]> 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: Guard ram_bytes_remaining against early call
[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
ca6011c2 594static void fill_source_migration_info(MigrationInfo *info)
5bb7910a 595{
17549e84
JQ		 596 MigrationState *s = migrate_get_current();
597
598 switch (s->state) {
31194731 599 case MIGRATION_STATUS_NONE:
17549e84 600 /* no migration has happened ever */
ca6011c2
AP		 601 /* do not overwrite destination migration status */
602 return;
17549e84 603 break;
31194731 604 case MIGRATION_STATUS_SETUP:
29ae8a41 605 info->has_status = true;
ed4fbd10 606 info->has_total_time = false;
29ae8a41 607 break;
31194731
HZ		 608 case MIGRATION_STATUS_ACTIVE:
609 case MIGRATION_STATUS_CANCELLING:
9ec055ae 610 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
31e06077
DDAG		 611 case MIGRATION_STATUS_PRE_SWITCHOVER:
612 case MIGRATION_STATUS_DEVICE:
c8f9f4f4 613 /* TODO add some postcopy stats */
9ec055ae
DDAG		 614 info->has_status = true;
615 info->has_total_time = true;
616 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
617 - s->total_time;
618 info->has_expected_downtime = true;
619 info->expected_downtime = s->expected_downtime;
620 info->has_setup_time = true;
621 info->setup_time = s->setup_time;
622
a22463a5 623 populate_ram_info(info, s);
930ac04c 624 populate_disk_info(info);
17549e84 625 break;
0b827d5e
HZ		 626 case MIGRATION_STATUS_COLO:
627 info->has_status = true;
628 /* TODO: display COLO specific information (checkpoint info etc.) */
629 break;
31194731 630 case MIGRATION_STATUS_COMPLETED:
791e7c82 631 info->has_status = true;
00c14997 632 info->has_total_time = true;
7aa939af 633 info->total_time = s->total_time;
9c5a9fcf
JQ		 634 info->has_downtime = true;
635 info->downtime = s->downtime;
ed4fbd10
MH		 636 info->has_setup_time = true;
637 info->setup_time = s->setup_time;
d5f8a570 638
a22463a5 639 populate_ram_info(info, s);
17549e84 640 break;
31194731 641 case MIGRATION_STATUS_FAILED:
791e7c82 642 info->has_status = true;
d59ce6f3
DB		 643 if (s->error) {
644 info->has_error_desc = true;
645 info->error_desc = g_strdup(error_get_pretty(s->error));
646 }
17549e84 647 break;
31194731 648 case MIGRATION_STATUS_CANCELLED:
791e7c82 649 info->has_status = true;
17549e84 650 break;
5bb7910a 651 }
cde63fbe 652 info->status = s->state;
5bb7910a
AL		 653}
654
4a84214e
PX		 655/**
656 * @migration_caps_check - check capability validity
657 *
658 * @cap_list: old capability list, array of bool
659 * @params: new capabilities to be applied soon
660 * @errp: set *errp if the check failed, with reason
661 *
662 * Returns true if check passed, otherwise false.
663 */
664static bool migrate_caps_check(bool *cap_list,
665 MigrationCapabilityStatusList *params,
666 Error **errp)
00458433 667{
00458433 668 MigrationCapabilityStatusList *cap;
4a84214e 669 bool old_postcopy_cap;
d7651f15 670 MigrationIncomingState *mis = migration_incoming_get_current();
00458433 671
4a84214e 672 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
00458433
OW		 673
674 for (cap = params; cap; cap = cap->next) {
4a84214e
PX		 675 cap_list[cap->value->capability] = cap->value->state;
676 }
677
ed1701c6 678#ifndef CONFIG_LIVE_BLOCK_MIGRATION
4a84214e
PX		 679 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
680 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
681 "block migration");
682 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
683 return false;
00458433 684 }
4a84214e 685#endif
53dd370c 686
4a84214e
PX		 687 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
688 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
53dd370c
DDAG		 689 /* The decompression threads asynchronously write into RAM
690 * rather than use the atomic copies needed to avoid
691 * userfaulting. It should be possible to fix the decompression
692 * threads for compatibility in future.
693 */
4a84214e
PX		 694 error_setg(errp, "Postcopy is not currently compatible "
695 "with compression");
696 return false;
53dd370c 697 }
4a84214e 698
096631bd
DDAG		 699 /* This check is reasonably expensive, so only when it's being
700 * set the first time, also it's only the destination that needs
701 * special support.
702 */
703 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
d7651f15 704 !postcopy_ram_supported_by_host(mis)) {
096631bd
DDAG		 705 /* postcopy_ram_supported_by_host will have emitted a more
706 * detailed message
707 */
4a84214e
PX		 708 error_setg(errp, "Postcopy is not supported");
709 return false;
096631bd 710 }
53dd370c 711 }
4a84214e
PX		 712
713 return true;
714}
715
ca6011c2
AP		 716static void fill_destination_migration_info(MigrationInfo *info)
717{
718 MigrationIncomingState *mis = migration_incoming_get_current();
719
720 switch (mis->state) {
721 case MIGRATION_STATUS_NONE:
722 return;
723 break;
724 case MIGRATION_STATUS_SETUP:
725 case MIGRATION_STATUS_CANCELLING:
726 case MIGRATION_STATUS_CANCELLED:
727 case MIGRATION_STATUS_ACTIVE:
728 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
729 case MIGRATION_STATUS_FAILED:
730 case MIGRATION_STATUS_COLO:
731 info->has_status = true;
732 break;
733 case MIGRATION_STATUS_COMPLETED:
734 info->has_status = true;
735 fill_destination_postcopy_migration_info(info);
736 break;
737 }
738 info->status = mis->state;
739}
740
741MigrationInfo *qmp_query_migrate(Error **errp)
742{
743 MigrationInfo *info = g_malloc0(sizeof(*info));
744
745 fill_destination_migration_info(info);
746 fill_source_migration_info(info);
747
748 return info;
749}
750
4a84214e
PX		 751void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
752 Error **errp)
753{
754 MigrationState *s = migrate_get_current();
755 MigrationCapabilityStatusList *cap;
756
757 if (migration_is_setup_or_active(s->state)) {
758 error_setg(errp, QERR_MIGRATION_ACTIVE);
759 return;
760 }
761
762 if (!migrate_caps_check(s->enabled_capabilities, params, errp)) {
763 return;
764 }
765
766 for (cap = params; cap; cap = cap->next) {
767 s->enabled_capabilities[cap->value->capability] = cap->value->state;
768 }
00458433
OW		 769}
770
16d063bc
PX		 771/*
772 * Check whether the parameters are valid. Error will be put into errp
773 * (if provided). Return true if valid, otherwise false.
