sm501: Fix hardware cursor color conversion
[qemu/rayw.git] / migration / migration.c
blobe1eaa97df4eb31a365880b3814f04ca49da373a0
1 /*
2 * QEMU live migration
4 * Copyright IBM, Corp. 2008
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
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.
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "migration/blocker.h"
20 #include "exec.h"
21 #include "fd.h"
22 #include "socket.h"
23 #include "rdma.h"
24 #include "ram.h"
25 #include "migration/global_state.h"
26 #include "migration/misc.h"
27 #include "migration.h"
28 #include "savevm.h"
29 #include "qemu-file-channel.h"
30 #include "qemu-file.h"
31 #include "migration/vmstate.h"
32 #include "block/block.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-commands-migration.h"
35 #include "qapi/qapi-events-migration.h"
36 #include "qapi/qmp/qerror.h"
37 #include "qapi/qmp/qnull.h"
38 #include "qemu/rcu.h"
39 #include "block.h"
40 #include "postcopy-ram.h"
41 #include "qemu/thread.h"
42 #include "trace.h"
43 #include "exec/target_page.h"
44 #include "io/channel-buffer.h"
45 #include "migration/colo.h"
46 #include "hw/boards.h"
47 #include "monitor/monitor.h"
49 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
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)
56 /* Time in milliseconds we are allowed to stop the source,
57 * for sending the last part */
58 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
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)
64 /* Default compression thread count */
65 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
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
69 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
70 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
71 /* Define default autoconverge cpu throttle migration parameters */
72 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
73 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
75 /* Migration XBZRLE default cache size */
76 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
78 /* The delay time (in ms) between two COLO checkpoints
79 * Note: Please change this default value to 10000 when we support hybrid mode.
81 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
82 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
83 #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
85 /* Background transfer rate for postcopy, 0 means unlimited, note
86 * that page requests can still exceed this limit.
88 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
90 static NotifierList migration_state_notifiers =
91 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
93 static bool deferred_incoming;
95 /* Messages sent on the return path from destination to source */
96 enum mig_rp_message_type {
97 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
98 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
99 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
101 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
102 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
103 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
104 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
106 MIG_RP_MSG_MAX
109 /* When we add fault tolerance, we could have several
110 migrations at once. For now we don't need to add
111 dynamic creation of migration */
113 static MigrationState *current_migration;
114 static MigrationIncomingState *current_incoming;
116 static bool migration_object_check(MigrationState *ms, Error **errp);
117 static int migration_maybe_pause(MigrationState *s,
118 int *current_active_state,
119 int new_state);
121 void migration_object_init(void)
123 MachineState *ms = MACHINE(qdev_get_machine());
124 Error *err = NULL;
126 /* This can only be called once. */
127 assert(!current_migration);
128 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
131 * Init the migrate incoming object as well no matter whether
132 * we'll use it or not.
134 assert(!current_incoming);
135 current_incoming = g_new0(MigrationIncomingState, 1);
136 current_incoming->state = MIGRATION_STATUS_NONE;
137 current_incoming->postcopy_remote_fds =
138 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
139 qemu_mutex_init(&current_incoming->rp_mutex);
140 qemu_event_init(&current_incoming->main_thread_load_event, false);
141 qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
142 qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
144 init_dirty_bitmap_incoming_migration();
146 if (!migration_object_check(current_migration, &err)) {
147 error_report_err(err);
148 exit(1);
152 * We cannot really do this in migration_instance_init() since at
153 * that time global properties are not yet applied, then this
154 * value will be definitely replaced by something else.
156 if (ms->enforce_config_section) {
157 current_migration->send_configuration = true;
161 void migration_object_finalize(void)
163 object_unref(OBJECT(current_migration));
166 /* For outgoing */
167 MigrationState *migrate_get_current(void)
169 /* This can only be called after the object created. */
170 assert(current_migration);
171 return current_migration;
174 MigrationIncomingState *migration_incoming_get_current(void)
176 assert(current_incoming);
177 return current_incoming;
180 void migration_incoming_state_destroy(void)
182 struct MigrationIncomingState *mis = migration_incoming_get_current();
184 if (mis->to_src_file) {
185 /* Tell source that we are done */
186 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
187 qemu_fclose(mis->to_src_file);
188 mis->to_src_file = NULL;
191 if (mis->from_src_file) {
192 qemu_fclose(mis->from_src_file);
193 mis->from_src_file = NULL;
195 if (mis->postcopy_remote_fds) {
196 g_array_free(mis->postcopy_remote_fds, TRUE);
197 mis->postcopy_remote_fds = NULL;
200 qemu_event_reset(&mis->main_thread_load_event);
203 static void migrate_generate_event(int new_state)
205 if (migrate_use_events()) {
206 qapi_event_send_migration(new_state, &error_abort);
210 static bool migrate_late_block_activate(void)
212 MigrationState *s;
214 s = migrate_get_current();
216 return s->enabled_capabilities[
217 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
221 * Called on -incoming with a defer: uri.
222 * The migration can be started later after any parameters have been
223 * changed.
225 static void deferred_incoming_migration(Error **errp)
227 if (deferred_incoming) {
228 error_setg(errp, "Incoming migration already deferred");
230 deferred_incoming = true;
234 * Send a message on the return channel back to the source
235 * of the migration.
237 static int migrate_send_rp_message(MigrationIncomingState *mis,
238 enum mig_rp_message_type message_type,
239 uint16_t len, void *data)
241 int ret = 0;
243 trace_migrate_send_rp_message((int)message_type, len);
244 qemu_mutex_lock(&mis->rp_mutex);
247 * It's possible that the file handle got lost due to network
248 * failures.
250 if (!mis->to_src_file) {
251 ret = -EIO;
252 goto error;
255 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
256 qemu_put_be16(mis->to_src_file, len);
257 qemu_put_buffer(mis->to_src_file, data, len);
258 qemu_fflush(mis->to_src_file);
260 /* It's possible that qemu file got error during sending */
261 ret = qemu_file_get_error(mis->to_src_file);
263 error:
264 qemu_mutex_unlock(&mis->rp_mutex);
265 return ret;
268 /* Request a range of pages from the source VM at the given
269 * start address.
270 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
271 * as the last request (a name must have been given previously)
272 * Start: Address offset within the RB
273 * Len: Length in bytes required - must be a multiple of pagesize
275 int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
276 ram_addr_t start, size_t len)
278 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
279 size_t msglen = 12; /* start + len */
280 enum mig_rp_message_type msg_type;
282 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
283 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
285 if (rbname) {
286 int rbname_len = strlen(rbname);
287 assert(rbname_len < 256);
289 bufc[msglen++] = rbname_len;
290 memcpy(bufc + msglen, rbname, rbname_len);
291 msglen += rbname_len;
292 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
293 } else {
294 msg_type = MIG_RP_MSG_REQ_PAGES;
297 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
300 void qemu_start_incoming_migration(const char *uri, Error **errp)
302 const char *p;
304 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
305 if (!strcmp(uri, "defer")) {
306 deferred_incoming_migration(errp);
307 } else if (strstart(uri, "tcp:", &p)) {
308 tcp_start_incoming_migration(p, errp);
309 #ifdef CONFIG_RDMA
310 } else if (strstart(uri, "rdma:", &p)) {
311 rdma_start_incoming_migration(p, errp);
312 #endif
313 } else if (strstart(uri, "exec:", &p)) {
314 exec_start_incoming_migration(p, errp);
315 } else if (strstart(uri, "unix:", &p)) {
316 unix_start_incoming_migration(p, errp);
317 } else if (strstart(uri, "fd:", &p)) {
318 fd_start_incoming_migration(p, errp);
319 } else {
320 error_setg(errp, "unknown migration protocol: %s", uri);
324 static void process_incoming_migration_bh(void *opaque)
326 Error *local_err = NULL;
327 MigrationIncomingState *mis = opaque;
329 /* If capability late_block_activate is set:
330 * Only fire up the block code now if we're going to restart the
331 * VM, else 'cont' will do it.
332 * This causes file locking to happen; so we don't want it to happen
333 * unless we really are starting the VM.
335 if (!migrate_late_block_activate() ||
336 (autostart && (!global_state_received() ||
337 global_state_get_runstate() == RUN_STATE_RUNNING))) {
338 /* Make sure all file formats flush their mutable metadata.
339 * If we get an error here, just don't restart the VM yet. */
340 bdrv_invalidate_cache_all(&local_err);
341 if (local_err) {
342 error_report_err(local_err);
343 local_err = NULL;
344 autostart = false;
349 * This must happen after all error conditions are dealt with and
350 * we're sure the VM is going to be running on this host.
352 qemu_announce_self();
354 if (multifd_load_cleanup(&local_err) != 0) {
355 error_report_err(local_err);
356 autostart = false;
358 /* If global state section was not received or we are in running
359 state, we need to obey autostart. Any other state is set with
360 runstate_set. */
362 dirty_bitmap_mig_before_vm_start();
364 if (!global_state_received() ||
365 global_state_get_runstate() == RUN_STATE_RUNNING) {
366 if (autostart) {
367 vm_start();
368 } else {
369 runstate_set(RUN_STATE_PAUSED);
371 } else {
372 runstate_set(global_state_get_runstate());
375 * This must happen after any state changes since as soon as an external
376 * observer sees this event they might start to prod at the VM assuming
377 * it's ready to use.
379 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
380 MIGRATION_STATUS_COMPLETED);
381 qemu_bh_delete(mis->bh);
382 migration_incoming_state_destroy();
385 static void process_incoming_migration_co(void *opaque)
387 MigrationIncomingState *mis = migration_incoming_get_current();
388 PostcopyState ps;
389 int ret;
391 assert(mis->from_src_file);
392 mis->largest_page_size = qemu_ram_pagesize_largest();
393 postcopy_state_set(POSTCOPY_INCOMING_NONE);
394 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
395 MIGRATION_STATUS_ACTIVE);
396 ret = qemu_loadvm_state(mis->from_src_file);
398 ps = postcopy_state_get();
399 trace_process_incoming_migration_co_end(ret, ps);
400 if (ps != POSTCOPY_INCOMING_NONE) {
401 if (ps == POSTCOPY_INCOMING_ADVISE) {
403 * Where a migration had postcopy enabled (and thus went to advise)
404 * but managed to complete within the precopy period, we can use
405 * the normal exit.
407 postcopy_ram_incoming_cleanup(mis);
408 } else if (ret >= 0) {
410 * Postcopy was started, cleanup should happen at the end of the
411 * postcopy thread.
