]> 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
migration: cleanup stats update into function
[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
1f895604
VSO		 135void migration_object_finalize(void)
136{
137 object_unref(OBJECT(current_migration));
138}
139
bca7856a 140/* For outgoing */
859bc756 141MigrationState *migrate_get_current(void)
17549e84 142{
e5cb7e76
PX		 143 /* This can only be called after the object created. */
144 assert(current_migration);
145 return current_migration;
17549e84
JQ		 146}
147
bca7856a
DDAG		 148MigrationIncomingState *migration_incoming_get_current(void)
149{
b4b076da
JQ		 150 static bool once;
151 static MigrationIncomingState mis_current;
bca7856a 152
b4b076da
JQ		 153 if (!once) {
154 mis_current.state = MIGRATION_STATUS_NONE;
155 memset(&mis_current, 0, sizeof(MigrationIncomingState));
b4b076da
JQ		 156 qemu_mutex_init(&mis_current.rp_mutex);
157 qemu_event_init(&mis_current.main_thread_load_event, false);
158 once = true;
159 }
160 return &mis_current;
bca7856a
DDAG		 161}
162
163void migration_incoming_state_destroy(void)
164{
b4b076da
JQ		 165 struct MigrationIncomingState *mis = migration_incoming_get_current();
166
3482655b 167 if (mis->to_src_file) {
660819b1
PX		 168 /* Tell source that we are done */
169 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
3482655b
PX		 170 qemu_fclose(mis->to_src_file);
171 mis->to_src_file = NULL;
172 }
173
660819b1
PX		 174 if (mis->from_src_file) {
175 qemu_fclose(mis->from_src_file);
176 mis->from_src_file = NULL;
177 }
178
5089e186 179 qemu_event_reset(&mis->main_thread_load_event);
bca7856a
DDAG		 180}
181
b05dc723
JQ		 182static void migrate_generate_event(int new_state)
183{
184 if (migrate_use_events()) {
185 qapi_event_send_migration(new_state, &error_abort);
b05dc723
JQ		 186 }
187}
188
adde220a
DDAG		 189/*
190 * Called on -incoming with a defer: uri.
191 * The migration can be started later after any parameters have been
192 * changed.
193 */
194static void deferred_incoming_migration(Error **errp)
195{
196 if (deferred_incoming) {
197 error_setg(errp, "Incoming migration already deferred");
198 }
199 deferred_incoming = true;
200}
201
da6f1790
JQ		 202/*
203 * Send a message on the return channel back to the source
204 * of the migration.
205 */
206static void migrate_send_rp_message(MigrationIncomingState *mis,
207 enum mig_rp_message_type message_type,
208 uint16_t len, void *data)
209{
210 trace_migrate_send_rp_message((int)message_type, len);
211 qemu_mutex_lock(&mis->rp_mutex);
212 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
213 qemu_put_be16(mis->to_src_file, len);
214 qemu_put_buffer(mis->to_src_file, data, len);
215 qemu_fflush(mis->to_src_file);
216 qemu_mutex_unlock(&mis->rp_mutex);
217}
218
1e2d90eb
DDAG		 219/* Request a range of pages from the source VM at the given
220 * start address.
221 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
222 * as the last request (a name must have been given previously)
223 * Start: Address offset within the RB
224 * Len: Length in bytes required - must be a multiple of pagesize
225 */
226void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
227 ram_addr_t start, size_t len)
228{
cb8d4c8f 229 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
1e2d90eb
DDAG		 230 size_t msglen = 12; /* start + len */
231
232 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
233 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
234
235 if (rbname) {
236 int rbname_len = strlen(rbname);
237 assert(rbname_len < 256);
238
239 bufc[msglen++] = rbname_len;
240 memcpy(bufc + msglen, rbname, rbname_len);
241 msglen += rbname_len;
242 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
243 } else {
244 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
245 }
246}
247
43eaae28 248void qemu_start_incoming_migration(const char *uri, Error **errp)
5bb7910a 249{
34c9dd8e
AL		 250 const char *p;
251
7cf1fe6d 252 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
adde220a
DDAG		 253 if (!strcmp(uri, "defer")) {
254 deferred_incoming_migration(errp);
255 } else if (strstart(uri, "tcp:", &p)) {
43eaae28 256 tcp_start_incoming_migration(p, errp);
2da776db 257#ifdef CONFIG_RDMA
adde220a 258 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 259 rdma_start_incoming_migration(p, errp);
260#endif
adde220a 261 } else if (strstart(uri, "exec:", &p)) {
43eaae28 262 exec_start_incoming_migration(p, errp);
adde220a 263 } else if (strstart(uri, "unix:", &p)) {
43eaae28 264 unix_start_incoming_migration(p, errp);
adde220a 265 } else if (strstart(uri, "fd:", &p)) {
43eaae28 266 fd_start_incoming_migration(p, errp);
adde220a 267 } else {
312fd5f2 268 error_setg(errp, "unknown migration protocol: %s", uri);
8ca5e801 269 }
5bb7910a
AL		 270}
271
0aa6aefc
DL		 272static void process_incoming_migration_bh(void *opaque)
273{
274 Error *local_err = NULL;
275 MigrationIncomingState *mis = opaque;
276
ace21a58
KW		 277 /* Make sure all file formats flush their mutable metadata.
278 * If we get an error here, just don't restart the VM yet. */
0aa6aefc 279 bdrv_invalidate_cache_all(&local_err);
d35ff5e6 280 if (local_err) {
ace21a58 281 error_report_err(local_err);
d35ff5e6
KW		 282 local_err = NULL;
283 autostart = false;
284 }
285
0aa6aefc
DL		 286 /*
287 * This must happen after all error conditions are dealt with and
288 * we're sure the VM is going to be running on this host.
289 */
290 qemu_announce_self();
291
f986c3d2
JQ		 292 if (multifd_load_cleanup(&local_err) != 0) {
293 error_report_err(local_err);
294 autostart = false;
295 }
0aa6aefc
DL		 296 /* If global state section was not received or we are in running
297 state, we need to obey autostart. Any other state is set with
298 runstate_set. */
299
300 if (!global_state_received() ||
301 global_state_get_runstate() == RUN_STATE_RUNNING) {
302 if (autostart) {
303 vm_start();
304 } else {
305 runstate_set(RUN_STATE_PAUSED);
306 }
307 } else {
308 runstate_set(global_state_get_runstate());
309 }
0aa6aefc
DL		 310 /*
311 * This must happen after any state changes since as soon as an external
312 * observer sees this event they might start to prod at the VM assuming
313 * it's ready to use.
314 */
315 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
316 MIGRATION_STATUS_COMPLETED);
317 qemu_bh_delete(mis->bh);
318 migration_incoming_state_destroy();
319}
320
82a4da79 321static void process_incoming_migration_co(void *opaque)
511c0231 322{
b4b076da 323 MigrationIncomingState *mis = migration_incoming_get_current();
e9bef235 324 PostcopyState ps;
1c12e1f5
PB		 325 int ret;
326
4f0fae7f 327 assert(mis->from_src_file);
67f11b5c 328 mis->largest_page_size = qemu_ram_pagesize_largest();
093e3c42 329 postcopy_state_set(POSTCOPY_INCOMING_NONE);
93d7af6f
HZ		 330 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
331 MIGRATION_STATUS_ACTIVE);
4f0fae7f 332 ret = qemu_loadvm_state(mis->from_src_file);
bca7856a 333
e9bef235
DDAG		 334 ps = postcopy_state_get();
335 trace_process_incoming_migration_co_end(ret, ps);
336 if (ps != POSTCOPY_INCOMING_NONE) {
337 if (ps == POSTCOPY_INCOMING_ADVISE) {
338 /*
339 * Where a migration had postcopy enabled (and thus went to advise)
340 * but managed to complete within the precopy period, we can use
341 * the normal exit.
342 */
343 postcopy_ram_incoming_cleanup(mis);
344 } else if (ret >= 0) {
345 /*
346 * Postcopy was started, cleanup should happen at the end of the
347 * postcopy thread.
348 */
349 trace_process_incoming_migration_co_postcopy_end_main();
350 return;
351 }
352 /* Else if something went wrong then just fall out of the normal exit */
353 }
354
25d0c16f
HZ		 355 /* we get COLO info, and know if we are in COLO mode */
356 if (!ret && migration_incoming_enable_colo()) {
357 mis->migration_incoming_co = qemu_coroutine_self();
358 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
359 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
360 mis->have_colo_incoming_thread = true;
361 qemu_coroutine_yield();
362
363 /* Wait checkpoint incoming thread exit before free resource */
364 qemu_thread_join(&mis->colo_incoming_thread);
365 }
366
1c12e1f5 367 if (ret < 0) {
f986c3d2
JQ		 368 Error *local_err = NULL;
369
93d7af6f
HZ		 370 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
371 MIGRATION_STATUS_FAILED);
db80face 372 error_report("load of migration failed: %s", strerror(-ret));
3a0f2cea 373 qemu_fclose(mis->from_src_file);
f986c3d2
JQ		 374 if (multifd_load_cleanup(&local_err) != 0) {
375 error_report_err(local_err);
376 }
4aead692 377 exit(EXIT_FAILURE);
511c0231 378 }
0aa6aefc
DL		 379 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
380 qemu_bh_schedule(mis->bh);
511c0231
JQ		 381}
382
e595a01a 383static void migration_incoming_setup(QEMUFile *f)
82a4da79 384{
4f0fae7f 385 MigrationIncomingState *mis = migration_incoming_get_current();
82a4da79 386
f986c3d2
JQ		 387 if (multifd_load_setup() != 0) {
388 /* We haven't been able to create multifd threads
389 nothing better to do */
390 exit(EXIT_FAILURE);
391 }
392
4f0fae7f
JQ		 393 if (!mis->from_src_file) {
394 mis->from_src_file = f;
395 }
06ad5135 396 qemu_file_set_blocking(f, false);
e595a01a
JQ		 397}
398
399static void migration_incoming_process(void)
400{
401 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
0b8b8753 402 qemu_coroutine_enter(co);
82a4da79
PB		 403}
404
e595a01a
JQ		 405void migration_fd_process_incoming(QEMUFile *f)
406{
407 migration_incoming_setup(f);
408 migration_incoming_process();
409}
410
4f0fae7f
JQ		 411void migration_ioc_process_incoming(QIOChannel *ioc)
412{
413 MigrationIncomingState *mis = migration_incoming_get_current();
414
415 if (!mis->from_src_file) {
416 QEMUFile *f = qemu_fopen_channel_input(ioc);
417 migration_fd_process_incoming(f);
418 }
419 /* We still only have a single channel. Nothing to do here yet */
420}
421
428d8908
JQ		 422/**
423 * @migration_has_all_channels: We have received all channels that we need
424 *
425 * Returns true when we have got connections to all the channels that
426 * we need for migration.
427 */
428bool migration_has_all_channels(void)
429{
430 return true;
431}
432
6decec93
DDAG		 433/*
434 * Send a 'SHUT' message on the return channel with the given value
435 * to indicate that we've finished with the RP. Non-0 value indicates
436 * error.
