]> 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
qdict qlist: Make most helper macros functions
[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"
e688df6b 33#include "qapi/error.h"
cc7a8ea7 34#include "qapi/qmp/qerror.h"
15280c36 35#include "qapi/qmp/qnull.h"
ab28bd23 36#include "qemu/rcu.h"
2c9e6fec 37#include "block.h"
be07b0ac 38#include "postcopy-ram.h"
766bd176 39#include "qemu/thread.h"
791e7c82 40#include "qmp-commands.h"
c09e5bb1 41#include "trace.h"
598cd2bd 42#include "qapi-event.h"
51180423 43#include "exec/target_page.h"
61b67d47 44#include "io/channel-buffer.h"
35a6ed4f 45#include "migration/colo.h"
4ffdb337 46#include "hw/boards.h"
9d18af93 47#include "monitor/monitor.h"
065e2813 48
dc325627 49#define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
5bb7910a 50
5b4e1eb7
JQ		 51/* Amount of time to allocate to each "chunk" of bandwidth-throttled
52 * data. */
53#define BUFFER_DELAY 100
54#define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
55
2ff30257
AA		 56/* Time in milliseconds we are allowed to stop the source,
57 * for sending the last part */
58#define DEFAULT_MIGRATE_SET_DOWNTIME 300
59
87c9cc1c
DHB		 60/* Maximum migrate downtime set to 2000 seconds */
61#define MAX_MIGRATE_DOWNTIME_SECONDS 2000
62#define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
63
8706d2d5
LL		 64/* Default compression thread count */
65#define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
3fcb38c2
LL		 66/* Default decompression thread count, usually decompression is at
67 * least 4 times as fast as compression.*/
68#define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
8706d2d5
LL		 69/*0: means nocompress, 1: best speed, ... 9: best compress ratio */
70#define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
1626fee3 71/* Define default autoconverge cpu throttle migration parameters */
d85a31d1
JH		 72#define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
73#define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
8706d2d5 74
17ad9b35 75/* Migration XBZRLE default cache size */
73af8dd8 76#define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
17ad9b35 77
68b53591
HZ		 78/* The delay time (in ms) between two COLO checkpoints
79 * Note: Please change this default value to 10000 when we support hybrid mode.
80 */
81#define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
4075fb1c 82#define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
0fb86605 83#define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
68b53591 84
99a0db9b
GH		 85static NotifierList migration_state_notifiers =
86 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
87
adde220a
DDAG		 88static bool deferred_incoming;
89
da6f1790
JQ		 90/* Messages sent on the return path from destination to source */
91enum mig_rp_message_type {
92 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
93 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
94 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
95
96 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
97 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
98
99 MIG_RP_MSG_MAX
100};
101
17549e84
JQ		 102/* When we add fault tolerance, we could have several
103 migrations at once. For now we don't need to add
104 dynamic creation of migration */
105
e5cb7e76
PX		 106static MigrationState *current_migration;
107
8b0b29dc 108static bool migration_object_check(MigrationState *ms, Error **errp);
0331c8ca
DDAG		 109static int migration_maybe_pause(MigrationState *s,
110 int *current_active_state,
111 int new_state);
8b0b29dc 112
e5cb7e76
PX		 113void migration_object_init(void)
114{
4ffdb337 115 MachineState *ms = MACHINE(qdev_get_machine());
8b0b29dc 116 Error *err = NULL;
4ffdb337 117
e5cb7e76
PX		 118 /* This can only be called once. */
119 assert(!current_migration);
120 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
4ffdb337 121
8b0b29dc
PX		 122 if (!migration_object_check(current_migration, &err)) {
123 error_report_err(err);
124 exit(1);
125 }
126
4ffdb337
PX		 127 /*
128 * We cannot really do this in migration_instance_init() since at
129 * that time global properties are not yet applied, then this
130 * value will be definitely replaced by something else.
131 */
132 if (ms->enforce_config_section) {
133 current_migration->send_configuration = true;
134 }
e5cb7e76
PX		 135}
136
1f895604
VSO		 137void migration_object_finalize(void)
138{
139 object_unref(OBJECT(current_migration));
140}
141
bca7856a 142/* For outgoing */
859bc756 143MigrationState *migrate_get_current(void)
17549e84 144{
e5cb7e76
PX		 145 /* This can only be called after the object created. */
146 assert(current_migration);
147 return current_migration;
17549e84
JQ		 148}
149
bca7856a
DDAG		 150MigrationIncomingState *migration_incoming_get_current(void)
151{
b4b076da
JQ		 152 static bool once;
153 static MigrationIncomingState mis_current;
bca7856a 154
b4b076da
JQ		 155 if (!once) {
156 mis_current.state = MIGRATION_STATUS_NONE;
157 memset(&mis_current, 0, sizeof(MigrationIncomingState));
b4b076da
JQ		 158 qemu_mutex_init(&mis_current.rp_mutex);
159 qemu_event_init(&mis_current.main_thread_load_event, false);
160 once = true;
161 }
162 return &mis_current;
bca7856a
DDAG		 163}
164
165void migration_incoming_state_destroy(void)
166{
b4b076da
JQ		 167 struct MigrationIncomingState *mis = migration_incoming_get_current();
168
3482655b 169 if (mis->to_src_file) {
660819b1
PX		 170 /* Tell source that we are done */
171 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
3482655b
PX		 172 qemu_fclose(mis->to_src_file);
173 mis->to_src_file = NULL;
174 }
175
660819b1
PX		 176 if (mis->from_src_file) {
177 qemu_fclose(mis->from_src_file);
178 mis->from_src_file = NULL;
179 }
180
5089e186 181 qemu_event_reset(&mis->main_thread_load_event);
bca7856a
DDAG		 182}
183
b05dc723
JQ		 184static void migrate_generate_event(int new_state)
185{
186 if (migrate_use_events()) {
187 qapi_event_send_migration(new_state, &error_abort);
b05dc723
JQ		 188 }
189}
190
adde220a
DDAG		 191/*
192 * Called on -incoming with a defer: uri.
193 * The migration can be started later after any parameters have been
194 * changed.
195 */
196static void deferred_incoming_migration(Error **errp)
197{
198 if (deferred_incoming) {
199 error_setg(errp, "Incoming migration already deferred");
200 }
201 deferred_incoming = true;
202}
203
da6f1790
JQ		 204/*
205 * Send a message on the return channel back to the source
206 * of the migration.
207 */
208static void migrate_send_rp_message(MigrationIncomingState *mis,
209 enum mig_rp_message_type message_type,
210 uint16_t len, void *data)
211{
212 trace_migrate_send_rp_message((int)message_type, len);
213 qemu_mutex_lock(&mis->rp_mutex);
214 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
215 qemu_put_be16(mis->to_src_file, len);
216 qemu_put_buffer(mis->to_src_file, data, len);
217 qemu_fflush(mis->to_src_file);
218 qemu_mutex_unlock(&mis->rp_mutex);
219}
220
1e2d90eb
DDAG		 221/* Request a range of pages from the source VM at the given
222 * start address.
223 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
224 * as the last request (a name must have been given previously)
225 * Start: Address offset within the RB
226 * Len: Length in bytes required - must be a multiple of pagesize
227 */
228void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
229 ram_addr_t start, size_t len)
230{
cb8d4c8f 231 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
1e2d90eb
DDAG		 232 size_t msglen = 12; /* start + len */
233
234 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
235 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
236
237 if (rbname) {
238 int rbname_len = strlen(rbname);
239 assert(rbname_len < 256);
240
241 bufc[msglen++] = rbname_len;
242 memcpy(bufc + msglen, rbname, rbname_len);
243 msglen += rbname_len;
244 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
245 } else {
246 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
247 }
248}
249
43eaae28 250void qemu_start_incoming_migration(const char *uri, Error **errp)
5bb7910a 251{
34c9dd8e
AL		 252 const char *p;
253
7cf1fe6d 254 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
adde220a
DDAG		 255 if (!strcmp(uri, "defer")) {
256 deferred_incoming_migration(errp);
257 } else if (strstart(uri, "tcp:", &p)) {
43eaae28 258 tcp_start_incoming_migration(p, errp);
2da776db 259#ifdef CONFIG_RDMA
adde220a 260 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 261 rdma_start_incoming_migration(p, errp);
262#endif
adde220a 263 } else if (strstart(uri, "exec:", &p)) {
43eaae28 264 exec_start_incoming_migration(p, errp);
adde220a 265 } else if (strstart(uri, "unix:", &p)) {
43eaae28 266 unix_start_incoming_migration(p, errp);
adde220a 267 } else if (strstart(uri, "fd:", &p)) {
43eaae28 268 fd_start_incoming_migration(p, errp);
adde220a 269 } else {
312fd5f2 270 error_setg(errp, "unknown migration protocol: %s", uri);
8ca5e801 271 }
5bb7910a
AL		 272}
273
0aa6aefc
DL		 274static void process_incoming_migration_bh(void *opaque)
275{
276 Error *local_err = NULL;
277 MigrationIncomingState *mis = opaque;
278
ace21a58
KW		 279 /* Make sure all file formats flush their mutable metadata.
280 * If we get an error here, just don't restart the VM yet. */
0aa6aefc 281 bdrv_invalidate_cache_all(&local_err);
d35ff5e6 282 if (local_err) {
ace21a58 283 error_report_err(local_err);
d35ff5e6
KW		 284 local_err = NULL;
285 autostart = false;
286 }
287
0aa6aefc
DL		 288 /*
289 * This must happen after all error conditions are dealt with and
290 * we're sure the VM is going to be running on this host.
291 */
292 qemu_announce_self();
293
f986c3d2
JQ		 294 if (multifd_load_cleanup(&local_err) != 0) {
295 error_report_err(local_err);
296 autostart = false;
297 }
0aa6aefc
DL		 298 /* If global state section was not received or we are in running
299 state, we need to obey autostart. Any other state is set with
300 runstate_set. */
301
302 if (!global_state_received() ||
303 global_state_get_runstate() == RUN_STATE_RUNNING) {
304 if (autostart) {
305 vm_start();
306 } else {
307 runstate_set(RUN_STATE_PAUSED);
308 }
309 } else {
310 runstate_set(global_state_get_runstate());
311 }
0aa6aefc
DL		 312 /*
313 * This must happen after any state changes since as soon as an external
314 * observer sees this event they might start to prod at the VM assuming
315 * it's ready to use.
316 */
317 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
318 MIGRATION_STATUS_COMPLETED);
319 qemu_bh_delete(mis->bh);
320 migration_incoming_state_destroy();
321}
322
82a4da79 323static void process_incoming_migration_co(void *opaque)
511c0231 324{
b4b076da 325 MigrationIncomingState *mis = migration_incoming_get_current();
e9bef235 326 PostcopyState ps;
1c12e1f5
PB		 327 int ret;
328
4f0fae7f 329 assert(mis->from_src_file);
67f11b5c 330 mis->largest_page_size = qemu_ram_pagesize_largest();
093e3c42 331 postcopy_state_set(POSTCOPY_INCOMING_NONE);
93d7af6f
HZ		 332 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
333 MIGRATION_STATUS_ACTIVE);
4f0fae7f 334 ret = qemu_loadvm_state(mis->from_src_file);
bca7856a 335
e9bef235
DDAG		 336 ps = postcopy_state_get();
337 trace_process_incoming_migration_co_end(ret, ps);
338 if (ps != POSTCOPY_INCOMING_NONE) {
339 if (ps == POSTCOPY_INCOMING_ADVISE) {
340 /*
341 * Where a migration had postcopy enabled (and thus went to advise)
342 * but managed to complete within the precopy period, we can use
343 * the normal exit.
344 */
345 postcopy_ram_incoming_cleanup(mis);
346 } else if (ret >= 0) {
347 /*
348 * Postcopy was started, cleanup should happen at the end of the
349 * postcopy thread.