413 trace_process_incoming_migration_co_postcopy_end_main();
414 return;
416 /* Else if something went wrong then just fall out of the normal exit */
419 /* we get COLO info, and know if we are in COLO mode */
420 if (!ret && migration_incoming_enable_colo()) {
421 mis->migration_incoming_co = qemu_coroutine_self();
422 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
423 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
424 mis->have_colo_incoming_thread = true;
425 qemu_coroutine_yield();
427 /* Wait checkpoint incoming thread exit before free resource */
428 qemu_thread_join(&mis->colo_incoming_thread);
431 if (ret < 0) {
432 Error *local_err = NULL;
434 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
435 MIGRATION_STATUS_FAILED);
436 error_report("load of migration failed: %s", strerror(-ret));
437 qemu_fclose(mis->from_src_file);
438 if (multifd_load_cleanup(&local_err) != 0) {
439 error_report_err(local_err);
441 exit(EXIT_FAILURE);
443 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
444 qemu_bh_schedule(mis->bh);
447 static void migration_incoming_setup(QEMUFile *f)
449 MigrationIncomingState *mis = migration_incoming_get_current();
451 if (multifd_load_setup() != 0) {
452 /* We haven't been able to create multifd threads
453 nothing better to do */
454 exit(EXIT_FAILURE);
457 if (!mis->from_src_file) {
458 mis->from_src_file = f;
460 qemu_file_set_blocking(f, false);
463 void migration_incoming_process(void)
465 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
466 qemu_coroutine_enter(co);
469 void migration_fd_process_incoming(QEMUFile *f)
471 MigrationIncomingState *mis = migration_incoming_get_current();
473 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
474 /* Resumed from a paused postcopy migration */
476 mis->from_src_file = f;
477 /* Postcopy has standalone thread to do vm load */
478 qemu_file_set_blocking(f, true);
480 /* Re-configure the return path */
481 mis->to_src_file = qemu_file_get_return_path(f);
483 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
484 MIGRATION_STATUS_POSTCOPY_RECOVER);
487 * Here, we only wake up the main loading thread (while the
488 * fault thread will still be waiting), so that we can receive
489 * commands from source now, and answer it if needed. The
490 * fault thread will be woken up afterwards until we are sure
491 * that source is ready to reply to page requests.
493 qemu_sem_post(&mis->postcopy_pause_sem_dst);
494 } else {
495 /* New incoming migration */
496 migration_incoming_setup(f);
497 migration_incoming_process();
501 void migration_ioc_process_incoming(QIOChannel *ioc)
503 MigrationIncomingState *mis = migration_incoming_get_current();
505 if (!mis->from_src_file) {
506 QEMUFile *f = qemu_fopen_channel_input(ioc);
507 migration_incoming_setup(f);
508 return;
510 multifd_recv_new_channel(ioc);
514 * @migration_has_all_channels: We have received all channels that we need
516 * Returns true when we have got connections to all the channels that
517 * we need for migration.
519 bool migration_has_all_channels(void)
521 bool all_channels;
523 all_channels = multifd_recv_all_channels_created();
525 return all_channels;
529 * Send a 'SHUT' message on the return channel with the given value
530 * to indicate that we've finished with the RP. Non-0 value indicates
531 * error.
533 void migrate_send_rp_shut(MigrationIncomingState *mis,
534 uint32_t value)
536 uint32_t buf;
538 buf = cpu_to_be32(value);
539 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
543 * Send a 'PONG' message on the return channel with the given value
544 * (normally in response to a 'PING')
546 void migrate_send_rp_pong(MigrationIncomingState *mis,
547 uint32_t value)
549 uint32_t buf;
551 buf = cpu_to_be32(value);
552 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
555 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
556 char *block_name)
558 char buf[512];
559 int len;
560 int64_t res;
563 * First, we send the header part. It contains only the len of
564 * idstr, and the idstr itself.
566 len = strlen(block_name);
567 buf[0] = len;
568 memcpy(buf + 1, block_name, len);
570 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
571 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
572 __func__);
573 return;
576 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
579 * Next, we dump the received bitmap to the stream.
581 * TODO: currently we are safe since we are the only one that is
582 * using the to_src_file handle (fault thread is still paused),
583 * and it's ok even not taking the mutex. However the best way is
584 * to take the lock before sending the message header, and release
585 * the lock after sending the bitmap.
587 qemu_mutex_lock(&mis->rp_mutex);
588 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
589 qemu_mutex_unlock(&mis->rp_mutex);
591 trace_migrate_send_rp_recv_bitmap(block_name, res);
594 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
596 uint32_t buf;
598 buf = cpu_to_be32(value);
599 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
602 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
604 MigrationCapabilityStatusList *head = NULL;
605 MigrationCapabilityStatusList *caps;
606 MigrationState *s = migrate_get_current();
607 int i;
609 caps = NULL; /* silence compiler warning */
610 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
611 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
612 if (i == MIGRATION_CAPABILITY_BLOCK) {
613 continue;
615 #endif
616 if (head == NULL) {
617 head = g_malloc0(sizeof(*caps));
618 caps = head;
619 } else {
620 caps->next = g_malloc0(sizeof(*caps));
621 caps = caps->next;
623 caps->value =
624 g_malloc(sizeof(*caps->value));
625 caps->value->capability = i;
626 caps->value->state = s->enabled_capabilities[i];
629 return head;
632 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
634 MigrationParameters *params;
635 MigrationState *s = migrate_get_current();
637 /* TODO use QAPI_CLONE() instead of duplicating it inline */
638 params = g_malloc0(sizeof(*params));
639 params->has_compress_level = true;
640 params->compress_level = s->parameters.compress_level;
641 params->has_compress_threads = true;
642 params->compress_threads = s->parameters.compress_threads;
643 params->has_decompress_threads = true;
644 params->decompress_threads = s->parameters.decompress_threads;
645 params->has_cpu_throttle_initial = true;
646 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
647 params->has_cpu_throttle_increment = true;
648 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
649 params->has_tls_creds = true;
650 params->tls_creds = g_strdup(s->parameters.tls_creds);
651 params->has_tls_hostname = true;
652 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
653 params->has_max_bandwidth = true;
654 params->max_bandwidth = s->parameters.max_bandwidth;
655 params->has_downtime_limit = true;
656 params->downtime_limit = s->parameters.downtime_limit;
657 params->has_x_checkpoint_delay = true;
658 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
659 params->has_block_incremental = true;
660 params->block_incremental = s->parameters.block_incremental;
661 params->has_x_multifd_channels = true;
662 params->x_multifd_channels = s->parameters.x_multifd_channels;
663 params->has_x_multifd_page_count = true;
664 params->x_multifd_page_count = s->parameters.x_multifd_page_count;
665 params->has_xbzrle_cache_size = true;
666 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
667 params->has_max_postcopy_bandwidth = true;
668 params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
670 return params;
674 * Return true if we're already in the middle of a migration
675 * (i.e. any of the active or setup states)
677 static bool migration_is_setup_or_active(int state)
679 switch (state) {
680 case MIGRATION_STATUS_ACTIVE:
681 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
682 case MIGRATION_STATUS_POSTCOPY_PAUSED:
683 case MIGRATION_STATUS_POSTCOPY_RECOVER:
684 case MIGRATION_STATUS_SETUP:
685 case MIGRATION_STATUS_PRE_SWITCHOVER:
686 case MIGRATION_STATUS_DEVICE:
687 return true;
689 default:
690 return false;
695 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
697 info->has_ram = true;
698 info->ram = g_malloc0(sizeof(*info->ram));
699 info->ram->transferred = ram_counters.transferred;
700 info->ram->total = ram_bytes_total();
701 info->ram->duplicate = ram_counters.duplicate;
702 /* legacy value. It is not used anymore */
703 info->ram->skipped = 0;
704 info->ram->normal = ram_counters.normal;
705 info->ram->normal_bytes = ram_counters.normal *
706 qemu_target_page_size();
707 info->ram->mbps = s->mbps;
708 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
709 info->ram->postcopy_requests = ram_counters.postcopy_requests;
710 info->ram->page_size = qemu_target_page_size();
712 if (migrate_use_xbzrle()) {
713 info->has_xbzrle_cache = true;
714 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
715 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
716 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
717 info->xbzrle_cache->pages = xbzrle_counters.pages;
718 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
719 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
720 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
723 if (cpu_throttle_active()) {
724 info->has_cpu_throttle_percentage = true;
725 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
728 if (s->state != MIGRATION_STATUS_COMPLETED) {
729 info->ram->remaining = ram_bytes_remaining();
730 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
734 static void populate_disk_info(MigrationInfo *info)
736 if (blk_mig_active()) {
737 info->has_disk = true;
738 info->disk = g_malloc0(sizeof(*info->disk));
739 info->disk->transferred = blk_mig_bytes_transferred();
740 info->disk->remaining = blk_mig_bytes_remaining();
741 info->disk->total = blk_mig_bytes_total();
745 static void fill_source_migration_info(MigrationInfo *info)
747 MigrationState *s = migrate_get_current();
749 switch (s->state) {
750 case MIGRATION_STATUS_NONE:
751 /* no migration has happened ever */
752 /* do not overwrite destination migration status */
753 return;
754 break;
755 case MIGRATION_STATUS_SETUP:
756 info->has_status = true;
757 info->has_total_time = false;
758 break;
759 case MIGRATION_STATUS_ACTIVE:
760 case MIGRATION_STATUS_CANCELLING:
761 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
762 case MIGRATION_STATUS_PRE_SWITCHOVER:
763 case MIGRATION_STATUS_DEVICE:
764 case MIGRATION_STATUS_POSTCOPY_PAUSED:
765 case MIGRATION_STATUS_POSTCOPY_RECOVER:
766 /* TODO add some postcopy stats */
767 info->has_status = true;
768 info->has_total_time = true;
769 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
770 - s->start_time;
771 info->has_expected_downtime = true;
772 info->expected_downtime = s->expected_downtime;
773 info->has_setup_time = true;
774 info->setup_time = s->setup_time;
776 populate_ram_info(info, s);
777 populate_disk_info(info);
778 break;
779 case MIGRATION_STATUS_COLO:
780 info->has_status = true;
781 /* TODO: display COLO specific information (checkpoint info etc.) */
782 break;
783 case MIGRATION_STATUS_COMPLETED:
784 info->has_status = true;
785 info->has_total_time = true;
786 info->total_time = s->total_time;
787 info->has_downtime = true;
788 info->downtime = s->downtime;
789 info->has_setup_time = true;
790 info->setup_time = s->setup_time;
792 populate_ram_info(info, s);
793 break;
794 case MIGRATION_STATUS_FAILED:
795 info->has_status = true;
796 if (s->error) {
797 info->has_error_desc = true;
798 info->error_desc = g_strdup(error_get_pretty(s->error));
800 break;
801 case MIGRATION_STATUS_CANCELLED:
802 info->has_status = true;
803 break;
805 info->status = s->state;
809 * @migration_caps_check - check capability validity
811 * @cap_list: old capability list, array of bool
812 * @params: new capabilities to be applied soon
813 * @errp: set *errp if the check failed, with reason
815 * Returns true if check passed, otherwise false.
