]> Git Repo - qemu.git/blame - migration/migration.c
projects / qemu.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Eliminate qapi/qmp/types.h
[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 <[email protected]>
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"
ab28bd23 35#include "qemu/rcu.h"
2c9e6fec 36#include "block.h"
be07b0ac 37#include "postcopy-ram.h"
766bd176 38#include "qemu/thread.h"
791e7c82 39#include "qmp-commands.h"
c09e5bb1 40#include "trace.h"
598cd2bd 41#include "qapi-event.h"
51180423 42#include "exec/target_page.h"
61b67d47 43#include "io/channel-buffer.h"
35a6ed4f 44#include "migration/colo.h"
4ffdb337 45#include "hw/boards.h"
9d18af93 46#include "monitor/monitor.h"
065e2813 47
dc325627 48#define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
5bb7910a 49
5b4e1eb7
JQ		 50/* Amount of time to allocate to each "chunk" of bandwidth-throttled
51 * data. */
52#define BUFFER_DELAY 100
53#define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
54
2ff30257
AA		 55/* Time in milliseconds we are allowed to stop the source,
56 * for sending the last part */
57#define DEFAULT_MIGRATE_SET_DOWNTIME 300
58
87c9cc1c
DHB		 59/* Maximum migrate downtime set to 2000 seconds */
60#define MAX_MIGRATE_DOWNTIME_SECONDS 2000
61#define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
62
8706d2d5
LL		 63/* Default compression thread count */
64#define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
3fcb38c2
LL		 65/* Default decompression thread count, usually decompression is at
66 * least 4 times as fast as compression.*/
67#define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
8706d2d5
LL		 68/*0: means nocompress, 1: best speed, ... 9: best compress ratio */
69#define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
1626fee3 70/* Define default autoconverge cpu throttle migration parameters */
d85a31d1
JH		 71#define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
72#define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
8706d2d5 73
17ad9b35 74/* Migration XBZRLE default cache size */
73af8dd8 75#define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
17ad9b35 76
68b53591
HZ		 77/* The delay time (in ms) between two COLO checkpoints
78 * Note: Please change this default value to 10000 when we support hybrid mode.
79 */
80#define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
4075fb1c 81#define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
0fb86605 82#define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
68b53591 83
99a0db9b
GH		 84static NotifierList migration_state_notifiers =
85 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
86
adde220a
DDAG		 87static bool deferred_incoming;
88
da6f1790
JQ		 89/* Messages sent on the return path from destination to source */
90enum mig_rp_message_type {
91 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
92 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
93 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
94
95 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
96 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
97
98 MIG_RP_MSG_MAX
99};
100
17549e84
JQ		 101/* When we add fault tolerance, we could have several
102 migrations at once. For now we don't need to add
103 dynamic creation of migration */
104
e5cb7e76
PX		 105static MigrationState *current_migration;
106
8b0b29dc 107static bool migration_object_check(MigrationState *ms, Error **errp);
0331c8ca
DDAG		 108static int migration_maybe_pause(MigrationState *s,
109 int *current_active_state,
110 int new_state);
8b0b29dc 111
e5cb7e76
PX		 112void migration_object_init(void)
113{
4ffdb337 114 MachineState *ms = MACHINE(qdev_get_machine());
8b0b29dc 115 Error *err = NULL;
4ffdb337 116
e5cb7e76
PX		 117 /* This can only be called once. */
118 assert(!current_migration);
119 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
4ffdb337 120
8b0b29dc
PX		 121 if (!migration_object_check(current_migration, &err)) {
122 error_report_err(err);
123 exit(1);
124 }
125
4ffdb337
PX		 126 /*
127 * We cannot really do this in migration_instance_init() since at
128 * that time global properties are not yet applied, then this
129 * value will be definitely replaced by something else.
130 */
131 if (ms->enforce_config_section) {
132 current_migration->send_configuration = true;
133 }
e5cb7e76
PX		 134}
135
1f895604
VSO		 136void migration_object_finalize(void)
137{
138 object_unref(OBJECT(current_migration));
139}
140
bca7856a 141/* For outgoing */
859bc756 142MigrationState *migrate_get_current(void)
17549e84 143{
e5cb7e76
PX		 144 /* This can only be called after the object created. */
145 assert(current_migration);
146 return current_migration;
17549e84
JQ		 147}
148
bca7856a
DDAG		 149MigrationIncomingState *migration_incoming_get_current(void)
150{
b4b076da
JQ		 151 static bool once;
152 static MigrationIncomingState mis_current;
bca7856a 153
b4b076da
JQ		 154 if (!once) {
155 mis_current.state = MIGRATION_STATUS_NONE;
156 memset(&mis_current, 0, sizeof(MigrationIncomingState));
b4b076da
JQ		 157 qemu_mutex_init(&mis_current.rp_mutex);
158 qemu_event_init(&mis_current.main_thread_load_event, false);
159 once = true;
160 }
161 return &mis_current;
bca7856a
DDAG		 162}
163
164void migration_incoming_state_destroy(void)
165{
b4b076da
JQ		 166 struct MigrationIncomingState *mis = migration_incoming_get_current();
167
3482655b 168 if (mis->to_src_file) {
660819b1
PX		 169 /* Tell source that we are done */
170 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
3482655b
PX		 171 qemu_fclose(mis->to_src_file);
172 mis->to_src_file = NULL;
173 }
174
660819b1
PX		 175 if (mis->from_src_file) {
176 qemu_fclose(mis->from_src_file);
177 mis->from_src_file = NULL;
178 }
179
5089e186 180 qemu_event_reset(&mis->main_thread_load_event);
bca7856a
DDAG		 181}
182
b05dc723
JQ		 183static void migrate_generate_event(int new_state)
184{
185 if (migrate_use_events()) {
186 qapi_event_send_migration(new_state, &error_abort);
b05dc723
JQ		 187 }
188}
189
adde220a
DDAG		 190/*
191 * Called on -incoming with a defer: uri.
192 * The migration can be started later after any parameters have been
193 * changed.
194 */
195static void deferred_incoming_migration(Error **errp)
196{
197 if (deferred_incoming) {
198 error_setg(errp, "Incoming migration already deferred");
199 }
200 deferred_incoming = true;
201}
202
da6f1790
JQ		 203/*
204 * Send a message on the return channel back to the source
205 * of the migration.
206 */
207static void migrate_send_rp_message(MigrationIncomingState *mis,
208 enum mig_rp_message_type message_type,
209 uint16_t len, void *data)
210{
211 trace_migrate_send_rp_message((int)message_type, len);
212 qemu_mutex_lock(&mis->rp_mutex);
213 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
214 qemu_put_be16(mis->to_src_file, len);
215 qemu_put_buffer(mis->to_src_file, data, len);
216 qemu_fflush(mis->to_src_file);
217 qemu_mutex_unlock(&mis->rp_mutex);
218}
219
1e2d90eb
DDAG		 220/* Request a range of pages from the source VM at the given
221 * start address.
222 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
223 * as the last request (a name must have been given previously)
224 * Start: Address offset within the RB
225 * Len: Length in bytes required - must be a multiple of pagesize
226 */
227void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
228 ram_addr_t start, size_t len)
229{
cb8d4c8f 230 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
1e2d90eb
DDAG		 231 size_t msglen = 12; /* start + len */
232
233 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
234 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
235
236 if (rbname) {
237 int rbname_len = strlen(rbname);
238 assert(rbname_len < 256);
239
240 bufc[msglen++] = rbname_len;
241 memcpy(bufc + msglen, rbname, rbname_len);
242 msglen += rbname_len;
243 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
244 } else {
245 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
246 }
247}
248
43eaae28 249void qemu_start_incoming_migration(const char *uri, Error **errp)
5bb7910a 250{
34c9dd8e
AL		 251 const char *p;
252
7cf1fe6d 253 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
adde220a
DDAG		 254 if (!strcmp(uri, "defer")) {
255 deferred_incoming_migration(errp);
256 } else if (strstart(uri, "tcp:", &p)) {
43eaae28 257 tcp_start_incoming_migration(p, errp);
2da776db 258#ifdef CONFIG_RDMA
adde220a 259 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 260 rdma_start_incoming_migration(p, errp);
261#endif
adde220a 262 } else if (strstart(uri, "exec:", &p)) {
43eaae28 263 exec_start_incoming_migration(p, errp);
adde220a 264 } else if (strstart(uri, "unix:", &p)) {
43eaae28 265 unix_start_incoming_migration(p, errp);
adde220a 266 } else if (strstart(uri, "fd:", &p)) {
43eaae28 267 fd_start_incoming_migration(p, errp);
adde220a 268 } else {
312fd5f2 269 error_setg(errp, "unknown migration protocol: %s", uri);
8ca5e801 270 }
5bb7910a
AL		 271}
272
0aa6aefc
DL		 273static void process_incoming_migration_bh(void *opaque)
274{
275 Error *local_err = NULL;
276 MigrationIncomingState *mis = opaque;
277
ace21a58
KW		 278 /* Make sure all file formats flush their mutable metadata.
279 * If we get an error here, just don't restart the VM yet. */
0aa6aefc 280 bdrv_invalidate_cache_all(&local_err);
d35ff5e6 281 if (local_err) {
ace21a58 282 error_report_err(local_err);
d35ff5e6
KW		 283 local_err = NULL;
284 autostart = false;
285 }
286
0aa6aefc
DL		 287 /*
288 * This must happen after all error conditions are dealt with and
289 * we're sure the VM is going to be running on this host.
290 */
291 qemu_announce_self();
292
f986c3d2
JQ		 293 if (multifd_load_cleanup(&local_err) != 0) {
294 error_report_err(local_err);
295 autostart = false;
296 }
0aa6aefc
DL		 297 /* If global state section was not received or we are in running
298 state, we need to obey autostart. Any other state is set with
299 runstate_set. */
300
301 if (!global_state_received() ||
302 global_state_get_runstate() == RUN_STATE_RUNNING) {
303 if (autostart) {
304 vm_start();
305 } else {
306 runstate_set(RUN_STATE_PAUSED);
307 }
308 } else {
309 runstate_set(global_state_get_runstate());
310 }
0aa6aefc
DL		 311 /*
312 * This must happen after any state changes since as soon as an external
313 * observer sees this event they might start to prod at the VM assuming
314 * it's ready to use.
315 */
316 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
317 MIGRATION_STATUS_COMPLETED);
318 qemu_bh_delete(mis->bh);
319 migration_incoming_state_destroy();
320}
321
82a4da79 322static void process_incoming_migration_co(void *opaque)
511c0231 323{
b4b076da 324 MigrationIncomingState *mis = migration_incoming_get_current();
e9bef235 325 PostcopyState ps;
1c12e1f5
PB		 326 int ret;
327
4f0fae7f 328 assert(mis->from_src_file);
67f11b5c 329 mis->largest_page_size = qemu_ram_pagesize_largest();
093e3c42 330 postcopy_state_set(POSTCOPY_INCOMING_NONE);
93d7af6f
HZ		 331 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
332 MIGRATION_STATUS_ACTIVE);
4f0fae7f 333 ret = qemu_loadvm_state(mis->from_src_file);
bca7856a 334
e9bef235
DDAG		 335 ps = postcopy_state_get();
336 trace_process_incoming_migration_co_end(ret, ps);
337 if (ps != POSTCOPY_INCOMING_NONE) {
338 if (ps == POSTCOPY_INCOMING_ADVISE) {
339 /*
340 * Where a migration had postcopy enabled (and thus went to advise)
341 * but managed to complete within the precopy period, we can use
342 * the normal exit.
343 */
344 postcopy_ram_incoming_cleanup(mis);
345 } else if (ret >= 0) {
346 /*
347 * Postcopy was started, cleanup should happen at the end of the
348 * postcopy thread.