350 */
351 trace_process_incoming_migration_co_postcopy_end_main();
352 return;
353 }
354 /* Else if something went wrong then just fall out of the normal exit */
355 }
356
25d0c16f
HZ		 357 /* we get COLO info, and know if we are in COLO mode */
358 if (!ret && migration_incoming_enable_colo()) {
359 mis->migration_incoming_co = qemu_coroutine_self();
360 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
361 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
362 mis->have_colo_incoming_thread = true;
363 qemu_coroutine_yield();
364
365 /* Wait checkpoint incoming thread exit before free resource */
366 qemu_thread_join(&mis->colo_incoming_thread);
367 }
368
1c12e1f5 369 if (ret < 0) {
f986c3d2
JQ		 370 Error *local_err = NULL;
371
93d7af6f
HZ		 372 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
373 MIGRATION_STATUS_FAILED);
db80face 374 error_report("load of migration failed: %s", strerror(-ret));
3a0f2cea 375 qemu_fclose(mis->from_src_file);
f986c3d2
JQ		 376 if (multifd_load_cleanup(&local_err) != 0) {
377 error_report_err(local_err);
378 }
4aead692 379 exit(EXIT_FAILURE);
511c0231 380 }
0aa6aefc
DL		 381 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
382 qemu_bh_schedule(mis->bh);
511c0231
JQ		 383}
384
e595a01a 385static void migration_incoming_setup(QEMUFile *f)
82a4da79 386{
4f0fae7f 387 MigrationIncomingState *mis = migration_incoming_get_current();
82a4da79 388
f986c3d2
JQ		 389 if (multifd_load_setup() != 0) {
390 /* We haven't been able to create multifd threads
391 nothing better to do */
392 exit(EXIT_FAILURE);
393 }
394
4f0fae7f
JQ		 395 if (!mis->from_src_file) {
396 mis->from_src_file = f;
397 }
06ad5135 398 qemu_file_set_blocking(f, false);
e595a01a
JQ		 399}
400
401static void migration_incoming_process(void)
402{
403 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
0b8b8753 404 qemu_coroutine_enter(co);
82a4da79
PB		 405}
406
e595a01a
JQ		 407void migration_fd_process_incoming(QEMUFile *f)
408{
409 migration_incoming_setup(f);
410 migration_incoming_process();
411}
412
4f0fae7f
JQ		 413void migration_ioc_process_incoming(QIOChannel *ioc)
414{
415 MigrationIncomingState *mis = migration_incoming_get_current();
416
417 if (!mis->from_src_file) {
418 QEMUFile *f = qemu_fopen_channel_input(ioc);
419 migration_fd_process_incoming(f);
420 }
421 /* We still only have a single channel. Nothing to do here yet */
422}
423
428d8908
JQ		 424/**
425 * @migration_has_all_channels: We have received all channels that we need
426 *
427 * Returns true when we have got connections to all the channels that
428 * we need for migration.
429 */
430bool migration_has_all_channels(void)
431{
432 return true;
433}
434
6decec93
DDAG		 435/*
436 * Send a 'SHUT' message on the return channel with the given value
437 * to indicate that we've finished with the RP. Non-0 value indicates
438 * error.
439 */
440void migrate_send_rp_shut(MigrationIncomingState *mis,
441 uint32_t value)
442{
443 uint32_t buf;
444
445 buf = cpu_to_be32(value);
446 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
447}
448
449/*
450 * Send a 'PONG' message on the return channel with the given value
451 * (normally in response to a 'PING')
452 */
453void migrate_send_rp_pong(MigrationIncomingState *mis,
454 uint32_t value)
455{
456 uint32_t buf;
457
458 buf = cpu_to_be32(value);
459 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
460}
461
bbf6da32
OW		 462MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
463{
464 MigrationCapabilityStatusList *head = NULL;
465 MigrationCapabilityStatusList *caps;
466 MigrationState *s = migrate_get_current();
467 int i;
468
387eedeb 469 caps = NULL; /* silence compiler warning */
7fb1cf16 470 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
ed1701c6
DDAG		 471#ifndef CONFIG_LIVE_BLOCK_MIGRATION
472 if (i == MIGRATION_CAPABILITY_BLOCK) {
473 continue;
474 }
475#endif
bbf6da32
OW		 476 if (head == NULL) {
477 head = g_malloc0(sizeof(*caps));
478 caps = head;
479 } else {
480 caps->next = g_malloc0(sizeof(*caps));
481 caps = caps->next;
482 }
483 caps->value =
484 g_malloc(sizeof(*caps->value));
485 caps->value->capability = i;
486 caps->value->state = s->enabled_capabilities[i];
487 }
488
489 return head;
490}
491
85de8323
LL		 492MigrationParameters *qmp_query_migrate_parameters(Error **errp)
493{
494 MigrationParameters *params;
495 MigrationState *s = migrate_get_current();
496
e87fae4c 497 /* TODO use QAPI_CLONE() instead of duplicating it inline */
85de8323 498 params = g_malloc0(sizeof(*params));
de63ab61 499 params->has_compress_level = true;
2594f56d 500 params->compress_level = s->parameters.compress_level;
de63ab61 501 params->has_compress_threads = true;
2594f56d 502 params->compress_threads = s->parameters.compress_threads;
de63ab61 503 params->has_decompress_threads = true;
2594f56d 504 params->decompress_threads = s->parameters.decompress_threads;
de63ab61 505 params->has_cpu_throttle_initial = true;
2594f56d 506 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
de63ab61 507 params->has_cpu_throttle_increment = true;
2594f56d 508 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
8cc99dcd 509 params->has_tls_creds = true;
69ef1f36 510 params->tls_creds = g_strdup(s->parameters.tls_creds);
8cc99dcd 511 params->has_tls_hostname = true;
69ef1f36 512 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
2ff30257
AA		 513 params->has_max_bandwidth = true;
514 params->max_bandwidth = s->parameters.max_bandwidth;
515 params->has_downtime_limit = true;
516 params->downtime_limit = s->parameters.downtime_limit;
fe39a4d4 517 params->has_x_checkpoint_delay = true;
68b53591 518 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
2833c59b
JQ		 519 params->has_block_incremental = true;
520 params->block_incremental = s->parameters.block_incremental;
4075fb1c
JQ		 521 params->has_x_multifd_channels = true;
522 params->x_multifd_channels = s->parameters.x_multifd_channels;
0fb86605
JQ		 523 params->has_x_multifd_page_count = true;
524 params->x_multifd_page_count = s->parameters.x_multifd_page_count;
73af8dd8
JQ		 525 params->has_xbzrle_cache_size = true;
526 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
85de8323
LL		 527
528 return params;
529}
530
f6844b99
DDAG		 531/*
532 * Return true if we're already in the middle of a migration
533 * (i.e. any of the active or setup states)
534 */
535static bool migration_is_setup_or_active(int state)
536{
537 switch (state) {
538 case MIGRATION_STATUS_ACTIVE:
9ec055ae 539 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
f6844b99 540 case MIGRATION_STATUS_SETUP:
31e06077
DDAG		 541 case MIGRATION_STATUS_PRE_SWITCHOVER:
542 case MIGRATION_STATUS_DEVICE:
f6844b99
DDAG		 543 return true;
544
545 default:
546 return false;
547
548 }
549}
550
a22463a5
DDAG		 551static void populate_ram_info(MigrationInfo *info, MigrationState *s)
552{
553 info->has_ram = true;
554 info->ram = g_malloc0(sizeof(*info->ram));
9360447d 555 info->ram->transferred = ram_counters.transferred;
a22463a5 556 info->ram->total = ram_bytes_total();
9360447d 557 info->ram->duplicate = ram_counters.duplicate;
bedf53c1
JQ		 558 /* legacy value. It is not used anymore */
559 info->ram->skipped = 0;
9360447d
JQ		 560 info->ram->normal = ram_counters.normal;
561 info->ram->normal_bytes = ram_counters.normal *
20afaed9 562 qemu_target_page_size();
a22463a5 563 info->ram->mbps = s->mbps;
9360447d
JQ		 564 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
565 info->ram->postcopy_requests = ram_counters.postcopy_requests;
030ce1f8 566 info->ram->page_size = qemu_target_page_size();
a22463a5 567
114f5aee
JQ		 568 if (migrate_use_xbzrle()) {
569 info->has_xbzrle_cache = true;
570 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
571 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
9360447d
JQ		 572 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
573 info->xbzrle_cache->pages = xbzrle_counters.pages;
574 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
575 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
576 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
114f5aee
JQ		 577 }
578
338182c8
JQ		 579 if (cpu_throttle_active()) {
580 info->has_cpu_throttle_percentage = true;
581 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
582 }
583
a22463a5
DDAG		 584 if (s->state != MIGRATION_STATUS_COMPLETED) {
585 info->ram->remaining = ram_bytes_remaining();
9360447d 586 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
a22463a5
DDAG		 587 }
588}
589
930ac04c
JQ		 590static void populate_disk_info(MigrationInfo *info)
591{
592 if (blk_mig_active()) {
593 info->has_disk = true;
594 info->disk = g_malloc0(sizeof(*info->disk));
595 info->disk->transferred = blk_mig_bytes_transferred();
596 info->disk->remaining = blk_mig_bytes_remaining();
597 info->disk->total = blk_mig_bytes_total();
598 }
599}
600
ee86981b 601MigrationInfo *qmp_query_migrate(Error **errp)
5bb7910a 602{
ee86981b 603 MigrationInfo *info = g_malloc0(sizeof(*info));
17549e84
JQ		 604 MigrationState *s = migrate_get_current();
605
606 switch (s->state) {
31194731 607 case MIGRATION_STATUS_NONE:
17549e84
JQ		 608 /* no migration has happened ever */
609 break;
31194731 610 case MIGRATION_STATUS_SETUP:
29ae8a41 611 info->has_status = true;
ed4fbd10 612 info->has_total_time = false;
29ae8a41 613 break;
31194731
HZ		 614 case MIGRATION_STATUS_ACTIVE:
615 case MIGRATION_STATUS_CANCELLING:
9ec055ae 616 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
31e06077
DDAG		 617 case MIGRATION_STATUS_PRE_SWITCHOVER:
618 case MIGRATION_STATUS_DEVICE:
c8f9f4f4 619 /* TODO add some postcopy stats */
9ec055ae
DDAG		 620 info->has_status = true;
621 info->has_total_time = true;
622 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
4af246a3 623 - s->start_time;
9ec055ae
DDAG		 624 info->has_expected_downtime = true;
625 info->expected_downtime = s->expected_downtime;
626 info->has_setup_time = true;
627 info->setup_time = s->setup_time;
628
a22463a5 629 populate_ram_info(info, s);
930ac04c 630 populate_disk_info(info);
17549e84 631 break;
0b827d5e
HZ		 632 case MIGRATION_STATUS_COLO:
633 info->has_status = true;
634 /* TODO: display COLO specific information (checkpoint info etc.) */
635 break;
31194731 636 case MIGRATION_STATUS_COMPLETED:
791e7c82 637 info->has_status = true;
00c14997 638 info->has_total_time = true;
7aa939af 639 info->total_time = s->total_time;
9c5a9fcf
JQ		 640 info->has_downtime = true;
641 info->downtime = s->downtime;
ed4fbd10
MH		 642 info->has_setup_time = true;
643 info->setup_time = s->setup_time;
d5f8a570 644
a22463a5 645 populate_ram_info(info, s);
17549e84 646 break;
31194731 647 case MIGRATION_STATUS_FAILED:
791e7c82 648 info->has_status = true;
d59ce6f3
DB		 649 if (s->error) {
650 info->has_error_desc = true;
651 info->error_desc = g_strdup(error_get_pretty(s->error));
652 }
17549e84 653 break;
31194731 654 case MIGRATION_STATUS_CANCELLED:
791e7c82 655 info->has_status = true;
17549e84 656 break;
5bb7910a 657 }
cde63fbe 658 info->status = s->state;
ee86981b
PM		 659
660 return info;
5bb7910a
AL		 661}
662
4a84214e
PX		 663/**
664 * @migration_caps_check - check capability validity
665 *
666 * @cap_list: old capability list, array of bool
667 * @params: new capabilities to be applied soon
668 * @errp: set *errp if the check failed, with reason
669 *
670 * Returns true if check passed, otherwise false.