817 static bool migrate_caps_check(bool *cap_list,
818 MigrationCapabilityStatusList *params,
819 Error **errp)
821 MigrationCapabilityStatusList *cap;
822 bool old_postcopy_cap;
823 MigrationIncomingState *mis = migration_incoming_get_current();
825 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
827 for (cap = params; cap; cap = cap->next) {
828 cap_list[cap->value->capability] = cap->value->state;
831 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
832 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
833 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
834 "block migration");
835 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
836 return false;
838 #endif
840 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
841 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
842 /* The decompression threads asynchronously write into RAM
843 * rather than use the atomic copies needed to avoid
844 * userfaulting. It should be possible to fix the decompression
845 * threads for compatibility in future.
847 error_setg(errp, "Postcopy is not currently compatible "
848 "with compression");
849 return false;
852 /* This check is reasonably expensive, so only when it's being
853 * set the first time, also it's only the destination that needs
854 * special support.
856 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
857 !postcopy_ram_supported_by_host(mis)) {
858 /* postcopy_ram_supported_by_host will have emitted a more
859 * detailed message
861 error_setg(errp, "Postcopy is not supported");
862 return false;
866 return true;
869 static void fill_destination_migration_info(MigrationInfo *info)
871 MigrationIncomingState *mis = migration_incoming_get_current();
873 switch (mis->state) {
874 case MIGRATION_STATUS_NONE:
875 return;
876 break;
877 case MIGRATION_STATUS_SETUP:
878 case MIGRATION_STATUS_CANCELLING:
879 case MIGRATION_STATUS_CANCELLED:
880 case MIGRATION_STATUS_ACTIVE:
881 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
882 case MIGRATION_STATUS_FAILED:
883 case MIGRATION_STATUS_COLO:
884 info->has_status = true;
885 break;
886 case MIGRATION_STATUS_COMPLETED:
887 info->has_status = true;
888 fill_destination_postcopy_migration_info(info);
889 break;
891 info->status = mis->state;
894 MigrationInfo *qmp_query_migrate(Error **errp)
896 MigrationInfo *info = g_malloc0(sizeof(*info));
898 fill_destination_migration_info(info);
899 fill_source_migration_info(info);
901 return info;
904 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
905 Error **errp)
907 MigrationState *s = migrate_get_current();
908 MigrationCapabilityStatusList *cap;
909 bool cap_list[MIGRATION_CAPABILITY__MAX];
911 if (migration_is_setup_or_active(s->state)) {
912 error_setg(errp, QERR_MIGRATION_ACTIVE);
913 return;
916 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
917 if (!migrate_caps_check(cap_list, params, errp)) {
918 return;
921 for (cap = params; cap; cap = cap->next) {
922 s->enabled_capabilities[cap->value->capability] = cap->value->state;
927 * Check whether the parameters are valid. Error will be put into errp
928 * (if provided). Return true if valid, otherwise false.
930 static bool migrate_params_check(MigrationParameters *params, Error **errp)
932 if (params->has_compress_level &&
933 (params->compress_level > 9)) {
934 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
935 "is invalid, it should be in the range of 0 to 9");
936 return false;
939 if (params->has_compress_threads && (params->compress_threads < 1)) {
940 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
941 "compress_threads",
942 "is invalid, it should be in the range of 1 to 255");
943 return false;
946 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
947 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
948 "decompress_threads",
949 "is invalid, it should be in the range of 1 to 255");
950 return false;
953 if (params->has_cpu_throttle_initial &&
954 (params->cpu_throttle_initial < 1 ||
955 params->cpu_throttle_initial > 99)) {
956 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
957 "cpu_throttle_initial",
958 "an integer in the range of 1 to 99");
959 return false;
962 if (params->has_cpu_throttle_increment &&
963 (params->cpu_throttle_increment < 1 ||
964 params->cpu_throttle_increment > 99)) {
965 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
966 "cpu_throttle_increment",
967 "an integer in the range of 1 to 99");
968 return false;
971 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
972 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
973 " range of 0 to %zu bytes/second", SIZE_MAX);
974 return false;
977 if (params->has_downtime_limit &&
978 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
979 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
980 "the range of 0 to %d milliseconds",
981 MAX_MIGRATE_DOWNTIME);
982 return false;
985 /* x_checkpoint_delay is now always positive */
987 if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) {
988 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
989 "multifd_channels",
990 "is invalid, it should be in the range of 1 to 255");
991 return false;
993 if (params->has_x_multifd_page_count &&
994 (params->x_multifd_page_count < 1 ||
995 params->x_multifd_page_count > 10000)) {
996 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
997 "multifd_page_count",
998 "is invalid, it should be in the range of 1 to 10000");
999 return false;
1002 if (params->has_xbzrle_cache_size &&
1003 (params->xbzrle_cache_size < qemu_target_page_size() ||
1004 !is_power_of_2(params->xbzrle_cache_size))) {
1005 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1006 "xbzrle_cache_size",
1007 "is invalid, it should be bigger than target page size"
1008 " and a power of two");
1009 return false;
1012 return true;
1015 static void migrate_params_test_apply(MigrateSetParameters *params,
1016 MigrationParameters *dest)
1018 *dest = migrate_get_current()->parameters;
1020 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1022 if (params->has_compress_level) {
1023 dest->compress_level = params->compress_level;
1026 if (params->has_compress_threads) {
1027 dest->compress_threads = params->compress_threads;
1030 if (params->has_decompress_threads) {
1031 dest->decompress_threads = params->decompress_threads;
1034 if (params->has_cpu_throttle_initial) {
1035 dest->cpu_throttle_initial = params->cpu_throttle_initial;
1038 if (params->has_cpu_throttle_increment) {
1039 dest->cpu_throttle_increment = params->cpu_throttle_increment;
1042 if (params->has_tls_creds) {
1043 assert(params->tls_creds->type == QTYPE_QSTRING);
1044 dest->tls_creds = g_strdup(params->tls_creds->u.s);
1047 if (params->has_tls_hostname) {
1048 assert(params->tls_hostname->type == QTYPE_QSTRING);
1049 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1052 if (params->has_max_bandwidth) {
1053 dest->max_bandwidth = params->max_bandwidth;
1056 if (params->has_downtime_limit) {
1057 dest->downtime_limit = params->downtime_limit;
1060 if (params->has_x_checkpoint_delay) {
1061 dest->x_checkpoint_delay = params->x_checkpoint_delay;
1064 if (params->has_block_incremental) {
1065 dest->block_incremental = params->block_incremental;
1067 if (params->has_x_multifd_channels) {
1068 dest->x_multifd_channels = params->x_multifd_channels;
1070 if (params->has_x_multifd_page_count) {
1071 dest->x_multifd_page_count = params->x_multifd_page_count;
1073 if (params->has_xbzrle_cache_size) {
1074 dest->xbzrle_cache_size = params->xbzrle_cache_size;
1076 if (params->has_max_postcopy_bandwidth) {
1077 dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1081 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1083 MigrationState *s = migrate_get_current();
1085 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1087 if (params->has_compress_level) {
1088 s->parameters.compress_level = params->compress_level;
1091 if (params->has_compress_threads) {
1092 s->parameters.compress_threads = params->compress_threads;
1095 if (params->has_decompress_threads) {
1096 s->parameters.decompress_threads = params->decompress_threads;
1099 if (params->has_cpu_throttle_initial) {
1100 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1103 if (params->has_cpu_throttle_increment) {
1104 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1107 if (params->has_tls_creds) {
1108 g_free(s->parameters.tls_creds);
1109 assert(params->tls_creds->type == QTYPE_QSTRING);
1110 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1113 if (params->has_tls_hostname) {
1114 g_free(s->parameters.tls_hostname);
1115 assert(params->tls_hostname->type == QTYPE_QSTRING);
1116 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1119 if (params->has_max_bandwidth) {
1120 s->parameters.max_bandwidth = params->max_bandwidth;
1121 if (s->to_dst_file) {
1122 qemu_file_set_rate_limit(s->to_dst_file,
1123 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1127 if (params->has_downtime_limit) {
1128 s->parameters.downtime_limit = params->downtime_limit;
1131 if (params->has_x_checkpoint_delay) {
1132 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1133 if (migration_in_colo_state()) {
1134 colo_checkpoint_notify(s);
1138 if (params->has_block_incremental) {
1139 s->parameters.block_incremental = params->block_incremental;
1141 if (params->has_x_multifd_channels) {
1142 s->parameters.x_multifd_channels = params->x_multifd_channels;
1144 if (params->has_x_multifd_page_count) {
1145 s->parameters.x_multifd_page_count = params->x_multifd_page_count;
1147 if (params->has_xbzrle_cache_size) {
1148 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1149 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1151 if (params->has_max_postcopy_bandwidth) {
1152 s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1156 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1158 MigrationParameters tmp;
1160 /* TODO Rewrite "" to null instead */
1161 if (params->has_tls_creds
1162 && params->tls_creds->type == QTYPE_QNULL) {
1163 qobject_unref(params->tls_creds->u.n);
1164 params->tls_creds->type = QTYPE_QSTRING;
1165 params->tls_creds->u.s = strdup("");
1167 /* TODO Rewrite "" to null instead */
1168 if (params->has_tls_hostname
1169 && params->tls_hostname->type == QTYPE_QNULL) {
1170 qobject_unref(params->tls_hostname->u.n);
1171 params->tls_hostname->type = QTYPE_QSTRING;
1172 params->tls_hostname->u.s = strdup("");
1175 migrate_params_test_apply(params, &tmp);
1177 if (!migrate_params_check(&tmp, errp)) {
1178 /* Invalid parameter */
1179 return;
1182 migrate_params_apply(params, errp);
1186 void qmp_migrate_start_postcopy(Error **errp)
1188 MigrationState *s = migrate_get_current();
1190 if (!migrate_postcopy()) {
1191 error_setg(errp, "Enable postcopy with migrate_set_capability before"
1192 " the start of migration");
1193 return;
1196 if (s->state == MIGRATION_STATUS_NONE) {
1197 error_setg(errp, "Postcopy must be started after migration has been"
1198 " started");
1199 return;
1202 * we don't error if migration has finished since that would be racy
1203 * with issuing this command.
