4 * Copyright IBM, Corp. 2008
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 "qemu/main-loop.h"
20 #include "migration/blocker.h"
24 #include "sysemu/runstate.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/cpu-throttle.h"
29 #include "ram-compress.h"
30 #include "migration/global_state.h"
31 #include "migration/misc.h"
32 #include "migration.h"
33 #include "migration-stats.h"
35 #include "qemu-file.h"
37 #include "migration/vmstate.h"
38 #include "block/block.h"
39 #include "qapi/error.h"
40 #include "qapi/clone-visitor.h"
41 #include "qapi/qapi-visit-migration.h"
42 #include "qapi/qapi-visit-sockets.h"
43 #include "qapi/qapi-commands-migration.h"
44 #include "qapi/qapi-events-migration.h"
45 #include "qapi/qmp/qerror.h"
46 #include "qapi/qmp/qnull.h"
49 #include "postcopy-ram.h"
50 #include "qemu/thread.h"
52 #include "exec/target_page.h"
53 #include "io/channel-buffer.h"
54 #include "io/channel-tls.h"
55 #include "migration/colo.h"
56 #include "hw/boards.h"
57 #include "monitor/monitor.h"
58 #include "net/announce.h"
59 #include "qemu/queue.h"
61 #include "threadinfo.h"
62 #include "qemu/yank.h"
63 #include "sysemu/cpus.h"
64 #include "yank_functions.h"
65 #include "sysemu/qtest.h"
67 #include "sysemu/dirtylimit.h"
69 static NotifierList migration_state_notifiers =
70 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
72 /* Messages sent on the return path from destination to source */
73 enum mig_rp_message_type {
74 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
75 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
76 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
78 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
79 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
80 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
81 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
82 MIG_RP_MSG_SWITCHOVER_ACK, /* Tell source it's OK to do switchover */
87 /* When we add fault tolerance, we could have several
88 migrations at once. For now we don't need to add
89 dynamic creation of migration */
91 static MigrationState *current_migration;
92 static MigrationIncomingState *current_incoming;
94 static GSList *migration_blockers;
96 static bool migration_object_check(MigrationState *ms, Error **errp);
97 static int migration_maybe_pause(MigrationState *s,
98 int *current_active_state,
100 static void migrate_fd_cancel(MigrationState *s);
102 static bool migration_needs_multiple_sockets(void)
104 return migrate_multifd() || migrate_postcopy_preempt();
107 static bool uri_supports_multi_channels(const char *uri)
109 return strstart(uri, "tcp:", NULL) || strstart(uri, "unix:", NULL) ||
110 strstart(uri, "vsock:", NULL);
114 migration_channels_and_uri_compatible(const char *uri, Error **errp)
116 if (migration_needs_multiple_sockets() &&
117 !uri_supports_multi_channels(uri)) {
118 error_setg(errp, "Migration requires multi-channel URIs (e.g. tcp)");
125 static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp)
127 uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp;
129 return (a > b) - (a < b);
132 void migration_object_init(void)
134 /* This can only be called once. */
135 assert(!current_migration);
136 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
139 * Init the migrate incoming object as well no matter whether
140 * we'll use it or not.
142 assert(!current_incoming);
143 current_incoming = g_new0(MigrationIncomingState, 1);
144 current_incoming->state = MIGRATION_STATUS_NONE;
145 current_incoming->postcopy_remote_fds =
146 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
147 qemu_mutex_init(¤t_incoming->rp_mutex);
148 qemu_mutex_init(¤t_incoming->postcopy_prio_thread_mutex);
149 qemu_event_init(¤t_incoming->main_thread_load_event, false);
150 qemu_sem_init(¤t_incoming->postcopy_pause_sem_dst, 0);
151 qemu_sem_init(¤t_incoming->postcopy_pause_sem_fault, 0);
152 qemu_sem_init(¤t_incoming->postcopy_pause_sem_fast_load, 0);
153 qemu_sem_init(¤t_incoming->postcopy_qemufile_dst_done, 0);
155 qemu_mutex_init(¤t_incoming->page_request_mutex);
156 current_incoming->page_requested = g_tree_new(page_request_addr_cmp);
158 migration_object_check(current_migration, &error_fatal);
162 dirty_bitmap_mig_init();
165 void migration_cancel(const Error *error)
168 migrate_set_error(current_migration, error);
170 if (migrate_dirty_limit()) {
171 qmp_cancel_vcpu_dirty_limit(false, -1, NULL);
173 migrate_fd_cancel(current_migration);
176 void migration_shutdown(void)
179 * When the QEMU main thread exit, the COLO thread
180 * may wait a semaphore. So, we should wakeup the
181 * COLO thread before migration shutdown.
185 * Cancel the current migration - that will (eventually)
186 * stop the migration using this structure
188 migration_cancel(NULL);
189 object_unref(OBJECT(current_migration));
192 * Cancel outgoing migration of dirty bitmaps. It should
193 * at least unref used block nodes.
195 dirty_bitmap_mig_cancel_outgoing();
198 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
199 * are non-critical data, and their loss never considered as
202 dirty_bitmap_mig_cancel_incoming();
206 MigrationState *migrate_get_current(void)
208 /* This can only be called after the object created. */
209 assert(current_migration);
210 return current_migration;
213 MigrationIncomingState *migration_incoming_get_current(void)
215 assert(current_incoming);
216 return current_incoming;
219 void migration_incoming_transport_cleanup(MigrationIncomingState *mis)
221 if (mis->socket_address_list) {
222 qapi_free_SocketAddressList(mis->socket_address_list);
223 mis->socket_address_list = NULL;
226 if (mis->transport_cleanup) {
227 mis->transport_cleanup(mis->transport_data);
228 mis->transport_data = mis->transport_cleanup = NULL;
232 void migration_incoming_state_destroy(void)
234 struct MigrationIncomingState *mis = migration_incoming_get_current();
236 multifd_load_cleanup();
237 compress_threads_load_cleanup();
239 if (mis->to_src_file) {
240 /* Tell source that we are done */
241 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
242 qemu_fclose(mis->to_src_file);
243 mis->to_src_file = NULL;
246 if (mis->from_src_file) {
247 migration_ioc_unregister_yank_from_file(mis->from_src_file);
248 qemu_fclose(mis->from_src_file);
249 mis->from_src_file = NULL;
251 if (mis->postcopy_remote_fds) {
252 g_array_free(mis->postcopy_remote_fds, TRUE);
253 mis->postcopy_remote_fds = NULL;
256 migration_incoming_transport_cleanup(mis);
257 qemu_event_reset(&mis->main_thread_load_event);
259 if (mis->page_requested) {
260 g_tree_destroy(mis->page_requested);
261 mis->page_requested = NULL;
264 if (mis->postcopy_qemufile_dst) {
265 migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst);
266 qemu_fclose(mis->postcopy_qemufile_dst);
267 mis->postcopy_qemufile_dst = NULL;
270 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
273 static void migrate_generate_event(int new_state)
275 if (migrate_events()) {
276 qapi_event_send_migration(new_state);
281 * Send a message on the return channel back to the source
284 static int migrate_send_rp_message(MigrationIncomingState *mis,
285 enum mig_rp_message_type message_type,
286 uint16_t len, void *data)
290 trace_migrate_send_rp_message((int)message_type, len);
291 QEMU_LOCK_GUARD(&mis->rp_mutex);
294 * It's possible that the file handle got lost due to network
297 if (!mis->to_src_file) {
302 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
303 qemu_put_be16(mis->to_src_file, len);
304 qemu_put_buffer(mis->to_src_file, data, len);
305 qemu_fflush(mis->to_src_file);
307 /* It's possible that qemu file got error during sending */
308 ret = qemu_file_get_error(mis->to_src_file);
313 /* Request one page from the source VM at the given start address.