437 */
438void migrate_send_rp_shut(MigrationIncomingState *mis,
439 uint32_t value)
440{
441 uint32_t buf;
442
443 buf = cpu_to_be32(value);
444 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
445}
446
447/*
448 * Send a 'PONG' message on the return channel with the given value
449 * (normally in response to a 'PING')
450 */
451void migrate_send_rp_pong(MigrationIncomingState *mis,
452 uint32_t value)
453{
454 uint32_t buf;
455
456 buf = cpu_to_be32(value);
457 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
458}
459
bbf6da32
OW		 460MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
461{
462 MigrationCapabilityStatusList *head = NULL;
463 MigrationCapabilityStatusList *caps;
464 MigrationState *s = migrate_get_current();
465 int i;
466
387eedeb 467 caps = NULL; /* silence compiler warning */
7fb1cf16 468 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
ed1701c6
DDAG		 469#ifndef CONFIG_LIVE_BLOCK_MIGRATION
470 if (i == MIGRATION_CAPABILITY_BLOCK) {
471 continue;
472 }
473#endif
bbf6da32
OW		 474 if (head == NULL) {
475 head = g_malloc0(sizeof(*caps));
476 caps = head;
477 } else {
478 caps->next = g_malloc0(sizeof(*caps));
479 caps = caps->next;
480 }
481 caps->value =
482 g_malloc(sizeof(*caps->value));
483 caps->value->capability = i;
484 caps->value->state = s->enabled_capabilities[i];
485 }
486
487 return head;
488}
489
85de8323
LL		 490MigrationParameters *qmp_query_migrate_parameters(Error **errp)
491{
492 MigrationParameters *params;
493 MigrationState *s = migrate_get_current();
494
e87fae4c 495 /* TODO use QAPI_CLONE() instead of duplicating it inline */
85de8323 496 params = g_malloc0(sizeof(*params));
de63ab61 497 params->has_compress_level = true;
2594f56d 498 params->compress_level = s->parameters.compress_level;
de63ab61 499 params->has_compress_threads = true;
2594f56d 500 params->compress_threads = s->parameters.compress_threads;
de63ab61 501 params->has_decompress_threads = true;
2594f56d 502 params->decompress_threads = s->parameters.decompress_threads;
de63ab61 503 params->has_cpu_throttle_initial = true;
2594f56d 504 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
de63ab61 505 params->has_cpu_throttle_increment = true;
2594f56d 506 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
8cc99dcd 507 params->has_tls_creds = true;
69ef1f36 508 params->tls_creds = g_strdup(s->parameters.tls_creds);
8cc99dcd 509 params->has_tls_hostname = true;
69ef1f36 510 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
2ff30257
AA		 511 params->has_max_bandwidth = true;
512 params->max_bandwidth = s->parameters.max_bandwidth;
513 params->has_downtime_limit = true;
514 params->downtime_limit = s->parameters.downtime_limit;
fe39a4d4 515 params->has_x_checkpoint_delay = true;
68b53591 516 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
2833c59b
JQ		 517 params->has_block_incremental = true;
518 params->block_incremental = s->parameters.block_incremental;
4075fb1c
JQ		 519 params->has_x_multifd_channels = true;
520 params->x_multifd_channels = s->parameters.x_multifd_channels;
0fb86605
JQ		 521 params->has_x_multifd_page_count = true;
522 params->x_multifd_page_count = s->parameters.x_multifd_page_count;
73af8dd8
JQ		 523 params->has_xbzrle_cache_size = true;
524 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
85de8323
LL		 525
526 return params;
527}
528
f6844b99
DDAG		 529/*
530 * Return true if we're already in the middle of a migration
531 * (i.e. any of the active or setup states)
532 */
533static bool migration_is_setup_or_active(int state)
534{
535 switch (state) {
536 case MIGRATION_STATUS_ACTIVE:
9ec055ae 537 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
f6844b99 538 case MIGRATION_STATUS_SETUP:
31e06077
DDAG		 539 case MIGRATION_STATUS_PRE_SWITCHOVER:
540 case MIGRATION_STATUS_DEVICE:
f6844b99
DDAG		 541 return true;
542
543 default:
544 return false;
545
546 }
547}
548
a22463a5
DDAG		 549static void populate_ram_info(MigrationInfo *info, MigrationState *s)
550{
551 info->has_ram = true;
552 info->ram = g_malloc0(sizeof(*info->ram));
9360447d 553 info->ram->transferred = ram_counters.transferred;
a22463a5 554 info->ram->total = ram_bytes_total();
9360447d 555 info->ram->duplicate = ram_counters.duplicate;
bedf53c1
JQ		 556 /* legacy value. It is not used anymore */
557 info->ram->skipped = 0;
9360447d
JQ		 558 info->ram->normal = ram_counters.normal;
559 info->ram->normal_bytes = ram_counters.normal *
20afaed9 560 qemu_target_page_size();
a22463a5 561 info->ram->mbps = s->mbps;
9360447d
JQ		 562 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
563 info->ram->postcopy_requests = ram_counters.postcopy_requests;
030ce1f8 564 info->ram->page_size = qemu_target_page_size();
a22463a5 565
114f5aee
JQ		 566 if (migrate_use_xbzrle()) {
567 info->has_xbzrle_cache = true;
568 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
569 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
9360447d
JQ		 570 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
571 info->xbzrle_cache->pages = xbzrle_counters.pages;
572 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
573 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
574 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
114f5aee
JQ		 575 }
576
338182c8
JQ		 577 if (cpu_throttle_active()) {
578 info->has_cpu_throttle_percentage = true;
579 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
580 }
581
a22463a5
DDAG		 582 if (s->state != MIGRATION_STATUS_COMPLETED) {
583 info->ram->remaining = ram_bytes_remaining();
9360447d 584 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
a22463a5
DDAG		 585 }
586}
587
930ac04c
JQ		 588static void populate_disk_info(MigrationInfo *info)
589{
590 if (blk_mig_active()) {
591 info->has_disk = true;
592 info->disk = g_malloc0(sizeof(*info->disk));
593 info->disk->transferred = blk_mig_bytes_transferred();
594 info->disk->remaining = blk_mig_bytes_remaining();
595 info->disk->total = blk_mig_bytes_total();
596 }
597}
598
ca6011c2 599static void fill_source_migration_info(MigrationInfo *info)
5bb7910a 600{
17549e84
JQ		 601 MigrationState *s = migrate_get_current();
602
603 switch (s->state) {
31194731 604 case MIGRATION_STATUS_NONE:
17549e84 605 /* no migration has happened ever */
ca6011c2
AP		 606 /* do not overwrite destination migration status */
607 return;
17549e84 608 break;
31194731 609 case MIGRATION_STATUS_SETUP:
29ae8a41 610 info->has_status = true;
ed4fbd10 611 info->has_total_time = false;
29ae8a41 612 break;
31194731
HZ		 613 case MIGRATION_STATUS_ACTIVE:
614 case MIGRATION_STATUS_CANCELLING:
9ec055ae 615 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
31e06077
DDAG		 616 case MIGRATION_STATUS_PRE_SWITCHOVER:
617 case MIGRATION_STATUS_DEVICE:
c8f9f4f4 618 /* TODO add some postcopy stats */
9ec055ae
DDAG		 619 info->has_status = true;
620 info->has_total_time = true;
621 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
4af246a3 622 - s->start_time;
9ec055ae
DDAG		 623 info->has_expected_downtime = true;
624 info->expected_downtime = s->expected_downtime;
625 info->has_setup_time = true;
626 info->setup_time = s->setup_time;
627
a22463a5 628 populate_ram_info(info, s);
930ac04c 629 populate_disk_info(info);
17549e84 630 break;
0b827d5e
HZ		 631 case MIGRATION_STATUS_COLO:
632 info->has_status = true;
633 /* TODO: display COLO specific information (checkpoint info etc.) */
634 break;
31194731 635 case MIGRATION_STATUS_COMPLETED:
791e7c82 636 info->has_status = true;
00c14997 637 info->has_total_time = true;
7aa939af 638 info->total_time = s->total_time;
9c5a9fcf
JQ		 639 info->has_downtime = true;
640 info->downtime = s->downtime;
ed4fbd10
MH		 641 info->has_setup_time = true;
642 info->setup_time = s->setup_time;
d5f8a570 643
a22463a5 644 populate_ram_info(info, s);
17549e84 645 break;
31194731 646 case MIGRATION_STATUS_FAILED:
791e7c82 647 info->has_status = true;
d59ce6f3
DB		 648 if (s->error) {
649 info->has_error_desc = true;
650 info->error_desc = g_strdup(error_get_pretty(s->error));
651 }
17549e84 652 break;
31194731 653 case MIGRATION_STATUS_CANCELLED:
791e7c82 654 info->has_status = true;
17549e84 655 break;
5bb7910a 656 }
cde63fbe 657 info->status = s->state;
5bb7910a
AL		 658}
659
4a84214e
PX		 660/**
661 * @migration_caps_check - check capability validity
662 *
663 * @cap_list: old capability list, array of bool
664 * @params: new capabilities to be applied soon
665 * @errp: set *errp if the check failed, with reason
666 *
667 * Returns true if check passed, otherwise false.
668 */
669static bool migrate_caps_check(bool *cap_list,
670 MigrationCapabilityStatusList *params,
671 Error **errp)
00458433 672{
00458433 673 MigrationCapabilityStatusList *cap;
4a84214e 674 bool old_postcopy_cap;
d7651f15 675 MigrationIncomingState *mis = migration_incoming_get_current();
00458433 676
4a84214e 677 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
00458433
OW		 678
679 for (cap = params; cap; cap = cap->next) {
4a84214e
PX		 680 cap_list[cap->value->capability] = cap->value->state;
681 }
682
ed1701c6 683#ifndef CONFIG_LIVE_BLOCK_MIGRATION
4a84214e
PX		 684 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
685 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
686 "block migration");
687 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
688 return false;
00458433 689 }
4a84214e 690#endif
53dd370c 691
4a84214e
PX		 692 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
693 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
53dd370c
DDAG		 694 /* The decompression threads asynchronously write into RAM
695 * rather than use the atomic copies needed to avoid
696 * userfaulting. It should be possible to fix the decompression
697 * threads for compatibility in future.
698 */
4a84214e
PX		 699 error_setg(errp, "Postcopy is not currently compatible "
700 "with compression");
701 return false;
53dd370c 702 }
4a84214e 703
096631bd
DDAG		 704 /* This check is reasonably expensive, so only when it's being
705 * set the first time, also it's only the destination that needs
706 * special support.
707 */
708 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
d7651f15 709 !postcopy_ram_supported_by_host(mis)) {
096631bd
DDAG		 710 /* postcopy_ram_supported_by_host will have emitted a more
711 * detailed message
712 */
4a84214e
PX		 713 error_setg(errp, "Postcopy is not supported");
714 return false;
096631bd 715 }
53dd370c 716 }
4a84214e
PX		 717
718 return true;
719}
720
ca6011c2
AP		 721static void fill_destination_migration_info(MigrationInfo *info)
722{
723 MigrationIncomingState *mis = migration_incoming_get_current();
724
725 switch (mis->state) {
726 case MIGRATION_STATUS_NONE:
727 return;
728 break;
729 case MIGRATION_STATUS_SETUP:
730 case MIGRATION_STATUS_CANCELLING:
731 case MIGRATION_STATUS_CANCELLED:
732 case MIGRATION_STATUS_ACTIVE:
733 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
734 case MIGRATION_STATUS_FAILED:
735 case MIGRATION_STATUS_COLO:
736 info->has_status = true;
737 break;
738 case MIGRATION_STATUS_COMPLETED:
739 info->has_status = true;
740 fill_destination_postcopy_migration_info(info);
741 break;
742 }
743 info->status = mis->state;
744}
745
746MigrationInfo *qmp_query_migrate(Error **errp)
747{
748 MigrationInfo *info = g_malloc0(sizeof(*info));
749
750 fill_destination_migration_info(info);
751 fill_source_migration_info(info);
752
753 return info;
754}
755
4a84214e
PX		 756void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
757 Error **errp)
758{
759 MigrationState *s = migrate_get_current();
760 MigrationCapabilityStatusList *cap;
761
762 if (migration_is_setup_or_active(s->state)) {
763 error_setg(errp, QERR_MIGRATION_ACTIVE);
764 return;
765 }
766
767 if (!migrate_caps_check(s->enabled_capabilities, params, errp)) {
768 return;
769 }
770
771 for (cap = params; cap; cap = cap->next) {
772 s->enabled_capabilities[cap->value->capability] = cap->value->state;
773 }
00458433
OW		 774}
775
16d063bc
PX		 776/*
777 * Check whether the parameters are valid. Error will be put into errp
778 * (if provided). Return true if valid, otherwise false.