1205 atomic_set(&s->start_postcopy, true);
1208 /* shared migration helpers */
1210 void migrate_set_state(int *state, int old_state, int new_state)
1212 assert(new_state < MIGRATION_STATUS__MAX);
1213 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1214 trace_migrate_set_state(MigrationStatus_str(new_state));
1215 migrate_generate_event(new_state);
1219 static MigrationCapabilityStatusList *migrate_cap_add(
1220 MigrationCapabilityStatusList *list,
1221 MigrationCapability index,
1222 bool state)
1224 MigrationCapabilityStatusList *cap;
1226 cap = g_new0(MigrationCapabilityStatusList, 1);
1227 cap->value = g_new0(MigrationCapabilityStatus, 1);
1228 cap->value->capability = index;
1229 cap->value->state = state;
1230 cap->next = list;
1232 return cap;
1235 void migrate_set_block_enabled(bool value, Error **errp)
1237 MigrationCapabilityStatusList *cap;
1239 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1240 qmp_migrate_set_capabilities(cap, errp);
1241 qapi_free_MigrationCapabilityStatusList(cap);
1244 static void migrate_set_block_incremental(MigrationState *s, bool value)
1246 s->parameters.block_incremental = value;
1249 static void block_cleanup_parameters(MigrationState *s)
1251 if (s->must_remove_block_options) {
1252 /* setting to false can never fail */
1253 migrate_set_block_enabled(false, &error_abort);
1254 migrate_set_block_incremental(s, false);
1255 s->must_remove_block_options = false;
1259 static void migrate_fd_cleanup(void *opaque)
1261 MigrationState *s = opaque;
1263 qemu_bh_delete(s->cleanup_bh);
1264 s->cleanup_bh = NULL;
1266 qemu_savevm_state_cleanup();
1268 if (s->to_dst_file) {
1269 Error *local_err = NULL;
1270 QEMUFile *tmp;
1272 trace_migrate_fd_cleanup();
1273 qemu_mutex_unlock_iothread();
1274 if (s->migration_thread_running) {
1275 qemu_thread_join(&s->thread);
1276 s->migration_thread_running = false;
1278 qemu_mutex_lock_iothread();
1280 if (multifd_save_cleanup(&local_err) != 0) {
1281 error_report_err(local_err);
1283 qemu_mutex_lock(&s->qemu_file_lock);
1284 tmp = s->to_dst_file;
1285 s->to_dst_file = NULL;
1286 qemu_mutex_unlock(&s->qemu_file_lock);
1288 * Close the file handle without the lock to make sure the
1289 * critical section won't block for long.
1291 qemu_fclose(tmp);
1294 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1295 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
1297 if (s->state == MIGRATION_STATUS_CANCELLING) {
1298 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1299 MIGRATION_STATUS_CANCELLED);
1302 if (s->error) {
1303 /* It is used on info migrate. We can't free it */
1304 error_report_err(error_copy(s->error));
1306 notifier_list_notify(&migration_state_notifiers, s);
1307 block_cleanup_parameters(s);
1310 void migrate_set_error(MigrationState *s, const Error *error)
1312 qemu_mutex_lock(&s->error_mutex);
1313 if (!s->error) {
1314 s->error = error_copy(error);
1316 qemu_mutex_unlock(&s->error_mutex);
1319 void migrate_fd_error(MigrationState *s, const Error *error)
1321 trace_migrate_fd_error(error_get_pretty(error));
1322 assert(s->to_dst_file == NULL);
1323 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1324 MIGRATION_STATUS_FAILED);
1325 migrate_set_error(s, error);
1328 static void migrate_fd_cancel(MigrationState *s)
1330 int old_state ;
1331 QEMUFile *f = migrate_get_current()->to_dst_file;
1332 trace_migrate_fd_cancel();
1334 if (s->rp_state.from_dst_file) {
1335 /* shutdown the rp socket, so causing the rp thread to shutdown */
1336 qemu_file_shutdown(s->rp_state.from_dst_file);
1339 do {
1340 old_state = s->state;
1341 if (!migration_is_setup_or_active(old_state)) {
1342 break;
1344 /* If the migration is paused, kick it out of the pause */
1345 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1346 qemu_sem_post(&s->pause_sem);
1348 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1349 } while (s->state != MIGRATION_STATUS_CANCELLING);
1352 * If we're unlucky the migration code might be stuck somewhere in a
1353 * send/write while the network has failed and is waiting to timeout;
1354 * if we've got shutdown(2) available then we can force it to quit.
1355 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1356 * called in a bh, so there is no race against this cancel.
1358 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1359 qemu_file_shutdown(f);
1361 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1362 Error *local_err = NULL;
1364 bdrv_invalidate_cache_all(&local_err);
1365 if (local_err) {
1366 error_report_err(local_err);
1367 } else {
1368 s->block_inactive = false;
1373 void add_migration_state_change_notifier(Notifier *notify)
1375 notifier_list_add(&migration_state_notifiers, notify);
1378 void remove_migration_state_change_notifier(Notifier *notify)
1380 notifier_remove(notify);
1383 bool migration_in_setup(MigrationState *s)
1385 return s->state == MIGRATION_STATUS_SETUP;
1388 bool migration_has_finished(MigrationState *s)
1390 return s->state == MIGRATION_STATUS_COMPLETED;
1393 bool migration_has_failed(MigrationState *s)
1395 return (s->state == MIGRATION_STATUS_CANCELLED ||
1396 s->state == MIGRATION_STATUS_FAILED);
1399 bool migration_in_postcopy(void)
1401 MigrationState *s = migrate_get_current();
1403 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1406 bool migration_in_postcopy_after_devices(MigrationState *s)
1408 return migration_in_postcopy() && s->postcopy_after_devices;
1411 bool migration_is_idle(void)
1413 MigrationState *s = migrate_get_current();
1415 switch (s->state) {
1416 case MIGRATION_STATUS_NONE:
1417 case MIGRATION_STATUS_CANCELLED:
1418 case MIGRATION_STATUS_COMPLETED:
1419 case MIGRATION_STATUS_FAILED:
1420 return true;
1421 case MIGRATION_STATUS_SETUP:
1422 case MIGRATION_STATUS_CANCELLING:
1423 case MIGRATION_STATUS_ACTIVE:
1424 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1425 case MIGRATION_STATUS_COLO:
1426 case MIGRATION_STATUS_PRE_SWITCHOVER:
1427 case MIGRATION_STATUS_DEVICE:
1428 return false;
1429 case MIGRATION_STATUS__MAX:
1430 g_assert_not_reached();
1433 return false;
1436 void migrate_init(MigrationState *s)
1439 * Reinitialise all migration state, except
1440 * parameters/capabilities that the user set, and
1441 * locks.
1443 s->bytes_xfer = 0;
1444 s->xfer_limit = 0;
1445 s->cleanup_bh = 0;
1446 s->to_dst_file = NULL;
1447 s->state = MIGRATION_STATUS_NONE;
1448 s->rp_state.from_dst_file = NULL;
1449 s->rp_state.error = false;
1450 s->mbps = 0.0;
1451 s->downtime = 0;
1452 s->expected_downtime = 0;
1453 s->setup_time = 0;
1454 s->start_postcopy = false;
1455 s->postcopy_after_devices = false;
1456 s->migration_thread_running = false;
1457 error_free(s->error);
1458 s->error = NULL;
1460 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1462 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1463 s->total_time = 0;
1464 s->vm_was_running = false;
1465 s->iteration_initial_bytes = 0;
1466 s->threshold_size = 0;
1469 static GSList *migration_blockers;
1471 int migrate_add_blocker(Error *reason, Error **errp)
1473 if (migrate_get_current()->only_migratable) {
1474 error_propagate(errp, error_copy(reason));
1475 error_prepend(errp, "disallowing migration blocker "
1476 "(--only_migratable) for: ");
1477 return -EACCES;
1480 if (migration_is_idle()) {
1481 migration_blockers = g_slist_prepend(migration_blockers, reason);
1482 return 0;
1485 error_propagate(errp, error_copy(reason));
1486 error_prepend(errp, "disallowing migration blocker (migration in "
1487 "progress) for: ");
1488 return -EBUSY;
1491 void migrate_del_blocker(Error *reason)
1493 migration_blockers = g_slist_remove(migration_blockers, reason);
1496 void qmp_migrate_incoming(const char *uri, Error **errp)
1498 Error *local_err = NULL;
1499 static bool once = true;
1501 if (!deferred_incoming) {
1502 error_setg(errp, "For use with '-incoming defer'");
1503 return;
1505 if (!once) {
1506 error_setg(errp, "The incoming migration has already been started");
1509 qemu_start_incoming_migration(uri, &local_err);
1511 if (local_err) {
1512 error_propagate(errp, local_err);
1513 return;
1516 once = false;
1519 void qmp_migrate_recover(const char *uri, Error **errp)
1521 MigrationIncomingState *mis = migration_incoming_get_current();
1523 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1524 error_setg(errp, "Migrate recover can only be run "
1525 "when postcopy is paused.");
1526 return;
1529 if (atomic_cmpxchg(&mis->postcopy_recover_triggered,
1530 false, true) == true) {
1531 error_setg(errp, "Migrate recovery is triggered already");
1532 return;
1536 * Note that this call will never start a real migration; it will
1537 * only re-setup the migration stream and poke existing migration
1538 * to continue using that newly established channel.