314 * rb: the RAMBlock to request the page in
315 * Start: Address offset within the RB
316 * Len: Length in bytes required - must be a multiple of pagesize
318 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
319 RAMBlock *rb, ram_addr_t start)
321 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
322 size_t msglen = 12; /* start + len */
323 size_t len = qemu_ram_pagesize(rb);
324 enum mig_rp_message_type msg_type;
328 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
329 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
332 * We maintain the last ramblock that we requested for page. Note that we
333 * don't need locking because this function will only be called within the
334 * postcopy ram fault thread.
336 if (rb != mis->last_rb) {
339 rbname = qemu_ram_get_idstr(rb);
340 rbname_len = strlen(rbname);
342 assert(rbname_len < 256);
344 bufc[msglen++] = rbname_len;
345 memcpy(bufc + msglen, rbname, rbname_len);
346 msglen += rbname_len;
347 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
349 msg_type = MIG_RP_MSG_REQ_PAGES;
352 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
355 int migrate_send_rp_req_pages(MigrationIncomingState *mis,
356 RAMBlock *rb, ram_addr_t start, uint64_t haddr)
358 void *aligned = (void *)(uintptr_t)ROUND_DOWN(haddr, qemu_ram_pagesize(rb));
359 bool received = false;
361 WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
362 received = ramblock_recv_bitmap_test_byte_offset(rb, start);
363 if (!received && !g_tree_lookup(mis->page_requested, aligned)) {
365 * The page has not been received, and it's not yet in the page
366 * request list. Queue it. Set the value of element to 1, so that
367 * things like g_tree_lookup() will return TRUE (1) when found.
369 g_tree_insert(mis->page_requested, aligned, (gpointer)1);
370 mis->page_requested_count++;
371 trace_postcopy_page_req_add(aligned, mis->page_requested_count);
376 * If the page is there, skip sending the message. We don't even need the
377 * lock because as long as the page arrived, it'll be there forever.
383 return migrate_send_rp_message_req_pages(mis, rb, start);
386 static bool migration_colo_enabled;
387 bool migration_incoming_colo_enabled(void)
389 return migration_colo_enabled;
392 void migration_incoming_disable_colo(void)
394 ram_block_discard_disable(false);
395 migration_colo_enabled = false;
398 int migration_incoming_enable_colo(void)
400 #ifndef CONFIG_REPLICATION
401 error_report("ENABLE_COLO command come in migration stream, but COLO "
402 "module is not built in");
406 if (!migrate_colo()) {
407 error_report("ENABLE_COLO command come in migration stream, but c-colo "
408 "capability is not set");
412 if (ram_block_discard_disable(true)) {
413 error_report("COLO: cannot disable RAM discard");
416 migration_colo_enabled = true;
420 void migrate_add_address(SocketAddress *address)
422 MigrationIncomingState *mis = migration_incoming_get_current();
424 QAPI_LIST_PREPEND(mis->socket_address_list,
425 QAPI_CLONE(SocketAddress, address));
428 static void qemu_start_incoming_migration(const char *uri, Error **errp)
430 const char *p = NULL;
432 /* URI is not suitable for migration? */
433 if (!migration_channels_and_uri_compatible(uri, errp)) {
437 qapi_event_send_migration(MIGRATION_STATUS_SETUP);
438 if (strstart(uri, "tcp:", &p) ||
439 strstart(uri, "unix:", NULL) ||
440 strstart(uri, "vsock:", NULL)) {
441 socket_start_incoming_migration(p ? p : uri, errp);
443 } else if (strstart(uri, "rdma:", &p)) {
444 rdma_start_incoming_migration(p, errp);
446 } else if (strstart(uri, "exec:", &p)) {
447 exec_start_incoming_migration(p, errp);
448 } else if (strstart(uri, "fd:", &p)) {
449 fd_start_incoming_migration(p, errp);
451 error_setg(errp, "unknown migration protocol: %s", uri);
455 static void process_incoming_migration_bh(void *opaque)
457 Error *local_err = NULL;
458 MigrationIncomingState *mis = opaque;
460 /* If capability late_block_activate is set:
461 * Only fire up the block code now if we're going to restart the
462 * VM, else 'cont' will do it.
463 * This causes file locking to happen; so we don't want it to happen
464 * unless we really are starting the VM.
466 if (!migrate_late_block_activate() ||
467 (autostart && (!global_state_received() ||
468 global_state_get_runstate() == RUN_STATE_RUNNING))) {
469 /* Make sure all file formats throw away their mutable metadata.
470 * If we get an error here, just don't restart the VM yet. */
471 bdrv_activate_all(&local_err);
473 error_report_err(local_err);
480 * This must happen after all error conditions are dealt with and
481 * we're sure the VM is going to be running on this host.
483 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
485 multifd_load_shutdown();
487 dirty_bitmap_mig_before_vm_start();
489 if (!global_state_received() ||
490 global_state_get_runstate() == RUN_STATE_RUNNING) {
494 runstate_set(RUN_STATE_PAUSED);
496 } else if (migration_incoming_colo_enabled()) {
497 migration_incoming_disable_colo();
500 runstate_set(global_state_get_runstate());
503 * This must happen after any state changes since as soon as an external
504 * observer sees this event they might start to prod at the VM assuming
507 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
508 MIGRATION_STATUS_COMPLETED);
509 qemu_bh_delete(mis->bh);
510 migration_incoming_state_destroy();
513 static void coroutine_fn
514 process_incoming_migration_co(void *opaque)
516 MigrationIncomingState *mis = migration_incoming_get_current();
520 assert(mis->from_src_file);
522 if (compress_threads_load_setup(mis->from_src_file)) {
523 error_report("Failed to setup decompress threads");
527 mis->largest_page_size = qemu_ram_pagesize_largest();
528 postcopy_state_set(POSTCOPY_INCOMING_NONE);
529 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
530 MIGRATION_STATUS_ACTIVE);
532 mis->loadvm_co = qemu_coroutine_self();
533 ret = qemu_loadvm_state(mis->from_src_file);
534 mis->loadvm_co = NULL;
536 ps = postcopy_state_get();
537 trace_process_incoming_migration_co_end(ret, ps);
538 if (ps != POSTCOPY_INCOMING_NONE) {
539 if (ps == POSTCOPY_INCOMING_ADVISE) {
541 * Where a migration had postcopy enabled (and thus went to advise)
542 * but managed to complete within the precopy period, we can use
545 postcopy_ram_incoming_cleanup(mis);
546 } else if (ret >= 0) {
548 * Postcopy was started, cleanup should happen at the end of the
551 trace_process_incoming_migration_co_postcopy_end_main();
554 /* Else if something went wrong then just fall out of the normal exit */
558 error_report("load of migration failed: %s", strerror(-ret));
562 if (colo_incoming_co() < 0) {
566 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
567 qemu_bh_schedule(mis->bh);
570 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
571 MIGRATION_STATUS_FAILED);
572 qemu_fclose(mis->from_src_file);
574 multifd_load_cleanup();
575 compress_threads_load_cleanup();
581 * migration_incoming_setup: Setup incoming migration
582 * @f: file for main migration channel
583 * @errp: where to put errors
585 * Returns: %true on success, %false on error.
