]> 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: add blocktime calculation into migration-test
[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"
9af23989
MA		 34#include "qapi/qapi-commands-migration.h"
35#include "qapi/qapi-events-migration.h"
cc7a8ea7 36#include "qapi/qmp/qerror.h"
15280c36 37#include "qapi/qmp/qnull.h"
ab28bd23 38#include "qemu/rcu.h"
2c9e6fec 39#include "block.h"
be07b0ac 40#include "postcopy-ram.h"
766bd176 41#include "qemu/thread.h"
c09e5bb1 42#include "trace.h"
51180423 43#include "exec/target_page.h"
61b67d47 44#include "io/channel-buffer.h"
35a6ed4f 45#include "migration/colo.h"
4ffdb337 46#include "hw/boards.h"
9d18af93 47#include "monitor/monitor.h"
065e2813 48
dc325627 49#define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
5bb7910a 50
5b4e1eb7
JQ		 51/* Amount of time to allocate to each "chunk" of bandwidth-throttled
52 * data. */
53#define BUFFER_DELAY 100
54#define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
55
2ff30257
AA		 56/* Time in milliseconds we are allowed to stop the source,
57 * for sending the last part */
58#define DEFAULT_MIGRATE_SET_DOWNTIME 300
59
87c9cc1c
DHB		 60/* Maximum migrate downtime set to 2000 seconds */
61#define MAX_MIGRATE_DOWNTIME_SECONDS 2000
62#define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
63
8706d2d5
LL		 64/* Default compression thread count */
65#define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
3fcb38c2
LL		 66/* Default decompression thread count, usually decompression is at
67 * least 4 times as fast as compression.*/
68#define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
8706d2d5
LL		 69/*0: means nocompress, 1: best speed, ... 9: best compress ratio */
70#define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
1626fee3 71/* Define default autoconverge cpu throttle migration parameters */
d85a31d1
JH		 72#define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
73#define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
8706d2d5 74
17ad9b35 75/* Migration XBZRLE default cache size */
73af8dd8 76#define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
17ad9b35 77
68b53591
HZ		 78/* The delay time (in ms) between two COLO checkpoints
79 * Note: Please change this default value to 10000 when we support hybrid mode.
80 */
81#define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
4075fb1c 82#define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
0fb86605 83#define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
68b53591 84
99a0db9b
GH		 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));
00fa4fc8
DDAG		 158 mis_current.postcopy_remote_fds = g_array_new(FALSE, TRUE,
159 sizeof(struct PostCopyFD));
b4b076da
JQ		 160 qemu_mutex_init(&mis_current.rp_mutex);
161 qemu_event_init(&mis_current.main_thread_load_event, false);
b35ebdf0
VSO		 162
163 init_dirty_bitmap_incoming_migration();
164
b4b076da
JQ		 165 once = true;
166 }
167 return &mis_current;
bca7856a
DDAG		 168}
169
170void migration_incoming_state_destroy(void)
171{
b4b076da
JQ		 172 struct MigrationIncomingState *mis = migration_incoming_get_current();
173
3482655b 174 if (mis->to_src_file) {
660819b1
PX		 175 /* Tell source that we are done */
176 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
3482655b
PX		 177 qemu_fclose(mis->to_src_file);
178 mis->to_src_file = NULL;
179 }
180
660819b1
PX		 181 if (mis->from_src_file) {
182 qemu_fclose(mis->from_src_file);
183 mis->from_src_file = NULL;
184 }
00fa4fc8
DDAG		 185 if (mis->postcopy_remote_fds) {
186 g_array_free(mis->postcopy_remote_fds, TRUE);
187 mis->postcopy_remote_fds = NULL;
188 }
660819b1 189
5089e186 190 qemu_event_reset(&mis->main_thread_load_event);
bca7856a
DDAG		 191}
192
b05dc723
JQ		 193static void migrate_generate_event(int new_state)
194{
195 if (migrate_use_events()) {
196 qapi_event_send_migration(new_state, &error_abort);
b05dc723
JQ		 197 }
198}
199
adde220a
DDAG		 200/*
201 * Called on -incoming with a defer: uri.
202 * The migration can be started later after any parameters have been
203 * changed.
204 */
205static void deferred_incoming_migration(Error **errp)
206{
207 if (deferred_incoming) {
208 error_setg(errp, "Incoming migration already deferred");
209 }
210 deferred_incoming = true;
211}
212
da6f1790
JQ		 213/*
214 * Send a message on the return channel back to the source
215 * of the migration.
216 */
d6208e35
PX		 217static int migrate_send_rp_message(MigrationIncomingState *mis,
218 enum mig_rp_message_type message_type,
219 uint16_t len, void *data)
da6f1790 220{
d6208e35
PX		 221 int ret = 0;
222
da6f1790
JQ		 223 trace_migrate_send_rp_message((int)message_type, len);
224 qemu_mutex_lock(&mis->rp_mutex);
d6208e35
PX		 225
226 /*
227 * It's possible that the file handle got lost due to network
228 * failures.
229 */
230 if (!mis->to_src_file) {
231 ret = -EIO;
232 goto error;
233 }
234
da6f1790
JQ		 235 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
236 qemu_put_be16(mis->to_src_file, len);
237 qemu_put_buffer(mis->to_src_file, data, len);
238 qemu_fflush(mis->to_src_file);
d6208e35
PX		 239
240 /* It's possible that qemu file got error during sending */
241 ret = qemu_file_get_error(mis->to_src_file);
242
243error:
da6f1790 244 qemu_mutex_unlock(&mis->rp_mutex);
d6208e35 245 return ret;
da6f1790
JQ		 246}
247
1e2d90eb
DDAG		 248/* Request a range of pages from the source VM at the given
249 * start address.
250 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
251 * as the last request (a name must have been given previously)
252 * Start: Address offset within the RB
253 * Len: Length in bytes required - must be a multiple of pagesize
254 */
d6208e35
PX		 255int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
256 ram_addr_t start, size_t len)
1e2d90eb 257{
cb8d4c8f 258 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
1e2d90eb 259 size_t msglen = 12; /* start + len */
d6208e35 260 enum mig_rp_message_type msg_type;
1e2d90eb
DDAG		 261
262 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
263 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
264
265 if (rbname) {
266 int rbname_len = strlen(rbname);
267 assert(rbname_len < 256);
268
269 bufc[msglen++] = rbname_len;
270 memcpy(bufc + msglen, rbname, rbname_len);
271 msglen += rbname_len;
d6208e35 272 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
1e2d90eb 273 } else {
d6208e35 274 msg_type = MIG_RP_MSG_REQ_PAGES;
1e2d90eb 275 }
d6208e35
PX		 276
277 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
1e2d90eb
DDAG		 278}
279
43eaae28 280void qemu_start_incoming_migration(const char *uri, Error **errp)
5bb7910a 281{
34c9dd8e
AL		 282 const char *p;
283
7cf1fe6d 284 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
adde220a
DDAG		 285 if (!strcmp(uri, "defer")) {
286 deferred_incoming_migration(errp);
287 } else if (strstart(uri, "tcp:", &p)) {
43eaae28 288 tcp_start_incoming_migration(p, errp);
2da776db 289#ifdef CONFIG_RDMA
adde220a 290 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 291 rdma_start_incoming_migration(p, errp);
292#endif
adde220a 293 } else if (strstart(uri, "exec:", &p)) {
43eaae28 294 exec_start_incoming_migration(p, errp);
adde220a 295 } else if (strstart(uri, "unix:", &p)) {
43eaae28 296 unix_start_incoming_migration(p, errp);
adde220a 297 } else if (strstart(uri, "fd:", &p)) {
43eaae28 298 fd_start_incoming_migration(p, errp);
adde220a 299 } else {
312fd5f2 300 error_setg(errp, "unknown migration protocol: %s", uri);
8ca5e801 301 }
5bb7910a
AL		 302}
303
0aa6aefc
DL		 304static void process_incoming_migration_bh(void *opaque)
305{
306 Error *local_err = NULL;
307 MigrationIncomingState *mis = opaque;
308
a18a73d7
DDAG		 309 /* Make sure all file formats flush their mutable metadata.
310 * If we get an error here, just don't restart the VM yet. */
311 bdrv_invalidate_cache_all(&local_err);
312 if (local_err) {
313 error_report_err(local_err);
314 local_err = NULL;
315 autostart = false;
d35ff5e6
KW		 316 }
317
0aa6aefc
DL		 318 /*
319 * This must happen after all error conditions are dealt with and
320 * we're sure the VM is going to be running on this host.
321 */
322 qemu_announce_self();
323
f986c3d2
JQ		 324 if (multifd_load_cleanup(&local_err) != 0) {
325 error_report_err(local_err);
326 autostart = false;
327 }
0aa6aefc
DL		 328 /* If global state section was not received or we are in running
329 state, we need to obey autostart. Any other state is set with
330 runstate_set. */
331
b35ebdf0
VSO		 332 dirty_bitmap_mig_before_vm_start();
333
0aa6aefc
DL		 334 if (!global_state_received() ||
335 global_state_get_runstate() == RUN_STATE_RUNNING) {
336 if (autostart) {
337 vm_start();
338 } else {
339 runstate_set(RUN_STATE_PAUSED);
340 }
341 } else {
342 runstate_set(global_state_get_runstate());
343 }
0aa6aefc
DL		 344 /*
345 * This must happen after any state changes since as soon as an external
346 * observer sees this event they might start to prod at the VM assuming
347 * it's ready to use.
348 */
349 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
350 MIGRATION_STATUS_COMPLETED);
351 qemu_bh_delete(mis->bh);
352 migration_incoming_state_destroy();
353}
354
82a4da79 355static void process_incoming_migration_co(void *opaque)
511c0231 356{
b4b076da 357 MigrationIncomingState *mis = migration_incoming_get_current();
e9bef235 358 PostcopyState ps;
1c12e1f5
PB		 359 int ret;
360
4f0fae7f 361 assert(mis->from_src_file);
67f11b5c 362 mis->largest_page_size = qemu_ram_pagesize_largest();
093e3c42 363 postcopy_state_set(POSTCOPY_INCOMING_NONE);
93d7af6f
HZ		 364 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
365 MIGRATION_STATUS_ACTIVE);
4f0fae7f 366 ret = qemu_loadvm_state(mis->from_src_file);
bca7856a 367
e9bef235
DDAG		 368 ps = postcopy_state_get();
369 trace_process_incoming_migration_co_end(ret, ps);
370 if (ps != POSTCOPY_INCOMING_NONE) {
371 if (ps == POSTCOPY_INCOMING_ADVISE) {
372 /*
373 * Where a migration had postcopy enabled (and thus went to advise)
374 * but managed to complete within the precopy period, we can use
375 * the normal exit.
376 */
377 postcopy_ram_incoming_cleanup(mis);
378 } else if (ret >= 0) {
379 /*
380 * Postcopy was started, cleanup should happen at the end of the
381 * postcopy thread.
