]> git.proxmox.com Git - mirror_qemu.git/blame - migration/migration.c
projects / mirror_qemu.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
blockjob: Make block_job_pause_all() keep a reference to the jobs
[mirror_qemu.git] / migration / migration.c
CommitLineData
5bb7910a
AL		 1/*
2 * QEMU live migration
3 *
4 * Copyright IBM, Corp. 2008
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
6b620ca3
PB		 12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
5bb7910a
AL		 14 */
15
1393a485 16#include "qemu/osdep.h"
f348b6d1 17#include "qemu/cutils.h"
d49b6836 18#include "qemu/error-report.h"
795c40b8 19#include "migration/blocker.h"
f4dbe1bf 20#include "exec.h"
7fcac4a2 21#include "fd.h"
61e8b148 22#include "socket.h"
e1a3ecee 23#include "rdma.h"
7b1e1a22 24#include "ram.h"
84a899de 25#include "migration/global_state.h"
c4b63b7c 26#include "migration/misc.h"
6666c96a 27#include "migration.h"
20a519a0 28#include "savevm.h"
40014d81 29#include "qemu-file-channel.h"
08a0aee1 30#include "qemu-file.h"
987772d9 31#include "migration/vmstate.h"
737e150e 32#include "block/block.h"
cc7a8ea7 33#include "qapi/qmp/qerror.h"
ab28bd23 34#include "qemu/rcu.h"
2c9e6fec 35#include "block.h"
be07b0ac 36#include "postcopy-ram.h"
766bd176 37#include "qemu/thread.h"
791e7c82 38#include "qmp-commands.h"
c09e5bb1 39#include "trace.h"
598cd2bd 40#include "qapi-event.h"
51180423 41#include "exec/target_page.h"
61b67d47 42#include "io/channel-buffer.h"
35a6ed4f 43#include "migration/colo.h"
4ffdb337 44#include "hw/boards.h"
9d18af93 45#include "monitor/monitor.h"
065e2813 46
dc325627 47#define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
5bb7910a 48
5b4e1eb7
JQ		 49/* Amount of time to allocate to each "chunk" of bandwidth-throttled
50 * data. */
51#define BUFFER_DELAY 100
52#define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
53
2ff30257
AA		 54/* Time in milliseconds we are allowed to stop the source,
55 * for sending the last part */
56#define DEFAULT_MIGRATE_SET_DOWNTIME 300
57
87c9cc1c
DHB		 58/* Maximum migrate downtime set to 2000 seconds */
59#define MAX_MIGRATE_DOWNTIME_SECONDS 2000
60#define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
61
8706d2d5
LL		 62/* Default compression thread count */
63#define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
3fcb38c2
LL		 64/* Default decompression thread count, usually decompression is at
65 * least 4 times as fast as compression.*/
66#define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
8706d2d5
LL		 67/*0: means nocompress, 1: best speed, ... 9: best compress ratio */
68#define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
1626fee3 69/* Define default autoconverge cpu throttle migration parameters */
d85a31d1
JH		 70#define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
71#define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
8706d2d5 72
17ad9b35 73/* Migration XBZRLE default cache size */
73af8dd8 74#define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
17ad9b35 75
68b53591
HZ		 76/* The delay time (in ms) between two COLO checkpoints
77 * Note: Please change this default value to 10000 when we support hybrid mode.
78 */
79#define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
4075fb1c 80#define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
0fb86605 81#define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
68b53591 82
99a0db9b
GH		 83static NotifierList migration_state_notifiers =
84 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
85
adde220a
DDAG		 86static bool deferred_incoming;
87
da6f1790
JQ		 88/* Messages sent on the return path from destination to source */
89enum mig_rp_message_type {
90 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
91 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
92 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
93
94 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
95 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
96
97 MIG_RP_MSG_MAX
98};
99
17549e84
JQ		 100/* When we add fault tolerance, we could have several
101 migrations at once. For now we don't need to add
102 dynamic creation of migration */
103
e5cb7e76
PX		 104static MigrationState *current_migration;
105
8b0b29dc 106static bool migration_object_check(MigrationState *ms, Error **errp);
0331c8ca
DDAG		 107static int migration_maybe_pause(MigrationState *s,
108 int *current_active_state,
109 int new_state);
8b0b29dc 110
e5cb7e76
PX		 111void migration_object_init(void)
112{
4ffdb337 113 MachineState *ms = MACHINE(qdev_get_machine());
8b0b29dc 114 Error *err = NULL;
4ffdb337 115
e5cb7e76
PX		 116 /* This can only be called once. */
117 assert(!current_migration);
118 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
4ffdb337 119
8b0b29dc
PX		 120 if (!migration_object_check(current_migration, &err)) {
121 error_report_err(err);
122 exit(1);
123 }
124
4ffdb337
PX		 125 /*
126 * We cannot really do this in migration_instance_init() since at
127 * that time global properties are not yet applied, then this
128 * value will be definitely replaced by something else.
129 */
130 if (ms->enforce_config_section) {
131 current_migration->send_configuration = true;
132 }
e5cb7e76
PX		 133}
134
bca7856a 135/* For outgoing */
859bc756 136MigrationState *migrate_get_current(void)
17549e84 137{
e5cb7e76
PX		 138 /* This can only be called after the object created. */
139 assert(current_migration);
140 return current_migration;
17549e84
JQ		 141}
142
bca7856a
DDAG		 143MigrationIncomingState *migration_incoming_get_current(void)
144{
b4b076da
JQ		 145 static bool once;
146 static MigrationIncomingState mis_current;
bca7856a 147
b4b076da
JQ		 148 if (!once) {
149 mis_current.state = MIGRATION_STATUS_NONE;
150 memset(&mis_current, 0, sizeof(MigrationIncomingState));
b4b076da
JQ		 151 qemu_mutex_init(&mis_current.rp_mutex);
152 qemu_event_init(&mis_current.main_thread_load_event, false);
153 once = true;
154 }
155 return &mis_current;
bca7856a
DDAG		 156}
157
158void migration_incoming_state_destroy(void)
159{
b4b076da
JQ		 160 struct MigrationIncomingState *mis = migration_incoming_get_current();
161
3482655b 162 if (mis->to_src_file) {
660819b1
PX		 163 /* Tell source that we are done */
164 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
3482655b
PX		 165 qemu_fclose(mis->to_src_file);
166 mis->to_src_file = NULL;
167 }
168
660819b1
PX		 169 if (mis->from_src_file) {
170 qemu_fclose(mis->from_src_file);
171 mis->from_src_file = NULL;
172 }
173
5089e186 174 qemu_event_reset(&mis->main_thread_load_event);
bca7856a
DDAG		 175}
176
b05dc723
JQ		 177static void migrate_generate_event(int new_state)
178{
179 if (migrate_use_events()) {
180 qapi_event_send_migration(new_state, &error_abort);
b05dc723
JQ		 181 }
182}
183
adde220a
DDAG		 184/*
185 * Called on -incoming with a defer: uri.
186 * The migration can be started later after any parameters have been
187 * changed.
188 */
189static void deferred_incoming_migration(Error **errp)
190{
191 if (deferred_incoming) {
192 error_setg(errp, "Incoming migration already deferred");
193 }
194 deferred_incoming = true;
195}
196
da6f1790
JQ		 197/*
198 * Send a message on the return channel back to the source
199 * of the migration.
200 */
201static void migrate_send_rp_message(MigrationIncomingState *mis,
202 enum mig_rp_message_type message_type,
203 uint16_t len, void *data)
204{
205 trace_migrate_send_rp_message((int)message_type, len);
206 qemu_mutex_lock(&mis->rp_mutex);
207 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
208 qemu_put_be16(mis->to_src_file, len);
209 qemu_put_buffer(mis->to_src_file, data, len);
210 qemu_fflush(mis->to_src_file);
211 qemu_mutex_unlock(&mis->rp_mutex);
212}
213
1e2d90eb
DDAG		 214/* Request a range of pages from the source VM at the given
215 * start address.
216 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
217 * as the last request (a name must have been given previously)
218 * Start: Address offset within the RB
219 * Len: Length in bytes required - must be a multiple of pagesize
220 */
221void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
222 ram_addr_t start, size_t len)
223{
cb8d4c8f 224 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
1e2d90eb
DDAG		 225 size_t msglen = 12; /* start + len */
226
227 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
228 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
229
230 if (rbname) {
231 int rbname_len = strlen(rbname);
232 assert(rbname_len < 256);
233
234 bufc[msglen++] = rbname_len;
235 memcpy(bufc + msglen, rbname, rbname_len);
236 msglen += rbname_len;
237 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
238 } else {
239 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
240 }
241}
242
43eaae28 243void qemu_start_incoming_migration(const char *uri, Error **errp)
5bb7910a 244{
34c9dd8e
AL		 245 const char *p;
246
7cf1fe6d 247 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
adde220a
DDAG		 248 if (!strcmp(uri, "defer")) {
249 deferred_incoming_migration(errp);
250 } else if (strstart(uri, "tcp:", &p)) {
43eaae28 251 tcp_start_incoming_migration(p, errp);
2da776db 252#ifdef CONFIG_RDMA
adde220a 253 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 254 rdma_start_incoming_migration(p, errp);
255#endif
adde220a 256 } else if (strstart(uri, "exec:", &p)) {
43eaae28 257 exec_start_incoming_migration(p, errp);
adde220a 258 } else if (strstart(uri, "unix:", &p)) {
43eaae28 259 unix_start_incoming_migration(p, errp);
adde220a 260 } else if (strstart(uri, "fd:", &p)) {
43eaae28 261 fd_start_incoming_migration(p, errp);
adde220a 262 } else {
312fd5f2 263 error_setg(errp, "unknown migration protocol: %s", uri);
8ca5e801 264 }
5bb7910a
AL		 265}
266
0aa6aefc
DL		 267static void process_incoming_migration_bh(void *opaque)
268{
269 Error *local_err = NULL;
270 MigrationIncomingState *mis = opaque;
271
ace21a58
KW		 272 /* Make sure all file formats flush their mutable metadata.
273 * If we get an error here, just don't restart the VM yet. */
0aa6aefc 274 bdrv_invalidate_cache_all(&local_err);
d35ff5e6 275 if (local_err) {
ace21a58 276 error_report_err(local_err);
d35ff5e6
KW		 277 local_err = NULL;
278 autostart = false;
279 }
280
0aa6aefc
DL		 281 /*
282 * This must happen after all error conditions are dealt with and
283 * we're sure the VM is going to be running on this host.
284 */
285 qemu_announce_self();
286
f986c3d2
JQ		 287 if (multifd_load_cleanup(&local_err) != 0) {
288 error_report_err(local_err);
289 autostart = false;
290 }
0aa6aefc
DL		 291 /* If global state section was not received or we are in running
292 state, we need to obey autostart. Any other state is set with
293 runstate_set. */
294
295 if (!global_state_received() ||
296 global_state_get_runstate() == RUN_STATE_RUNNING) {
297 if (autostart) {
298 vm_start();
299 } else {
300 runstate_set(RUN_STATE_PAUSED);
301 }
302 } else {
303 runstate_set(global_state_get_runstate());
304 }
0aa6aefc
DL		 305 /*
306 * This must happen after any state changes since as soon as an external
307 * observer sees this event they might start to prod at the VM assuming
308 * it's ready to use.
309 */
310 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
311 MIGRATION_STATUS_COMPLETED);
312 qemu_bh_delete(mis->bh);
313 migration_incoming_state_destroy();
314}
315
82a4da79 316static void process_incoming_migration_co(void *opaque)
511c0231 317{
b4b076da 318 MigrationIncomingState *mis = migration_incoming_get_current();
e9bef235 319 PostcopyState ps;
1c12e1f5
PB		 320 int ret;
321
4f0fae7f 322 assert(mis->from_src_file);
67f11b5c 323 mis->largest_page_size = qemu_ram_pagesize_largest();
093e3c42 324 postcopy_state_set(POSTCOPY_INCOMING_NONE);
93d7af6f
HZ		 325 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
326 MIGRATION_STATUS_ACTIVE);
4f0fae7f 327 ret = qemu_loadvm_state(mis->from_src_file);
bca7856a 328
e9bef235
DDAG		 329 ps = postcopy_state_get();
330 trace_process_incoming_migration_co_end(ret, ps);
331 if (ps != POSTCOPY_INCOMING_NONE) {
332 if (ps == POSTCOPY_INCOMING_ADVISE) {
333 /*
334 * Where a migration had postcopy enabled (and thus went to advise)
335 * but managed to complete within the precopy period, we can use
336 * the normal exit.