774 */
775static bool migrate_params_check(MigrationParameters *params, Error **errp)
85de8323 776{
7f375e04 777 if (params->has_compress_level &&
741d4086 778 (params->compress_level > 9)) {
c6bd8c70
MA		 779 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
780 "is invalid, it should be in the range of 0 to 9");
16d063bc 781 return false;
85de8323 782 }
16d063bc 783
741d4086 784 if (params->has_compress_threads && (params->compress_threads < 1)) {
c6bd8c70
MA		 785 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
786 "compress_threads",
787 "is invalid, it should be in the range of 1 to 255");
16d063bc 788 return false;
85de8323 789 }
16d063bc 790
741d4086 791 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
c6bd8c70
MA		 792 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
793 "decompress_threads",
794 "is invalid, it should be in the range of 1 to 255");
16d063bc 795 return false;
85de8323 796 }
16d063bc 797
7f375e04
EB		 798 if (params->has_cpu_throttle_initial &&
799 (params->cpu_throttle_initial < 1 ||
800 params->cpu_throttle_initial > 99)) {
1626fee3 801 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 802 "cpu_throttle_initial",
1626fee3 803 "an integer in the range of 1 to 99");
16d063bc 804 return false;
1626fee3 805 }
16d063bc 806
7f375e04
EB		 807 if (params->has_cpu_throttle_increment &&
808 (params->cpu_throttle_increment < 1 ||
809 params->cpu_throttle_increment > 99)) {
1626fee3 810 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 811 "cpu_throttle_increment",
1626fee3 812 "an integer in the range of 1 to 99");
16d063bc 813 return false;
1626fee3 814 }
16d063bc 815
741d4086 816 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
2ff30257
AA		 817 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
818 " range of 0 to %zu bytes/second", SIZE_MAX);
16d063bc 819 return false;
2ff30257 820 }
16d063bc 821
2ff30257 822 if (params->has_downtime_limit &&
741d4086 823 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
87c9cc1c
DHB		 824 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
825 "the range of 0 to %d milliseconds",
826 MAX_MIGRATE_DOWNTIME);
16d063bc 827 return false;
2ff30257 828 }
16d063bc 829
741d4086
JQ		 830 /* x_checkpoint_delay is now always positive */
831
832 if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) {
4075fb1c
JQ		 833 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
834 "multifd_channels",
835 "is invalid, it should be in the range of 1 to 255");
836 return false;
837 }
0fb86605 838 if (params->has_x_multifd_page_count &&
741d4086
JQ		 839 (params->x_multifd_page_count < 1 ||
840 params->x_multifd_page_count > 10000)) {
0fb86605
JQ		 841 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
842 "multifd_page_count",
843 "is invalid, it should be in the range of 1 to 10000");
844 return false;
845 }
16d063bc 846
73af8dd8
JQ		 847 if (params->has_xbzrle_cache_size &&
848 (params->xbzrle_cache_size < qemu_target_page_size() ||
849 !is_power_of_2(params->xbzrle_cache_size))) {
850 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
851 "xbzrle_cache_size",
852 "is invalid, it should be bigger than target page size"
853 " and a power of two");
854 return false;
855 }
856
16d063bc
PX		 857 return true;
858}
859
1bda8b3c
MA		 860static void migrate_params_test_apply(MigrateSetParameters *params,
861 MigrationParameters *dest)
862{
863 *dest = migrate_get_current()->parameters;
864
865 /* TODO use QAPI_CLONE() instead of duplicating it inline */
866
867 if (params->has_compress_level) {
868 dest->compress_level = params->compress_level;
869 }
870
871 if (params->has_compress_threads) {
872 dest->compress_threads = params->compress_threads;
873 }
874
875 if (params->has_decompress_threads) {
876 dest->decompress_threads = params->decompress_threads;
877 }
878
879 if (params->has_cpu_throttle_initial) {
880 dest->cpu_throttle_initial = params->cpu_throttle_initial;
881 }
882
883 if (params->has_cpu_throttle_increment) {
884 dest->cpu_throttle_increment = params->cpu_throttle_increment;
885 }
886
887 if (params->has_tls_creds) {
01fa5598
MA		 888 assert(params->tls_creds->type == QTYPE_QSTRING);
889 dest->tls_creds = g_strdup(params->tls_creds->u.s);
1bda8b3c
MA		 890 }
891
892 if (params->has_tls_hostname) {
01fa5598
MA		 893 assert(params->tls_hostname->type == QTYPE_QSTRING);
894 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1bda8b3c
MA		 895 }
896
897 if (params->has_max_bandwidth) {
898 dest->max_bandwidth = params->max_bandwidth;
899 }
900
901 if (params->has_downtime_limit) {
902 dest->downtime_limit = params->downtime_limit;
903 }
904
905 if (params->has_x_checkpoint_delay) {
906 dest->x_checkpoint_delay = params->x_checkpoint_delay;
907 }
908
909 if (params->has_block_incremental) {
910 dest->block_incremental = params->block_incremental;
911 }
5e7577a1
JQ		 912 if (params->has_x_multifd_channels) {
913 dest->x_multifd_channels = params->x_multifd_channels;
914 }
915 if (params->has_x_multifd_page_count) {
916 dest->x_multifd_page_count = params->x_multifd_page_count;
917 }
73af8dd8
JQ		 918 if (params->has_xbzrle_cache_size) {
919 dest->xbzrle_cache_size = params->xbzrle_cache_size;
920 }
1bda8b3c
MA		 921}
922
73af8dd8 923static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
16d063bc
PX		 924{
925 MigrationState *s = migrate_get_current();
926
e87fae4c
MA		 927 /* TODO use QAPI_CLONE() instead of duplicating it inline */
928
7f375e04
EB		 929 if (params->has_compress_level) {
930 s->parameters.compress_level = params->compress_level;
85de8323 931 }
476c72aa 932
7f375e04
EB		 933 if (params->has_compress_threads) {
934 s->parameters.compress_threads = params->compress_threads;
85de8323 935 }
476c72aa 936
7f375e04
EB		 937 if (params->has_decompress_threads) {
938 s->parameters.decompress_threads = params->decompress_threads;
85de8323 939 }
476c72aa 940
7f375e04
EB		 941 if (params->has_cpu_throttle_initial) {
942 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1626fee3 943 }
476c72aa 944
7f375e04
EB		 945 if (params->has_cpu_throttle_increment) {
946 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1626fee3 947 }
476c72aa 948
7f375e04 949 if (params->has_tls_creds) {
69ef1f36 950 g_free(s->parameters.tls_creds);
01fa5598
MA		 951 assert(params->tls_creds->type == QTYPE_QSTRING);
952 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
69ef1f36 953 }
476c72aa 954
7f375e04 955 if (params->has_tls_hostname) {
69ef1f36 956 g_free(s->parameters.tls_hostname);
01fa5598
MA		 957 assert(params->tls_hostname->type == QTYPE_QSTRING);
958 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
69ef1f36 959 }
476c72aa 960
2ff30257
AA		 961 if (params->has_max_bandwidth) {
962 s->parameters.max_bandwidth = params->max_bandwidth;
963 if (s->to_dst_file) {
964 qemu_file_set_rate_limit(s->to_dst_file,
965 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
966 }
967 }
476c72aa 968
2ff30257
AA		 969 if (params->has_downtime_limit) {
970 s->parameters.downtime_limit = params->downtime_limit;
971 }
68b53591
HZ		 972
973 if (params->has_x_checkpoint_delay) {
974 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
479125d5
HZ		 975 if (migration_in_colo_state()) {
976 colo_checkpoint_notify(s);
977 }
68b53591 978 }
476c72aa 979
2833c59b
JQ		 980 if (params->has_block_incremental) {
981 s->parameters.block_incremental = params->block_incremental;
982 }
4075fb1c
JQ		 983 if (params->has_x_multifd_channels) {
984 s->parameters.x_multifd_channels = params->x_multifd_channels;
985 }
0fb86605
JQ		 986 if (params->has_x_multifd_page_count) {
987 s->parameters.x_multifd_page_count = params->x_multifd_page_count;
988 }
73af8dd8
JQ		 989 if (params->has_xbzrle_cache_size) {
990 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
991 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
992 }
85de8323
LL		 993}
994
1bda8b3c 995void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
476c72aa 996{
1bda8b3c
MA		 997 MigrationParameters tmp;
998
01fa5598
MA		 999 /* TODO Rewrite "" to null instead */
1000 if (params->has_tls_creds
1001 && params->tls_creds->type == QTYPE_QNULL) {
1002 QDECREF(params->tls_creds->u.n);
1003 params->tls_creds->type = QTYPE_QSTRING;
1004 params->tls_creds->u.s = strdup("");
1005 }
1006 /* TODO Rewrite "" to null instead */
1007 if (params->has_tls_hostname
1008 && params->tls_hostname->type == QTYPE_QNULL) {
1009 QDECREF(params->tls_hostname->u.n);
1010 params->tls_hostname->type = QTYPE_QSTRING;
1011 params->tls_hostname->u.s = strdup("");
1012 }
1013
1bda8b3c
MA		 1014 migrate_params_test_apply(params, &tmp);
1015
1016 if (!migrate_params_check(&tmp, errp)) {
476c72aa
PX		 1017 /* Invalid parameter */
1018 return;
1019 }
1020
73af8dd8 1021 migrate_params_apply(params, errp);
476c72aa
PX		 1022}
1023
2594f56d 1024
4886a1bc
DDAG		 1025void qmp_migrate_start_postcopy(Error **errp)
1026{
1027 MigrationState *s = migrate_get_current();
1028
1029 if (!migrate_postcopy_ram()) {
a54d340b 1030 error_setg(errp, "Enable postcopy with migrate_set_capability before"
4886a1bc
DDAG		 1031 " the start of migration");
1032 return;
1033 }
1034
1035 if (s->state == MIGRATION_STATUS_NONE) {
1036 error_setg(errp, "Postcopy must be started after migration has been"
1037 " started");
1038 return;
1039 }
1040 /*
1041 * we don't error if migration has finished since that would be racy
1042 * with issuing this command.