587 static bool migration_incoming_setup(QEMUFile *f, Error **errp)
589 MigrationIncomingState *mis = migration_incoming_get_current();
591 if (!mis->from_src_file) {
592 mis->from_src_file = f;
594 qemu_file_set_blocking(f, false);
598 void migration_incoming_process(void)
600 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
601 qemu_coroutine_enter(co);
604 /* Returns true if recovered from a paused migration, otherwise false */
605 static bool postcopy_try_recover(void)
607 MigrationIncomingState *mis = migration_incoming_get_current();
609 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
610 /* Resumed from a paused postcopy migration */
612 /* This should be set already in migration_incoming_setup() */
613 assert(mis->from_src_file);
614 /* Postcopy has standalone thread to do vm load */
615 qemu_file_set_blocking(mis->from_src_file, true);
617 /* Re-configure the return path */
618 mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
620 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
621 MIGRATION_STATUS_POSTCOPY_RECOVER);
624 * Here, we only wake up the main loading thread (while the
625 * rest threads will still be waiting), so that we can receive
626 * commands from source now, and answer it if needed. The
627 * rest threads will be woken up afterwards until we are sure
628 * that source is ready to reply to page requests.
630 qemu_sem_post(&mis->postcopy_pause_sem_dst);
637 void migration_fd_process_incoming(QEMUFile *f, Error **errp)
639 if (!migration_incoming_setup(f, errp)) {
642 if (postcopy_try_recover()) {
645 migration_incoming_process();
649 * Returns true when we want to start a new incoming migration process,
652 static bool migration_should_start_incoming(bool main_channel)
654 /* Multifd doesn't start unless all channels are established */
655 if (migrate_multifd()) {
656 return migration_has_all_channels();
659 /* Preempt channel only starts when the main channel is created */
660 if (migrate_postcopy_preempt()) {
665 * For all the rest types of migration, we should only reach here when
666 * it's the main channel that's being created, and we should always
667 * proceed with this channel.
669 assert(main_channel);
673 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
675 MigrationIncomingState *mis = migration_incoming_get_current();
676 Error *local_err = NULL;
678 bool default_channel = true;
679 uint32_t channel_magic = 0;
682 if (migrate_multifd() && !migrate_postcopy_ram() &&
683 qio_channel_has_feature(ioc, QIO_CHANNEL_FEATURE_READ_MSG_PEEK)) {
685 * With multiple channels, it is possible that we receive channels
686 * out of order on destination side, causing incorrect mapping of
687 * source channels on destination side. Check channel MAGIC to
688 * decide type of channel. Please note this is best effort, postcopy
689 * preempt channel does not send any magic number so avoid it for
690 * postcopy live migration. Also tls live migration already does
691 * tls handshake while initializing main channel so with tls this
692 * issue is not possible.
694 ret = migration_channel_read_peek(ioc, (void *)&channel_magic,
695 sizeof(channel_magic), &local_err);
698 error_propagate(errp, local_err);
702 default_channel = (channel_magic == cpu_to_be32(QEMU_VM_FILE_MAGIC));
704 default_channel = !mis->from_src_file;
707 if (multifd_load_setup(errp) != 0) {
708 error_setg(errp, "Failed to setup multifd channels");
712 if (default_channel) {
713 f = qemu_file_new_input(ioc);
715 if (!migration_incoming_setup(f, errp)) {
719 /* Multiple connections */
720 assert(migration_needs_multiple_sockets());
721 if (migrate_multifd()) {
722 multifd_recv_new_channel(ioc, &local_err);
724 assert(migrate_postcopy_preempt());
725 f = qemu_file_new_input(ioc);
726 postcopy_preempt_new_channel(mis, f);
729 error_propagate(errp, local_err);
734 if (migration_should_start_incoming(default_channel)) {
735 /* If it's a recovery, we're done */
736 if (postcopy_try_recover()) {
739 migration_incoming_process();
744 * @migration_has_all_channels: We have received all channels that we need
746 * Returns true when we have got connections to all the channels that
747 * we need for migration.
749 bool migration_has_all_channels(void)
751 MigrationIncomingState *mis = migration_incoming_get_current();
753 if (!mis->from_src_file) {
757 if (migrate_multifd()) {
758 return multifd_recv_all_channels_created();
761 if (migrate_postcopy_preempt()) {
762 return mis->postcopy_qemufile_dst != NULL;
768 int migrate_send_rp_switchover_ack(MigrationIncomingState *mis)
770 return migrate_send_rp_message(mis, MIG_RP_MSG_SWITCHOVER_ACK, 0, NULL);
774 * Send a 'SHUT' message on the return channel with the given value
775 * to indicate that we've finished with the RP. Non-0 value indicates
778 void migrate_send_rp_shut(MigrationIncomingState *mis,
783 buf = cpu_to_be32(value);
784 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
788 * Send a 'PONG' message on the return channel with the given value
789 * (normally in response to a 'PING')
791 void migrate_send_rp_pong(MigrationIncomingState *mis,
796 buf = cpu_to_be32(value);
797 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
800 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
808 * First, we send the header part. It contains only the len of
809 * idstr, and the idstr itself.
811 len = strlen(block_name);
813 memcpy(buf + 1, block_name, len);
815 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
816 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
821 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
824 * Next, we dump the received bitmap to the stream.
826 * TODO: currently we are safe since we are the only one that is
827 * using the to_src_file handle (fault thread is still paused),
828 * and it's ok even not taking the mutex. However the best way is
829 * to take the lock before sending the message header, and release
830 * the lock after sending the bitmap.