337 */
338 postcopy_ram_incoming_cleanup(mis);
339 } else if (ret >= 0) {
340 /*
341 * Postcopy was started, cleanup should happen at the end of the
342 * postcopy thread.
343 */
344 trace_process_incoming_migration_co_postcopy_end_main();
345 return;
346 }
347 /* Else if something went wrong then just fall out of the normal exit */
348 }
349
25d0c16f
HZ		 350 /* we get COLO info, and know if we are in COLO mode */
351 if (!ret && migration_incoming_enable_colo()) {
352 mis->migration_incoming_co = qemu_coroutine_self();
353 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
354 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
355 mis->have_colo_incoming_thread = true;
356 qemu_coroutine_yield();
357
358 /* Wait checkpoint incoming thread exit before free resource */
359 qemu_thread_join(&mis->colo_incoming_thread);
360 }
361
1c12e1f5 362 if (ret < 0) {
f986c3d2
JQ		 363 Error *local_err = NULL;
364
93d7af6f
HZ		 365 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
366 MIGRATION_STATUS_FAILED);
db80face 367 error_report("load of migration failed: %s", strerror(-ret));
3a0f2cea 368 qemu_fclose(mis->from_src_file);
f986c3d2
JQ		 369 if (multifd_load_cleanup(&local_err) != 0) {
370 error_report_err(local_err);
371 }
4aead692 372 exit(EXIT_FAILURE);
511c0231 373 }
0aa6aefc
DL		 374 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
375 qemu_bh_schedule(mis->bh);
511c0231
JQ		 376}
377
e595a01a 378static void migration_incoming_setup(QEMUFile *f)
82a4da79 379{
4f0fae7f 380 MigrationIncomingState *mis = migration_incoming_get_current();
82a4da79 381
f986c3d2
JQ		 382 if (multifd_load_setup() != 0) {
383 /* We haven't been able to create multifd threads
384 nothing better to do */
385 exit(EXIT_FAILURE);
386 }
387
4f0fae7f
JQ		 388 if (!mis->from_src_file) {
389 mis->from_src_file = f;
390 }
06ad5135 391 qemu_file_set_blocking(f, false);
e595a01a
JQ		 392}
393
394static void migration_incoming_process(void)
395{
396 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
0b8b8753 397 qemu_coroutine_enter(co);
82a4da79
PB		 398}
399
e595a01a
JQ		 400void migration_fd_process_incoming(QEMUFile *f)
401{
402 migration_incoming_setup(f);
403 migration_incoming_process();
404}
405
4f0fae7f
JQ		 406void migration_ioc_process_incoming(QIOChannel *ioc)
407{
408 MigrationIncomingState *mis = migration_incoming_get_current();
409
410 if (!mis->from_src_file) {
411 QEMUFile *f = qemu_fopen_channel_input(ioc);
412 migration_fd_process_incoming(f);
413 }
414 /* We still only have a single channel. Nothing to do here yet */
415}
416
428d8908
JQ		 417/**
418 * @migration_has_all_channels: We have received all channels that we need
419 *
420 * Returns true when we have got connections to all the channels that
421 * we need for migration.
422 */
423bool migration_has_all_channels(void)
424{
425 return true;
426}
427
6decec93
DDAG		 428/*
429 * Send a 'SHUT' message on the return channel with the given value
430 * to indicate that we've finished with the RP. Non-0 value indicates
431 * error.
432 */
433void migrate_send_rp_shut(MigrationIncomingState *mis,
434 uint32_t value)
435{
436 uint32_t buf;
437
438 buf = cpu_to_be32(value);
439 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
440}
441
442/*
443 * Send a 'PONG' message on the return channel with the given value
444 * (normally in response to a 'PING')
445 */
446void migrate_send_rp_pong(MigrationIncomingState *mis,
447 uint32_t value)
448{
449 uint32_t buf;
450
451 buf = cpu_to_be32(value);
452 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
453}
454
bbf6da32
OW		 455MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
456{
457 MigrationCapabilityStatusList *head = NULL;
458 MigrationCapabilityStatusList *caps;
459 MigrationState *s = migrate_get_current();
460 int i;
461
387eedeb 462 caps = NULL; /* silence compiler warning */
7fb1cf16 463 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
ed1701c6
DDAG		 464#ifndef CONFIG_LIVE_BLOCK_MIGRATION
465 if (i == MIGRATION_CAPABILITY_BLOCK) {
466 continue;
467 }
468#endif
bbf6da32
OW		 469 if (head == NULL) {
470 head = g_malloc0(sizeof(*caps));
471 caps = head;
472 } else {
473 caps->next = g_malloc0(sizeof(*caps));
474 caps = caps->next;
475 }
476 caps->value =
477 g_malloc(sizeof(*caps->value));
478 caps->value->capability = i;
479 caps->value->state = s->enabled_capabilities[i];
480 }
481
482 return head;
483}
484
85de8323
LL		 485MigrationParameters *qmp_query_migrate_parameters(Error **errp)
486{
487 MigrationParameters *params;
488 MigrationState *s = migrate_get_current();
489
e87fae4c 490 /* TODO use QAPI_CLONE() instead of duplicating it inline */
85de8323 491 params = g_malloc0(sizeof(*params));
de63ab61 492 params->has_compress_level = true;
2594f56d 493 params->compress_level = s->parameters.compress_level;
de63ab61 494 params->has_compress_threads = true;
2594f56d 495 params->compress_threads = s->parameters.compress_threads;
de63ab61 496 params->has_decompress_threads = true;
2594f56d 497 params->decompress_threads = s->parameters.decompress_threads;
de63ab61 498 params->has_cpu_throttle_initial = true;
2594f56d 499 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
de63ab61 500 params->has_cpu_throttle_increment = true;
2594f56d 501 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
8cc99dcd 502 params->has_tls_creds = true;
69ef1f36 503 params->tls_creds = g_strdup(s->parameters.tls_creds);
8cc99dcd 504 params->has_tls_hostname = true;
69ef1f36 505 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
2ff30257
AA		 506 params->has_max_bandwidth = true;
507 params->max_bandwidth = s->parameters.max_bandwidth;
508 params->has_downtime_limit = true;
509 params->downtime_limit = s->parameters.downtime_limit;
fe39a4d4 510 params->has_x_checkpoint_delay = true;
68b53591 511 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
2833c59b
JQ		 512 params->has_block_incremental = true;
513 params->block_incremental = s->parameters.block_incremental;
4075fb1c
JQ		 514 params->has_x_multifd_channels = true;
515 params->x_multifd_channels = s->parameters.x_multifd_channels;
0fb86605
JQ		 516 params->has_x_multifd_page_count = true;
517 params->x_multifd_page_count = s->parameters.x_multifd_page_count;
73af8dd8
JQ		 518 params->has_xbzrle_cache_size = true;
519 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
85de8323
LL		 520
521 return params;
522}
523
f6844b99
DDAG		 524/*
525 * Return true if we're already in the middle of a migration
526 * (i.e. any of the active or setup states)
527 */
528static bool migration_is_setup_or_active(int state)
529{
530 switch (state) {
531 case MIGRATION_STATUS_ACTIVE:
9ec055ae 532 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
f6844b99 533 case MIGRATION_STATUS_SETUP:
31e06077
DDAG		 534 case MIGRATION_STATUS_PRE_SWITCHOVER:
535 case MIGRATION_STATUS_DEVICE:
f6844b99
DDAG		 536 return true;
537
538 default:
539 return false;
540
541 }
542}
543
a22463a5
DDAG		 544static void populate_ram_info(MigrationInfo *info, MigrationState *s)
545{
546 info->has_ram = true;
547 info->ram = g_malloc0(sizeof(*info->ram));
9360447d 548 info->ram->transferred = ram_counters.transferred;
a22463a5 549 info->ram->total = ram_bytes_total();
9360447d 550 info->ram->duplicate = ram_counters.duplicate;
bedf53c1
JQ		 551 /* legacy value. It is not used anymore */
552 info->ram->skipped = 0;
9360447d
JQ		 553 info->ram->normal = ram_counters.normal;
554 info->ram->normal_bytes = ram_counters.normal *
20afaed9 555 qemu_target_page_size();
a22463a5 556 info->ram->mbps = s->mbps;
9360447d
JQ		 557 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
558 info->ram->postcopy_requests = ram_counters.postcopy_requests;
030ce1f8 559 info->ram->page_size = qemu_target_page_size();
a22463a5 560
114f5aee
JQ		 561 if (migrate_use_xbzrle()) {
562 info->has_xbzrle_cache = true;
563 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
564 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
9360447d
JQ		 565 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
566 info->xbzrle_cache->pages = xbzrle_counters.pages;
567 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
568 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
569 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
114f5aee
JQ		 570 }
571
338182c8
JQ		 572 if (cpu_throttle_active()) {
573 info->has_cpu_throttle_percentage = true;
574 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
575 }
576
a22463a5
DDAG		 577 if (s->state != MIGRATION_STATUS_COMPLETED) {
578 info->ram->remaining = ram_bytes_remaining();
9360447d 579 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
a22463a5
DDAG		 580 }
581}
582
930ac04c
JQ		 583static void populate_disk_info(MigrationInfo *info)
584{
585 if (blk_mig_active()) {
586 info->has_disk = true;
587 info->disk = g_malloc0(sizeof(*info->disk));
588 info->disk->transferred = blk_mig_bytes_transferred();
589 info->disk->remaining = blk_mig_bytes_remaining();
590 info->disk->total = blk_mig_bytes_total();
591 }
592}
593
791e7c82 594MigrationInfo *qmp_query_migrate(Error **errp)
5bb7910a 595{
791e7c82 596 MigrationInfo *info = g_malloc0(sizeof(*info));
17549e84
JQ		 597 MigrationState *s = migrate_get_current();
598
599 switch (s->state) {
31194731 600 case MIGRATION_STATUS_NONE:
17549e84
JQ		 601 /* no migration has happened ever */
602 break;
31194731 603 case MIGRATION_STATUS_SETUP:
29ae8a41 604 info->has_status = true;
ed4fbd10 605 info->has_total_time = false;
29ae8a41 606 break;
31194731
HZ		 607 case MIGRATION_STATUS_ACTIVE:
608 case MIGRATION_STATUS_CANCELLING:
9ec055ae 609 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
31e06077
DDAG		 610 case MIGRATION_STATUS_PRE_SWITCHOVER:
611 case MIGRATION_STATUS_DEVICE:
c8f9f4f4 612 /* TODO add some postcopy stats */
9ec055ae
DDAG		 613 info->has_status = true;
614 info->has_total_time = true;
615 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
616 - s->total_time;
617 info->has_expected_downtime = true;
618 info->expected_downtime = s->expected_downtime;
619 info->has_setup_time = true;
620 info->setup_time = s->setup_time;
621
a22463a5 622 populate_ram_info(info, s);
930ac04c 623 populate_disk_info(info);
17549e84 624 break;
0b827d5e
HZ		 625 case MIGRATION_STATUS_COLO:
626 info->has_status = true;
627 /* TODO: display COLO specific information (checkpoint info etc.) */
628 break;
31194731 629 case MIGRATION_STATUS_COMPLETED:
791e7c82 630 info->has_status = true;
00c14997 631 info->has_total_time = true;
7aa939af 632 info->total_time = s->total_time;
9c5a9fcf
JQ		 633 info->has_downtime = true;
634 info->downtime = s->downtime;
ed4fbd10
MH		 635 info->has_setup_time = true;
636 info->setup_time = s->setup_time;
d5f8a570 637
a22463a5 638 populate_ram_info(info, s);
17549e84 639 break;
31194731 640 case MIGRATION_STATUS_FAILED:
791e7c82 641 info->has_status = true;
d59ce6f3
DB		 642 if (s->error) {
643 info->has_error_desc = true;
644 info->error_desc = g_strdup(error_get_pretty(s->error));
645 }
17549e84 646 break;
31194731 647 case MIGRATION_STATUS_CANCELLED:
791e7c82 648 info->has_status = true;
17549e84 649 break;
5bb7910a 650 }
cde63fbe 651 info->status = s->state;
791e7c82
LC		 652
653 return info;
5bb7910a
AL		 654}
655
4a84214e
PX		 656/**
657 * @migration_caps_check - check capability validity
658 *
659 * @cap_list: old capability list, array of bool
660 * @params: new capabilities to be applied soon
661 * @errp: set *errp if the check failed, with reason
662 *
663 * Returns true if check passed, otherwise false.