1043 */
1044 atomic_set(&s->start_postcopy, true);
1045}
1046
065e2813
AL		 1047/* shared migration helpers */
1048
48781e5b 1049void migrate_set_state(int *state, int old_state, int new_state)
51cf4c1a 1050{
a31fedee 1051 assert(new_state < MIGRATION_STATUS__MAX);
48781e5b 1052 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
a31fedee 1053 trace_migrate_set_state(MigrationStatus_str(new_state));
b05dc723 1054 migrate_generate_event(new_state);
51cf4c1a
Z		 1055 }
1056}
1057
4e4a3d3a
PX		 1058static MigrationCapabilityStatusList *migrate_cap_add(
1059 MigrationCapabilityStatusList *list,
1060 MigrationCapability index,
1061 bool state)
2833c59b
JQ		 1062{
1063 MigrationCapabilityStatusList *cap;
1064
1065 cap = g_new0(MigrationCapabilityStatusList, 1);
1066 cap->value = g_new0(MigrationCapabilityStatus, 1);
4e4a3d3a
PX		 1067 cap->value->capability = index;
1068 cap->value->state = state;
1069 cap->next = list;
1070
1071 return cap;
1072}
1073
1074void migrate_set_block_enabled(bool value, Error **errp)
1075{
1076 MigrationCapabilityStatusList *cap;
1077
1078 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
2833c59b
JQ		 1079 qmp_migrate_set_capabilities(cap, errp);
1080 qapi_free_MigrationCapabilityStatusList(cap);
1081}
1082
1083static void migrate_set_block_incremental(MigrationState *s, bool value)
1084{
1085 s->parameters.block_incremental = value;
1086}
1087
1088static void block_cleanup_parameters(MigrationState *s)
1089{
1090 if (s->must_remove_block_options) {
1091 /* setting to false can never fail */
1092 migrate_set_block_enabled(false, &error_abort);
1093 migrate_set_block_incremental(s, false);
1094 s->must_remove_block_options = false;
1095 }
1096}
1097
bb1fadc4 1098static void migrate_fd_cleanup(void *opaque)
065e2813 1099{
bb1fadc4
PB		 1100 MigrationState *s = opaque;
1101
1102 qemu_bh_delete(s->cleanup_bh);
1103 s->cleanup_bh = NULL;
1104
89a02a9f 1105 if (s->to_dst_file) {
f986c3d2
JQ		 1106 Error *local_err = NULL;
1107
9013dca5 1108 trace_migrate_fd_cleanup();
404a7c05 1109 qemu_mutex_unlock_iothread();
1d34e4bf
DDAG		 1110 if (s->migration_thread_running) {
1111 qemu_thread_join(&s->thread);
1112 s->migration_thread_running = false;
1113 }
404a7c05
PB		 1114 qemu_mutex_lock_iothread();
1115
f986c3d2
JQ		 1116 if (multifd_save_cleanup(&local_err) != 0) {
1117 error_report_err(local_err);
1118 }
89a02a9f
HZ		 1119 qemu_fclose(s->to_dst_file);
1120 s->to_dst_file = NULL;
065e2813
AL		 1121 }
1122
9ec055ae
DDAG		 1123 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1124 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
7a2c1721 1125
94f5a437 1126 if (s->state == MIGRATION_STATUS_CANCELLING) {
48781e5b 1127 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
94f5a437 1128 MIGRATION_STATUS_CANCELLED);
7a2c1721 1129 }
a3fa1d78 1130
87db1a7d
JQ		 1131 if (s->error) {
1132 /* It is used on info migrate. We can't free it */
1133 error_report_err(error_copy(s->error));
1134 }
a3fa1d78 1135 notifier_list_notify(&migration_state_notifiers, s);
2833c59b 1136 block_cleanup_parameters(s);
065e2813
AL		 1137}
1138
87db1a7d
JQ		 1139void migrate_set_error(MigrationState *s, const Error *error)
1140{
1141 qemu_mutex_lock(&s->error_mutex);
1142 if (!s->error) {
1143 s->error = error_copy(error);
1144 }
1145 qemu_mutex_unlock(&s->error_mutex);
1146}
1147
d59ce6f3 1148void migrate_fd_error(MigrationState *s, const Error *error)
065e2813 1149{
25174055 1150 trace_migrate_fd_error(error_get_pretty(error));
89a02a9f 1151 assert(s->to_dst_file == NULL);
48781e5b
HZ		 1152 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1153 MIGRATION_STATUS_FAILED);
87db1a7d 1154 migrate_set_error(s, error);
bb1fadc4 1155 notifier_list_notify(&migration_state_notifiers, s);
2833c59b 1156 block_cleanup_parameters(s);
458cf28e
JQ		 1157}
1158
0edda1c4 1159static void migrate_fd_cancel(MigrationState *s)
065e2813 1160{
6f2b811a 1161 int old_state ;
89a02a9f 1162 QEMUFile *f = migrate_get_current()->to_dst_file;
9013dca5 1163 trace_migrate_fd_cancel();
065e2813 1164
70b20477
DDAG		 1165 if (s->rp_state.from_dst_file) {
1166 /* shutdown the rp socket, so causing the rp thread to shutdown */
1167 qemu_file_shutdown(s->rp_state.from_dst_file);
1168 }
1169
6f2b811a
Z		 1170 do {
1171 old_state = s->state;
f6844b99 1172 if (!migration_is_setup_or_active(old_state)) {
6f2b811a
Z		 1173 break;
1174 }
a7b36b48
DDAG		 1175 /* If the migration is paused, kick it out of the pause */
1176 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1177 qemu_sem_post(&s->pause_sem);
1178 }
48781e5b 1179 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
31194731 1180 } while (s->state != MIGRATION_STATUS_CANCELLING);
a26ba26e
DDAG		 1181
1182 /*
1183 * If we're unlucky the migration code might be stuck somewhere in a
1184 * send/write while the network has failed and is waiting to timeout;
1185 * if we've got shutdown(2) available then we can force it to quit.
1186 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1187 * called in a bh, so there is no race against this cancel.
1188 */
31194731 1189 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
a26ba26e
DDAG		 1190 qemu_file_shutdown(f);
1191 }
1d2acc31
HZ		 1192 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1193 Error *local_err = NULL;
1194
1195 bdrv_invalidate_cache_all(&local_err);
1196 if (local_err) {
1197 error_report_err(local_err);
1198 } else {
1199 s->block_inactive = false;
1200 }
1201 }
2833c59b 1202 block_cleanup_parameters(s);
065e2813
AL		 1203}
1204
99a0db9b
GH		 1205void add_migration_state_change_notifier(Notifier *notify)
1206{
1207 notifier_list_add(&migration_state_notifiers, notify);
1208}
1209
1210void remove_migration_state_change_notifier(Notifier *notify)
1211{
31552529 1212 notifier_remove(notify);
99a0db9b
GH		 1213}
1214
02edd2e7 1215bool migration_in_setup(MigrationState *s)
afe2df69 1216{
31194731 1217 return s->state == MIGRATION_STATUS_SETUP;
afe2df69
GH		 1218}
1219
7073693b 1220bool migration_has_finished(MigrationState *s)
99a0db9b 1221{
31194731 1222 return s->state == MIGRATION_STATUS_COMPLETED;
99a0db9b 1223}
0edda1c4 1224
afe2df69
GH		 1225bool migration_has_failed(MigrationState *s)
1226{
31194731
HZ		 1227 return (s->state == MIGRATION_STATUS_CANCELLED ||
1228 s->state == MIGRATION_STATUS_FAILED);
afe2df69
GH		 1229}
1230
5727309d 1231bool migration_in_postcopy(void)
9ec055ae 1232{
5727309d
JQ		 1233 MigrationState *s = migrate_get_current();
1234
9ec055ae
DDAG		 1235 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1236}
1237
b82fc321
DDAG		 1238bool migration_in_postcopy_after_devices(MigrationState *s)
1239{
5727309d 1240 return migration_in_postcopy() && s->postcopy_after_devices;
b82fc321
DDAG		 1241}
1242
fab35005 1243bool migration_is_idle(void)
fe44dc91 1244{
fab35005 1245 MigrationState *s = migrate_get_current();
fe44dc91
AA		 1246
1247 switch (s->state) {
1248 case MIGRATION_STATUS_NONE:
1249 case MIGRATION_STATUS_CANCELLED:
1250 case MIGRATION_STATUS_COMPLETED:
1251 case MIGRATION_STATUS_FAILED:
1252 return true;
1253 case MIGRATION_STATUS_SETUP:
1254 case MIGRATION_STATUS_CANCELLING:
1255 case MIGRATION_STATUS_ACTIVE:
1256 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1257 case MIGRATION_STATUS_COLO:
31e06077
DDAG		 1258 case MIGRATION_STATUS_PRE_SWITCHOVER:
1259 case MIGRATION_STATUS_DEVICE:
fe44dc91
AA		 1260 return false;
1261 case MIGRATION_STATUS__MAX:
1262 g_assert_not_reached();
1263 }
1264
1265 return false;
1266}
1267
a0762d9e 1268MigrationState *migrate_init(void)
0edda1c4 1269{
17549e84 1270 MigrationState *s = migrate_get_current();
bbf6da32 1271
389775d1
DDAG		 1272 /*
1273 * Reinitialise all migration state, except
1274 * parameters/capabilities that the user set, and
1275 * locks.