779 */
780static bool migrate_params_check(MigrationParameters *params, Error **errp)
85de8323 781{
7f375e04 782 if (params->has_compress_level &&
741d4086 783 (params->compress_level > 9)) {
c6bd8c70
MA		 784 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
785 "is invalid, it should be in the range of 0 to 9");
16d063bc 786 return false;
85de8323 787 }
16d063bc 788
741d4086 789 if (params->has_compress_threads && (params->compress_threads < 1)) {
c6bd8c70
MA		 790 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
791 "compress_threads",
792 "is invalid, it should be in the range of 1 to 255");
16d063bc 793 return false;
85de8323 794 }
16d063bc 795
741d4086 796 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
c6bd8c70
MA		 797 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
798 "decompress_threads",
799 "is invalid, it should be in the range of 1 to 255");
16d063bc 800 return false;
85de8323 801 }
16d063bc 802
7f375e04
EB		 803 if (params->has_cpu_throttle_initial &&
804 (params->cpu_throttle_initial < 1 ||
805 params->cpu_throttle_initial > 99)) {
1626fee3 806 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 807 "cpu_throttle_initial",
1626fee3 808 "an integer in the range of 1 to 99");
16d063bc 809 return false;
1626fee3 810 }
16d063bc 811
7f375e04
EB		 812 if (params->has_cpu_throttle_increment &&
813 (params->cpu_throttle_increment < 1 ||
814 params->cpu_throttle_increment > 99)) {
1626fee3 815 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 816 "cpu_throttle_increment",
1626fee3 817 "an integer in the range of 1 to 99");
16d063bc 818 return false;
1626fee3 819 }
16d063bc 820
741d4086 821 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
2ff30257
AA		 822 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
823 " range of 0 to %zu bytes/second", SIZE_MAX);
16d063bc 824 return false;
2ff30257 825 }
16d063bc 826
2ff30257 827 if (params->has_downtime_limit &&
741d4086 828 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
87c9cc1c
DHB		 829 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
830 "the range of 0 to %d milliseconds",
831 MAX_MIGRATE_DOWNTIME);
16d063bc 832 return false;
2ff30257 833 }
16d063bc 834
741d4086
JQ		 835 /* x_checkpoint_delay is now always positive */
836
837 if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) {
4075fb1c
JQ		 838 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
839 "multifd_channels",
840 "is invalid, it should be in the range of 1 to 255");
841 return false;
842 }
0fb86605 843 if (params->has_x_multifd_page_count &&
741d4086
JQ		 844 (params->x_multifd_page_count < 1 ||
845 params->x_multifd_page_count > 10000)) {
0fb86605
JQ		 846 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
847 "multifd_page_count",
848 "is invalid, it should be in the range of 1 to 10000");
849 return false;
850 }
16d063bc 851
73af8dd8
JQ		 852 if (params->has_xbzrle_cache_size &&
853 (params->xbzrle_cache_size < qemu_target_page_size() ||
854 !is_power_of_2(params->xbzrle_cache_size))) {
855 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
856 "xbzrle_cache_size",
857 "is invalid, it should be bigger than target page size"
858 " and a power of two");
859 return false;
860 }
861
16d063bc
PX		 862 return true;
863}
864
1bda8b3c
MA		 865static void migrate_params_test_apply(MigrateSetParameters *params,
866 MigrationParameters *dest)
867{
868 *dest = migrate_get_current()->parameters;
869
870 /* TODO use QAPI_CLONE() instead of duplicating it inline */
871
872 if (params->has_compress_level) {
873 dest->compress_level = params->compress_level;
874 }
875
876 if (params->has_compress_threads) {
877 dest->compress_threads = params->compress_threads;
878 }
879
880 if (params->has_decompress_threads) {
881 dest->decompress_threads = params->decompress_threads;
882 }
883
884 if (params->has_cpu_throttle_initial) {
885 dest->cpu_throttle_initial = params->cpu_throttle_initial;
886 }
887
888 if (params->has_cpu_throttle_increment) {
889 dest->cpu_throttle_increment = params->cpu_throttle_increment;
890 }
891
892 if (params->has_tls_creds) {
01fa5598
MA		 893 assert(params->tls_creds->type == QTYPE_QSTRING);
894 dest->tls_creds = g_strdup(params->tls_creds->u.s);
1bda8b3c
MA		 895 }
896
897 if (params->has_tls_hostname) {
01fa5598
MA		 898 assert(params->tls_hostname->type == QTYPE_QSTRING);
899 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1bda8b3c
MA		 900 }
901
902 if (params->has_max_bandwidth) {
903 dest->max_bandwidth = params->max_bandwidth;
904 }
905
906 if (params->has_downtime_limit) {
907 dest->downtime_limit = params->downtime_limit;
908 }
909
910 if (params->has_x_checkpoint_delay) {
911 dest->x_checkpoint_delay = params->x_checkpoint_delay;
912 }
913
914 if (params->has_block_incremental) {
915 dest->block_incremental = params->block_incremental;
916 }
5e7577a1
JQ		 917 if (params->has_x_multifd_channels) {
918 dest->x_multifd_channels = params->x_multifd_channels;
919 }
920 if (params->has_x_multifd_page_count) {
921 dest->x_multifd_page_count = params->x_multifd_page_count;
922 }
73af8dd8
JQ		 923 if (params->has_xbzrle_cache_size) {
924 dest->xbzrle_cache_size = params->xbzrle_cache_size;
925 }
1bda8b3c
MA		 926}
927
73af8dd8 928static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
16d063bc
PX		 929{
930 MigrationState *s = migrate_get_current();
931
e87fae4c
MA		 932 /* TODO use QAPI_CLONE() instead of duplicating it inline */
933
7f375e04
EB		 934 if (params->has_compress_level) {
935 s->parameters.compress_level = params->compress_level;
85de8323 936 }
476c72aa 937
7f375e04
EB		 938 if (params->has_compress_threads) {
939 s->parameters.compress_threads = params->compress_threads;
85de8323 940 }
476c72aa 941
7f375e04
EB		 942 if (params->has_decompress_threads) {
943 s->parameters.decompress_threads = params->decompress_threads;
85de8323 944 }
476c72aa 945
7f375e04
EB		 946 if (params->has_cpu_throttle_initial) {
947 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1626fee3 948 }
476c72aa 949
7f375e04
EB		 950 if (params->has_cpu_throttle_increment) {
951 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1626fee3 952 }
476c72aa 953
7f375e04 954 if (params->has_tls_creds) {
69ef1f36 955 g_free(s->parameters.tls_creds);
01fa5598
MA		 956 assert(params->tls_creds->type == QTYPE_QSTRING);
957 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
69ef1f36 958 }
476c72aa 959
7f375e04 960 if (params->has_tls_hostname) {
69ef1f36 961 g_free(s->parameters.tls_hostname);
01fa5598
MA		 962 assert(params->tls_hostname->type == QTYPE_QSTRING);
963 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
69ef1f36 964 }
476c72aa 965
2ff30257
AA		 966 if (params->has_max_bandwidth) {
967 s->parameters.max_bandwidth = params->max_bandwidth;
968 if (s->to_dst_file) {
969 qemu_file_set_rate_limit(s->to_dst_file,
970 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
971 }
972 }
476c72aa 973
2ff30257
AA		 974 if (params->has_downtime_limit) {
975 s->parameters.downtime_limit = params->downtime_limit;
976 }
68b53591
HZ		 977
978 if (params->has_x_checkpoint_delay) {
979 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
479125d5
HZ		 980 if (migration_in_colo_state()) {
981 colo_checkpoint_notify(s);
982 }
68b53591 983 }
476c72aa 984
2833c59b
JQ		 985 if (params->has_block_incremental) {
986 s->parameters.block_incremental = params->block_incremental;
987 }
4075fb1c
JQ		 988 if (params->has_x_multifd_channels) {
989 s->parameters.x_multifd_channels = params->x_multifd_channels;
990 }
0fb86605
JQ		 991 if (params->has_x_multifd_page_count) {
992 s->parameters.x_multifd_page_count = params->x_multifd_page_count;
993 }
73af8dd8
JQ		 994 if (params->has_xbzrle_cache_size) {
995 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
996 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
997 }
85de8323
LL		 998}
999
1bda8b3c 1000void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
476c72aa 1001{
1bda8b3c
MA		 1002 MigrationParameters tmp;
1003
01fa5598
MA		 1004 /* TODO Rewrite "" to null instead */
1005 if (params->has_tls_creds
1006 && params->tls_creds->type == QTYPE_QNULL) {
1007 QDECREF(params->tls_creds->u.n);
1008 params->tls_creds->type = QTYPE_QSTRING;
1009 params->tls_creds->u.s = strdup("");
1010 }
1011 /* TODO Rewrite "" to null instead */
1012 if (params->has_tls_hostname
1013 && params->tls_hostname->type == QTYPE_QNULL) {
1014 QDECREF(params->tls_hostname->u.n);
1015 params->tls_hostname->type = QTYPE_QSTRING;
1016 params->tls_hostname->u.s = strdup("");
1017 }
1018
1bda8b3c
MA		 1019 migrate_params_test_apply(params, &tmp);
1020
1021 if (!migrate_params_check(&tmp, errp)) {
476c72aa
PX		 1022 /* Invalid parameter */
1023 return;
1024 }
1025
73af8dd8 1026 migrate_params_apply(params, errp);
476c72aa
PX		 1027}
1028
2594f56d 1029
4886a1bc
DDAG		 1030void qmp_migrate_start_postcopy(Error **errp)
1031{
1032 MigrationState *s = migrate_get_current();
1033
1034 if (!migrate_postcopy_ram()) {
a54d340b 1035 error_setg(errp, "Enable postcopy with migrate_set_capability before"
4886a1bc
DDAG		 1036 " the start of migration");
1037 return;
1038 }
1039
1040 if (s->state == MIGRATION_STATUS_NONE) {
1041 error_setg(errp, "Postcopy must be started after migration has been"
1042 " started");
1043 return;
1044 }
1045 /*
1046 * we don't error if migration has finished since that would be racy
1047 * with issuing this command.