382 */
383 trace_process_incoming_migration_co_postcopy_end_main();
384 return;
385 }
386 /* Else if something went wrong then just fall out of the normal exit */
387 }
388
25d0c16f
HZ		 389 /* we get COLO info, and know if we are in COLO mode */
390 if (!ret && migration_incoming_enable_colo()) {
391 mis->migration_incoming_co = qemu_coroutine_self();
392 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
393 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
394 mis->have_colo_incoming_thread = true;
395 qemu_coroutine_yield();
396
397 /* Wait checkpoint incoming thread exit before free resource */
398 qemu_thread_join(&mis->colo_incoming_thread);
399 }
400
1c12e1f5 401 if (ret < 0) {
f986c3d2
JQ		 402 Error *local_err = NULL;
403
93d7af6f
HZ		 404 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
405 MIGRATION_STATUS_FAILED);
db80face 406 error_report("load of migration failed: %s", strerror(-ret));
3a0f2cea 407 qemu_fclose(mis->from_src_file);
f986c3d2
JQ		 408 if (multifd_load_cleanup(&local_err) != 0) {
409 error_report_err(local_err);
410 }
4aead692 411 exit(EXIT_FAILURE);
511c0231 412 }
0aa6aefc
DL		 413 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
414 qemu_bh_schedule(mis->bh);
511c0231
JQ		 415}
416
e595a01a 417static void migration_incoming_setup(QEMUFile *f)
82a4da79 418{
4f0fae7f 419 MigrationIncomingState *mis = migration_incoming_get_current();
82a4da79 420
f986c3d2
JQ		 421 if (multifd_load_setup() != 0) {
422 /* We haven't been able to create multifd threads
423 nothing better to do */
424 exit(EXIT_FAILURE);
425 }
426
4f0fae7f
JQ		 427 if (!mis->from_src_file) {
428 mis->from_src_file = f;
429 }
06ad5135 430 qemu_file_set_blocking(f, false);
e595a01a
JQ		 431}
432
433static void migration_incoming_process(void)
434{
435 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
0b8b8753 436 qemu_coroutine_enter(co);
82a4da79
PB		 437}
438
e595a01a
JQ		 439void migration_fd_process_incoming(QEMUFile *f)
440{
441 migration_incoming_setup(f);
442 migration_incoming_process();
443}
444
4f0fae7f
JQ		 445void migration_ioc_process_incoming(QIOChannel *ioc)
446{
447 MigrationIncomingState *mis = migration_incoming_get_current();
448
449 if (!mis->from_src_file) {
450 QEMUFile *f = qemu_fopen_channel_input(ioc);
451 migration_fd_process_incoming(f);
452 }
453 /* We still only have a single channel. Nothing to do here yet */
454}
455
428d8908
JQ		 456/**
457 * @migration_has_all_channels: We have received all channels that we need
458 *
459 * Returns true when we have got connections to all the channels that
460 * we need for migration.
461 */
462bool migration_has_all_channels(void)
463{
464 return true;
465}
466
6decec93
DDAG		 467/*
468 * Send a 'SHUT' message on the return channel with the given value
469 * to indicate that we've finished with the RP. Non-0 value indicates
470 * error.
471 */
472void migrate_send_rp_shut(MigrationIncomingState *mis,
473 uint32_t value)
474{
475 uint32_t buf;
476
477 buf = cpu_to_be32(value);
478 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
479}
480
481/*
482 * Send a 'PONG' message on the return channel with the given value
483 * (normally in response to a 'PING')
484 */
485void migrate_send_rp_pong(MigrationIncomingState *mis,
486 uint32_t value)
487{
488 uint32_t buf;
489
490 buf = cpu_to_be32(value);
491 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
492}
493
bbf6da32
OW		 494MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
495{
496 MigrationCapabilityStatusList *head = NULL;
497 MigrationCapabilityStatusList *caps;
498 MigrationState *s = migrate_get_current();
499 int i;
500
387eedeb 501 caps = NULL; /* silence compiler warning */
7fb1cf16 502 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
ed1701c6
DDAG		 503#ifndef CONFIG_LIVE_BLOCK_MIGRATION
504 if (i == MIGRATION_CAPABILITY_BLOCK) {
505 continue;
506 }
507#endif
bbf6da32
OW		 508 if (head == NULL) {
509 head = g_malloc0(sizeof(*caps));
510 caps = head;
511 } else {
512 caps->next = g_malloc0(sizeof(*caps));
513 caps = caps->next;
514 }
515 caps->value =
516 g_malloc(sizeof(*caps->value));
517 caps->value->capability = i;
518 caps->value->state = s->enabled_capabilities[i];
519 }
520
521 return head;
522}
523
85de8323
LL		 524MigrationParameters *qmp_query_migrate_parameters(Error **errp)
525{
526 MigrationParameters *params;
527 MigrationState *s = migrate_get_current();
528
e87fae4c 529 /* TODO use QAPI_CLONE() instead of duplicating it inline */
85de8323 530 params = g_malloc0(sizeof(*params));
de63ab61 531 params->has_compress_level = true;
2594f56d 532 params->compress_level = s->parameters.compress_level;
de63ab61 533 params->has_compress_threads = true;
2594f56d 534 params->compress_threads = s->parameters.compress_threads;
de63ab61 535 params->has_decompress_threads = true;
2594f56d 536 params->decompress_threads = s->parameters.decompress_threads;
de63ab61 537 params->has_cpu_throttle_initial = true;
2594f56d 538 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
de63ab61 539 params->has_cpu_throttle_increment = true;
2594f56d 540 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
8cc99dcd 541 params->has_tls_creds = true;
69ef1f36 542 params->tls_creds = g_strdup(s->parameters.tls_creds);
8cc99dcd 543 params->has_tls_hostname = true;
69ef1f36 544 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
2ff30257
AA		 545 params->has_max_bandwidth = true;
546 params->max_bandwidth = s->parameters.max_bandwidth;
547 params->has_downtime_limit = true;
548 params->downtime_limit = s->parameters.downtime_limit;
fe39a4d4 549 params->has_x_checkpoint_delay = true;
68b53591 550 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
2833c59b
JQ		 551 params->has_block_incremental = true;
552 params->block_incremental = s->parameters.block_incremental;
4075fb1c
JQ		 553 params->has_x_multifd_channels = true;
554 params->x_multifd_channels = s->parameters.x_multifd_channels;
0fb86605
JQ		 555 params->has_x_multifd_page_count = true;
556 params->x_multifd_page_count = s->parameters.x_multifd_page_count;
73af8dd8
JQ		 557 params->has_xbzrle_cache_size = true;
558 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
85de8323
LL		 559
560 return params;
561}
562
f6844b99
DDAG		 563/*
564 * Return true if we're already in the middle of a migration
565 * (i.e. any of the active or setup states)
566 */
567static bool migration_is_setup_or_active(int state)
568{
569 switch (state) {
570 case MIGRATION_STATUS_ACTIVE:
9ec055ae 571 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
f6844b99 572 case MIGRATION_STATUS_SETUP:
31e06077
DDAG		 573 case MIGRATION_STATUS_PRE_SWITCHOVER:
574 case MIGRATION_STATUS_DEVICE:
f6844b99
DDAG		 575 return true;
576
577 default:
578 return false;
579
580 }
581}
582
a22463a5
DDAG		 583static void populate_ram_info(MigrationInfo *info, MigrationState *s)
584{
585 info->has_ram = true;
586 info->ram = g_malloc0(sizeof(*info->ram));
9360447d 587 info->ram->transferred = ram_counters.transferred;
a22463a5 588 info->ram->total = ram_bytes_total();
9360447d 589 info->ram->duplicate = ram_counters.duplicate;
bedf53c1
JQ		 590 /* legacy value. It is not used anymore */
591 info->ram->skipped = 0;
9360447d
JQ		 592 info->ram->normal = ram_counters.normal;
593 info->ram->normal_bytes = ram_counters.normal *
20afaed9 594 qemu_target_page_size();
a22463a5 595 info->ram->mbps = s->mbps;
9360447d
JQ		 596 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
597 info->ram->postcopy_requests = ram_counters.postcopy_requests;
030ce1f8 598 info->ram->page_size = qemu_target_page_size();
a22463a5 599
114f5aee
JQ		 600 if (migrate_use_xbzrle()) {
601 info->has_xbzrle_cache = true;
602 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
603 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
9360447d
JQ		 604 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
605 info->xbzrle_cache->pages = xbzrle_counters.pages;
606 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
607 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
608 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
114f5aee
JQ		 609 }
610
338182c8
JQ		 611 if (cpu_throttle_active()) {
612 info->has_cpu_throttle_percentage = true;
613 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
614 }
615
a22463a5
DDAG		 616 if (s->state != MIGRATION_STATUS_COMPLETED) {
617 info->ram->remaining = ram_bytes_remaining();
9360447d 618 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
a22463a5
DDAG		 619 }
620}
621
930ac04c
JQ		 622static void populate_disk_info(MigrationInfo *info)
623{
624 if (blk_mig_active()) {
625 info->has_disk = true;
626 info->disk = g_malloc0(sizeof(*info->disk));
627 info->disk->transferred = blk_mig_bytes_transferred();
628 info->disk->remaining = blk_mig_bytes_remaining();
629 info->disk->total = blk_mig_bytes_total();
630 }
631}
632
ee86981b 633MigrationInfo *qmp_query_migrate(Error **errp)
5bb7910a 634{
ee86981b 635 MigrationInfo *info = g_malloc0(sizeof(*info));
17549e84
JQ		 636 MigrationState *s = migrate_get_current();
637
638 switch (s->state) {
31194731 639 case MIGRATION_STATUS_NONE:
17549e84
JQ		 640 /* no migration has happened ever */
641 break;
31194731 642 case MIGRATION_STATUS_SETUP:
29ae8a41 643 info->has_status = true;
ed4fbd10 644 info->has_total_time = false;
29ae8a41 645 break;
31194731
HZ		 646 case MIGRATION_STATUS_ACTIVE:
647 case MIGRATION_STATUS_CANCELLING:
9ec055ae 648 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
31e06077
DDAG		 649 case MIGRATION_STATUS_PRE_SWITCHOVER:
650 case MIGRATION_STATUS_DEVICE:
c8f9f4f4 651 /* TODO add some postcopy stats */
9ec055ae
DDAG		 652 info->has_status = true;
653 info->has_total_time = true;
654 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
4af246a3 655 - s->start_time;
9ec055ae
DDAG		 656 info->has_expected_downtime = true;
657 info->expected_downtime = s->expected_downtime;
658 info->has_setup_time = true;
659 info->setup_time = s->setup_time;
660
a22463a5 661 populate_ram_info(info, s);
930ac04c 662 populate_disk_info(info);
17549e84 663 break;
0b827d5e
HZ		 664 case MIGRATION_STATUS_COLO:
665 info->has_status = true;
666 /* TODO: display COLO specific information (checkpoint info etc.) */
667 break;
31194731 668 case MIGRATION_STATUS_COMPLETED:
791e7c82 669 info->has_status = true;
00c14997 670 info->has_total_time = true;
7aa939af 671 info->total_time = s->total_time;
9c5a9fcf
JQ		 672 info->has_downtime = true;
673 info->downtime = s->downtime;
ed4fbd10
MH		 674 info->has_setup_time = true;
675 info->setup_time = s->setup_time;
d5f8a570 676
a22463a5 677 populate_ram_info(info, s);
17549e84 678 break;
31194731 679 case MIGRATION_STATUS_FAILED:
791e7c82 680 info->has_status = true;
d59ce6f3
DB		 681 if (s->error) {
682 info->has_error_desc = true;
683 info->error_desc = g_strdup(error_get_pretty(s->error));
684 }
17549e84 685 break;
31194731 686 case MIGRATION_STATUS_CANCELLED:
791e7c82 687 info->has_status = true;
17549e84 688 break;
5bb7910a 689 }
cde63fbe 690 info->status = s->state;
ee86981b
PM		 691
692 return info;
5bb7910a
AL		 693}
694
4a84214e
PX		 695/**
696 * @migration_caps_check - check capability validity
697 *
698 * @cap_list: old capability list, array of bool
699 * @params: new capabilities to be applied soon
700 * @errp: set *errp if the check failed, with reason
701 *
702 * Returns true if check passed, otherwise false.
703 */
704static bool migrate_caps_check(bool *cap_list,
705 MigrationCapabilityStatusList *params,
706 Error **errp)
00458433 707{
00458433 708 MigrationCapabilityStatusList *cap;
4a84214e 709 bool old_postcopy_cap;
d7651f15 710 MigrationIncomingState *mis = migration_incoming_get_current();
00458433 711
4a84214e 712 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
00458433
OW		 713
714 for (cap = params; cap; cap = cap->next) {
4a84214e
PX		 715 cap_list[cap->value->capability] = cap->value->state;
716 }
717
ed1701c6 718#ifndef CONFIG_LIVE_BLOCK_MIGRATION
4a84214e
PX		 719 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
720 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
721 "block migration");
722 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
723 return false;
00458433 724 }
4a84214e 725#endif
53dd370c 726
4a84214e
PX		 727 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
728 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
53dd370c
DDAG		 729 /* The decompression threads asynchronously write into RAM
730 * rather than use the atomic copies needed to avoid
731 * userfaulting. It should be possible to fix the decompression
732 * threads for compatibility in future.