1540 qemu_start_incoming_migration(uri, errp);
1543 void qmp_migrate_pause(Error **errp)
1545 MigrationState *ms = migrate_get_current();
1546 MigrationIncomingState *mis = migration_incoming_get_current();
1547 int ret;
1549 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1550 /* Source side, during postcopy */
1551 qemu_mutex_lock(&ms->qemu_file_lock);
1552 ret = qemu_file_shutdown(ms->to_dst_file);
1553 qemu_mutex_unlock(&ms->qemu_file_lock);
1554 if (ret) {
1555 error_setg(errp, "Failed to pause source migration");
1557 return;
1560 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1561 ret = qemu_file_shutdown(mis->from_src_file);
1562 if (ret) {
1563 error_setg(errp, "Failed to pause destination migration");
1565 return;
1568 error_setg(errp, "migrate-pause is currently only supported "
1569 "during postcopy-active state");
1572 bool migration_is_blocked(Error **errp)
1574 if (qemu_savevm_state_blocked(errp)) {
1575 return true;
1578 if (migration_blockers) {
1579 error_propagate(errp, error_copy(migration_blockers->data));
1580 return true;
1583 return false;
1586 /* Returns true if continue to migrate, or false if error detected */
1587 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
1588 bool resume, Error **errp)
1590 Error *local_err = NULL;
1592 if (resume) {
1593 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1594 error_setg(errp, "Cannot resume if there is no "
1595 "paused migration");
1596 return false;
1598 /* This is a resume, skip init status */
1599 return true;
1602 if (migration_is_setup_or_active(s->state) ||
1603 s->state == MIGRATION_STATUS_CANCELLING ||
1604 s->state == MIGRATION_STATUS_COLO) {
1605 error_setg(errp, QERR_MIGRATION_ACTIVE);
1606 return false;
1609 if (runstate_check(RUN_STATE_INMIGRATE)) {
1610 error_setg(errp, "Guest is waiting for an incoming migration");
1611 return false;
1614 if (migration_is_blocked(errp)) {
1615 return false;
1618 if (blk || blk_inc) {
1619 if (migrate_use_block() || migrate_use_block_incremental()) {
1620 error_setg(errp, "Command options are incompatible with "
1621 "current migration capabilities");
1622 return false;
1624 migrate_set_block_enabled(true, &local_err);
1625 if (local_err) {
1626 error_propagate(errp, local_err);
1627 return false;
1629 s->must_remove_block_options = true;
1632 if (blk_inc) {
1633 migrate_set_block_incremental(s, true);
1636 migrate_init(s);
1638 return true;
1641 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1642 bool has_inc, bool inc, bool has_detach, bool detach,
1643 bool has_resume, bool resume, Error **errp)
1645 Error *local_err = NULL;
1646 MigrationState *s = migrate_get_current();
1647 const char *p;
1649 if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
1650 has_resume && resume, errp)) {
1651 /* Error detected, put into errp */
1652 return;
1655 if (strstart(uri, "tcp:", &p)) {
1656 tcp_start_outgoing_migration(s, p, &local_err);
1657 #ifdef CONFIG_RDMA
1658 } else if (strstart(uri, "rdma:", &p)) {
1659 rdma_start_outgoing_migration(s, p, &local_err);
1660 #endif
1661 } else if (strstart(uri, "exec:", &p)) {
1662 exec_start_outgoing_migration(s, p, &local_err);
1663 } else if (strstart(uri, "unix:", &p)) {
1664 unix_start_outgoing_migration(s, p, &local_err);
1665 } else if (strstart(uri, "fd:", &p)) {
1666 fd_start_outgoing_migration(s, p, &local_err);
1667 } else {
1668 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1669 "a valid migration protocol");
1670 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1671 MIGRATION_STATUS_FAILED);
1672 block_cleanup_parameters(s);
1673 return;
1676 if (local_err) {
1677 migrate_fd_error(s, local_err);
1678 error_propagate(errp, local_err);
1679 return;
1683 void qmp_migrate_cancel(Error **errp)
1685 migrate_fd_cancel(migrate_get_current());
1688 void qmp_migrate_continue(MigrationStatus state, Error **errp)
1690 MigrationState *s = migrate_get_current();
1691 if (s->state != state) {
1692 error_setg(errp, "Migration not in expected state: %s",
1693 MigrationStatus_str(s->state));
1694 return;
1696 qemu_sem_post(&s->pause_sem);
1699 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1701 MigrateSetParameters p = {
1702 .has_xbzrle_cache_size = true,
1703 .xbzrle_cache_size = value,
1706 qmp_migrate_set_parameters(&p, errp);
1709 int64_t qmp_query_migrate_cache_size(Error **errp)
1711 return migrate_xbzrle_cache_size();
1714 void qmp_migrate_set_speed(int64_t value, Error **errp)
1716 MigrateSetParameters p = {
1717 .has_max_bandwidth = true,
1718 .max_bandwidth = value,
1721 qmp_migrate_set_parameters(&p, errp);
1724 void qmp_migrate_set_downtime(double value, Error **errp)
1726 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1727 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1728 "the range of 0 to %d seconds",
1729 MAX_MIGRATE_DOWNTIME_SECONDS);
1730 return;
1733 value *= 1000; /* Convert to milliseconds */
1734 value = MAX(0, MIN(INT64_MAX, value));
1736 MigrateSetParameters p = {
1737 .has_downtime_limit = true,
1738 .downtime_limit = value,
1741 qmp_migrate_set_parameters(&p, errp);
1744 bool migrate_release_ram(void)
1746 MigrationState *s;
1748 s = migrate_get_current();
1750 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1753 bool migrate_postcopy_ram(void)
1755 MigrationState *s;
1757 s = migrate_get_current();
1759 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1762 bool migrate_postcopy(void)
1764 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1767 bool migrate_auto_converge(void)
1769 MigrationState *s;
1771 s = migrate_get_current();
1773 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1776 bool migrate_zero_blocks(void)
1778 MigrationState *s;
1780 s = migrate_get_current();
1782 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1785 bool migrate_postcopy_blocktime(void)
1787 MigrationState *s;
1789 s = migrate_get_current();
1791 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
1794 bool migrate_use_compression(void)
1796 MigrationState *s;
1798 s = migrate_get_current();
1800 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1803 int migrate_compress_level(void)
1805 MigrationState *s;
1807 s = migrate_get_current();
1809 return s->parameters.compress_level;
1812 int migrate_compress_threads(void)
1814 MigrationState *s;
1816 s = migrate_get_current();
1818 return s->parameters.compress_threads;
1821 int migrate_decompress_threads(void)
1823 MigrationState *s;
1825 s = migrate_get_current();
1827 return s->parameters.decompress_threads;
1830 bool migrate_dirty_bitmaps(void)
1832 MigrationState *s;
1834 s = migrate_get_current();
1836 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
1839 bool migrate_use_events(void)
1841 MigrationState *s;
1843 s = migrate_get_current();
1845 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1848 bool migrate_use_multifd(void)
1850 MigrationState *s;
1852 s = migrate_get_current();
1854 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1857 bool migrate_pause_before_switchover(void)
1859 MigrationState *s;
1861 s = migrate_get_current();
1863 return s->enabled_capabilities[
1864 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
1867 int migrate_multifd_channels(void)
1869 MigrationState *s;
1871 s = migrate_get_current();
1873 return s->parameters.x_multifd_channels;
1876 int migrate_multifd_page_count(void)
1878 MigrationState *s;
1880 s = migrate_get_current();
1882 return s->parameters.x_multifd_page_count;
1885 int migrate_use_xbzrle(void)
1887 MigrationState *s;
1889 s = migrate_get_current();
1891 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1894 int64_t migrate_xbzrle_cache_size(void)
1896 MigrationState *s;
1898 s = migrate_get_current();
1900 return s->parameters.xbzrle_cache_size;
1903 static int64_t migrate_max_postcopy_bandwidth(void)
1905 MigrationState *s;
1907 s = migrate_get_current();
1909 return s->parameters.max_postcopy_bandwidth;
1913 bool migrate_use_block(void)
1915 MigrationState *s;
1917 s = migrate_get_current();
1919 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1922 bool migrate_use_return_path(void)
1924 MigrationState *s;
1926 s = migrate_get_current();
1928 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1931 bool migrate_use_block_incremental(void)
1933 MigrationState *s;
1935 s = migrate_get_current();
1937 return s->parameters.block_incremental;
1940 /* migration thread support */
1942 * Something bad happened to the RP stream, mark an error
1943 * The caller shall print or trace something to indicate why
1945 static void mark_source_rp_bad(MigrationState *s)
1947 s->rp_state.error = true;
1950 static struct rp_cmd_args {
1951 ssize_t len; /* -1 = variable */
1952 const char *name;
1953 } rp_cmd_args[] = {
1954 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1955 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1956 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1957 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1958 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
1959 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
1960 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
1961 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1965 * Process a request for pages received on the return path,
1966 * We're allowed to send more than requested (e.g. to round to our page size)
1967 * and we don't need to send pages that have already been sent.
1969 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1970 ram_addr_t start, size_t len)
1972 long our_host_ps = getpagesize();
1974 trace_migrate_handle_rp_req_pages(rbname, start, len);
1977 * Since we currently insist on matching page sizes, just sanity check
1978 * we're being asked for whole host pages.
1980 if (start & (our_host_ps-1) ||
1981 (len & (our_host_ps-1))) {
1982 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1983 " len: %zd", __func__, start, len);
1984 mark_source_rp_bad(ms);
1985 return;
1988 if (ram_save_queue_pages(rbname, start, len)) {
1989 mark_source_rp_bad(ms);
1993 /* Return true to retry, false to quit */
1994 static bool postcopy_pause_return_path_thread(MigrationState *s)
1996 trace_postcopy_pause_return_path();
1998 qemu_sem_wait(&s->postcopy_pause_rp_sem);
2000 trace_postcopy_pause_return_path_continued();
2002 return true;
2005 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2007 RAMBlock *block = qemu_ram_block_by_name(block_name);
2009 if (!block) {
2010 error_report("%s: invalid block name '%s'", __func__, block_name);
2011 return -EINVAL;
2014 /* Fetch the received bitmap and refresh the dirty bitmap */
2015 return ram_dirty_bitmap_reload(s, block);
2018 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2020 trace_source_return_path_thread_resume_ack(value);
2022 if (value != MIGRATION_RESUME_ACK_VALUE) {
2023 error_report("%s: illegal resume_ack value %"PRIu32,
2024 __func__, value);
2025 return -1;
2028 /* Now both sides are active. */
2029 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2030 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2032 /* Notify send thread that time to continue send pages */
2033 qemu_sem_post(&s->rp_state.rp_sem);
2035 return 0;
2039 * Handles messages sent on the return path towards the source VM
2042 static void *source_return_path_thread(void *opaque)
2044 MigrationState *ms = opaque;
2045 QEMUFile *rp = ms->rp_state.from_dst_file;
2046 uint16_t header_len, header_type;
2047 uint8_t buf[512];
2048 uint32_t tmp32, sibling_error;
2049 ram_addr_t start = 0; /* =0 to silence warning */
2050 size_t len = 0, expected_len;
2051 int res;
2053 trace_source_return_path_thread_entry();
2055 retry:
2056 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2057 migration_is_setup_or_active(ms->state)) {
2058 trace_source_return_path_thread_loop_top();
2059 header_type = qemu_get_be16(rp);
2060 header_len = qemu_get_be16(rp);
2062 if (qemu_file_get_error(rp)) {
2063 mark_source_rp_bad(ms);
2064 goto out;
2067 if (header_type >= MIG_RP_MSG_MAX ||
2068 header_type == MIG_RP_MSG_INVALID) {
2069 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2070 header_type, header_len);
2071 mark_source_rp_bad(ms);
2072 goto out;
2075 if ((rp_cmd_args[header_type].len != -1 &&
2076 header_len != rp_cmd_args[header_type].len) ||
2077 header_len > sizeof(buf)) {
2078 error_report("RP: Received '%s' message (0x%04x) with"
2079 "incorrect length %d expecting %zu",
2080 rp_cmd_args[header_type].name, header_type, header_len,
2081 (size_t)rp_cmd_args[header_type].len);
2082 mark_source_rp_bad(ms);
2083 goto out;
2086 /* We know we've got a valid header by this point */
2087 res = qemu_get_buffer(rp, buf, header_len);
2088 if (res != header_len) {
2089 error_report("RP: Failed reading data for message 0x%04x"
2090 " read %d expected %d",
2091 header_type, res, header_len);
2092 mark_source_rp_bad(ms);
2093 goto out;
2096 /* OK, we have the message and the data */
2097 switch (header_type) {
2098 case MIG_RP_MSG_SHUT:
2099 sibling_error = ldl_be_p(buf);
2100 trace_source_return_path_thread_shut(sibling_error);
2101 if (sibling_error) {
2102 error_report("RP: Sibling indicated error %d", sibling_error);
2103 mark_source_rp_bad(ms);
2106 * We'll let the main thread deal with closing the RP
2107 * we could do a shutdown(2) on it, but we're the only user
2108 * anyway, so there's nothing gained.