1048 */
1049 atomic_set(&s->start_postcopy, true);
1050}
1051
065e2813
AL		 1052/* shared migration helpers */
1053
48781e5b 1054void migrate_set_state(int *state, int old_state, int new_state)
51cf4c1a 1055{
a31fedee 1056 assert(new_state < MIGRATION_STATUS__MAX);
48781e5b 1057 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
a31fedee 1058 trace_migrate_set_state(MigrationStatus_str(new_state));
b05dc723 1059 migrate_generate_event(new_state);
51cf4c1a
Z		 1060 }
1061}
1062
4e4a3d3a
PX		 1063static MigrationCapabilityStatusList *migrate_cap_add(
1064 MigrationCapabilityStatusList *list,
1065 MigrationCapability index,
1066 bool state)
2833c59b
JQ		 1067{
1068 MigrationCapabilityStatusList *cap;
1069
1070 cap = g_new0(MigrationCapabilityStatusList, 1);
1071 cap->value = g_new0(MigrationCapabilityStatus, 1);
4e4a3d3a
PX		 1072 cap->value->capability = index;
1073 cap->value->state = state;
1074 cap->next = list;
1075
1076 return cap;
1077}
1078
1079void migrate_set_block_enabled(bool value, Error **errp)
1080{
1081 MigrationCapabilityStatusList *cap;
1082
1083 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
2833c59b
JQ		 1084 qmp_migrate_set_capabilities(cap, errp);
1085 qapi_free_MigrationCapabilityStatusList(cap);
1086}
1087
1088static void migrate_set_block_incremental(MigrationState *s, bool value)
1089{
1090 s->parameters.block_incremental = value;
1091}
1092
1093static void block_cleanup_parameters(MigrationState *s)
1094{
1095 if (s->must_remove_block_options) {
1096 /* setting to false can never fail */
1097 migrate_set_block_enabled(false, &error_abort);
1098 migrate_set_block_incremental(s, false);
1099 s->must_remove_block_options = false;
1100 }
1101}
1102
bb1fadc4 1103static void migrate_fd_cleanup(void *opaque)
065e2813 1104{
bb1fadc4
PB		 1105 MigrationState *s = opaque;
1106
1107 qemu_bh_delete(s->cleanup_bh);
1108 s->cleanup_bh = NULL;
1109
0ceccd85
PX		 1110 qemu_savevm_state_cleanup();
1111
89a02a9f 1112 if (s->to_dst_file) {
f986c3d2
JQ		 1113 Error *local_err = NULL;
1114
9013dca5 1115 trace_migrate_fd_cleanup();
404a7c05 1116 qemu_mutex_unlock_iothread();
1d34e4bf
DDAG		 1117 if (s->migration_thread_running) {
1118 qemu_thread_join(&s->thread);
1119 s->migration_thread_running = false;
1120 }
404a7c05
PB		 1121 qemu_mutex_lock_iothread();
1122
f986c3d2
JQ		 1123 if (multifd_save_cleanup(&local_err) != 0) {
1124 error_report_err(local_err);
1125 }
89a02a9f
HZ		 1126 qemu_fclose(s->to_dst_file);
1127 s->to_dst_file = NULL;
065e2813
AL		 1128 }
1129
9ec055ae
DDAG		 1130 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1131 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
7a2c1721 1132
94f5a437 1133 if (s->state == MIGRATION_STATUS_CANCELLING) {
48781e5b 1134 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
94f5a437 1135 MIGRATION_STATUS_CANCELLED);
7a2c1721 1136 }
a3fa1d78 1137
87db1a7d
JQ		 1138 if (s->error) {
1139 /* It is used on info migrate. We can't free it */
1140 error_report_err(error_copy(s->error));
1141 }
a3fa1d78 1142 notifier_list_notify(&migration_state_notifiers, s);
2833c59b 1143 block_cleanup_parameters(s);
065e2813
AL		 1144}
1145
87db1a7d
JQ		 1146void migrate_set_error(MigrationState *s, const Error *error)
1147{
1148 qemu_mutex_lock(&s->error_mutex);
1149 if (!s->error) {
1150 s->error = error_copy(error);
1151 }
1152 qemu_mutex_unlock(&s->error_mutex);
1153}
1154
d59ce6f3 1155void migrate_fd_error(MigrationState *s, const Error *error)
065e2813 1156{
25174055 1157 trace_migrate_fd_error(error_get_pretty(error));
89a02a9f 1158 assert(s->to_dst_file == NULL);
48781e5b
HZ		 1159 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1160 MIGRATION_STATUS_FAILED);
87db1a7d 1161 migrate_set_error(s, error);
bb1fadc4 1162 notifier_list_notify(&migration_state_notifiers, s);
2833c59b 1163 block_cleanup_parameters(s);
458cf28e
JQ		 1164}
1165
0edda1c4 1166static void migrate_fd_cancel(MigrationState *s)
065e2813 1167{
6f2b811a 1168 int old_state ;
89a02a9f 1169 QEMUFile *f = migrate_get_current()->to_dst_file;
9013dca5 1170 trace_migrate_fd_cancel();
065e2813 1171
70b20477
DDAG		 1172 if (s->rp_state.from_dst_file) {
1173 /* shutdown the rp socket, so causing the rp thread to shutdown */
1174 qemu_file_shutdown(s->rp_state.from_dst_file);
1175 }
1176
6f2b811a
Z		 1177 do {
1178 old_state = s->state;
f6844b99 1179 if (!migration_is_setup_or_active(old_state)) {
6f2b811a
Z		 1180 break;
1181 }
a7b36b48
DDAG		 1182 /* If the migration is paused, kick it out of the pause */
1183 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1184 qemu_sem_post(&s->pause_sem);
1185 }
48781e5b 1186 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
31194731 1187 } while (s->state != MIGRATION_STATUS_CANCELLING);
a26ba26e
DDAG		 1188
1189 /*
1190 * If we're unlucky the migration code might be stuck somewhere in a
1191 * send/write while the network has failed and is waiting to timeout;
1192 * if we've got shutdown(2) available then we can force it to quit.
1193 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1194 * called in a bh, so there is no race against this cancel.
1195 */
31194731 1196 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
a26ba26e
DDAG		 1197 qemu_file_shutdown(f);
1198 }
1d2acc31
HZ		 1199 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1200 Error *local_err = NULL;
1201
1202 bdrv_invalidate_cache_all(&local_err);
1203 if (local_err) {
1204 error_report_err(local_err);
1205 } else {
1206 s->block_inactive = false;
1207 }
1208 }
2833c59b 1209 block_cleanup_parameters(s);
065e2813
AL		 1210}
1211
99a0db9b
GH		 1212void add_migration_state_change_notifier(Notifier *notify)
1213{
1214 notifier_list_add(&migration_state_notifiers, notify);
1215}
1216
1217void remove_migration_state_change_notifier(Notifier *notify)
1218{
31552529 1219 notifier_remove(notify);
99a0db9b
GH		 1220}
1221
02edd2e7 1222bool migration_in_setup(MigrationState *s)
afe2df69 1223{
31194731 1224 return s->state == MIGRATION_STATUS_SETUP;
afe2df69
GH		 1225}
1226
7073693b 1227bool migration_has_finished(MigrationState *s)
99a0db9b 1228{
31194731 1229 return s->state == MIGRATION_STATUS_COMPLETED;
99a0db9b 1230}
0edda1c4 1231
afe2df69
GH		 1232bool migration_has_failed(MigrationState *s)
1233{
31194731
HZ		 1234 return (s->state == MIGRATION_STATUS_CANCELLED ||
1235 s->state == MIGRATION_STATUS_FAILED);
afe2df69
GH		 1236}
1237
5727309d 1238bool migration_in_postcopy(void)
9ec055ae 1239{
5727309d
JQ		 1240 MigrationState *s = migrate_get_current();
1241
9ec055ae
DDAG		 1242 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1243}
1244
b82fc321
DDAG		 1245bool migration_in_postcopy_after_devices(MigrationState *s)
1246{
5727309d 1247 return migration_in_postcopy() && s->postcopy_after_devices;
b82fc321
DDAG		 1248}
1249
fab35005 1250bool migration_is_idle(void)
fe44dc91 1251{
fab35005 1252 MigrationState *s = migrate_get_current();
fe44dc91
AA		 1253
1254 switch (s->state) {
1255 case MIGRATION_STATUS_NONE:
1256 case MIGRATION_STATUS_CANCELLED:
1257 case MIGRATION_STATUS_COMPLETED:
1258 case MIGRATION_STATUS_FAILED:
1259 return true;
1260 case MIGRATION_STATUS_SETUP:
1261 case MIGRATION_STATUS_CANCELLING:
1262 case MIGRATION_STATUS_ACTIVE:
1263 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1264 case MIGRATION_STATUS_COLO:
31e06077
DDAG		 1265 case MIGRATION_STATUS_PRE_SWITCHOVER:
1266 case MIGRATION_STATUS_DEVICE:
fe44dc91
AA		 1267 return false;
1268 case MIGRATION_STATUS__MAX:
1269 g_assert_not_reached();
1270 }
1271
1272 return false;
1273}
1274
a0762d9e 1275MigrationState *migrate_init(void)
0edda1c4 1276{
17549e84 1277 MigrationState *s = migrate_get_current();
bbf6da32 1278
389775d1
DDAG		 1279 /*
1280 * Reinitialise all migration state, except
1281 * parameters/capabilities that the user set, and
1282 * locks.
