 | |
2431 | if (pending_size && pending_size >= s->threshold_size) { | |
2432 | /* Still a significant amount to transfer */ | |
2433 | if (migrate_postcopy() && !in_postcopy && | |
47995026 | 2434 | pend_pre <= s->threshold_size && |
2ad87305 PX |
2435 | atomic_read(&s->start_postcopy)) { |
2436 | if (postcopy_start(s)) { | |
2437 | error_report("%s: postcopy failed to start", __func__); | |
2438 | } | |
2439 | return MIG_ITERATE_SKIP; | |
2440 | } | |
2441 | /* Just another iteration step */ | |
2442 | qemu_savevm_state_iterate(s->to_dst_file, | |
2443 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); | |
2444 | } else { | |
2445 | trace_migration_thread_low_pending(pending_size); | |
2446 | migration_completion(s); | |
2447 | return MIG_ITERATE_BREAK; | |
2448 | } | |
2449 | ||
2450 | return MIG_ITERATE_RESUME; | |
2451 | } | |
2452 | ||
199aa6d4 PX |
2453 | static void migration_iteration_finish(MigrationState *s) |
2454 | { | |
2455 | /* If we enabled cpu throttling for auto-converge, turn it off. */ | |
2456 | cpu_throttle_stop(); | |
2457 | ||
2458 | qemu_mutex_lock_iothread(); | |
2459 | switch (s->state) { | |
2460 | case MIGRATION_STATUS_COMPLETED: | |
2461 | migration_calculate_complete(s); | |
2462 | runstate_set(RUN_STATE_POSTMIGRATE); | |
2463 | break; | |
2464 | ||
2465 | case MIGRATION_STATUS_ACTIVE: | |
2466 | /* | |
2467 | * We should really assert here, but since it's during | |
2468 | * migration, let's try to reduce the usage of assertions. | |
2469 | */ | |
2470 | if (!migrate_colo_enabled()) { | |
2471 | error_report("%s: critical error: calling COLO code without " | |
2472 | "COLO enabled", __func__); | |
2473 | } | |
2474 | migrate_start_colo_process(s); | |
2475 | /* | |
2476 | * Fixme: we will run VM in COLO no matter its old running state. | |
2477 | * After exited COLO, we will keep running. | |
2478 | */ | |
2479 | s->vm_was_running = true; | |
2480 | /* Fallthrough */ | |
2481 | case MIGRATION_STATUS_FAILED: | |
2482 | case MIGRATION_STATUS_CANCELLED: | |
2483 | if (s->vm_was_running) { | |
2484 | vm_start(); | |
2485 | } else { | |
2486 | if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { | |
2487 | runstate_set(RUN_STATE_POSTMIGRATE); | |
2488 | } | |
2489 | } | |
2490 | break; | |
2491 | ||
2492 | default: | |
2493 | /* Should not reach here, but if so, forgive the VM. */ | |
2494 | error_report("%s: Unknown ending state %d", __func__, s->state); | |
2495 | break; | |
2496 | } | |
2497 | qemu_bh_schedule(s->cleanup_bh); | |
2498 | qemu_mutex_unlock_iothread(); | |
2499 | } | |
2500 | ||
70b20477 DDAG |
2501 | /* |
2502 | * Master migration thread on the source VM. | |
2503 | * It drives the migration and pumps the data down the outgoing channel. | |
2504 | */ | |
5f496a1b | 2505 | static void *migration_thread(void *opaque) |
0d82d0e8 | 2506 | { |
9848a404 | 2507 | MigrationState *s = opaque; |
bc72ad67 | 2508 | int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); |
b23c2ade | 2509 | MigThrError thr_error; |
76f5933a | 2510 | |
ab28bd23 PB |
2511 | rcu_register_thread(); |
2512 | ||
b15df1ae PX |
2513 | s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
2514 | ||
89a02a9f | 2515 | qemu_savevm_state_header(s->to_dst_file); |
1d34e4bf | 2516 | |
62a02658 PX |
2517 | /* |
2518 | * If we opened the return path, we need to make sure dst has it | |
2519 | * opened as well. | |
2520 | */ | |
2521 | if (s->rp_state.from_dst_file) { | |