832 qemu_mutex_lock(&mis->rp_mutex);
833 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
834 qemu_mutex_unlock(&mis->rp_mutex);
836 trace_migrate_send_rp_recv_bitmap(block_name, res);
839 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
843 buf = cpu_to_be32(value);
844 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
848 * Return true if we're already in the middle of a migration
849 * (i.e. any of the active or setup states)
851 bool migration_is_setup_or_active(int state)
854 case MIGRATION_STATUS_ACTIVE:
855 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
856 case MIGRATION_STATUS_POSTCOPY_PAUSED:
857 case MIGRATION_STATUS_POSTCOPY_RECOVER:
858 case MIGRATION_STATUS_SETUP:
859 case MIGRATION_STATUS_PRE_SWITCHOVER:
860 case MIGRATION_STATUS_DEVICE:
861 case MIGRATION_STATUS_WAIT_UNPLUG:
862 case MIGRATION_STATUS_COLO:
871 bool migration_is_running(int state)
874 case MIGRATION_STATUS_ACTIVE:
875 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
876 case MIGRATION_STATUS_POSTCOPY_PAUSED:
877 case MIGRATION_STATUS_POSTCOPY_RECOVER:
878 case MIGRATION_STATUS_SETUP:
879 case MIGRATION_STATUS_PRE_SWITCHOVER:
880 case MIGRATION_STATUS_DEVICE:
881 case MIGRATION_STATUS_WAIT_UNPLUG:
882 case MIGRATION_STATUS_CANCELLING:
891 static bool migrate_show_downtime(MigrationState *s)
893 return (s->state == MIGRATION_STATUS_COMPLETED) || migration_in_postcopy();
896 static void populate_time_info(MigrationInfo *info, MigrationState *s)
898 info->has_status = true;
899 info->has_setup_time = true;
900 info->setup_time = s->setup_time;
902 if (s->state == MIGRATION_STATUS_COMPLETED) {
903 info->has_total_time = true;
904 info->total_time = s->total_time;
906 info->has_total_time = true;
907 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
911 if (migrate_show_downtime(s)) {
912 info->has_downtime = true;
913 info->downtime = s->downtime;
915 info->has_expected_downtime = true;
916 info->expected_downtime = s->expected_downtime;
920 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
922 size_t page_size = qemu_target_page_size();
924 info->ram = g_malloc0(sizeof(*info->ram));
925 info->ram->transferred = stat64_get(&mig_stats.transferred);
926 info->ram->total = ram_bytes_total();
927 info->ram->duplicate = stat64_get(&mig_stats.zero_pages);
928 /* legacy value. It is not used anymore */
929 info->ram->skipped = 0;
930 info->ram->normal = stat64_get(&mig_stats.normal_pages);
931 info->ram->normal_bytes = info->ram->normal * page_size;
932 info->ram->mbps = s->mbps;
933 info->ram->dirty_sync_count =
934 stat64_get(&mig_stats.dirty_sync_count);
935 info->ram->dirty_sync_missed_zero_copy =
936 stat64_get(&mig_stats.dirty_sync_missed_zero_copy);
937 info->ram->postcopy_requests =
938 stat64_get(&mig_stats.postcopy_requests);
939 info->ram->page_size = page_size;
940 info->ram->multifd_bytes = stat64_get(&mig_stats.multifd_bytes);
941 info->ram->pages_per_second = s->pages_per_second;
942 info->ram->precopy_bytes = stat64_get(&mig_stats.precopy_bytes);
943 info->ram->downtime_bytes = stat64_get(&mig_stats.downtime_bytes);
944 info->ram->postcopy_bytes = stat64_get(&mig_stats.postcopy_bytes);
946 if (migrate_xbzrle()) {
947 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
948 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
949 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
950 info->xbzrle_cache->pages = xbzrle_counters.pages;
951 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
952 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
953 info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate;
954 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
957 if (migrate_compress()) {
958 info->compression = g_malloc0(sizeof(*info->compression));
959 info->compression->pages = compression_counters.pages;
960 info->compression->busy = compression_counters.busy;
961 info->compression->busy_rate = compression_counters.busy_rate;
962 info->compression->compressed_size =
963 compression_counters.compressed_size;
964 info->compression->compression_rate =
965 compression_counters.compression_rate;
968 if (cpu_throttle_active()) {
969 info->has_cpu_throttle_percentage = true;
970 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
973 if (s->state != MIGRATION_STATUS_COMPLETED) {
974 info->ram->remaining = ram_bytes_remaining();
975 info->ram->dirty_pages_rate =
976 stat64_get(&mig_stats.dirty_pages_rate);
979 if (migrate_dirty_limit() && dirtylimit_in_service()) {
980 info->has_dirty_limit_throttle_time_per_round = true;
981 info->dirty_limit_throttle_time_per_round =
982 dirtylimit_throttle_time_per_round();
984 info->has_dirty_limit_ring_full_time = true;
985 info->dirty_limit_ring_full_time = dirtylimit_ring_full_time();
989 static void populate_disk_info(MigrationInfo *info)
991 if (blk_mig_active()) {
992 info->disk = g_malloc0(sizeof(*info->disk));
993 info->disk->transferred = blk_mig_bytes_transferred();
994 info->disk->remaining = blk_mig_bytes_remaining();
995 info->disk->total = blk_mig_bytes_total();
999 static void fill_source_migration_info(MigrationInfo *info)
1001 MigrationState *s = migrate_get_current();
1002 int state = qatomic_read(&s->state);
1003 GSList *cur_blocker = migration_blockers;
1005 info->blocked_reasons = NULL;
1008 * There are two types of reasons a migration might be blocked;
1009 * a) devices marked in VMState as non-migratable, and
1010 * b) Explicit migration blockers
1011 * We need to add both of them here.
1013 qemu_savevm_non_migratable_list(&info->blocked_reasons);
1015 while (cur_blocker) {
1016 QAPI_LIST_PREPEND(info->blocked_reasons,
1017 g_strdup(error_get_pretty(cur_blocker->data)));
1018 cur_blocker = g_slist_next(cur_blocker);
1020 info->has_blocked_reasons = info->blocked_reasons != NULL;
1023 case MIGRATION_STATUS_NONE:
1024 /* no migration has happened ever */
1025 /* do not overwrite destination migration status */
1027 case MIGRATION_STATUS_SETUP:
1028 info->has_status = true;
1029 info->has_total_time = false;
1031 case MIGRATION_STATUS_ACTIVE:
1032 case MIGRATION_STATUS_CANCELLING:
1033 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1034 case MIGRATION_STATUS_PRE_SWITCHOVER:
1035 case MIGRATION_STATUS_DEVICE:
1036 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1037 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1038 /* TODO add some postcopy stats */
1039 populate_time_info(info, s);
1040 populate_ram_info(info, s);
1041 populate_disk_info(info);
1042 populate_vfio_info(info);
1044 case MIGRATION_STATUS_COLO:
1045 info->has_status = true;
1046 /* TODO: display COLO specific information (checkpoint info etc.) */
1048 case MIGRATION_STATUS_COMPLETED:
1049 populate_time_info(info, s);
1050 populate_ram_info(info, s);
1051 populate_vfio_info(info);
1053 case MIGRATION_STATUS_FAILED:
1054 info->has_status = true;
1056 info->error_desc = g_strdup(error_get_pretty(s->error));
1059 case MIGRATION_STATUS_CANCELLED:
1060 info->has_status = true;
1062 case MIGRATION_STATUS_WAIT_UNPLUG:
1063 info->has_status = true;
1066 info->status = state;
1069 static void fill_destination_migration_info(MigrationInfo *info)
1071 MigrationIncomingState *mis = migration_incoming_get_current();
1073 if (mis->socket_address_list) {
1074 info->has_socket_address = true;
1075 info->socket_address =
1076 QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1079 switch (mis->state) {
1080 case MIGRATION_STATUS_NONE:
1082 case MIGRATION_STATUS_SETUP:
1083 case MIGRATION_STATUS_CANCELLING:
1084 case MIGRATION_STATUS_CANCELLED:
1085 case MIGRATION_STATUS_ACTIVE:
1086 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1087 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1088 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1089 case MIGRATION_STATUS_FAILED:
1090 case MIGRATION_STATUS_COLO:
1091 info->has_status = true;
1093 case MIGRATION_STATUS_COMPLETED:
1094 info->has_status = true;
1095 fill_destination_postcopy_migration_info(info);
1098 info->status = mis->state;
1101 MigrationInfo *qmp_query_migrate(Error **errp)
1103 MigrationInfo *info = g_malloc0(sizeof(*info));
1105 fill_destination_migration_info(info);
1106 fill_source_migration_info(info);
1111 void qmp_migrate_start_postcopy(Error **errp)
1113 MigrationState *s = migrate_get_current();
1115 if (!migrate_postcopy()) {
1116 error_setg(errp, "Enable postcopy with migrate_set_capability before"
1117 " the start of migration");
1121 if (s->state == MIGRATION_STATUS_NONE) {
1122 error_setg(errp, "Postcopy must be started after migration has been"
1127 * we don't error if migration has finished since that would be racy
1128 * with issuing this command.