733 */
4a84214e
PX		 734 error_setg(errp, "Postcopy is not currently compatible "
735 "with compression");
736 return false;
53dd370c 737 }
4a84214e 738
096631bd
DDAG		 739 /* This check is reasonably expensive, so only when it's being
740 * set the first time, also it's only the destination that needs
741 * special support.
742 */
743 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
d7651f15 744 !postcopy_ram_supported_by_host(mis)) {
096631bd
DDAG		 745 /* postcopy_ram_supported_by_host will have emitted a more
746 * detailed message
747 */
4a84214e
PX		 748 error_setg(errp, "Postcopy is not supported");
749 return false;
096631bd 750 }
53dd370c 751 }
4a84214e
PX		 752
753 return true;
754}
755
756void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
757 Error **errp)
758{
759 MigrationState *s = migrate_get_current();
760 MigrationCapabilityStatusList *cap;
dd0ee30c 761 bool cap_list[MIGRATION_CAPABILITY__MAX];
4a84214e
PX		 762
763 if (migration_is_setup_or_active(s->state)) {
764 error_setg(errp, QERR_MIGRATION_ACTIVE);
765 return;
766 }
767
dd0ee30c
PX		 768 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
769 if (!migrate_caps_check(cap_list, params, errp)) {
4a84214e
PX		 770 return;
771 }
772
773 for (cap = params; cap; cap = cap->next) {
774 s->enabled_capabilities[cap->value->capability] = cap->value->state;
775 }
00458433
OW		 776}
777
16d063bc
PX		 778/*
779 * Check whether the parameters are valid. Error will be put into errp
780 * (if provided). Return true if valid, otherwise false.
781 */
782static bool migrate_params_check(MigrationParameters *params, Error **errp)
85de8323 783{
7f375e04 784 if (params->has_compress_level &&
741d4086 785 (params->compress_level > 9)) {
c6bd8c70
MA		 786 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
787 "is invalid, it should be in the range of 0 to 9");
16d063bc 788 return false;
85de8323 789 }
16d063bc 790
741d4086 791 if (params->has_compress_threads && (params->compress_threads < 1)) {
c6bd8c70
MA		 792 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
793 "compress_threads",
794 "is invalid, it should be in the range of 1 to 255");
16d063bc 795 return false;
85de8323 796 }
16d063bc 797
741d4086 798 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
c6bd8c70
MA		 799 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
800 "decompress_threads",
801 "is invalid, it should be in the range of 1 to 255");
16d063bc 802 return false;
85de8323 803 }
16d063bc 804
7f375e04
EB		 805 if (params->has_cpu_throttle_initial &&
806 (params->cpu_throttle_initial < 1 ||
807 params->cpu_throttle_initial > 99)) {
1626fee3 808 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 809 "cpu_throttle_initial",
1626fee3 810 "an integer in the range of 1 to 99");
16d063bc 811 return false;
1626fee3 812 }
16d063bc 813
7f375e04
EB		 814 if (params->has_cpu_throttle_increment &&
815 (params->cpu_throttle_increment < 1 ||
816 params->cpu_throttle_increment > 99)) {
1626fee3 817 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 818 "cpu_throttle_increment",
1626fee3 819 "an integer in the range of 1 to 99");
16d063bc 820 return false;
1626fee3 821 }
16d063bc 822
741d4086 823 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
2ff30257
AA		 824 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
825 " range of 0 to %zu bytes/second", SIZE_MAX);
16d063bc 826 return false;
2ff30257 827 }
16d063bc 828
2ff30257 829 if (params->has_downtime_limit &&
741d4086 830 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
87c9cc1c
DHB		 831 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
832 "the range of 0 to %d milliseconds",
833 MAX_MIGRATE_DOWNTIME);
16d063bc 834 return false;
2ff30257 835 }
16d063bc 836
741d4086
JQ		 837 /* x_checkpoint_delay is now always positive */
838
839 if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) {
4075fb1c
JQ		 840 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
841 "multifd_channels",
842 "is invalid, it should be in the range of 1 to 255");
843 return false;
844 }
0fb86605 845 if (params->has_x_multifd_page_count &&
741d4086
JQ		 846 (params->x_multifd_page_count < 1 ||
847 params->x_multifd_page_count > 10000)) {
0fb86605
JQ		 848 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
849 "multifd_page_count",
850 "is invalid, it should be in the range of 1 to 10000");
851 return false;
852 }
16d063bc 853
73af8dd8
JQ		 854 if (params->has_xbzrle_cache_size &&
855 (params->xbzrle_cache_size < qemu_target_page_size() ||
856 !is_power_of_2(params->xbzrle_cache_size))) {
857 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
858 "xbzrle_cache_size",
859 "is invalid, it should be bigger than target page size"
860 " and a power of two");
861 return false;
862 }
863
16d063bc
PX		 864 return true;
865}
866
1bda8b3c
MA		 867static void migrate_params_test_apply(MigrateSetParameters *params,
868 MigrationParameters *dest)
869{
870 *dest = migrate_get_current()->parameters;
871
872 /* TODO use QAPI_CLONE() instead of duplicating it inline */
873
874 if (params->has_compress_level) {
875 dest->compress_level = params->compress_level;
876 }
877
878 if (params->has_compress_threads) {
879 dest->compress_threads = params->compress_threads;
880 }
881
882 if (params->has_decompress_threads) {
883 dest->decompress_threads = params->decompress_threads;
884 }
885
886 if (params->has_cpu_throttle_initial) {
887 dest->cpu_throttle_initial = params->cpu_throttle_initial;
888 }
889
890 if (params->has_cpu_throttle_increment) {
891 dest->cpu_throttle_increment = params->cpu_throttle_increment;
892 }
893
894 if (params->has_tls_creds) {
01fa5598
MA		 895 assert(params->tls_creds->type == QTYPE_QSTRING);
896 dest->tls_creds = g_strdup(params->tls_creds->u.s);
1bda8b3c
MA		 897 }
898
899 if (params->has_tls_hostname) {
01fa5598
MA		 900 assert(params->tls_hostname->type == QTYPE_QSTRING);
901 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1bda8b3c
MA		 902 }
903
904 if (params->has_max_bandwidth) {
905 dest->max_bandwidth = params->max_bandwidth;
906 }
907
908 if (params->has_downtime_limit) {
909 dest->downtime_limit = params->downtime_limit;
910 }
911
912 if (params->has_x_checkpoint_delay) {
913 dest->x_checkpoint_delay = params->x_checkpoint_delay;
914 }
915
916 if (params->has_block_incremental) {
917 dest->block_incremental = params->block_incremental;
918 }
5e7577a1
JQ		 919 if (params->has_x_multifd_channels) {
920 dest->x_multifd_channels = params->x_multifd_channels;
921 }
922 if (params->has_x_multifd_page_count) {
923 dest->x_multifd_page_count = params->x_multifd_page_count;
924 }
73af8dd8
JQ		 925 if (params->has_xbzrle_cache_size) {
926 dest->xbzrle_cache_size = params->xbzrle_cache_size;
927 }
1bda8b3c
MA		 928}
929
73af8dd8 930static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
16d063bc
PX		 931{
932 MigrationState *s = migrate_get_current();
933
e87fae4c
MA		 934 /* TODO use QAPI_CLONE() instead of duplicating it inline */
935
7f375e04
EB		 936 if (params->has_compress_level) {
937 s->parameters.compress_level = params->compress_level;
85de8323 938 }
476c72aa 939
7f375e04
EB		 940 if (params->has_compress_threads) {
941 s->parameters.compress_threads = params->compress_threads;
85de8323 942 }
476c72aa 943
7f375e04
EB		 944 if (params->has_decompress_threads) {
945 s->parameters.decompress_threads = params->decompress_threads;
85de8323 946 }
476c72aa 947
7f375e04
EB		 948 if (params->has_cpu_throttle_initial) {
949 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1626fee3 950 }
476c72aa 951
7f375e04
EB		 952 if (params->has_cpu_throttle_increment) {
953 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1626fee3 954 }
476c72aa 955
7f375e04 956 if (params->has_tls_creds) {
69ef1f36 957 g_free(s->parameters.tls_creds);
01fa5598
MA		 958 assert(params->tls_creds->type == QTYPE_QSTRING);
959 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
69ef1f36 960 }
476c72aa 961
7f375e04 962 if (params->has_tls_hostname) {
69ef1f36 963 g_free(s->parameters.tls_hostname);
01fa5598
MA		 964 assert(params->tls_hostname->type == QTYPE_QSTRING);
965 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
69ef1f36 966 }
476c72aa 967
2ff30257
AA		 968 if (params->has_max_bandwidth) {
969 s->parameters.max_bandwidth = params->max_bandwidth;
970 if (s->to_dst_file) {
971 qemu_file_set_rate_limit(s->to_dst_file,
972 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
973 }
974 }
476c72aa 975
2ff30257
AA		 976 if (params->has_downtime_limit) {
977 s->parameters.downtime_limit = params->downtime_limit;
978 }
68b53591
HZ		 979
980 if (params->has_x_checkpoint_delay) {
981 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
479125d5
HZ		 982 if (migration_in_colo_state()) {
983 colo_checkpoint_notify(s);
984 }
68b53591 985 }
476c72aa 986
2833c59b
JQ		 987 if (params->has_block_incremental) {
988 s->parameters.block_incremental = params->block_incremental;
989 }
4075fb1c
JQ		 990 if (params->has_x_multifd_channels) {
991 s->parameters.x_multifd_channels = params->x_multifd_channels;
992 }
0fb86605
JQ		 993 if (params->has_x_multifd_page_count) {
994 s->parameters.x_multifd_page_count = params->x_multifd_page_count;
995 }
73af8dd8
JQ		 996 if (params->has_xbzrle_cache_size) {
997 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
998 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
999 }
85de8323
LL		 1000}
1001
1bda8b3c 1002void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
476c72aa 1003{
1bda8b3c
MA		 1004 MigrationParameters tmp;
1005
01fa5598
MA		 1006 /* TODO Rewrite "" to null instead */
1007 if (params->has_tls_creds
1008 && params->tls_creds->type == QTYPE_QNULL) {
1009 QDECREF(params->tls_creds->u.n);
1010 params->tls_creds->type = QTYPE_QSTRING;
1011 params->tls_creds->u.s = strdup("");
1012 }
1013 /* TODO Rewrite "" to null instead */
1014 if (params->has_tls_hostname
1015 && params->tls_hostname->type == QTYPE_QNULL) {
1016 QDECREF(params->tls_hostname->u.n);
1017 params->tls_hostname->type = QTYPE_QSTRING;
1018 params->tls_hostname->u.s = strdup("");
1019 }
1020
1bda8b3c
MA		 1021 migrate_params_test_apply(params, &tmp);
1022
1023 if (!migrate_params_check(&tmp, errp)) {
476c72aa
PX		 1024 /* Invalid parameter */
1025 return;
1026 }
1027
73af8dd8 1028 migrate_params_apply(params, errp);
476c72aa
PX		 1029}
1030
2594f56d 1031
4886a1bc
DDAG		 1032void qmp_migrate_start_postcopy(Error **errp)
1033{
1034 MigrationState *s = migrate_get_current();
1035
16b0fd32 1036 if (!migrate_postcopy()) {
a54d340b 1037 error_setg(errp, "Enable postcopy with migrate_set_capability before"
4886a1bc
DDAG		 1038 " the start of migration");
1039 return;
1040 }
1041
1042 if (s->state == MIGRATION_STATUS_NONE) {
1043 error_setg(errp, "Postcopy must be started after migration has been"
1044 " started");
1045 return;
1046 }
1047 /*
1048 * we don't error if migration has finished since that would be racy
1049 * with issuing this command.