664 */
665static bool migrate_caps_check(bool *cap_list,
666 MigrationCapabilityStatusList *params,
667 Error **errp)
00458433 668{
00458433 669 MigrationCapabilityStatusList *cap;
4a84214e 670 bool old_postcopy_cap;
d7651f15 671 MigrationIncomingState *mis = migration_incoming_get_current();
00458433 672
4a84214e 673 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
00458433
OW		 674
675 for (cap = params; cap; cap = cap->next) {
4a84214e
PX		 676 cap_list[cap->value->capability] = cap->value->state;
677 }
678
ed1701c6 679#ifndef CONFIG_LIVE_BLOCK_MIGRATION
4a84214e
PX		 680 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
681 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
682 "block migration");
683 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
684 return false;
00458433 685 }
4a84214e 686#endif
53dd370c 687
4a84214e
PX		 688 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
689 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
53dd370c
DDAG		 690 /* The decompression threads asynchronously write into RAM
691 * rather than use the atomic copies needed to avoid
692 * userfaulting. It should be possible to fix the decompression
693 * threads for compatibility in future.
694 */
4a84214e
PX		 695 error_setg(errp, "Postcopy is not currently compatible "
696 "with compression");
697 return false;
53dd370c 698 }
4a84214e 699
096631bd
DDAG		 700 /* This check is reasonably expensive, so only when it's being
701 * set the first time, also it's only the destination that needs
702 * special support.
703 */
704 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
d7651f15 705 !postcopy_ram_supported_by_host(mis)) {
096631bd
DDAG		 706 /* postcopy_ram_supported_by_host will have emitted a more
707 * detailed message
708 */
4a84214e
PX		 709 error_setg(errp, "Postcopy is not supported");
710 return false;
096631bd 711 }
53dd370c 712 }
4a84214e
PX		 713
714 return true;
715}
716
717void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
718 Error **errp)
719{
720 MigrationState *s = migrate_get_current();
721 MigrationCapabilityStatusList *cap;
722
723 if (migration_is_setup_or_active(s->state)) {
724 error_setg(errp, QERR_MIGRATION_ACTIVE);
725 return;
726 }
727
728 if (!migrate_caps_check(s->enabled_capabilities, params, errp)) {
729 return;
730 }
731
732 for (cap = params; cap; cap = cap->next) {
733 s->enabled_capabilities[cap->value->capability] = cap->value->state;
734 }
00458433
OW		 735}
736
16d063bc
PX		 737/*
738 * Check whether the parameters are valid. Error will be put into errp
739 * (if provided). Return true if valid, otherwise false.
740 */
741static bool migrate_params_check(MigrationParameters *params, Error **errp)
85de8323 742{
7f375e04
EB		 743 if (params->has_compress_level &&
744 (params->compress_level < 0 || params->compress_level > 9)) {
c6bd8c70
MA		 745 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
746 "is invalid, it should be in the range of 0 to 9");
16d063bc 747 return false;
85de8323 748 }
16d063bc 749
7f375e04
EB		 750 if (params->has_compress_threads &&
751 (params->compress_threads < 1 || params->compress_threads > 255)) {
c6bd8c70
MA		 752 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
753 "compress_threads",
754 "is invalid, it should be in the range of 1 to 255");
16d063bc 755 return false;
85de8323 756 }
16d063bc 757
7f375e04
EB		 758 if (params->has_decompress_threads &&
759 (params->decompress_threads < 1 || params->decompress_threads > 255)) {
c6bd8c70
MA		 760 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
761 "decompress_threads",
762 "is invalid, it should be in the range of 1 to 255");
16d063bc 763 return false;
85de8323 764 }
16d063bc 765
7f375e04
EB		 766 if (params->has_cpu_throttle_initial &&
767 (params->cpu_throttle_initial < 1 ||
768 params->cpu_throttle_initial > 99)) {
1626fee3 769 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 770 "cpu_throttle_initial",
1626fee3 771 "an integer in the range of 1 to 99");
16d063bc 772 return false;
1626fee3 773 }
16d063bc 774
7f375e04
EB		 775 if (params->has_cpu_throttle_increment &&
776 (params->cpu_throttle_increment < 1 ||
777 params->cpu_throttle_increment > 99)) {
1626fee3 778 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 779 "cpu_throttle_increment",
1626fee3 780 "an integer in the range of 1 to 99");
16d063bc 781 return false;
1626fee3 782 }
16d063bc 783
2ff30257
AA		 784 if (params->has_max_bandwidth &&
785 (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
786 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
787 " range of 0 to %zu bytes/second", SIZE_MAX);
16d063bc 788 return false;
2ff30257 789 }
16d063bc 790
2ff30257 791 if (params->has_downtime_limit &&
87c9cc1c
DHB		 792 (params->downtime_limit < 0 ||
793 params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
794 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
795 "the range of 0 to %d milliseconds",
796 MAX_MIGRATE_DOWNTIME);
16d063bc 797 return false;
2ff30257 798 }
16d063bc 799
68b53591
HZ		 800 if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
801 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
802 "x_checkpoint_delay",
803 "is invalid, it should be positive");
16d063bc
PX		 804 return false;
805 }
4075fb1c
JQ		 806 if (params->has_x_multifd_channels &&
807 (params->x_multifd_channels < 1 || params->x_multifd_channels > 255)) {
808 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
809 "multifd_channels",
810 "is invalid, it should be in the range of 1 to 255");
811 return false;
812 }
0fb86605
JQ		 813 if (params->has_x_multifd_page_count &&
814 (params->x_multifd_page_count < 1 ||
815 params->x_multifd_page_count > 10000)) {
816 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
817 "multifd_page_count",
818 "is invalid, it should be in the range of 1 to 10000");
819 return false;
820 }
16d063bc 821
73af8dd8
JQ		 822 if (params->has_xbzrle_cache_size &&
823 (params->xbzrle_cache_size < qemu_target_page_size() ||
824 !is_power_of_2(params->xbzrle_cache_size))) {
825 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
826 "xbzrle_cache_size",
827 "is invalid, it should be bigger than target page size"
828 " and a power of two");
829 return false;
830 }
831
16d063bc
PX		 832 return true;
833}
834
1bda8b3c
MA		 835static void migrate_params_test_apply(MigrateSetParameters *params,
836 MigrationParameters *dest)
837{
838 *dest = migrate_get_current()->parameters;
839
840 /* TODO use QAPI_CLONE() instead of duplicating it inline */
841
842 if (params->has_compress_level) {
843 dest->compress_level = params->compress_level;
844 }
845
846 if (params->has_compress_threads) {
847 dest->compress_threads = params->compress_threads;
848 }
849
850 if (params->has_decompress_threads) {
851 dest->decompress_threads = params->decompress_threads;
852 }
853
854 if (params->has_cpu_throttle_initial) {
855 dest->cpu_throttle_initial = params->cpu_throttle_initial;
856 }
857
858 if (params->has_cpu_throttle_increment) {
859 dest->cpu_throttle_increment = params->cpu_throttle_increment;
860 }
861
862 if (params->has_tls_creds) {
01fa5598
MA		 863 assert(params->tls_creds->type == QTYPE_QSTRING);
864 dest->tls_creds = g_strdup(params->tls_creds->u.s);
1bda8b3c
MA		 865 }
866
867 if (params->has_tls_hostname) {
01fa5598
MA		 868 assert(params->tls_hostname->type == QTYPE_QSTRING);
869 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1bda8b3c
MA		 870 }
871
872 if (params->has_max_bandwidth) {
873 dest->max_bandwidth = params->max_bandwidth;
874 }
875
876 if (params->has_downtime_limit) {
877 dest->downtime_limit = params->downtime_limit;
878 }
879
880 if (params->has_x_checkpoint_delay) {
881 dest->x_checkpoint_delay = params->x_checkpoint_delay;
882 }
883
884 if (params->has_block_incremental) {
885 dest->block_incremental = params->block_incremental;
886 }
5e7577a1
JQ		 887 if (params->has_x_multifd_channels) {
888 dest->x_multifd_channels = params->x_multifd_channels;
889 }
890 if (params->has_x_multifd_page_count) {
891 dest->x_multifd_page_count = params->x_multifd_page_count;
892 }
73af8dd8
JQ		 893 if (params->has_xbzrle_cache_size) {
894 dest->xbzrle_cache_size = params->xbzrle_cache_size;
895 }
1bda8b3c
MA		 896}
897
73af8dd8 898static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
16d063bc
PX		 899{
900 MigrationState *s = migrate_get_current();
901
e87fae4c
MA		 902 /* TODO use QAPI_CLONE() instead of duplicating it inline */
903
7f375e04
EB		 904 if (params->has_compress_level) {
905 s->parameters.compress_level = params->compress_level;
85de8323 906 }
476c72aa 907
7f375e04
EB		 908 if (params->has_compress_threads) {
909 s->parameters.compress_threads = params->compress_threads;
85de8323 910 }
476c72aa 911
7f375e04
EB		 912 if (params->has_decompress_threads) {
913 s->parameters.decompress_threads = params->decompress_threads;
85de8323 914 }
476c72aa 915
7f375e04
EB		 916 if (params->has_cpu_throttle_initial) {
917 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1626fee3 918 }
476c72aa 919
7f375e04
EB		 920 if (params->has_cpu_throttle_increment) {
921 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1626fee3 922 }
476c72aa 923
7f375e04 924 if (params->has_tls_creds) {
69ef1f36 925 g_free(s->parameters.tls_creds);
01fa5598
MA		 926 assert(params->tls_creds->type == QTYPE_QSTRING);
927 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
69ef1f36 928 }
476c72aa 929
7f375e04 930 if (params->has_tls_hostname) {
69ef1f36 931 g_free(s->parameters.tls_hostname);
01fa5598
MA		 932 assert(params->tls_hostname->type == QTYPE_QSTRING);
933 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
69ef1f36 934 }
476c72aa 935
2ff30257
AA		 936 if (params->has_max_bandwidth) {
937 s->parameters.max_bandwidth = params->max_bandwidth;
938 if (s->to_dst_file) {
939 qemu_file_set_rate_limit(s->to_dst_file,
940 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
941 }
942 }
476c72aa 943
2ff30257
AA		 944 if (params->has_downtime_limit) {
945 s->parameters.downtime_limit = params->downtime_limit;
946 }
68b53591
HZ		 947
948 if (params->has_x_checkpoint_delay) {
949 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
479125d5
HZ		 950 if (migration_in_colo_state()) {
951 colo_checkpoint_notify(s);
952 }
68b53591 953 }
476c72aa 954
2833c59b
JQ		 955 if (params->has_block_incremental) {
956 s->parameters.block_incremental = params->block_incremental;
957 }
4075fb1c
JQ		 958 if (params->has_x_multifd_channels) {
959 s->parameters.x_multifd_channels = params->x_multifd_channels;
960 }
0fb86605
JQ		 961 if (params->has_x_multifd_page_count) {
962 s->parameters.x_multifd_page_count = params->x_multifd_page_count;
963 }
73af8dd8
JQ		 964 if (params->has_xbzrle_cache_size) {
965 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
966 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
967 }
85de8323
LL		 968}
969
1bda8b3c 970void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
476c72aa 971{
1bda8b3c
MA		 972 MigrationParameters tmp;
973
01fa5598
MA		 974 /* TODO Rewrite "" to null instead */
975 if (params->has_tls_creds
976 && params->tls_creds->type == QTYPE_QNULL) {
977 QDECREF(params->tls_creds->u.n);
978 params->tls_creds->type = QTYPE_QSTRING;
979 params->tls_creds->u.s = strdup("");
980 }
981 /* TODO Rewrite "" to null instead */
982 if (params->has_tls_hostname
983 && params->tls_hostname->type == QTYPE_QNULL) {
984 QDECREF(params->tls_hostname->u.n);
985 params->tls_hostname->type = QTYPE_QSTRING;
986 params->tls_hostname->u.s = strdup("");
987 }
988
1bda8b3c
MA		 989 migrate_params_test_apply(params, &tmp);
990
991 if (!migrate_params_check(&tmp, errp)) {
476c72aa
PX		 992 /* Invalid parameter */
993 return;
994 }
995
73af8dd8 996 migrate_params_apply(params, errp);
476c72aa
PX		 997}
998
2594f56d 999
4886a1bc
DDAG		 1000void qmp_migrate_start_postcopy(Error **errp)
1001{
1002 MigrationState *s = migrate_get_current();
1003
1004 if (!migrate_postcopy_ram()) {
a54d340b 1005 error_setg(errp, "Enable postcopy with migrate_set_capability before"
4886a1bc
DDAG		 1006 " the start of migration");
1007 return;
1008 }
1009
1010 if (s->state == MIGRATION_STATUS_NONE) {
1011 error_setg(errp, "Postcopy must be started after migration has been"
1012 " started");
1013 return;
1014 }
1015 /*
1016 * we don't error if migration has finished since that would be racy
1017 * with issuing this command.