1283 */
1284 s->bytes_xfer = 0;
1285 s->xfer_limit = 0;
1286 s->cleanup_bh = 0;
89a02a9f 1287 s->to_dst_file = NULL;
389775d1 1288 s->state = MIGRATION_STATUS_NONE;
389775d1
DDAG		 1289 s->rp_state.from_dst_file = NULL;
1290 s->rp_state.error = false;
1291 s->mbps = 0.0;
1292 s->downtime = 0;
1293 s->expected_downtime = 0;
389775d1 1294 s->setup_time = 0;
389775d1 1295 s->start_postcopy = false;
b82fc321 1296 s->postcopy_after_devices = false;
389775d1 1297 s->migration_thread_running = false;
d59ce6f3
DB		 1298 error_free(s->error);
1299 s->error = NULL;
389775d1 1300
48781e5b 1301 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
0edda1c4 1302
4af246a3
PX		 1303 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1304 s->total_time = 0;
7287cbd4 1305 s->vm_was_running = false;
b15df1ae
PX		 1306 s->iteration_initial_bytes = 0;
1307 s->threshold_size = 0;
0edda1c4
JQ		 1308 return s;
1309}
cab30143 1310
fa2756b7
AL		 1311static GSList *migration_blockers;
1312
fe44dc91 1313int migrate_add_blocker(Error *reason, Error **errp)
fa2756b7 1314{
3df663e5 1315 if (migrate_get_current()->only_migratable) {
b67b8c3a
AA		 1316 error_propagate(errp, error_copy(reason));
1317 error_prepend(errp, "disallowing migration blocker "
1318 "(--only_migratable) for: ");
1319 return -EACCES;
1320 }
1321
fab35005 1322 if (migration_is_idle()) {
fe44dc91
AA		 1323 migration_blockers = g_slist_prepend(migration_blockers, reason);
1324 return 0;
1325 }
1326
1327 error_propagate(errp, error_copy(reason));
1328 error_prepend(errp, "disallowing migration blocker (migration in "
1329 "progress) for: ");
1330 return -EBUSY;
fa2756b7
AL		 1331}
1332
1333void migrate_del_blocker(Error *reason)
1334{
1335 migration_blockers = g_slist_remove(migration_blockers, reason);
1336}
1337
bf1ae1f4
DDAG		 1338void qmp_migrate_incoming(const char *uri, Error **errp)
1339{
1340 Error *local_err = NULL;
4debb5f5 1341 static bool once = true;
bf1ae1f4
DDAG		 1342
1343 if (!deferred_incoming) {
4debb5f5 1344 error_setg(errp, "For use with '-incoming defer'");
bf1ae1f4
DDAG		 1345 return;
1346 }
4debb5f5
DDAG		 1347 if (!once) {
1348 error_setg(errp, "The incoming migration has already been started");
1349 }
bf1ae1f4
DDAG		 1350
1351 qemu_start_incoming_migration(uri, &local_err);
1352
1353 if (local_err) {
1354 error_propagate(errp, local_err);
1355 return;
1356 }
1357
4debb5f5 1358 once = false;
bf1ae1f4
DDAG		 1359}
1360
24f3902b
GK		 1361bool migration_is_blocked(Error **errp)
1362{
1363 if (qemu_savevm_state_blocked(errp)) {
1364 return true;
1365 }
1366
1367 if (migration_blockers) {
250561e1 1368 error_propagate(errp, error_copy(migration_blockers->data));
24f3902b
GK		 1369 return true;
1370 }
1371
1372 return false;
1373}
1374
e1c37d0e
LC		 1375void qmp_migrate(const char *uri, bool has_blk, bool blk,
1376 bool has_inc, bool inc, bool has_detach, bool detach,
1377 Error **errp)
cab30143 1378{
be7059cd 1379 Error *local_err = NULL;
17549e84 1380 MigrationState *s = migrate_get_current();
cab30143 1381 const char *p;
cab30143 1382
f6844b99 1383 if (migration_is_setup_or_active(s->state) ||
0b827d5e
HZ		 1384 s->state == MIGRATION_STATUS_CANCELLING ||
1385 s->state == MIGRATION_STATUS_COLO) {
c6bd8c70 1386 error_setg(errp, QERR_MIGRATION_ACTIVE);
e1c37d0e 1387 return;
cab30143 1388 }
ca99993a
DDAG		 1389 if (runstate_check(RUN_STATE_INMIGRATE)) {
1390 error_setg(errp, "Guest is waiting for an incoming migration");
1391 return;
1392 }
1393
24f3902b 1394 if (migration_is_blocked(errp)) {
e1c37d0e 1395 return;
fa2756b7
AL		 1396 }
1397
2833c59b
JQ		 1398 if ((has_blk && blk) || (has_inc && inc)) {
1399 if (migrate_use_block() || migrate_use_block_incremental()) {
1400 error_setg(errp, "Command options are incompatible with "
1401 "current migration capabilities");
1402 return;
1403 }
1404 migrate_set_block_enabled(true, &local_err);
1405 if (local_err) {
1406 error_propagate(errp, local_err);
1407 return;
1408 }
1409 s->must_remove_block_options = true;
1410 }
1411
1412 if (has_inc && inc) {
1413 migrate_set_block_incremental(s, true);
1414 }
1415
a0762d9e 1416 s = migrate_init();
cab30143
JQ		 1417
1418 if (strstart(uri, "tcp:", &p)) {
f37afb5a 1419 tcp_start_outgoing_migration(s, p, &local_err);
2da776db 1420#ifdef CONFIG_RDMA
41310c68 1421 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 1422 rdma_start_outgoing_migration(s, p, &local_err);
1423#endif
cab30143 1424 } else if (strstart(uri, "exec:", &p)) {
f37afb5a 1425 exec_start_outgoing_migration(s, p, &local_err);
cab30143 1426 } else if (strstart(uri, "unix:", &p)) {
f37afb5a 1427 unix_start_outgoing_migration(s, p, &local_err);
cab30143 1428 } else if (strstart(uri, "fd:", &p)) {
f37afb5a 1429 fd_start_outgoing_migration(s, p, &local_err);
99a0db9b 1430 } else {
c6bd8c70
MA		 1431 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1432 "a valid migration protocol");
48781e5b
HZ		 1433 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1434 MIGRATION_STATUS_FAILED);
e1c37d0e 1435 return;
cab30143
JQ		 1436 }
1437
f37afb5a 1438 if (local_err) {
d59ce6f3 1439 migrate_fd_error(s, local_err);
f37afb5a 1440 error_propagate(errp, local_err);
e1c37d0e 1441 return;
1299c631 1442 }
cab30143
JQ		 1443}
1444
6cdedb07 1445void qmp_migrate_cancel(Error **errp)
cab30143 1446{
17549e84 1447 migrate_fd_cancel(migrate_get_current());
cab30143
JQ		 1448}
1449
89cfc02c
DDAG		 1450void qmp_migrate_continue(MigrationStatus state, Error **errp)
1451{
1452 MigrationState *s = migrate_get_current();
1453 if (s->state != state) {
1454 error_setg(errp, "Migration not in expected state: %s",
1455 MigrationStatus_str(s->state));
1456 return;
1457 }
1458 qemu_sem_post(&s->pause_sem);
1459}
1460
9e1ba4cc
OW		 1461void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1462{
73af8dd8
JQ		 1463 MigrateSetParameters p = {
1464 .has_xbzrle_cache_size = true,
1465 .xbzrle_cache_size = value,
1466 };
c91e681a 1467
73af8dd8 1468 qmp_migrate_set_parameters(&p, errp);
9e1ba4cc
OW		 1469}
1470
1471int64_t qmp_query_migrate_cache_size(Error **errp)
1472{
1473 return migrate_xbzrle_cache_size();
1474}
1475
3dc85383 1476void qmp_migrate_set_speed(int64_t value, Error **errp)
cab30143 1477{
1bda8b3c 1478 MigrateSetParameters p = {
2ff30257
AA		 1479 .has_max_bandwidth = true,
1480 .max_bandwidth = value,
1481 };
cab30143 1482
2ff30257 1483 qmp_migrate_set_parameters(&p, errp);
cab30143
JQ		 1484}
1485
4f0a993b 1486void qmp_migrate_set_downtime(double value, Error **errp)
cab30143 1487{
87c9cc1c
DHB		 1488 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1489 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1490 "the range of 0 to %d seconds",
1491 MAX_MIGRATE_DOWNTIME_SECONDS);
1492 return;
1493 }
1494
2ff30257
AA		 1495 value *= 1000; /* Convert to milliseconds */
1496 value = MAX(0, MIN(INT64_MAX, value));
1497
1bda8b3c 1498 MigrateSetParameters p = {
2ff30257
AA		 1499 .has_downtime_limit = true,
1500 .downtime_limit = value,
1501 };
1502
1503 qmp_migrate_set_parameters(&p, errp);
99a0db9b 1504}
17ad9b35 1505
53f09a10
PB		 1506bool migrate_release_ram(void)
1507{
1508 MigrationState *s;
1509
1510 s = migrate_get_current();
1511
1512 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1513}
1514
53dd370c
DDAG		 1515bool migrate_postcopy_ram(void)
1516{
1517 MigrationState *s;
1518
1519 s = migrate_get_current();
1520
32c3db5b 1521 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
53dd370c
DDAG		 1522}
1523
58110f0a
VSO		 1524bool migrate_postcopy(void)
1525{
1526 return migrate_postcopy_ram();
1527}
1528
bde1e2ec
CV		 1529bool migrate_auto_converge(void)
1530{
1531 MigrationState *s;
1532
1533 s = migrate_get_current();
1534
1535 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1536}
1537
323004a3
PL		 1538bool migrate_zero_blocks(void)
1539{
1540 MigrationState *s;
1541
1542 s = migrate_get_current();
1543
1544 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1545}
1546
31bf06a9
AP		 1547bool migrate_postcopy_blocktime(void)
1548{
1549 MigrationState *s;
1550
1551 s = migrate_get_current();
1552
1553 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
1554}
1555
8706d2d5
LL		 1556bool migrate_use_compression(void)
1557{
dde4e694
LL		 1558 MigrationState *s;
1559
1560 s = migrate_get_current();
1561
1562 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
8706d2d5
LL		 1563}
1564
1565int migrate_compress_level(void)
1566{
1567 MigrationState *s;
1568
1569 s = migrate_get_current();
1570
2594f56d 1571 return s->parameters.compress_level;
8706d2d5
LL		 1572}
1573
1574int migrate_compress_threads(void)
1575{
1576 MigrationState *s;
1577
1578 s = migrate_get_current();
1579
2594f56d 1580 return s->parameters.compress_threads;
8706d2d5
LL		 1581}
1582
3fcb38c2
LL		 1583int migrate_decompress_threads(void)
1584{
1585 MigrationState *s;
1586
1587 s = migrate_get_current();
1588
2594f56d 1589 return s->parameters.decompress_threads;
3fcb38c2
LL		 1590}
1591
b05dc723
JQ		 1592bool migrate_use_events(void)
1593{
1594 MigrationState *s;
1595
1596 s = migrate_get_current();
1597
1598 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1599}
1600
30126bbf
JQ		 1601bool migrate_use_multifd(void)
1602{
1603 MigrationState *s;
1604
1605 s = migrate_get_current();
1606
1607 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1608}
1609
93fbd031
DDAG		 1610bool migrate_pause_before_switchover(void)
1611{
1612 MigrationState *s;
1613
1614 s = migrate_get_current();
1615
1616 return s->enabled_capabilities[
1617 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
1618}
1619
4075fb1c
JQ		 1620int migrate_multifd_channels(void)
1621{
1622 MigrationState *s;
1623
1624 s = migrate_get_current();
1625
1626 return s->parameters.x_multifd_channels;
1627}
1628
0fb86605
JQ		 1629int migrate_multifd_page_count(void)
1630{
1631 MigrationState *s;
1632
1633 s = migrate_get_current();
1634
1635 return s->parameters.x_multifd_page_count;
1636}
1637
17ad9b35
OW		 1638int migrate_use_xbzrle(void)
1639{
1640 MigrationState *s;
1641
1642 s = migrate_get_current();
1643
1644 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1645}
1646
1647int64_t migrate_xbzrle_cache_size(void)
1648{
1649 MigrationState *s;
1650
1651 s = migrate_get_current();
1652
73af8dd8 1653 return s->parameters.xbzrle_cache_size;
17ad9b35 1654}
0d82d0e8 1655
2833c59b
JQ		 1656bool migrate_use_block(void)
1657{
1658 MigrationState *s;
1659
1660 s = migrate_get_current();
1661
1662 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1663}
1664
c788ada8
PX		 1665bool migrate_use_return_path(void)
1666{
1667 MigrationState *s;
1668
1669 s = migrate_get_current();
1670
1671 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1672}
1673
2833c59b
JQ		 1674bool migrate_use_block_incremental(void)
1675{
1676 MigrationState *s;
1677
1678 s = migrate_get_current();
1679
1680 return s->parameters.block_incremental;
1681}
1682
70b20477
DDAG		 1683/* migration thread support */
1684/*
1685 * Something bad happened to the RP stream, mark an error
1686 * The caller shall print or trace something to indicate why
1687 */
1688static void mark_source_rp_bad(MigrationState *s)
1689{
1690 s->rp_state.error = true;
1691}
1692
1693static struct rp_cmd_args {
1694 ssize_t len; /* -1 = variable */
1695 const char *name;
1696} rp_cmd_args[] = {
1697 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1698 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1699 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1e2d90eb
DDAG		 1700 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1701 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
70b20477
DDAG		 1702 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1703};
1704
1e2d90eb
DDAG		 1705/*
1706 * Process a request for pages received on the return path,
1707 * We're allowed to send more than requested (e.g. to round to our page size)
1708 * and we don't need to send pages that have already been sent.
1709 */
1710static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1711 ram_addr_t start, size_t len)
1712{
6c595cde
DDAG		 1713 long our_host_ps = getpagesize();
1714
1e2d90eb 1715 trace_migrate_handle_rp_req_pages(rbname, start, len);
6c595cde
DDAG		 1716
1717 /*
1718 * Since we currently insist on matching page sizes, just sanity check
1719 * we're being asked for whole host pages.