349 */
350 trace_process_incoming_migration_co_postcopy_end_main();
351 return;
352 }
353 /* Else if something went wrong then just fall out of the normal exit */
354 }
355
25d0c16f
HZ		 356 /* we get COLO info, and know if we are in COLO mode */
357 if (!ret && migration_incoming_enable_colo()) {
358 mis->migration_incoming_co = qemu_coroutine_self();
359 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
360 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
361 mis->have_colo_incoming_thread = true;
362 qemu_coroutine_yield();
363
364 /* Wait checkpoint incoming thread exit before free resource */
365 qemu_thread_join(&mis->colo_incoming_thread);
366 }
367
1c12e1f5 368 if (ret < 0) {
f986c3d2
JQ		 369 Error *local_err = NULL;
370
93d7af6f
HZ		 371 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
372 MIGRATION_STATUS_FAILED);
db80face 373 error_report("load of migration failed: %s", strerror(-ret));
3a0f2cea 374 qemu_fclose(mis->from_src_file);
f986c3d2
JQ		 375 if (multifd_load_cleanup(&local_err) != 0) {
376 error_report_err(local_err);
377 }
4aead692 378 exit(EXIT_FAILURE);
511c0231 379 }
0aa6aefc
DL		 380 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
381 qemu_bh_schedule(mis->bh);
511c0231
JQ		 382}
383
e595a01a 384static void migration_incoming_setup(QEMUFile *f)
82a4da79 385{
4f0fae7f 386 MigrationIncomingState *mis = migration_incoming_get_current();
82a4da79 387
f986c3d2
JQ		 388 if (multifd_load_setup() != 0) {
389 /* We haven't been able to create multifd threads
390 nothing better to do */
391 exit(EXIT_FAILURE);
392 }
393
4f0fae7f
JQ		 394 if (!mis->from_src_file) {
395 mis->from_src_file = f;
396 }
06ad5135 397 qemu_file_set_blocking(f, false);
e595a01a
JQ		 398}
399
400static void migration_incoming_process(void)
401{
402 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
0b8b8753 403 qemu_coroutine_enter(co);
82a4da79
PB		 404}
405
e595a01a
JQ		 406void migration_fd_process_incoming(QEMUFile *f)
407{
408 migration_incoming_setup(f);
409 migration_incoming_process();
410}
411
4f0fae7f
JQ		 412void migration_ioc_process_incoming(QIOChannel *ioc)
413{
414 MigrationIncomingState *mis = migration_incoming_get_current();
415
416 if (!mis->from_src_file) {
417 QEMUFile *f = qemu_fopen_channel_input(ioc);
418 migration_fd_process_incoming(f);
419 }
420 /* We still only have a single channel. Nothing to do here yet */
421}
422
428d8908
JQ		 423/**
424 * @migration_has_all_channels: We have received all channels that we need
425 *
426 * Returns true when we have got connections to all the channels that
427 * we need for migration.
428 */
429bool migration_has_all_channels(void)
430{
431 return true;
432}
433
6decec93
DDAG		 434/*
435 * Send a 'SHUT' message on the return channel with the given value
436 * to indicate that we've finished with the RP. Non-0 value indicates
437 * error.
438 */
439void migrate_send_rp_shut(MigrationIncomingState *mis,
440 uint32_t value)
441{
442 uint32_t buf;
443
444 buf = cpu_to_be32(value);
445 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
446}
447
448/*
449 * Send a 'PONG' message on the return channel with the given value
450 * (normally in response to a 'PING')
451 */
452void migrate_send_rp_pong(MigrationIncomingState *mis,
453 uint32_t value)
454{
455 uint32_t buf;
456
457 buf = cpu_to_be32(value);
458 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
459}
460
bbf6da32
OW		 461MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
462{
463 MigrationCapabilityStatusList *head = NULL;
464 MigrationCapabilityStatusList *caps;
465 MigrationState *s = migrate_get_current();
466 int i;
467
387eedeb 468 caps = NULL; /* silence compiler warning */
7fb1cf16 469 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
ed1701c6
DDAG		 470#ifndef CONFIG_LIVE_BLOCK_MIGRATION
471 if (i == MIGRATION_CAPABILITY_BLOCK) {
472 continue;
473 }
474#endif
bbf6da32
OW		 475 if (head == NULL) {
476 head = g_malloc0(sizeof(*caps));
477 caps = head;
478 } else {
479 caps->next = g_malloc0(sizeof(*caps));
480 caps = caps->next;
481 }
482 caps->value =
483 g_malloc(sizeof(*caps->value));
484 caps->value->capability = i;
485 caps->value->state = s->enabled_capabilities[i];
486 }
487
488 return head;
489}
490
85de8323
LL		 491MigrationParameters *qmp_query_migrate_parameters(Error **errp)
492{
493 MigrationParameters *params;
494 MigrationState *s = migrate_get_current();
495
e87fae4c 496 /* TODO use QAPI_CLONE() instead of duplicating it inline */
85de8323 497 params = g_malloc0(sizeof(*params));
de63ab61 498 params->has_compress_level = true;
2594f56d 499 params->compress_level = s->parameters.compress_level;
de63ab61 500 params->has_compress_threads = true;
2594f56d 501 params->compress_threads = s->parameters.compress_threads;
de63ab61 502 params->has_decompress_threads = true;
2594f56d 503 params->decompress_threads = s->parameters.decompress_threads;
de63ab61 504 params->has_cpu_throttle_initial = true;
2594f56d 505 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
de63ab61 506 params->has_cpu_throttle_increment = true;
2594f56d 507 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
8cc99dcd 508 params->has_tls_creds = true;
69ef1f36 509 params->tls_creds = g_strdup(s->parameters.tls_creds);
8cc99dcd 510 params->has_tls_hostname = true;
69ef1f36 511 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
2ff30257
AA		 512 params->has_max_bandwidth = true;
513 params->max_bandwidth = s->parameters.max_bandwidth;
514 params->has_downtime_limit = true;
515 params->downtime_limit = s->parameters.downtime_limit;
fe39a4d4 516 params->has_x_checkpoint_delay = true;
68b53591 517 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
2833c59b
JQ		 518 params->has_block_incremental = true;
519 params->block_incremental = s->parameters.block_incremental;
4075fb1c
JQ		 520 params->has_x_multifd_channels = true;
521 params->x_multifd_channels = s->parameters.x_multifd_channels;
0fb86605
JQ		 522 params->has_x_multifd_page_count = true;
523 params->x_multifd_page_count = s->parameters.x_multifd_page_count;
73af8dd8
JQ		 524 params->has_xbzrle_cache_size = true;
525 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
85de8323
LL		 526
527 return params;
528}
529
f6844b99
DDAG		 530/*
531 * Return true if we're already in the middle of a migration
532 * (i.e. any of the active or setup states)
533 */
534static bool migration_is_setup_or_active(int state)
535{
536 switch (state) {
537 case MIGRATION_STATUS_ACTIVE:
9ec055ae 538 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
f6844b99 539 case MIGRATION_STATUS_SETUP:
31e06077
DDAG		 540 case MIGRATION_STATUS_PRE_SWITCHOVER:
541 case MIGRATION_STATUS_DEVICE:
f6844b99
DDAG		 542 return true;
543
544 default:
545 return false;
546
547 }
548}
549
a22463a5
DDAG		 550static void populate_ram_info(MigrationInfo *info, MigrationState *s)
551{
552 info->has_ram = true;
553 info->ram = g_malloc0(sizeof(*info->ram));
9360447d 554 info->ram->transferred = ram_counters.transferred;
a22463a5 555 info->ram->total = ram_bytes_total();
9360447d 556 info->ram->duplicate = ram_counters.duplicate;
bedf53c1
JQ		 557 /* legacy value. It is not used anymore */
558 info->ram->skipped = 0;
9360447d
JQ		 559 info->ram->normal = ram_counters.normal;
560 info->ram->normal_bytes = ram_counters.normal *
20afaed9 561 qemu_target_page_size();
a22463a5 562 info->ram->mbps = s->mbps;
9360447d
JQ		 563 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
564 info->ram->postcopy_requests = ram_counters.postcopy_requests;
030ce1f8 565 info->ram->page_size = qemu_target_page_size();
a22463a5 566
114f5aee
JQ		 567 if (migrate_use_xbzrle()) {
568 info->has_xbzrle_cache = true;
569 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
570 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
9360447d
JQ		 571 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
572 info->xbzrle_cache->pages = xbzrle_counters.pages;
573 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
574 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
575 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
114f5aee
JQ		 576 }
577
338182c8
JQ		 578 if (cpu_throttle_active()) {
579 info->has_cpu_throttle_percentage = true;
580 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
581 }
582
a22463a5
DDAG		 583 if (s->state != MIGRATION_STATUS_COMPLETED) {
584 info->ram->remaining = ram_bytes_remaining();
9360447d 585 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
a22463a5
DDAG		 586 }
587}
588
930ac04c
JQ		 589static void populate_disk_info(MigrationInfo *info)
590{
591 if (blk_mig_active()) {
592 info->has_disk = true;
593 info->disk = g_malloc0(sizeof(*info->disk));
594 info->disk->transferred = blk_mig_bytes_transferred();
595 info->disk->remaining = blk_mig_bytes_remaining();
596 info->disk->total = blk_mig_bytes_total();
597 }
598}
599
ee86981b 600MigrationInfo *qmp_query_migrate(Error **errp)
5bb7910a 601{
ee86981b 602 MigrationInfo *info = g_malloc0(sizeof(*info));
17549e84
JQ		 603 MigrationState *s = migrate_get_current();
604
605 switch (s->state) {
31194731 606 case MIGRATION_STATUS_NONE:
17549e84
JQ		 607 /* no migration has happened ever */
608 break;
31194731 609 case MIGRATION_STATUS_SETUP:
29ae8a41 610 info->has_status = true;
ed4fbd10 611 info->has_total_time = false;
29ae8a41 612 break;
31194731
HZ		 613 case MIGRATION_STATUS_ACTIVE:
614 case MIGRATION_STATUS_CANCELLING:
9ec055ae 615 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
31e06077
DDAG		 616 case MIGRATION_STATUS_PRE_SWITCHOVER:
617 case MIGRATION_STATUS_DEVICE:
c8f9f4f4 618 /* TODO add some postcopy stats */
9ec055ae
DDAG		 619 info->has_status = true;
620 info->has_total_time = true;
621 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
4af246a3 622 - s->start_time;
9ec055ae
DDAG		 623 info->has_expected_downtime = true;
624 info->expected_downtime = s->expected_downtime;
625 info->has_setup_time = true;
626 info->setup_time = s->setup_time;
627
a22463a5 628 populate_ram_info(info, s);
930ac04c 629 populate_disk_info(info);
17549e84 630 break;
0b827d5e
HZ		 631 case MIGRATION_STATUS_COLO:
632 info->has_status = true;
633 /* TODO: display COLO specific information (checkpoint info etc.) */
634 break;
31194731 635 case MIGRATION_STATUS_COMPLETED:
791e7c82 636 info->has_status = true;
00c14997 637 info->has_total_time = true;
7aa939af 638 info->total_time = s->total_time;
9c5a9fcf
JQ		 639 info->has_downtime = true;
640 info->downtime = s->downtime;
ed4fbd10
MH		 641 info->has_setup_time = true;
642 info->setup_time = s->setup_time;
d5f8a570 643
a22463a5 644 populate_ram_info(info, s);
17549e84 645 break;
31194731 646 case MIGRATION_STATUS_FAILED:
791e7c82 647 info->has_status = true;
d59ce6f3
DB		 648 if (s->error) {
649 info->has_error_desc = true;
650 info->error_desc = g_strdup(error_get_pretty(s->error));
651 }
17549e84 652 break;
31194731 653 case MIGRATION_STATUS_CANCELLED:
791e7c82 654 info->has_status = true;
17549e84 655 break;
5bb7910a 656 }
cde63fbe 657 info->status = s->state;
ee86981b
PM		 658
659 return info;
5bb7910a
AL		 660}
661
4a84214e
PX		 662/**
663 * @migration_caps_check - check capability validity
664 *
665 * @cap_list: old capability list, array of bool
666 * @params: new capabilities to be applied soon
667 * @errp: set *errp if the check failed, with reason
668 *
669 * Returns true if check passed, otherwise false.