1018 */
1019 atomic_set(&s->start_postcopy, true);
1020}
1021
065e2813
AL		 1022/* shared migration helpers */
1023
48781e5b 1024void migrate_set_state(int *state, int old_state, int new_state)
51cf4c1a 1025{
a31fedee 1026 assert(new_state < MIGRATION_STATUS__MAX);
48781e5b 1027 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
a31fedee 1028 trace_migrate_set_state(MigrationStatus_str(new_state));
b05dc723 1029 migrate_generate_event(new_state);
51cf4c1a
Z		 1030 }
1031}
1032
4e4a3d3a
PX		 1033static MigrationCapabilityStatusList *migrate_cap_add(
1034 MigrationCapabilityStatusList *list,
1035 MigrationCapability index,
1036 bool state)
2833c59b
JQ		 1037{
1038 MigrationCapabilityStatusList *cap;
1039
1040 cap = g_new0(MigrationCapabilityStatusList, 1);
1041 cap->value = g_new0(MigrationCapabilityStatus, 1);
4e4a3d3a
PX		 1042 cap->value->capability = index;
1043 cap->value->state = state;
1044 cap->next = list;
1045
1046 return cap;
1047}
1048
1049void migrate_set_block_enabled(bool value, Error **errp)
1050{
1051 MigrationCapabilityStatusList *cap;
1052
1053 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
2833c59b
JQ		 1054 qmp_migrate_set_capabilities(cap, errp);
1055 qapi_free_MigrationCapabilityStatusList(cap);
1056}
1057
1058static void migrate_set_block_incremental(MigrationState *s, bool value)
1059{
1060 s->parameters.block_incremental = value;
1061}
1062
1063static void block_cleanup_parameters(MigrationState *s)
1064{
1065 if (s->must_remove_block_options) {
1066 /* setting to false can never fail */
1067 migrate_set_block_enabled(false, &error_abort);
1068 migrate_set_block_incremental(s, false);
1069 s->must_remove_block_options = false;
1070 }
1071}
1072
bb1fadc4 1073static void migrate_fd_cleanup(void *opaque)
065e2813 1074{
bb1fadc4
PB		 1075 MigrationState *s = opaque;
1076
1077 qemu_bh_delete(s->cleanup_bh);
1078 s->cleanup_bh = NULL;
1079
89a02a9f 1080 if (s->to_dst_file) {
f986c3d2
JQ		 1081 Error *local_err = NULL;
1082
9013dca5 1083 trace_migrate_fd_cleanup();
404a7c05 1084 qemu_mutex_unlock_iothread();
1d34e4bf
DDAG		 1085 if (s->migration_thread_running) {
1086 qemu_thread_join(&s->thread);
1087 s->migration_thread_running = false;
1088 }
404a7c05
PB		 1089 qemu_mutex_lock_iothread();
1090
f986c3d2
JQ		 1091 if (multifd_save_cleanup(&local_err) != 0) {
1092 error_report_err(local_err);
1093 }
89a02a9f
HZ		 1094 qemu_fclose(s->to_dst_file);
1095 s->to_dst_file = NULL;
065e2813
AL		 1096 }
1097
9ec055ae
DDAG		 1098 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1099 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
7a2c1721 1100
94f5a437 1101 if (s->state == MIGRATION_STATUS_CANCELLING) {
48781e5b 1102 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
94f5a437 1103 MIGRATION_STATUS_CANCELLED);
7a2c1721 1104 }
a3fa1d78 1105
87db1a7d
JQ		 1106 if (s->error) {
1107 /* It is used on info migrate. We can't free it */
1108 error_report_err(error_copy(s->error));
1109 }
a3fa1d78 1110 notifier_list_notify(&migration_state_notifiers, s);
2833c59b 1111 block_cleanup_parameters(s);
065e2813
AL		 1112}
1113
87db1a7d
JQ		 1114void migrate_set_error(MigrationState *s, const Error *error)
1115{
1116 qemu_mutex_lock(&s->error_mutex);
1117 if (!s->error) {
1118 s->error = error_copy(error);
1119 }
1120 qemu_mutex_unlock(&s->error_mutex);
1121}
1122
d59ce6f3 1123void migrate_fd_error(MigrationState *s, const Error *error)
065e2813 1124{
25174055 1125 trace_migrate_fd_error(error_get_pretty(error));
89a02a9f 1126 assert(s->to_dst_file == NULL);
48781e5b
HZ		 1127 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1128 MIGRATION_STATUS_FAILED);
87db1a7d 1129 migrate_set_error(s, error);
bb1fadc4 1130 notifier_list_notify(&migration_state_notifiers, s);
2833c59b 1131 block_cleanup_parameters(s);
458cf28e
JQ		 1132}
1133
0edda1c4 1134static void migrate_fd_cancel(MigrationState *s)
065e2813 1135{
6f2b811a 1136 int old_state ;
89a02a9f 1137 QEMUFile *f = migrate_get_current()->to_dst_file;
9013dca5 1138 trace_migrate_fd_cancel();
065e2813 1139
70b20477
DDAG		 1140 if (s->rp_state.from_dst_file) {
1141 /* shutdown the rp socket, so causing the rp thread to shutdown */
1142 qemu_file_shutdown(s->rp_state.from_dst_file);
1143 }
1144
6f2b811a
Z		 1145 do {
1146 old_state = s->state;
f6844b99 1147 if (!migration_is_setup_or_active(old_state)) {
6f2b811a
Z		 1148 break;
1149 }
a7b36b48
DDAG		 1150 /* If the migration is paused, kick it out of the pause */
1151 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1152 qemu_sem_post(&s->pause_sem);
1153 }
48781e5b 1154 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
31194731 1155 } while (s->state != MIGRATION_STATUS_CANCELLING);
a26ba26e
DDAG		 1156
1157 /*
1158 * If we're unlucky the migration code might be stuck somewhere in a
1159 * send/write while the network has failed and is waiting to timeout;
1160 * if we've got shutdown(2) available then we can force it to quit.
1161 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1162 * called in a bh, so there is no race against this cancel.
1163 */
31194731 1164 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
a26ba26e
DDAG		 1165 qemu_file_shutdown(f);
1166 }
1d2acc31
HZ		 1167 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1168 Error *local_err = NULL;
1169
1170 bdrv_invalidate_cache_all(&local_err);
1171 if (local_err) {
1172 error_report_err(local_err);
1173 } else {
1174 s->block_inactive = false;
1175 }
1176 }
2833c59b 1177 block_cleanup_parameters(s);
065e2813
AL		 1178}
1179
99a0db9b
GH		 1180void add_migration_state_change_notifier(Notifier *notify)
1181{
1182 notifier_list_add(&migration_state_notifiers, notify);
1183}
1184
1185void remove_migration_state_change_notifier(Notifier *notify)
1186{
31552529 1187 notifier_remove(notify);
99a0db9b
GH		 1188}
1189
02edd2e7 1190bool migration_in_setup(MigrationState *s)
afe2df69 1191{
31194731 1192 return s->state == MIGRATION_STATUS_SETUP;
afe2df69
GH		 1193}
1194
7073693b 1195bool migration_has_finished(MigrationState *s)
99a0db9b 1196{
31194731 1197 return s->state == MIGRATION_STATUS_COMPLETED;
99a0db9b 1198}
0edda1c4 1199
afe2df69
GH		 1200bool migration_has_failed(MigrationState *s)
1201{
31194731
HZ		 1202 return (s->state == MIGRATION_STATUS_CANCELLED ||
1203 s->state == MIGRATION_STATUS_FAILED);
afe2df69
GH		 1204}
1205
5727309d 1206bool migration_in_postcopy(void)
9ec055ae 1207{
5727309d
JQ		 1208 MigrationState *s = migrate_get_current();
1209
9ec055ae
DDAG		 1210 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1211}
1212
b82fc321
DDAG		 1213bool migration_in_postcopy_after_devices(MigrationState *s)
1214{
5727309d 1215 return migration_in_postcopy() && s->postcopy_after_devices;
b82fc321
DDAG		 1216}
1217
fab35005 1218bool migration_is_idle(void)
fe44dc91 1219{
fab35005 1220 MigrationState *s = migrate_get_current();
fe44dc91
AA		 1221
1222 switch (s->state) {
1223 case MIGRATION_STATUS_NONE:
1224 case MIGRATION_STATUS_CANCELLED:
1225 case MIGRATION_STATUS_COMPLETED:
1226 case MIGRATION_STATUS_FAILED:
1227 return true;
1228 case MIGRATION_STATUS_SETUP:
1229 case MIGRATION_STATUS_CANCELLING:
1230 case MIGRATION_STATUS_ACTIVE:
1231 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1232 case MIGRATION_STATUS_COLO:
31e06077
DDAG		 1233 case MIGRATION_STATUS_PRE_SWITCHOVER:
1234 case MIGRATION_STATUS_DEVICE:
fe44dc91
AA		 1235 return false;
1236 case MIGRATION_STATUS__MAX:
1237 g_assert_not_reached();
1238 }
1239
1240 return false;
1241}
1242
a0762d9e 1243MigrationState *migrate_init(void)
0edda1c4 1244{
17549e84 1245 MigrationState *s = migrate_get_current();
bbf6da32 1246
389775d1
DDAG		 1247 /*
1248 * Reinitialise all migration state, except
1249 * parameters/capabilities that the user set, and
1250 * locks.