1d34e4bf | 2522 | /* Now tell the dest that it should open its end so it can reply */ |
89a02a9f | 2523 | qemu_savevm_send_open_return_path(s->to_dst_file); |
1d34e4bf DDAG |
2524 | |
2525 | /* And do a ping that will make stuff easier to debug */ | |
89a02a9f | 2526 | qemu_savevm_send_ping(s->to_dst_file, 1); |
0425dc97 | 2527 | } |
1d34e4bf | 2528 | |
58110f0a | 2529 | if (migrate_postcopy()) { |
1d34e4bf DDAG |
2530 | /* |
2531 | * Tell the destination that we *might* want to do postcopy later; | |
2532 | * if the other end can't do postcopy it should fail now, nice and | |
2533 | * early. | |
2534 | */ | |
89a02a9f | 2535 | qemu_savevm_send_postcopy_advise(s->to_dst_file); |
1d34e4bf DDAG |
2536 | } |
2537 | ||
9907e842 | 2538 | qemu_savevm_state_setup(s->to_dst_file); |
0d82d0e8 | 2539 | |
bc72ad67 | 2540 | s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; |
48781e5b HZ |
2541 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
2542 | MIGRATION_STATUS_ACTIVE); | |
29ae8a41 | 2543 | |
9ec055ae DDAG |
2544 | trace_migration_thread_setup_complete(); |
2545 | ||
2546 | while (s->state == MIGRATION_STATUS_ACTIVE || | |
2547 | s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { | |
a3e879cd | 2548 | int64_t current_time; |
0d82d0e8 | 2549 | |
89a02a9f | 2550 | if (!qemu_file_rate_limit(s->to_dst_file)) { |
2ad87305 PX |
2551 | MigIterateState iter_state = migration_iteration_run(s); |
2552 | if (iter_state == MIG_ITERATE_SKIP) { | |
2553 | continue; | |
2554 | } else if (iter_state == MIG_ITERATE_BREAK) { | |
09f6c85e | 2555 | break; |
c369f40d JQ |
2556 | } |
2557 | } | |
f4410a5d | 2558 | |
b23c2ade PX |
2559 | /* |
2560 | * Try to detect any kind of failures, and see whether we | |
2561 | * should stop the migration now. | |
2562 | */ | |
2563 | thr_error = migration_detect_error(s); | |
2564 | if (thr_error == MIG_THR_ERR_FATAL) { | |
2565 | /* Stop migration */ | |
fd45ee2c | 2566 | break; |
b23c2ade PX |
2567 | } else if (thr_error == MIG_THR_ERR_RECOVERED) { |
2568 | /* | |
2569 | * Just recovered from a e.g. network failure, reset all | |
2570 | * the local variables. This is important to avoid | |
2571 | * breaking transferred_bytes and bandwidth calculation | |
2572 | */ | |
2573 | s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); | |
2574 | s->iteration_initial_bytes = 0; | |
fd45ee2c | 2575 | } |
b15df1ae | 2576 | |
bc72ad67 | 2577 | current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); |
0d82d0e8 | 2578 | |
b15df1ae PX |
2579 | migration_update_counters(s, current_time); |
2580 | ||
89a02a9f | 2581 | if (qemu_file_rate_limit(s->to_dst_file)) { |
0d82d0e8 | 2582 | /* usleep expects microseconds */ |
b15df1ae PX |
2583 | g_usleep((s->iteration_start_time + BUFFER_DELAY - |
2584 | current_time) * 1000); | |
0d82d0e8 | 2585 | } |
a3fa1d78 PB |
2586 | } |
2587 | ||
1d34e4bf | 2588 | trace_migration_thread_after_loop(); |
199aa6d4 | 2589 | migration_iteration_finish(s); |
ab28bd23 | 2590 | rcu_unregister_thread(); |
0d82d0e8 JQ |
2591 | return NULL; |
2592 | } | |
2593 | ||
cce8040b | 2594 | void migrate_fd_connect(MigrationState *s, Error *error_in) |
0d82d0e8 | 2595 | { |
2ff30257 | 2596 | s->expected_downtime = s->parameters.downtime_limit; |
bb1fadc4 | 2597 | s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s); |
cce8040b DDAG |
2598 | if (error_in) { |
2599 | migrate_fd_error(s, error_in); | |
2600 | migrate_fd_cleanup(s); | |