670 */
671static bool migrate_caps_check(bool *cap_list,
672 MigrationCapabilityStatusList *params,
673 Error **errp)
00458433 674{
00458433 675 MigrationCapabilityStatusList *cap;
4a84214e 676 bool old_postcopy_cap;
d7651f15 677 MigrationIncomingState *mis = migration_incoming_get_current();
00458433 678
4a84214e 679 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
00458433
OW		 680
681 for (cap = params; cap; cap = cap->next) {
4a84214e
PX		 682 cap_list[cap->value->capability] = cap->value->state;
683 }
684
ed1701c6 685#ifndef CONFIG_LIVE_BLOCK_MIGRATION
4a84214e
PX		 686 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
687 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
688 "block migration");
689 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
690 return false;
00458433 691 }
4a84214e 692#endif
53dd370c 693
4a84214e
PX		 694 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
695 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
53dd370c
DDAG		 696 /* The decompression threads asynchronously write into RAM
697 * rather than use the atomic copies needed to avoid
698 * userfaulting. It should be possible to fix the decompression
699 * threads for compatibility in future.
700 */
4a84214e
PX		 701 error_setg(errp, "Postcopy is not currently compatible "
702 "with compression");
703 return false;
53dd370c 704 }
4a84214e 705
096631bd
DDAG		 706 /* This check is reasonably expensive, so only when it's being
707 * set the first time, also it's only the destination that needs
708 * special support.
709 */
710 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
d7651f15 711 !postcopy_ram_supported_by_host(mis)) {
096631bd
DDAG		 712 /* postcopy_ram_supported_by_host will have emitted a more
713 * detailed message
714 */
4a84214e
PX		 715 error_setg(errp, "Postcopy is not supported");
716 return false;
096631bd 717 }
53dd370c 718 }
4a84214e
PX		 719
720 return true;
721}
722
723void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
724 Error **errp)
725{
726 MigrationState *s = migrate_get_current();
727 MigrationCapabilityStatusList *cap;
728
729 if (migration_is_setup_or_active(s->state)) {
730 error_setg(errp, QERR_MIGRATION_ACTIVE);
731 return;
732 }
733
734 if (!migrate_caps_check(s->enabled_capabilities, params, errp)) {
735 return;
736 }
737
738 for (cap = params; cap; cap = cap->next) {
739 s->enabled_capabilities[cap->value->capability] = cap->value->state;
740 }
00458433
OW		 741}
742
16d063bc
PX		 743/*
744 * Check whether the parameters are valid. Error will be put into errp
745 * (if provided). Return true if valid, otherwise false.
746 */
747static bool migrate_params_check(MigrationParameters *params, Error **errp)
85de8323 748{
7f375e04 749 if (params->has_compress_level &&
741d4086 750 (params->compress_level > 9)) {
c6bd8c70
MA		 751 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
752 "is invalid, it should be in the range of 0 to 9");
16d063bc 753 return false;
85de8323 754 }
16d063bc 755
741d4086 756 if (params->has_compress_threads && (params->compress_threads < 1)) {
c6bd8c70
MA		 757 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
758 "compress_threads",
759 "is invalid, it should be in the range of 1 to 255");
16d063bc 760 return false;
85de8323 761 }
16d063bc 762
741d4086 763 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
c6bd8c70
MA		 764 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
765 "decompress_threads",
766 "is invalid, it should be in the range of 1 to 255");
16d063bc 767 return false;
85de8323 768 }
16d063bc 769
7f375e04
EB		 770 if (params->has_cpu_throttle_initial &&
771 (params->cpu_throttle_initial < 1 ||
772 params->cpu_throttle_initial > 99)) {
1626fee3 773 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 774 "cpu_throttle_initial",
1626fee3 775 "an integer in the range of 1 to 99");
16d063bc 776 return false;
1626fee3 777 }
16d063bc 778
7f375e04
EB		 779 if (params->has_cpu_throttle_increment &&
780 (params->cpu_throttle_increment < 1 ||
781 params->cpu_throttle_increment > 99)) {
1626fee3 782 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
d85a31d1 783 "cpu_throttle_increment",
1626fee3 784 "an integer in the range of 1 to 99");
16d063bc 785 return false;
1626fee3 786 }
16d063bc 787
741d4086 788 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
2ff30257
AA		 789 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
790 " range of 0 to %zu bytes/second", SIZE_MAX);
16d063bc 791 return false;
2ff30257 792 }
16d063bc 793
2ff30257 794 if (params->has_downtime_limit &&
741d4086 795 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
87c9cc1c
DHB		 796 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
797 "the range of 0 to %d milliseconds",
798 MAX_MIGRATE_DOWNTIME);
16d063bc 799 return false;
2ff30257 800 }
16d063bc 801
741d4086
JQ		 802 /* x_checkpoint_delay is now always positive */
803
804 if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) {
4075fb1c
JQ		 805 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
806 "multifd_channels",
807 "is invalid, it should be in the range of 1 to 255");
808 return false;
809 }
0fb86605 810 if (params->has_x_multifd_page_count &&
741d4086
JQ		 811 (params->x_multifd_page_count < 1 ||
812 params->x_multifd_page_count > 10000)) {
0fb86605
JQ		 813 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
814 "multifd_page_count",
815 "is invalid, it should be in the range of 1 to 10000");
816 return false;
817 }
16d063bc 818
73af8dd8
JQ		 819 if (params->has_xbzrle_cache_size &&
820 (params->xbzrle_cache_size < qemu_target_page_size() ||
821 !is_power_of_2(params->xbzrle_cache_size))) {
822 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
823 "xbzrle_cache_size",
824 "is invalid, it should be bigger than target page size"
825 " and a power of two");
826 return false;
827 }
828
16d063bc
PX		 829 return true;
830}
831
1bda8b3c
MA		 832static void migrate_params_test_apply(MigrateSetParameters *params,
833 MigrationParameters *dest)
834{
835 *dest = migrate_get_current()->parameters;
836
837 /* TODO use QAPI_CLONE() instead of duplicating it inline */
838
839 if (params->has_compress_level) {
840 dest->compress_level = params->compress_level;
841 }
842
843 if (params->has_compress_threads) {
844 dest->compress_threads = params->compress_threads;
845 }
846
847 if (params->has_decompress_threads) {
848 dest->decompress_threads = params->decompress_threads;
849 }
850
851 if (params->has_cpu_throttle_initial) {
852 dest->cpu_throttle_initial = params->cpu_throttle_initial;
853 }
854
855 if (params->has_cpu_throttle_increment) {
856 dest->cpu_throttle_increment = params->cpu_throttle_increment;
857 }
858
859 if (params->has_tls_creds) {
01fa5598
MA		 860 assert(params->tls_creds->type == QTYPE_QSTRING);
861 dest->tls_creds = g_strdup(params->tls_creds->u.s);
1bda8b3c
MA		 862 }
863
864 if (params->has_tls_hostname) {
01fa5598
MA		 865 assert(params->tls_hostname->type == QTYPE_QSTRING);
866 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1bda8b3c
MA		 867 }
868
869 if (params->has_max_bandwidth) {
870 dest->max_bandwidth = params->max_bandwidth;
871 }
872
873 if (params->has_downtime_limit) {
874 dest->downtime_limit = params->downtime_limit;
875 }
876
877 if (params->has_x_checkpoint_delay) {
878 dest->x_checkpoint_delay = params->x_checkpoint_delay;
879 }
880
881 if (params->has_block_incremental) {
882 dest->block_incremental = params->block_incremental;
883 }
5e7577a1
JQ		 884 if (params->has_x_multifd_channels) {
885 dest->x_multifd_channels = params->x_multifd_channels;
886 }
887 if (params->has_x_multifd_page_count) {
888 dest->x_multifd_page_count = params->x_multifd_page_count;
889 }
73af8dd8
JQ		 890 if (params->has_xbzrle_cache_size) {
891 dest->xbzrle_cache_size = params->xbzrle_cache_size;
892 }
1bda8b3c
MA		 893}
894
73af8dd8 895static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
16d063bc
PX		 896{
897 MigrationState *s = migrate_get_current();
898
e87fae4c
MA		 899 /* TODO use QAPI_CLONE() instead of duplicating it inline */
900
7f375e04
EB		 901 if (params->has_compress_level) {
902 s->parameters.compress_level = params->compress_level;
85de8323 903 }
476c72aa 904
7f375e04
EB		 905 if (params->has_compress_threads) {
906 s->parameters.compress_threads = params->compress_threads;
85de8323 907 }
476c72aa 908
7f375e04
EB		 909 if (params->has_decompress_threads) {
910 s->parameters.decompress_threads = params->decompress_threads;
85de8323 911 }
476c72aa 912
7f375e04
EB		 913 if (params->has_cpu_throttle_initial) {
914 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1626fee3 915 }
476c72aa 916
7f375e04
EB		 917 if (params->has_cpu_throttle_increment) {
918 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1626fee3 919 }
476c72aa 920
7f375e04 921 if (params->has_tls_creds) {
69ef1f36 922 g_free(s->parameters.tls_creds);
01fa5598
MA		 923 assert(params->tls_creds->type == QTYPE_QSTRING);
924 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
69ef1f36 925 }
476c72aa 926
7f375e04 927 if (params->has_tls_hostname) {
69ef1f36 928 g_free(s->parameters.tls_hostname);
01fa5598
MA		 929 assert(params->tls_hostname->type == QTYPE_QSTRING);
930 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
69ef1f36 931 }
476c72aa 932
2ff30257
AA		 933 if (params->has_max_bandwidth) {
934 s->parameters.max_bandwidth = params->max_bandwidth;
935 if (s->to_dst_file) {
936 qemu_file_set_rate_limit(s->to_dst_file,
937 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
938 }
939 }
476c72aa 940
2ff30257
AA		 941 if (params->has_downtime_limit) {
942 s->parameters.downtime_limit = params->downtime_limit;
943 }
68b53591
HZ		 944
945 if (params->has_x_checkpoint_delay) {
946 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
479125d5
HZ		 947 if (migration_in_colo_state()) {
948 colo_checkpoint_notify(s);
949 }
68b53591 950 }
476c72aa 951
2833c59b
JQ		 952 if (params->has_block_incremental) {
953 s->parameters.block_incremental = params->block_incremental;
954 }
4075fb1c
JQ		 955 if (params->has_x_multifd_channels) {
956 s->parameters.x_multifd_channels = params->x_multifd_channels;
957 }
0fb86605
JQ		 958 if (params->has_x_multifd_page_count) {
959 s->parameters.x_multifd_page_count = params->x_multifd_page_count;
960 }
73af8dd8
JQ		 961 if (params->has_xbzrle_cache_size) {
962 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
963 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
964 }
85de8323
LL		 965}
966
1bda8b3c 967void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
476c72aa 968{
1bda8b3c
MA		 969 MigrationParameters tmp;
970
01fa5598
MA		 971 /* TODO Rewrite "" to null instead */
972 if (params->has_tls_creds
973 && params->tls_creds->type == QTYPE_QNULL) {
974 QDECREF(params->tls_creds->u.n);
975 params->tls_creds->type = QTYPE_QSTRING;
976 params->tls_creds->u.s = strdup("");
977 }
978 /* TODO Rewrite "" to null instead */
979 if (params->has_tls_hostname
980 && params->tls_hostname->type == QTYPE_QNULL) {
981 QDECREF(params->tls_hostname->u.n);
982 params->tls_hostname->type = QTYPE_QSTRING;
983 params->tls_hostname->u.s = strdup("");
984 }
985
1bda8b3c
MA		 986 migrate_params_test_apply(params, &tmp);
987
988 if (!migrate_params_check(&tmp, errp)) {
476c72aa
PX		 989 /* Invalid parameter */
990 return;
991 }
992
73af8dd8 993 migrate_params_apply(params, errp);
476c72aa
PX		 994}
995
2594f56d 996
4886a1bc
DDAG		 997void qmp_migrate_start_postcopy(Error **errp)
998{
999 MigrationState *s = migrate_get_current();
1000
1001 if (!migrate_postcopy_ram()) {
a54d340b 1002 error_setg(errp, "Enable postcopy with migrate_set_capability before"
4886a1bc
DDAG		 1003 " the start of migration");
1004 return;
1005 }
1006
1007 if (s->state == MIGRATION_STATUS_NONE) {
1008 error_setg(errp, "Postcopy must be started after migration has been"
1009 " started");
1010 return;
1011 }
1012 /*
1013 * we don't error if migration has finished since that would be racy
1014 * with issuing this command.