1720 */
1721 if (start & (our_host_ps-1) ||
1722 (len & (our_host_ps-1))) {
1723 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1724 " len: %zd", __func__, start, len);
1725 mark_source_rp_bad(ms);
1726 return;
1727 }
1728
96506894 1729 if (ram_save_queue_pages(rbname, start, len)) {
6c595cde
DDAG		 1730 mark_source_rp_bad(ms);
1731 }
1e2d90eb
DDAG		 1732}
1733
70b20477
DDAG		 1734/*
1735 * Handles messages sent on the return path towards the source VM
1736 *
1737 */
1738static void *source_return_path_thread(void *opaque)
1739{
1740 MigrationState *ms = opaque;
1741 QEMUFile *rp = ms->rp_state.from_dst_file;
1742 uint16_t header_len, header_type;
568b01ca 1743 uint8_t buf[512];
70b20477 1744 uint32_t tmp32, sibling_error;
1e2d90eb
DDAG		 1745 ram_addr_t start = 0; /* =0 to silence warning */
1746 size_t len = 0, expected_len;
70b20477
DDAG		 1747 int res;
1748
1749 trace_source_return_path_thread_entry();
1750 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1751 migration_is_setup_or_active(ms->state)) {
1752 trace_source_return_path_thread_loop_top();
1753 header_type = qemu_get_be16(rp);
1754 header_len = qemu_get_be16(rp);
1755
1756 if (header_type >= MIG_RP_MSG_MAX ||
1757 header_type == MIG_RP_MSG_INVALID) {
1758 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1759 header_type, header_len);
1760 mark_source_rp_bad(ms);
1761 goto out;
1762 }
1763
1764 if ((rp_cmd_args[header_type].len != -1 &&
1765 header_len != rp_cmd_args[header_type].len) ||
568b01ca 1766 header_len > sizeof(buf)) {
70b20477
DDAG		 1767 error_report("RP: Received '%s' message (0x%04x) with"
1768 "incorrect length %d expecting %zu",
1769 rp_cmd_args[header_type].name, header_type, header_len,
1770 (size_t)rp_cmd_args[header_type].len);
1771 mark_source_rp_bad(ms);
1772 goto out;
1773 }
1774
1775 /* We know we've got a valid header by this point */
1776 res = qemu_get_buffer(rp, buf, header_len);
1777 if (res != header_len) {
1778 error_report("RP: Failed reading data for message 0x%04x"
1779 " read %d expected %d",
1780 header_type, res, header_len);
1781 mark_source_rp_bad(ms);
1782 goto out;
1783 }
1784
1785 /* OK, we have the message and the data */
1786 switch (header_type) {
1787 case MIG_RP_MSG_SHUT:
4d885131 1788 sibling_error = ldl_be_p(buf);
70b20477
DDAG		 1789 trace_source_return_path_thread_shut(sibling_error);
1790 if (sibling_error) {
1791 error_report("RP: Sibling indicated error %d", sibling_error);
1792 mark_source_rp_bad(ms);
1793 }
1794 /*
1795 * We'll let the main thread deal with closing the RP
1796 * we could do a shutdown(2) on it, but we're the only user
1797 * anyway, so there's nothing gained.
1798 */
1799 goto out;
1800
1801 case MIG_RP_MSG_PONG:
4d885131 1802 tmp32 = ldl_be_p(buf);
70b20477
DDAG		 1803 trace_source_return_path_thread_pong(tmp32);
1804 break;
1805
1e2d90eb 1806 case MIG_RP_MSG_REQ_PAGES:
4d885131
PM		 1807 start = ldq_be_p(buf);
1808 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 1809 migrate_handle_rp_req_pages(ms, NULL, start, len);
1810 break;
1811
1812 case MIG_RP_MSG_REQ_PAGES_ID:
1813 expected_len = 12 + 1; /* header + termination */
1814
1815 if (header_len >= expected_len) {
4d885131
PM		 1816 start = ldq_be_p(buf);
1817 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 1818 /* Now we expect an idstr */
1819 tmp32 = buf[12]; /* Length of the following idstr */
1820 buf[13 + tmp32] = '\0';
1821 expected_len += tmp32;
1822 }
1823 if (header_len != expected_len) {
1824 error_report("RP: Req_Page_id with length %d expecting %zd",
1825 header_len, expected_len);
1826 mark_source_rp_bad(ms);
1827 goto out;
1828 }
1829 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1830 break;
1831
70b20477
DDAG		 1832 default:
1833 break;
1834 }
1835 }
5df5416e 1836 if (qemu_file_get_error(rp)) {
70b20477
DDAG		 1837 trace_source_return_path_thread_bad_end();
1838 mark_source_rp_bad(ms);
1839 }
1840
1841 trace_source_return_path_thread_end();
1842out:
1843 ms->rp_state.from_dst_file = NULL;
1844 qemu_fclose(rp);
1845 return NULL;
1846}
1847
70b20477
DDAG		 1848static int open_return_path_on_source(MigrationState *ms)
1849{
1850
89a02a9f 1851 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
70b20477
DDAG		 1852 if (!ms->rp_state.from_dst_file) {
1853 return -1;
1854 }
1855
1856 trace_open_return_path_on_source();
1857 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1858 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1859
1860 trace_open_return_path_on_source_continue();
1861
1862 return 0;
1863}
1864
70b20477
DDAG		 1865/* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1866static int await_return_path_close_on_source(MigrationState *ms)
1867{
1868 /*
1869 * If this is a normal exit then the destination will send a SHUT and the
1870 * rp_thread will exit, however if there's an error we need to cause
1871 * it to exit.
1872 */
89a02a9f 1873 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
70b20477
DDAG		 1874 /*
1875 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1876 * waiting for the destination.
1877 */
1878 qemu_file_shutdown(ms->rp_state.from_dst_file);
1879 mark_source_rp_bad(ms);
1880 }
1881 trace_await_return_path_close_on_source_joining();
1882 qemu_thread_join(&ms->rp_state.rp_thread);
1883 trace_await_return_path_close_on_source_close();
1884 return ms->rp_state.error;
1885}
1886
1d34e4bf
DDAG		 1887/*
1888 * Switch from normal iteration to postcopy
1889 * Returns non-0 on error
1890 */
7287cbd4 1891static int postcopy_start(MigrationState *ms)
1d34e4bf
DDAG		 1892{
1893 int ret;
61b67d47
DB		 1894 QIOChannelBuffer *bioc;
1895 QEMUFile *fb;
1d34e4bf 1896 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
ef8d6488 1897 bool restart_block = false;
0331c8ca
DDAG		 1898 int cur_state = MIGRATION_STATUS_ACTIVE;
1899 if (!migrate_pause_before_switchover()) {
1900 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1901 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1902 }
1d34e4bf
DDAG		 1903
1904 trace_postcopy_start();
1905 qemu_mutex_lock_iothread();
1906 trace_postcopy_start_set_run();
1907
1908 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1d34e4bf
DDAG		 1909 global_state_store();
1910 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
76b1c7fe
KW		 1911 if (ret < 0) {
1912 goto fail;
1913 }
1d34e4bf 1914
0331c8ca
DDAG		 1915 ret = migration_maybe_pause(ms, &cur_state,
1916 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1917 if (ret < 0) {
1918 goto fail;
1919 }
1920
76b1c7fe 1921 ret = bdrv_inactivate_all();
1d34e4bf
DDAG		 1922 if (ret < 0) {
1923 goto fail;
1924 }
ef8d6488 1925 restart_block = true;
1d34e4bf 1926
1c0d249d
DDAG		 1927 /*
1928 * Cause any non-postcopiable, but iterative devices to
1929 * send out their final data.
1930 */
a1fbe750 1931 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1c0d249d 1932
1d34e4bf
DDAG		 1933 /*
1934 * in Finish migrate and with the io-lock held everything should
1935 * be quiet, but we've potentially still got dirty pages and we
1936 * need to tell the destination to throw any pages it's already received
1937 * that are dirty
1938 */
58110f0a
VSO		 1939 if (migrate_postcopy_ram()) {
1940 if (ram_postcopy_send_discard_bitmap(ms)) {
1941 error_report("postcopy send discard bitmap failed");
1942 goto fail;
1943 }
1d34e4bf
DDAG		 1944 }
1945
1946 /*
1947 * send rest of state - note things that are doing postcopy
1948 * will notice we're in POSTCOPY_ACTIVE and not actually
1949 * wrap their state up here
1950 */
89a02a9f 1951 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
58110f0a
VSO		 1952 if (migrate_postcopy_ram()) {
1953 /* Ping just for debugging, helps line traces up */
1954 qemu_savevm_send_ping(ms->to_dst_file, 2);
1955 }
1d34e4bf
DDAG		 1956
1957 /*
1958 * While loading the device state we may trigger page transfer
1959 * requests and the fd must be free to process those, and thus
1960 * the destination must read the whole device state off the fd before
1961 * it starts processing it. Unfortunately the ad-hoc migration format
1962 * doesn't allow the destination to know the size to read without fully
1963 * parsing it through each devices load-state code (especially the open
1964 * coded devices that use get/put).
1965 * So we wrap the device state up in a package with a length at the start;
1966 * to do this we use a qemu_buf to hold the whole of the device state.
1967 */
61b67d47 1968 bioc = qio_channel_buffer_new(4096);
6f01f136 1969 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
61b67d47
DB		 1970 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1971 object_unref(OBJECT(bioc));
1d34e4bf 1972
c76201ab
DDAG		 1973 /*
1974 * Make sure the receiver can get incoming pages before we send the rest
1975 * of the state
1976 */
1977 qemu_savevm_send_postcopy_listen(fb);
1978
a1fbe750 1979 qemu_savevm_state_complete_precopy(fb, false, false);
58110f0a
VSO		 1980 if (migrate_postcopy_ram()) {
1981 qemu_savevm_send_ping(fb, 3);
1982 }
1d34e4bf
DDAG		 1983
1984 qemu_savevm_send_postcopy_run(fb);
1985
1986 /* <><> end of stuff going into the package */
1d34e4bf 1987
ef8d6488
DDAG		 1988 /* Last point of recovery; as soon as we send the package the destination
1989 * can open devices and potentially start running.
1990 * Lets just check again we've not got any errors.
1991 */
1992 ret = qemu_file_get_error(ms->to_dst_file);
1993 if (ret) {
1994 error_report("postcopy_start: Migration stream errored (pre package)");
1995 goto fail_closefb;
1996 }
1997
1998 restart_block = false;
1999
1d34e4bf 2000 /* Now send that blob */
61b67d47 2001 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1d34e4bf
DDAG		 2002 goto fail_closefb;
2003 }
2004 qemu_fclose(fb);
b82fc321
DDAG		 2005
2006 /* Send a notify to give a chance for anything that needs to happen
2007 * at the transition to postcopy and after the device state; in particular
2008 * spice needs to trigger a transition now
2009 */
2010 ms->postcopy_after_devices = true;
2011 notifier_list_notify(&migration_state_notifiers, ms);
2012
1d34e4bf
DDAG		 2013 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2014
2015 qemu_mutex_unlock_iothread();
2016
58110f0a
VSO		 2017 if (migrate_postcopy_ram()) {
2018 /*
2019 * Although this ping is just for debug, it could potentially be
2020 * used for getting a better measurement of downtime at the source.
2021 */
2022 qemu_savevm_send_ping(ms->to_dst_file, 4);
2023 }
1d34e4bf 2024
ced1c616
PB		 2025 if (migrate_release_ram()) {
2026 ram_postcopy_migrated_memory_release(ms);
2027 }
2028
89a02a9f 2029 ret = qemu_file_get_error(ms->to_dst_file);
1d34e4bf
DDAG		 2030 if (ret) {
2031 error_report("postcopy_start: Migration stream errored");
48781e5b 2032 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf
DDAG		 2033 MIGRATION_STATUS_FAILED);
2034 }
2035
2036 return ret;
2037
2038fail_closefb:
2039 qemu_fclose(fb);
2040fail:
48781e5b 2041 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf 2042 MIGRATION_STATUS_FAILED);
ef8d6488
DDAG		 2043 if (restart_block) {
2044 /* A failure happened early enough that we know the destination hasn't
2045 * accessed block devices, so we're safe to recover.