2601 | return; | |
2602 | } | |
0d82d0e8 | 2603 | |
9e4d2b98 | 2604 | qemu_file_set_blocking(s->to_dst_file, true); |
89a02a9f | 2605 | qemu_file_set_rate_limit(s->to_dst_file, |
2ff30257 | 2606 | s->parameters.max_bandwidth / XFER_LIMIT_RATIO); |
442773ce | 2607 | |
9287ac27 SH |
2608 | /* Notify before starting migration thread */ |
2609 | notifier_list_notify(&migration_state_notifiers, s); | |
2610 | ||
1d34e4bf | 2611 | /* |
c788ada8 PX |
2612 | * Open the return path. For postcopy, it is used exclusively. For |
2613 | * precopy, only if user specified "return-path" capability would | |
2614 | * QEMU uses the return path. | |
1d34e4bf | 2615 | */ |
c788ada8 | 2616 | if (migrate_postcopy_ram() || migrate_use_return_path()) { |
1d34e4bf DDAG |
2617 | if (open_return_path_on_source(s)) { |
2618 | error_report("Unable to open return-path for postcopy"); | |
48781e5b | 2619 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, |
1d34e4bf DDAG |
2620 | MIGRATION_STATUS_FAILED); |
2621 | migrate_fd_cleanup(s); | |
2622 | return; | |
2623 | } | |
2624 | } | |
2625 | ||
f986c3d2 JQ |
2626 | if (multifd_save_setup() != 0) { |
2627 | migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, | |
2628 | MIGRATION_STATUS_FAILED); | |
2629 | migrate_fd_cleanup(s); | |
2630 | return; | |
2631 | } | |
009fad7f | 2632 | qemu_thread_create(&s->thread, "live_migration", migration_thread, s, |
bb1fadc4 | 2633 | QEMU_THREAD_JOINABLE); |
1d34e4bf | 2634 | s->migration_thread_running = true; |
0d82d0e8 | 2635 | } |
093e3c42 | 2636 | |
9d18af93 PX |
2637 | void migration_global_dump(Monitor *mon) |
2638 | { | |
2639 | MigrationState *ms = migrate_get_current(); | |
2640 | ||
6f0f6428 JQ |
2641 | monitor_printf(mon, "globals:\n"); |
2642 | monitor_printf(mon, "store-global-state: %s\n", | |
2643 | ms->store_global_state ? "on" : "off"); | |
2644 | monitor_printf(mon, "only-migratable: %s\n", | |
2645 | ms->only_migratable ? "on" : "off"); | |
2646 | monitor_printf(mon, "send-configuration: %s\n", | |
2647 | ms->send_configuration ? "on" : "off"); | |
2648 | monitor_printf(mon, "send-section-footer: %s\n", | |
2649 | ms->send_section_footer ? "on" : "off"); | |
9d18af93 PX |
2650 | } |
2651 | ||
20814758 PX |
2652 | #define DEFINE_PROP_MIG_CAP(name, x) \ |
2653 | DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false) | |
2654 | ||
5272298c PX |
2655 | static Property migration_properties[] = { |
2656 | DEFINE_PROP_BOOL("store-global-state", MigrationState, | |
2657 | store_global_state, true), | |
3df663e5 | 2658 | DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false), |
71dd4c1a PX |
2659 | DEFINE_PROP_BOOL("send-configuration", MigrationState, |
2660 | send_configuration, true), | |
15c38503 PX |
2661 | DEFINE_PROP_BOOL("send-section-footer", MigrationState, |
2662 | send_section_footer, true), | |
89632faf PX |
2663 | |
2664 | /* Migration parameters */ | |
741d4086 | 2665 | DEFINE_PROP_UINT8("x-compress-level", MigrationState, |
89632faf PX |
2666 | parameters.compress_level, |
2667 | DEFAULT_MIGRATE_COMPRESS_LEVEL), | |
741d4086 | 2668 | DEFINE_PROP_UINT8("x-compress-threads", MigrationState, |
89632faf PX |
2669 | parameters.compress_threads, |
2670 | DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT), | |
741d4086 | 2671 | DEFINE_PROP_UINT8("x-decompress-threads", MigrationState, |
89632faf PX |
2672 | parameters.decompress_threads, |