1015 */
1016 atomic_set(&s->start_postcopy, true);
1017}
1018
065e2813
AL		 1019/* shared migration helpers */
1020
48781e5b 1021void migrate_set_state(int *state, int old_state, int new_state)
51cf4c1a 1022{
a31fedee 1023 assert(new_state < MIGRATION_STATUS__MAX);
48781e5b 1024 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
a31fedee 1025 trace_migrate_set_state(MigrationStatus_str(new_state));
b05dc723 1026 migrate_generate_event(new_state);
51cf4c1a
Z		 1027 }
1028}
1029
4e4a3d3a
PX		 1030static MigrationCapabilityStatusList *migrate_cap_add(
1031 MigrationCapabilityStatusList *list,
1032 MigrationCapability index,
1033 bool state)
2833c59b
JQ		 1034{
1035 MigrationCapabilityStatusList *cap;
1036
1037 cap = g_new0(MigrationCapabilityStatusList, 1);
1038 cap->value = g_new0(MigrationCapabilityStatus, 1);
4e4a3d3a
PX		 1039 cap->value->capability = index;
1040 cap->value->state = state;
1041 cap->next = list;
1042
1043 return cap;
1044}
1045
1046void migrate_set_block_enabled(bool value, Error **errp)
1047{
1048 MigrationCapabilityStatusList *cap;
1049
1050 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
2833c59b
JQ		 1051 qmp_migrate_set_capabilities(cap, errp);
1052 qapi_free_MigrationCapabilityStatusList(cap);
1053}
1054
1055static void migrate_set_block_incremental(MigrationState *s, bool value)
1056{
1057 s->parameters.block_incremental = value;
1058}
1059
1060static void block_cleanup_parameters(MigrationState *s)
1061{
1062 if (s->must_remove_block_options) {
1063 /* setting to false can never fail */
1064 migrate_set_block_enabled(false, &error_abort);
1065 migrate_set_block_incremental(s, false);
1066 s->must_remove_block_options = false;
1067 }
1068}
1069
bb1fadc4 1070static void migrate_fd_cleanup(void *opaque)
065e2813 1071{
bb1fadc4
PB		 1072 MigrationState *s = opaque;
1073
1074 qemu_bh_delete(s->cleanup_bh);
1075 s->cleanup_bh = NULL;
1076
0ceccd85
PX		 1077 qemu_savevm_state_cleanup();
1078
89a02a9f 1079 if (s->to_dst_file) {
f986c3d2
JQ		 1080 Error *local_err = NULL;
1081
9013dca5 1082 trace_migrate_fd_cleanup();
404a7c05 1083 qemu_mutex_unlock_iothread();
1d34e4bf
DDAG		 1084 if (s->migration_thread_running) {
1085 qemu_thread_join(&s->thread);
1086 s->migration_thread_running = false;
1087 }
404a7c05
PB		 1088 qemu_mutex_lock_iothread();
1089
f986c3d2
JQ		 1090 if (multifd_save_cleanup(&local_err) != 0) {
1091 error_report_err(local_err);
1092 }
89a02a9f
HZ		 1093 qemu_fclose(s->to_dst_file);
1094 s->to_dst_file = NULL;
065e2813
AL		 1095 }
1096
9ec055ae
DDAG		 1097 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1098 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
7a2c1721 1099
94f5a437 1100 if (s->state == MIGRATION_STATUS_CANCELLING) {
48781e5b 1101 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
94f5a437 1102 MIGRATION_STATUS_CANCELLED);
7a2c1721 1103 }
a3fa1d78 1104
87db1a7d
JQ		 1105 if (s->error) {
1106 /* It is used on info migrate. We can't free it */
1107 error_report_err(error_copy(s->error));
1108 }
a3fa1d78 1109 notifier_list_notify(&migration_state_notifiers, s);
2833c59b 1110 block_cleanup_parameters(s);
065e2813
AL		 1111}
1112
87db1a7d
JQ		 1113void migrate_set_error(MigrationState *s, const Error *error)
1114{
1115 qemu_mutex_lock(&s->error_mutex);
1116 if (!s->error) {
1117 s->error = error_copy(error);
1118 }
1119 qemu_mutex_unlock(&s->error_mutex);
1120}
1121
d59ce6f3 1122void migrate_fd_error(MigrationState *s, const Error *error)
065e2813 1123{
25174055 1124 trace_migrate_fd_error(error_get_pretty(error));
89a02a9f 1125 assert(s->to_dst_file == NULL);
48781e5b
HZ		 1126 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1127 MIGRATION_STATUS_FAILED);
87db1a7d 1128 migrate_set_error(s, error);
458cf28e
JQ		 1129}
1130
0edda1c4 1131static void migrate_fd_cancel(MigrationState *s)
065e2813 1132{
6f2b811a 1133 int old_state ;
89a02a9f 1134 QEMUFile *f = migrate_get_current()->to_dst_file;
9013dca5 1135 trace_migrate_fd_cancel();
065e2813 1136
70b20477
DDAG		 1137 if (s->rp_state.from_dst_file) {
1138 /* shutdown the rp socket, so causing the rp thread to shutdown */
1139 qemu_file_shutdown(s->rp_state.from_dst_file);
1140 }
1141
6f2b811a
Z		 1142 do {
1143 old_state = s->state;
f6844b99 1144 if (!migration_is_setup_or_active(old_state)) {
6f2b811a
Z		 1145 break;
1146 }
a7b36b48
DDAG		 1147 /* If the migration is paused, kick it out of the pause */
1148 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1149 qemu_sem_post(&s->pause_sem);
1150 }
48781e5b 1151 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
31194731 1152 } while (s->state != MIGRATION_STATUS_CANCELLING);
a26ba26e
DDAG		 1153
1154 /*
1155 * If we're unlucky the migration code might be stuck somewhere in a
1156 * send/write while the network has failed and is waiting to timeout;
1157 * if we've got shutdown(2) available then we can force it to quit.
1158 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1159 * called in a bh, so there is no race against this cancel.
1160 */
31194731 1161 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
a26ba26e
DDAG		 1162 qemu_file_shutdown(f);
1163 }
1d2acc31
HZ		 1164 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1165 Error *local_err = NULL;
1166
1167 bdrv_invalidate_cache_all(&local_err);
1168 if (local_err) {
1169 error_report_err(local_err);
1170 } else {
1171 s->block_inactive = false;
1172 }
1173 }
065e2813
AL		 1174}
1175
99a0db9b
GH		 1176void add_migration_state_change_notifier(Notifier *notify)
1177{
1178 notifier_list_add(&migration_state_notifiers, notify);
1179}
1180
1181void remove_migration_state_change_notifier(Notifier *notify)
1182{
31552529 1183 notifier_remove(notify);
99a0db9b
GH		 1184}
1185
02edd2e7 1186bool migration_in_setup(MigrationState *s)
afe2df69 1187{
31194731 1188 return s->state == MIGRATION_STATUS_SETUP;
afe2df69
GH		 1189}
1190
7073693b 1191bool migration_has_finished(MigrationState *s)
99a0db9b 1192{
31194731 1193 return s->state == MIGRATION_STATUS_COMPLETED;
99a0db9b 1194}
0edda1c4 1195
afe2df69
GH		 1196bool migration_has_failed(MigrationState *s)
1197{
31194731
HZ		 1198 return (s->state == MIGRATION_STATUS_CANCELLED ||
1199 s->state == MIGRATION_STATUS_FAILED);
afe2df69
GH		 1200}
1201
5727309d 1202bool migration_in_postcopy(void)
9ec055ae 1203{
5727309d
JQ		 1204 MigrationState *s = migrate_get_current();
1205
9ec055ae
DDAG		 1206 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1207}
1208
b82fc321
DDAG		 1209bool migration_in_postcopy_after_devices(MigrationState *s)
1210{
5727309d 1211 return migration_in_postcopy() && s->postcopy_after_devices;
b82fc321
DDAG		 1212}
1213
fab35005 1214bool migration_is_idle(void)
fe44dc91 1215{
fab35005 1216 MigrationState *s = migrate_get_current();
fe44dc91
AA		 1217
1218 switch (s->state) {
1219 case MIGRATION_STATUS_NONE:
1220 case MIGRATION_STATUS_CANCELLED:
1221 case MIGRATION_STATUS_COMPLETED:
1222 case MIGRATION_STATUS_FAILED:
1223 return true;
1224 case MIGRATION_STATUS_SETUP:
1225 case MIGRATION_STATUS_CANCELLING:
1226 case MIGRATION_STATUS_ACTIVE:
1227 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1228 case MIGRATION_STATUS_COLO:
31e06077
DDAG		 1229 case MIGRATION_STATUS_PRE_SWITCHOVER:
1230 case MIGRATION_STATUS_DEVICE:
fe44dc91
AA		 1231 return false;
1232 case MIGRATION_STATUS__MAX:
1233 g_assert_not_reached();
1234 }
1235
1236 return false;
1237}
1238
a0762d9e 1239MigrationState *migrate_init(void)
0edda1c4 1240{
17549e84 1241 MigrationState *s = migrate_get_current();
bbf6da32 1242
389775d1
DDAG		 1243 /*
1244 * Reinitialise all migration state, except
1245 * parameters/capabilities that the user set, and
1246 * locks.