1050 */
1051 atomic_set(&s->start_postcopy, true);
1052}
1053
065e2813
AL		 1054/* shared migration helpers */
1055
48781e5b 1056void migrate_set_state(int *state, int old_state, int new_state)
51cf4c1a 1057{
a31fedee 1058 assert(new_state < MIGRATION_STATUS__MAX);
48781e5b 1059 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
a31fedee 1060 trace_migrate_set_state(MigrationStatus_str(new_state));
b05dc723 1061 migrate_generate_event(new_state);
51cf4c1a
Z		 1062 }
1063}
1064
4e4a3d3a
PX		 1065static MigrationCapabilityStatusList *migrate_cap_add(
1066 MigrationCapabilityStatusList *list,
1067 MigrationCapability index,
1068 bool state)
2833c59b
JQ		 1069{
1070 MigrationCapabilityStatusList *cap;
1071
1072 cap = g_new0(MigrationCapabilityStatusList, 1);
1073 cap->value = g_new0(MigrationCapabilityStatus, 1);
4e4a3d3a
PX		 1074 cap->value->capability = index;
1075 cap->value->state = state;
1076 cap->next = list;
1077
1078 return cap;
1079}
1080
1081void migrate_set_block_enabled(bool value, Error **errp)
1082{
1083 MigrationCapabilityStatusList *cap;
1084
1085 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
2833c59b
JQ		 1086 qmp_migrate_set_capabilities(cap, errp);
1087 qapi_free_MigrationCapabilityStatusList(cap);
1088}
1089
1090static void migrate_set_block_incremental(MigrationState *s, bool value)
1091{
1092 s->parameters.block_incremental = value;
1093}
1094
1095static void block_cleanup_parameters(MigrationState *s)
1096{
1097 if (s->must_remove_block_options) {
1098 /* setting to false can never fail */
1099 migrate_set_block_enabled(false, &error_abort);
1100 migrate_set_block_incremental(s, false);
1101 s->must_remove_block_options = false;
1102 }
1103}
1104
bb1fadc4 1105static void migrate_fd_cleanup(void *opaque)
065e2813 1106{
bb1fadc4
PB		 1107 MigrationState *s = opaque;
1108
1109 qemu_bh_delete(s->cleanup_bh);
1110 s->cleanup_bh = NULL;
1111
0ceccd85
PX		 1112 qemu_savevm_state_cleanup();
1113
89a02a9f 1114 if (s->to_dst_file) {
f986c3d2
JQ		 1115 Error *local_err = NULL;
1116
9013dca5 1117 trace_migrate_fd_cleanup();
404a7c05 1118 qemu_mutex_unlock_iothread();
1d34e4bf
DDAG		 1119 if (s->migration_thread_running) {
1120 qemu_thread_join(&s->thread);
1121 s->migration_thread_running = false;
1122 }
404a7c05
PB		 1123 qemu_mutex_lock_iothread();
1124
f986c3d2
JQ		 1125 if (multifd_save_cleanup(&local_err) != 0) {
1126 error_report_err(local_err);
1127 }
89a02a9f
HZ		 1128 qemu_fclose(s->to_dst_file);
1129 s->to_dst_file = NULL;
065e2813
AL		 1130 }
1131
9ec055ae
DDAG		 1132 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1133 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
7a2c1721 1134
94f5a437 1135 if (s->state == MIGRATION_STATUS_CANCELLING) {
48781e5b 1136 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
94f5a437 1137 MIGRATION_STATUS_CANCELLED);
7a2c1721 1138 }
a3fa1d78 1139
87db1a7d
JQ		 1140 if (s->error) {
1141 /* It is used on info migrate. We can't free it */
1142 error_report_err(error_copy(s->error));
1143 }
a3fa1d78 1144 notifier_list_notify(&migration_state_notifiers, s);
2833c59b 1145 block_cleanup_parameters(s);
065e2813
AL		 1146}
1147
87db1a7d
JQ		 1148void migrate_set_error(MigrationState *s, const Error *error)
1149{
1150 qemu_mutex_lock(&s->error_mutex);
1151 if (!s->error) {
1152 s->error = error_copy(error);
1153 }
1154 qemu_mutex_unlock(&s->error_mutex);
1155}
1156
d59ce6f3 1157void migrate_fd_error(MigrationState *s, const Error *error)
065e2813 1158{
25174055 1159 trace_migrate_fd_error(error_get_pretty(error));
89a02a9f 1160 assert(s->to_dst_file == NULL);
48781e5b
HZ		 1161 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1162 MIGRATION_STATUS_FAILED);
87db1a7d 1163 migrate_set_error(s, error);
458cf28e
JQ		 1164}
1165
0edda1c4 1166static void migrate_fd_cancel(MigrationState *s)
065e2813 1167{
6f2b811a 1168 int old_state ;
89a02a9f 1169 QEMUFile *f = migrate_get_current()->to_dst_file;
9013dca5 1170 trace_migrate_fd_cancel();
065e2813 1171
70b20477
DDAG		 1172 if (s->rp_state.from_dst_file) {
1173 /* shutdown the rp socket, so causing the rp thread to shutdown */
1174 qemu_file_shutdown(s->rp_state.from_dst_file);
1175 }
1176
6f2b811a
Z		 1177 do {
1178 old_state = s->state;
f6844b99 1179 if (!migration_is_setup_or_active(old_state)) {
6f2b811a
Z		 1180 break;
1181 }
a7b36b48
DDAG		 1182 /* If the migration is paused, kick it out of the pause */
1183 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1184 qemu_sem_post(&s->pause_sem);
1185 }
48781e5b 1186 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
31194731 1187 } while (s->state != MIGRATION_STATUS_CANCELLING);
a26ba26e
DDAG		 1188
1189 /*
1190 * If we're unlucky the migration code might be stuck somewhere in a
1191 * send/write while the network has failed and is waiting to timeout;
1192 * if we've got shutdown(2) available then we can force it to quit.
1193 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1194 * called in a bh, so there is no race against this cancel.
1195 */
31194731 1196 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
a26ba26e
DDAG		 1197 qemu_file_shutdown(f);
1198 }
1d2acc31
HZ		 1199 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1200 Error *local_err = NULL;
1201
1202 bdrv_invalidate_cache_all(&local_err);
1203 if (local_err) {
1204 error_report_err(local_err);
1205 } else {
1206 s->block_inactive = false;
1207 }
1208 }
065e2813
AL		 1209}
1210
99a0db9b
GH		 1211void add_migration_state_change_notifier(Notifier *notify)
1212{
1213 notifier_list_add(&migration_state_notifiers, notify);
1214}
1215
1216void remove_migration_state_change_notifier(Notifier *notify)
1217{
31552529 1218 notifier_remove(notify);
99a0db9b
GH		 1219}
1220
02edd2e7 1221bool migration_in_setup(MigrationState *s)
afe2df69 1222{
31194731 1223 return s->state == MIGRATION_STATUS_SETUP;
afe2df69
GH		 1224}
1225
7073693b 1226bool migration_has_finished(MigrationState *s)
99a0db9b 1227{
31194731 1228 return s->state == MIGRATION_STATUS_COMPLETED;
99a0db9b 1229}
0edda1c4 1230
afe2df69
GH		 1231bool migration_has_failed(MigrationState *s)
1232{
31194731
HZ		 1233 return (s->state == MIGRATION_STATUS_CANCELLED ||
1234 s->state == MIGRATION_STATUS_FAILED);
afe2df69
GH		 1235}
1236
5727309d 1237bool migration_in_postcopy(void)
9ec055ae 1238{
5727309d
JQ		 1239 MigrationState *s = migrate_get_current();
1240
9ec055ae
DDAG		 1241 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1242}
1243
b82fc321
DDAG		 1244bool migration_in_postcopy_after_devices(MigrationState *s)
1245{
5727309d 1246 return migration_in_postcopy() && s->postcopy_after_devices;
b82fc321
DDAG		 1247}
1248
fab35005 1249bool migration_is_idle(void)
fe44dc91 1250{
fab35005 1251 MigrationState *s = migrate_get_current();
fe44dc91
AA		 1252
1253 switch (s->state) {
1254 case MIGRATION_STATUS_NONE:
1255 case MIGRATION_STATUS_CANCELLED:
1256 case MIGRATION_STATUS_COMPLETED:
1257 case MIGRATION_STATUS_FAILED:
1258 return true;
1259 case MIGRATION_STATUS_SETUP:
1260 case MIGRATION_STATUS_CANCELLING:
1261 case MIGRATION_STATUS_ACTIVE:
1262 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1263 case MIGRATION_STATUS_COLO:
31e06077
DDAG		 1264 case MIGRATION_STATUS_PRE_SWITCHOVER:
1265 case MIGRATION_STATUS_DEVICE:
fe44dc91
AA		 1266 return false;
1267 case MIGRATION_STATUS__MAX:
1268 g_assert_not_reached();
1269 }
1270
1271 return false;
1272}
1273
3e0c8050 1274void migrate_init(MigrationState *s)
0edda1c4 1275{
389775d1
DDAG		 1276 /*
1277 * Reinitialise all migration state, except
1278 * parameters/capabilities that the user set, and
1279 * locks.