1276 */
1277 s->bytes_xfer = 0;
1278 s->xfer_limit = 0;
1279 s->cleanup_bh = 0;
89a02a9f 1280 s->to_dst_file = NULL;
389775d1 1281 s->state = MIGRATION_STATUS_NONE;
389775d1
DDAG		 1282 s->rp_state.from_dst_file = NULL;
1283 s->rp_state.error = false;
1284 s->mbps = 0.0;
1285 s->downtime = 0;
1286 s->expected_downtime = 0;
389775d1 1287 s->setup_time = 0;
389775d1 1288 s->start_postcopy = false;
b82fc321 1289 s->postcopy_after_devices = false;
389775d1 1290 s->migration_thread_running = false;
d59ce6f3
DB		 1291 error_free(s->error);
1292 s->error = NULL;
389775d1 1293
48781e5b 1294 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
0edda1c4 1295
bc72ad67 1296 s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
0edda1c4
JQ		 1297 return s;
1298}
cab30143 1299
fa2756b7
AL		 1300static GSList *migration_blockers;
1301
fe44dc91 1302int migrate_add_blocker(Error *reason, Error **errp)
fa2756b7 1303{
3df663e5 1304 if (migrate_get_current()->only_migratable) {
b67b8c3a
AA		 1305 error_propagate(errp, error_copy(reason));
1306 error_prepend(errp, "disallowing migration blocker "
1307 "(--only_migratable) for: ");
1308 return -EACCES;
1309 }
1310
fab35005 1311 if (migration_is_idle()) {
fe44dc91
AA		 1312 migration_blockers = g_slist_prepend(migration_blockers, reason);
1313 return 0;
1314 }
1315
1316 error_propagate(errp, error_copy(reason));
1317 error_prepend(errp, "disallowing migration blocker (migration in "
1318 "progress) for: ");
1319 return -EBUSY;
fa2756b7
AL		 1320}
1321
1322void migrate_del_blocker(Error *reason)
1323{
1324 migration_blockers = g_slist_remove(migration_blockers, reason);
1325}
1326
bf1ae1f4
DDAG		 1327void qmp_migrate_incoming(const char *uri, Error **errp)
1328{
1329 Error *local_err = NULL;
4debb5f5 1330 static bool once = true;
bf1ae1f4
DDAG		 1331
1332 if (!deferred_incoming) {
4debb5f5 1333 error_setg(errp, "For use with '-incoming defer'");
bf1ae1f4
DDAG		 1334 return;
1335 }
4debb5f5
DDAG		 1336 if (!once) {
1337 error_setg(errp, "The incoming migration has already been started");
1338 }
bf1ae1f4
DDAG		 1339
1340 qemu_start_incoming_migration(uri, &local_err);
1341
1342 if (local_err) {
1343 error_propagate(errp, local_err);
1344 return;
1345 }
1346
4debb5f5 1347 once = false;
bf1ae1f4
DDAG		 1348}
1349
24f3902b
GK		 1350bool migration_is_blocked(Error **errp)
1351{
1352 if (qemu_savevm_state_blocked(errp)) {
1353 return true;
1354 }
1355
1356 if (migration_blockers) {
250561e1 1357 error_propagate(errp, error_copy(migration_blockers->data));
24f3902b
GK		 1358 return true;
1359 }
1360
1361 return false;
1362}
1363
e1c37d0e
LC		 1364void qmp_migrate(const char *uri, bool has_blk, bool blk,
1365 bool has_inc, bool inc, bool has_detach, bool detach,
1366 Error **errp)
cab30143 1367{
be7059cd 1368 Error *local_err = NULL;
17549e84 1369 MigrationState *s = migrate_get_current();
cab30143 1370 const char *p;
cab30143 1371
f6844b99 1372 if (migration_is_setup_or_active(s->state) ||
0b827d5e
HZ		 1373 s->state == MIGRATION_STATUS_CANCELLING ||
1374 s->state == MIGRATION_STATUS_COLO) {
c6bd8c70 1375 error_setg(errp, QERR_MIGRATION_ACTIVE);
e1c37d0e 1376 return;
cab30143 1377 }
ca99993a
DDAG		 1378 if (runstate_check(RUN_STATE_INMIGRATE)) {
1379 error_setg(errp, "Guest is waiting for an incoming migration");
1380 return;
1381 }
1382
24f3902b 1383 if (migration_is_blocked(errp)) {
e1c37d0e 1384 return;
fa2756b7
AL		 1385 }
1386
2833c59b
JQ		 1387 if ((has_blk && blk) || (has_inc && inc)) {
1388 if (migrate_use_block() || migrate_use_block_incremental()) {
1389 error_setg(errp, "Command options are incompatible with "
1390 "current migration capabilities");
1391 return;
1392 }
1393 migrate_set_block_enabled(true, &local_err);
1394 if (local_err) {
1395 error_propagate(errp, local_err);
1396 return;
1397 }
1398 s->must_remove_block_options = true;
1399 }
1400
1401 if (has_inc && inc) {
1402 migrate_set_block_incremental(s, true);
1403 }
1404
a0762d9e 1405 s = migrate_init();
cab30143
JQ		 1406
1407 if (strstart(uri, "tcp:", &p)) {
f37afb5a 1408 tcp_start_outgoing_migration(s, p, &local_err);
2da776db 1409#ifdef CONFIG_RDMA
41310c68 1410 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 1411 rdma_start_outgoing_migration(s, p, &local_err);
1412#endif
cab30143 1413 } else if (strstart(uri, "exec:", &p)) {
f37afb5a 1414 exec_start_outgoing_migration(s, p, &local_err);
cab30143 1415 } else if (strstart(uri, "unix:", &p)) {
f37afb5a 1416 unix_start_outgoing_migration(s, p, &local_err);
cab30143 1417 } else if (strstart(uri, "fd:", &p)) {
f37afb5a 1418 fd_start_outgoing_migration(s, p, &local_err);
99a0db9b 1419 } else {
c6bd8c70
MA		 1420 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1421 "a valid migration protocol");
48781e5b
HZ		 1422 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1423 MIGRATION_STATUS_FAILED);
e1c37d0e 1424 return;
cab30143
JQ		 1425 }
1426
f37afb5a 1427 if (local_err) {
d59ce6f3 1428 migrate_fd_error(s, local_err);
f37afb5a 1429 error_propagate(errp, local_err);
e1c37d0e 1430 return;
1299c631 1431 }
cab30143
JQ		 1432}
1433
6cdedb07 1434void qmp_migrate_cancel(Error **errp)
cab30143 1435{
17549e84 1436 migrate_fd_cancel(migrate_get_current());
cab30143
JQ		 1437}
1438
89cfc02c
DDAG		 1439void qmp_migrate_continue(MigrationStatus state, Error **errp)
1440{
1441 MigrationState *s = migrate_get_current();
1442 if (s->state != state) {
1443 error_setg(errp, "Migration not in expected state: %s",
1444 MigrationStatus_str(s->state));
1445 return;
1446 }
1447 qemu_sem_post(&s->pause_sem);
1448}
1449
9e1ba4cc
OW		 1450void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1451{
73af8dd8
JQ		 1452 MigrateSetParameters p = {
1453 .has_xbzrle_cache_size = true,
1454 .xbzrle_cache_size = value,
1455 };
c91e681a 1456
73af8dd8 1457 qmp_migrate_set_parameters(&p, errp);
9e1ba4cc
OW		 1458}
1459
1460int64_t qmp_query_migrate_cache_size(Error **errp)
1461{
1462 return migrate_xbzrle_cache_size();
1463}
1464
3dc85383 1465void qmp_migrate_set_speed(int64_t value, Error **errp)
cab30143 1466{
1bda8b3c 1467 MigrateSetParameters p = {
2ff30257
AA		 1468 .has_max_bandwidth = true,
1469 .max_bandwidth = value,
1470 };
cab30143 1471
2ff30257 1472 qmp_migrate_set_parameters(&p, errp);
cab30143
JQ		 1473}
1474
4f0a993b 1475void qmp_migrate_set_downtime(double value, Error **errp)
cab30143 1476{
87c9cc1c
DHB		 1477 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1478 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1479 "the range of 0 to %d seconds",
1480 MAX_MIGRATE_DOWNTIME_SECONDS);
1481 return;
1482 }
1483
2ff30257
AA		 1484 value *= 1000; /* Convert to milliseconds */
1485 value = MAX(0, MIN(INT64_MAX, value));
1486
1bda8b3c 1487 MigrateSetParameters p = {
2ff30257
AA		 1488 .has_downtime_limit = true,
1489 .downtime_limit = value,
1490 };
1491
1492 qmp_migrate_set_parameters(&p, errp);
99a0db9b 1493}
17ad9b35 1494
53f09a10
PB		 1495bool migrate_release_ram(void)
1496{
1497 MigrationState *s;
1498
1499 s = migrate_get_current();
1500
1501 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1502}
1503
53dd370c
DDAG		 1504bool migrate_postcopy_ram(void)
1505{
1506 MigrationState *s;
1507
1508 s = migrate_get_current();
1509
32c3db5b 1510 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
53dd370c
DDAG		 1511}
1512
58110f0a
VSO		 1513bool migrate_postcopy(void)
1514{
1515 return migrate_postcopy_ram();
1516}
1517
bde1e2ec
CV		 1518bool migrate_auto_converge(void)
1519{
1520 MigrationState *s;
1521
1522 s = migrate_get_current();
1523
1524 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1525}
1526
323004a3
PL		 1527bool migrate_zero_blocks(void)
1528{
1529 MigrationState *s;
1530
1531 s = migrate_get_current();
1532
1533 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1534}
1535
31bf06a9
AP		 1536bool migrate_postcopy_blocktime(void)
1537{
1538 MigrationState *s;
1539
1540 s = migrate_get_current();
1541
1542 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
1543}
1544
8706d2d5
LL		 1545bool migrate_use_compression(void)
1546{
dde4e694
LL		 1547 MigrationState *s;
1548
1549 s = migrate_get_current();
1550
1551 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
8706d2d5
LL		 1552}
1553
1554int migrate_compress_level(void)
1555{
1556 MigrationState *s;
1557
1558 s = migrate_get_current();
1559
2594f56d 1560 return s->parameters.compress_level;
8706d2d5
LL		 1561}
1562
1563int migrate_compress_threads(void)
1564{
1565 MigrationState *s;
1566
1567 s = migrate_get_current();
1568
2594f56d 1569 return s->parameters.compress_threads;
8706d2d5
LL		 1570}
1571
3fcb38c2
LL		 1572int migrate_decompress_threads(void)
1573{
1574 MigrationState *s;
1575
1576 s = migrate_get_current();
1577
2594f56d 1578 return s->parameters.decompress_threads;
3fcb38c2
LL		 1579}
1580
b05dc723
JQ		 1581bool migrate_use_events(void)
1582{
1583 MigrationState *s;
1584
1585 s = migrate_get_current();
1586
1587 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1588}
1589
30126bbf
JQ		 1590bool migrate_use_multifd(void)
1591{
1592 MigrationState *s;
1593
1594 s = migrate_get_current();
1595
1596 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1597}
1598
93fbd031
DDAG		 1599bool migrate_pause_before_switchover(void)
1600{
1601 MigrationState *s;
1602
1603 s = migrate_get_current();
1604
1605 return s->enabled_capabilities[
1606 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
1607}
1608
4075fb1c
JQ		 1609int migrate_multifd_channels(void)
1610{
1611 MigrationState *s;
1612
1613 s = migrate_get_current();
1614
1615 return s->parameters.x_multifd_channels;
1616}
1617
0fb86605
JQ		 1618int migrate_multifd_page_count(void)
1619{
1620 MigrationState *s;
1621
1622 s = migrate_get_current();
1623
1624 return s->parameters.x_multifd_page_count;
1625}
1626
17ad9b35
OW		 1627int migrate_use_xbzrle(void)
1628{
1629 MigrationState *s;
1630
1631 s = migrate_get_current();
1632
1633 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1634}
1635
1636int64_t migrate_xbzrle_cache_size(void)
1637{
1638 MigrationState *s;
1639
1640 s = migrate_get_current();
1641
73af8dd8 1642 return s->parameters.xbzrle_cache_size;
17ad9b35 1643}
0d82d0e8 1644
2833c59b
JQ		 1645bool migrate_use_block(void)
1646{
1647 MigrationState *s;
1648
1649 s = migrate_get_current();
1650
1651 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1652}
1653
c788ada8
PX		 1654bool migrate_use_return_path(void)
1655{
1656 MigrationState *s;
1657
1658 s = migrate_get_current();
1659
1660 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1661}
1662
2833c59b
JQ		 1663bool migrate_use_block_incremental(void)
1664{
1665 MigrationState *s;
1666
1667 s = migrate_get_current();
1668
1669 return s->parameters.block_incremental;
1670}
1671
70b20477
DDAG		 1672/* migration thread support */
1673/*
1674 * Something bad happened to the RP stream, mark an error
1675 * The caller shall print or trace something to indicate why
1676 */
1677static void mark_source_rp_bad(MigrationState *s)
1678{
1679 s->rp_state.error = true;
1680}
1681
1682static struct rp_cmd_args {
1683 ssize_t len; /* -1 = variable */
1684 const char *name;
1685} rp_cmd_args[] = {
1686 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1687 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1688 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1e2d90eb
DDAG		 1689 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1690 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
70b20477
DDAG		 1691 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1692};
1693
1e2d90eb
DDAG		 1694/*
1695 * Process a request for pages received on the return path,
1696 * We're allowed to send more than requested (e.g. to round to our page size)
1697 * and we don't need to send pages that have already been sent.