671 */
672static bool migrate_caps_check(bool *cap_list,
673 MigrationCapabilityStatusList *params,
674 Error **errp)
00458433 675{
00458433 676 MigrationCapabilityStatusList *cap;
4a84214e 677 bool old_postcopy_cap;
d7651f15 678 MigrationIncomingState *mis = migration_incoming_get_current();
00458433 679
4a84214e 680 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
00458433
OW		 681
682 for (cap = params; cap; cap = cap->next) {
4a84214e
PX		 683 cap_list[cap->value->capability] = cap->value->state;
684 }
685
ed1701c6 686#ifndef CONFIG_LIVE_BLOCK_MIGRATION
4a84214e
PX		 687 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
688 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
689 "block migration");
690 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
691 return false;
00458433 692 }
4a84214e 693#endif
53dd370c 694
4a84214e
PX		 695 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
696 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
53dd370c
DDAG		 697 /* The decompression threads asynchronously write into RAM
698 * rather than use the atomic copies needed to avoid
699 * userfaulting. It should be possible to fix the decompression
700 * threads for compatibility in future.
701 */
4a84214e
PX		 702 error_setg(errp, "Postcopy is not currently compatible "
703 "with compression");
704 return false;
53dd370c 705 }
4a84214e 706
096631bd
DDAG		 707 /* This check is reasonably expensive, so only when it's being
708 * set the first time, also it's only the destination that needs
709 * special support.
710 */
711 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
d7651f15 712 !postcopy_ram_supported_by_host(mis)) {
096631bd
DDAG		 713 /* postcopy_ram_supported_by_host will have emitted a more
714 * detailed message
715 */
4a84214e
PX		 716 error_setg(errp, "Postcopy is not supported");
717 return false;
096631bd 718 }
53dd370c 719 }
4a84214e
PX		 720
721 return true;
722}
723
724void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
725 Error **errp)
726{
727 MigrationState *s = migrate_get_current();
728 MigrationCapabilityStatusList *cap;
729
730 if (migration_is_setup_or_active(s->state)) {
731 error_setg(errp, QERR_MIGRATION_ACTIVE);
732 return;
733 }
734
735 if (!migrate_caps_check(s->enabled_capabilities, params, errp)) {
736 return;
737 }
738
739 for (cap = params; cap; cap = cap->next) {
740 s->enabled_capabilities[cap->value->capability] = cap->value->state;
741 }
00458433
OW		 742}
743
16d063bc
PX		 744/*
745 * Check whether the parameters are valid. Error will be put into errp
746 * (if provided). Return true if valid, otherwise false.
747 */
748static bool migrate_params_check(MigrationParameters *params, Error **errp)
85de8323 749{
7f375e04 750 if (params->has_compress_level &&
741d4086 751 (params->compress_level > 9)) {
c6bd8c70
MA		 752 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
753 "is invalid, it should be in the range of 0 to 9");
16d063bc 754 return false;
85de8323 755 }
16d063bc 756
741d4086 757 if (params->has_compress_threads && (params->compress_threads < 1)) {
c6bd8c70
MA		 758 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
759 "compress_threads",
760 "is invalid, it should be in the range of 1 to 255");
16d063bc 761 return false;
85de8323 762 }
16d063bc 763
741d4086 764 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
c6bd8c70
MA		 765 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
766 "decompress_threads",
767 "is invalid, it should be in the range of 1 to 255");
16d063bc 768 return false;
85de8323 769 }
16d063bc 770
7f375e04
EB		 771 if (params->has_cpu_throttle_initial &&
772 (params->cpu_throttle_initial < 1 ||
773 params->cpu_throttle_initial > 99)) {
1626fee3 774 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 775 "cpu_throttle_initial",
1626fee3 776 "an integer in the range of 1 to 99");
16d063bc 777 return false;
1626fee3 778 }
16d063bc 779
7f375e04
EB		 780 if (params->has_cpu_throttle_increment &&
781 (params->cpu_throttle_increment < 1 ||
782 params->cpu_throttle_increment > 99)) {
1626fee3 783 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 784 "cpu_throttle_increment",
1626fee3 785 "an integer in the range of 1 to 99");
16d063bc 786 return false;
1626fee3 787 }
16d063bc 788
741d4086 789 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
2ff30257
AA		 790 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
791 " range of 0 to %zu bytes/second", SIZE_MAX);
16d063bc 792 return false;
2ff30257 793 }
16d063bc 794
2ff30257 795 if (params->has_downtime_limit &&
741d4086 796 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
87c9cc1c
DHB		 797 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
798 "the range of 0 to %d milliseconds",
799 MAX_MIGRATE_DOWNTIME);
16d063bc 800 return false;
2ff30257 801 }
16d063bc 802
741d4086
JQ		 803 /* x_checkpoint_delay is now always positive */
804
805 if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) {
4075fb1c
JQ		 806 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
807 "multifd_channels",
808 "is invalid, it should be in the range of 1 to 255");
809 return false;
810 }
0fb86605 811 if (params->has_x_multifd_page_count &&
741d4086
JQ		 812 (params->x_multifd_page_count < 1 ||
813 params->x_multifd_page_count > 10000)) {
0fb86605
JQ		 814 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
815 "multifd_page_count",
816 "is invalid, it should be in the range of 1 to 10000");
817 return false;
818 }
16d063bc 819
73af8dd8
JQ		 820 if (params->has_xbzrle_cache_size &&
821 (params->xbzrle_cache_size < qemu_target_page_size() ||
822 !is_power_of_2(params->xbzrle_cache_size))) {
823 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
824 "xbzrle_cache_size",
825 "is invalid, it should be bigger than target page size"
826 " and a power of two");
827 return false;
828 }
829
16d063bc
PX		 830 return true;
831}
832
1bda8b3c
MA		 833static void migrate_params_test_apply(MigrateSetParameters *params,
834 MigrationParameters *dest)
835{
836 *dest = migrate_get_current()->parameters;
837
838 /* TODO use QAPI_CLONE() instead of duplicating it inline */
839
840 if (params->has_compress_level) {
841 dest->compress_level = params->compress_level;
842 }
843
844 if (params->has_compress_threads) {
845 dest->compress_threads = params->compress_threads;
846 }
847
848 if (params->has_decompress_threads) {
849 dest->decompress_threads = params->decompress_threads;
850 }
851
852 if (params->has_cpu_throttle_initial) {
853 dest->cpu_throttle_initial = params->cpu_throttle_initial;
854 }
855
856 if (params->has_cpu_throttle_increment) {
857 dest->cpu_throttle_increment = params->cpu_throttle_increment;
858 }
859
860 if (params->has_tls_creds) {
01fa5598
MA		 861 assert(params->tls_creds->type == QTYPE_QSTRING);
862 dest->tls_creds = g_strdup(params->tls_creds->u.s);
1bda8b3c
MA		 863 }
864
865 if (params->has_tls_hostname) {
01fa5598
MA		 866 assert(params->tls_hostname->type == QTYPE_QSTRING);
867 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1bda8b3c
MA		 868 }
869
870 if (params->has_max_bandwidth) {
871 dest->max_bandwidth = params->max_bandwidth;
872 }
873
874 if (params->has_downtime_limit) {
875 dest->downtime_limit = params->downtime_limit;
876 }
877
878 if (params->has_x_checkpoint_delay) {
879 dest->x_checkpoint_delay = params->x_checkpoint_delay;
880 }
881
882 if (params->has_block_incremental) {
883 dest->block_incremental = params->block_incremental;
884 }
5e7577a1
JQ		 885 if (params->has_x_multifd_channels) {
886 dest->x_multifd_channels = params->x_multifd_channels;
887 }
888 if (params->has_x_multifd_page_count) {
889 dest->x_multifd_page_count = params->x_multifd_page_count;
890 }
73af8dd8
JQ		 891 if (params->has_xbzrle_cache_size) {
892 dest->xbzrle_cache_size = params->xbzrle_cache_size;
893 }
1bda8b3c
MA		 894}
895
73af8dd8 896static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
16d063bc
PX		 897{
898 MigrationState *s = migrate_get_current();
899
e87fae4c
MA		 900 /* TODO use QAPI_CLONE() instead of duplicating it inline */
901
7f375e04
EB		 902 if (params->has_compress_level) {
903 s->parameters.compress_level = params->compress_level;
85de8323 904 }
476c72aa 905
7f375e04
EB		 906 if (params->has_compress_threads) {
907 s->parameters.compress_threads = params->compress_threads;
85de8323 908 }
476c72aa 909
7f375e04
EB		 910 if (params->has_decompress_threads) {
911 s->parameters.decompress_threads = params->decompress_threads;
85de8323 912 }
476c72aa 913
7f375e04
EB		 914 if (params->has_cpu_throttle_initial) {
915 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1626fee3 916 }
476c72aa 917
7f375e04
EB		 918 if (params->has_cpu_throttle_increment) {
919 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1626fee3 920 }
476c72aa 921
7f375e04 922 if (params->has_tls_creds) {
69ef1f36 923 g_free(s->parameters.tls_creds);
01fa5598
MA		 924 assert(params->tls_creds->type == QTYPE_QSTRING);
925 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
69ef1f36 926 }
476c72aa 927
7f375e04 928 if (params->has_tls_hostname) {
69ef1f36 929 g_free(s->parameters.tls_hostname);
01fa5598
MA		 930 assert(params->tls_hostname->type == QTYPE_QSTRING);
931 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
69ef1f36 932 }
476c72aa 933
2ff30257
AA		 934 if (params->has_max_bandwidth) {
935 s->parameters.max_bandwidth = params->max_bandwidth;
936 if (s->to_dst_file) {
937 qemu_file_set_rate_limit(s->to_dst_file,
938 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
939 }
940 }
476c72aa 941
2ff30257
AA		 942 if (params->has_downtime_limit) {
943 s->parameters.downtime_limit = params->downtime_limit;
944 }
68b53591
HZ		 945
946 if (params->has_x_checkpoint_delay) {
947 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
479125d5
HZ		 948 if (migration_in_colo_state()) {
949 colo_checkpoint_notify(s);
950 }
68b53591 951 }
476c72aa 952
2833c59b
JQ		 953 if (params->has_block_incremental) {
954 s->parameters.block_incremental = params->block_incremental;
955 }
4075fb1c
JQ		 956 if (params->has_x_multifd_channels) {
957 s->parameters.x_multifd_channels = params->x_multifd_channels;
958 }
0fb86605
JQ		 959 if (params->has_x_multifd_page_count) {
960 s->parameters.x_multifd_page_count = params->x_multifd_page_count;
961 }
73af8dd8
JQ		 962 if (params->has_xbzrle_cache_size) {
963 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
964 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
965 }
85de8323
LL		 966}
967
1bda8b3c 968void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
476c72aa 969{
1bda8b3c
MA		 970 MigrationParameters tmp;
971
01fa5598
MA		 972 /* TODO Rewrite "" to null instead */
973 if (params->has_tls_creds
974 && params->tls_creds->type == QTYPE_QNULL) {
975 QDECREF(params->tls_creds->u.n);
976 params->tls_creds->type = QTYPE_QSTRING;
977 params->tls_creds->u.s = strdup("");
978 }
979 /* TODO Rewrite "" to null instead */
980 if (params->has_tls_hostname
981 && params->tls_hostname->type == QTYPE_QNULL) {
982 QDECREF(params->tls_hostname->u.n);
983 params->tls_hostname->type = QTYPE_QSTRING;
984 params->tls_hostname->u.s = strdup("");
985 }
986
1bda8b3c
MA		 987 migrate_params_test_apply(params, &tmp);
988
989 if (!migrate_params_check(&tmp, errp)) {
476c72aa
PX		 990 /* Invalid parameter */
991 return;
992 }
993
73af8dd8 994 migrate_params_apply(params, errp);
476c72aa
PX		 995}
996
2594f56d 997
4886a1bc
DDAG		 998void qmp_migrate_start_postcopy(Error **errp)
999{
1000 MigrationState *s = migrate_get_current();
1001
1002 if (!migrate_postcopy_ram()) {
a54d340b 1003 error_setg(errp, "Enable postcopy with migrate_set_capability before"
4886a1bc
DDAG		 1004 " the start of migration");
1005 return;
1006 }
1007
1008 if (s->state == MIGRATION_STATUS_NONE) {
1009 error_setg(errp, "Postcopy must be started after migration has been"
1010 " started");
1011 return;
1012 }
1013 /*
1014 * we don't error if migration has finished since that would be racy
1015 * with issuing this command.