1251 */
1252 s->bytes_xfer = 0;
1253 s->xfer_limit = 0;
1254 s->cleanup_bh = 0;
89a02a9f 1255 s->to_dst_file = NULL;
389775d1 1256 s->state = MIGRATION_STATUS_NONE;
389775d1
DDAG		 1257 s->rp_state.from_dst_file = NULL;
1258 s->rp_state.error = false;
1259 s->mbps = 0.0;
1260 s->downtime = 0;
1261 s->expected_downtime = 0;
389775d1 1262 s->setup_time = 0;
389775d1 1263 s->start_postcopy = false;
b82fc321 1264 s->postcopy_after_devices = false;
389775d1 1265 s->migration_thread_running = false;
d59ce6f3
DB		 1266 error_free(s->error);
1267 s->error = NULL;
389775d1 1268
48781e5b 1269 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
0edda1c4 1270
bc72ad67 1271 s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
0edda1c4
JQ		 1272 return s;
1273}
cab30143 1274
fa2756b7
AL		 1275static GSList *migration_blockers;
1276
fe44dc91 1277int migrate_add_blocker(Error *reason, Error **errp)
fa2756b7 1278{
3df663e5 1279 if (migrate_get_current()->only_migratable) {
b67b8c3a
AA		 1280 error_propagate(errp, error_copy(reason));
1281 error_prepend(errp, "disallowing migration blocker "
1282 "(--only_migratable) for: ");
1283 return -EACCES;
1284 }
1285
fab35005 1286 if (migration_is_idle()) {
fe44dc91
AA		 1287 migration_blockers = g_slist_prepend(migration_blockers, reason);
1288 return 0;
1289 }
1290
1291 error_propagate(errp, error_copy(reason));
1292 error_prepend(errp, "disallowing migration blocker (migration in "
1293 "progress) for: ");
1294 return -EBUSY;
fa2756b7
AL		 1295}
1296
1297void migrate_del_blocker(Error *reason)
1298{
1299 migration_blockers = g_slist_remove(migration_blockers, reason);
1300}
1301
bf1ae1f4
DDAG		 1302void qmp_migrate_incoming(const char *uri, Error **errp)
1303{
1304 Error *local_err = NULL;
4debb5f5 1305 static bool once = true;
bf1ae1f4
DDAG		 1306
1307 if (!deferred_incoming) {
4debb5f5 1308 error_setg(errp, "For use with '-incoming defer'");
bf1ae1f4
DDAG		 1309 return;
1310 }
4debb5f5
DDAG		 1311 if (!once) {
1312 error_setg(errp, "The incoming migration has already been started");
1313 }
bf1ae1f4
DDAG		 1314
1315 qemu_start_incoming_migration(uri, &local_err);
1316
1317 if (local_err) {
1318 error_propagate(errp, local_err);
1319 return;
1320 }
1321
4debb5f5 1322 once = false;
bf1ae1f4
DDAG		 1323}
1324
24f3902b
GK		 1325bool migration_is_blocked(Error **errp)
1326{
1327 if (qemu_savevm_state_blocked(errp)) {
1328 return true;
1329 }
1330
1331 if (migration_blockers) {
250561e1 1332 error_propagate(errp, error_copy(migration_blockers->data));
24f3902b
GK		 1333 return true;
1334 }
1335
1336 return false;
1337}
1338
e1c37d0e
LC		 1339void qmp_migrate(const char *uri, bool has_blk, bool blk,
1340 bool has_inc, bool inc, bool has_detach, bool detach,
1341 Error **errp)
cab30143 1342{
be7059cd 1343 Error *local_err = NULL;
17549e84 1344 MigrationState *s = migrate_get_current();
cab30143 1345 const char *p;
cab30143 1346
f6844b99 1347 if (migration_is_setup_or_active(s->state) ||
0b827d5e
HZ		 1348 s->state == MIGRATION_STATUS_CANCELLING ||
1349 s->state == MIGRATION_STATUS_COLO) {
c6bd8c70 1350 error_setg(errp, QERR_MIGRATION_ACTIVE);
e1c37d0e 1351 return;
cab30143 1352 }
ca99993a
DDAG		 1353 if (runstate_check(RUN_STATE_INMIGRATE)) {
1354 error_setg(errp, "Guest is waiting for an incoming migration");
1355 return;
1356 }
1357
24f3902b 1358 if (migration_is_blocked(errp)) {
e1c37d0e 1359 return;
fa2756b7
AL		 1360 }
1361
2833c59b
JQ		 1362 if ((has_blk && blk) || (has_inc && inc)) {
1363 if (migrate_use_block() || migrate_use_block_incremental()) {
1364 error_setg(errp, "Command options are incompatible with "
1365 "current migration capabilities");
1366 return;
1367 }
1368 migrate_set_block_enabled(true, &local_err);
1369 if (local_err) {
1370 error_propagate(errp, local_err);
1371 return;
1372 }
1373 s->must_remove_block_options = true;
1374 }
1375
1376 if (has_inc && inc) {
1377 migrate_set_block_incremental(s, true);
1378 }
1379
a0762d9e 1380 s = migrate_init();
cab30143
JQ		 1381
1382 if (strstart(uri, "tcp:", &p)) {
f37afb5a 1383 tcp_start_outgoing_migration(s, p, &local_err);
2da776db 1384#ifdef CONFIG_RDMA
41310c68 1385 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 1386 rdma_start_outgoing_migration(s, p, &local_err);
1387#endif
cab30143 1388 } else if (strstart(uri, "exec:", &p)) {
f37afb5a 1389 exec_start_outgoing_migration(s, p, &local_err);
cab30143 1390 } else if (strstart(uri, "unix:", &p)) {
f37afb5a 1391 unix_start_outgoing_migration(s, p, &local_err);
cab30143 1392 } else if (strstart(uri, "fd:", &p)) {
f37afb5a 1393 fd_start_outgoing_migration(s, p, &local_err);
99a0db9b 1394 } else {
c6bd8c70
MA		 1395 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1396 "a valid migration protocol");
48781e5b
HZ		 1397 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1398 MIGRATION_STATUS_FAILED);
e1c37d0e 1399 return;
cab30143
JQ		 1400 }
1401
f37afb5a 1402 if (local_err) {
d59ce6f3 1403 migrate_fd_error(s, local_err);
f37afb5a 1404 error_propagate(errp, local_err);
e1c37d0e 1405 return;
1299c631 1406 }
cab30143
JQ		 1407}
1408
6cdedb07 1409void qmp_migrate_cancel(Error **errp)
cab30143 1410{
17549e84 1411 migrate_fd_cancel(migrate_get_current());
cab30143
JQ		 1412}
1413
89cfc02c
DDAG		 1414void qmp_migrate_continue(MigrationStatus state, Error **errp)
1415{
1416 MigrationState *s = migrate_get_current();
1417 if (s->state != state) {
1418 error_setg(errp, "Migration not in expected state: %s",
1419 MigrationStatus_str(s->state));
1420 return;
1421 }
1422 qemu_sem_post(&s->pause_sem);
1423}
1424
9e1ba4cc
OW		 1425void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1426{
73af8dd8
JQ		 1427 MigrateSetParameters p = {
1428 .has_xbzrle_cache_size = true,
1429 .xbzrle_cache_size = value,
1430 };
c91e681a 1431
73af8dd8 1432 qmp_migrate_set_parameters(&p, errp);
9e1ba4cc
OW		 1433}
1434
1435int64_t qmp_query_migrate_cache_size(Error **errp)
1436{
1437 return migrate_xbzrle_cache_size();
1438}
1439
3dc85383 1440void qmp_migrate_set_speed(int64_t value, Error **errp)
cab30143 1441{
1bda8b3c 1442 MigrateSetParameters p = {
2ff30257
AA		 1443 .has_max_bandwidth = true,
1444 .max_bandwidth = value,
1445 };
cab30143 1446
2ff30257 1447 qmp_migrate_set_parameters(&p, errp);
cab30143
JQ		 1448}
1449
4f0a993b 1450void qmp_migrate_set_downtime(double value, Error **errp)
cab30143 1451{
87c9cc1c
DHB		 1452 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1453 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1454 "the range of 0 to %d seconds",
1455 MAX_MIGRATE_DOWNTIME_SECONDS);
1456 return;
1457 }
1458
2ff30257
AA		 1459 value *= 1000; /* Convert to milliseconds */
1460 value = MAX(0, MIN(INT64_MAX, value));
1461
1bda8b3c 1462 MigrateSetParameters p = {
2ff30257
AA		 1463 .has_downtime_limit = true,
1464 .downtime_limit = value,
1465 };
1466
1467 qmp_migrate_set_parameters(&p, errp);
99a0db9b 1468}
17ad9b35 1469
53f09a10
PB		 1470bool migrate_release_ram(void)
1471{
1472 MigrationState *s;
1473
1474 s = migrate_get_current();
1475
1476 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1477}
1478
53dd370c
DDAG		 1479bool migrate_postcopy_ram(void)
1480{
1481 MigrationState *s;
1482
1483 s = migrate_get_current();
1484
32c3db5b 1485 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
53dd370c
DDAG		 1486}
1487
58110f0a
VSO		 1488bool migrate_postcopy(void)
1489{
1490 return migrate_postcopy_ram();
1491}
1492
bde1e2ec
CV		 1493bool migrate_auto_converge(void)
1494{
1495 MigrationState *s;
1496
1497 s = migrate_get_current();
1498
1499 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1500}
1501
323004a3
PL		 1502bool migrate_zero_blocks(void)
1503{
1504 MigrationState *s;
1505
1506 s = migrate_get_current();
1507
1508 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1509}
1510
8706d2d5
LL		 1511bool migrate_use_compression(void)
1512{
dde4e694
LL		 1513 MigrationState *s;
1514
1515 s = migrate_get_current();
1516
1517 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
8706d2d5
LL		 1518}
1519
1520int migrate_compress_level(void)
1521{
1522 MigrationState *s;
1523
1524 s = migrate_get_current();
1525
2594f56d 1526 return s->parameters.compress_level;
8706d2d5
LL		 1527}
1528
1529int migrate_compress_threads(void)
1530{
1531 MigrationState *s;
1532
1533 s = migrate_get_current();
1534
2594f56d 1535 return s->parameters.compress_threads;
8706d2d5
LL		 1536}
1537
3fcb38c2
LL		 1538int migrate_decompress_threads(void)
1539{
1540 MigrationState *s;
1541
1542 s = migrate_get_current();
1543
2594f56d 1544 return s->parameters.decompress_threads;
3fcb38c2
LL		 1545}
1546
b05dc723
JQ		 1547bool migrate_use_events(void)
1548{
1549 MigrationState *s;
1550
1551 s = migrate_get_current();
1552
1553 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1554}
1555
30126bbf
JQ		 1556bool migrate_use_multifd(void)
1557{
1558 MigrationState *s;
1559
1560 s = migrate_get_current();
1561
1562 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1563}
1564
93fbd031
DDAG		 1565bool migrate_pause_before_switchover(void)
1566{
1567 MigrationState *s;
1568
1569 s = migrate_get_current();
1570
1571 return s->enabled_capabilities[
1572 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
1573}
1574
4075fb1c
JQ		 1575int migrate_multifd_channels(void)
1576{
1577 MigrationState *s;
1578
1579 s = migrate_get_current();
1580
1581 return s->parameters.x_multifd_channels;
1582}
1583
0fb86605
JQ		 1584int migrate_multifd_page_count(void)
1585{
1586 MigrationState *s;
1587
1588 s = migrate_get_current();
1589
1590 return s->parameters.x_multifd_page_count;
1591}
1592
17ad9b35
OW		 1593int migrate_use_xbzrle(void)
1594{
1595 MigrationState *s;
1596
1597 s = migrate_get_current();
1598
1599 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1600}
1601
1602int64_t migrate_xbzrle_cache_size(void)
1603{
1604 MigrationState *s;
1605
1606 s = migrate_get_current();
1607
73af8dd8 1608 return s->parameters.xbzrle_cache_size;
17ad9b35 1609}
0d82d0e8 1610
2833c59b
JQ		 1611bool migrate_use_block(void)
1612{
1613 MigrationState *s;
1614
1615 s = migrate_get_current();
1616
1617 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1618}
1619
c788ada8
PX		 1620bool migrate_use_return_path(void)
1621{
1622 MigrationState *s;
1623
1624 s = migrate_get_current();
1625
1626 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1627}
1628
2833c59b
JQ		 1629bool migrate_use_block_incremental(void)
1630{
1631 MigrationState *s;
1632
1633 s = migrate_get_current();
1634
1635 return s->parameters.block_incremental;
1636}
1637
70b20477
DDAG		 1638/* migration thread support */
1639/*
1640 * Something bad happened to the RP stream, mark an error
1641 * The caller shall print or trace something to indicate why
1642 */
1643static void mark_source_rp_bad(MigrationState *s)
1644{
1645 s->rp_state.error = true;
1646}
1647
1648static struct rp_cmd_args {
1649 ssize_t len; /* -1 = variable */
1650 const char *name;
1651} rp_cmd_args[] = {
1652 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1653 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1654 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1e2d90eb
DDAG		 1655 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1656 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
70b20477
DDAG		 1657 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1658};
1659
1e2d90eb
DDAG		 1660/*
1661 * Process a request for pages received on the return path,
1662 * We're allowed to send more than requested (e.g. to round to our page size)
1663 * and we don't need to send pages that have already been sent.