1280 */
1281 s->bytes_xfer = 0;
1282 s->xfer_limit = 0;
1283 s->cleanup_bh = 0;
89a02a9f 1284 s->to_dst_file = NULL;
389775d1 1285 s->state = MIGRATION_STATUS_NONE;
389775d1
DDAG		 1286 s->rp_state.from_dst_file = NULL;
1287 s->rp_state.error = false;
1288 s->mbps = 0.0;
1289 s->downtime = 0;
1290 s->expected_downtime = 0;
389775d1 1291 s->setup_time = 0;
389775d1 1292 s->start_postcopy = false;
b82fc321 1293 s->postcopy_after_devices = false;
389775d1 1294 s->migration_thread_running = false;
d59ce6f3
DB		 1295 error_free(s->error);
1296 s->error = NULL;
389775d1 1297
48781e5b 1298 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
0edda1c4 1299
4af246a3
PX		 1300 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1301 s->total_time = 0;
7287cbd4 1302 s->vm_was_running = false;
b15df1ae
PX		 1303 s->iteration_initial_bytes = 0;
1304 s->threshold_size = 0;
0edda1c4 1305}
cab30143 1306
fa2756b7
AL		 1307static GSList *migration_blockers;
1308
fe44dc91 1309int migrate_add_blocker(Error *reason, Error **errp)
fa2756b7 1310{
3df663e5 1311 if (migrate_get_current()->only_migratable) {
b67b8c3a
AA		 1312 error_propagate(errp, error_copy(reason));
1313 error_prepend(errp, "disallowing migration blocker "
1314 "(--only_migratable) for: ");
1315 return -EACCES;
1316 }
1317
fab35005 1318 if (migration_is_idle()) {
fe44dc91
AA		 1319 migration_blockers = g_slist_prepend(migration_blockers, reason);
1320 return 0;
1321 }
1322
1323 error_propagate(errp, error_copy(reason));
1324 error_prepend(errp, "disallowing migration blocker (migration in "
1325 "progress) for: ");
1326 return -EBUSY;
fa2756b7
AL		 1327}
1328
1329void migrate_del_blocker(Error *reason)
1330{
1331 migration_blockers = g_slist_remove(migration_blockers, reason);
1332}
1333
bf1ae1f4
DDAG		 1334void qmp_migrate_incoming(const char *uri, Error **errp)
1335{
1336 Error *local_err = NULL;
4debb5f5 1337 static bool once = true;
bf1ae1f4
DDAG		 1338
1339 if (!deferred_incoming) {
4debb5f5 1340 error_setg(errp, "For use with '-incoming defer'");
bf1ae1f4
DDAG		 1341 return;
1342 }
4debb5f5
DDAG		 1343 if (!once) {
1344 error_setg(errp, "The incoming migration has already been started");
1345 }
bf1ae1f4
DDAG		 1346
1347 qemu_start_incoming_migration(uri, &local_err);
1348
1349 if (local_err) {
1350 error_propagate(errp, local_err);
1351 return;
1352 }
1353
4debb5f5 1354 once = false;
bf1ae1f4
DDAG		 1355}
1356
24f3902b
GK		 1357bool migration_is_blocked(Error **errp)
1358{
1359 if (qemu_savevm_state_blocked(errp)) {
1360 return true;
1361 }
1362
1363 if (migration_blockers) {
250561e1 1364 error_propagate(errp, error_copy(migration_blockers->data));
24f3902b
GK		 1365 return true;
1366 }
1367
1368 return false;
1369}
1370
e1c37d0e
LC		 1371void qmp_migrate(const char *uri, bool has_blk, bool blk,
1372 bool has_inc, bool inc, bool has_detach, bool detach,
1373 Error **errp)
cab30143 1374{
be7059cd 1375 Error *local_err = NULL;
17549e84 1376 MigrationState *s = migrate_get_current();
cab30143 1377 const char *p;
cab30143 1378
f6844b99 1379 if (migration_is_setup_or_active(s->state) ||
0b827d5e
HZ		 1380 s->state == MIGRATION_STATUS_CANCELLING ||
1381 s->state == MIGRATION_STATUS_COLO) {
c6bd8c70 1382 error_setg(errp, QERR_MIGRATION_ACTIVE);
e1c37d0e 1383 return;
cab30143 1384 }
ca99993a
DDAG		 1385 if (runstate_check(RUN_STATE_INMIGRATE)) {
1386 error_setg(errp, "Guest is waiting for an incoming migration");
1387 return;
1388 }
1389
24f3902b 1390 if (migration_is_blocked(errp)) {
e1c37d0e 1391 return;
fa2756b7
AL		 1392 }
1393
2833c59b
JQ		 1394 if ((has_blk && blk) || (has_inc && inc)) {
1395 if (migrate_use_block() || migrate_use_block_incremental()) {
1396 error_setg(errp, "Command options are incompatible with "
1397 "current migration capabilities");
1398 return;
1399 }
1400 migrate_set_block_enabled(true, &local_err);
1401 if (local_err) {
1402 error_propagate(errp, local_err);
1403 return;
1404 }
1405 s->must_remove_block_options = true;
1406 }
1407
1408 if (has_inc && inc) {
1409 migrate_set_block_incremental(s, true);
1410 }
1411
3e0c8050 1412 migrate_init(s);
cab30143
JQ		 1413
1414 if (strstart(uri, "tcp:", &p)) {
f37afb5a 1415 tcp_start_outgoing_migration(s, p, &local_err);
2da776db 1416#ifdef CONFIG_RDMA
41310c68 1417 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 1418 rdma_start_outgoing_migration(s, p, &local_err);
1419#endif
cab30143 1420 } else if (strstart(uri, "exec:", &p)) {
f37afb5a 1421 exec_start_outgoing_migration(s, p, &local_err);
cab30143 1422 } else if (strstart(uri, "unix:", &p)) {
f37afb5a 1423 unix_start_outgoing_migration(s, p, &local_err);
cab30143 1424 } else if (strstart(uri, "fd:", &p)) {
f37afb5a 1425 fd_start_outgoing_migration(s, p, &local_err);
99a0db9b 1426 } else {
c6bd8c70
MA		 1427 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1428 "a valid migration protocol");
48781e5b
HZ		 1429 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1430 MIGRATION_STATUS_FAILED);
09576e74 1431 block_cleanup_parameters(s);
e1c37d0e 1432 return;
cab30143
JQ		 1433 }
1434
f37afb5a 1435 if (local_err) {
d59ce6f3 1436 migrate_fd_error(s, local_err);
f37afb5a 1437 error_propagate(errp, local_err);
e1c37d0e 1438 return;
1299c631 1439 }
cab30143
JQ		 1440}
1441
6cdedb07 1442void qmp_migrate_cancel(Error **errp)
cab30143 1443{
17549e84 1444 migrate_fd_cancel(migrate_get_current());
cab30143
JQ		 1445}
1446
89cfc02c
DDAG		 1447void qmp_migrate_continue(MigrationStatus state, Error **errp)
1448{
1449 MigrationState *s = migrate_get_current();
1450 if (s->state != state) {
1451 error_setg(errp, "Migration not in expected state: %s",
1452 MigrationStatus_str(s->state));
1453 return;
1454 }
1455 qemu_sem_post(&s->pause_sem);
1456}
1457
9e1ba4cc
OW		 1458void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1459{
73af8dd8
JQ		 1460 MigrateSetParameters p = {
1461 .has_xbzrle_cache_size = true,
1462 .xbzrle_cache_size = value,
1463 };
c91e681a 1464
73af8dd8 1465 qmp_migrate_set_parameters(&p, errp);
9e1ba4cc
OW		 1466}
1467
1468int64_t qmp_query_migrate_cache_size(Error **errp)
1469{
1470 return migrate_xbzrle_cache_size();
1471}
1472
3dc85383 1473void qmp_migrate_set_speed(int64_t value, Error **errp)
cab30143 1474{
1bda8b3c 1475 MigrateSetParameters p = {
2ff30257
AA		 1476 .has_max_bandwidth = true,
1477 .max_bandwidth = value,
1478 };
cab30143 1479
2ff30257 1480 qmp_migrate_set_parameters(&p, errp);
cab30143
JQ		 1481}
1482
4f0a993b 1483void qmp_migrate_set_downtime(double value, Error **errp)
cab30143 1484{
87c9cc1c
DHB		 1485 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1486 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1487 "the range of 0 to %d seconds",
1488 MAX_MIGRATE_DOWNTIME_SECONDS);
1489 return;
1490 }
1491
2ff30257
AA		 1492 value *= 1000; /* Convert to milliseconds */
1493 value = MAX(0, MIN(INT64_MAX, value));
1494
1bda8b3c 1495 MigrateSetParameters p = {
2ff30257
AA		 1496 .has_downtime_limit = true,
1497 .downtime_limit = value,
1498 };
1499
1500 qmp_migrate_set_parameters(&p, errp);
99a0db9b 1501}
17ad9b35 1502
53f09a10
PB		 1503bool migrate_release_ram(void)
1504{
1505 MigrationState *s;
1506
1507 s = migrate_get_current();
1508
1509 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1510}
1511
53dd370c
DDAG		 1512bool migrate_postcopy_ram(void)
1513{
1514 MigrationState *s;
1515
1516 s = migrate_get_current();
1517
32c3db5b 1518 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
53dd370c
DDAG		 1519}
1520
58110f0a
VSO		 1521bool migrate_postcopy(void)
1522{
dd6bb914 1523 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
58110f0a
VSO		 1524}
1525
bde1e2ec
CV		 1526bool migrate_auto_converge(void)
1527{
1528 MigrationState *s;
1529
1530 s = migrate_get_current();
1531
1532 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1533}
1534
323004a3
PL		 1535bool migrate_zero_blocks(void)
1536{
1537 MigrationState *s;
1538
1539 s = migrate_get_current();
1540
1541 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1542}
1543
f22f928e
AP		 1544bool migrate_postcopy_blocktime(void)
1545{
1546 MigrationState *s;
1547
1548 s = migrate_get_current();
1549
1550 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
1551}
1552
8706d2d5
LL		 1553bool migrate_use_compression(void)
1554{
dde4e694
LL		 1555 MigrationState *s;
1556
1557 s = migrate_get_current();
1558
1559 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
8706d2d5
LL		 1560}
1561
1562int migrate_compress_level(void)
1563{
1564 MigrationState *s;
1565
1566 s = migrate_get_current();
1567
2594f56d 1568 return s->parameters.compress_level;
8706d2d5
LL		 1569}
1570
1571int migrate_compress_threads(void)
1572{
1573 MigrationState *s;
1574
1575 s = migrate_get_current();
1576
2594f56d 1577 return s->parameters.compress_threads;
8706d2d5
LL		 1578}
1579
3fcb38c2
LL		 1580int migrate_decompress_threads(void)
1581{
1582 MigrationState *s;
1583
1584 s = migrate_get_current();
1585
2594f56d 1586 return s->parameters.decompress_threads;
3fcb38c2
LL		 1587}
1588
55efc8c2
VSO		 1589bool migrate_dirty_bitmaps(void)
1590{
1591 MigrationState *s;
1592
1593 s = migrate_get_current();
1594
1595 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
1596}
1597
b05dc723
JQ		 1598bool migrate_use_events(void)
1599{
1600 MigrationState *s;
1601
1602 s = migrate_get_current();
1603
1604 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1605}
1606
30126bbf
JQ		 1607bool migrate_use_multifd(void)
1608{
1609 MigrationState *s;
1610
1611 s = migrate_get_current();
1612
1613 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1614}
1615
93fbd031
DDAG		 1616bool migrate_pause_before_switchover(void)
1617{
1618 MigrationState *s;
1619
1620 s = migrate_get_current();
1621
1622 return s->enabled_capabilities[
1623 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
1624}
1625
4075fb1c
JQ		 1626int migrate_multifd_channels(void)
1627{
1628 MigrationState *s;
1629
1630 s = migrate_get_current();
1631
1632 return s->parameters.x_multifd_channels;
1633}
1634
0fb86605
JQ		 1635int migrate_multifd_page_count(void)
1636{
1637 MigrationState *s;
1638
1639 s = migrate_get_current();
1640
1641 return s->parameters.x_multifd_page_count;
1642}
1643
17ad9b35
OW		 1644int migrate_use_xbzrle(void)
1645{
1646 MigrationState *s;
1647
1648 s = migrate_get_current();
1649
1650 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1651}
1652
1653int64_t migrate_xbzrle_cache_size(void)
1654{
1655 MigrationState *s;
1656
1657 s = migrate_get_current();
1658
73af8dd8 1659 return s->parameters.xbzrle_cache_size;
17ad9b35 1660}
0d82d0e8 1661
2833c59b
JQ		 1662bool migrate_use_block(void)
1663{
1664 MigrationState *s;
1665
1666 s = migrate_get_current();
1667
1668 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1669}
1670
c788ada8
PX		 1671bool migrate_use_return_path(void)
1672{
1673 MigrationState *s;
1674
1675 s = migrate_get_current();
1676
1677 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1678}
1679
2833c59b
JQ		 1680bool migrate_use_block_incremental(void)
1681{
1682 MigrationState *s;
1683
1684 s = migrate_get_current();
1685
1686 return s->parameters.block_incremental;
1687}
1688
70b20477
DDAG		 1689/* migration thread support */
1690/*
1691 * Something bad happened to the RP stream, mark an error
1692 * The caller shall print or trace something to indicate why
1693 */
1694static void mark_source_rp_bad(MigrationState *s)
1695{
1696 s->rp_state.error = true;
1697}
1698
1699static struct rp_cmd_args {
1700 ssize_t len; /* -1 = variable */
1701 const char *name;
1702} rp_cmd_args[] = {
1703 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1704 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1705 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1e2d90eb
DDAG		 1706 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1707 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
70b20477
DDAG		 1708 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1709};
1710
1e2d90eb
DDAG		 1711/*
1712 * Process a request for pages received on the return path,
1713 * We're allowed to send more than requested (e.g. to round to our page size)
1714 * and we don't need to send pages that have already been sent.
1715 */
1716static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1717 ram_addr_t start, size_t len)
1718{
6c595cde
DDAG		 1719 long our_host_ps = getpagesize();
1720
1e2d90eb 1721 trace_migrate_handle_rp_req_pages(rbname, start, len);
6c595cde
DDAG		 1722
1723 /*
1724 * Since we currently insist on matching page sizes, just sanity check
1725 * we're being asked for whole host pages.