1016 */
1017 atomic_set(&s->start_postcopy, true);
1018}
1019
065e2813
AL		 1020/* shared migration helpers */
1021
48781e5b 1022void migrate_set_state(int *state, int old_state, int new_state)
51cf4c1a 1023{
a31fedee 1024 assert(new_state < MIGRATION_STATUS__MAX);
48781e5b 1025 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
a31fedee 1026 trace_migrate_set_state(MigrationStatus_str(new_state));
b05dc723 1027 migrate_generate_event(new_state);
51cf4c1a
Z		 1028 }
1029}
1030
4e4a3d3a
PX		 1031static MigrationCapabilityStatusList *migrate_cap_add(
1032 MigrationCapabilityStatusList *list,
1033 MigrationCapability index,
1034 bool state)
2833c59b
JQ		 1035{
1036 MigrationCapabilityStatusList *cap;
1037
1038 cap = g_new0(MigrationCapabilityStatusList, 1);
1039 cap->value = g_new0(MigrationCapabilityStatus, 1);
4e4a3d3a
PX		 1040 cap->value->capability = index;
1041 cap->value->state = state;
1042 cap->next = list;
1043
1044 return cap;
1045}
1046
1047void migrate_set_block_enabled(bool value, Error **errp)
1048{
1049 MigrationCapabilityStatusList *cap;
1050
1051 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
2833c59b
JQ		 1052 qmp_migrate_set_capabilities(cap, errp);
1053 qapi_free_MigrationCapabilityStatusList(cap);
1054}
1055
1056static void migrate_set_block_incremental(MigrationState *s, bool value)
1057{
1058 s->parameters.block_incremental = value;
1059}
1060
1061static void block_cleanup_parameters(MigrationState *s)
1062{
1063 if (s->must_remove_block_options) {
1064 /* setting to false can never fail */
1065 migrate_set_block_enabled(false, &error_abort);
1066 migrate_set_block_incremental(s, false);
1067 s->must_remove_block_options = false;
1068 }
1069}
1070
bb1fadc4 1071static void migrate_fd_cleanup(void *opaque)
065e2813 1072{
bb1fadc4
PB		 1073 MigrationState *s = opaque;
1074
1075 qemu_bh_delete(s->cleanup_bh);
1076 s->cleanup_bh = NULL;
1077
0ceccd85
PX		 1078 qemu_savevm_state_cleanup();
1079
89a02a9f 1080 if (s->to_dst_file) {
f986c3d2
JQ		 1081 Error *local_err = NULL;
1082
9013dca5 1083 trace_migrate_fd_cleanup();
404a7c05 1084 qemu_mutex_unlock_iothread();
1d34e4bf
DDAG		 1085 if (s->migration_thread_running) {
1086 qemu_thread_join(&s->thread);
1087 s->migration_thread_running = false;
1088 }
404a7c05
PB		 1089 qemu_mutex_lock_iothread();
1090
f986c3d2
JQ		 1091 if (multifd_save_cleanup(&local_err) != 0) {
1092 error_report_err(local_err);
1093 }
89a02a9f
HZ		 1094 qemu_fclose(s->to_dst_file);
1095 s->to_dst_file = NULL;
065e2813
AL		 1096 }
1097
9ec055ae
DDAG		 1098 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1099 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
7a2c1721 1100
94f5a437 1101 if (s->state == MIGRATION_STATUS_CANCELLING) {
48781e5b 1102 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
94f5a437 1103 MIGRATION_STATUS_CANCELLED);
7a2c1721 1104 }
a3fa1d78 1105
87db1a7d
JQ		 1106 if (s->error) {
1107 /* It is used on info migrate. We can't free it */
1108 error_report_err(error_copy(s->error));
1109 }
a3fa1d78 1110 notifier_list_notify(&migration_state_notifiers, s);
2833c59b 1111 block_cleanup_parameters(s);
065e2813
AL		 1112}
1113
87db1a7d
JQ		 1114void migrate_set_error(MigrationState *s, const Error *error)
1115{
1116 qemu_mutex_lock(&s->error_mutex);
1117 if (!s->error) {
1118 s->error = error_copy(error);
1119 }
1120 qemu_mutex_unlock(&s->error_mutex);
1121}
1122
d59ce6f3 1123void migrate_fd_error(MigrationState *s, const Error *error)
065e2813 1124{
25174055 1125 trace_migrate_fd_error(error_get_pretty(error));
89a02a9f 1126 assert(s->to_dst_file == NULL);
48781e5b
HZ		 1127 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1128 MIGRATION_STATUS_FAILED);
87db1a7d 1129 migrate_set_error(s, error);
458cf28e
JQ		 1130}
1131
0edda1c4 1132static void migrate_fd_cancel(MigrationState *s)
065e2813 1133{
6f2b811a 1134 int old_state ;
89a02a9f 1135 QEMUFile *f = migrate_get_current()->to_dst_file;
9013dca5 1136 trace_migrate_fd_cancel();
065e2813 1137
70b20477
DDAG		 1138 if (s->rp_state.from_dst_file) {
1139 /* shutdown the rp socket, so causing the rp thread to shutdown */
1140 qemu_file_shutdown(s->rp_state.from_dst_file);
1141 }
1142
6f2b811a
Z		 1143 do {
1144 old_state = s->state;
f6844b99 1145 if (!migration_is_setup_or_active(old_state)) {
6f2b811a
Z		 1146 break;
1147 }
a7b36b48
DDAG		 1148 /* If the migration is paused, kick it out of the pause */
1149 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1150 qemu_sem_post(&s->pause_sem);
1151 }
48781e5b 1152 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
31194731 1153 } while (s->state != MIGRATION_STATUS_CANCELLING);
a26ba26e
DDAG		 1154
1155 /*
1156 * If we're unlucky the migration code might be stuck somewhere in a
1157 * send/write while the network has failed and is waiting to timeout;
1158 * if we've got shutdown(2) available then we can force it to quit.
1159 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1160 * called in a bh, so there is no race against this cancel.
1161 */
31194731 1162 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
a26ba26e
DDAG		 1163 qemu_file_shutdown(f);
1164 }
1d2acc31
HZ		 1165 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1166 Error *local_err = NULL;
1167
1168 bdrv_invalidate_cache_all(&local_err);
1169 if (local_err) {
1170 error_report_err(local_err);
1171 } else {
1172 s->block_inactive = false;
1173 }
1174 }
065e2813
AL		 1175}
1176
99a0db9b
GH		 1177void add_migration_state_change_notifier(Notifier *notify)
1178{
1179 notifier_list_add(&migration_state_notifiers, notify);
1180}
1181
1182void remove_migration_state_change_notifier(Notifier *notify)
1183{
31552529 1184 notifier_remove(notify);
99a0db9b
GH		 1185}
1186
02edd2e7 1187bool migration_in_setup(MigrationState *s)
afe2df69 1188{
31194731 1189 return s->state == MIGRATION_STATUS_SETUP;
afe2df69
GH		 1190}
1191
7073693b 1192bool migration_has_finished(MigrationState *s)
99a0db9b 1193{
31194731 1194 return s->state == MIGRATION_STATUS_COMPLETED;
99a0db9b 1195}
0edda1c4 1196
afe2df69
GH		 1197bool migration_has_failed(MigrationState *s)
1198{
31194731
HZ		 1199 return (s->state == MIGRATION_STATUS_CANCELLED ||
1200 s->state == MIGRATION_STATUS_FAILED);
afe2df69
GH		 1201}
1202
5727309d 1203bool migration_in_postcopy(void)
9ec055ae 1204{
5727309d
JQ		 1205 MigrationState *s = migrate_get_current();
1206
9ec055ae
DDAG		 1207 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1208}
1209
b82fc321
DDAG		 1210bool migration_in_postcopy_after_devices(MigrationState *s)
1211{
5727309d 1212 return migration_in_postcopy() && s->postcopy_after_devices;
b82fc321
DDAG		 1213}
1214
fab35005 1215bool migration_is_idle(void)
fe44dc91 1216{
fab35005 1217 MigrationState *s = migrate_get_current();
fe44dc91
AA		 1218
1219 switch (s->state) {
1220 case MIGRATION_STATUS_NONE:
1221 case MIGRATION_STATUS_CANCELLED:
1222 case MIGRATION_STATUS_COMPLETED:
1223 case MIGRATION_STATUS_FAILED:
1224 return true;
1225 case MIGRATION_STATUS_SETUP:
1226 case MIGRATION_STATUS_CANCELLING:
1227 case MIGRATION_STATUS_ACTIVE:
1228 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1229 case MIGRATION_STATUS_COLO:
31e06077
DDAG		 1230 case MIGRATION_STATUS_PRE_SWITCHOVER:
1231 case MIGRATION_STATUS_DEVICE:
fe44dc91
AA		 1232 return false;
1233 case MIGRATION_STATUS__MAX:
1234 g_assert_not_reached();
1235 }
1236
1237 return false;
1238}
1239
a0762d9e 1240MigrationState *migrate_init(void)
0edda1c4 1241{
17549e84 1242 MigrationState *s = migrate_get_current();
bbf6da32 1243
389775d1
DDAG		 1244 /*
1245 * Reinitialise all migration state, except
1246 * parameters/capabilities that the user set, and
1247 * locks.