2110 goto out;
2112 case MIG_RP_MSG_PONG:
2113 tmp32 = ldl_be_p(buf);
2114 trace_source_return_path_thread_pong(tmp32);
2115 break;
2117 case MIG_RP_MSG_REQ_PAGES:
2118 start = ldq_be_p(buf);
2119 len = ldl_be_p(buf + 8);
2120 migrate_handle_rp_req_pages(ms, NULL, start, len);
2121 break;
2123 case MIG_RP_MSG_REQ_PAGES_ID:
2124 expected_len = 12 + 1; /* header + termination */
2126 if (header_len >= expected_len) {
2127 start = ldq_be_p(buf);
2128 len = ldl_be_p(buf + 8);
2129 /* Now we expect an idstr */
2130 tmp32 = buf[12]; /* Length of the following idstr */
2131 buf[13 + tmp32] = '\0';
2132 expected_len += tmp32;
2134 if (header_len != expected_len) {
2135 error_report("RP: Req_Page_id with length %d expecting %zd",
2136 header_len, expected_len);
2137 mark_source_rp_bad(ms);
2138 goto out;
2140 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2141 break;
2143 case MIG_RP_MSG_RECV_BITMAP:
2144 if (header_len < 1) {
2145 error_report("%s: missing block name", __func__);
2146 mark_source_rp_bad(ms);
2147 goto out;
2149 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2150 buf[buf[0] + 1] = '\0';
2151 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2152 mark_source_rp_bad(ms);
2153 goto out;
2155 break;
2157 case MIG_RP_MSG_RESUME_ACK:
2158 tmp32 = ldl_be_p(buf);
2159 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2160 mark_source_rp_bad(ms);
2161 goto out;
2163 break;
2165 default:
2166 break;
2170 out:
2171 res = qemu_file_get_error(rp);
2172 if (res) {
2173 if (res == -EIO) {
2175 * Maybe there is something we can do: it looks like a
2176 * network down issue, and we pause for a recovery.
2178 if (postcopy_pause_return_path_thread(ms)) {
2179 /* Reload rp, reset the rest */
2180 rp = ms->rp_state.from_dst_file;
2181 ms->rp_state.error = false;
2182 goto retry;
2186 trace_source_return_path_thread_bad_end();
2187 mark_source_rp_bad(ms);
2190 trace_source_return_path_thread_end();
2191 ms->rp_state.from_dst_file = NULL;
2192 qemu_fclose(rp);
2193 return NULL;
2196 static int open_return_path_on_source(MigrationState *ms,
2197 bool create_thread)
2200 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2201 if (!ms->rp_state.from_dst_file) {
2202 return -1;
2205 trace_open_return_path_on_source();
2207 if (!create_thread) {
2208 /* We're done */
2209 return 0;
2212 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2213 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2215 trace_open_return_path_on_source_continue();
2217 return 0;
2220 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2221 static int await_return_path_close_on_source(MigrationState *ms)
2224 * If this is a normal exit then the destination will send a SHUT and the
2225 * rp_thread will exit, however if there's an error we need to cause
2226 * it to exit.
2228 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2230 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2231 * waiting for the destination.
2233 qemu_file_shutdown(ms->rp_state.from_dst_file);
2234 mark_source_rp_bad(ms);
2236 trace_await_return_path_close_on_source_joining();
2237 qemu_thread_join(&ms->rp_state.rp_thread);
2238 trace_await_return_path_close_on_source_close();
2239 return ms->rp_state.error;
2243 * Switch from normal iteration to postcopy
2244 * Returns non-0 on error
2246 static int postcopy_start(MigrationState *ms)
2248 int ret;
2249 QIOChannelBuffer *bioc;
2250 QEMUFile *fb;
2251 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2252 int64_t bandwidth = migrate_max_postcopy_bandwidth();
2253 bool restart_block = false;
2254 int cur_state = MIGRATION_STATUS_ACTIVE;
2255 if (!migrate_pause_before_switchover()) {
2256 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2257 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2260 trace_postcopy_start();
2261 qemu_mutex_lock_iothread();
2262 trace_postcopy_start_set_run();
2264 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
2265 global_state_store();
2266 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2267 if (ret < 0) {
2268 goto fail;
2271 ret = migration_maybe_pause(ms, &cur_state,
2272 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2273 if (ret < 0) {
2274 goto fail;
2277 ret = bdrv_inactivate_all();
2278 if (ret < 0) {
2279 goto fail;
2281 restart_block = true;
2284 * Cause any non-postcopiable, but iterative devices to
2285 * send out their final data.
2287 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2290 * in Finish migrate and with the io-lock held everything should
2291 * be quiet, but we've potentially still got dirty pages and we
2292 * need to tell the destination to throw any pages it's already received
2293 * that are dirty
2295 if (migrate_postcopy_ram()) {
2296 if (ram_postcopy_send_discard_bitmap(ms)) {
2297 error_report("postcopy send discard bitmap failed");
2298 goto fail;
2303 * send rest of state - note things that are doing postcopy
2304 * will notice we're in POSTCOPY_ACTIVE and not actually
2305 * wrap their state up here
2307 /* 0 max-postcopy-bandwidth means unlimited */
2308 if (!bandwidth) {
2309 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2310 } else {
2311 qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2313 if (migrate_postcopy_ram()) {
2314 /* Ping just for debugging, helps line traces up */
2315 qemu_savevm_send_ping(ms->to_dst_file, 2);
2319 * While loading the device state we may trigger page transfer
2320 * requests and the fd must be free to process those, and thus
2321 * the destination must read the whole device state off the fd before
2322 * it starts processing it. Unfortunately the ad-hoc migration format
2323 * doesn't allow the destination to know the size to read without fully
2324 * parsing it through each devices load-state code (especially the open
2325 * coded devices that use get/put).
2326 * So we wrap the device state up in a package with a length at the start;
2327 * to do this we use a qemu_buf to hold the whole of the device state.
2329 bioc = qio_channel_buffer_new(4096);
2330 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2331 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2332 object_unref(OBJECT(bioc));
2335 * Make sure the receiver can get incoming pages before we send the rest
2336 * of the state
2338 qemu_savevm_send_postcopy_listen(fb);
2340 qemu_savevm_state_complete_precopy(fb, false, false);
2341 if (migrate_postcopy_ram()) {
2342 qemu_savevm_send_ping(fb, 3);
2345 qemu_savevm_send_postcopy_run(fb);
2347 /* <><> end of stuff going into the package */
2349 /* Last point of recovery; as soon as we send the package the destination
2350 * can open devices and potentially start running.
2351 * Lets just check again we've not got any errors.
2353 ret = qemu_file_get_error(ms->to_dst_file);
2354 if (ret) {
2355 error_report("postcopy_start: Migration stream errored (pre package)");
2356 goto fail_closefb;
2359 restart_block = false;
2361 /* Now send that blob */
2362 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2363 goto fail_closefb;
2365 qemu_fclose(fb);
2367 /* Send a notify to give a chance for anything that needs to happen
2368 * at the transition to postcopy and after the device state; in particular
2369 * spice needs to trigger a transition now
2371 ms->postcopy_after_devices = true;
2372 notifier_list_notify(&migration_state_notifiers, ms);
2374 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2376 qemu_mutex_unlock_iothread();
2378 if (migrate_postcopy_ram()) {
2380 * Although this ping is just for debug, it could potentially be
2381 * used for getting a better measurement of downtime at the source.
2383 qemu_savevm_send_ping(ms->to_dst_file, 4);
2386 if (migrate_release_ram()) {
2387 ram_postcopy_migrated_memory_release(ms);
2390 ret = qemu_file_get_error(ms->to_dst_file);
2391 if (ret) {
2392 error_report("postcopy_start: Migration stream errored");
2393 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2394 MIGRATION_STATUS_FAILED);
2397 return ret;
2399 fail_closefb:
2400 qemu_fclose(fb);
2401 fail:
2402 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2403 MIGRATION_STATUS_FAILED);
2404 if (restart_block) {
2405 /* A failure happened early enough that we know the destination hasn't
2406 * accessed block devices, so we're safe to recover.
2408 Error *local_err = NULL;
2410 bdrv_invalidate_cache_all(&local_err);
2411 if (local_err) {
2412 error_report_err(local_err);
2415 qemu_mutex_unlock_iothread();
2416 return -1;
2420 * migration_maybe_pause: Pause if required to by
2421 * migrate_pause_before_switchover called with the iothread locked
2422 * Returns: 0 on success
2424 static int migration_maybe_pause(MigrationState *s,
2425 int *current_active_state,
2426 int new_state)
2428 if (!migrate_pause_before_switchover()) {
2429 return 0;
2432 /* Since leaving this state is not atomic with posting the semaphore
2433 * it's possible that someone could have issued multiple migrate_continue
2434 * and the semaphore is incorrectly positive at this point;
2435 * the docs say it's undefined to reinit a semaphore that's already
2436 * init'd, so use timedwait to eat up any existing posts.
2438 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2439 /* This block intentionally left blank */
2442 qemu_mutex_unlock_iothread();
2443 migrate_set_state(&s->state, *current_active_state,
2444 MIGRATION_STATUS_PRE_SWITCHOVER);
2445 qemu_sem_wait(&s->pause_sem);
2446 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2447 new_state);
2448 *current_active_state = new_state;
2449 qemu_mutex_lock_iothread();
2451 return s->state == new_state ? 0 : -EINVAL;
2455 * migration_completion: Used by migration_thread when there's not much left.
2456 * The caller 'breaks' the loop when this returns.