1726 */
1727 if (start & (our_host_ps-1) ||
1728 (len & (our_host_ps-1))) {
1729 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1730 " len: %zd", __func__, start, len);
1731 mark_source_rp_bad(ms);
1732 return;
1733 }
1734
96506894 1735 if (ram_save_queue_pages(rbname, start, len)) {
6c595cde
DDAG		 1736 mark_source_rp_bad(ms);
1737 }
1e2d90eb
DDAG		 1738}
1739
70b20477
DDAG		 1740/*
1741 * Handles messages sent on the return path towards the source VM
1742 *
1743 */
1744static void *source_return_path_thread(void *opaque)
1745{
1746 MigrationState *ms = opaque;
1747 QEMUFile *rp = ms->rp_state.from_dst_file;
1748 uint16_t header_len, header_type;
568b01ca 1749 uint8_t buf[512];
70b20477 1750 uint32_t tmp32, sibling_error;
1e2d90eb
DDAG		 1751 ram_addr_t start = 0; /* =0 to silence warning */
1752 size_t len = 0, expected_len;
70b20477
DDAG		 1753 int res;
1754
1755 trace_source_return_path_thread_entry();
1756 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1757 migration_is_setup_or_active(ms->state)) {
1758 trace_source_return_path_thread_loop_top();
1759 header_type = qemu_get_be16(rp);
1760 header_len = qemu_get_be16(rp);
1761
7a9ddfbf
PX		 1762 if (qemu_file_get_error(rp)) {
1763 mark_source_rp_bad(ms);
1764 goto out;
1765 }
1766
70b20477
DDAG		 1767 if (header_type >= MIG_RP_MSG_MAX ||
1768 header_type == MIG_RP_MSG_INVALID) {
1769 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1770 header_type, header_len);
1771 mark_source_rp_bad(ms);
1772 goto out;
1773 }
1774
1775 if ((rp_cmd_args[header_type].len != -1 &&
1776 header_len != rp_cmd_args[header_type].len) ||
568b01ca 1777 header_len > sizeof(buf)) {
70b20477
DDAG		 1778 error_report("RP: Received '%s' message (0x%04x) with"
1779 "incorrect length %d expecting %zu",
1780 rp_cmd_args[header_type].name, header_type, header_len,
1781 (size_t)rp_cmd_args[header_type].len);
1782 mark_source_rp_bad(ms);
1783 goto out;
1784 }
1785
1786 /* We know we've got a valid header by this point */
1787 res = qemu_get_buffer(rp, buf, header_len);
1788 if (res != header_len) {
1789 error_report("RP: Failed reading data for message 0x%04x"
1790 " read %d expected %d",
1791 header_type, res, header_len);
1792 mark_source_rp_bad(ms);
1793 goto out;
1794 }
1795
1796 /* OK, we have the message and the data */
1797 switch (header_type) {
1798 case MIG_RP_MSG_SHUT:
4d885131 1799 sibling_error = ldl_be_p(buf);
70b20477
DDAG		 1800 trace_source_return_path_thread_shut(sibling_error);
1801 if (sibling_error) {
1802 error_report("RP: Sibling indicated error %d", sibling_error);
1803 mark_source_rp_bad(ms);
1804 }
1805 /*
1806 * We'll let the main thread deal with closing the RP
1807 * we could do a shutdown(2) on it, but we're the only user
1808 * anyway, so there's nothing gained.
1809 */
1810 goto out;
1811
1812 case MIG_RP_MSG_PONG:
4d885131 1813 tmp32 = ldl_be_p(buf);
70b20477
DDAG		 1814 trace_source_return_path_thread_pong(tmp32);
1815 break;
1816
1e2d90eb 1817 case MIG_RP_MSG_REQ_PAGES:
4d885131
PM		 1818 start = ldq_be_p(buf);
1819 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 1820 migrate_handle_rp_req_pages(ms, NULL, start, len);
1821 break;
1822
1823 case MIG_RP_MSG_REQ_PAGES_ID:
1824 expected_len = 12 + 1; /* header + termination */
1825
1826 if (header_len >= expected_len) {
4d885131
PM		 1827 start = ldq_be_p(buf);
1828 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 1829 /* Now we expect an idstr */
1830 tmp32 = buf[12]; /* Length of the following idstr */
1831 buf[13 + tmp32] = '\0';
1832 expected_len += tmp32;
1833 }
1834 if (header_len != expected_len) {
1835 error_report("RP: Req_Page_id with length %d expecting %zd",
1836 header_len, expected_len);
1837 mark_source_rp_bad(ms);
1838 goto out;
1839 }
1840 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1841 break;
1842
70b20477
DDAG		 1843 default:
1844 break;
1845 }
1846 }
5df5416e 1847 if (qemu_file_get_error(rp)) {
70b20477
DDAG		 1848 trace_source_return_path_thread_bad_end();
1849 mark_source_rp_bad(ms);
1850 }
1851
1852 trace_source_return_path_thread_end();
1853out:
1854 ms->rp_state.from_dst_file = NULL;
1855 qemu_fclose(rp);
1856 return NULL;
1857}
1858
70b20477
DDAG		 1859static int open_return_path_on_source(MigrationState *ms)
1860{
1861
89a02a9f 1862 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
70b20477
DDAG		 1863 if (!ms->rp_state.from_dst_file) {
1864 return -1;
1865 }
1866
1867 trace_open_return_path_on_source();
1868 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1869 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1870
1871 trace_open_return_path_on_source_continue();
1872
1873 return 0;
1874}
1875
70b20477
DDAG		 1876/* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1877static int await_return_path_close_on_source(MigrationState *ms)
1878{
1879 /*
1880 * If this is a normal exit then the destination will send a SHUT and the
1881 * rp_thread will exit, however if there's an error we need to cause
1882 * it to exit.
1883 */
89a02a9f 1884 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
70b20477
DDAG		 1885 /*
1886 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1887 * waiting for the destination.
1888 */
1889 qemu_file_shutdown(ms->rp_state.from_dst_file);
1890 mark_source_rp_bad(ms);
1891 }
1892 trace_await_return_path_close_on_source_joining();
1893 qemu_thread_join(&ms->rp_state.rp_thread);
1894 trace_await_return_path_close_on_source_close();
1895 return ms->rp_state.error;
1896}
1897
1d34e4bf
DDAG		 1898/*
1899 * Switch from normal iteration to postcopy
1900 * Returns non-0 on error
1901 */
7287cbd4 1902static int postcopy_start(MigrationState *ms)
1d34e4bf
DDAG		 1903{
1904 int ret;
61b67d47
DB		 1905 QIOChannelBuffer *bioc;
1906 QEMUFile *fb;
1d34e4bf 1907 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
ef8d6488 1908 bool restart_block = false;
0331c8ca
DDAG		 1909 int cur_state = MIGRATION_STATUS_ACTIVE;
1910 if (!migrate_pause_before_switchover()) {
1911 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1912 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1913 }
1d34e4bf
DDAG		 1914
1915 trace_postcopy_start();
1916 qemu_mutex_lock_iothread();
1917 trace_postcopy_start_set_run();
1918
1919 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1d34e4bf
DDAG		 1920 global_state_store();
1921 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
76b1c7fe
KW		 1922 if (ret < 0) {
1923 goto fail;
1924 }
1d34e4bf 1925
0331c8ca
DDAG		 1926 ret = migration_maybe_pause(ms, &cur_state,
1927 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1928 if (ret < 0) {
1929 goto fail;
1930 }
1931
76b1c7fe 1932 ret = bdrv_inactivate_all();
1d34e4bf
DDAG		 1933 if (ret < 0) {
1934 goto fail;
1935 }
ef8d6488 1936 restart_block = true;
1d34e4bf 1937
1c0d249d
DDAG		 1938 /*
1939 * Cause any non-postcopiable, but iterative devices to
1940 * send out their final data.
1941 */
a1fbe750 1942 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1c0d249d 1943
1d34e4bf
DDAG		 1944 /*
1945 * in Finish migrate and with the io-lock held everything should
1946 * be quiet, but we've potentially still got dirty pages and we
1947 * need to tell the destination to throw any pages it's already received
1948 * that are dirty
1949 */
58110f0a
VSO		 1950 if (migrate_postcopy_ram()) {
1951 if (ram_postcopy_send_discard_bitmap(ms)) {
1952 error_report("postcopy send discard bitmap failed");
1953 goto fail;
1954 }
1d34e4bf
DDAG		 1955 }
1956
1957 /*
1958 * send rest of state - note things that are doing postcopy
1959 * will notice we're in POSTCOPY_ACTIVE and not actually
1960 * wrap their state up here
1961 */
89a02a9f 1962 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
58110f0a
VSO		 1963 if (migrate_postcopy_ram()) {
1964 /* Ping just for debugging, helps line traces up */
1965 qemu_savevm_send_ping(ms->to_dst_file, 2);
1966 }
1d34e4bf
DDAG		 1967
1968 /*
1969 * While loading the device state we may trigger page transfer
1970 * requests and the fd must be free to process those, and thus
1971 * the destination must read the whole device state off the fd before
1972 * it starts processing it. Unfortunately the ad-hoc migration format
1973 * doesn't allow the destination to know the size to read without fully
1974 * parsing it through each devices load-state code (especially the open
1975 * coded devices that use get/put).
1976 * So we wrap the device state up in a package with a length at the start;
1977 * to do this we use a qemu_buf to hold the whole of the device state.
1978 */
61b67d47 1979 bioc = qio_channel_buffer_new(4096);
6f01f136 1980 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
61b67d47
DB		 1981 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1982 object_unref(OBJECT(bioc));
1d34e4bf 1983
c76201ab
DDAG		 1984 /*
1985 * Make sure the receiver can get incoming pages before we send the rest
1986 * of the state
1987 */
1988 qemu_savevm_send_postcopy_listen(fb);
1989
a1fbe750 1990 qemu_savevm_state_complete_precopy(fb, false, false);
58110f0a
VSO		 1991 if (migrate_postcopy_ram()) {
1992 qemu_savevm_send_ping(fb, 3);
1993 }
1d34e4bf
DDAG		 1994
1995 qemu_savevm_send_postcopy_run(fb);
1996
1997 /* <><> end of stuff going into the package */
1d34e4bf 1998
ef8d6488
DDAG		 1999 /* Last point of recovery; as soon as we send the package the destination
2000 * can open devices and potentially start running.
2001 * Lets just check again we've not got any errors.
2002 */
2003 ret = qemu_file_get_error(ms->to_dst_file);
2004 if (ret) {
2005 error_report("postcopy_start: Migration stream errored (pre package)");
2006 goto fail_closefb;
2007 }
2008
2009 restart_block = false;
2010
1d34e4bf 2011 /* Now send that blob */
61b67d47 2012 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1d34e4bf
DDAG		 2013 goto fail_closefb;
2014 }
2015 qemu_fclose(fb);
b82fc321
DDAG		 2016
2017 /* Send a notify to give a chance for anything that needs to happen
2018 * at the transition to postcopy and after the device state; in particular
2019 * spice needs to trigger a transition now
2020 */
2021 ms->postcopy_after_devices = true;
2022 notifier_list_notify(&migration_state_notifiers, ms);
2023
1d34e4bf
DDAG		 2024 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2025
2026 qemu_mutex_unlock_iothread();
2027
58110f0a
VSO		 2028 if (migrate_postcopy_ram()) {
2029 /*
2030 * Although this ping is just for debug, it could potentially be
2031 * used for getting a better measurement of downtime at the source.
2032 */
2033 qemu_savevm_send_ping(ms->to_dst_file, 4);
2034 }
1d34e4bf 2035
ced1c616
PB		 2036 if (migrate_release_ram()) {
2037 ram_postcopy_migrated_memory_release(ms);
2038 }
2039
89a02a9f 2040 ret = qemu_file_get_error(ms->to_dst_file);
1d34e4bf
DDAG		 2041 if (ret) {
2042 error_report("postcopy_start: Migration stream errored");
48781e5b 2043 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf
DDAG		 2044 MIGRATION_STATUS_FAILED);
2045 }
2046
2047 return ret;
2048
2049fail_closefb:
2050 qemu_fclose(fb);
2051fail:
48781e5b 2052 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf 2053 MIGRATION_STATUS_FAILED);
ef8d6488
DDAG		 2054 if (restart_block) {
2055 /* A failure happened early enough that we know the destination hasn't
2056 * accessed block devices, so we're safe to recover.