2046 */
2047 Error *local_err = NULL;
2048
2049 bdrv_invalidate_cache_all(&local_err);
2050 if (local_err) {
2051 error_report_err(local_err);
2052 }
2053 }
1d34e4bf
DDAG		 2054 qemu_mutex_unlock_iothread();
2055 return -1;
2056}
2057
e91d8951
DDAG		 2058/**
2059 * migration_maybe_pause: Pause if required to by
2060 * migrate_pause_before_switchover called with the iothread locked
2061 * Returns: 0 on success
2062 */
0331c8ca
DDAG		 2063static int migration_maybe_pause(MigrationState *s,
2064 int *current_active_state,
2065 int new_state)
e91d8951
DDAG		 2066{
2067 if (!migrate_pause_before_switchover()) {
2068 return 0;
2069 }
2070
2071 /* Since leaving this state is not atomic with posting the semaphore
2072 * it's possible that someone could have issued multiple migrate_continue
2073 * and the semaphore is incorrectly positive at this point;
2074 * the docs say it's undefined to reinit a semaphore that's already
2075 * init'd, so use timedwait to eat up any existing posts.
2076 */
2077 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2078 /* This block intentionally left blank */
2079 }
2080
2081 qemu_mutex_unlock_iothread();
2082 migrate_set_state(&s->state, *current_active_state,
2083 MIGRATION_STATUS_PRE_SWITCHOVER);
2084 qemu_sem_wait(&s->pause_sem);
2085 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
0331c8ca
DDAG		 2086 new_state);
2087 *current_active_state = new_state;
e91d8951
DDAG		 2088 qemu_mutex_lock_iothread();
2089
0331c8ca 2090 return s->state == new_state ? 0 : -EINVAL;
e91d8951
DDAG		 2091}
2092
09f6c85e
DDAG		 2093/**
2094 * migration_completion: Used by migration_thread when there's not much left.
2095 * The caller 'breaks' the loop when this returns.
2096 *
2097 * @s: Current migration state
36f48567 2098 * @current_active_state: The migration state we expect to be in
09f6c85e 2099 */
64909f97 2100static void migration_completion(MigrationState *s, int current_active_state)
09f6c85e
DDAG		 2101{
2102 int ret;
2103
b10ac0c4
DDAG		 2104 if (s->state == MIGRATION_STATUS_ACTIVE) {
2105 qemu_mutex_lock_iothread();
64909f97 2106 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
b10ac0c4 2107 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
7287cbd4 2108 s->vm_was_running = runstate_is_running();
b10ac0c4
DDAG		 2109 ret = global_state_store();
2110
2111 if (!ret) {
a1fbe750 2112 bool inactivate = !migrate_colo_enabled();
b10ac0c4 2113 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
e91d8951 2114 if (ret >= 0) {
0331c8ca
DDAG		 2115 ret = migration_maybe_pause(s, &current_active_state,
2116 MIGRATION_STATUS_DEVICE);
e91d8951 2117 }
f07fa4cb
KW		 2118 if (ret >= 0) {
2119 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
a1fbe750
FZ		 2120 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2121 inactivate);
f07fa4cb 2122 }
a1fbe750
FZ		 2123 if (inactivate && ret >= 0) {
2124 s->block_inactive = true;
b10ac0c4
DDAG		 2125 }
2126 }
2127 qemu_mutex_unlock_iothread();
09f6c85e 2128
b10ac0c4
DDAG		 2129 if (ret < 0) {
2130 goto fail;
09f6c85e 2131 }
b10ac0c4
DDAG		 2132 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2133 trace_migration_completion_postcopy_end();
2134
89a02a9f 2135 qemu_savevm_state_complete_postcopy(s->to_dst_file);
b10ac0c4 2136 trace_migration_completion_postcopy_end_after_complete();
09f6c85e 2137 }
09f6c85e 2138
b10ac0c4
DDAG		 2139 /*
2140 * If rp was opened we must clean up the thread before
2141 * cleaning everything else up (since if there are no failures
2142 * it will wait for the destination to send it's status in
2143 * a SHUT command).
b10ac0c4 2144 */
0425dc97 2145 if (s->rp_state.from_dst_file) {
b10ac0c4 2146 int rp_error;
0425dc97 2147 trace_migration_return_path_end_before();
b10ac0c4 2148 rp_error = await_return_path_close_on_source(s);
0425dc97 2149 trace_migration_return_path_end_after(rp_error);
b10ac0c4 2150 if (rp_error) {
fe904ea8 2151 goto fail_invalidate;
b10ac0c4 2152 }
09f6c85e
DDAG		 2153 }
2154
89a02a9f 2155 if (qemu_file_get_error(s->to_dst_file)) {
09f6c85e 2156 trace_migration_completion_file_err();
fe904ea8 2157 goto fail_invalidate;
09f6c85e
DDAG		 2158 }
2159
0b827d5e
HZ		 2160 if (!migrate_colo_enabled()) {
2161 migrate_set_state(&s->state, current_active_state,
2162 MIGRATION_STATUS_COMPLETED);
2163 }
2164
09f6c85e
DDAG		 2165 return;
2166
fe904ea8
GK		 2167fail_invalidate:
2168 /* If not doing postcopy, vm_start() will be called: let's regain
2169 * control on images.
2170 */
2171 if (s->state == MIGRATION_STATUS_ACTIVE) {
2172 Error *local_err = NULL;
2173
1d2acc31 2174 qemu_mutex_lock_iothread();
fe904ea8
GK		 2175 bdrv_invalidate_cache_all(&local_err);
2176 if (local_err) {
2177 error_report_err(local_err);
1d2acc31
HZ		 2178 } else {
2179 s->block_inactive = false;
fe904ea8 2180 }
1d2acc31 2181 qemu_mutex_unlock_iothread();
fe904ea8
GK		 2182 }
2183
09f6c85e 2184fail:
48781e5b
HZ		 2185 migrate_set_state(&s->state, current_active_state,
2186 MIGRATION_STATUS_FAILED);
09f6c85e
DDAG		 2187}
2188
35a6ed4f
HZ		 2189bool migrate_colo_enabled(void)
2190{
2191 MigrationState *s = migrate_get_current();
2192 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2193}
2194
cf011f08
PX		 2195static void migration_calculate_complete(MigrationState *s)
2196{
2197 uint64_t bytes = qemu_ftell(s->to_dst_file);
2198 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2199
2200 s->total_time = end_time - s->start_time;
2201 if (!s->downtime) {
2202 /*
2203 * It's still not set, so we are precopy migration. For
2204 * postcopy, downtime is calculated during postcopy_start().
2205 */
2206 s->downtime = end_time - s->downtime_start;
2207 }
2208
2209 if (s->total_time) {
2210 s->mbps = ((double) bytes * 8.0) / s->total_time / 1000;
2211 }
2212}
2213
b15df1ae
PX		 2214static void migration_update_counters(MigrationState *s,
2215 int64_t current_time)
2216{
2217 uint64_t transferred, time_spent;
2218 int64_t threshold_size;
2219 double bandwidth;
2220
2221 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
2222 return;
2223 }
2224
2225 transferred = qemu_ftell(s->to_dst_file) - s->iteration_initial_bytes;
2226 time_spent = current_time - s->iteration_start_time;
2227 bandwidth = (double)transferred / time_spent;
2228 threshold_size = bandwidth * s->parameters.downtime_limit;
2229
2230 s->mbps = (((double) transferred * 8.0) /
2231 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2232
2233 /*
2234 * if we haven't sent anything, we don't want to
2235 * recalculate. 10000 is a small enough number for our purposes
2236 */
2237 if (ram_counters.dirty_pages_rate && transferred > 10000) {
2238 s->expected_downtime = ram_counters.dirty_pages_rate *
2239 qemu_target_page_size() / bandwidth;
2240 }
2241
2242 qemu_file_reset_rate_limit(s->to_dst_file);
2243
2244 s->iteration_start_time = current_time;
2245 s->iteration_initial_bytes = qemu_ftell(s->to_dst_file);
2246
2247 trace_migrate_transferred(transferred, time_spent,
2248 bandwidth, threshold_size);
2249}
2250
70b20477
DDAG		 2251/*
2252 * Master migration thread on the source VM.
2253 * It drives the migration and pumps the data down the outgoing channel.
2254 */
5f496a1b 2255static void *migration_thread(void *opaque)
0d82d0e8 2256{
9848a404 2257 MigrationState *s = opaque;
bc72ad67 2258 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1d34e4bf
DDAG		 2259 bool entered_postcopy = false;
2260 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
2261 enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
76f5933a 2262
ab28bd23
PB		 2263 rcu_register_thread();
2264
b15df1ae
PX		 2265 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2266
89a02a9f 2267 qemu_savevm_state_header(s->to_dst_file);
1d34e4bf 2268
62a02658
PX		 2269 /*
2270 * If we opened the return path, we need to make sure dst has it
2271 * opened as well.
2272 */
2273 if (s->rp_state.from_dst_file) {
1d34e4bf 2274 /* Now tell the dest that it should open its end so it can reply */
89a02a9f 2275 qemu_savevm_send_open_return_path(s->to_dst_file);
1d34e4bf
DDAG		 2276
2277 /* And do a ping that will make stuff easier to debug */
89a02a9f 2278 qemu_savevm_send_ping(s->to_dst_file, 1);
0425dc97 2279 }
1d34e4bf 2280
58110f0a 2281 if (migrate_postcopy()) {
1d34e4bf
DDAG		 2282 /*
2283 * Tell the destination that we *might* want to do postcopy later;
2284 * if the other end can't do postcopy it should fail now, nice and
2285 * early.
2286 */
89a02a9f 2287 qemu_savevm_send_postcopy_advise(s->to_dst_file);
1d34e4bf
DDAG		 2288 }
2289
9907e842 2290 qemu_savevm_state_setup(s->to_dst_file);
0d82d0e8 2291
bc72ad67 2292 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
48781e5b
HZ		 2293 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2294 MIGRATION_STATUS_ACTIVE);
29ae8a41 2295
9ec055ae
DDAG		 2296 trace_migration_thread_setup_complete();
2297
2298 while (s->state == MIGRATION_STATUS_ACTIVE ||
2299 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
a3e879cd 2300 int64_t current_time;
c369f40d 2301 uint64_t pending_size;
0d82d0e8 2302
89a02a9f 2303 if (!qemu_file_rate_limit(s->to_dst_file)) {
c31b098f
DDAG		 2304 uint64_t pend_post, pend_nonpost;
2305
b15df1ae 2306 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size,
faec066a 2307 &pend_nonpost, &pend_post);
c31b098f 2308 pending_size = pend_nonpost + pend_post;
b15df1ae 2309 trace_migrate_pending(pending_size, s->threshold_size,
c31b098f 2310 pend_post, pend_nonpost);
b15df1ae 2311 if (pending_size && pending_size >= s->threshold_size) {
1d34e4bf
DDAG		 2312 /* Still a significant amount to transfer */
2313
58110f0a 2314 if (migrate_postcopy() &&
1d34e4bf 2315 s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
b15df1ae 2316 pend_nonpost <= s->threshold_size &&
1d34e4bf
DDAG		 2317 atomic_read(&s->start_postcopy)) {
2318
7287cbd4 2319 if (!postcopy_start(s)) {
1d34e4bf
DDAG		 2320 current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
2321 entered_postcopy = true;
2322 }
2323
2324 continue;
2325 }
2326 /* Just another iteration step */
89a02a9f 2327 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
c369f40d 2328 } else {
09f6c85e 2329 trace_migration_thread_low_pending(pending_size);
64909f97 2330 migration_completion(s, current_active_state);
09f6c85e 2331 break;
c369f40d
JQ		 2332 }
2333 }
f4410a5d 2334
89a02a9f 2335 if (qemu_file_get_error(s->to_dst_file)) {
48781e5b
HZ		 2336 migrate_set_state(&s->state, current_active_state,
2337 MIGRATION_STATUS_FAILED);
1d34e4bf 2338 trace_migration_thread_file_err();
fd45ee2c
PB		 2339 break;
2340 }
b15df1ae 2341
bc72ad67 2342 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
0d82d0e8 2343
b15df1ae
PX		 2344 migration_update_counters(s, current_time);
2345
89a02a9f 2346 if (qemu_file_rate_limit(s->to_dst_file)) {
0d82d0e8 2347 /* usleep expects microseconds */
b15df1ae
PX		 2348 g_usleep((s->iteration_start_time + BUFFER_DELAY -
2349 current_time) * 1000);
0d82d0e8 2350 }
a3fa1d78
PB		 2351 }
2352
1d34e4bf 2353 trace_migration_thread_after_loop();
070afca2
JH		 2354 /* If we enabled cpu throttling for auto-converge, turn it off. */
2355 cpu_throttle_stop();
2356
f4410a5d 2357 qemu_mutex_lock_iothread();
39b9e179
PX		 2358 switch (s->state) {
2359 case MIGRATION_STATUS_COMPLETED:
cf011f08 2360 migration_calculate_complete(s);
a3fa1d78 2361 runstate_set(RUN_STATE_POSTMIGRATE);
39b9e179
PX		 2362 break;
2363
2364 case MIGRATION_STATUS_ACTIVE:
2365 /*
2366 * We should really assert here, but since it's during
2367 * migration, let's try to reduce the usage of assertions.