1130 qatomic_set(&s->start_postcopy, true);
1133 /* shared migration helpers */
1135 void migrate_set_state(int *state, int old_state, int new_state)
1137 assert(new_state < MIGRATION_STATUS__MAX);
1138 if (qatomic_cmpxchg(state, old_state, new_state) == old_state) {
1139 trace_migrate_set_state(MigrationStatus_str(new_state));
1140 migrate_generate_event(new_state);
1144 static void migrate_fd_cleanup(MigrationState *s)
1146 qemu_bh_delete(s->cleanup_bh);
1147 s->cleanup_bh = NULL;
1149 g_free(s->hostname);
1151 json_writer_free(s->vmdesc);
1154 qemu_savevm_state_cleanup();
1156 if (s->to_dst_file) {
1159 trace_migrate_fd_cleanup();
1160 qemu_mutex_unlock_iothread();
1161 if (s->migration_thread_running) {
1162 qemu_thread_join(&s->thread);
1163 s->migration_thread_running = false;
1165 qemu_mutex_lock_iothread();
1167 multifd_save_cleanup();
1168 qemu_mutex_lock(&s->qemu_file_lock);
1169 tmp = s->to_dst_file;
1170 s->to_dst_file = NULL;
1171 qemu_mutex_unlock(&s->qemu_file_lock);
1173 * Close the file handle without the lock to make sure the
1174 * critical section won't block for long.
1176 migration_ioc_unregister_yank_from_file(tmp);
1180 if (s->postcopy_qemufile_src) {
1181 migration_ioc_unregister_yank_from_file(s->postcopy_qemufile_src);
1182 qemu_fclose(s->postcopy_qemufile_src);
1183 s->postcopy_qemufile_src = NULL;
1186 assert(!migration_is_active(s));
1188 if (s->state == MIGRATION_STATUS_CANCELLING) {
1189 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1190 MIGRATION_STATUS_CANCELLED);
1194 /* It is used on info migrate. We can't free it */
1195 error_report_err(error_copy(s->error));
1197 notifier_list_notify(&migration_state_notifiers, s);
1198 block_cleanup_parameters();
1199 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
1202 static void migrate_fd_cleanup_schedule(MigrationState *s)
1205 * Ref the state for bh, because it may be called when
1206 * there're already no other refs
1208 object_ref(OBJECT(s));
1209 qemu_bh_schedule(s->cleanup_bh);
1212 static void migrate_fd_cleanup_bh(void *opaque)
1214 MigrationState *s = opaque;
1215 migrate_fd_cleanup(s);
1216 object_unref(OBJECT(s));
1219 void migrate_set_error(MigrationState *s, const Error *error)
1221 QEMU_LOCK_GUARD(&s->error_mutex);
1223 s->error = error_copy(error);
1227 static void migrate_error_free(MigrationState *s)
1229 QEMU_LOCK_GUARD(&s->error_mutex);
1231 error_free(s->error);
1236 static void migrate_fd_error(MigrationState *s, const Error *error)
1238 trace_migrate_fd_error(error_get_pretty(error));
1239 assert(s->to_dst_file == NULL);
1240 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1241 MIGRATION_STATUS_FAILED);
1242 migrate_set_error(s, error);
1245 static void migrate_fd_cancel(MigrationState *s)
1248 QEMUFile *f = migrate_get_current()->to_dst_file;
1249 trace_migrate_fd_cancel();
1251 WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) {
1252 if (s->rp_state.from_dst_file) {
1253 /* shutdown the rp socket, so causing the rp thread to shutdown */
1254 qemu_file_shutdown(s->rp_state.from_dst_file);
1259 old_state = s->state;
1260 if (!migration_is_running(old_state)) {
1263 /* If the migration is paused, kick it out of the pause */
1264 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1265 qemu_sem_post(&s->pause_sem);
1267 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1268 } while (s->state != MIGRATION_STATUS_CANCELLING);
1271 * If we're unlucky the migration code might be stuck somewhere in a
1272 * send/write while the network has failed and is waiting to timeout;
1273 * if we've got shutdown(2) available then we can force it to quit.
1274 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1275 * called in a bh, so there is no race against this cancel.
1277 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1278 qemu_file_shutdown(f);
1280 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1281 Error *local_err = NULL;
1283 bdrv_activate_all(&local_err);
1285 error_report_err(local_err);
1287 s->block_inactive = false;
1292 void add_migration_state_change_notifier(Notifier *notify)
1294 notifier_list_add(&migration_state_notifiers, notify);
1297 void remove_migration_state_change_notifier(Notifier *notify)
1299 notifier_remove(notify);
1302 bool migration_in_setup(MigrationState *s)
1304 return s->state == MIGRATION_STATUS_SETUP;
1307 bool migration_has_finished(MigrationState *s)
1309 return s->state == MIGRATION_STATUS_COMPLETED;
1312 bool migration_has_failed(MigrationState *s)
1314 return (s->state == MIGRATION_STATUS_CANCELLED ||
1315 s->state == MIGRATION_STATUS_FAILED);
1318 bool migration_in_postcopy(void)
1320 MigrationState *s = migrate_get_current();
1323 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1324 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1325 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1332 bool migration_in_postcopy_after_devices(MigrationState *s)
1334 return migration_in_postcopy() && s->postcopy_after_devices;
1337 bool migration_in_incoming_postcopy(void)
1339 PostcopyState ps = postcopy_state_get();
1341 return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END;
1344 bool migration_incoming_postcopy_advised(void)
1346 PostcopyState ps = postcopy_state_get();
1348 return ps >= POSTCOPY_INCOMING_ADVISE && ps < POSTCOPY_INCOMING_END;
1351 bool migration_in_bg_snapshot(void)
1353 MigrationState *s = migrate_get_current();
1355 return migrate_background_snapshot() &&
1356 migration_is_setup_or_active(s->state);
1359 bool migration_is_idle(void)
1361 MigrationState *s = current_migration;
1368 case MIGRATION_STATUS_NONE:
1369 case MIGRATION_STATUS_CANCELLED:
1370 case MIGRATION_STATUS_COMPLETED:
1371 case MIGRATION_STATUS_FAILED:
1373 case MIGRATION_STATUS_SETUP:
1374 case MIGRATION_STATUS_CANCELLING:
1375 case MIGRATION_STATUS_ACTIVE:
1376 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1377 case MIGRATION_STATUS_COLO:
1378 case MIGRATION_STATUS_PRE_SWITCHOVER:
1379 case MIGRATION_STATUS_DEVICE:
1380 case MIGRATION_STATUS_WAIT_UNPLUG:
1382 case MIGRATION_STATUS__MAX:
1383 g_assert_not_reached();
1389 bool migration_is_active(MigrationState *s)
1391 return (s->state == MIGRATION_STATUS_ACTIVE ||
1392 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1395 void migrate_init(MigrationState *s)
1398 * Reinitialise all migration state, except
1399 * parameters/capabilities that the user set, and
1404 s->to_dst_file = NULL;
1405 s->state = MIGRATION_STATUS_NONE;
1406 s->rp_state.from_dst_file = NULL;
1407 s->rp_state.error = false;
1409 s->pages_per_second = 0.0;
1411 s->expected_downtime = 0;
1413 s->start_postcopy = false;
1414 s->postcopy_after_devices = false;
1415 s->migration_thread_running = false;
1416 error_free(s->error);
1420 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1422 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1424 s->vm_old_state = -1;
1425 s->iteration_initial_bytes = 0;
1426 s->threshold_size = 0;
1427 s->switchover_acked = false;
1430 int migrate_add_blocker_internal(Error *reason, Error **errp)
1432 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
1433 if (runstate_check(RUN_STATE_SAVE_VM) || !migration_is_idle()) {
1434 error_propagate_prepend(errp, error_copy(reason),
1435 "disallowing migration blocker "
1436 "(migration/snapshot in progress) for: ");
1440 migration_blockers = g_slist_prepend(migration_blockers, reason);
1444 int migrate_add_blocker(Error *reason, Error **errp)
1446 if (only_migratable) {
1447 error_propagate_prepend(errp, error_copy(reason),
1448 "disallowing migration blocker "
1449 "(--only-migratable) for: ");