2458 * @s: Current migration state
2460 static void migration_completion(MigrationState *s)
2462 int ret;
2463 int current_active_state = s->state;
2465 if (s->state == MIGRATION_STATUS_ACTIVE) {
2466 qemu_mutex_lock_iothread();
2467 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2468 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
2469 s->vm_was_running = runstate_is_running();
2470 ret = global_state_store();
2472 if (!ret) {
2473 bool inactivate = !migrate_colo_enabled();
2474 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2475 if (ret >= 0) {
2476 ret = migration_maybe_pause(s, &current_active_state,
2477 MIGRATION_STATUS_DEVICE);
2479 if (ret >= 0) {
2480 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2481 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2482 inactivate);
2484 if (inactivate && ret >= 0) {
2485 s->block_inactive = true;
2488 qemu_mutex_unlock_iothread();
2490 if (ret < 0) {
2491 goto fail;
2493 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2494 trace_migration_completion_postcopy_end();
2496 qemu_savevm_state_complete_postcopy(s->to_dst_file);
2497 trace_migration_completion_postcopy_end_after_complete();
2501 * If rp was opened we must clean up the thread before
2502 * cleaning everything else up (since if there are no failures
2503 * it will wait for the destination to send it's status in
2504 * a SHUT command).
2506 if (s->rp_state.from_dst_file) {
2507 int rp_error;
2508 trace_migration_return_path_end_before();
2509 rp_error = await_return_path_close_on_source(s);
2510 trace_migration_return_path_end_after(rp_error);
2511 if (rp_error) {
2512 goto fail_invalidate;
2516 if (qemu_file_get_error(s->to_dst_file)) {
2517 trace_migration_completion_file_err();
2518 goto fail_invalidate;
2521 if (!migrate_colo_enabled()) {
2522 migrate_set_state(&s->state, current_active_state,
2523 MIGRATION_STATUS_COMPLETED);
2526 return;
2528 fail_invalidate:
2529 /* If not doing postcopy, vm_start() will be called: let's regain
2530 * control on images.
2532 if (s->state == MIGRATION_STATUS_ACTIVE ||
2533 s->state == MIGRATION_STATUS_DEVICE) {
2534 Error *local_err = NULL;
2536 qemu_mutex_lock_iothread();
2537 bdrv_invalidate_cache_all(&local_err);
2538 if (local_err) {
2539 error_report_err(local_err);
2540 } else {
2541 s->block_inactive = false;
2543 qemu_mutex_unlock_iothread();
2546 fail:
2547 migrate_set_state(&s->state, current_active_state,
2548 MIGRATION_STATUS_FAILED);
2551 bool migrate_colo_enabled(void)
2553 MigrationState *s = migrate_get_current();
2554 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2557 typedef enum MigThrError {
2558 /* No error detected */
2559 MIG_THR_ERR_NONE = 0,
2560 /* Detected error, but resumed successfully */
2561 MIG_THR_ERR_RECOVERED = 1,
2562 /* Detected fatal error, need to exit */
2563 MIG_THR_ERR_FATAL = 2,
2564 } MigThrError;
2566 static int postcopy_resume_handshake(MigrationState *s)
2568 qemu_savevm_send_postcopy_resume(s->to_dst_file);
2570 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2571 qemu_sem_wait(&s->rp_state.rp_sem);
2574 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2575 return 0;
2578 return -1;
2581 /* Return zero if success, or <0 for error */
2582 static int postcopy_do_resume(MigrationState *s)
2584 int ret;
2587 * Call all the resume_prepare() hooks, so that modules can be
2588 * ready for the migration resume.
2590 ret = qemu_savevm_state_resume_prepare(s);
2591 if (ret) {
2592 error_report("%s: resume_prepare() failure detected: %d",
2593 __func__, ret);
2594 return ret;
2598 * Last handshake with destination on the resume (destination will
2599 * switch to postcopy-active afterwards)
2601 ret = postcopy_resume_handshake(s);
2602 if (ret) {
2603 error_report("%s: handshake failed: %d", __func__, ret);
2604 return ret;
2607 return 0;
2611 * We don't return until we are in a safe state to continue current
2612 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2613 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2615 static MigThrError postcopy_pause(MigrationState *s)
2617 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2619 while (true) {
2620 QEMUFile *file;
2622 migrate_set_state(&s->state, s->state,
2623 MIGRATION_STATUS_POSTCOPY_PAUSED);
2625 /* Current channel is possibly broken. Release it. */
2626 assert(s->to_dst_file);
2627 qemu_mutex_lock(&s->qemu_file_lock);
2628 file = s->to_dst_file;
2629 s->to_dst_file = NULL;
2630 qemu_mutex_unlock(&s->qemu_file_lock);
2632 qemu_file_shutdown(file);
2633 qemu_fclose(file);
2635 error_report("Detected IO failure for postcopy. "
2636 "Migration paused.");
2639 * We wait until things fixed up. Then someone will setup the
2640 * status back for us.
2642 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2643 qemu_sem_wait(&s->postcopy_pause_sem);
2646 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2647 /* Woken up by a recover procedure. Give it a shot */
2650 * Firstly, let's wake up the return path now, with a new
2651 * return path channel.
2653 qemu_sem_post(&s->postcopy_pause_rp_sem);
2655 /* Do the resume logic */
2656 if (postcopy_do_resume(s) == 0) {
2657 /* Let's continue! */
2658 trace_postcopy_pause_continued();
2659 return MIG_THR_ERR_RECOVERED;
2660 } else {
2662 * Something wrong happened during the recovery, let's
2663 * pause again. Pause is always better than throwing
2664 * data away.
2666 continue;
2668 } else {
2669 /* This is not right... Time to quit. */
2670 return MIG_THR_ERR_FATAL;
2675 static MigThrError migration_detect_error(MigrationState *s)
2677 int ret;
2679 /* Try to detect any file errors */
2680 ret = qemu_file_get_error(s->to_dst_file);
2682 if (!ret) {
2683 /* Everything is fine */
2684 return MIG_THR_ERR_NONE;
2687 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
2689 * For postcopy, we allow the network to be down for a
2690 * while. After that, it can be continued by a
2691 * recovery phase.
2693 return postcopy_pause(s);
2694 } else {
2696 * For precopy (or postcopy with error outside IO), we fail
2697 * with no time.
2699 migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
2700 trace_migration_thread_file_err();
2702 /* Time to stop the migration, now. */
2703 return MIG_THR_ERR_FATAL;
2707 static void migration_calculate_complete(MigrationState *s)
2709 uint64_t bytes = qemu_ftell(s->to_dst_file);
2710 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2712 s->total_time = end_time - s->start_time;
2713 if (!s->downtime) {
2715 * It's still not set, so we are precopy migration. For
2716 * postcopy, downtime is calculated during postcopy_start().
2718 s->downtime = end_time - s->downtime_start;
2721 if (s->total_time) {
2722 s->mbps = ((double) bytes * 8.0) / s->total_time / 1000;
2726 static void migration_update_counters(MigrationState *s,
2727 int64_t current_time)
2729 uint64_t transferred, time_spent;
2730 double bandwidth;
2732 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
2733 return;
2736 transferred = qemu_ftell(s->to_dst_file) - s->iteration_initial_bytes;
2737 time_spent = current_time - s->iteration_start_time;
2738 bandwidth = (double)transferred / time_spent;
2739 s->threshold_size = bandwidth * s->parameters.downtime_limit;
2741 s->mbps = (((double) transferred * 8.0) /
2742 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2745 * if we haven't sent anything, we don't want to
2746 * recalculate. 10000 is a small enough number for our purposes
2748 if (ram_counters.dirty_pages_rate && transferred > 10000) {
2749 s->expected_downtime = ram_counters.remaining / bandwidth;
2752 qemu_file_reset_rate_limit(s->to_dst_file);
2754 s->iteration_start_time = current_time;
2755 s->iteration_initial_bytes = qemu_ftell(s->to_dst_file);
2757 trace_migrate_transferred(transferred, time_spent,
2758 bandwidth, s->threshold_size);
2761 /* Migration thread iteration status */
2762 typedef enum {
2763 MIG_ITERATE_RESUME, /* Resume current iteration */
2764 MIG_ITERATE_SKIP, /* Skip current iteration */
2765 MIG_ITERATE_BREAK, /* Break the loop */
2766 } MigIterateState;
2769 * Return true if continue to the next iteration directly, false
2770 * otherwise.
2772 static MigIterateState migration_iteration_run(MigrationState *s)
2774 uint64_t pending_size, pend_pre, pend_compat, pend_post;
2775 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
2777 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
2778 &pend_compat, &pend_post);
2779 pending_size = pend_pre + pend_compat + pend_post;
2781 trace_migrate_pending(pending_size, s->threshold_size,
2782 pend_pre, pend_compat, pend_post);
2784 if (pending_size && pending_size >= s->threshold_size) {
2785 /* Still a significant amount to transfer */
2786 if (migrate_postcopy() && !in_postcopy &&
2787 pend_pre <= s->threshold_size &&
2788 atomic_read(&s->start_postcopy)) {
2789 if (postcopy_start(s)) {
2790 error_report("%s: postcopy failed to start", __func__);
2792 return MIG_ITERATE_SKIP;
2794 /* Just another iteration step */
2795 qemu_savevm_state_iterate(s->to_dst_file,
2796 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2797 } else {
2798 trace_migration_thread_low_pending(pending_size);
2799 migration_completion(s);
2800 return MIG_ITERATE_BREAK;
2803 return MIG_ITERATE_RESUME;
2806 static void migration_iteration_finish(MigrationState *s)
2808 /* If we enabled cpu throttling for auto-converge, turn it off. */
2809 cpu_throttle_stop();
2811 qemu_mutex_lock_iothread();
2812 switch (s->state) {
2813 case MIGRATION_STATUS_COMPLETED:
2814 migration_calculate_complete(s);
2815 runstate_set(RUN_STATE_POSTMIGRATE);
2816 break;
2818 case MIGRATION_STATUS_ACTIVE:
2820 * We should really assert here, but since it's during
2821 * migration, let's try to reduce the usage of assertions.
2823 if (!migrate_colo_enabled()) {
2824 error_report("%s: critical error: calling COLO code without "
2825 "COLO enabled", __func__);
2827 migrate_start_colo_process(s);
2829 * Fixme: we will run VM in COLO no matter its old running state.
2830 * After exited COLO, we will keep running.
2832 s->vm_was_running = true;
2833 /* Fallthrough */
2834 case MIGRATION_STATUS_FAILED:
2835 case MIGRATION_STATUS_CANCELLED:
2836 if (s->vm_was_running) {
2837 vm_start();
2838 } else {
2839 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2840 runstate_set(RUN_STATE_POSTMIGRATE);
2843 break;
2845 default:
2846 /* Should not reach here, but if so, forgive the VM. */
2847 error_report("%s: Unknown ending state %d", __func__, s->state);
2848 break;
2850 qemu_bh_schedule(s->cleanup_bh);
2851 qemu_mutex_unlock_iothread();
2854 void migration_make_urgent_request(void)
2856 qemu_sem_post(&migrate_get_current()->rate_limit_sem);
2859 void migration_consume_urgent_request(void)
2861 qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
2865 * Master migration thread on the source VM.
2866 * It drives the migration and pumps the data down the outgoing channel.