1698 */
1699static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1700 ram_addr_t start, size_t len)
1701{
6c595cde
DDAG		 1702 long our_host_ps = getpagesize();
1703
1e2d90eb 1704 trace_migrate_handle_rp_req_pages(rbname, start, len);
6c595cde
DDAG		 1705
1706 /*
1707 * Since we currently insist on matching page sizes, just sanity check
1708 * we're being asked for whole host pages.
1709 */
1710 if (start & (our_host_ps-1) ||
1711 (len & (our_host_ps-1))) {
1712 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1713 " len: %zd", __func__, start, len);
1714 mark_source_rp_bad(ms);
1715 return;
1716 }
1717
96506894 1718 if (ram_save_queue_pages(rbname, start, len)) {
6c595cde
DDAG		 1719 mark_source_rp_bad(ms);
1720 }
1e2d90eb
DDAG		 1721}
1722
70b20477
DDAG		 1723/*
1724 * Handles messages sent on the return path towards the source VM
1725 *
1726 */
1727static void *source_return_path_thread(void *opaque)
1728{
1729 MigrationState *ms = opaque;
1730 QEMUFile *rp = ms->rp_state.from_dst_file;
1731 uint16_t header_len, header_type;
568b01ca 1732 uint8_t buf[512];
70b20477 1733 uint32_t tmp32, sibling_error;
1e2d90eb
DDAG		 1734 ram_addr_t start = 0; /* =0 to silence warning */
1735 size_t len = 0, expected_len;
70b20477
DDAG		 1736 int res;
1737
1738 trace_source_return_path_thread_entry();
1739 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1740 migration_is_setup_or_active(ms->state)) {
1741 trace_source_return_path_thread_loop_top();
1742 header_type = qemu_get_be16(rp);
1743 header_len = qemu_get_be16(rp);
1744
1745 if (header_type >= MIG_RP_MSG_MAX ||
1746 header_type == MIG_RP_MSG_INVALID) {
1747 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1748 header_type, header_len);
1749 mark_source_rp_bad(ms);
1750 goto out;
1751 }
1752
1753 if ((rp_cmd_args[header_type].len != -1 &&
1754 header_len != rp_cmd_args[header_type].len) ||
568b01ca 1755 header_len > sizeof(buf)) {
70b20477
DDAG		 1756 error_report("RP: Received '%s' message (0x%04x) with"
1757 "incorrect length %d expecting %zu",
1758 rp_cmd_args[header_type].name, header_type, header_len,
1759 (size_t)rp_cmd_args[header_type].len);
1760 mark_source_rp_bad(ms);
1761 goto out;
1762 }
1763
1764 /* We know we've got a valid header by this point */
1765 res = qemu_get_buffer(rp, buf, header_len);
1766 if (res != header_len) {
1767 error_report("RP: Failed reading data for message 0x%04x"
1768 " read %d expected %d",
1769 header_type, res, header_len);
1770 mark_source_rp_bad(ms);
1771 goto out;
1772 }
1773
1774 /* OK, we have the message and the data */
1775 switch (header_type) {
1776 case MIG_RP_MSG_SHUT:
4d885131 1777 sibling_error = ldl_be_p(buf);
70b20477
DDAG		 1778 trace_source_return_path_thread_shut(sibling_error);
1779 if (sibling_error) {
1780 error_report("RP: Sibling indicated error %d", sibling_error);
1781 mark_source_rp_bad(ms);
1782 }
1783 /*
1784 * We'll let the main thread deal with closing the RP
1785 * we could do a shutdown(2) on it, but we're the only user
1786 * anyway, so there's nothing gained.
1787 */
1788 goto out;
1789
1790 case MIG_RP_MSG_PONG:
4d885131 1791 tmp32 = ldl_be_p(buf);
70b20477
DDAG		 1792 trace_source_return_path_thread_pong(tmp32);
1793 break;
1794
1e2d90eb 1795 case MIG_RP_MSG_REQ_PAGES:
4d885131
PM		 1796 start = ldq_be_p(buf);
1797 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 1798 migrate_handle_rp_req_pages(ms, NULL, start, len);
1799 break;
1800
1801 case MIG_RP_MSG_REQ_PAGES_ID:
1802 expected_len = 12 + 1; /* header + termination */
1803
1804 if (header_len >= expected_len) {
4d885131
PM		 1805 start = ldq_be_p(buf);
1806 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 1807 /* Now we expect an idstr */
1808 tmp32 = buf[12]; /* Length of the following idstr */
1809 buf[13 + tmp32] = '\0';
1810 expected_len += tmp32;
1811 }
1812 if (header_len != expected_len) {
1813 error_report("RP: Req_Page_id with length %d expecting %zd",
1814 header_len, expected_len);
1815 mark_source_rp_bad(ms);
1816 goto out;
1817 }
1818 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1819 break;
1820
70b20477
DDAG		 1821 default:
1822 break;
1823 }
1824 }
5df5416e 1825 if (qemu_file_get_error(rp)) {
70b20477
DDAG		 1826 trace_source_return_path_thread_bad_end();
1827 mark_source_rp_bad(ms);
1828 }
1829
1830 trace_source_return_path_thread_end();
1831out:
1832 ms->rp_state.from_dst_file = NULL;
1833 qemu_fclose(rp);
1834 return NULL;
1835}
1836
70b20477
DDAG		 1837static int open_return_path_on_source(MigrationState *ms)
1838{
1839
89a02a9f 1840 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
70b20477
DDAG		 1841 if (!ms->rp_state.from_dst_file) {
1842 return -1;
1843 }
1844
1845 trace_open_return_path_on_source();
1846 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1847 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1848
1849 trace_open_return_path_on_source_continue();
1850
1851 return 0;
1852}
1853
70b20477
DDAG		 1854/* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1855static int await_return_path_close_on_source(MigrationState *ms)
1856{
1857 /*
1858 * If this is a normal exit then the destination will send a SHUT and the
1859 * rp_thread will exit, however if there's an error we need to cause
1860 * it to exit.
1861 */
89a02a9f 1862 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
70b20477
DDAG		 1863 /*
1864 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1865 * waiting for the destination.
1866 */
1867 qemu_file_shutdown(ms->rp_state.from_dst_file);
1868 mark_source_rp_bad(ms);
1869 }
1870 trace_await_return_path_close_on_source_joining();
1871 qemu_thread_join(&ms->rp_state.rp_thread);
1872 trace_await_return_path_close_on_source_close();
1873 return ms->rp_state.error;
1874}
1875
1d34e4bf
DDAG		 1876/*
1877 * Switch from normal iteration to postcopy
1878 * Returns non-0 on error
1879 */
1880static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1881{
1882 int ret;
61b67d47
DB		 1883 QIOChannelBuffer *bioc;
1884 QEMUFile *fb;
1d34e4bf 1885 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
ef8d6488 1886 bool restart_block = false;
0331c8ca
DDAG		 1887 int cur_state = MIGRATION_STATUS_ACTIVE;
1888 if (!migrate_pause_before_switchover()) {
1889 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1890 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1891 }
1d34e4bf
DDAG		 1892
1893 trace_postcopy_start();
1894 qemu_mutex_lock_iothread();
1895 trace_postcopy_start_set_run();
1896
1897 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1898 *old_vm_running = runstate_is_running();
1899 global_state_store();
1900 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
76b1c7fe
KW		 1901 if (ret < 0) {
1902 goto fail;
1903 }
1d34e4bf 1904
0331c8ca
DDAG		 1905 ret = migration_maybe_pause(ms, &cur_state,
1906 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1907 if (ret < 0) {
1908 goto fail;
1909 }
1910
76b1c7fe 1911 ret = bdrv_inactivate_all();
1d34e4bf
DDAG		 1912 if (ret < 0) {
1913 goto fail;
1914 }
ef8d6488 1915 restart_block = true;
1d34e4bf 1916
1c0d249d
DDAG		 1917 /*
1918 * Cause any non-postcopiable, but iterative devices to
1919 * send out their final data.