1248 */
1249 s->bytes_xfer = 0;
1250 s->xfer_limit = 0;
1251 s->cleanup_bh = 0;
89a02a9f 1252 s->to_dst_file = NULL;
389775d1 1253 s->state = MIGRATION_STATUS_NONE;
389775d1
DDAG		 1254 s->rp_state.from_dst_file = NULL;
1255 s->rp_state.error = false;
1256 s->mbps = 0.0;
1257 s->downtime = 0;
1258 s->expected_downtime = 0;
389775d1 1259 s->setup_time = 0;
389775d1 1260 s->start_postcopy = false;
b82fc321 1261 s->postcopy_after_devices = false;
389775d1 1262 s->migration_thread_running = false;
d59ce6f3
DB		 1263 error_free(s->error);
1264 s->error = NULL;
389775d1 1265
48781e5b 1266 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
0edda1c4 1267
4af246a3
PX		 1268 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1269 s->total_time = 0;
7287cbd4 1270 s->vm_was_running = false;
b15df1ae
PX		 1271 s->iteration_initial_bytes = 0;
1272 s->threshold_size = 0;
0edda1c4
JQ		 1273 return s;
1274}
cab30143 1275
fa2756b7
AL		 1276static GSList *migration_blockers;
1277
fe44dc91 1278int migrate_add_blocker(Error *reason, Error **errp)
fa2756b7 1279{
3df663e5 1280 if (migrate_get_current()->only_migratable) {
b67b8c3a
AA		 1281 error_propagate(errp, error_copy(reason));
1282 error_prepend(errp, "disallowing migration blocker "
1283 "(--only_migratable) for: ");
1284 return -EACCES;
1285 }
1286
fab35005 1287 if (migration_is_idle()) {
fe44dc91
AA		 1288 migration_blockers = g_slist_prepend(migration_blockers, reason);
1289 return 0;
1290 }
1291
1292 error_propagate(errp, error_copy(reason));
1293 error_prepend(errp, "disallowing migration blocker (migration in "
1294 "progress) for: ");
1295 return -EBUSY;
fa2756b7
AL		 1296}
1297
1298void migrate_del_blocker(Error *reason)
1299{
1300 migration_blockers = g_slist_remove(migration_blockers, reason);
1301}
1302
bf1ae1f4
DDAG		 1303void qmp_migrate_incoming(const char *uri, Error **errp)
1304{
1305 Error *local_err = NULL;
4debb5f5 1306 static bool once = true;
bf1ae1f4
DDAG		 1307
1308 if (!deferred_incoming) {
4debb5f5 1309 error_setg(errp, "For use with '-incoming defer'");
bf1ae1f4
DDAG		 1310 return;
1311 }
4debb5f5
DDAG		 1312 if (!once) {
1313 error_setg(errp, "The incoming migration has already been started");
1314 }
bf1ae1f4
DDAG		 1315
1316 qemu_start_incoming_migration(uri, &local_err);
1317
1318 if (local_err) {
1319 error_propagate(errp, local_err);
1320 return;
1321 }
1322
4debb5f5 1323 once = false;
bf1ae1f4
DDAG		 1324}
1325
24f3902b
GK		 1326bool migration_is_blocked(Error **errp)
1327{
1328 if (qemu_savevm_state_blocked(errp)) {
1329 return true;
1330 }
1331
1332 if (migration_blockers) {
250561e1 1333 error_propagate(errp, error_copy(migration_blockers->data));
24f3902b
GK		 1334 return true;
1335 }
1336
1337 return false;
1338}
1339
e1c37d0e
LC		 1340void qmp_migrate(const char *uri, bool has_blk, bool blk,
1341 bool has_inc, bool inc, bool has_detach, bool detach,
1342 Error **errp)
cab30143 1343{
be7059cd 1344 Error *local_err = NULL;
17549e84 1345 MigrationState *s = migrate_get_current();
cab30143 1346 const char *p;
cab30143 1347
f6844b99 1348 if (migration_is_setup_or_active(s->state) ||
0b827d5e
HZ		 1349 s->state == MIGRATION_STATUS_CANCELLING ||
1350 s->state == MIGRATION_STATUS_COLO) {
c6bd8c70 1351 error_setg(errp, QERR_MIGRATION_ACTIVE);
e1c37d0e 1352 return;
cab30143 1353 }
ca99993a
DDAG		 1354 if (runstate_check(RUN_STATE_INMIGRATE)) {
1355 error_setg(errp, "Guest is waiting for an incoming migration");
1356 return;
1357 }
1358
24f3902b 1359 if (migration_is_blocked(errp)) {
e1c37d0e 1360 return;
fa2756b7
AL		 1361 }
1362
2833c59b
JQ		 1363 if ((has_blk && blk) || (has_inc && inc)) {
1364 if (migrate_use_block() || migrate_use_block_incremental()) {
1365 error_setg(errp, "Command options are incompatible with "
1366 "current migration capabilities");
1367 return;
1368 }
1369 migrate_set_block_enabled(true, &local_err);
1370 if (local_err) {
1371 error_propagate(errp, local_err);
1372 return;
1373 }
1374 s->must_remove_block_options = true;
1375 }
1376
1377 if (has_inc && inc) {
1378 migrate_set_block_incremental(s, true);
1379 }
1380
a0762d9e 1381 s = migrate_init();
cab30143
JQ		 1382
1383 if (strstart(uri, "tcp:", &p)) {
f37afb5a 1384 tcp_start_outgoing_migration(s, p, &local_err);
2da776db 1385#ifdef CONFIG_RDMA
41310c68 1386 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 1387 rdma_start_outgoing_migration(s, p, &local_err);
1388#endif
cab30143 1389 } else if (strstart(uri, "exec:", &p)) {
f37afb5a 1390 exec_start_outgoing_migration(s, p, &local_err);
cab30143 1391 } else if (strstart(uri, "unix:", &p)) {
f37afb5a 1392 unix_start_outgoing_migration(s, p, &local_err);
cab30143 1393 } else if (strstart(uri, "fd:", &p)) {
f37afb5a 1394 fd_start_outgoing_migration(s, p, &local_err);
99a0db9b 1395 } else {
c6bd8c70
MA		 1396 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1397 "a valid migration protocol");
48781e5b
HZ		 1398 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1399 MIGRATION_STATUS_FAILED);
e1c37d0e 1400 return;
cab30143
JQ		 1401 }
1402
f37afb5a 1403 if (local_err) {
d59ce6f3 1404 migrate_fd_error(s, local_err);
f37afb5a 1405 error_propagate(errp, local_err);
e1c37d0e 1406 return;
1299c631 1407 }
cab30143
JQ		 1408}
1409
6cdedb07 1410void qmp_migrate_cancel(Error **errp)
cab30143 1411{
17549e84 1412 migrate_fd_cancel(migrate_get_current());
cab30143
JQ		 1413}
1414
89cfc02c
DDAG		 1415void qmp_migrate_continue(MigrationStatus state, Error **errp)
1416{
1417 MigrationState *s = migrate_get_current();
1418 if (s->state != state) {
1419 error_setg(errp, "Migration not in expected state: %s",
1420 MigrationStatus_str(s->state));
1421 return;
1422 }
1423 qemu_sem_post(&s->pause_sem);
1424}
1425
9e1ba4cc
OW		 1426void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1427{
73af8dd8
JQ		 1428 MigrateSetParameters p = {
1429 .has_xbzrle_cache_size = true,
1430 .xbzrle_cache_size = value,
1431 };
c91e681a 1432
73af8dd8 1433 qmp_migrate_set_parameters(&p, errp);
9e1ba4cc
OW		 1434}
1435
1436int64_t qmp_query_migrate_cache_size(Error **errp)
1437{
1438 return migrate_xbzrle_cache_size();
1439}
1440
3dc85383 1441void qmp_migrate_set_speed(int64_t value, Error **errp)
cab30143 1442{
1bda8b3c 1443 MigrateSetParameters p = {
2ff30257
AA		 1444 .has_max_bandwidth = true,
1445 .max_bandwidth = value,
1446 };
cab30143 1447
2ff30257 1448 qmp_migrate_set_parameters(&p, errp);
cab30143
JQ		 1449}
1450
4f0a993b 1451void qmp_migrate_set_downtime(double value, Error **errp)
cab30143 1452{
87c9cc1c
DHB		 1453 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1454 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1455 "the range of 0 to %d seconds",
1456 MAX_MIGRATE_DOWNTIME_SECONDS);
1457 return;
1458 }
1459
2ff30257
AA		 1460 value *= 1000; /* Convert to milliseconds */
1461 value = MAX(0, MIN(INT64_MAX, value));
1462
1bda8b3c 1463 MigrateSetParameters p = {
2ff30257
AA		 1464 .has_downtime_limit = true,
1465 .downtime_limit = value,
1466 };
1467
1468 qmp_migrate_set_parameters(&p, errp);
99a0db9b 1469}
17ad9b35 1470
53f09a10
PB		 1471bool migrate_release_ram(void)
1472{
1473 MigrationState *s;
1474
1475 s = migrate_get_current();
1476
1477 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1478}
1479
53dd370c
DDAG		 1480bool migrate_postcopy_ram(void)
1481{
1482 MigrationState *s;
1483
1484 s = migrate_get_current();
1485
32c3db5b 1486 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
53dd370c
DDAG		 1487}
1488
58110f0a
VSO		 1489bool migrate_postcopy(void)
1490{
1491 return migrate_postcopy_ram();
1492}
1493
bde1e2ec
CV		 1494bool migrate_auto_converge(void)
1495{
1496 MigrationState *s;
1497
1498 s = migrate_get_current();
1499
1500 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1501}
1502
323004a3
PL		 1503bool migrate_zero_blocks(void)
1504{
1505 MigrationState *s;
1506
1507 s = migrate_get_current();
1508
1509 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1510}
1511
8706d2d5
LL		 1512bool migrate_use_compression(void)
1513{
dde4e694
LL		 1514 MigrationState *s;
1515
1516 s = migrate_get_current();
1517
1518 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
8706d2d5
LL		 1519}
1520
1521int migrate_compress_level(void)
1522{
1523 MigrationState *s;
1524
1525 s = migrate_get_current();
1526
2594f56d 1527 return s->parameters.compress_level;
8706d2d5
LL		 1528}
1529
1530int migrate_compress_threads(void)
1531{
1532 MigrationState *s;
1533
1534 s = migrate_get_current();
1535
2594f56d 1536 return s->parameters.compress_threads;
8706d2d5
LL		 1537}
1538
3fcb38c2
LL		 1539int migrate_decompress_threads(void)
1540{
1541 MigrationState *s;
1542
1543 s = migrate_get_current();
1544
2594f56d 1545 return s->parameters.decompress_threads;
3fcb38c2
LL		 1546}
1547
b05dc723
JQ		 1548bool migrate_use_events(void)
1549{
1550 MigrationState *s;
1551
1552 s = migrate_get_current();
1553
1554 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1555}
1556
30126bbf
JQ		 1557bool migrate_use_multifd(void)
1558{
1559 MigrationState *s;
1560
1561 s = migrate_get_current();
1562
1563 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1564}
1565
93fbd031
DDAG		 1566bool migrate_pause_before_switchover(void)
1567{
1568 MigrationState *s;
1569
1570 s = migrate_get_current();
1571
1572 return s->enabled_capabilities[
1573 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
1574}
1575
4075fb1c
JQ		 1576int migrate_multifd_channels(void)
1577{
1578 MigrationState *s;
1579
1580 s = migrate_get_current();
1581
1582 return s->parameters.x_multifd_channels;
1583}
1584
0fb86605
JQ		 1585int migrate_multifd_page_count(void)
1586{
1587 MigrationState *s;
1588
1589 s = migrate_get_current();
1590
1591 return s->parameters.x_multifd_page_count;
1592}
1593
17ad9b35
OW		 1594int migrate_use_xbzrle(void)
1595{
1596 MigrationState *s;
1597
1598 s = migrate_get_current();
1599
1600 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1601}
1602
1603int64_t migrate_xbzrle_cache_size(void)
1604{
1605 MigrationState *s;
1606
1607 s = migrate_get_current();
1608
73af8dd8 1609 return s->parameters.xbzrle_cache_size;
17ad9b35 1610}
0d82d0e8 1611
2833c59b
JQ		 1612bool migrate_use_block(void)
1613{
1614 MigrationState *s;
1615
1616 s = migrate_get_current();
1617
1618 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1619}
1620
c788ada8
PX		 1621bool migrate_use_return_path(void)
1622{
1623 MigrationState *s;
1624
1625 s = migrate_get_current();
1626
1627 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1628}
1629
2833c59b
JQ		 1630bool migrate_use_block_incremental(void)
1631{
1632 MigrationState *s;
1633
1634 s = migrate_get_current();
1635
1636 return s->parameters.block_incremental;
1637}
1638
70b20477
DDAG		 1639/* migration thread support */
1640/*
1641 * Something bad happened to the RP stream, mark an error
1642 * The caller shall print or trace something to indicate why
1643 */
1644static void mark_source_rp_bad(MigrationState *s)
1645{
1646 s->rp_state.error = true;
1647}
1648
1649static struct rp_cmd_args {
1650 ssize_t len; /* -1 = variable */
1651 const char *name;
1652} rp_cmd_args[] = {
1653 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1654 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1655 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1e2d90eb
DDAG		 1656 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1657 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
70b20477
DDAG		 1658 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1659};
1660
1e2d90eb
DDAG		 1661/*
1662 * Process a request for pages received on the return path,
1663 * We're allowed to send more than requested (e.g. to round to our page size)
1664 * and we don't need to send pages that have already been sent.
1665 */
1666static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1667 ram_addr_t start, size_t len)
1668{
6c595cde
DDAG		 1669 long our_host_ps = getpagesize();
1670
1e2d90eb 1671 trace_migrate_handle_rp_req_pages(rbname, start, len);
6c595cde
DDAG		 1672
1673 /*
1674 * Since we currently insist on matching page sizes, just sanity check
1675 * we're being asked for whole host pages.