1664 */
1665static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1666 ram_addr_t start, size_t len)
1667{
6c595cde
DDAG		 1668 long our_host_ps = getpagesize();
1669
1e2d90eb 1670 trace_migrate_handle_rp_req_pages(rbname, start, len);
6c595cde
DDAG		 1671
1672 /*
1673 * Since we currently insist on matching page sizes, just sanity check
1674 * we're being asked for whole host pages.
1675 */
1676 if (start & (our_host_ps-1) ||
1677 (len & (our_host_ps-1))) {
1678 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1679 " len: %zd", __func__, start, len);
1680 mark_source_rp_bad(ms);
1681 return;
1682 }
1683
96506894 1684 if (ram_save_queue_pages(rbname, start, len)) {
6c595cde
DDAG		 1685 mark_source_rp_bad(ms);
1686 }
1e2d90eb
DDAG		 1687}
1688
70b20477
DDAG		 1689/*
1690 * Handles messages sent on the return path towards the source VM
1691 *
1692 */
1693static void *source_return_path_thread(void *opaque)
1694{
1695 MigrationState *ms = opaque;
1696 QEMUFile *rp = ms->rp_state.from_dst_file;
1697 uint16_t header_len, header_type;
568b01ca 1698 uint8_t buf[512];
70b20477 1699 uint32_t tmp32, sibling_error;
1e2d90eb
DDAG		 1700 ram_addr_t start = 0; /* =0 to silence warning */
1701 size_t len = 0, expected_len;
70b20477
DDAG		 1702 int res;
1703
1704 trace_source_return_path_thread_entry();
1705 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1706 migration_is_setup_or_active(ms->state)) {
1707 trace_source_return_path_thread_loop_top();
1708 header_type = qemu_get_be16(rp);
1709 header_len = qemu_get_be16(rp);
1710
1711 if (header_type >= MIG_RP_MSG_MAX ||
1712 header_type == MIG_RP_MSG_INVALID) {
1713 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1714 header_type, header_len);
1715 mark_source_rp_bad(ms);
1716 goto out;
1717 }
1718
1719 if ((rp_cmd_args[header_type].len != -1 &&
1720 header_len != rp_cmd_args[header_type].len) ||
568b01ca 1721 header_len > sizeof(buf)) {
70b20477
DDAG		 1722 error_report("RP: Received '%s' message (0x%04x) with"
1723 "incorrect length %d expecting %zu",
1724 rp_cmd_args[header_type].name, header_type, header_len,
1725 (size_t)rp_cmd_args[header_type].len);
1726 mark_source_rp_bad(ms);
1727 goto out;
1728 }
1729
1730 /* We know we've got a valid header by this point */
1731 res = qemu_get_buffer(rp, buf, header_len);
1732 if (res != header_len) {
1733 error_report("RP: Failed reading data for message 0x%04x"
1734 " read %d expected %d",
1735 header_type, res, header_len);
1736 mark_source_rp_bad(ms);
1737 goto out;
1738 }
1739
1740 /* OK, we have the message and the data */
1741 switch (header_type) {
1742 case MIG_RP_MSG_SHUT:
4d885131 1743 sibling_error = ldl_be_p(buf);
70b20477
DDAG		 1744 trace_source_return_path_thread_shut(sibling_error);
1745 if (sibling_error) {
1746 error_report("RP: Sibling indicated error %d", sibling_error);
1747 mark_source_rp_bad(ms);
1748 }
1749 /*
1750 * We'll let the main thread deal with closing the RP
1751 * we could do a shutdown(2) on it, but we're the only user
1752 * anyway, so there's nothing gained.
1753 */
1754 goto out;
1755
1756 case MIG_RP_MSG_PONG:
4d885131 1757 tmp32 = ldl_be_p(buf);
70b20477
DDAG		 1758 trace_source_return_path_thread_pong(tmp32);
1759 break;
1760
1e2d90eb 1761 case MIG_RP_MSG_REQ_PAGES:
4d885131
PM		 1762 start = ldq_be_p(buf);
1763 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 1764 migrate_handle_rp_req_pages(ms, NULL, start, len);
1765 break;
1766
1767 case MIG_RP_MSG_REQ_PAGES_ID:
1768 expected_len = 12 + 1; /* header + termination */
1769
1770 if (header_len >= expected_len) {
4d885131
PM		 1771 start = ldq_be_p(buf);
1772 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 1773 /* Now we expect an idstr */
1774 tmp32 = buf[12]; /* Length of the following idstr */
1775 buf[13 + tmp32] = '\0';
1776 expected_len += tmp32;
1777 }
1778 if (header_len != expected_len) {
1779 error_report("RP: Req_Page_id with length %d expecting %zd",
1780 header_len, expected_len);
1781 mark_source_rp_bad(ms);
1782 goto out;
1783 }
1784 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1785 break;
1786
70b20477
DDAG		 1787 default:
1788 break;
1789 }
1790 }
5df5416e 1791 if (qemu_file_get_error(rp)) {
70b20477
DDAG		 1792 trace_source_return_path_thread_bad_end();
1793 mark_source_rp_bad(ms);
1794 }
1795
1796 trace_source_return_path_thread_end();
1797out:
1798 ms->rp_state.from_dst_file = NULL;
1799 qemu_fclose(rp);
1800 return NULL;
1801}
1802
70b20477
DDAG		 1803static int open_return_path_on_source(MigrationState *ms)
1804{
1805
89a02a9f 1806 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
70b20477
DDAG		 1807 if (!ms->rp_state.from_dst_file) {
1808 return -1;
1809 }
1810
1811 trace_open_return_path_on_source();
1812 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1813 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1814
1815 trace_open_return_path_on_source_continue();
1816
1817 return 0;
1818}
1819
70b20477
DDAG		 1820/* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1821static int await_return_path_close_on_source(MigrationState *ms)
1822{
1823 /*
1824 * If this is a normal exit then the destination will send a SHUT and the
1825 * rp_thread will exit, however if there's an error we need to cause
1826 * it to exit.
1827 */
89a02a9f 1828 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
70b20477
DDAG		 1829 /*
1830 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1831 * waiting for the destination.
1832 */
1833 qemu_file_shutdown(ms->rp_state.from_dst_file);
1834 mark_source_rp_bad(ms);
1835 }
1836 trace_await_return_path_close_on_source_joining();
1837 qemu_thread_join(&ms->rp_state.rp_thread);
1838 trace_await_return_path_close_on_source_close();
1839 return ms->rp_state.error;
1840}
1841
1d34e4bf
DDAG		 1842/*
1843 * Switch from normal iteration to postcopy
1844 * Returns non-0 on error
1845 */
1846static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1847{
1848 int ret;
61b67d47
DB		 1849 QIOChannelBuffer *bioc;
1850 QEMUFile *fb;
1d34e4bf 1851 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
ef8d6488 1852 bool restart_block = false;
0331c8ca
DDAG		 1853 int cur_state = MIGRATION_STATUS_ACTIVE;
1854 if (!migrate_pause_before_switchover()) {
1855 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1856 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1857 }
1d34e4bf
DDAG		 1858
1859 trace_postcopy_start();
1860 qemu_mutex_lock_iothread();
1861 trace_postcopy_start_set_run();
1862
1863 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1864 *old_vm_running = runstate_is_running();
1865 global_state_store();
1866 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
76b1c7fe
KW		 1867 if (ret < 0) {
1868 goto fail;
1869 }
1d34e4bf 1870
0331c8ca
DDAG		 1871 ret = migration_maybe_pause(ms, &cur_state,
1872 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1873 if (ret < 0) {
1874 goto fail;
1875 }
1876
76b1c7fe 1877 ret = bdrv_inactivate_all();
1d34e4bf
DDAG		 1878 if (ret < 0) {
1879 goto fail;
1880 }
ef8d6488 1881 restart_block = true;
1d34e4bf 1882
1c0d249d
DDAG		 1883 /*
1884 * Cause any non-postcopiable, but iterative devices to
1885 * send out their final data.
1886 */
a1fbe750 1887 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1c0d249d 1888
1d34e4bf
DDAG		 1889 /*
1890 * in Finish migrate and with the io-lock held everything should
1891 * be quiet, but we've potentially still got dirty pages and we
1892 * need to tell the destination to throw any pages it's already received
1893 * that are dirty
1894 */
58110f0a
VSO		 1895 if (migrate_postcopy_ram()) {
1896 if (ram_postcopy_send_discard_bitmap(ms)) {
1897 error_report("postcopy send discard bitmap failed");
1898 goto fail;
1899 }
1d34e4bf
DDAG		 1900 }
1901
1902 /*
1903 * send rest of state - note things that are doing postcopy
1904 * will notice we're in POSTCOPY_ACTIVE and not actually
1905 * wrap their state up here
1906 */
89a02a9f 1907 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
58110f0a
VSO		 1908 if (migrate_postcopy_ram()) {
1909 /* Ping just for debugging, helps line traces up */
1910 qemu_savevm_send_ping(ms->to_dst_file, 2);
1911 }
1d34e4bf
DDAG		 1912
1913 /*
1914 * While loading the device state we may trigger page transfer
1915 * requests and the fd must be free to process those, and thus
1916 * the destination must read the whole device state off the fd before
1917 * it starts processing it. Unfortunately the ad-hoc migration format
1918 * doesn't allow the destination to know the size to read without fully
1919 * parsing it through each devices load-state code (especially the open
1920 * coded devices that use get/put).
1921 * So we wrap the device state up in a package with a length at the start;
1922 * to do this we use a qemu_buf to hold the whole of the device state.
1923 */
61b67d47 1924 bioc = qio_channel_buffer_new(4096);
6f01f136 1925 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
61b67d47
DB		 1926 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1927 object_unref(OBJECT(bioc));
1d34e4bf 1928
c76201ab
DDAG		 1929 /*
1930 * Make sure the receiver can get incoming pages before we send the rest
1931 * of the state
1932 */
1933 qemu_savevm_send_postcopy_listen(fb);
1934
a1fbe750 1935 qemu_savevm_state_complete_precopy(fb, false, false);
58110f0a
VSO		 1936 if (migrate_postcopy_ram()) {
1937 qemu_savevm_send_ping(fb, 3);
1938 }
1d34e4bf
DDAG		 1939
1940 qemu_savevm_send_postcopy_run(fb);
1941
1942 /* <><> end of stuff going into the package */
1d34e4bf 1943
ef8d6488
DDAG		 1944 /* Last point of recovery; as soon as we send the package the destination
1945 * can open devices and potentially start running.
1946 * Lets just check again we've not got any errors.
1947 */
1948 ret = qemu_file_get_error(ms->to_dst_file);
1949 if (ret) {
1950 error_report("postcopy_start: Migration stream errored (pre package)");
1951 goto fail_closefb;
1952 }
1953
1954 restart_block = false;
1955
1d34e4bf 1956 /* Now send that blob */
61b67d47 1957 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1d34e4bf
DDAG		 1958 goto fail_closefb;
1959 }
1960 qemu_fclose(fb);
b82fc321
DDAG		 1961
1962 /* Send a notify to give a chance for anything that needs to happen
1963 * at the transition to postcopy and after the device state; in particular
1964 * spice needs to trigger a transition now
1965 */
1966 ms->postcopy_after_devices = true;
1967 notifier_list_notify(&migration_state_notifiers, ms);
1968
1d34e4bf
DDAG		 1969 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1970
1971 qemu_mutex_unlock_iothread();
1972
58110f0a
VSO		 1973 if (migrate_postcopy_ram()) {
1974 /*
1975 * Although this ping is just for debug, it could potentially be
1976 * used for getting a better measurement of downtime at the source.