1920 */
a1fbe750 1921 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1c0d249d 1922
1d34e4bf
DDAG		 1923 /*
1924 * in Finish migrate and with the io-lock held everything should
1925 * be quiet, but we've potentially still got dirty pages and we
1926 * need to tell the destination to throw any pages it's already received
1927 * that are dirty
1928 */
58110f0a
VSO		 1929 if (migrate_postcopy_ram()) {
1930 if (ram_postcopy_send_discard_bitmap(ms)) {
1931 error_report("postcopy send discard bitmap failed");
1932 goto fail;
1933 }
1d34e4bf
DDAG		 1934 }
1935
1936 /*
1937 * send rest of state - note things that are doing postcopy
1938 * will notice we're in POSTCOPY_ACTIVE and not actually
1939 * wrap their state up here
1940 */
89a02a9f 1941 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
58110f0a
VSO		 1942 if (migrate_postcopy_ram()) {
1943 /* Ping just for debugging, helps line traces up */
1944 qemu_savevm_send_ping(ms->to_dst_file, 2);
1945 }
1d34e4bf
DDAG		 1946
1947 /*
1948 * While loading the device state we may trigger page transfer
1949 * requests and the fd must be free to process those, and thus
1950 * the destination must read the whole device state off the fd before
1951 * it starts processing it. Unfortunately the ad-hoc migration format
1952 * doesn't allow the destination to know the size to read without fully
1953 * parsing it through each devices load-state code (especially the open
1954 * coded devices that use get/put).
1955 * So we wrap the device state up in a package with a length at the start;
1956 * to do this we use a qemu_buf to hold the whole of the device state.
1957 */
61b67d47 1958 bioc = qio_channel_buffer_new(4096);
6f01f136 1959 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
61b67d47
DB		 1960 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1961 object_unref(OBJECT(bioc));
1d34e4bf 1962
c76201ab
DDAG		 1963 /*
1964 * Make sure the receiver can get incoming pages before we send the rest
1965 * of the state
1966 */
1967 qemu_savevm_send_postcopy_listen(fb);
1968
a1fbe750 1969 qemu_savevm_state_complete_precopy(fb, false, false);
58110f0a
VSO		 1970 if (migrate_postcopy_ram()) {
1971 qemu_savevm_send_ping(fb, 3);
1972 }
1d34e4bf
DDAG		 1973
1974 qemu_savevm_send_postcopy_run(fb);
1975
1976 /* <><> end of stuff going into the package */
1d34e4bf 1977
ef8d6488
DDAG		 1978 /* Last point of recovery; as soon as we send the package the destination
1979 * can open devices and potentially start running.
1980 * Lets just check again we've not got any errors.
1981 */
1982 ret = qemu_file_get_error(ms->to_dst_file);
1983 if (ret) {
1984 error_report("postcopy_start: Migration stream errored (pre package)");
1985 goto fail_closefb;
1986 }
1987
1988 restart_block = false;
1989
1d34e4bf 1990 /* Now send that blob */
61b67d47 1991 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1d34e4bf
DDAG		 1992 goto fail_closefb;
1993 }
1994 qemu_fclose(fb);
b82fc321
DDAG		 1995
1996 /* Send a notify to give a chance for anything that needs to happen
1997 * at the transition to postcopy and after the device state; in particular
1998 * spice needs to trigger a transition now
1999 */
2000 ms->postcopy_after_devices = true;
2001 notifier_list_notify(&migration_state_notifiers, ms);
2002
1d34e4bf
DDAG		 2003 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2004
2005 qemu_mutex_unlock_iothread();
2006
58110f0a
VSO		 2007 if (migrate_postcopy_ram()) {
2008 /*
2009 * Although this ping is just for debug, it could potentially be
2010 * used for getting a better measurement of downtime at the source.
2011 */
2012 qemu_savevm_send_ping(ms->to_dst_file, 4);
2013 }
1d34e4bf 2014
ced1c616
PB		 2015 if (migrate_release_ram()) {
2016 ram_postcopy_migrated_memory_release(ms);
2017 }
2018
89a02a9f 2019 ret = qemu_file_get_error(ms->to_dst_file);
1d34e4bf
DDAG		 2020 if (ret) {
2021 error_report("postcopy_start: Migration stream errored");
48781e5b 2022 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf
DDAG		 2023 MIGRATION_STATUS_FAILED);
2024 }
2025
2026 return ret;
2027
2028fail_closefb:
2029 qemu_fclose(fb);
2030fail:
48781e5b 2031 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf 2032 MIGRATION_STATUS_FAILED);
ef8d6488
DDAG		 2033 if (restart_block) {
2034 /* A failure happened early enough that we know the destination hasn't
2035 * accessed block devices, so we're safe to recover.
2036 */
2037 Error *local_err = NULL;
2038
2039 bdrv_invalidate_cache_all(&local_err);
2040 if (local_err) {
2041 error_report_err(local_err);
2042 }
2043 }
1d34e4bf
DDAG		 2044 qemu_mutex_unlock_iothread();
2045 return -1;
2046}
2047
e91d8951
DDAG		 2048/**
2049 * migration_maybe_pause: Pause if required to by
2050 * migrate_pause_before_switchover called with the iothread locked
2051 * Returns: 0 on success
2052 */
0331c8ca
DDAG		 2053static int migration_maybe_pause(MigrationState *s,
2054 int *current_active_state,
2055 int new_state)
e91d8951
DDAG		 2056{
2057 if (!migrate_pause_before_switchover()) {
2058 return 0;
2059 }
2060
2061 /* Since leaving this state is not atomic with posting the semaphore
2062 * it's possible that someone could have issued multiple migrate_continue
2063 * and the semaphore is incorrectly positive at this point;
2064 * the docs say it's undefined to reinit a semaphore that's already
2065 * init'd, so use timedwait to eat up any existing posts.
2066 */
2067 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2068 /* This block intentionally left blank */
2069 }
2070
2071 qemu_mutex_unlock_iothread();
2072 migrate_set_state(&s->state, *current_active_state,
2073 MIGRATION_STATUS_PRE_SWITCHOVER);
2074 qemu_sem_wait(&s->pause_sem);
2075 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
0331c8ca
DDAG		 2076 new_state);
2077 *current_active_state = new_state;
e91d8951
DDAG		 2078 qemu_mutex_lock_iothread();
2079
0331c8ca 2080 return s->state == new_state ? 0 : -EINVAL;
e91d8951
DDAG		 2081}
2082
09f6c85e
DDAG		 2083/**
2084 * migration_completion: Used by migration_thread when there's not much left.
2085 * The caller 'breaks' the loop when this returns.
2086 *
2087 * @s: Current migration state
36f48567 2088 * @current_active_state: The migration state we expect to be in
09f6c85e
DDAG		 2089 * @*old_vm_running: Pointer to old_vm_running flag
2090 * @*start_time: Pointer to time to update
2091 */
36f48567
DDAG		 2092static void migration_completion(MigrationState *s, int current_active_state,
2093 bool *old_vm_running,
09f6c85e
DDAG		 2094 int64_t *start_time)
2095{
2096 int ret;
2097
b10ac0c4
DDAG		 2098 if (s->state == MIGRATION_STATUS_ACTIVE) {
2099 qemu_mutex_lock_iothread();
2100 *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2101 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
2102 *old_vm_running = runstate_is_running();
2103 ret = global_state_store();
2104
2105 if (!ret) {
a1fbe750 2106 bool inactivate = !migrate_colo_enabled();
b10ac0c4 2107 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
e91d8951 2108 if (ret >= 0) {
0331c8ca
DDAG		 2109 ret = migration_maybe_pause(s, &current_active_state,
2110 MIGRATION_STATUS_DEVICE);
e91d8951 2111 }
f07fa4cb
KW		 2112 if (ret >= 0) {
2113 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
a1fbe750
FZ		 2114 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2115 inactivate);
f07fa4cb 2116 }
a1fbe750
FZ		 2117 if (inactivate && ret >= 0) {
2118 s->block_inactive = true;
b10ac0c4
DDAG		 2119 }
2120 }
2121 qemu_mutex_unlock_iothread();
09f6c85e 2122
b10ac0c4
DDAG		 2123 if (ret < 0) {
2124 goto fail;
09f6c85e 2125 }
b10ac0c4
DDAG		 2126 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2127 trace_migration_completion_postcopy_end();
2128
89a02a9f 2129 qemu_savevm_state_complete_postcopy(s->to_dst_file);
b10ac0c4 2130 trace_migration_completion_postcopy_end_after_complete();
09f6c85e 2131 }
09f6c85e 2132
b10ac0c4
DDAG		 2133 /*
2134 * If rp was opened we must clean up the thread before
2135 * cleaning everything else up (since if there are no failures
2136 * it will wait for the destination to send it's status in
2137 * a SHUT command).
b10ac0c4 2138 */
0425dc97 2139 if (s->rp_state.from_dst_file) {
b10ac0c4 2140 int rp_error;
0425dc97 2141 trace_migration_return_path_end_before();
b10ac0c4 2142 rp_error = await_return_path_close_on_source(s);
0425dc97 2143 trace_migration_return_path_end_after(rp_error);
b10ac0c4 2144 if (rp_error) {
fe904ea8 2145 goto fail_invalidate;
b10ac0c4 2146 }
09f6c85e
DDAG		 2147 }
2148
89a02a9f 2149 if (qemu_file_get_error(s->to_dst_file)) {
09f6c85e 2150 trace_migration_completion_file_err();
fe904ea8 2151 goto fail_invalidate;
09f6c85e
DDAG		 2152 }
2153
0b827d5e
HZ		 2154 if (!migrate_colo_enabled()) {
2155 migrate_set_state(&s->state, current_active_state,
2156 MIGRATION_STATUS_COMPLETED);
2157 }
2158
09f6c85e
DDAG		 2159 return;
2160
fe904ea8
GK		 2161fail_invalidate:
2162 /* If not doing postcopy, vm_start() will be called: let's regain
2163 * control on images.