1247 */
1248 s->bytes_xfer = 0;
1249 s->xfer_limit = 0;
1250 s->cleanup_bh = 0;
89a02a9f 1251 s->to_dst_file = NULL;
389775d1 1252 s->state = MIGRATION_STATUS_NONE;
389775d1
DDAG		 1253 s->rp_state.from_dst_file = NULL;
1254 s->rp_state.error = false;
1255 s->mbps = 0.0;
1256 s->downtime = 0;
1257 s->expected_downtime = 0;
389775d1 1258 s->setup_time = 0;
389775d1 1259 s->start_postcopy = false;
b82fc321 1260 s->postcopy_after_devices = false;
389775d1 1261 s->migration_thread_running = false;
d59ce6f3
DB		 1262 error_free(s->error);
1263 s->error = NULL;
389775d1 1264
48781e5b 1265 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
0edda1c4 1266
4af246a3
PX		 1267 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1268 s->total_time = 0;
7287cbd4 1269 s->vm_was_running = false;
b15df1ae
PX		 1270 s->iteration_initial_bytes = 0;
1271 s->threshold_size = 0;
0edda1c4
JQ		 1272 return s;
1273}
cab30143 1274
fa2756b7
AL		 1275static GSList *migration_blockers;
1276
fe44dc91 1277int migrate_add_blocker(Error *reason, Error **errp)
fa2756b7 1278{
3df663e5 1279 if (migrate_get_current()->only_migratable) {
b67b8c3a
AA		 1280 error_propagate(errp, error_copy(reason));
1281 error_prepend(errp, "disallowing migration blocker "
1282 "(--only_migratable) for: ");
1283 return -EACCES;
1284 }
1285
fab35005 1286 if (migration_is_idle()) {
fe44dc91
AA		 1287 migration_blockers = g_slist_prepend(migration_blockers, reason);
1288 return 0;
1289 }
1290
1291 error_propagate(errp, error_copy(reason));
1292 error_prepend(errp, "disallowing migration blocker (migration in "
1293 "progress) for: ");
1294 return -EBUSY;
fa2756b7
AL		 1295}
1296
1297void migrate_del_blocker(Error *reason)
1298{
1299 migration_blockers = g_slist_remove(migration_blockers, reason);
1300}
1301
bf1ae1f4
DDAG		 1302void qmp_migrate_incoming(const char *uri, Error **errp)
1303{
1304 Error *local_err = NULL;
4debb5f5 1305 static bool once = true;
bf1ae1f4
DDAG		 1306
1307 if (!deferred_incoming) {
4debb5f5 1308 error_setg(errp, "For use with '-incoming defer'");
bf1ae1f4
DDAG		 1309 return;
1310 }
4debb5f5
DDAG		 1311 if (!once) {
1312 error_setg(errp, "The incoming migration has already been started");
1313 }
bf1ae1f4
DDAG		 1314
1315 qemu_start_incoming_migration(uri, &local_err);
1316
1317 if (local_err) {
1318 error_propagate(errp, local_err);
1319 return;
1320 }
1321
4debb5f5 1322 once = false;
bf1ae1f4
DDAG		 1323}
1324
24f3902b
GK		 1325bool migration_is_blocked(Error **errp)
1326{
1327 if (qemu_savevm_state_blocked(errp)) {
1328 return true;
1329 }
1330
1331 if (migration_blockers) {
250561e1 1332 error_propagate(errp, error_copy(migration_blockers->data));
24f3902b
GK		 1333 return true;
1334 }
1335
1336 return false;
1337}
1338
e1c37d0e
LC		 1339void qmp_migrate(const char *uri, bool has_blk, bool blk,
1340 bool has_inc, bool inc, bool has_detach, bool detach,
1341 Error **errp)
cab30143 1342{
be7059cd 1343 Error *local_err = NULL;
17549e84 1344 MigrationState *s = migrate_get_current();
cab30143 1345 const char *p;
cab30143 1346
f6844b99 1347 if (migration_is_setup_or_active(s->state) ||
0b827d5e
HZ		 1348 s->state == MIGRATION_STATUS_CANCELLING ||
1349 s->state == MIGRATION_STATUS_COLO) {
c6bd8c70 1350 error_setg(errp, QERR_MIGRATION_ACTIVE);
e1c37d0e 1351 return;
cab30143 1352 }
ca99993a
DDAG		 1353 if (runstate_check(RUN_STATE_INMIGRATE)) {
1354 error_setg(errp, "Guest is waiting for an incoming migration");
1355 return;
1356 }
1357
24f3902b 1358 if (migration_is_blocked(errp)) {
e1c37d0e 1359 return;
fa2756b7
AL		 1360 }
1361
2833c59b
JQ		 1362 if ((has_blk && blk) || (has_inc && inc)) {
1363 if (migrate_use_block() || migrate_use_block_incremental()) {
1364 error_setg(errp, "Command options are incompatible with "
1365 "current migration capabilities");
1366 return;
1367 }
1368 migrate_set_block_enabled(true, &local_err);
1369 if (local_err) {
1370 error_propagate(errp, local_err);
1371 return;
1372 }
1373 s->must_remove_block_options = true;
1374 }
1375
1376 if (has_inc && inc) {
1377 migrate_set_block_incremental(s, true);
1378 }
1379
a0762d9e 1380 s = migrate_init();
cab30143
JQ		 1381
1382 if (strstart(uri, "tcp:", &p)) {
f37afb5a 1383 tcp_start_outgoing_migration(s, p, &local_err);
2da776db 1384#ifdef CONFIG_RDMA
41310c68 1385 } else if (strstart(uri, "rdma:", &p)) {
2da776db
MH		 1386 rdma_start_outgoing_migration(s, p, &local_err);
1387#endif
cab30143 1388 } else if (strstart(uri, "exec:", &p)) {
f37afb5a 1389 exec_start_outgoing_migration(s, p, &local_err);
cab30143 1390 } else if (strstart(uri, "unix:", &p)) {
f37afb5a 1391 unix_start_outgoing_migration(s, p, &local_err);
cab30143 1392 } else if (strstart(uri, "fd:", &p)) {
f37afb5a 1393 fd_start_outgoing_migration(s, p, &local_err);
99a0db9b 1394 } else {
c6bd8c70
MA		 1395 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1396 "a valid migration protocol");
48781e5b
HZ		 1397 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1398 MIGRATION_STATUS_FAILED);
e1c37d0e 1399 return;
cab30143
JQ		 1400 }
1401
f37afb5a 1402 if (local_err) {
d59ce6f3 1403 migrate_fd_error(s, local_err);
f37afb5a 1404 error_propagate(errp, local_err);
e1c37d0e 1405 return;
1299c631 1406 }
cab30143
JQ		 1407}
1408
6cdedb07 1409void qmp_migrate_cancel(Error **errp)
cab30143 1410{
17549e84 1411 migrate_fd_cancel(migrate_get_current());
cab30143
JQ		 1412}
1413
89cfc02c
DDAG		 1414void qmp_migrate_continue(MigrationStatus state, Error **errp)
1415{
1416 MigrationState *s = migrate_get_current();
1417 if (s->state != state) {
1418 error_setg(errp, "Migration not in expected state: %s",
1419 MigrationStatus_str(s->state));
1420 return;
1421 }
1422 qemu_sem_post(&s->pause_sem);
1423}
1424
9e1ba4cc
OW		 1425void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1426{
73af8dd8
JQ		 1427 MigrateSetParameters p = {
1428 .has_xbzrle_cache_size = true,
1429 .xbzrle_cache_size = value,
1430 };
c91e681a 1431
73af8dd8 1432 qmp_migrate_set_parameters(&p, errp);
9e1ba4cc
OW		 1433}
1434
1435int64_t qmp_query_migrate_cache_size(Error **errp)
1436{
1437 return migrate_xbzrle_cache_size();
1438}
1439
3dc85383 1440void qmp_migrate_set_speed(int64_t value, Error **errp)
cab30143 1441{
1bda8b3c 1442 MigrateSetParameters p = {
2ff30257
AA		 1443 .has_max_bandwidth = true,
1444 .max_bandwidth = value,
1445 };
cab30143 1446
2ff30257 1447 qmp_migrate_set_parameters(&p, errp);
cab30143
JQ		 1448}
1449
4f0a993b 1450void qmp_migrate_set_downtime(double value, Error **errp)
cab30143 1451{
87c9cc1c
DHB		 1452 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1453 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1454 "the range of 0 to %d seconds",
1455 MAX_MIGRATE_DOWNTIME_SECONDS);
1456 return;
1457 }
1458
2ff30257
AA		 1459 value *= 1000; /* Convert to milliseconds */
1460 value = MAX(0, MIN(INT64_MAX, value));
1461
1bda8b3c 1462 MigrateSetParameters p = {
2ff30257
AA		 1463 .has_downtime_limit = true,
1464 .downtime_limit = value,
1465 };
1466
1467 qmp_migrate_set_parameters(&p, errp);
99a0db9b 1468}
17ad9b35 1469
53f09a10
PB		 1470bool migrate_release_ram(void)
1471{
1472 MigrationState *s;
1473
1474 s = migrate_get_current();
1475
1476 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1477}
1478
53dd370c
DDAG		 1479bool migrate_postcopy_ram(void)
1480{
1481 MigrationState *s;
1482
1483 s = migrate_get_current();
1484
32c3db5b 1485 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
53dd370c
DDAG		 1486}
1487
58110f0a
VSO		 1488bool migrate_postcopy(void)
1489{
1490 return migrate_postcopy_ram();
1491}
1492
bde1e2ec
CV		 1493bool migrate_auto_converge(void)
1494{
1495 MigrationState *s;
1496
1497 s = migrate_get_current();
1498
1499 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1500}
1501
323004a3
PL		 1502bool migrate_zero_blocks(void)
1503{
1504 MigrationState *s;
1505
1506 s = migrate_get_current();
1507
1508 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1509}
1510
8706d2d5
LL		 1511bool migrate_use_compression(void)
1512{
dde4e694
LL		 1513 MigrationState *s;
1514
1515 s = migrate_get_current();
1516
1517 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
8706d2d5
LL		 1518}
1519
1520int migrate_compress_level(void)
1521{
1522 MigrationState *s;
1523
1524 s = migrate_get_current();
1525
2594f56d 1526 return s->parameters.compress_level;
8706d2d5
LL		 1527}
1528
1529int migrate_compress_threads(void)
1530{
1531 MigrationState *s;
1532
1533 s = migrate_get_current();
1534
2594f56d 1535 return s->parameters.compress_threads;
8706d2d5
LL		 1536}
1537
3fcb38c2
LL		 1538int migrate_decompress_threads(void)
1539{
1540 MigrationState *s;
1541
1542 s = migrate_get_current();
1543
2594f56d 1544 return s->parameters.decompress_threads;
3fcb38c2
LL		 1545}
1546
b05dc723
JQ		 1547bool migrate_use_events(void)
1548{
1549 MigrationState *s;
1550
1551 s = migrate_get_current();
1552
1553 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1554}
1555
30126bbf
JQ		 1556bool migrate_use_multifd(void)
1557{
1558 MigrationState *s;
1559
1560 s = migrate_get_current();
1561
1562 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1563}
1564
93fbd031
DDAG		 1565bool migrate_pause_before_switchover(void)
1566{
1567 MigrationState *s;
1568
1569 s = migrate_get_current();
1570
1571 return s->enabled_capabilities[
1572 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
1573}
1574
4075fb1c
JQ		 1575int migrate_multifd_channels(void)
1576{
1577 MigrationState *s;
1578
1579 s = migrate_get_current();
1580
1581 return s->parameters.x_multifd_channels;
1582}
1583
0fb86605
JQ		 1584int migrate_multifd_page_count(void)
1585{
1586 MigrationState *s;
1587
1588 s = migrate_get_current();
1589
1590 return s->parameters.x_multifd_page_count;
1591}
1592
17ad9b35
OW		 1593int migrate_use_xbzrle(void)
1594{
1595 MigrationState *s;
1596
1597 s = migrate_get_current();
1598
1599 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1600}
1601
1602int64_t migrate_xbzrle_cache_size(void)
1603{
1604 MigrationState *s;
1605
1606 s = migrate_get_current();
1607
73af8dd8 1608 return s->parameters.xbzrle_cache_size;
17ad9b35 1609}
0d82d0e8 1610
2833c59b
JQ		 1611bool migrate_use_block(void)
1612{
1613 MigrationState *s;
1614
1615 s = migrate_get_current();
1616
1617 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1618}
1619
c788ada8
PX		 1620bool migrate_use_return_path(void)
1621{
1622 MigrationState *s;
1623
1624 s = migrate_get_current();
1625
1626 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1627}
1628
2833c59b
JQ		 1629bool migrate_use_block_incremental(void)
1630{
1631 MigrationState *s;
1632
1633 s = migrate_get_current();
1634
1635 return s->parameters.block_incremental;
1636}
1637
70b20477
DDAG		 1638/* migration thread support */
1639/*
1640 * Something bad happened to the RP stream, mark an error
1641 * The caller shall print or trace something to indicate why
1642 */
1643static void mark_source_rp_bad(MigrationState *s)
1644{
1645 s->rp_state.error = true;
1646}
1647
1648static struct rp_cmd_args {
1649 ssize_t len; /* -1 = variable */
1650 const char *name;
1651} rp_cmd_args[] = {
1652 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1653 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1654 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1e2d90eb
DDAG		 1655 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1656 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
70b20477
DDAG		 1657 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1658};
1659
1e2d90eb
DDAG		 1660/*
1661 * Process a request for pages received on the return path,
1662 * We're allowed to send more than requested (e.g. to round to our page size)
1663 * and we don't need to send pages that have already been sent.
1664 */
1665static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1666 ram_addr_t start, size_t len)
1667{
6c595cde
DDAG		 1668 long our_host_ps = getpagesize();
1669
1e2d90eb 1670 trace_migrate_handle_rp_req_pages(rbname, start, len);
6c595cde
DDAG		 1671
1672 /*
1673 * Since we currently insist on matching page sizes, just sanity check
1674 * we're being asked for whole host pages.