1453 return migrate_add_blocker_internal(reason, errp);
1456 void migrate_del_blocker(Error *reason)
1458 migration_blockers = g_slist_remove(migration_blockers, reason);
1461 void qmp_migrate_incoming(const char *uri, Error **errp)
1463 Error *local_err = NULL;
1464 static bool once = true;
1467 error_setg(errp, "The incoming migration has already been started");
1470 if (!runstate_check(RUN_STATE_INMIGRATE)) {
1471 error_setg(errp, "'-incoming' was not specified on the command line");
1475 if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
1479 qemu_start_incoming_migration(uri, &local_err);
1482 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
1483 error_propagate(errp, local_err);
1490 void qmp_migrate_recover(const char *uri, Error **errp)
1492 MigrationIncomingState *mis = migration_incoming_get_current();
1495 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
1496 * callers (no one should ignore a recover failure); if there is, it's a
1497 * programming error.
1501 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1502 error_setg(errp, "Migrate recover can only be run "
1503 "when postcopy is paused.");
1507 /* If there's an existing transport, release it */
1508 migration_incoming_transport_cleanup(mis);
1511 * Note that this call will never start a real migration; it will
1512 * only re-setup the migration stream and poke existing migration
1513 * to continue using that newly established channel.
1515 qemu_start_incoming_migration(uri, errp);
1518 void qmp_migrate_pause(Error **errp)
1520 MigrationState *ms = migrate_get_current();
1521 MigrationIncomingState *mis = migration_incoming_get_current();
1524 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1525 /* Source side, during postcopy */
1526 qemu_mutex_lock(&ms->qemu_file_lock);
1527 ret = qemu_file_shutdown(ms->to_dst_file);
1528 qemu_mutex_unlock(&ms->qemu_file_lock);
1530 error_setg(errp, "Failed to pause source migration");
1535 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1536 ret = qemu_file_shutdown(mis->from_src_file);
1538 error_setg(errp, "Failed to pause destination migration");
1543 error_setg(errp, "migrate-pause is currently only supported "
1544 "during postcopy-active state");
1547 bool migration_is_blocked(Error **errp)
1549 if (qemu_savevm_state_blocked(errp)) {
1553 if (migration_blockers) {
1554 error_propagate(errp, error_copy(migration_blockers->data));
1561 /* Returns true if continue to migrate, or false if error detected */
1562 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
1563 bool resume, Error **errp)
1565 Error *local_err = NULL;
1568 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1569 error_setg(errp, "Cannot resume if there is no "
1570 "paused migration");
1575 * Postcopy recovery won't work well with release-ram
1576 * capability since release-ram will drop the page buffer as
1577 * long as the page is put into the send buffer. So if there
1578 * is a network failure happened, any page buffers that have
1579 * not yet reached the destination VM but have already been
1580 * sent from the source VM will be lost forever. Let's refuse
1581 * the client from resuming such a postcopy migration.
1582 * Luckily release-ram was designed to only be used when src
1583 * and destination VMs are on the same host, so it should be
1586 if (migrate_release_ram()) {
1587 error_setg(errp, "Postcopy recovery cannot work "
1588 "when release-ram capability is set");
1592 /* This is a resume, skip init status */
1596 if (migration_is_running(s->state)) {
1597 error_setg(errp, QERR_MIGRATION_ACTIVE);
1601 if (runstate_check(RUN_STATE_INMIGRATE)) {
1602 error_setg(errp, "Guest is waiting for an incoming migration");
1606 if (runstate_check(RUN_STATE_POSTMIGRATE)) {
1607 error_setg(errp, "Can't migrate the vm that was paused due to "
1608 "previous migration");
1612 if (migration_is_blocked(errp)) {
1616 if (blk || blk_inc) {
1617 if (migrate_colo()) {
1618 error_setg(errp, "No disk migration is required in COLO mode");
1621 if (migrate_block() || migrate_block_incremental()) {
1622 error_setg(errp, "Command options are incompatible with "
1623 "current migration capabilities");
1626 if (!migrate_cap_set(MIGRATION_CAPABILITY_BLOCK, true, &local_err)) {
1627 error_propagate(errp, local_err);
1630 s->must_remove_block_options = true;
1634 migrate_set_block_incremental(true);
1639 * set mig_stats compression_counters memory to zero for a
1642 memset(&mig_stats, 0, sizeof(mig_stats));
1643 memset(&compression_counters, 0, sizeof(compression_counters));
1644 reset_vfio_bytes_transferred();
1649 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1650 bool has_inc, bool inc, bool has_detach, bool detach,
1651 bool has_resume, bool resume, Error **errp)
1653 bool resume_requested;
1654 Error *local_err = NULL;
1655 MigrationState *s = migrate_get_current();
1656 const char *p = NULL;
1658 /* URI is not suitable for migration? */
1659 if (!migration_channels_and_uri_compatible(uri, errp)) {
1663 resume_requested = has_resume && resume;
1664 if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
1665 resume_requested, errp)) {
1666 /* Error detected, put into errp */
1670 if (!resume_requested) {
1671 if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
1676 if (strstart(uri, "tcp:", &p) ||
1677 strstart(uri, "unix:", NULL) ||
1678 strstart(uri, "vsock:", NULL)) {
1679 socket_start_outgoing_migration(s, p ? p : uri, &local_err);
1681 } else if (strstart(uri, "rdma:", &p)) {
1682 rdma_start_outgoing_migration(s, p, &local_err);
1684 } else if (strstart(uri, "exec:", &p)) {
1685 exec_start_outgoing_migration(s, p, &local_err);
1686 } else if (strstart(uri, "fd:", &p)) {
1687 fd_start_outgoing_migration(s, p, &local_err);
1689 if (!resume_requested) {
1690 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
1692 error_setg(&local_err, QERR_INVALID_PARAMETER_VALUE, "uri",
1693 "a valid migration protocol");
1694 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1695 MIGRATION_STATUS_FAILED);
1696 block_cleanup_parameters();
1701 if (!resume_requested) {
1702 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
1704 migrate_fd_error(s, local_err);
1705 error_propagate(errp, local_err);
1710 void qmp_migrate_cancel(Error **errp)
1712 migration_cancel(NULL);
1715 void qmp_migrate_continue(MigrationStatus state, Error **errp)
1717 MigrationState *s = migrate_get_current();
1718 if (s->state != state) {
1719 error_setg(errp, "Migration not in expected state: %s",
1720 MigrationStatus_str(s->state));
1723 qemu_sem_post(&s->pause_sem);
1726 /* migration thread support */
1728 * Something bad happened to the RP stream, mark an error
1729 * The caller shall print or trace something to indicate why
1731 static void mark_source_rp_bad(MigrationState *s)
1733 s->rp_state.error = true;
1736 static struct rp_cmd_args {
1737 ssize_t len; /* -1 = variable */
1740 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1741 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1742 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1743 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1744 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
1745 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
1746 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
1747 [MIG_RP_MSG_SWITCHOVER_ACK] = { .len = 0, .name = "SWITCHOVER_ACK" },
1748 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1752 * Process a request for pages received on the return path,
1753 * We're allowed to send more than requested (e.g. to round to our page size)
1754 * and we don't need to send pages that have already been sent.