2868 static void *migration_thread(void *opaque)
2870 MigrationState *s = opaque;
2871 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
2872 MigThrError thr_error;
2873 bool urgent = false;
2875 rcu_register_thread();
2877 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2879 qemu_savevm_state_header(s->to_dst_file);
2882 * If we opened the return path, we need to make sure dst has it
2883 * opened as well.
2885 if (s->rp_state.from_dst_file) {
2886 /* Now tell the dest that it should open its end so it can reply */
2887 qemu_savevm_send_open_return_path(s->to_dst_file);
2889 /* And do a ping that will make stuff easier to debug */
2890 qemu_savevm_send_ping(s->to_dst_file, 1);
2893 if (migrate_postcopy()) {
2895 * Tell the destination that we *might* want to do postcopy later;
2896 * if the other end can't do postcopy it should fail now, nice and
2897 * early.
2899 qemu_savevm_send_postcopy_advise(s->to_dst_file);
2902 qemu_savevm_state_setup(s->to_dst_file);
2904 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
2905 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2906 MIGRATION_STATUS_ACTIVE);
2908 trace_migration_thread_setup_complete();
2910 while (s->state == MIGRATION_STATUS_ACTIVE ||
2911 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2912 int64_t current_time;
2914 if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
2915 MigIterateState iter_state = migration_iteration_run(s);
2916 if (iter_state == MIG_ITERATE_SKIP) {
2917 continue;
2918 } else if (iter_state == MIG_ITERATE_BREAK) {
2919 break;
2924 * Try to detect any kind of failures, and see whether we
2925 * should stop the migration now.
2927 thr_error = migration_detect_error(s);
2928 if (thr_error == MIG_THR_ERR_FATAL) {
2929 /* Stop migration */
2930 break;
2931 } else if (thr_error == MIG_THR_ERR_RECOVERED) {
2933 * Just recovered from a e.g. network failure, reset all
2934 * the local variables. This is important to avoid
2935 * breaking transferred_bytes and bandwidth calculation
2937 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2938 s->iteration_initial_bytes = 0;
2941 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2943 migration_update_counters(s, current_time);
2945 urgent = false;
2946 if (qemu_file_rate_limit(s->to_dst_file)) {
2947 /* Wait for a delay to do rate limiting OR
2948 * something urgent to post the semaphore.
2950 int ms = s->iteration_start_time + BUFFER_DELAY - current_time;
2951 trace_migration_thread_ratelimit_pre(ms);
2952 if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
2953 /* We were worken by one or more urgent things but
2954 * the timedwait will have consumed one of them.
2955 * The service routine for the urgent wake will dec
2956 * the semaphore itself for each item it consumes,
2957 * so add this one we just eat back.
2959 qemu_sem_post(&s->rate_limit_sem);
2960 urgent = true;
2962 trace_migration_thread_ratelimit_post(urgent);
2966 trace_migration_thread_after_loop();
2967 migration_iteration_finish(s);
2968 rcu_unregister_thread();
2969 return NULL;
2972 void migrate_fd_connect(MigrationState *s, Error *error_in)
2974 int64_t rate_limit;
2975 bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
2977 s->expected_downtime = s->parameters.downtime_limit;
2978 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2979 if (error_in) {
2980 migrate_fd_error(s, error_in);
2981 migrate_fd_cleanup(s);
2982 return;
2985 if (resume) {
2986 /* This is a resumed migration */
2987 rate_limit = INT64_MAX;
2988 } else {
2989 /* This is a fresh new migration */
2990 rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
2991 s->expected_downtime = s->parameters.downtime_limit;
2992 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2994 /* Notify before starting migration thread */
2995 notifier_list_notify(&migration_state_notifiers, s);
2998 qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
2999 qemu_file_set_blocking(s->to_dst_file, true);
3002 * Open the return path. For postcopy, it is used exclusively. For
3003 * precopy, only if user specified "return-path" capability would
3004 * QEMU uses the return path.
3006 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3007 if (open_return_path_on_source(s, !resume)) {
3008 error_report("Unable to open return-path for postcopy");
3009 migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3010 migrate_fd_cleanup(s);
3011 return;
3015 if (resume) {
3016 /* Wakeup the main migration thread to do the recovery */
3017 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3018 MIGRATION_STATUS_POSTCOPY_RECOVER);
3019 qemu_sem_post(&s->postcopy_pause_sem);
3020 return;
3023 if (multifd_save_setup() != 0) {
3024 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3025 MIGRATION_STATUS_FAILED);
3026 migrate_fd_cleanup(s);
3027 return;
3029 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
3030 QEMU_THREAD_JOINABLE);
3031 s->migration_thread_running = true;
3034 void migration_global_dump(Monitor *mon)
3036 MigrationState *ms = migrate_get_current();
3038 monitor_printf(mon, "globals:\n");
3039 monitor_printf(mon, "store-global-state: %s\n",
3040 ms->store_global_state ? "on" : "off");
3041 monitor_printf(mon, "only-migratable: %s\n",
3042 ms->only_migratable ? "on" : "off");
3043 monitor_printf(mon, "send-configuration: %s\n",
3044 ms->send_configuration ? "on" : "off");
3045 monitor_printf(mon, "send-section-footer: %s\n",
3046 ms->send_section_footer ? "on" : "off");
3047 monitor_printf(mon, "decompress-error-check: %s\n",
3048 ms->decompress_error_check ? "on" : "off");
3051 #define DEFINE_PROP_MIG_CAP(name, x) \
3052 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3054 static Property migration_properties[] = {
3055 DEFINE_PROP_BOOL("store-global-state", MigrationState,
3056 store_global_state, true),
3057 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
3058 DEFINE_PROP_BOOL("send-configuration", MigrationState,
3059 send_configuration, true),
3060 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
3061 send_section_footer, true),
3062 DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
3063 decompress_error_check, true),
3065 /* Migration parameters */
3066 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
3067 parameters.compress_level,
3068 DEFAULT_MIGRATE_COMPRESS_LEVEL),
3069 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
3070 parameters.compress_threads,
3071 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
3072 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
3073 parameters.decompress_threads,
3074 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
3075 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
3076 parameters.cpu_throttle_initial,
3077 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
3078 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
3079 parameters.cpu_throttle_increment,
3080 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
3081 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
3082 parameters.max_bandwidth, MAX_THROTTLE),
3083 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
3084 parameters.downtime_limit,
3085 DEFAULT_MIGRATE_SET_DOWNTIME),
3086 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
3087 parameters.x_checkpoint_delay,
3088 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
3089 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState,
3090 parameters.x_multifd_channels,
3091 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
3092 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState,
3093 parameters.x_multifd_page_count,
3094 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
3095 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
3096 parameters.xbzrle_cache_size,
3097 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
3098 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
3099 parameters.max_postcopy_bandwidth,
3100 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
3102 /* Migration capabilities */
3103 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
3104 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
3105 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
3106 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
3107 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
3108 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
3109 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
3110 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
3111 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
3112 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
3113 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
3114 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
3116 DEFINE_PROP_END_OF_LIST(),
3119 static void migration_class_init(ObjectClass *klass, void *data)
3121 DeviceClass *dc = DEVICE_CLASS(klass);
3123 dc->user_creatable = false;
3124 dc->props = migration_properties;
3127 static void migration_instance_finalize(Object *obj)
3129 MigrationState *ms = MIGRATION_OBJ(obj);
3130 MigrationParameters *params = &ms->parameters;
3132 qemu_mutex_destroy(&ms->error_mutex);
3133 qemu_mutex_destroy(&ms->qemu_file_lock);
3134 g_free(params->tls_hostname);
3135 g_free(params->tls_creds);
3136 qemu_sem_destroy(&ms->rate_limit_sem);
3137 qemu_sem_destroy(&ms->pause_sem);
3138 qemu_sem_destroy(&ms->postcopy_pause_sem);
3139 qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
3140 qemu_sem_destroy(&ms->rp_state.rp_sem);
3141 error_free(ms->error);
3144 static void migration_instance_init(Object *obj)
3146 MigrationState *ms = MIGRATION_OBJ(obj);
3147 MigrationParameters *params = &ms->parameters;
3149 ms->state = MIGRATION_STATUS_NONE;
3150 ms->mbps = -1;
3151 qemu_sem_init(&ms->pause_sem, 0);
3152 qemu_mutex_init(&ms->error_mutex);
3154 params->tls_hostname = g_strdup("");
3155 params->tls_creds = g_strdup("");
3157 /* Set has_* up only for parameter checks */
3158 params->has_compress_level = true;
3159 params->has_compress_threads = true;
3160 params->has_decompress_threads = true;
3161 params->has_cpu_throttle_initial = true;
3162 params->has_cpu_throttle_increment = true;
3163 params->has_max_bandwidth = true;
3164 params->has_downtime_limit = true;
3165 params->has_x_checkpoint_delay = true;
3166 params->has_block_incremental = true;
3167 params->has_x_multifd_channels = true;
3168 params->has_x_multifd_page_count = true;
3169 params->has_xbzrle_cache_size = true;
3170 params->has_max_postcopy_bandwidth = true;
3172 qemu_sem_init(&ms->postcopy_pause_sem, 0);
3173 qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
3174 qemu_sem_init(&ms->rp_state.rp_sem, 0);
3175 qemu_sem_init(&ms->rate_limit_sem, 0);
3176 qemu_mutex_init(&ms->qemu_file_lock);
3180 * Return true if check pass, false otherwise. Error will be put
3181 * inside errp if provided.
3183 static bool migration_object_check(MigrationState *ms, Error **errp)
3185 MigrationCapabilityStatusList *head = NULL;
3186 /* Assuming all off */
3187 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
3188 int i;
3190 if (!migrate_params_check(&ms->parameters, errp)) {
3191 return false;
3194 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
3195 if (ms->enabled_capabilities[i]) {
3196 head = migrate_cap_add(head, i, true);
3200 ret = migrate_caps_check(cap_list, head, errp);
3202 /* It works with head == NULL */
3203 qapi_free_MigrationCapabilityStatusList(head);
3205 return ret;
3208 static const TypeInfo migration_type = {
3209 .name = TYPE_MIGRATION,
3211 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3212 * not created using qdev_create(), it is not attached to the qdev
3213 * device tree, and it is never realized.
3215 * TODO: Make this TYPE_OBJECT once QOM provides something like
3216 * TYPE_DEVICE's "-global" properties.
3218 .parent = TYPE_DEVICE,
3219 .class_init = migration_class_init,
3220 .class_size = sizeof(MigrationClass),
3221 .instance_size = sizeof(MigrationState),
3222 .instance_init = migration_instance_init,
3223 .instance_finalize = migration_instance_finalize,
3226 static void register_migration_types(void)
3228 type_register_static(&migration_type);
3231 type_init(register_migration_types);