1977 */
1978 qemu_savevm_send_ping(ms->to_dst_file, 4);
1979 }
1d34e4bf 1980
ced1c616
PB		 1981 if (migrate_release_ram()) {
1982 ram_postcopy_migrated_memory_release(ms);
1983 }
1984
89a02a9f 1985 ret = qemu_file_get_error(ms->to_dst_file);
1d34e4bf
DDAG		 1986 if (ret) {
1987 error_report("postcopy_start: Migration stream errored");
48781e5b 1988 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf
DDAG		 1989 MIGRATION_STATUS_FAILED);
1990 }
1991
1992 return ret;
1993
1994fail_closefb:
1995 qemu_fclose(fb);
1996fail:
48781e5b 1997 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf 1998 MIGRATION_STATUS_FAILED);
ef8d6488
DDAG		 1999 if (restart_block) {
2000 /* A failure happened early enough that we know the destination hasn't
2001 * accessed block devices, so we're safe to recover.
2002 */
2003 Error *local_err = NULL;
2004
2005 bdrv_invalidate_cache_all(&local_err);
2006 if (local_err) {
2007 error_report_err(local_err);
2008 }
2009 }
1d34e4bf
DDAG		 2010 qemu_mutex_unlock_iothread();
2011 return -1;
2012}
2013
e91d8951
DDAG		 2014/**
2015 * migration_maybe_pause: Pause if required to by
2016 * migrate_pause_before_switchover called with the iothread locked
2017 * Returns: 0 on success
2018 */
0331c8ca
DDAG		 2019static int migration_maybe_pause(MigrationState *s,
2020 int *current_active_state,
2021 int new_state)
e91d8951
DDAG		 2022{
2023 if (!migrate_pause_before_switchover()) {
2024 return 0;
2025 }
2026
2027 /* Since leaving this state is not atomic with posting the semaphore
2028 * it's possible that someone could have issued multiple migrate_continue
2029 * and the semaphore is incorrectly positive at this point;
2030 * the docs say it's undefined to reinit a semaphore that's already
2031 * init'd, so use timedwait to eat up any existing posts.
2032 */
2033 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2034 /* This block intentionally left blank */
2035 }
2036
2037 qemu_mutex_unlock_iothread();
2038 migrate_set_state(&s->state, *current_active_state,
2039 MIGRATION_STATUS_PRE_SWITCHOVER);
2040 qemu_sem_wait(&s->pause_sem);
2041 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
0331c8ca
DDAG		 2042 new_state);
2043 *current_active_state = new_state;
e91d8951
DDAG		 2044 qemu_mutex_lock_iothread();
2045
0331c8ca 2046 return s->state == new_state ? 0 : -EINVAL;
e91d8951
DDAG		 2047}
2048
09f6c85e
DDAG		 2049/**
2050 * migration_completion: Used by migration_thread when there's not much left.
2051 * The caller 'breaks' the loop when this returns.
2052 *
2053 * @s: Current migration state
36f48567 2054 * @current_active_state: The migration state we expect to be in
09f6c85e
DDAG		 2055 * @*old_vm_running: Pointer to old_vm_running flag
2056 * @*start_time: Pointer to time to update
2057 */
36f48567
DDAG		 2058static void migration_completion(MigrationState *s, int current_active_state,
2059 bool *old_vm_running,
09f6c85e
DDAG		 2060 int64_t *start_time)
2061{
2062 int ret;
2063
b10ac0c4
DDAG		 2064 if (s->state == MIGRATION_STATUS_ACTIVE) {
2065 qemu_mutex_lock_iothread();
2066 *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2067 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
2068 *old_vm_running = runstate_is_running();
2069 ret = global_state_store();
2070
2071 if (!ret) {
a1fbe750 2072 bool inactivate = !migrate_colo_enabled();
b10ac0c4 2073 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
e91d8951 2074 if (ret >= 0) {
0331c8ca
DDAG		 2075 ret = migration_maybe_pause(s, &current_active_state,
2076 MIGRATION_STATUS_DEVICE);
e91d8951 2077 }
f07fa4cb
KW		 2078 if (ret >= 0) {
2079 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
a1fbe750
FZ		 2080 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2081 inactivate);
f07fa4cb 2082 }
a1fbe750
FZ		 2083 if (inactivate && ret >= 0) {
2084 s->block_inactive = true;
b10ac0c4
DDAG		 2085 }
2086 }
2087 qemu_mutex_unlock_iothread();
09f6c85e 2088
b10ac0c4
DDAG		 2089 if (ret < 0) {
2090 goto fail;
09f6c85e 2091 }
b10ac0c4
DDAG		 2092 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2093 trace_migration_completion_postcopy_end();
2094
89a02a9f 2095 qemu_savevm_state_complete_postcopy(s->to_dst_file);
b10ac0c4 2096 trace_migration_completion_postcopy_end_after_complete();
09f6c85e 2097 }
09f6c85e 2098
b10ac0c4
DDAG		 2099 /*
2100 * If rp was opened we must clean up the thread before
2101 * cleaning everything else up (since if there are no failures
2102 * it will wait for the destination to send it's status in
2103 * a SHUT command).
b10ac0c4 2104 */
0425dc97 2105 if (s->rp_state.from_dst_file) {
b10ac0c4 2106 int rp_error;
0425dc97 2107 trace_migration_return_path_end_before();
b10ac0c4 2108 rp_error = await_return_path_close_on_source(s);
0425dc97 2109 trace_migration_return_path_end_after(rp_error);
b10ac0c4 2110 if (rp_error) {
fe904ea8 2111 goto fail_invalidate;
b10ac0c4 2112 }
09f6c85e
DDAG		 2113 }
2114
89a02a9f 2115 if (qemu_file_get_error(s->to_dst_file)) {
09f6c85e 2116 trace_migration_completion_file_err();
fe904ea8 2117 goto fail_invalidate;
09f6c85e
DDAG		 2118 }
2119
0b827d5e
HZ		 2120 if (!migrate_colo_enabled()) {
2121 migrate_set_state(&s->state, current_active_state,
2122 MIGRATION_STATUS_COMPLETED);
2123 }
2124
09f6c85e
DDAG		 2125 return;
2126
fe904ea8
GK		 2127fail_invalidate:
2128 /* If not doing postcopy, vm_start() will be called: let's regain
2129 * control on images.
2130 */
2131 if (s->state == MIGRATION_STATUS_ACTIVE) {
2132 Error *local_err = NULL;
2133
1d2acc31 2134 qemu_mutex_lock_iothread();
fe904ea8
GK		 2135 bdrv_invalidate_cache_all(&local_err);
2136 if (local_err) {
2137 error_report_err(local_err);
1d2acc31
HZ		 2138 } else {
2139 s->block_inactive = false;
fe904ea8 2140 }
1d2acc31 2141 qemu_mutex_unlock_iothread();
fe904ea8
GK		 2142 }
2143
09f6c85e 2144fail:
48781e5b
HZ		 2145 migrate_set_state(&s->state, current_active_state,
2146 MIGRATION_STATUS_FAILED);
09f6c85e
DDAG		 2147}
2148
35a6ed4f
HZ		 2149bool migrate_colo_enabled(void)
2150{
2151 MigrationState *s = migrate_get_current();
2152 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2153}
2154
70b20477
DDAG		 2155/*
2156 * Master migration thread on the source VM.
2157 * It drives the migration and pumps the data down the outgoing channel.
2158 */
5f496a1b 2159static void *migration_thread(void *opaque)
0d82d0e8 2160{
9848a404 2161 MigrationState *s = opaque;
1d34e4bf 2162 /* Used by the bandwidth calcs, updated later */
bc72ad67
AB		 2163 int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2164 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
be7172e2 2165 int64_t initial_bytes = 0;
faec066a
PX		 2166 /*
2167 * The final stage happens when the remaining data is smaller than
2168 * this threshold; it's calculated from the requested downtime and
2169 * measured bandwidth
2170 */
2171 int64_t threshold_size = 0;
a3fa1d78 2172 int64_t start_time = initial_time;
94f5a437 2173 int64_t end_time;
a3fa1d78 2174 bool old_vm_running = false;
1d34e4bf
DDAG		 2175 bool entered_postcopy = false;
2176 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
2177 enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
0b827d5e 2178 bool enable_colo = migrate_colo_enabled();
76f5933a 2179
ab28bd23
PB		 2180 rcu_register_thread();
2181
89a02a9f 2182 qemu_savevm_state_header(s->to_dst_file);
1d34e4bf 2183
62a02658
PX		 2184 /*
2185 * If we opened the return path, we need to make sure dst has it
2186 * opened as well.
2187 */
2188 if (s->rp_state.from_dst_file) {
1d34e4bf 2189 /* Now tell the dest that it should open its end so it can reply */
89a02a9f 2190 qemu_savevm_send_open_return_path(s->to_dst_file);
1d34e4bf
DDAG		 2191
2192 /* And do a ping that will make stuff easier to debug */
89a02a9f 2193 qemu_savevm_send_ping(s->to_dst_file, 1);
0425dc97 2194 }
1d34e4bf 2195
58110f0a 2196 if (migrate_postcopy()) {
1d34e4bf
DDAG		 2197 /*
2198 * Tell the destination that we *might* want to do postcopy later;
2199 * if the other end can't do postcopy it should fail now, nice and
2200 * early.
2201 */
89a02a9f 2202 qemu_savevm_send_postcopy_advise(s->to_dst_file);
1d34e4bf
DDAG		 2203 }
2204
9907e842 2205 qemu_savevm_state_setup(s->to_dst_file);
0d82d0e8 2206
bc72ad67 2207 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
48781e5b
HZ		 2208 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2209 MIGRATION_STATUS_ACTIVE);
29ae8a41 2210
9ec055ae
DDAG		 2211 trace_migration_thread_setup_complete();
2212
2213 while (s->state == MIGRATION_STATUS_ACTIVE ||
2214 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
a3e879cd 2215 int64_t current_time;
c369f40d 2216 uint64_t pending_size;
0d82d0e8 2217
89a02a9f 2218 if (!qemu_file_rate_limit(s->to_dst_file)) {
c31b098f
DDAG		 2219 uint64_t pend_post, pend_nonpost;
2220
faec066a
PX		 2221 qemu_savevm_state_pending(s->to_dst_file, threshold_size,
2222 &pend_nonpost, &pend_post);
c31b098f 2223 pending_size = pend_nonpost + pend_post;
faec066a 2224 trace_migrate_pending(pending_size, threshold_size,
c31b098f 2225 pend_post, pend_nonpost);
faec066a 2226 if (pending_size && pending_size >= threshold_size) {
1d34e4bf
DDAG		 2227 /* Still a significant amount to transfer */
2228
58110f0a 2229 if (migrate_postcopy() &&
1d34e4bf 2230 s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
faec066a 2231 pend_nonpost <= threshold_size &&
1d34e4bf
DDAG		 2232 atomic_read(&s->start_postcopy)) {
2233
2234 if (!postcopy_start(s, &old_vm_running)) {
2235 current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
2236 entered_postcopy = true;
2237 }
2238
2239 continue;
2240 }
2241 /* Just another iteration step */
89a02a9f 2242 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
c369f40d 2243 } else {
09f6c85e 2244 trace_migration_thread_low_pending(pending_size);
1d34e4bf 2245 migration_completion(s, current_active_state,
36f48567 2246 &old_vm_running, &start_time);
09f6c85e 2247 break;
c369f40d
JQ		 2248 }
2249 }
f4410a5d 2250
89a02a9f 2251 if (qemu_file_get_error(s->to_dst_file)) {
48781e5b
HZ		 2252 migrate_set_state(&s->state, current_active_state,
2253 MIGRATION_STATUS_FAILED);
1d34e4bf 2254 trace_migration_thread_file_err();
fd45ee2c
PB		 2255 break;
2256 }
bc72ad67 2257 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
0d82d0e8 2258 if (current_time >= initial_time + BUFFER_DELAY) {
89a02a9f
HZ		 2259 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
2260 initial_bytes;
77417f10 2261 uint64_t time_spent = current_time - initial_time;
a694ee34 2262 double bandwidth = (double)transferred_bytes / time_spent;
faec066a 2263 threshold_size = bandwidth * s->parameters.downtime_limit;
0d82d0e8 2264
5b648de0
WY		 2265 s->mbps = (((double) transferred_bytes * 8.0) /
2266 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
7e114f8c 2267
9013dca5 2268 trace_migrate_transferred(transferred_bytes, time_spent,
faec066a 2269 bandwidth, threshold_size);
90f8ae72
JQ		 2270 /* if we haven't sent anything, we don't want to recalculate
2271 10000 is a small enough number for our purposes */
9360447d
JQ		 2272 if (ram_counters.dirty_pages_rate && transferred_bytes > 10000) {
2273 s->expected_downtime = ram_counters.dirty_pages_rate *
20afaed9 2274 qemu_target_page_size() / bandwidth;
90f8ae72 2275 }
0d82d0e8 2276
89a02a9f 2277 qemu_file_reset_rate_limit(s->to_dst_file);
0d82d0e8 2278 initial_time = current_time;
89a02a9f 2279 initial_bytes = qemu_ftell(s->to_dst_file);
0d82d0e8 2280 }
89a02a9f 2281 if (qemu_file_rate_limit(s->to_dst_file)) {
0d82d0e8
JQ		 2282 /* usleep expects microseconds */
2283 g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
2284 }
a3fa1d78
PB		 2285 }
2286
1d34e4bf 2287 trace_migration_thread_after_loop();
070afca2
JH		 2288 /* If we enabled cpu throttling for auto-converge, turn it off. */
2289 cpu_throttle_stop();
94f5a437 2290 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
070afca2 2291
f4410a5d 2292 qemu_mutex_lock_iothread();
0b827d5e
HZ		 2293 /*
2294 * The resource has been allocated by migration will be reused in COLO
2295 * process, so don't release them.