1676 */
1677 if (start & (our_host_ps-1) ||
1678 (len & (our_host_ps-1))) {
1679 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1680 " len: %zd", __func__, start, len);
1681 mark_source_rp_bad(ms);
1682 return;
1683 }
1684
96506894 1685 if (ram_save_queue_pages(rbname, start, len)) {
6c595cde
DDAG		 1686 mark_source_rp_bad(ms);
1687 }
1e2d90eb
DDAG		 1688}
1689
70b20477
DDAG		 1690/*
1691 * Handles messages sent on the return path towards the source VM
1692 *
1693 */
1694static void *source_return_path_thread(void *opaque)
1695{
1696 MigrationState *ms = opaque;
1697 QEMUFile *rp = ms->rp_state.from_dst_file;
1698 uint16_t header_len, header_type;
568b01ca 1699 uint8_t buf[512];
70b20477 1700 uint32_t tmp32, sibling_error;
1e2d90eb
DDAG		 1701 ram_addr_t start = 0; /* =0 to silence warning */
1702 size_t len = 0, expected_len;
70b20477
DDAG		 1703 int res;
1704
1705 trace_source_return_path_thread_entry();
1706 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1707 migration_is_setup_or_active(ms->state)) {
1708 trace_source_return_path_thread_loop_top();
1709 header_type = qemu_get_be16(rp);
1710 header_len = qemu_get_be16(rp);
1711
1712 if (header_type >= MIG_RP_MSG_MAX ||
1713 header_type == MIG_RP_MSG_INVALID) {
1714 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1715 header_type, header_len);
1716 mark_source_rp_bad(ms);
1717 goto out;
1718 }
1719
1720 if ((rp_cmd_args[header_type].len != -1 &&
1721 header_len != rp_cmd_args[header_type].len) ||
568b01ca 1722 header_len > sizeof(buf)) {
70b20477
DDAG		 1723 error_report("RP: Received '%s' message (0x%04x) with"
1724 "incorrect length %d expecting %zu",
1725 rp_cmd_args[header_type].name, header_type, header_len,
1726 (size_t)rp_cmd_args[header_type].len);
1727 mark_source_rp_bad(ms);
1728 goto out;
1729 }
1730
1731 /* We know we've got a valid header by this point */
1732 res = qemu_get_buffer(rp, buf, header_len);
1733 if (res != header_len) {
1734 error_report("RP: Failed reading data for message 0x%04x"
1735 " read %d expected %d",
1736 header_type, res, header_len);
1737 mark_source_rp_bad(ms);
1738 goto out;
1739 }
1740
1741 /* OK, we have the message and the data */
1742 switch (header_type) {
1743 case MIG_RP_MSG_SHUT:
4d885131 1744 sibling_error = ldl_be_p(buf);
70b20477
DDAG		 1745 trace_source_return_path_thread_shut(sibling_error);
1746 if (sibling_error) {
1747 error_report("RP: Sibling indicated error %d", sibling_error);
1748 mark_source_rp_bad(ms);
1749 }
1750 /*
1751 * We'll let the main thread deal with closing the RP
1752 * we could do a shutdown(2) on it, but we're the only user
1753 * anyway, so there's nothing gained.
1754 */
1755 goto out;
1756
1757 case MIG_RP_MSG_PONG:
4d885131 1758 tmp32 = ldl_be_p(buf);
70b20477
DDAG		 1759 trace_source_return_path_thread_pong(tmp32);
1760 break;
1761
1e2d90eb 1762 case MIG_RP_MSG_REQ_PAGES:
4d885131
PM		 1763 start = ldq_be_p(buf);
1764 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 1765 migrate_handle_rp_req_pages(ms, NULL, start, len);
1766 break;
1767
1768 case MIG_RP_MSG_REQ_PAGES_ID:
1769 expected_len = 12 + 1; /* header + termination */
1770
1771 if (header_len >= expected_len) {
4d885131
PM		 1772 start = ldq_be_p(buf);
1773 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 1774 /* Now we expect an idstr */
1775 tmp32 = buf[12]; /* Length of the following idstr */
1776 buf[13 + tmp32] = '\0';
1777 expected_len += tmp32;
1778 }
1779 if (header_len != expected_len) {
1780 error_report("RP: Req_Page_id with length %d expecting %zd",
1781 header_len, expected_len);
1782 mark_source_rp_bad(ms);
1783 goto out;
1784 }
1785 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1786 break;
1787
70b20477
DDAG		 1788 default:
1789 break;
1790 }
1791 }
5df5416e 1792 if (qemu_file_get_error(rp)) {
70b20477
DDAG		 1793 trace_source_return_path_thread_bad_end();
1794 mark_source_rp_bad(ms);
1795 }
1796
1797 trace_source_return_path_thread_end();
1798out:
1799 ms->rp_state.from_dst_file = NULL;
1800 qemu_fclose(rp);
1801 return NULL;
1802}
1803
70b20477
DDAG		 1804static int open_return_path_on_source(MigrationState *ms)
1805{
1806
89a02a9f 1807 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
70b20477
DDAG		 1808 if (!ms->rp_state.from_dst_file) {
1809 return -1;
1810 }
1811
1812 trace_open_return_path_on_source();
1813 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1814 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1815
1816 trace_open_return_path_on_source_continue();
1817
1818 return 0;
1819}
1820
70b20477
DDAG		 1821/* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1822static int await_return_path_close_on_source(MigrationState *ms)
1823{
1824 /*
1825 * If this is a normal exit then the destination will send a SHUT and the
1826 * rp_thread will exit, however if there's an error we need to cause
1827 * it to exit.
1828 */
89a02a9f 1829 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
70b20477
DDAG		 1830 /*
1831 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1832 * waiting for the destination.
1833 */
1834 qemu_file_shutdown(ms->rp_state.from_dst_file);
1835 mark_source_rp_bad(ms);
1836 }
1837 trace_await_return_path_close_on_source_joining();
1838 qemu_thread_join(&ms->rp_state.rp_thread);
1839 trace_await_return_path_close_on_source_close();
1840 return ms->rp_state.error;
1841}
1842
1d34e4bf
DDAG		 1843/*
1844 * Switch from normal iteration to postcopy
1845 * Returns non-0 on error
1846 */
7287cbd4 1847static int postcopy_start(MigrationState *ms)
1d34e4bf
DDAG		 1848{
1849 int ret;
61b67d47
DB		 1850 QIOChannelBuffer *bioc;
1851 QEMUFile *fb;
1d34e4bf 1852 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
ef8d6488 1853 bool restart_block = false;
0331c8ca
DDAG		 1854 int cur_state = MIGRATION_STATUS_ACTIVE;
1855 if (!migrate_pause_before_switchover()) {
1856 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1857 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1858 }
1d34e4bf
DDAG		 1859
1860 trace_postcopy_start();
1861 qemu_mutex_lock_iothread();
1862 trace_postcopy_start_set_run();
1863
1864 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1d34e4bf
DDAG		 1865 global_state_store();
1866 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
76b1c7fe
KW		 1867 if (ret < 0) {
1868 goto fail;
1869 }
1d34e4bf 1870
0331c8ca
DDAG		 1871 ret = migration_maybe_pause(ms, &cur_state,
1872 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1873 if (ret < 0) {
1874 goto fail;
1875 }
1876
76b1c7fe 1877 ret = bdrv_inactivate_all();
1d34e4bf
DDAG		 1878 if (ret < 0) {
1879 goto fail;
1880 }
ef8d6488 1881 restart_block = true;
1d34e4bf 1882
1c0d249d
DDAG		 1883 /*
1884 * Cause any non-postcopiable, but iterative devices to
1885 * send out their final data.
1886 */
a1fbe750 1887 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1c0d249d 1888
1d34e4bf
DDAG		 1889 /*
1890 * in Finish migrate and with the io-lock held everything should
1891 * be quiet, but we've potentially still got dirty pages and we
1892 * need to tell the destination to throw any pages it's already received
1893 * that are dirty
1894 */
58110f0a
VSO		 1895 if (migrate_postcopy_ram()) {
1896 if (ram_postcopy_send_discard_bitmap(ms)) {
1897 error_report("postcopy send discard bitmap failed");
1898 goto fail;
1899 }
1d34e4bf
DDAG		 1900 }
1901
1902 /*
1903 * send rest of state - note things that are doing postcopy
1904 * will notice we're in POSTCOPY_ACTIVE and not actually
1905 * wrap their state up here
1906 */
89a02a9f 1907 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
58110f0a
VSO		 1908 if (migrate_postcopy_ram()) {
1909 /* Ping just for debugging, helps line traces up */
1910 qemu_savevm_send_ping(ms->to_dst_file, 2);
1911 }
1d34e4bf
DDAG		 1912
1913 /*
1914 * While loading the device state we may trigger page transfer
1915 * requests and the fd must be free to process those, and thus
1916 * the destination must read the whole device state off the fd before
1917 * it starts processing it. Unfortunately the ad-hoc migration format
1918 * doesn't allow the destination to know the size to read without fully
1919 * parsing it through each devices load-state code (especially the open
1920 * coded devices that use get/put).
1921 * So we wrap the device state up in a package with a length at the start;
1922 * to do this we use a qemu_buf to hold the whole of the device state.
1923 */
61b67d47 1924 bioc = qio_channel_buffer_new(4096);
6f01f136 1925 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
61b67d47
DB		 1926 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1927 object_unref(OBJECT(bioc));
1d34e4bf 1928
c76201ab
DDAG		 1929 /*
1930 * Make sure the receiver can get incoming pages before we send the rest
1931 * of the state
1932 */
1933 qemu_savevm_send_postcopy_listen(fb);
1934
a1fbe750 1935 qemu_savevm_state_complete_precopy(fb, false, false);
58110f0a
VSO		 1936 if (migrate_postcopy_ram()) {
1937 qemu_savevm_send_ping(fb, 3);
1938 }
1d34e4bf
DDAG		 1939
1940 qemu_savevm_send_postcopy_run(fb);
1941
1942 /* <><> end of stuff going into the package */
1d34e4bf 1943
ef8d6488
DDAG		 1944 /* Last point of recovery; as soon as we send the package the destination
1945 * can open devices and potentially start running.
1946 * Lets just check again we've not got any errors.
1947 */
1948 ret = qemu_file_get_error(ms->to_dst_file);
1949 if (ret) {
1950 error_report("postcopy_start: Migration stream errored (pre package)");
1951 goto fail_closefb;
1952 }
1953
1954 restart_block = false;
1955
1d34e4bf 1956 /* Now send that blob */
61b67d47 1957 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1d34e4bf
DDAG		 1958 goto fail_closefb;
1959 }
1960 qemu_fclose(fb);
b82fc321
DDAG		 1961
1962 /* Send a notify to give a chance for anything that needs to happen
1963 * at the transition to postcopy and after the device state; in particular
1964 * spice needs to trigger a transition now
1965 */
1966 ms->postcopy_after_devices = true;
1967 notifier_list_notify(&migration_state_notifiers, ms);
1968
1d34e4bf
DDAG		 1969 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1970
1971 qemu_mutex_unlock_iothread();
1972
58110f0a
VSO		 1973 if (migrate_postcopy_ram()) {
1974 /*
1975 * Although this ping is just for debug, it could potentially be
1976 * used for getting a better measurement of downtime at the source.
1977 */
1978 qemu_savevm_send_ping(ms->to_dst_file, 4);
1979 }
1d34e4bf 1980
ced1c616
PB		 1981 if (migrate_release_ram()) {
1982 ram_postcopy_migrated_memory_release(ms);
1983 }
1984
89a02a9f 1985 ret = qemu_file_get_error(ms->to_dst_file);
1d34e4bf
DDAG		 1986 if (ret) {
1987 error_report("postcopy_start: Migration stream errored");
48781e5b 1988 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf
DDAG		 1989 MIGRATION_STATUS_FAILED);
1990 }
1991
1992 return ret;
1993
1994fail_closefb:
1995 qemu_fclose(fb);
1996fail:
48781e5b 1997 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf 1998 MIGRATION_STATUS_FAILED);
ef8d6488
DDAG		 1999 if (restart_block) {
2000 /* A failure happened early enough that we know the destination hasn't
2001 * accessed block devices, so we're safe to recover.