2164 */
2165 if (s->state == MIGRATION_STATUS_ACTIVE) {
2166 Error *local_err = NULL;
2167
1d2acc31 2168 qemu_mutex_lock_iothread();
fe904ea8
GK		 2169 bdrv_invalidate_cache_all(&local_err);
2170 if (local_err) {
2171 error_report_err(local_err);
1d2acc31
HZ		 2172 } else {
2173 s->block_inactive = false;
fe904ea8 2174 }
1d2acc31 2175 qemu_mutex_unlock_iothread();
fe904ea8
GK		 2176 }
2177
09f6c85e 2178fail:
48781e5b
HZ		 2179 migrate_set_state(&s->state, current_active_state,
2180 MIGRATION_STATUS_FAILED);
09f6c85e
DDAG		 2181}
2182
35a6ed4f
HZ		 2183bool migrate_colo_enabled(void)
2184{
2185 MigrationState *s = migrate_get_current();
2186 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2187}
2188
70b20477
DDAG		 2189/*
2190 * Master migration thread on the source VM.
2191 * It drives the migration and pumps the data down the outgoing channel.
2192 */
5f496a1b 2193static void *migration_thread(void *opaque)
0d82d0e8 2194{
9848a404 2195 MigrationState *s = opaque;
1d34e4bf 2196 /* Used by the bandwidth calcs, updated later */
bc72ad67
AB		 2197 int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2198 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
be7172e2 2199 int64_t initial_bytes = 0;
faec066a
PX		 2200 /*
2201 * The final stage happens when the remaining data is smaller than
2202 * this threshold; it's calculated from the requested downtime and
2203 * measured bandwidth
2204 */
2205 int64_t threshold_size = 0;
a3fa1d78 2206 int64_t start_time = initial_time;
94f5a437 2207 int64_t end_time;
a3fa1d78 2208 bool old_vm_running = false;
1d34e4bf
DDAG		 2209 bool entered_postcopy = false;
2210 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
2211 enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
0b827d5e 2212 bool enable_colo = migrate_colo_enabled();
76f5933a 2213
ab28bd23
PB		 2214 rcu_register_thread();
2215
89a02a9f 2216 qemu_savevm_state_header(s->to_dst_file);
1d34e4bf 2217
62a02658
PX		 2218 /*
2219 * If we opened the return path, we need to make sure dst has it
2220 * opened as well.
2221 */
2222 if (s->rp_state.from_dst_file) {
1d34e4bf 2223 /* Now tell the dest that it should open its end so it can reply */
89a02a9f 2224 qemu_savevm_send_open_return_path(s->to_dst_file);
1d34e4bf
DDAG		 2225
2226 /* And do a ping that will make stuff easier to debug */
89a02a9f 2227 qemu_savevm_send_ping(s->to_dst_file, 1);
0425dc97 2228 }
1d34e4bf 2229
58110f0a 2230 if (migrate_postcopy()) {
1d34e4bf
DDAG		 2231 /*
2232 * Tell the destination that we *might* want to do postcopy later;
2233 * if the other end can't do postcopy it should fail now, nice and
2234 * early.
2235 */
89a02a9f 2236 qemu_savevm_send_postcopy_advise(s->to_dst_file);
1d34e4bf
DDAG		 2237 }
2238
9907e842 2239 qemu_savevm_state_setup(s->to_dst_file);
0d82d0e8 2240
bc72ad67 2241 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
48781e5b
HZ		 2242 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2243 MIGRATION_STATUS_ACTIVE);
29ae8a41 2244
9ec055ae
DDAG		 2245 trace_migration_thread_setup_complete();
2246
2247 while (s->state == MIGRATION_STATUS_ACTIVE ||
2248 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
a3e879cd 2249 int64_t current_time;
c369f40d 2250 uint64_t pending_size;
0d82d0e8 2251
89a02a9f 2252 if (!qemu_file_rate_limit(s->to_dst_file)) {
c31b098f
DDAG		 2253 uint64_t pend_post, pend_nonpost;
2254
faec066a
PX		 2255 qemu_savevm_state_pending(s->to_dst_file, threshold_size,
2256 &pend_nonpost, &pend_post);
c31b098f 2257 pending_size = pend_nonpost + pend_post;
faec066a 2258 trace_migrate_pending(pending_size, threshold_size,
c31b098f 2259 pend_post, pend_nonpost);
faec066a 2260 if (pending_size && pending_size >= threshold_size) {
1d34e4bf
DDAG		 2261 /* Still a significant amount to transfer */
2262
58110f0a 2263 if (migrate_postcopy() &&
1d34e4bf 2264 s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
faec066a 2265 pend_nonpost <= threshold_size &&
1d34e4bf
DDAG		 2266 atomic_read(&s->start_postcopy)) {
2267
2268 if (!postcopy_start(s, &old_vm_running)) {
2269 current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
2270 entered_postcopy = true;
2271 }
2272
2273 continue;
2274 }
2275 /* Just another iteration step */
89a02a9f 2276 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
c369f40d 2277 } else {
09f6c85e 2278 trace_migration_thread_low_pending(pending_size);
1d34e4bf 2279 migration_completion(s, current_active_state,
36f48567 2280 &old_vm_running, &start_time);
09f6c85e 2281 break;
c369f40d
JQ		 2282 }
2283 }
f4410a5d 2284
89a02a9f 2285 if (qemu_file_get_error(s->to_dst_file)) {
48781e5b
HZ		 2286 migrate_set_state(&s->state, current_active_state,
2287 MIGRATION_STATUS_FAILED);
1d34e4bf 2288 trace_migration_thread_file_err();
fd45ee2c
PB		 2289 break;
2290 }
bc72ad67 2291 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
0d82d0e8 2292 if (current_time >= initial_time + BUFFER_DELAY) {
89a02a9f
HZ		 2293 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
2294 initial_bytes;
77417f10 2295 uint64_t time_spent = current_time - initial_time;
a694ee34 2296 double bandwidth = (double)transferred_bytes / time_spent;
faec066a 2297 threshold_size = bandwidth * s->parameters.downtime_limit;
0d82d0e8 2298
5b648de0
WY		 2299 s->mbps = (((double) transferred_bytes * 8.0) /
2300 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
7e114f8c 2301
9013dca5 2302 trace_migrate_transferred(transferred_bytes, time_spent,
faec066a 2303 bandwidth, threshold_size);
90f8ae72
JQ		 2304 /* if we haven't sent anything, we don't want to recalculate
2305 10000 is a small enough number for our purposes */
9360447d
JQ		 2306 if (ram_counters.dirty_pages_rate && transferred_bytes > 10000) {
2307 s->expected_downtime = ram_counters.dirty_pages_rate *
20afaed9 2308 qemu_target_page_size() / bandwidth;
90f8ae72 2309 }
0d82d0e8 2310
89a02a9f 2311 qemu_file_reset_rate_limit(s->to_dst_file);
0d82d0e8 2312 initial_time = current_time;
89a02a9f 2313 initial_bytes = qemu_ftell(s->to_dst_file);
0d82d0e8 2314 }
89a02a9f 2315 if (qemu_file_rate_limit(s->to_dst_file)) {
0d82d0e8
JQ		 2316 /* usleep expects microseconds */
2317 g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
2318 }
a3fa1d78
PB		 2319 }
2320
1d34e4bf 2321 trace_migration_thread_after_loop();
070afca2
JH		 2322 /* If we enabled cpu throttling for auto-converge, turn it off. */
2323 cpu_throttle_stop();
94f5a437 2324 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
070afca2 2325
f4410a5d 2326 qemu_mutex_lock_iothread();
0b827d5e
HZ		 2327 /*
2328 * The resource has been allocated by migration will be reused in COLO
2329 * process, so don't release them.
2330 */
2331 if (!enable_colo) {
2332 qemu_savevm_state_cleanup();
2333 }
31194731 2334 if (s->state == MIGRATION_STATUS_COMPLETED) {
89a02a9f 2335 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
a3fa1d78 2336 s->total_time = end_time - s->total_time;
1d34e4bf
DDAG		 2337 if (!entered_postcopy) {
2338 s->downtime = end_time - start_time;
2339 }
d6ed7312
PL		 2340 if (s->total_time) {
2341 s->mbps = (((double) transferred_bytes * 8.0) /
2342 ((double) s->total_time)) / 1000;
2343 }
a3fa1d78
PB		 2344 runstate_set(RUN_STATE_POSTMIGRATE);
2345 } else {
0b827d5e
HZ		 2346 if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
2347 migrate_start_colo_process(s);
2348 qemu_savevm_state_cleanup();
2349 /*
2350 * Fixme: we will run VM in COLO no matter its old running state.