1675 */
1676 if (start & (our_host_ps-1) ||
1677 (len & (our_host_ps-1))) {
1678 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1679 " len: %zd", __func__, start, len);
1680 mark_source_rp_bad(ms);
1681 return;
1682 }
1683
96506894 1684 if (ram_save_queue_pages(rbname, start, len)) {
6c595cde
DDAG		 1685 mark_source_rp_bad(ms);
1686 }
1e2d90eb
DDAG		 1687}
1688
70b20477
DDAG		 1689/*
1690 * Handles messages sent on the return path towards the source VM
1691 *
1692 */
1693static void *source_return_path_thread(void *opaque)
1694{
1695 MigrationState *ms = opaque;
1696 QEMUFile *rp = ms->rp_state.from_dst_file;
1697 uint16_t header_len, header_type;
568b01ca 1698 uint8_t buf[512];
70b20477 1699 uint32_t tmp32, sibling_error;
1e2d90eb
DDAG		 1700 ram_addr_t start = 0; /* =0 to silence warning */
1701 size_t len = 0, expected_len;
70b20477
DDAG		 1702 int res;
1703
1704 trace_source_return_path_thread_entry();
1705 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1706 migration_is_setup_or_active(ms->state)) {
1707 trace_source_return_path_thread_loop_top();
1708 header_type = qemu_get_be16(rp);
1709 header_len = qemu_get_be16(rp);
1710
1711 if (header_type >= MIG_RP_MSG_MAX ||
1712 header_type == MIG_RP_MSG_INVALID) {
1713 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1714 header_type, header_len);
1715 mark_source_rp_bad(ms);
1716 goto out;
1717 }
1718
1719 if ((rp_cmd_args[header_type].len != -1 &&
1720 header_len != rp_cmd_args[header_type].len) ||
568b01ca 1721 header_len > sizeof(buf)) {
70b20477
DDAG		 1722 error_report("RP: Received '%s' message (0x%04x) with"
1723 "incorrect length %d expecting %zu",
1724 rp_cmd_args[header_type].name, header_type, header_len,
1725 (size_t)rp_cmd_args[header_type].len);
1726 mark_source_rp_bad(ms);
1727 goto out;
1728 }
1729
1730 /* We know we've got a valid header by this point */
1731 res = qemu_get_buffer(rp, buf, header_len);
1732 if (res != header_len) {
1733 error_report("RP: Failed reading data for message 0x%04x"
1734 " read %d expected %d",
1735 header_type, res, header_len);
1736 mark_source_rp_bad(ms);
1737 goto out;
1738 }
1739
1740 /* OK, we have the message and the data */
1741 switch (header_type) {
1742 case MIG_RP_MSG_SHUT:
4d885131 1743 sibling_error = ldl_be_p(buf);
70b20477
DDAG		 1744 trace_source_return_path_thread_shut(sibling_error);
1745 if (sibling_error) {
1746 error_report("RP: Sibling indicated error %d", sibling_error);
1747 mark_source_rp_bad(ms);
1748 }
1749 /*
1750 * We'll let the main thread deal with closing the RP
1751 * we could do a shutdown(2) on it, but we're the only user
1752 * anyway, so there's nothing gained.
1753 */
1754 goto out;
1755
1756 case MIG_RP_MSG_PONG:
4d885131 1757 tmp32 = ldl_be_p(buf);
70b20477
DDAG		 1758 trace_source_return_path_thread_pong(tmp32);
1759 break;
1760
1e2d90eb 1761 case MIG_RP_MSG_REQ_PAGES:
4d885131
PM		 1762 start = ldq_be_p(buf);
1763 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 1764 migrate_handle_rp_req_pages(ms, NULL, start, len);
1765 break;
1766
1767 case MIG_RP_MSG_REQ_PAGES_ID:
1768 expected_len = 12 + 1; /* header + termination */
1769
1770 if (header_len >= expected_len) {
4d885131
PM		 1771 start = ldq_be_p(buf);
1772 len = ldl_be_p(buf + 8);
1e2d90eb
DDAG		 1773 /* Now we expect an idstr */
1774 tmp32 = buf[12]; /* Length of the following idstr */
1775 buf[13 + tmp32] = '\0';
1776 expected_len += tmp32;
1777 }
1778 if (header_len != expected_len) {
1779 error_report("RP: Req_Page_id with length %d expecting %zd",
1780 header_len, expected_len);
1781 mark_source_rp_bad(ms);
1782 goto out;
1783 }
1784 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1785 break;
1786
70b20477
DDAG		 1787 default:
1788 break;
1789 }
1790 }
5df5416e 1791 if (qemu_file_get_error(rp)) {
70b20477
DDAG		 1792 trace_source_return_path_thread_bad_end();
1793 mark_source_rp_bad(ms);
1794 }
1795
1796 trace_source_return_path_thread_end();
1797out:
1798 ms->rp_state.from_dst_file = NULL;
1799 qemu_fclose(rp);
1800 return NULL;
1801}
1802
70b20477
DDAG		 1803static int open_return_path_on_source(MigrationState *ms)
1804{
1805
89a02a9f 1806 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
70b20477
DDAG		 1807 if (!ms->rp_state.from_dst_file) {
1808 return -1;
1809 }
1810
1811 trace_open_return_path_on_source();
1812 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1813 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1814
1815 trace_open_return_path_on_source_continue();
1816
1817 return 0;
1818}
1819
70b20477
DDAG		 1820/* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1821static int await_return_path_close_on_source(MigrationState *ms)
1822{
1823 /*
1824 * If this is a normal exit then the destination will send a SHUT and the
1825 * rp_thread will exit, however if there's an error we need to cause
1826 * it to exit.
1827 */
89a02a9f 1828 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
70b20477
DDAG		 1829 /*
1830 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1831 * waiting for the destination.
1832 */
1833 qemu_file_shutdown(ms->rp_state.from_dst_file);
1834 mark_source_rp_bad(ms);
1835 }
1836 trace_await_return_path_close_on_source_joining();
1837 qemu_thread_join(&ms->rp_state.rp_thread);
1838 trace_await_return_path_close_on_source_close();
1839 return ms->rp_state.error;
1840}
1841
1d34e4bf
DDAG		 1842/*
1843 * Switch from normal iteration to postcopy
1844 * Returns non-0 on error
1845 */
7287cbd4 1846static int postcopy_start(MigrationState *ms)
1d34e4bf
DDAG		 1847{
1848 int ret;
61b67d47
DB		 1849 QIOChannelBuffer *bioc;
1850 QEMUFile *fb;
1d34e4bf 1851 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
ef8d6488 1852 bool restart_block = false;
0331c8ca
DDAG		 1853 int cur_state = MIGRATION_STATUS_ACTIVE;
1854 if (!migrate_pause_before_switchover()) {
1855 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1856 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1857 }
1d34e4bf
DDAG		 1858
1859 trace_postcopy_start();
1860 qemu_mutex_lock_iothread();
1861 trace_postcopy_start_set_run();
1862
1863 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1d34e4bf
DDAG		 1864 global_state_store();
1865 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
76b1c7fe
KW		 1866 if (ret < 0) {
1867 goto fail;
1868 }
1d34e4bf 1869
0331c8ca
DDAG		 1870 ret = migration_maybe_pause(ms, &cur_state,
1871 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1872 if (ret < 0) {
1873 goto fail;
1874 }
1875
76b1c7fe 1876 ret = bdrv_inactivate_all();
1d34e4bf
DDAG		 1877 if (ret < 0) {
1878 goto fail;
1879 }
ef8d6488 1880 restart_block = true;
1d34e4bf 1881
1c0d249d
DDAG		 1882 /*
1883 * Cause any non-postcopiable, but iterative devices to
1884 * send out their final data.
1885 */
a1fbe750 1886 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1c0d249d 1887
1d34e4bf
DDAG		 1888 /*
1889 * in Finish migrate and with the io-lock held everything should
1890 * be quiet, but we've potentially still got dirty pages and we
1891 * need to tell the destination to throw any pages it's already received
1892 * that are dirty
1893 */
58110f0a
VSO		 1894 if (migrate_postcopy_ram()) {
1895 if (ram_postcopy_send_discard_bitmap(ms)) {
1896 error_report("postcopy send discard bitmap failed");
1897 goto fail;
1898 }
1d34e4bf
DDAG		 1899 }
1900
1901 /*
1902 * send rest of state - note things that are doing postcopy
1903 * will notice we're in POSTCOPY_ACTIVE and not actually
1904 * wrap their state up here
1905 */
89a02a9f 1906 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
58110f0a
VSO		 1907 if (migrate_postcopy_ram()) {
1908 /* Ping just for debugging, helps line traces up */
1909 qemu_savevm_send_ping(ms->to_dst_file, 2);
1910 }
1d34e4bf
DDAG		 1911
1912 /*
1913 * While loading the device state we may trigger page transfer
1914 * requests and the fd must be free to process those, and thus
1915 * the destination must read the whole device state off the fd before
1916 * it starts processing it. Unfortunately the ad-hoc migration format
1917 * doesn't allow the destination to know the size to read without fully
1918 * parsing it through each devices load-state code (especially the open
1919 * coded devices that use get/put).
1920 * So we wrap the device state up in a package with a length at the start;
1921 * to do this we use a qemu_buf to hold the whole of the device state.
1922 */
61b67d47 1923 bioc = qio_channel_buffer_new(4096);
6f01f136 1924 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
61b67d47
DB		 1925 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1926 object_unref(OBJECT(bioc));
1d34e4bf 1927
c76201ab
DDAG		 1928 /*
1929 * Make sure the receiver can get incoming pages before we send the rest
1930 * of the state
1931 */
1932 qemu_savevm_send_postcopy_listen(fb);
1933
a1fbe750 1934 qemu_savevm_state_complete_precopy(fb, false, false);
58110f0a
VSO		 1935 if (migrate_postcopy_ram()) {
1936 qemu_savevm_send_ping(fb, 3);
1937 }
1d34e4bf
DDAG		 1938
1939 qemu_savevm_send_postcopy_run(fb);
1940
1941 /* <><> end of stuff going into the package */
1d34e4bf 1942
ef8d6488
DDAG		 1943 /* Last point of recovery; as soon as we send the package the destination
1944 * can open devices and potentially start running.
1945 * Lets just check again we've not got any errors.
1946 */
1947 ret = qemu_file_get_error(ms->to_dst_file);
1948 if (ret) {
1949 error_report("postcopy_start: Migration stream errored (pre package)");
1950 goto fail_closefb;
1951 }
1952
1953 restart_block = false;
1954
1d34e4bf 1955 /* Now send that blob */
61b67d47 1956 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1d34e4bf
DDAG		 1957 goto fail_closefb;
1958 }
1959 qemu_fclose(fb);
b82fc321
DDAG		 1960
1961 /* Send a notify to give a chance for anything that needs to happen
1962 * at the transition to postcopy and after the device state; in particular
1963 * spice needs to trigger a transition now
1964 */
1965 ms->postcopy_after_devices = true;
1966 notifier_list_notify(&migration_state_notifiers, ms);
1967
1d34e4bf
DDAG		 1968 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1969
1970 qemu_mutex_unlock_iothread();
1971
58110f0a
VSO		 1972 if (migrate_postcopy_ram()) {
1973 /*
1974 * Although this ping is just for debug, it could potentially be
1975 * used for getting a better measurement of downtime at the source.
1976 */
1977 qemu_savevm_send_ping(ms->to_dst_file, 4);
1978 }
1d34e4bf 1979
ced1c616
PB		 1980 if (migrate_release_ram()) {
1981 ram_postcopy_migrated_memory_release(ms);
1982 }
1983
89a02a9f 1984 ret = qemu_file_get_error(ms->to_dst_file);
1d34e4bf
DDAG		 1985 if (ret) {
1986 error_report("postcopy_start: Migration stream errored");
48781e5b 1987 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf
DDAG		 1988 MIGRATION_STATUS_FAILED);
1989 }
1990
1991 return ret;
1992
1993fail_closefb:
1994 qemu_fclose(fb);
1995fail:
48781e5b 1996 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1d34e4bf 1997 MIGRATION_STATUS_FAILED);
ef8d6488
DDAG		 1998 if (restart_block) {
1999 /* A failure happened early enough that we know the destination hasn't
2000 * accessed block devices, so we're safe to recover.