2002 */
2003 Error *local_err = NULL;
2004
2005 bdrv_invalidate_cache_all(&local_err);
2006 if (local_err) {
2007 error_report_err(local_err);
2008 }
2009 }
1d34e4bf
DDAG		 2010 qemu_mutex_unlock_iothread();
2011 return -1;
2012}
2013
e91d8951
DDAG		 2014/**
2015 * migration_maybe_pause: Pause if required to by
2016 * migrate_pause_before_switchover called with the iothread locked
2017 * Returns: 0 on success
2018 */
0331c8ca
DDAG		 2019static int migration_maybe_pause(MigrationState *s,
2020 int *current_active_state,
2021 int new_state)
e91d8951
DDAG		 2022{
2023 if (!migrate_pause_before_switchover()) {
2024 return 0;
2025 }
2026
2027 /* Since leaving this state is not atomic with posting the semaphore
2028 * it's possible that someone could have issued multiple migrate_continue
2029 * and the semaphore is incorrectly positive at this point;
2030 * the docs say it's undefined to reinit a semaphore that's already
2031 * init'd, so use timedwait to eat up any existing posts.
2032 */
2033 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2034 /* This block intentionally left blank */
2035 }
2036
2037 qemu_mutex_unlock_iothread();
2038 migrate_set_state(&s->state, *current_active_state,
2039 MIGRATION_STATUS_PRE_SWITCHOVER);
2040 qemu_sem_wait(&s->pause_sem);
2041 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
0331c8ca
DDAG		 2042 new_state);
2043 *current_active_state = new_state;
e91d8951
DDAG		 2044 qemu_mutex_lock_iothread();
2045
0331c8ca 2046 return s->state == new_state ? 0 : -EINVAL;
e91d8951
DDAG		 2047}
2048
09f6c85e
DDAG		 2049/**
2050 * migration_completion: Used by migration_thread when there's not much left.
2051 * The caller 'breaks' the loop when this returns.
2052 *
2053 * @s: Current migration state
09f6c85e 2054 */
2ad87305 2055static void migration_completion(MigrationState *s)
09f6c85e
DDAG		 2056{
2057 int ret;
2ad87305 2058 int current_active_state = s->state;
09f6c85e 2059
b10ac0c4
DDAG		 2060 if (s->state == MIGRATION_STATUS_ACTIVE) {
2061 qemu_mutex_lock_iothread();
64909f97 2062 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
b10ac0c4 2063 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
7287cbd4 2064 s->vm_was_running = runstate_is_running();
b10ac0c4
DDAG		 2065 ret = global_state_store();
2066
2067 if (!ret) {
a1fbe750 2068 bool inactivate = !migrate_colo_enabled();
b10ac0c4 2069 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
e91d8951 2070 if (ret >= 0) {
0331c8ca
DDAG		 2071 ret = migration_maybe_pause(s, &current_active_state,
2072 MIGRATION_STATUS_DEVICE);
e91d8951 2073 }
f07fa4cb
KW		 2074 if (ret >= 0) {
2075 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
a1fbe750
FZ		 2076 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2077 inactivate);
f07fa4cb 2078 }
a1fbe750
FZ		 2079 if (inactivate && ret >= 0) {
2080 s->block_inactive = true;
b10ac0c4
DDAG		 2081 }
2082 }
2083 qemu_mutex_unlock_iothread();
09f6c85e 2084
b10ac0c4
DDAG		 2085 if (ret < 0) {
2086 goto fail;
09f6c85e 2087 }
b10ac0c4
DDAG		 2088 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2089 trace_migration_completion_postcopy_end();
2090
89a02a9f 2091 qemu_savevm_state_complete_postcopy(s->to_dst_file);
b10ac0c4 2092 trace_migration_completion_postcopy_end_after_complete();
09f6c85e 2093 }
09f6c85e 2094
b10ac0c4
DDAG		 2095 /*
2096 * If rp was opened we must clean up the thread before
2097 * cleaning everything else up (since if there are no failures
2098 * it will wait for the destination to send it's status in
2099 * a SHUT command).
b10ac0c4 2100 */
0425dc97 2101 if (s->rp_state.from_dst_file) {
b10ac0c4 2102 int rp_error;
0425dc97 2103 trace_migration_return_path_end_before();
b10ac0c4 2104 rp_error = await_return_path_close_on_source(s);
0425dc97 2105 trace_migration_return_path_end_after(rp_error);
b10ac0c4 2106 if (rp_error) {
fe904ea8 2107 goto fail_invalidate;
b10ac0c4 2108 }
09f6c85e
DDAG		 2109 }
2110
89a02a9f 2111 if (qemu_file_get_error(s->to_dst_file)) {
09f6c85e 2112 trace_migration_completion_file_err();
fe904ea8 2113 goto fail_invalidate;
09f6c85e
DDAG		 2114 }
2115
0b827d5e
HZ		 2116 if (!migrate_colo_enabled()) {
2117 migrate_set_state(&s->state, current_active_state,
2118 MIGRATION_STATUS_COMPLETED);
2119 }
2120
09f6c85e
DDAG		 2121 return;
2122
fe904ea8
GK		 2123fail_invalidate:
2124 /* If not doing postcopy, vm_start() will be called: let's regain
2125 * control on images.
2126 */
6039dd5b
DDAG		 2127 if (s->state == MIGRATION_STATUS_ACTIVE ||
2128 s->state == MIGRATION_STATUS_DEVICE) {
fe904ea8
GK		 2129 Error *local_err = NULL;
2130
1d2acc31 2131 qemu_mutex_lock_iothread();
fe904ea8
GK		 2132 bdrv_invalidate_cache_all(&local_err);
2133 if (local_err) {
2134 error_report_err(local_err);
1d2acc31
HZ		 2135 } else {
2136 s->block_inactive = false;
fe904ea8 2137 }
1d2acc31 2138 qemu_mutex_unlock_iothread();
fe904ea8
GK		 2139 }
2140
09f6c85e 2141fail:
48781e5b
HZ		 2142 migrate_set_state(&s->state, current_active_state,
2143 MIGRATION_STATUS_FAILED);
09f6c85e
DDAG		 2144}
2145
35a6ed4f
HZ		 2146bool migrate_colo_enabled(void)
2147{
2148 MigrationState *s = migrate_get_current();
2149 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2150}
2151
cf011f08
PX		 2152static void migration_calculate_complete(MigrationState *s)
2153{
2154 uint64_t bytes = qemu_ftell(s->to_dst_file);
2155 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2156
2157 s->total_time = end_time - s->start_time;
2158 if (!s->downtime) {
2159 /*
2160 * It's still not set, so we are precopy migration. For
2161 * postcopy, downtime is calculated during postcopy_start().
2162 */
2163 s->downtime = end_time - s->downtime_start;
2164 }
2165
2166 if (s->total_time) {
2167 s->mbps = ((double) bytes * 8.0) / s->total_time / 1000;
2168 }
2169}
2170
b15df1ae
PX		 2171static void migration_update_counters(MigrationState *s,
2172 int64_t current_time)
2173{
2174 uint64_t transferred, time_spent;
b15df1ae
PX		 2175 double bandwidth;
2176
2177 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
2178 return;
2179 }
2180
2181 transferred = qemu_ftell(s->to_dst_file) - s->iteration_initial_bytes;
2182 time_spent = current_time - s->iteration_start_time;
2183 bandwidth = (double)transferred / time_spent;
0781c1ed 2184 s->threshold_size = bandwidth * s->parameters.downtime_limit;
b15df1ae
PX		 2185
2186 s->mbps = (((double) transferred * 8.0) /
2187 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2188
2189 /*
2190 * if we haven't sent anything, we don't want to
2191 * recalculate. 10000 is a small enough number for our purposes
2192 */
2193 if (ram_counters.dirty_pages_rate && transferred > 10000) {
2194 s->expected_downtime = ram_counters.dirty_pages_rate *
2195 qemu_target_page_size() / bandwidth;
2196 }
2197
2198 qemu_file_reset_rate_limit(s->to_dst_file);
2199
2200 s->iteration_start_time = current_time;
2201 s->iteration_initial_bytes = qemu_ftell(s->to_dst_file);
2202
2203 trace_migrate_transferred(transferred, time_spent,
0781c1ed 2204 bandwidth, s->threshold_size);
b15df1ae
PX		 2205}
2206
2ad87305
PX		 2207/* Migration thread iteration status */
2208typedef enum {
2209 MIG_ITERATE_RESUME, /* Resume current iteration */
2210 MIG_ITERATE_SKIP, /* Skip current iteration */
2211 MIG_ITERATE_BREAK, /* Break the loop */
2212} MigIterateState;
2213
2214/*
2215 * Return true if continue to the next iteration directly, false
2216 * otherwise.
2217 */
2218static MigIterateState migration_iteration_run(MigrationState *s)
2219{
2220 uint64_t pending_size, pend_post, pend_nonpost;
2221 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
2222
2223 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size,
2224 &pend_nonpost, &pend_post);
2225 pending_size = pend_nonpost + pend_post;
2226
2227 trace_migrate_pending(pending_size, s->threshold_size,
2228 pend_post, pend_nonpost);
2229
2230 if (pending_size && pending_size >= s->threshold_size) {
2231 /* Still a significant amount to transfer */
2232 if (migrate_postcopy() && !in_postcopy &&
2233 pend_nonpost <= s->threshold_size &&
2234 atomic_read(&s->start_postcopy)) {
2235 if (postcopy_start(s)) {
2236 error_report("%s: postcopy failed to start", __func__);
2237 }
2238 return MIG_ITERATE_SKIP;
2239 }
2240 /* Just another iteration step */
2241 qemu_savevm_state_iterate(s->to_dst_file,
2242 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2243 } else {
2244 trace_migration_thread_low_pending(pending_size);
2245 migration_completion(s);
2246 return MIG_ITERATE_BREAK;
2247 }
2248
2249 return MIG_ITERATE_RESUME;
2250}
2251
199aa6d4
PX		 2252static void migration_iteration_finish(MigrationState *s)
2253{
2254 /* If we enabled cpu throttling for auto-converge, turn it off. */
2255 cpu_throttle_stop();
2256
2257 qemu_mutex_lock_iothread();
2258 switch (s->state) {
2259 case MIGRATION_STATUS_COMPLETED:
2260 migration_calculate_complete(s);
2261 runstate_set(RUN_STATE_POSTMIGRATE);
2262 break;
2263
2264 case MIGRATION_STATUS_ACTIVE:
2265 /*
2266 * We should really assert here, but since it's during
2267 * migration, let's try to reduce the usage of assertions.
2268 */
2269 if (!migrate_colo_enabled()) {
2270 error_report("%s: critical error: calling COLO code without "
2271 "COLO enabled", __func__);
2272 }
2273 migrate_start_colo_process(s);
2274 /*
2275 * Fixme: we will run VM in COLO no matter its old running state.
2276 * After exited COLO, we will keep running.
2277 */
2278 s->vm_was_running = true;
2279 /* Fallthrough */
2280 case MIGRATION_STATUS_FAILED:
2281 case MIGRATION_STATUS_CANCELLED:
2282 if (s->vm_was_running) {
2283 vm_start();
2284 } else {
2285 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2286 runstate_set(RUN_STATE_POSTMIGRATE);
2287 }
2288 }
2289 break;
2290
2291 default:
2292 /* Should not reach here, but if so, forgive the VM. */
2293 error_report("%s: Unknown ending state %d", __func__, s->state);
2294 break;
2295 }
2296 qemu_bh_schedule(s->cleanup_bh);
2297 qemu_mutex_unlock_iothread();
2298}
2299
70b20477
DDAG		 2300/*
2301 * Master migration thread on the source VM.
2302 * It drives the migration and pumps the data down the outgoing channel.
2303 */
5f496a1b 2304static void *migration_thread(void *opaque)
0d82d0e8 2305{
9848a404 2306 MigrationState *s = opaque;
bc72ad67 2307 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
76f5933a 2308
ab28bd23
PB		 2309 rcu_register_thread();
2310
b15df1ae
PX		 2311 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2312
89a02a9f 2313 qemu_savevm_state_header(s->to_dst_file);
1d34e4bf 2314
62a02658
PX		 2315 /*
2316 * If we opened the return path, we need to make sure dst has it
2317 * opened as well.