2057 */
2058 Error *local_err = NULL;
2059
2060 bdrv_invalidate_cache_all(&local_err);
2061 if (local_err) {
2062 error_report_err(local_err);
2063 }
2064 }
1d34e4bf
DDAG		 2065 qemu_mutex_unlock_iothread();
2066 return -1;
2067}
2068
e91d8951
DDAG		 2069/**
2070 * migration_maybe_pause: Pause if required to by
2071 * migrate_pause_before_switchover called with the iothread locked
2072 * Returns: 0 on success
2073 */
0331c8ca
DDAG		 2074static int migration_maybe_pause(MigrationState *s,
2075 int *current_active_state,
2076 int new_state)
e91d8951
DDAG		 2077{
2078 if (!migrate_pause_before_switchover()) {
2079 return 0;
2080 }
2081
2082 /* Since leaving this state is not atomic with posting the semaphore
2083 * it's possible that someone could have issued multiple migrate_continue
2084 * and the semaphore is incorrectly positive at this point;
2085 * the docs say it's undefined to reinit a semaphore that's already
2086 * init'd, so use timedwait to eat up any existing posts.
2087 */
2088 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2089 /* This block intentionally left blank */
2090 }
2091
2092 qemu_mutex_unlock_iothread();
2093 migrate_set_state(&s->state, *current_active_state,
2094 MIGRATION_STATUS_PRE_SWITCHOVER);
2095 qemu_sem_wait(&s->pause_sem);
2096 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
0331c8ca
DDAG		 2097 new_state);
2098 *current_active_state = new_state;
e91d8951
DDAG		 2099 qemu_mutex_lock_iothread();
2100
0331c8ca 2101 return s->state == new_state ? 0 : -EINVAL;
e91d8951
DDAG		 2102}
2103
09f6c85e
DDAG		 2104/**
2105 * migration_completion: Used by migration_thread when there's not much left.
2106 * The caller 'breaks' the loop when this returns.
2107 *
2108 * @s: Current migration state
09f6c85e 2109 */
2ad87305 2110static void migration_completion(MigrationState *s)
09f6c85e
DDAG		 2111{
2112 int ret;
2ad87305 2113 int current_active_state = s->state;
09f6c85e 2114
b10ac0c4
DDAG		 2115 if (s->state == MIGRATION_STATUS_ACTIVE) {
2116 qemu_mutex_lock_iothread();
64909f97 2117 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
b10ac0c4 2118 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
7287cbd4 2119 s->vm_was_running = runstate_is_running();
b10ac0c4
DDAG		 2120 ret = global_state_store();
2121
2122 if (!ret) {
a1fbe750 2123 bool inactivate = !migrate_colo_enabled();
b10ac0c4 2124 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
e91d8951 2125 if (ret >= 0) {
0331c8ca
DDAG		 2126 ret = migration_maybe_pause(s, &current_active_state,
2127 MIGRATION_STATUS_DEVICE);
e91d8951 2128 }
f07fa4cb
KW		 2129 if (ret >= 0) {
2130 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
a1fbe750
FZ		 2131 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2132 inactivate);
f07fa4cb 2133 }
a1fbe750
FZ		 2134 if (inactivate && ret >= 0) {
2135 s->block_inactive = true;
b10ac0c4
DDAG		 2136 }
2137 }
2138 qemu_mutex_unlock_iothread();
09f6c85e 2139
b10ac0c4
DDAG		 2140 if (ret < 0) {
2141 goto fail;
09f6c85e 2142 }
b10ac0c4
DDAG		 2143 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2144 trace_migration_completion_postcopy_end();
2145
89a02a9f 2146 qemu_savevm_state_complete_postcopy(s->to_dst_file);
b10ac0c4 2147 trace_migration_completion_postcopy_end_after_complete();
09f6c85e 2148 }
09f6c85e 2149
b10ac0c4
DDAG		 2150 /*
2151 * If rp was opened we must clean up the thread before
2152 * cleaning everything else up (since if there are no failures
2153 * it will wait for the destination to send it's status in
2154 * a SHUT command).
b10ac0c4 2155 */
0425dc97 2156 if (s->rp_state.from_dst_file) {
b10ac0c4 2157 int rp_error;
0425dc97 2158 trace_migration_return_path_end_before();
b10ac0c4 2159 rp_error = await_return_path_close_on_source(s);
0425dc97 2160 trace_migration_return_path_end_after(rp_error);
b10ac0c4 2161 if (rp_error) {
fe904ea8 2162 goto fail_invalidate;
b10ac0c4 2163 }
09f6c85e
DDAG		 2164 }
2165
89a02a9f 2166 if (qemu_file_get_error(s->to_dst_file)) {
09f6c85e 2167 trace_migration_completion_file_err();
fe904ea8 2168 goto fail_invalidate;
09f6c85e
DDAG		 2169 }
2170
0b827d5e
HZ		 2171 if (!migrate_colo_enabled()) {
2172 migrate_set_state(&s->state, current_active_state,
2173 MIGRATION_STATUS_COMPLETED);
2174 }
2175
09f6c85e
DDAG		 2176 return;
2177
fe904ea8
GK		 2178fail_invalidate:
2179 /* If not doing postcopy, vm_start() will be called: let's regain
2180 * control on images.
2181 */
6039dd5b
DDAG		 2182 if (s->state == MIGRATION_STATUS_ACTIVE ||
2183 s->state == MIGRATION_STATUS_DEVICE) {
fe904ea8
GK		 2184 Error *local_err = NULL;
2185
1d2acc31 2186 qemu_mutex_lock_iothread();
fe904ea8
GK		 2187 bdrv_invalidate_cache_all(&local_err);
2188 if (local_err) {
2189 error_report_err(local_err);
1d2acc31
HZ		 2190 } else {
2191 s->block_inactive = false;
fe904ea8 2192 }
1d2acc31 2193 qemu_mutex_unlock_iothread();
fe904ea8
GK		 2194 }
2195
09f6c85e 2196fail:
48781e5b
HZ		 2197 migrate_set_state(&s->state, current_active_state,
2198 MIGRATION_STATUS_FAILED);
09f6c85e
DDAG		 2199}
2200
35a6ed4f
HZ		 2201bool migrate_colo_enabled(void)
2202{
2203 MigrationState *s = migrate_get_current();
2204 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2205}
2206
cf011f08
PX		 2207static void migration_calculate_complete(MigrationState *s)
2208{
2209 uint64_t bytes = qemu_ftell(s->to_dst_file);
2210 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2211
2212 s->total_time = end_time - s->start_time;
2213 if (!s->downtime) {
2214 /*
2215 * It's still not set, so we are precopy migration. For
2216 * postcopy, downtime is calculated during postcopy_start().
2217 */
2218 s->downtime = end_time - s->downtime_start;
2219 }
2220
2221 if (s->total_time) {
2222 s->mbps = ((double) bytes * 8.0) / s->total_time / 1000;
2223 }
2224}
2225
b15df1ae
PX		 2226static void migration_update_counters(MigrationState *s,
2227 int64_t current_time)
2228{
2229 uint64_t transferred, time_spent;
b15df1ae
PX		 2230 double bandwidth;
2231
2232 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
2233 return;
2234 }
2235
2236 transferred = qemu_ftell(s->to_dst_file) - s->iteration_initial_bytes;
2237 time_spent = current_time - s->iteration_start_time;
2238 bandwidth = (double)transferred / time_spent;
0781c1ed 2239 s->threshold_size = bandwidth * s->parameters.downtime_limit;
b15df1ae
PX		 2240
2241 s->mbps = (((double) transferred * 8.0) /
2242 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2243
2244 /*
2245 * if we haven't sent anything, we don't want to
2246 * recalculate. 10000 is a small enough number for our purposes
2247 */
2248 if (ram_counters.dirty_pages_rate && transferred > 10000) {
2249 s->expected_downtime = ram_counters.dirty_pages_rate *
2250 qemu_target_page_size() / bandwidth;
2251 }
2252
2253 qemu_file_reset_rate_limit(s->to_dst_file);
2254
2255 s->iteration_start_time = current_time;
2256 s->iteration_initial_bytes = qemu_ftell(s->to_dst_file);
2257
2258 trace_migrate_transferred(transferred, time_spent,
0781c1ed 2259 bandwidth, s->threshold_size);
b15df1ae
PX		 2260}
2261
2ad87305
PX		 2262/* Migration thread iteration status */
2263typedef enum {
2264 MIG_ITERATE_RESUME, /* Resume current iteration */
2265 MIG_ITERATE_SKIP, /* Skip current iteration */
2266 MIG_ITERATE_BREAK, /* Break the loop */
2267} MigIterateState;
2268
2269/*
2270 * Return true if continue to the next iteration directly, false
2271 * otherwise.
2272 */
2273static MigIterateState migration_iteration_run(MigrationState *s)
2274{
47995026 2275 uint64_t pending_size, pend_pre, pend_compat, pend_post;
2ad87305
PX		 2276 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
2277
47995026
VSO		 2278 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
2279 &pend_compat, &pend_post);
2280 pending_size = pend_pre + pend_compat + pend_post;
2ad87305
PX		 2281
2282 trace_migrate_pending(pending_size, s->threshold_size,
47995026 2283 pend_pre, pend_compat, pend_post);
2ad87305
PX		 2284
2285 if (pending_size && pending_size >= s->threshold_size) {
2286 /* Still a significant amount to transfer */
2287 if (migrate_postcopy() && !in_postcopy &&
47995026 2288 pend_pre <= s->threshold_size &&
2ad87305
PX		 2289 atomic_read(&s->start_postcopy)) {
2290 if (postcopy_start(s)) {
2291 error_report("%s: postcopy failed to start", __func__);
2292 }
2293 return MIG_ITERATE_SKIP;
2294 }
2295 /* Just another iteration step */
2296 qemu_savevm_state_iterate(s->to_dst_file,
2297 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2298 } else {
2299 trace_migration_thread_low_pending(pending_size);
2300 migration_completion(s);
2301 return MIG_ITERATE_BREAK;
2302 }
2303
2304 return MIG_ITERATE_RESUME;
2305}
2306
199aa6d4
PX		 2307static void migration_iteration_finish(MigrationState *s)
2308{
2309 /* If we enabled cpu throttling for auto-converge, turn it off. */
2310 cpu_throttle_stop();
2311
2312 qemu_mutex_lock_iothread();
2313 switch (s->state) {
2314 case MIGRATION_STATUS_COMPLETED:
2315 migration_calculate_complete(s);
2316 runstate_set(RUN_STATE_POSTMIGRATE);
2317 break;
2318
2319 case MIGRATION_STATUS_ACTIVE:
2320 /*
2321 * We should really assert here, but since it's during
2322 * migration, let's try to reduce the usage of assertions.
2323 */
2324 if (!migrate_colo_enabled()) {
2325 error_report("%s: critical error: calling COLO code without "
2326 "COLO enabled", __func__);
2327 }
2328 migrate_start_colo_process(s);
2329 /*
2330 * Fixme: we will run VM in COLO no matter its old running state.
2331 * After exited COLO, we will keep running.
2332 */
2333 s->vm_was_running = true;
2334 /* Fallthrough */
2335 case MIGRATION_STATUS_FAILED:
2336 case MIGRATION_STATUS_CANCELLED:
2337 if (s->vm_was_running) {
2338 vm_start();
2339 } else {
2340 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2341 runstate_set(RUN_STATE_POSTMIGRATE);
2342 }
2343 }
2344 break;
2345
2346 default:
2347 /* Should not reach here, but if so, forgive the VM. */
2348 error_report("%s: Unknown ending state %d", __func__, s->state);
2349 break;
2350 }
2351 qemu_bh_schedule(s->cleanup_bh);
2352 qemu_mutex_unlock_iothread();
2353}
2354
70b20477
DDAG		 2355/*
2356 * Master migration thread on the source VM.
2357 * It drives the migration and pumps the data down the outgoing channel.
2358 */
5f496a1b 2359static void *migration_thread(void *opaque)
0d82d0e8 2360{
9848a404 2361 MigrationState *s = opaque;
bc72ad67 2362 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
76f5933a 2363
ab28bd23
PB		 2364 rcu_register_thread();
2365
b15df1ae
PX		 2366 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2367
89a02a9f 2368 qemu_savevm_state_header(s->to_dst_file);
1d34e4bf 2369
62a02658
PX		 2370 /*
2371 * If we opened the return path, we need to make sure dst has it
2372 * opened as well.