2001 */
2002 Error *local_err = NULL;
2003
2004 bdrv_invalidate_cache_all(&local_err);
2005 if (local_err) {
2006 error_report_err(local_err);
2007 }
2008 }
1d34e4bf
DDAG		 2009 qemu_mutex_unlock_iothread();
2010 return -1;
2011}
2012
e91d8951
DDAG		 2013/**
2014 * migration_maybe_pause: Pause if required to by
2015 * migrate_pause_before_switchover called with the iothread locked
2016 * Returns: 0 on success
2017 */
0331c8ca
DDAG		 2018static int migration_maybe_pause(MigrationState *s,
2019 int *current_active_state,
2020 int new_state)
e91d8951
DDAG		 2021{
2022 if (!migrate_pause_before_switchover()) {
2023 return 0;
2024 }
2025
2026 /* Since leaving this state is not atomic with posting the semaphore
2027 * it's possible that someone could have issued multiple migrate_continue
2028 * and the semaphore is incorrectly positive at this point;
2029 * the docs say it's undefined to reinit a semaphore that's already
2030 * init'd, so use timedwait to eat up any existing posts.
2031 */
2032 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2033 /* This block intentionally left blank */
2034 }
2035
2036 qemu_mutex_unlock_iothread();
2037 migrate_set_state(&s->state, *current_active_state,
2038 MIGRATION_STATUS_PRE_SWITCHOVER);
2039 qemu_sem_wait(&s->pause_sem);
2040 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
0331c8ca
DDAG		 2041 new_state);
2042 *current_active_state = new_state;
e91d8951
DDAG		 2043 qemu_mutex_lock_iothread();
2044
0331c8ca 2045 return s->state == new_state ? 0 : -EINVAL;
e91d8951
DDAG		 2046}
2047
09f6c85e
DDAG		 2048/**
2049 * migration_completion: Used by migration_thread when there's not much left.
2050 * The caller 'breaks' the loop when this returns.
2051 *
2052 * @s: Current migration state
09f6c85e 2053 */
2ad87305 2054static void migration_completion(MigrationState *s)
09f6c85e
DDAG		 2055{
2056 int ret;
2ad87305 2057 int current_active_state = s->state;
09f6c85e 2058
b10ac0c4
DDAG		 2059 if (s->state == MIGRATION_STATUS_ACTIVE) {
2060 qemu_mutex_lock_iothread();
64909f97 2061 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
b10ac0c4 2062 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
7287cbd4 2063 s->vm_was_running = runstate_is_running();
b10ac0c4
DDAG		 2064 ret = global_state_store();
2065
2066 if (!ret) {
a1fbe750 2067 bool inactivate = !migrate_colo_enabled();
b10ac0c4 2068 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
e91d8951 2069 if (ret >= 0) {
0331c8ca
DDAG		 2070 ret = migration_maybe_pause(s, &current_active_state,
2071 MIGRATION_STATUS_DEVICE);
e91d8951 2072 }
f07fa4cb
KW		 2073 if (ret >= 0) {
2074 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
a1fbe750
FZ		 2075 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2076 inactivate);
f07fa4cb 2077 }
a1fbe750
FZ		 2078 if (inactivate && ret >= 0) {
2079 s->block_inactive = true;
b10ac0c4
DDAG		 2080 }
2081 }
2082 qemu_mutex_unlock_iothread();
09f6c85e 2083
b10ac0c4
DDAG		 2084 if (ret < 0) {
2085 goto fail;
09f6c85e 2086 }
b10ac0c4
DDAG		 2087 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2088 trace_migration_completion_postcopy_end();
2089
89a02a9f 2090 qemu_savevm_state_complete_postcopy(s->to_dst_file);
b10ac0c4 2091 trace_migration_completion_postcopy_end_after_complete();
09f6c85e 2092 }
09f6c85e 2093
b10ac0c4
DDAG		 2094 /*
2095 * If rp was opened we must clean up the thread before
2096 * cleaning everything else up (since if there are no failures
2097 * it will wait for the destination to send it's status in
2098 * a SHUT command).
b10ac0c4 2099 */
0425dc97 2100 if (s->rp_state.from_dst_file) {
b10ac0c4 2101 int rp_error;
0425dc97 2102 trace_migration_return_path_end_before();
b10ac0c4 2103 rp_error = await_return_path_close_on_source(s);
0425dc97 2104 trace_migration_return_path_end_after(rp_error);
b10ac0c4 2105 if (rp_error) {
fe904ea8 2106 goto fail_invalidate;
b10ac0c4 2107 }
09f6c85e
DDAG		 2108 }
2109
89a02a9f 2110 if (qemu_file_get_error(s->to_dst_file)) {
09f6c85e 2111 trace_migration_completion_file_err();
fe904ea8 2112 goto fail_invalidate;
09f6c85e
DDAG		 2113 }
2114
0b827d5e
HZ		 2115 if (!migrate_colo_enabled()) {
2116 migrate_set_state(&s->state, current_active_state,
2117 MIGRATION_STATUS_COMPLETED);
2118 }
2119
09f6c85e
DDAG		 2120 return;
2121
fe904ea8
GK		 2122fail_invalidate:
2123 /* If not doing postcopy, vm_start() will be called: let's regain
2124 * control on images.
2125 */
6039dd5b
DDAG		 2126 if (s->state == MIGRATION_STATUS_ACTIVE ||
2127 s->state == MIGRATION_STATUS_DEVICE) {
fe904ea8
GK		 2128 Error *local_err = NULL;
2129
1d2acc31 2130 qemu_mutex_lock_iothread();
fe904ea8
GK		 2131 bdrv_invalidate_cache_all(&local_err);
2132 if (local_err) {
2133 error_report_err(local_err);
1d2acc31
HZ		 2134 } else {
2135 s->block_inactive = false;
fe904ea8 2136 }
1d2acc31 2137 qemu_mutex_unlock_iothread();
fe904ea8
GK		 2138 }
2139
09f6c85e 2140fail:
48781e5b
HZ		 2141 migrate_set_state(&s->state, current_active_state,
2142 MIGRATION_STATUS_FAILED);
09f6c85e
DDAG		 2143}
2144
35a6ed4f
HZ		 2145bool migrate_colo_enabled(void)
2146{
2147 MigrationState *s = migrate_get_current();
2148 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2149}
2150
cf011f08
PX		 2151static void migration_calculate_complete(MigrationState *s)
2152{
2153 uint64_t bytes = qemu_ftell(s->to_dst_file);
2154 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2155
2156 s->total_time = end_time - s->start_time;
2157 if (!s->downtime) {
2158 /*
2159 * It's still not set, so we are precopy migration. For
2160 * postcopy, downtime is calculated during postcopy_start().
2161 */
2162 s->downtime = end_time - s->downtime_start;
2163 }
2164
2165 if (s->total_time) {
2166 s->mbps = ((double) bytes * 8.0) / s->total_time / 1000;
2167 }
2168}
2169
b15df1ae
PX		 2170static void migration_update_counters(MigrationState *s,
2171 int64_t current_time)
2172{
2173 uint64_t transferred, time_spent;
b15df1ae
PX		 2174 double bandwidth;
2175
2176 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
2177 return;
2178 }
2179
2180 transferred = qemu_ftell(s->to_dst_file) - s->iteration_initial_bytes;
2181 time_spent = current_time - s->iteration_start_time;
2182 bandwidth = (double)transferred / time_spent;
0781c1ed 2183 s->threshold_size = bandwidth * s->parameters.downtime_limit;
b15df1ae
PX		 2184
2185 s->mbps = (((double) transferred * 8.0) /
2186 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2187
2188 /*
2189 * if we haven't sent anything, we don't want to
2190 * recalculate. 10000 is a small enough number for our purposes
2191 */
2192 if (ram_counters.dirty_pages_rate && transferred > 10000) {
2193 s->expected_downtime = ram_counters.dirty_pages_rate *
2194 qemu_target_page_size() / bandwidth;
2195 }
2196
2197 qemu_file_reset_rate_limit(s->to_dst_file);
2198
2199 s->iteration_start_time = current_time;
2200 s->iteration_initial_bytes = qemu_ftell(s->to_dst_file);
2201
2202 trace_migrate_transferred(transferred, time_spent,
0781c1ed 2203 bandwidth, s->threshold_size);
b15df1ae
PX		 2204}
2205
2ad87305
PX		 2206/* Migration thread iteration status */
2207typedef enum {
2208 MIG_ITERATE_RESUME, /* Resume current iteration */
2209 MIG_ITERATE_SKIP, /* Skip current iteration */
2210 MIG_ITERATE_BREAK, /* Break the loop */
2211} MigIterateState;
2212
2213/*
2214 * Return true if continue to the next iteration directly, false
2215 * otherwise.
2216 */
2217static MigIterateState migration_iteration_run(MigrationState *s)
2218{
2219 uint64_t pending_size, pend_post, pend_nonpost;
2220 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
2221
2222 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size,
2223 &pend_nonpost, &pend_post);
2224 pending_size = pend_nonpost + pend_post;
2225
2226 trace_migrate_pending(pending_size, s->threshold_size,
2227 pend_post, pend_nonpost);
2228
2229 if (pending_size && pending_size >= s->threshold_size) {
2230 /* Still a significant amount to transfer */
2231 if (migrate_postcopy() && !in_postcopy &&
2232 pend_nonpost <= s->threshold_size &&
2233 atomic_read(&s->start_postcopy)) {
2234 if (postcopy_start(s)) {
2235 error_report("%s: postcopy failed to start", __func__);
2236 }
2237 return MIG_ITERATE_SKIP;
2238 }
2239 /* Just another iteration step */
2240 qemu_savevm_state_iterate(s->to_dst_file,
2241 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2242 } else {
2243 trace_migration_thread_low_pending(pending_size);
2244 migration_completion(s);
2245 return MIG_ITERATE_BREAK;
2246 }
2247
2248 return MIG_ITERATE_RESUME;
2249}
2250
199aa6d4
PX		 2251static void migration_iteration_finish(MigrationState *s)
2252{
2253 /* If we enabled cpu throttling for auto-converge, turn it off. */
2254 cpu_throttle_stop();
2255
2256 qemu_mutex_lock_iothread();
2257 switch (s->state) {
2258 case MIGRATION_STATUS_COMPLETED:
2259 migration_calculate_complete(s);
2260 runstate_set(RUN_STATE_POSTMIGRATE);
2261 break;
2262
2263 case MIGRATION_STATUS_ACTIVE:
2264 /*
2265 * We should really assert here, but since it's during
2266 * migration, let's try to reduce the usage of assertions.
2267 */
2268 if (!migrate_colo_enabled()) {
2269 error_report("%s: critical error: calling COLO code without "
2270 "COLO enabled", __func__);
2271 }
2272 migrate_start_colo_process(s);
2273 /*
2274 * Fixme: we will run VM in COLO no matter its old running state.
2275 * After exited COLO, we will keep running.
2276 */
2277 s->vm_was_running = true;
2278 /* Fallthrough */
2279 case MIGRATION_STATUS_FAILED:
2280 case MIGRATION_STATUS_CANCELLED:
2281 if (s->vm_was_running) {
2282 vm_start();
2283 } else {
2284 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2285 runstate_set(RUN_STATE_POSTMIGRATE);
2286 }
2287 }
2288 break;
2289
2290 default:
2291 /* Should not reach here, but if so, forgive the VM. */
2292 error_report("%s: Unknown ending state %d", __func__, s->state);
2293 break;
2294 }
2295 qemu_bh_schedule(s->cleanup_bh);
2296 qemu_mutex_unlock_iothread();
2297}
2298
70b20477
DDAG		 2299/*
2300 * Master migration thread on the source VM.
2301 * It drives the migration and pumps the data down the outgoing channel.
2302 */
5f496a1b 2303static void *migration_thread(void *opaque)
0d82d0e8 2304{
9848a404 2305 MigrationState *s = opaque;
bc72ad67 2306 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
76f5933a 2307
ab28bd23
PB		 2308 rcu_register_thread();
2309
b15df1ae
PX		 2310 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2311
89a02a9f 2312 qemu_savevm_state_header(s->to_dst_file);
1d34e4bf 2313
62a02658
PX		 2314 /*
2315 * If we opened the return path, we need to make sure dst has it
2316 * opened as well.