2318 */
2319 if (s->rp_state.from_dst_file) {
1d34e4bf 2320 /* Now tell the dest that it should open its end so it can reply */
89a02a9f 2321 qemu_savevm_send_open_return_path(s->to_dst_file);
1d34e4bf
DDAG		 2322
2323 /* And do a ping that will make stuff easier to debug */
89a02a9f 2324 qemu_savevm_send_ping(s->to_dst_file, 1);
0425dc97 2325 }
1d34e4bf 2326
58110f0a 2327 if (migrate_postcopy()) {
1d34e4bf
DDAG		 2328 /*
2329 * Tell the destination that we *might* want to do postcopy later;
2330 * if the other end can't do postcopy it should fail now, nice and
2331 * early.
2332 */
89a02a9f 2333 qemu_savevm_send_postcopy_advise(s->to_dst_file);
1d34e4bf
DDAG		 2334 }
2335
9907e842 2336 qemu_savevm_state_setup(s->to_dst_file);
0d82d0e8 2337
bc72ad67 2338 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
48781e5b
HZ		 2339 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2340 MIGRATION_STATUS_ACTIVE);
29ae8a41 2341
9ec055ae
DDAG		 2342 trace_migration_thread_setup_complete();
2343
2344 while (s->state == MIGRATION_STATUS_ACTIVE ||
2345 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
a3e879cd 2346 int64_t current_time;
0d82d0e8 2347
89a02a9f 2348 if (!qemu_file_rate_limit(s->to_dst_file)) {
2ad87305
PX		 2349 MigIterateState iter_state = migration_iteration_run(s);
2350 if (iter_state == MIG_ITERATE_SKIP) {
2351 continue;
2352 } else if (iter_state == MIG_ITERATE_BREAK) {
09f6c85e 2353 break;
c369f40d
JQ		 2354 }
2355 }
f4410a5d 2356
89a02a9f 2357 if (qemu_file_get_error(s->to_dst_file)) {
2ad87305
PX		 2358 if (migration_is_setup_or_active(s->state)) {
2359 migrate_set_state(&s->state, s->state,
2360 MIGRATION_STATUS_FAILED);
2361 }
1d34e4bf 2362 trace_migration_thread_file_err();
fd45ee2c
PB		 2363 break;
2364 }
b15df1ae 2365
bc72ad67 2366 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
0d82d0e8 2367
b15df1ae
PX		 2368 migration_update_counters(s, current_time);
2369
89a02a9f 2370 if (qemu_file_rate_limit(s->to_dst_file)) {
0d82d0e8 2371 /* usleep expects microseconds */
b15df1ae
PX		 2372 g_usleep((s->iteration_start_time + BUFFER_DELAY -
2373 current_time) * 1000);
0d82d0e8 2374 }
a3fa1d78
PB		 2375 }
2376
1d34e4bf 2377 trace_migration_thread_after_loop();
199aa6d4 2378 migration_iteration_finish(s);
ab28bd23 2379 rcu_unregister_thread();
0d82d0e8
JQ		 2380 return NULL;
2381}
2382
cce8040b 2383void migrate_fd_connect(MigrationState *s, Error *error_in)
0d82d0e8 2384{
2ff30257 2385 s->expected_downtime = s->parameters.downtime_limit;
bb1fadc4 2386 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
cce8040b
DDAG		 2387 if (error_in) {
2388 migrate_fd_error(s, error_in);
2389 migrate_fd_cleanup(s);
2390 return;
2391 }
0d82d0e8 2392
9e4d2b98 2393 qemu_file_set_blocking(s->to_dst_file, true);
89a02a9f 2394 qemu_file_set_rate_limit(s->to_dst_file,
2ff30257 2395 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
442773ce 2396
9287ac27
SH		 2397 /* Notify before starting migration thread */
2398 notifier_list_notify(&migration_state_notifiers, s);
2399
1d34e4bf 2400 /*
c788ada8
PX		 2401 * Open the return path. For postcopy, it is used exclusively. For
2402 * precopy, only if user specified "return-path" capability would
2403 * QEMU uses the return path.
1d34e4bf 2404 */
c788ada8 2405 if (migrate_postcopy_ram() || migrate_use_return_path()) {
1d34e4bf
DDAG		 2406 if (open_return_path_on_source(s)) {
2407 error_report("Unable to open return-path for postcopy");
48781e5b 2408 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1d34e4bf
DDAG		 2409 MIGRATION_STATUS_FAILED);
2410 migrate_fd_cleanup(s);
2411 return;
2412 }
2413 }
2414
f986c3d2
JQ		 2415 if (multifd_save_setup() != 0) {
2416 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2417 MIGRATION_STATUS_FAILED);
2418 migrate_fd_cleanup(s);
2419 return;
2420 }
009fad7f 2421 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
bb1fadc4 2422 QEMU_THREAD_JOINABLE);
1d34e4bf 2423 s->migration_thread_running = true;
0d82d0e8 2424}
093e3c42 2425
9d18af93
PX		 2426void migration_global_dump(Monitor *mon)
2427{
2428 MigrationState *ms = migrate_get_current();
2429
6f0f6428
JQ		 2430 monitor_printf(mon, "globals:\n");
2431 monitor_printf(mon, "store-global-state: %s\n",
2432 ms->store_global_state ? "on" : "off");
2433 monitor_printf(mon, "only-migratable: %s\n",
2434 ms->only_migratable ? "on" : "off");
2435 monitor_printf(mon, "send-configuration: %s\n",
2436 ms->send_configuration ? "on" : "off");
2437 monitor_printf(mon, "send-section-footer: %s\n",
2438 ms->send_section_footer ? "on" : "off");
9d18af93
PX		 2439}
2440
20814758
PX		 2441#define DEFINE_PROP_MIG_CAP(name, x) \
2442 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2443
5272298c
PX		 2444static Property migration_properties[] = {
2445 DEFINE_PROP_BOOL("store-global-state", MigrationState,
2446 store_global_state, true),
3df663e5 2447 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
71dd4c1a
PX		 2448 DEFINE_PROP_BOOL("send-configuration", MigrationState,
2449 send_configuration, true),
15c38503
PX		 2450 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2451 send_section_footer, true),
89632faf
PX		 2452
2453 /* Migration parameters */
741d4086 2454 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
89632faf
PX		 2455 parameters.compress_level,
2456 DEFAULT_MIGRATE_COMPRESS_LEVEL),
741d4086 2457 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
89632faf
PX		 2458 parameters.compress_threads,
2459 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
741d4086 2460 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
89632faf
PX		 2461 parameters.decompress_threads,
2462 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
741d4086 2463 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
89632faf
PX		 2464 parameters.cpu_throttle_initial,
2465 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
741d4086 2466 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
89632faf
PX		 2467 parameters.cpu_throttle_increment,
2468 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
741d4086 2469 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
89632faf 2470 parameters.max_bandwidth, MAX_THROTTLE),
741d4086 2471 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
89632faf
PX		 2472 parameters.downtime_limit,
2473 DEFAULT_MIGRATE_SET_DOWNTIME),
741d4086 2474 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
89632faf
PX		 2475 parameters.x_checkpoint_delay,
2476 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
741d4086 2477 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState,
4075fb1c
JQ		 2478 parameters.x_multifd_channels,
2479 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
741d4086 2480 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState,
0fb86605
JQ		 2481 parameters.x_multifd_page_count,
2482 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
73af8dd8
JQ		 2483 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
2484 parameters.xbzrle_cache_size,
2485 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
20814758
PX		 2486
2487 /* Migration capabilities */
2488 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2489 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2490 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2491 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2492 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2493 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2494 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2495 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2496 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2497 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2498 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
30126bbf 2499 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
20814758 2500
5272298c
PX		 2501 DEFINE_PROP_END_OF_LIST(),
2502};
2503
e5cb7e76
PX		 2504static void migration_class_init(ObjectClass *klass, void *data)
2505{
2506 DeviceClass *dc = DEVICE_CLASS(klass);
2507
2508 dc->user_creatable = false;
5272298c 2509 dc->props = migration_properties;
e5cb7e76
PX		 2510}
2511
b91bf5e4
MAL		 2512static void migration_instance_finalize(Object *obj)
2513{
2514 MigrationState *ms = MIGRATION_OBJ(obj);
2515 MigrationParameters *params = &ms->parameters;
2516
87db1a7d 2517 qemu_mutex_destroy(&ms->error_mutex);
b91bf5e4
MAL		 2518 g_free(params->tls_hostname);
2519 g_free(params->tls_creds);
e91d8951 2520 qemu_sem_destroy(&ms->pause_sem);
b91bf5e4
MAL		 2521}
2522
e5cb7e76
PX		 2523static void migration_instance_init(Object *obj)
2524{
2525 MigrationState *ms = MIGRATION_OBJ(obj);
8b0b29dc 2526 MigrationParameters *params = &ms->parameters;
e5cb7e76
PX		 2527
2528 ms->state = MIGRATION_STATUS_NONE;
e5cb7e76 2529 ms->mbps = -1;
e91d8951 2530 qemu_sem_init(&ms->pause_sem, 0);
87db1a7d 2531 qemu_mutex_init(&ms->error_mutex);
8b0b29dc
PX		 2532
2533 params->tls_hostname = g_strdup("");
2534 params->tls_creds = g_strdup("");
2535
2536 /* Set has_* up only for parameter checks */
2537 params->has_compress_level = true;
2538 params->has_compress_threads = true;
2539 params->has_decompress_threads = true;
2540 params->has_cpu_throttle_initial = true;
2541 params->has_cpu_throttle_increment = true;
2542 params->has_max_bandwidth = true;
2543 params->has_downtime_limit = true;
2544 params->has_x_checkpoint_delay = true;
2545 params->has_block_incremental = true;
4075fb1c 2546 params->has_x_multifd_channels = true;
0fb86605 2547 params->has_x_multifd_page_count = true;
73af8dd8 2548 params->has_xbzrle_cache_size = true;
8b0b29dc
PX		 2549}
2550
2551/*
2552 * Return true if check pass, false otherwise. Error will be put
2553 * inside errp if provided.
2554 */
2555static bool migration_object_check(MigrationState *ms, Error **errp)
2556{
6b19a7d9
PX		 2557 MigrationCapabilityStatusList *head = NULL;
2558 /* Assuming all off */
2559 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2560 int i;
2561
8b0b29dc
PX		 2562 if (!migrate_params_check(&ms->parameters, errp)) {
2563 return false;
2564 }
2565
6b19a7d9
PX		 2566 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2567 if (ms->enabled_capabilities[i]) {
2568 head = migrate_cap_add(head, i, true);
2569 }
2570 }
2571
2572 ret = migrate_caps_check(cap_list, head, errp);
2573
2574 /* It works with head == NULL */
2575 qapi_free_MigrationCapabilityStatusList(head);
2576
2577 return ret;
e5cb7e76
PX		 2578}
2579
2580static const TypeInfo migration_type = {
2581 .name = TYPE_MIGRATION,
01f6e14c 2582 /*
c8d3ff38
PX		 2583 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2584 * not created using qdev_create(), it is not attached to the qdev
2585 * device tree, and it is never realized.
2586 *
2587 * TODO: Make this TYPE_OBJECT once QOM provides something like
2588 * TYPE_DEVICE's "-global" properties.
01f6e14c 2589 */
e5cb7e76
PX		 2590 .parent = TYPE_DEVICE,
2591 .class_init = migration_class_init,
2592 .class_size = sizeof(MigrationClass),
2593 .instance_size = sizeof(MigrationState),
2594 .instance_init = migration_instance_init,
b91bf5e4 2595 .instance_finalize = migration_instance_finalize,
e5cb7e76
PX		 2596};
2597
2598static void register_migration_types(void)
2599{
2600 type_register_static(&migration_type);
2601}
2602
2603type_init(register_migration_types);
QEMU mirror