2351 * After exited COLO, we will keep running.
2352 */
2353 old_vm_running = true;
2354 }
1d34e4bf 2355 if (old_vm_running && !entered_postcopy) {
a3fa1d78 2356 vm_start();
42da5550
DDAG		 2357 } else {
2358 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2359 runstate_set(RUN_STATE_POSTMIGRATE);
2360 }
dba433c0 2361 }
0d82d0e8 2362 }
bb1fadc4 2363 qemu_bh_schedule(s->cleanup_bh);
dba433c0 2364 qemu_mutex_unlock_iothread();
f4410a5d 2365
ab28bd23 2366 rcu_unregister_thread();
0d82d0e8
JQ		 2367 return NULL;
2368}
2369
9848a404 2370void migrate_fd_connect(MigrationState *s)
0d82d0e8 2371{
2ff30257 2372 s->expected_downtime = s->parameters.downtime_limit;
bb1fadc4 2373 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
0d82d0e8 2374
9e4d2b98 2375 qemu_file_set_blocking(s->to_dst_file, true);
89a02a9f 2376 qemu_file_set_rate_limit(s->to_dst_file,
2ff30257 2377 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
442773ce 2378
9287ac27
SH		 2379 /* Notify before starting migration thread */
2380 notifier_list_notify(&migration_state_notifiers, s);
2381
1d34e4bf 2382 /*
c788ada8
PX		 2383 * Open the return path. For postcopy, it is used exclusively. For
2384 * precopy, only if user specified "return-path" capability would
2385 * QEMU uses the return path.
1d34e4bf 2386 */
c788ada8 2387 if (migrate_postcopy_ram() || migrate_use_return_path()) {
1d34e4bf
DDAG		 2388 if (open_return_path_on_source(s)) {
2389 error_report("Unable to open return-path for postcopy");
48781e5b 2390 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1d34e4bf
DDAG		 2391 MIGRATION_STATUS_FAILED);
2392 migrate_fd_cleanup(s);
2393 return;
2394 }
2395 }
2396
f986c3d2
JQ		 2397 if (multifd_save_setup() != 0) {
2398 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2399 MIGRATION_STATUS_FAILED);
2400 migrate_fd_cleanup(s);
2401 return;
2402 }
009fad7f 2403 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
bb1fadc4 2404 QEMU_THREAD_JOINABLE);
1d34e4bf 2405 s->migration_thread_running = true;
0d82d0e8 2406}
093e3c42 2407
9d18af93
PX		 2408void migration_global_dump(Monitor *mon)
2409{
2410 MigrationState *ms = migrate_get_current();
2411
6f0f6428
JQ		 2412 monitor_printf(mon, "globals:\n");
2413 monitor_printf(mon, "store-global-state: %s\n",
2414 ms->store_global_state ? "on" : "off");
2415 monitor_printf(mon, "only-migratable: %s\n",
2416 ms->only_migratable ? "on" : "off");
2417 monitor_printf(mon, "send-configuration: %s\n",
2418 ms->send_configuration ? "on" : "off");
2419 monitor_printf(mon, "send-section-footer: %s\n",
2420 ms->send_section_footer ? "on" : "off");
9d18af93
PX		 2421}
2422
20814758
PX		 2423#define DEFINE_PROP_MIG_CAP(name, x) \
2424 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2425
5272298c
PX		 2426static Property migration_properties[] = {
2427 DEFINE_PROP_BOOL("store-global-state", MigrationState,
2428 store_global_state, true),
3df663e5 2429 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
71dd4c1a
PX		 2430 DEFINE_PROP_BOOL("send-configuration", MigrationState,
2431 send_configuration, true),
15c38503
PX		 2432 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2433 send_section_footer, true),
89632faf
PX		 2434
2435 /* Migration parameters */
741d4086 2436 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
89632faf
PX		 2437 parameters.compress_level,
2438 DEFAULT_MIGRATE_COMPRESS_LEVEL),
741d4086 2439 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
89632faf
PX		 2440 parameters.compress_threads,
2441 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
741d4086 2442 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
89632faf
PX		 2443 parameters.decompress_threads,
2444 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
741d4086 2445 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
89632faf
PX		 2446 parameters.cpu_throttle_initial,
2447 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
741d4086 2448 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
89632faf
PX		 2449 parameters.cpu_throttle_increment,
2450 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
741d4086 2451 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
89632faf 2452 parameters.max_bandwidth, MAX_THROTTLE),
741d4086 2453 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
89632faf
PX		 2454 parameters.downtime_limit,
2455 DEFAULT_MIGRATE_SET_DOWNTIME),
741d4086 2456 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
89632faf
PX		 2457 parameters.x_checkpoint_delay,
2458 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
741d4086 2459 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState,
4075fb1c
JQ		 2460 parameters.x_multifd_channels,
2461 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
741d4086 2462 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState,
0fb86605
JQ		 2463 parameters.x_multifd_page_count,
2464 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
73af8dd8
JQ		 2465 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
2466 parameters.xbzrle_cache_size,
2467 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
20814758
PX		 2468
2469 /* Migration capabilities */
2470 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2471 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2472 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2473 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2474 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2475 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2476 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2477 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2478 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2479 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2480 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
30126bbf 2481 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
20814758 2482
5272298c
PX		 2483 DEFINE_PROP_END_OF_LIST(),
2484};
2485
e5cb7e76
PX		 2486static void migration_class_init(ObjectClass *klass, void *data)
2487{
2488 DeviceClass *dc = DEVICE_CLASS(klass);
2489
2490 dc->user_creatable = false;
5272298c 2491 dc->props = migration_properties;
e5cb7e76
PX		 2492}
2493
b91bf5e4
MAL		 2494static void migration_instance_finalize(Object *obj)
2495{
2496 MigrationState *ms = MIGRATION_OBJ(obj);
2497 MigrationParameters *params = &ms->parameters;
2498
87db1a7d 2499 qemu_mutex_destroy(&ms->error_mutex);
b91bf5e4
MAL		 2500 g_free(params->tls_hostname);
2501 g_free(params->tls_creds);
e91d8951 2502 qemu_sem_destroy(&ms->pause_sem);
b91bf5e4
MAL		 2503}
2504
e5cb7e76
PX		 2505static void migration_instance_init(Object *obj)
2506{
2507 MigrationState *ms = MIGRATION_OBJ(obj);
8b0b29dc 2508 MigrationParameters *params = &ms->parameters;
e5cb7e76
PX		 2509
2510 ms->state = MIGRATION_STATUS_NONE;
e5cb7e76 2511 ms->mbps = -1;
e91d8951 2512 qemu_sem_init(&ms->pause_sem, 0);
87db1a7d 2513 qemu_mutex_init(&ms->error_mutex);
8b0b29dc
PX		 2514
2515 params->tls_hostname = g_strdup("");
2516 params->tls_creds = g_strdup("");
2517
2518 /* Set has_* up only for parameter checks */
2519 params->has_compress_level = true;
2520 params->has_compress_threads = true;
2521 params->has_decompress_threads = true;
2522 params->has_cpu_throttle_initial = true;
2523 params->has_cpu_throttle_increment = true;
2524 params->has_max_bandwidth = true;
2525 params->has_downtime_limit = true;
2526 params->has_x_checkpoint_delay = true;
2527 params->has_block_incremental = true;
4075fb1c 2528 params->has_x_multifd_channels = true;
0fb86605 2529 params->has_x_multifd_page_count = true;
73af8dd8 2530 params->has_xbzrle_cache_size = true;
8b0b29dc
PX		 2531}
2532
2533/*
2534 * Return true if check pass, false otherwise. Error will be put
2535 * inside errp if provided.
2536 */
2537static bool migration_object_check(MigrationState *ms, Error **errp)
2538{
6b19a7d9
PX		 2539 MigrationCapabilityStatusList *head = NULL;
2540 /* Assuming all off */
2541 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2542 int i;
2543
8b0b29dc
PX		 2544 if (!migrate_params_check(&ms->parameters, errp)) {
2545 return false;
2546 }
2547
6b19a7d9
PX		 2548 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2549 if (ms->enabled_capabilities[i]) {
2550 head = migrate_cap_add(head, i, true);
2551 }
2552 }
2553
2554 ret = migrate_caps_check(cap_list, head, errp);
2555
2556 /* It works with head == NULL */
2557 qapi_free_MigrationCapabilityStatusList(head);
2558
2559 return ret;
e5cb7e76
PX		 2560}
2561
2562static const TypeInfo migration_type = {
2563 .name = TYPE_MIGRATION,
01f6e14c 2564 /*
c8d3ff38
PX		 2565 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2566 * not created using qdev_create(), it is not attached to the qdev
2567 * device tree, and it is never realized.
2568 *
2569 * TODO: Make this TYPE_OBJECT once QOM provides something like
2570 * TYPE_DEVICE's "-global" properties.
01f6e14c 2571 */
e5cb7e76
PX		 2572 .parent = TYPE_DEVICE,
2573 .class_init = migration_class_init,
2574 .class_size = sizeof(MigrationClass),
2575 .instance_size = sizeof(MigrationState),
2576 .instance_init = migration_instance_init,
b91bf5e4 2577 .instance_finalize = migration_instance_finalize,
e5cb7e76
PX		 2578};
2579
2580static void register_migration_types(void)
2581{
2582 type_register_static(&migration_type);
2583}
2584
2585type_init(register_migration_types);
QEMU mirror