2673 | DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT), | |
741d4086 | 2674 | DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState, |
89632faf PX |
2675 | parameters.cpu_throttle_initial, |
2676 | DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL), | |
741d4086 | 2677 | DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState, |
89632faf PX |
2678 | parameters.cpu_throttle_increment, |
2679 | DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT), | |
741d4086 | 2680 | DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState, |
89632faf | 2681 | parameters.max_bandwidth, MAX_THROTTLE), |
741d4086 | 2682 | DEFINE_PROP_UINT64("x-downtime-limit", MigrationState, |
89632faf PX |
2683 | parameters.downtime_limit, |
2684 | DEFAULT_MIGRATE_SET_DOWNTIME), | |
741d4086 | 2685 | DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState, |
89632faf PX |
2686 | parameters.x_checkpoint_delay, |
2687 | DEFAULT_MIGRATE_X_CHECKPOINT_DELAY), | |
741d4086 | 2688 | DEFINE_PROP_UINT8("x-multifd-channels", MigrationState, |
4075fb1c JQ |
2689 | parameters.x_multifd_channels, |
2690 | DEFAULT_MIGRATE_MULTIFD_CHANNELS), | |
741d4086 | 2691 | DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState, |
0fb86605 JQ |
2692 | parameters.x_multifd_page_count, |
2693 | DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT), | |
73af8dd8 JQ |
2694 | DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState, |
2695 | parameters.xbzrle_cache_size, | |
2696 | DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE), | |
20814758 PX |
2697 | |
2698 | /* Migration capabilities */ | |
2699 | DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE), | |
2700 | DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL), | |
2701 | DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE), | |
2702 | DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS), | |
2703 | DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS), | |
2704 | DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS), | |
2705 | DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM), | |
2706 | DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO), | |
2707 | DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM), | |
2708 | DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK), | |
2709 | DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH), | |
30126bbf | 2710 | DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD), |
20814758 | 2711 | |
5272298c PX |
2712 | DEFINE_PROP_END_OF_LIST(), |
2713 | }; | |
2714 | ||
e5cb7e76 PX |
2715 | static void migration_class_init(ObjectClass *klass, void *data) |
2716 | { | |
2717 | DeviceClass *dc = DEVICE_CLASS(klass); | |
2718 | ||
2719 | dc->user_creatable = false; | |
5272298c | 2720 | dc->props = migration_properties; |
e5cb7e76 PX |
2721 | } |
2722 | ||
b91bf5e4 MAL |
2723 | static void migration_instance_finalize(Object *obj) |
2724 | { | |
2725 | MigrationState *ms = MIGRATION_OBJ(obj); | |
2726 | MigrationParameters *params = &ms->parameters; | |
2727 | ||
87db1a7d | 2728 | qemu_mutex_destroy(&ms->error_mutex); |
b91bf5e4 MAL |
2729 | g_free(params->tls_hostname); |
2730 | g_free(params->tls_creds); | |
e91d8951 | 2731 | qemu_sem_destroy(&ms->pause_sem); |
b23c2ade | 2732 | qemu_sem_destroy(&ms->postcopy_pause_sem); |
14b1742e | 2733 | qemu_sem_destroy(&ms->postcopy_pause_rp_sem); |
ab105cc1 | 2734 | error_free(ms->error); |
b91bf5e4 MAL |
2735 | } |