1756 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1757 ram_addr_t start, size_t len)
1759 long our_host_ps = qemu_real_host_page_size();
1761 trace_migrate_handle_rp_req_pages(rbname, start, len);
1764 * Since we currently insist on matching page sizes, just sanity check
1765 * we're being asked for whole host pages.
1767 if (!QEMU_IS_ALIGNED(start, our_host_ps) ||
1768 !QEMU_IS_ALIGNED(len, our_host_ps)) {
1769 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1770 " len: %zd", __func__, start, len);
1771 mark_source_rp_bad(ms);
1775 if (ram_save_queue_pages(rbname, start, len)) {
1776 mark_source_rp_bad(ms);
1780 /* Return true to retry, false to quit */
1781 static bool postcopy_pause_return_path_thread(MigrationState *s)
1783 trace_postcopy_pause_return_path();
1785 qemu_sem_wait(&s->postcopy_pause_rp_sem);
1787 trace_postcopy_pause_return_path_continued();
1792 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
1794 RAMBlock *block = qemu_ram_block_by_name(block_name);
1797 error_report("%s: invalid block name '%s'", __func__, block_name);
1801 /* Fetch the received bitmap and refresh the dirty bitmap */
1802 return ram_dirty_bitmap_reload(s, block);
1805 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
1807 trace_source_return_path_thread_resume_ack(value);
1809 if (value != MIGRATION_RESUME_ACK_VALUE) {
1810 error_report("%s: illegal resume_ack value %"PRIu32,
1815 /* Now both sides are active. */
1816 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
1817 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1819 /* Notify send thread that time to continue send pages */
1820 qemu_sem_post(&s->rp_state.rp_sem);
1826 * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if
1827 * existed) in a safe way.
1829 static void migration_release_dst_files(MigrationState *ms)
1833 WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) {
1835 * Reset the from_dst_file pointer first before releasing it, as we
1836 * can't block within lock section
1838 file = ms->rp_state.from_dst_file;
1839 ms->rp_state.from_dst_file = NULL;
1843 * Do the same to postcopy fast path socket too if there is. No
1844 * locking needed because this qemufile should only be managed by
1845 * return path thread.
1847 if (ms->postcopy_qemufile_src) {
1848 migration_ioc_unregister_yank_from_file(ms->postcopy_qemufile_src);
1849 qemu_file_shutdown(ms->postcopy_qemufile_src);
1850 qemu_fclose(ms->postcopy_qemufile_src);
1851 ms->postcopy_qemufile_src = NULL;
1858 * Handles messages sent on the return path towards the source VM
1861 static void *source_return_path_thread(void *opaque)
1863 MigrationState *ms = opaque;
1864 QEMUFile *rp = ms->rp_state.from_dst_file;
1865 uint16_t header_len, header_type;
1867 uint32_t tmp32, sibling_error;
1868 ram_addr_t start = 0; /* =0 to silence warning */
1869 size_t len = 0, expected_len;
1872 trace_source_return_path_thread_entry();
1873 rcu_register_thread();
1876 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1877 migration_is_setup_or_active(ms->state)) {
1878 trace_source_return_path_thread_loop_top();
1879 header_type = qemu_get_be16(rp);
1880 header_len = qemu_get_be16(rp);
1882 if (qemu_file_get_error(rp)) {
1883 mark_source_rp_bad(ms);
1887 if (header_type >= MIG_RP_MSG_MAX ||
1888 header_type == MIG_RP_MSG_INVALID) {
1889 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1890 header_type, header_len);
1891 mark_source_rp_bad(ms);
1895 if ((rp_cmd_args[header_type].len != -1 &&
1896 header_len != rp_cmd_args[header_type].len) ||
1897 header_len > sizeof(buf)) {
1898 error_report("RP: Received '%s' message (0x%04x) with"
1899 "incorrect length %d expecting %zu",
1900 rp_cmd_args[header_type].name, header_type, header_len,
1901 (size_t)rp_cmd_args[header_type].len);
1902 mark_source_rp_bad(ms);
1906 /* We know we've got a valid header by this point */
1907 res = qemu_get_buffer(rp, buf, header_len);
1908 if (res != header_len) {
1909 error_report("RP: Failed reading data for message 0x%04x"
1910 " read %d expected %d",
1911 header_type, res, header_len);
1912 mark_source_rp_bad(ms);
1916 /* OK, we have the message and the data */
1917 switch (header_type) {
1918 case MIG_RP_MSG_SHUT:
1919 sibling_error = ldl_be_p(buf);
1920 trace_source_return_path_thread_shut(sibling_error);
1921 if (sibling_error) {
1922 error_report("RP: Sibling indicated error %d", sibling_error);
1923 mark_source_rp_bad(ms);
1926 * We'll let the main thread deal with closing the RP
1927 * we could do a shutdown(2) on it, but we're the only user
1928 * anyway, so there's nothing gained.