2317 */
2318 if (s->rp_state.from_dst_file) {
1d34e4bf 2319 /* Now tell the dest that it should open its end so it can reply */
89a02a9f 2320 qemu_savevm_send_open_return_path(s->to_dst_file);
1d34e4bf
DDAG		 2321
2322 /* And do a ping that will make stuff easier to debug */
89a02a9f 2323 qemu_savevm_send_ping(s->to_dst_file, 1);
0425dc97 2324 }
1d34e4bf 2325
58110f0a 2326 if (migrate_postcopy()) {
1d34e4bf
DDAG		 2327 /*
2328 * Tell the destination that we *might* want to do postcopy later;
2329 * if the other end can't do postcopy it should fail now, nice and
2330 * early.
2331 */
89a02a9f 2332 qemu_savevm_send_postcopy_advise(s->to_dst_file);
1d34e4bf
DDAG		 2333 }
2334
9907e842 2335 qemu_savevm_state_setup(s->to_dst_file);
0d82d0e8 2336
bc72ad67 2337 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
48781e5b
HZ		 2338 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2339 MIGRATION_STATUS_ACTIVE);
29ae8a41 2340
9ec055ae
DDAG		 2341 trace_migration_thread_setup_complete();
2342
2343 while (s->state == MIGRATION_STATUS_ACTIVE ||
2344 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
a3e879cd 2345 int64_t current_time;
0d82d0e8 2346
89a02a9f 2347 if (!qemu_file_rate_limit(s->to_dst_file)) {
2ad87305
PX		 2348 MigIterateState iter_state = migration_iteration_run(s);
2349 if (iter_state == MIG_ITERATE_SKIP) {
2350 continue;
2351 } else if (iter_state == MIG_ITERATE_BREAK) {
09f6c85e 2352 break;
c369f40d
JQ		 2353 }
2354 }
f4410a5d 2355
89a02a9f 2356 if (qemu_file_get_error(s->to_dst_file)) {
2ad87305
PX		 2357 if (migration_is_setup_or_active(s->state)) {
2358 migrate_set_state(&s->state, s->state,
2359 MIGRATION_STATUS_FAILED);
2360 }
1d34e4bf 2361 trace_migration_thread_file_err();
fd45ee2c
PB		 2362 break;
2363 }
b15df1ae 2364
bc72ad67 2365 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
0d82d0e8 2366
b15df1ae
PX		 2367 migration_update_counters(s, current_time);
2368
89a02a9f 2369 if (qemu_file_rate_limit(s->to_dst_file)) {
0d82d0e8 2370 /* usleep expects microseconds */
b15df1ae
PX		 2371 g_usleep((s->iteration_start_time + BUFFER_DELAY -
2372 current_time) * 1000);
0d82d0e8 2373 }
a3fa1d78
PB		 2374 }
2375
1d34e4bf 2376 trace_migration_thread_after_loop();
199aa6d4 2377 migration_iteration_finish(s);
ab28bd23 2378 rcu_unregister_thread();
0d82d0e8
JQ		 2379 return NULL;
2380}
2381
cce8040b 2382void migrate_fd_connect(MigrationState *s, Error *error_in)
0d82d0e8 2383{
2ff30257 2384 s->expected_downtime = s->parameters.downtime_limit;
bb1fadc4 2385 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
cce8040b
DDAG		 2386 if (error_in) {
2387 migrate_fd_error(s, error_in);
2388 migrate_fd_cleanup(s);
2389 return;
2390 }
0d82d0e8 2391
9e4d2b98 2392 qemu_file_set_blocking(s->to_dst_file, true);
89a02a9f 2393 qemu_file_set_rate_limit(s->to_dst_file,
2ff30257 2394 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
442773ce 2395
9287ac27
SH		 2396 /* Notify before starting migration thread */
2397 notifier_list_notify(&migration_state_notifiers, s);
2398
1d34e4bf 2399 /*
c788ada8
PX		 2400 * Open the return path. For postcopy, it is used exclusively. For
2401 * precopy, only if user specified "return-path" capability would
2402 * QEMU uses the return path.
1d34e4bf 2403 */
c788ada8 2404 if (migrate_postcopy_ram() || migrate_use_return_path()) {
1d34e4bf
DDAG		 2405 if (open_return_path_on_source(s)) {
2406 error_report("Unable to open return-path for postcopy");
48781e5b 2407 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1d34e4bf
DDAG		 2408 MIGRATION_STATUS_FAILED);
2409 migrate_fd_cleanup(s);
2410 return;
2411 }
2412 }
2413
f986c3d2
JQ		 2414 if (multifd_save_setup() != 0) {
2415 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2416 MIGRATION_STATUS_FAILED);
2417 migrate_fd_cleanup(s);
2418 return;
2419 }
009fad7f 2420 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
bb1fadc4 2421 QEMU_THREAD_JOINABLE);
1d34e4bf 2422 s->migration_thread_running = true;
0d82d0e8 2423}
093e3c42 2424
9d18af93
PX		 2425void migration_global_dump(Monitor *mon)
2426{
2427 MigrationState *ms = migrate_get_current();
2428
6f0f6428
JQ		 2429 monitor_printf(mon, "globals:\n");
2430 monitor_printf(mon, "store-global-state: %s\n",
2431 ms->store_global_state ? "on" : "off");
2432 monitor_printf(mon, "only-migratable: %s\n",
2433 ms->only_migratable ? "on" : "off");
2434 monitor_printf(mon, "send-configuration: %s\n",
2435 ms->send_configuration ? "on" : "off");
2436 monitor_printf(mon, "send-section-footer: %s\n",
2437 ms->send_section_footer ? "on" : "off");
9d18af93
PX		 2438}
2439
20814758
PX		 2440#define DEFINE_PROP_MIG_CAP(name, x) \
2441 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2442
5272298c
PX		 2443static Property migration_properties[] = {
2444 DEFINE_PROP_BOOL("store-global-state", MigrationState,
2445 store_global_state, true),
3df663e5 2446 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
71dd4c1a
PX		 2447 DEFINE_PROP_BOOL("send-configuration", MigrationState,
2448 send_configuration, true),
15c38503
PX		 2449 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2450 send_section_footer, true),
89632faf
PX		 2451
2452 /* Migration parameters */
741d4086 2453 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
89632faf
PX		 2454 parameters.compress_level,
2455 DEFAULT_MIGRATE_COMPRESS_LEVEL),
741d4086 2456 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
89632faf
PX		 2457 parameters.compress_threads,
2458 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
741d4086 2459 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
89632faf
PX		 2460 parameters.decompress_threads,
2461 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
741d4086 2462 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
89632faf
PX		 2463 parameters.cpu_throttle_initial,
2464 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
741d4086 2465 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
89632faf
PX		 2466 parameters.cpu_throttle_increment,
2467 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
741d4086 2468 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
89632faf 2469 parameters.max_bandwidth, MAX_THROTTLE),
741d4086 2470 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
89632faf
PX		 2471 parameters.downtime_limit,
2472 DEFAULT_MIGRATE_SET_DOWNTIME),
741d4086 2473 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
89632faf
PX		 2474 parameters.x_checkpoint_delay,
2475 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
741d4086 2476 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState,
4075fb1c
JQ		 2477 parameters.x_multifd_channels,
2478 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
741d4086 2479 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState,
0fb86605
JQ		 2480 parameters.x_multifd_page_count,
2481 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
73af8dd8
JQ		 2482 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
2483 parameters.xbzrle_cache_size,
2484 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
20814758
PX		 2485
2486 /* Migration capabilities */
2487 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2488 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2489 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2490 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2491 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2492 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2493 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2494 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2495 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2496 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2497 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
30126bbf 2498 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
20814758 2499
5272298c
PX		 2500 DEFINE_PROP_END_OF_LIST(),
2501};
2502
e5cb7e76
PX		 2503static void migration_class_init(ObjectClass *klass, void *data)
2504{
2505 DeviceClass *dc = DEVICE_CLASS(klass);
2506
2507 dc->user_creatable = false;
5272298c 2508 dc->props = migration_properties;
e5cb7e76
PX		 2509}
2510
b91bf5e4
MAL		 2511static void migration_instance_finalize(Object *obj)
2512{
2513 MigrationState *ms = MIGRATION_OBJ(obj);
2514 MigrationParameters *params = &ms->parameters;
2515
87db1a7d 2516 qemu_mutex_destroy(&ms->error_mutex);
b91bf5e4
MAL		 2517 g_free(params->tls_hostname);
2518 g_free(params->tls_creds);
e91d8951 2519 qemu_sem_destroy(&ms->pause_sem);
b91bf5e4
MAL		 2520}
2521
e5cb7e76
PX		 2522static void migration_instance_init(Object *obj)
2523{
2524 MigrationState *ms = MIGRATION_OBJ(obj);
8b0b29dc 2525 MigrationParameters *params = &ms->parameters;
e5cb7e76
PX		 2526
2527 ms->state = MIGRATION_STATUS_NONE;
e5cb7e76 2528 ms->mbps = -1;
e91d8951 2529 qemu_sem_init(&ms->pause_sem, 0);
87db1a7d 2530 qemu_mutex_init(&ms->error_mutex);
8b0b29dc
PX		 2531
2532 params->tls_hostname = g_strdup("");
2533 params->tls_creds = g_strdup("");
2534
2535 /* Set has_* up only for parameter checks */
2536 params->has_compress_level = true;
2537 params->has_compress_threads = true;
2538 params->has_decompress_threads = true;
2539 params->has_cpu_throttle_initial = true;
2540 params->has_cpu_throttle_increment = true;
2541 params->has_max_bandwidth = true;
2542 params->has_downtime_limit = true;
2543 params->has_x_checkpoint_delay = true;
2544 params->has_block_incremental = true;
4075fb1c 2545 params->has_x_multifd_channels = true;
0fb86605 2546 params->has_x_multifd_page_count = true;
73af8dd8 2547 params->has_xbzrle_cache_size = true;
8b0b29dc
PX		 2548}
2549
2550/*
2551 * Return true if check pass, false otherwise. Error will be put
2552 * inside errp if provided.
2553 */
2554static bool migration_object_check(MigrationState *ms, Error **errp)
2555{
6b19a7d9
PX		 2556 MigrationCapabilityStatusList *head = NULL;
2557 /* Assuming all off */
2558 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2559 int i;
2560
8b0b29dc
PX		 2561 if (!migrate_params_check(&ms->parameters, errp)) {
2562 return false;
2563 }
2564
6b19a7d9
PX		 2565 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2566 if (ms->enabled_capabilities[i]) {
2567 head = migrate_cap_add(head, i, true);
2568 }
2569 }
2570
2571 ret = migrate_caps_check(cap_list, head, errp);
2572
2573 /* It works with head == NULL */
2574 qapi_free_MigrationCapabilityStatusList(head);
2575
2576 return ret;
e5cb7e76
PX		 2577}
2578
2579static const TypeInfo migration_type = {
2580 .name = TYPE_MIGRATION,
01f6e14c 2581 /*
c8d3ff38
PX		 2582 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2583 * not created using qdev_create(), it is not attached to the qdev
2584 * device tree, and it is never realized.
2585 *
2586 * TODO: Make this TYPE_OBJECT once QOM provides something like
2587 * TYPE_DEVICE's "-global" properties.
01f6e14c 2588 */
e5cb7e76
PX		 2589 .parent = TYPE_DEVICE,
2590 .class_init = migration_class_init,
2591 .class_size = sizeof(MigrationClass),
2592 .instance_size = sizeof(MigrationState),
2593 .instance_init = migration_instance_init,
b91bf5e4 2594 .instance_finalize = migration_instance_finalize,
e5cb7e76
PX		 2595};
2596
2597static void register_migration_types(void)
2598{
2599 type_register_static(&migration_type);
2600}
2601
2602type_init(register_migration_types);
Empty description
This page took 1.097405 seconds and 4 git commands to generate.