2736 | ||
e5cb7e76 PX |
2737 | static void migration_instance_init(Object *obj) |
2738 | { | |
2739 | MigrationState *ms = MIGRATION_OBJ(obj); | |
8b0b29dc | 2740 | MigrationParameters *params = &ms->parameters; |
e5cb7e76 PX |
2741 | |
2742 | ms->state = MIGRATION_STATUS_NONE; | |
e5cb7e76 | 2743 | ms->mbps = -1; |
e91d8951 | 2744 | qemu_sem_init(&ms->pause_sem, 0); |
87db1a7d | 2745 | qemu_mutex_init(&ms->error_mutex); |
8b0b29dc PX |
2746 | |
2747 | params->tls_hostname = g_strdup(""); | |
2748 | params->tls_creds = g_strdup(""); | |
2749 | ||
2750 | /* Set has_* up only for parameter checks */ | |
2751 | params->has_compress_level = true; | |
2752 | params->has_compress_threads = true; | |
2753 | params->has_decompress_threads = true; | |
2754 | params->has_cpu_throttle_initial = true; | |
2755 | params->has_cpu_throttle_increment = true; | |
2756 | params->has_max_bandwidth = true; | |
2757 | params->has_downtime_limit = true; | |
2758 | params->has_x_checkpoint_delay = true; | |
2759 | params->has_block_incremental = true; | |
4075fb1c | 2760 | params->has_x_multifd_channels = true; |
0fb86605 | 2761 | params->has_x_multifd_page_count = true; |
73af8dd8 | 2762 | params->has_xbzrle_cache_size = true; |
b23c2ade PX |
2763 | |
2764 | qemu_sem_init(&ms->postcopy_pause_sem, 0); | |
14b1742e | 2765 | qemu_sem_init(&ms->postcopy_pause_rp_sem, 0); |
8b0b29dc PX |
2766 | } |
2767 | ||
2768 | /* | |
2769 | * Return true if check pass, false otherwise. Error will be put | |
2770 | * inside errp if provided. | |
2771 | */ | |
2772 | static bool migration_object_check(MigrationState *ms, Error **errp) | |
2773 | { | |
6b19a7d9 PX |
2774 | MigrationCapabilityStatusList *head = NULL; |
2775 | /* Assuming all off */ | |
2776 | bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret; | |
2777 | int i; | |
2778 | ||
8b0b29dc PX |
2779 | if (!migrate_params_check(&ms->parameters, errp)) { |
2780 | return false; | |
2781 | } | |
2782 | ||
6b19a7d9 PX |
2783 | for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { |
2784 | if (ms->enabled_capabilities[i]) { | |
2785 | head = migrate_cap_add(head, i, true); | |
2786 | } | |
2787 | } | |
2788 | ||
2789 | ret = migrate_caps_check(cap_list, head, errp); | |
2790 | ||
2791 | /* It works with head == NULL */ | |
2792 | qapi_free_MigrationCapabilityStatusList(head); | |
2793 | ||
2794 | return ret; | |
e5cb7e76 PX |
2795 | } |
2796 | ||
2797 | static const TypeInfo migration_type = { | |
2798 | .name = TYPE_MIGRATION, | |
01f6e14c | 2799 | /* |
c8d3ff38 PX |
2800 | * NOTE: TYPE_MIGRATION is not really a device, as the object is |
2801 | * not created using qdev_create(), it is not attached to the qdev | |
2802 | * device tree, and it is never realized. | |
2803 | * | |
2804 | * TODO: Make this TYPE_OBJECT once QOM provides something like | |
2805 | * TYPE_DEVICE's "-global" properties. | |
01f6e14c | 2806 | */ |
e5cb7e76 PX |
2807 | .parent = TYPE_DEVICE, |
2808 | .class_init = migration_class_init, | |
2809 | .class_size = sizeof(MigrationClass), | |
2810 | .instance_size = sizeof(MigrationState), | |
2811 | .instance_init = migration_instance_init, | |
b91bf5e4 | 2812 | .instance_finalize = migration_instance_finalize, |
e5cb7e76 PX |
2813 | }; |
2814 | ||
2815 | static void register_migration_types(void) | |
2816 | { | |
2817 | type_register_static(&migration_type); | |
2818 | } | |
2819 | ||
2820 | type_init(register_migration_types); |