2373 */
2374 if (s->rp_state.from_dst_file) {
1d34e4bf 2375 /* Now tell the dest that it should open its end so it can reply */
89a02a9f 2376 qemu_savevm_send_open_return_path(s->to_dst_file);
1d34e4bf
DDAG		 2377
2378 /* And do a ping that will make stuff easier to debug */
89a02a9f 2379 qemu_savevm_send_ping(s->to_dst_file, 1);
0425dc97 2380 }
1d34e4bf 2381
58110f0a 2382 if (migrate_postcopy()) {
1d34e4bf
DDAG		 2383 /*
2384 * Tell the destination that we *might* want to do postcopy later;
2385 * if the other end can't do postcopy it should fail now, nice and
2386 * early.
2387 */
89a02a9f 2388 qemu_savevm_send_postcopy_advise(s->to_dst_file);
1d34e4bf
DDAG		 2389 }
2390
9907e842 2391 qemu_savevm_state_setup(s->to_dst_file);
0d82d0e8 2392
bc72ad67 2393 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
48781e5b
HZ		 2394 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2395 MIGRATION_STATUS_ACTIVE);
29ae8a41 2396
9ec055ae
DDAG		 2397 trace_migration_thread_setup_complete();
2398
2399 while (s->state == MIGRATION_STATUS_ACTIVE ||
2400 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
a3e879cd 2401 int64_t current_time;
0d82d0e8 2402
89a02a9f 2403 if (!qemu_file_rate_limit(s->to_dst_file)) {
2ad87305
PX		 2404 MigIterateState iter_state = migration_iteration_run(s);
2405 if (iter_state == MIG_ITERATE_SKIP) {
2406 continue;
2407 } else if (iter_state == MIG_ITERATE_BREAK) {
09f6c85e 2408 break;
c369f40d
JQ		 2409 }
2410 }
f4410a5d 2411
89a02a9f 2412 if (qemu_file_get_error(s->to_dst_file)) {
2ad87305
PX		 2413 if (migration_is_setup_or_active(s->state)) {
2414 migrate_set_state(&s->state, s->state,
2415 MIGRATION_STATUS_FAILED);
2416 }
1d34e4bf 2417 trace_migration_thread_file_err();
fd45ee2c
PB		 2418 break;
2419 }
b15df1ae 2420
bc72ad67 2421 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
0d82d0e8 2422
b15df1ae
PX		 2423 migration_update_counters(s, current_time);
2424
89a02a9f 2425 if (qemu_file_rate_limit(s->to_dst_file)) {
0d82d0e8 2426 /* usleep expects microseconds */
b15df1ae
PX		 2427 g_usleep((s->iteration_start_time + BUFFER_DELAY -
2428 current_time) * 1000);
0d82d0e8 2429 }
a3fa1d78
PB		 2430 }
2431
1d34e4bf 2432 trace_migration_thread_after_loop();
199aa6d4 2433 migration_iteration_finish(s);
ab28bd23 2434 rcu_unregister_thread();
0d82d0e8
JQ		 2435 return NULL;
2436}
2437
cce8040b 2438void migrate_fd_connect(MigrationState *s, Error *error_in)
0d82d0e8 2439{
2ff30257 2440 s->expected_downtime = s->parameters.downtime_limit;
bb1fadc4 2441 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
cce8040b
DDAG		 2442 if (error_in) {
2443 migrate_fd_error(s, error_in);
2444 migrate_fd_cleanup(s);
2445 return;
2446 }
0d82d0e8 2447
9e4d2b98 2448 qemu_file_set_blocking(s->to_dst_file, true);
89a02a9f 2449 qemu_file_set_rate_limit(s->to_dst_file,
2ff30257 2450 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
442773ce 2451
9287ac27
SH		 2452 /* Notify before starting migration thread */
2453 notifier_list_notify(&migration_state_notifiers, s);
2454
1d34e4bf 2455 /*
c788ada8
PX		 2456 * Open the return path. For postcopy, it is used exclusively. For
2457 * precopy, only if user specified "return-path" capability would
2458 * QEMU uses the return path.
1d34e4bf 2459 */
c788ada8 2460 if (migrate_postcopy_ram() || migrate_use_return_path()) {
1d34e4bf
DDAG		 2461 if (open_return_path_on_source(s)) {
2462 error_report("Unable to open return-path for postcopy");
48781e5b 2463 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1d34e4bf
DDAG		 2464 MIGRATION_STATUS_FAILED);
2465 migrate_fd_cleanup(s);
2466 return;
2467 }
2468 }
2469
f986c3d2
JQ		 2470 if (multifd_save_setup() != 0) {
2471 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2472 MIGRATION_STATUS_FAILED);
2473 migrate_fd_cleanup(s);
2474 return;
2475 }
009fad7f 2476 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
bb1fadc4 2477 QEMU_THREAD_JOINABLE);
1d34e4bf 2478 s->migration_thread_running = true;
0d82d0e8 2479}
093e3c42 2480
9d18af93
PX		 2481void migration_global_dump(Monitor *mon)
2482{
2483 MigrationState *ms = migrate_get_current();
2484
6f0f6428
JQ		 2485 monitor_printf(mon, "globals:\n");
2486 monitor_printf(mon, "store-global-state: %s\n",
2487 ms->store_global_state ? "on" : "off");
2488 monitor_printf(mon, "only-migratable: %s\n",
2489 ms->only_migratable ? "on" : "off");
2490 monitor_printf(mon, "send-configuration: %s\n",
2491 ms->send_configuration ? "on" : "off");
2492 monitor_printf(mon, "send-section-footer: %s\n",
2493 ms->send_section_footer ? "on" : "off");
9d18af93
PX		 2494}
2495
20814758
PX		 2496#define DEFINE_PROP_MIG_CAP(name, x) \
2497 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2498
5272298c
PX		 2499static Property migration_properties[] = {
2500 DEFINE_PROP_BOOL("store-global-state", MigrationState,
2501 store_global_state, true),
3df663e5 2502 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
71dd4c1a
PX		 2503 DEFINE_PROP_BOOL("send-configuration", MigrationState,
2504 send_configuration, true),
15c38503
PX		 2505 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2506 send_section_footer, true),
89632faf
PX		 2507
2508 /* Migration parameters */
741d4086 2509 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
89632faf
PX		 2510 parameters.compress_level,
2511 DEFAULT_MIGRATE_COMPRESS_LEVEL),
741d4086 2512 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
89632faf
PX		 2513 parameters.compress_threads,
2514 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
741d4086 2515 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
89632faf
PX		 2516 parameters.decompress_threads,
2517 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
741d4086 2518 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
89632faf
PX		 2519 parameters.cpu_throttle_initial,
2520 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
741d4086 2521 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
89632faf
PX		 2522 parameters.cpu_throttle_increment,
2523 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
741d4086 2524 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
89632faf 2525 parameters.max_bandwidth, MAX_THROTTLE),
741d4086 2526 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
89632faf
PX		 2527 parameters.downtime_limit,
2528 DEFAULT_MIGRATE_SET_DOWNTIME),
741d4086 2529 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
89632faf
PX		 2530 parameters.x_checkpoint_delay,
2531 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
741d4086 2532 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState,
4075fb1c
JQ		 2533 parameters.x_multifd_channels,
2534 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
741d4086 2535 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState,
0fb86605
JQ		 2536 parameters.x_multifd_page_count,
2537 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
73af8dd8
JQ		 2538 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
2539 parameters.xbzrle_cache_size,
2540 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
20814758
PX		 2541
2542 /* Migration capabilities */
2543 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2544 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2545 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2546 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2547 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2548 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2549 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2550 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2551 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2552 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2553 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
30126bbf 2554 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
20814758 2555
5272298c
PX		 2556 DEFINE_PROP_END_OF_LIST(),
2557};
2558
e5cb7e76
PX		 2559static void migration_class_init(ObjectClass *klass, void *data)
2560{
2561 DeviceClass *dc = DEVICE_CLASS(klass);
2562
2563 dc->user_creatable = false;
5272298c 2564 dc->props = migration_properties;
e5cb7e76
PX		 2565}
2566
b91bf5e4
MAL		 2567static void migration_instance_finalize(Object *obj)
2568{
2569 MigrationState *ms = MIGRATION_OBJ(obj);
2570 MigrationParameters *params = &ms->parameters;
2571
87db1a7d 2572 qemu_mutex_destroy(&ms->error_mutex);
b91bf5e4
MAL		 2573 g_free(params->tls_hostname);
2574 g_free(params->tls_creds);
e91d8951 2575 qemu_sem_destroy(&ms->pause_sem);
ab105cc1 2576 error_free(ms->error);
b91bf5e4
MAL		 2577}
2578
e5cb7e76
PX		 2579static void migration_instance_init(Object *obj)
2580{
2581 MigrationState *ms = MIGRATION_OBJ(obj);
8b0b29dc 2582 MigrationParameters *params = &ms->parameters;
e5cb7e76
PX		 2583
2584 ms->state = MIGRATION_STATUS_NONE;
e5cb7e76 2585 ms->mbps = -1;
e91d8951 2586 qemu_sem_init(&ms->pause_sem, 0);
87db1a7d 2587 qemu_mutex_init(&ms->error_mutex);
8b0b29dc
PX		 2588
2589 params->tls_hostname = g_strdup("");
2590 params->tls_creds = g_strdup("");
2591
2592 /* Set has_* up only for parameter checks */
2593 params->has_compress_level = true;
2594 params->has_compress_threads = true;
2595 params->has_decompress_threads = true;
2596 params->has_cpu_throttle_initial = true;
2597 params->has_cpu_throttle_increment = true;
2598 params->has_max_bandwidth = true;
2599 params->has_downtime_limit = true;
2600 params->has_x_checkpoint_delay = true;
2601 params->has_block_incremental = true;
4075fb1c 2602 params->has_x_multifd_channels = true;
0fb86605 2603 params->has_x_multifd_page_count = true;
73af8dd8 2604 params->has_xbzrle_cache_size = true;
8b0b29dc
PX		 2605}
2606
2607/*
2608 * Return true if check pass, false otherwise. Error will be put
2609 * inside errp if provided.
2610 */
2611static bool migration_object_check(MigrationState *ms, Error **errp)
2612{
6b19a7d9
PX		 2613 MigrationCapabilityStatusList *head = NULL;
2614 /* Assuming all off */
2615 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2616 int i;
2617
8b0b29dc
PX		 2618 if (!migrate_params_check(&ms->parameters, errp)) {
2619 return false;
2620 }
2621
6b19a7d9
PX		 2622 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2623 if (ms->enabled_capabilities[i]) {
2624 head = migrate_cap_add(head, i, true);
2625 }
2626 }
2627
2628 ret = migrate_caps_check(cap_list, head, errp);
2629
2630 /* It works with head == NULL */
2631 qapi_free_MigrationCapabilityStatusList(head);
2632
2633 return ret;
e5cb7e76
PX		 2634}
2635
2636static const TypeInfo migration_type = {
2637 .name = TYPE_MIGRATION,
01f6e14c 2638 /*
c8d3ff38
PX		 2639 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2640 * not created using qdev_create(), it is not attached to the qdev
2641 * device tree, and it is never realized.
2642 *
2643 * TODO: Make this TYPE_OBJECT once QOM provides something like
2644 * TYPE_DEVICE's "-global" properties.
01f6e14c 2645 */
e5cb7e76
PX		 2646 .parent = TYPE_DEVICE,
2647 .class_init = migration_class_init,
2648 .class_size = sizeof(MigrationClass),
2649 .instance_size = sizeof(MigrationState),
2650 .instance_init = migration_instance_init,
b91bf5e4 2651 .instance_finalize = migration_instance_finalize,
e5cb7e76
PX		 2652};
2653
2654static void register_migration_types(void)
2655{
2656 type_register_static(&migration_type);
2657}
2658
2659type_init(register_migration_types);
QEMU mirror