2368 */
2369 if (!migrate_colo_enabled()) {
2370 error_report("%s: critical error: calling COLO code without "
2371 "COLO enabled", __func__);
0b827d5e 2372 }
39b9e179
PX		 2373 migrate_start_colo_process(s);
2374 /*
2375 * Fixme: we will run VM in COLO no matter its old running state.
2376 * After exited COLO, we will keep running.
2377 */
2378 s->vm_was_running = true;
2379 /* Fallthrough */
2380 case MIGRATION_STATUS_FAILED:
2381 case MIGRATION_STATUS_CANCELLED:
7287cbd4 2382 if (s->vm_was_running) {
a3fa1d78 2383 vm_start();
42da5550
DDAG		 2384 } else {
2385 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2386 runstate_set(RUN_STATE_POSTMIGRATE);
2387 }
dba433c0 2388 }
39b9e179
PX		 2389 break;
2390
2391 default:
2392 /* Should not reach here, but if so, forgive the VM. */
2393 error_report("%s: Unknown ending state %d", __func__, s->state);
2394 break;
0d82d0e8 2395 }
bb1fadc4 2396 qemu_bh_schedule(s->cleanup_bh);
dba433c0 2397 qemu_mutex_unlock_iothread();
f4410a5d 2398
ab28bd23 2399 rcu_unregister_thread();
0d82d0e8
JQ		 2400 return NULL;
2401}
2402
9848a404 2403void migrate_fd_connect(MigrationState *s)
0d82d0e8 2404{
2ff30257 2405 s->expected_downtime = s->parameters.downtime_limit;
bb1fadc4 2406 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
0d82d0e8 2407
9e4d2b98 2408 qemu_file_set_blocking(s->to_dst_file, true);
89a02a9f 2409 qemu_file_set_rate_limit(s->to_dst_file,
2ff30257 2410 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
442773ce 2411
9287ac27
SH		 2412 /* Notify before starting migration thread */
2413 notifier_list_notify(&migration_state_notifiers, s);
2414
1d34e4bf 2415 /*
c788ada8
PX		 2416 * Open the return path. For postcopy, it is used exclusively. For
2417 * precopy, only if user specified "return-path" capability would
2418 * QEMU uses the return path.
1d34e4bf 2419 */
c788ada8 2420 if (migrate_postcopy_ram() || migrate_use_return_path()) {
1d34e4bf
DDAG		 2421 if (open_return_path_on_source(s)) {
2422 error_report("Unable to open return-path for postcopy");
48781e5b 2423 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1d34e4bf
DDAG		 2424 MIGRATION_STATUS_FAILED);
2425 migrate_fd_cleanup(s);
2426 return;
2427 }
2428 }
2429
f986c3d2
JQ		 2430 if (multifd_save_setup() != 0) {
2431 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2432 MIGRATION_STATUS_FAILED);
2433 migrate_fd_cleanup(s);
2434 return;
2435 }
009fad7f 2436 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
bb1fadc4 2437 QEMU_THREAD_JOINABLE);
1d34e4bf 2438 s->migration_thread_running = true;
0d82d0e8 2439}
093e3c42 2440
9d18af93
PX		 2441void migration_global_dump(Monitor *mon)
2442{
2443 MigrationState *ms = migrate_get_current();
2444
6f0f6428
JQ		 2445 monitor_printf(mon, "globals:\n");
2446 monitor_printf(mon, "store-global-state: %s\n",
2447 ms->store_global_state ? "on" : "off");
2448 monitor_printf(mon, "only-migratable: %s\n",
2449 ms->only_migratable ? "on" : "off");
2450 monitor_printf(mon, "send-configuration: %s\n",
2451 ms->send_configuration ? "on" : "off");
2452 monitor_printf(mon, "send-section-footer: %s\n",
2453 ms->send_section_footer ? "on" : "off");
9d18af93
PX		 2454}
2455
20814758
PX		 2456#define DEFINE_PROP_MIG_CAP(name, x) \
2457 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2458
5272298c
PX		 2459static Property migration_properties[] = {
2460 DEFINE_PROP_BOOL("store-global-state", MigrationState,
2461 store_global_state, true),
3df663e5 2462 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
71dd4c1a
PX		 2463 DEFINE_PROP_BOOL("send-configuration", MigrationState,
2464 send_configuration, true),
15c38503
PX		 2465 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2466 send_section_footer, true),
89632faf
PX		 2467
2468 /* Migration parameters */
741d4086 2469 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
89632faf
PX		 2470 parameters.compress_level,
2471 DEFAULT_MIGRATE_COMPRESS_LEVEL),
741d4086 2472 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
89632faf
PX		 2473 parameters.compress_threads,
2474 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
741d4086 2475 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
89632faf
PX		 2476 parameters.decompress_threads,
2477 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
741d4086 2478 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
89632faf
PX		 2479 parameters.cpu_throttle_initial,
2480 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
741d4086 2481 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
89632faf
PX		 2482 parameters.cpu_throttle_increment,
2483 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
741d4086 2484 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
89632faf 2485 parameters.max_bandwidth, MAX_THROTTLE),
741d4086 2486 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
89632faf
PX		 2487 parameters.downtime_limit,
2488 DEFAULT_MIGRATE_SET_DOWNTIME),
741d4086 2489 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
89632faf
PX		 2490 parameters.x_checkpoint_delay,
2491 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
741d4086 2492 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState,
4075fb1c
JQ		 2493 parameters.x_multifd_channels,
2494 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
741d4086 2495 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState,
0fb86605
JQ		 2496 parameters.x_multifd_page_count,
2497 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
73af8dd8
JQ		 2498 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
2499 parameters.xbzrle_cache_size,
2500 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
20814758
PX		 2501
2502 /* Migration capabilities */
2503 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2504 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2505 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2506 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2507 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2508 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2509 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2510 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2511 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2512 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2513 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
30126bbf 2514 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
20814758 2515
5272298c
PX		 2516 DEFINE_PROP_END_OF_LIST(),
2517};
2518
e5cb7e76
PX		 2519static void migration_class_init(ObjectClass *klass, void *data)
2520{
2521 DeviceClass *dc = DEVICE_CLASS(klass);
2522
2523 dc->user_creatable = false;
5272298c 2524 dc->props = migration_properties;
e5cb7e76
PX		 2525}
2526
b91bf5e4
MAL		 2527static void migration_instance_finalize(Object *obj)
2528{
2529 MigrationState *ms = MIGRATION_OBJ(obj);
2530 MigrationParameters *params = &ms->parameters;
2531
87db1a7d 2532 qemu_mutex_destroy(&ms->error_mutex);
b91bf5e4
MAL		 2533 g_free(params->tls_hostname);
2534 g_free(params->tls_creds);
e91d8951 2535 qemu_sem_destroy(&ms->pause_sem);
b91bf5e4
MAL		 2536}
2537
e5cb7e76
PX		 2538static void migration_instance_init(Object *obj)
2539{
2540 MigrationState *ms = MIGRATION_OBJ(obj);
8b0b29dc 2541 MigrationParameters *params = &ms->parameters;
e5cb7e76
PX		 2542
2543 ms->state = MIGRATION_STATUS_NONE;
e5cb7e76 2544 ms->mbps = -1;
e91d8951 2545 qemu_sem_init(&ms->pause_sem, 0);
87db1a7d 2546 qemu_mutex_init(&ms->error_mutex);
8b0b29dc
PX		 2547
2548 params->tls_hostname = g_strdup("");
2549 params->tls_creds = g_strdup("");
2550
2551 /* Set has_* up only for parameter checks */
2552 params->has_compress_level = true;
2553 params->has_compress_threads = true;
2554 params->has_decompress_threads = true;
2555 params->has_cpu_throttle_initial = true;
2556 params->has_cpu_throttle_increment = true;
2557 params->has_max_bandwidth = true;
2558 params->has_downtime_limit = true;
2559 params->has_x_checkpoint_delay = true;
2560 params->has_block_incremental = true;
4075fb1c 2561 params->has_x_multifd_channels = true;
0fb86605 2562 params->has_x_multifd_page_count = true;
73af8dd8 2563 params->has_xbzrle_cache_size = true;
8b0b29dc
PX		 2564}
2565
2566/*
2567 * Return true if check pass, false otherwise. Error will be put
2568 * inside errp if provided.
2569 */
2570static bool migration_object_check(MigrationState *ms, Error **errp)
2571{
6b19a7d9
PX		 2572 MigrationCapabilityStatusList *head = NULL;
2573 /* Assuming all off */
2574 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2575 int i;
2576
8b0b29dc
PX		 2577 if (!migrate_params_check(&ms->parameters, errp)) {
2578 return false;
2579 }
2580
6b19a7d9
PX		 2581 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2582 if (ms->enabled_capabilities[i]) {
2583 head = migrate_cap_add(head, i, true);
2584 }
2585 }
2586
2587 ret = migrate_caps_check(cap_list, head, errp);
2588
2589 /* It works with head == NULL */
2590 qapi_free_MigrationCapabilityStatusList(head);
2591
2592 return ret;
e5cb7e76
PX		 2593}
2594
2595static const TypeInfo migration_type = {
2596 .name = TYPE_MIGRATION,
01f6e14c 2597 /*
c8d3ff38
PX		 2598 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2599 * not created using qdev_create(), it is not attached to the qdev
2600 * device tree, and it is never realized.
2601 *
2602 * TODO: Make this TYPE_OBJECT once QOM provides something like
2603 * TYPE_DEVICE's "-global" properties.
01f6e14c 2604 */
e5cb7e76
PX		 2605 .parent = TYPE_DEVICE,
2606 .class_init = migration_class_init,
2607 .class_size = sizeof(MigrationClass),
2608 .instance_size = sizeof(MigrationState),
2609 .instance_init = migration_instance_init,
b91bf5e4 2610 .instance_finalize = migration_instance_finalize,
e5cb7e76
PX		 2611};
2612
2613static void register_migration_types(void)
2614{
2615 type_register_static(&migration_type);
2616}
2617
2618type_init(register_migration_types);
QEMU mirror