2296 */
2297 if (!enable_colo) {
2298 qemu_savevm_state_cleanup();
2299 }
31194731 2300 if (s->state == MIGRATION_STATUS_COMPLETED) {
89a02a9f 2301 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
a3fa1d78 2302 s->total_time = end_time - s->total_time;
1d34e4bf
DDAG		 2303 if (!entered_postcopy) {
2304 s->downtime = end_time - start_time;
2305 }
d6ed7312
PL		 2306 if (s->total_time) {
2307 s->mbps = (((double) transferred_bytes * 8.0) /
2308 ((double) s->total_time)) / 1000;
2309 }
a3fa1d78
PB		 2310 runstate_set(RUN_STATE_POSTMIGRATE);
2311 } else {
0b827d5e
HZ		 2312 if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
2313 migrate_start_colo_process(s);
2314 qemu_savevm_state_cleanup();
2315 /*
2316 * Fixme: we will run VM in COLO no matter its old running state.
2317 * After exited COLO, we will keep running.
2318 */
2319 old_vm_running = true;
2320 }
1d34e4bf 2321 if (old_vm_running && !entered_postcopy) {
a3fa1d78 2322 vm_start();
42da5550
DDAG		 2323 } else {
2324 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2325 runstate_set(RUN_STATE_POSTMIGRATE);
2326 }
dba433c0 2327 }
0d82d0e8 2328 }
bb1fadc4 2329 qemu_bh_schedule(s->cleanup_bh);
dba433c0 2330 qemu_mutex_unlock_iothread();
f4410a5d 2331
ab28bd23 2332 rcu_unregister_thread();
0d82d0e8
JQ		 2333 return NULL;
2334}
2335
9848a404 2336void migrate_fd_connect(MigrationState *s)
0d82d0e8 2337{
2ff30257 2338 s->expected_downtime = s->parameters.downtime_limit;
bb1fadc4 2339 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
0d82d0e8 2340
9e4d2b98 2341 qemu_file_set_blocking(s->to_dst_file, true);
89a02a9f 2342 qemu_file_set_rate_limit(s->to_dst_file,
2ff30257 2343 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
442773ce 2344
9287ac27
SH		 2345 /* Notify before starting migration thread */
2346 notifier_list_notify(&migration_state_notifiers, s);
2347
1d34e4bf 2348 /*
c788ada8
PX		 2349 * Open the return path. For postcopy, it is used exclusively. For
2350 * precopy, only if user specified "return-path" capability would
2351 * QEMU uses the return path.
1d34e4bf 2352 */
c788ada8 2353 if (migrate_postcopy_ram() || migrate_use_return_path()) {
1d34e4bf
DDAG		 2354 if (open_return_path_on_source(s)) {
2355 error_report("Unable to open return-path for postcopy");
48781e5b 2356 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1d34e4bf
DDAG		 2357 MIGRATION_STATUS_FAILED);
2358 migrate_fd_cleanup(s);
2359 return;
2360 }
2361 }
2362
f986c3d2
JQ		 2363 if (multifd_save_setup() != 0) {
2364 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2365 MIGRATION_STATUS_FAILED);
2366 migrate_fd_cleanup(s);
2367 return;
2368 }
009fad7f 2369 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
bb1fadc4 2370 QEMU_THREAD_JOINABLE);
1d34e4bf 2371 s->migration_thread_running = true;
0d82d0e8 2372}
093e3c42 2373
9d18af93
PX		 2374void migration_global_dump(Monitor *mon)
2375{
2376 MigrationState *ms = migrate_get_current();
2377
2378 monitor_printf(mon, "globals: store-global-state=%d, only_migratable=%d, "
2379 "send-configuration=%d, send-section-footer=%d\n",
2380 ms->store_global_state, ms->only_migratable,
2381 ms->send_configuration, ms->send_section_footer);
2382}
2383
20814758
PX		 2384#define DEFINE_PROP_MIG_CAP(name, x) \
2385 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2386
5272298c
PX		 2387static Property migration_properties[] = {
2388 DEFINE_PROP_BOOL("store-global-state", MigrationState,
2389 store_global_state, true),
3df663e5 2390 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
71dd4c1a
PX		 2391 DEFINE_PROP_BOOL("send-configuration", MigrationState,
2392 send_configuration, true),
15c38503
PX		 2393 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2394 send_section_footer, true),
89632faf
PX		 2395
2396 /* Migration parameters */
2397 DEFINE_PROP_INT64("x-compress-level", MigrationState,
2398 parameters.compress_level,
2399 DEFAULT_MIGRATE_COMPRESS_LEVEL),
2400 DEFINE_PROP_INT64("x-compress-threads", MigrationState,
2401 parameters.compress_threads,
2402 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
2403 DEFINE_PROP_INT64("x-decompress-threads", MigrationState,
2404 parameters.decompress_threads,
2405 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
2406 DEFINE_PROP_INT64("x-cpu-throttle-initial", MigrationState,
2407 parameters.cpu_throttle_initial,
2408 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
2409 DEFINE_PROP_INT64("x-cpu-throttle-increment", MigrationState,
2410 parameters.cpu_throttle_increment,
2411 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
2412 DEFINE_PROP_INT64("x-max-bandwidth", MigrationState,
2413 parameters.max_bandwidth, MAX_THROTTLE),
2414 DEFINE_PROP_INT64("x-downtime-limit", MigrationState,
2415 parameters.downtime_limit,
2416 DEFAULT_MIGRATE_SET_DOWNTIME),
2417 DEFINE_PROP_INT64("x-checkpoint-delay", MigrationState,
2418 parameters.x_checkpoint_delay,
2419 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
4075fb1c
JQ		 2420 DEFINE_PROP_INT64("x-multifd-channels", MigrationState,
2421 parameters.x_multifd_channels,
2422 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
0fb86605
JQ		 2423 DEFINE_PROP_INT64("x-multifd-page-count", MigrationState,
2424 parameters.x_multifd_page_count,
2425 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
73af8dd8
JQ		 2426 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
2427 parameters.xbzrle_cache_size,
2428 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
20814758
PX		 2429
2430 /* Migration capabilities */
2431 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2432 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2433 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2434 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2435 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2436 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2437 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2438 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2439 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2440 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2441 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
30126bbf 2442 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
20814758 2443
5272298c
PX		 2444 DEFINE_PROP_END_OF_LIST(),
2445};
2446
e5cb7e76
PX		 2447static void migration_class_init(ObjectClass *klass, void *data)
2448{
2449 DeviceClass *dc = DEVICE_CLASS(klass);
2450
2451 dc->user_creatable = false;
5272298c 2452 dc->props = migration_properties;
e5cb7e76
PX		 2453}
2454
b91bf5e4
MAL		 2455static void migration_instance_finalize(Object *obj)
2456{
2457 MigrationState *ms = MIGRATION_OBJ(obj);
2458 MigrationParameters *params = &ms->parameters;
2459
87db1a7d 2460 qemu_mutex_destroy(&ms->error_mutex);
b91bf5e4
MAL		 2461 g_free(params->tls_hostname);
2462 g_free(params->tls_creds);
e91d8951 2463 qemu_sem_destroy(&ms->pause_sem);
b91bf5e4
MAL		 2464}
2465
e5cb7e76
PX		 2466static void migration_instance_init(Object *obj)
2467{
2468 MigrationState *ms = MIGRATION_OBJ(obj);
8b0b29dc 2469 MigrationParameters *params = &ms->parameters;
e5cb7e76
PX		 2470
2471 ms->state = MIGRATION_STATUS_NONE;
e5cb7e76 2472 ms->mbps = -1;
e91d8951 2473 qemu_sem_init(&ms->pause_sem, 0);
87db1a7d 2474 qemu_mutex_init(&ms->error_mutex);
8b0b29dc
PX		 2475
2476 params->tls_hostname = g_strdup("");
2477 params->tls_creds = g_strdup("");
2478
2479 /* Set has_* up only for parameter checks */
2480 params->has_compress_level = true;
2481 params->has_compress_threads = true;
2482 params->has_decompress_threads = true;
2483 params->has_cpu_throttle_initial = true;
2484 params->has_cpu_throttle_increment = true;
2485 params->has_max_bandwidth = true;
2486 params->has_downtime_limit = true;
2487 params->has_x_checkpoint_delay = true;
2488 params->has_block_incremental = true;
4075fb1c 2489 params->has_x_multifd_channels = true;
0fb86605 2490 params->has_x_multifd_page_count = true;
73af8dd8 2491 params->has_xbzrle_cache_size = true;
8b0b29dc
PX		 2492}
2493
2494/*
2495 * Return true if check pass, false otherwise. Error will be put
2496 * inside errp if provided.
2497 */
2498static bool migration_object_check(MigrationState *ms, Error **errp)
2499{
6b19a7d9
PX		 2500 MigrationCapabilityStatusList *head = NULL;
2501 /* Assuming all off */
2502 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2503 int i;
2504
8b0b29dc
PX		 2505 if (!migrate_params_check(&ms->parameters, errp)) {
2506 return false;
2507 }
2508
6b19a7d9
PX		 2509 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2510 if (ms->enabled_capabilities[i]) {
2511 head = migrate_cap_add(head, i, true);
2512 }
2513 }
2514
2515 ret = migrate_caps_check(cap_list, head, errp);
2516
2517 /* It works with head == NULL */
2518 qapi_free_MigrationCapabilityStatusList(head);
2519
2520 return ret;
e5cb7e76
PX		 2521}
2522
2523static const TypeInfo migration_type = {
2524 .name = TYPE_MIGRATION,
01f6e14c 2525 /*
c8d3ff38
PX		 2526 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2527 * not created using qdev_create(), it is not attached to the qdev
2528 * device tree, and it is never realized.
2529 *
2530 * TODO: Make this TYPE_OBJECT once QOM provides something like
2531 * TYPE_DEVICE's "-global" properties.
01f6e14c 2532 */
e5cb7e76
PX		 2533 .parent = TYPE_DEVICE,
2534 .class_init = migration_class_init,
2535 .class_size = sizeof(MigrationClass),
2536 .instance_size = sizeof(MigrationState),
2537 .instance_init = migration_instance_init,
b91bf5e4 2538 .instance_finalize = migration_instance_finalize,
e5cb7e76
PX		 2539};
2540
2541static void register_migration_types(void)
2542{
2543 type_register_static(&migration_type);
2544}
2545
2546type_init(register_migration_types);
QEMU mirror