1932 case MIG_RP_MSG_PONG:
1933 tmp32 = ldl_be_p(buf);
1934 trace_source_return_path_thread_pong(tmp32);
1935 qemu_sem_post(&ms->rp_state.rp_pong_acks);
1938 case MIG_RP_MSG_REQ_PAGES:
1939 start = ldq_be_p(buf);
1940 len = ldl_be_p(buf + 8);
1941 migrate_handle_rp_req_pages(ms, NULL, start, len);
1944 case MIG_RP_MSG_REQ_PAGES_ID:
1945 expected_len = 12 + 1; /* header + termination */
1947 if (header_len >= expected_len) {
1948 start = ldq_be_p(buf);
1949 len = ldl_be_p(buf + 8);
1950 /* Now we expect an idstr */
1951 tmp32 = buf[12]; /* Length of the following idstr */
1952 buf[13 + tmp32] = '\0';
1953 expected_len += tmp32;
1955 if (header_len != expected_len) {
1956 error_report("RP: Req_Page_id with length %d expecting %zd",
1957 header_len, expected_len);
1958 mark_source_rp_bad(ms);
1961 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1964 case MIG_RP_MSG_RECV_BITMAP:
1965 if (header_len < 1) {
1966 error_report("%s: missing block name", __func__);
1967 mark_source_rp_bad(ms);
1970 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
1971 buf[buf[0] + 1] = '\0';
1972 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
1973 mark_source_rp_bad(ms);
1978 case MIG_RP_MSG_RESUME_ACK:
1979 tmp32 = ldl_be_p(buf);
1980 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
1981 mark_source_rp_bad(ms);
1986 case MIG_RP_MSG_SWITCHOVER_ACK:
1987 ms->switchover_acked = true;
1988 trace_source_return_path_thread_switchover_acked();
1997 res = qemu_file_get_error(rp);
1999 if (res && migration_in_postcopy()) {
2001 * Maybe there is something we can do: it looks like a
2002 * network down issue, and we pause for a recovery.
2004 migration_release_dst_files(ms);
2006 if (postcopy_pause_return_path_thread(ms)) {
2008 * Reload rp, reset the rest. Referencing it is safe since
2009 * it's reset only by us above, or when migration completes
2011 rp = ms->rp_state.from_dst_file;
2012 ms->rp_state.error = false;
2017 trace_source_return_path_thread_bad_end();
2018 mark_source_rp_bad(ms);
2021 trace_source_return_path_thread_end();
2022 migration_release_dst_files(ms);
2023 rcu_unregister_thread();
2027 static int open_return_path_on_source(MigrationState *ms,
2030 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2031 if (!ms->rp_state.from_dst_file) {
2035 trace_open_return_path_on_source();
2037 if (!create_thread) {
2042 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2043 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2044 ms->rp_state.rp_thread_created = true;
2046 trace_open_return_path_on_source_continue();
2051 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2052 static int await_return_path_close_on_source(MigrationState *ms)
2055 * If this is a normal exit then the destination will send a SHUT and the
2056 * rp_thread will exit, however if there's an error we need to cause
2059 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2061 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2062 * waiting for the destination.
2064 qemu_file_shutdown(ms->rp_state.from_dst_file);
2065 mark_source_rp_bad(ms);
2067 trace_await_return_path_close_on_source_joining();
2068 qemu_thread_join(&ms->rp_state.rp_thread);
2069 ms->rp_state.rp_thread_created = false;
2070 trace_await_return_path_close_on_source_close();
2071 return ms->rp_state.error;
2075 migration_wait_main_channel(MigrationState *ms)
2077 /* Wait until one PONG message received */
2078 qemu_sem_wait(&ms->rp_state.rp_pong_acks);
2082 * Switch from normal iteration to postcopy
2083 * Returns non-0 on error
2085 static int postcopy_start(MigrationState *ms, Error **errp)
2088 QIOChannelBuffer *bioc;
2090 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2091 uint64_t bandwidth = migrate_max_postcopy_bandwidth();
2092 bool restart_block = false;
2093 int cur_state = MIGRATION_STATUS_ACTIVE;
2095 if (migrate_postcopy_preempt()) {
2096 migration_wait_main_channel(ms);
2097 if (postcopy_preempt_establish_channel(ms)) {
2098 migrate_set_state(&ms->state, ms->state, MIGRATION_STATUS_FAILED);
2103 if (!migrate_pause_before_switchover()) {
2104 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2105 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2108 trace_postcopy_start();
2109 qemu_mutex_lock_iothread();
2110 trace_postcopy_start_set_run();
2112 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2113 global_state_store();
2114 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2119 ret = migration_maybe_pause(ms, &cur_state,
2120 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2125 ret = bdrv_inactivate_all();
2129 restart_block = true;
2132 * Cause any non-postcopiable, but iterative devices to
2133 * send out their final data.
2135 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2138 * in Finish migrate and with the io-lock held everything should
2139 * be quiet, but we've potentially still got dirty pages and we
2140 * need to tell the destination to throw any pages it's already received
2143 if (migrate_postcopy_ram()) {
2144 ram_postcopy_send_discard_bitmap(ms);
2148 * send rest of state - note things that are doing postcopy
2149 * will notice we're in POSTCOPY_ACTIVE and not actually
2150 * wrap their state up here
2152 migration_rate_set(bandwidth);
2153 if (migrate_postcopy_ram()) {
2154 /* Ping just for debugging, helps line traces up */
2155 qemu_savevm_send_ping(ms->to_dst_file, 2);
2159 * While loading the device state we may trigger page transfer
2160 * requests and the fd must be free to process those, and thus
2161 * the destination must read the whole device state off the fd before
2162 * it starts processing it. Unfortunately the ad-hoc migration format
2163 * doesn't allow the destination to know the size to read without fully
2164 * parsing it through each devices load-state code (especially the open
2165 * coded devices that use get/put).
2166 * So we wrap the device state up in a package with a length at the start;
2167 * to do this we use a qemu_buf to hold the whole of the device state.
2169 bioc = qio_channel_buffer_new(4096);
2170 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2171 fb = qemu_file_new_output(QIO_CHANNEL(bioc));
2172 object_unref(OBJECT(bioc));
2175 * Make sure the receiver can get incoming pages before we send the rest
2178 qemu_savevm_send_postcopy_listen(fb);
2180 qemu_savevm_state_complete_precopy(fb, false, false);
2181 if (migrate_postcopy_ram()) {
2182 qemu_savevm_send_ping(fb, 3);
2185 qemu_savevm_send_postcopy_run(fb);
2187 /* <><> end of stuff going into the package */
2189 /* Last point of recovery; as soon as we send the package the destination
2190 * can open devices and potentially start running.
2191 * Lets just check again we've not got any errors.
2193 ret = qemu_file_get_error(ms->to_dst_file);
2195 error_setg(errp, "postcopy_start: Migration stream errored (pre package)");
2199 restart_block = false;
2201 /* Now send that blob */
2202 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2207 /* Send a notify to give a chance for anything that needs to happen
2208 * at the transition to postcopy and after the device state; in particular
2209 * spice needs to trigger a transition now
2211 ms->postcopy_after_devices = true;
2212 notifier_list_notify(&migration_state_notifiers, ms);
2214 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2216 qemu_mutex_unlock_iothread();
2218 if (migrate_postcopy_ram()) {
2220 * Although this ping is just for debug, it could potentially be
2221 * used for getting a better measurement of downtime at the source.
2223 qemu_savevm_send_ping(ms->to_dst_file, 4);
2226 if (migrate_release_ram()) {
2227 ram_postcopy_migrated_memory_release(ms);
2230 ret = qemu_file_get_error(ms->to_dst_file);
2232 error_setg(errp, "postcopy_start: Migration stream errored");
2233 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2234 MIGRATION_STATUS_FAILED);
2237 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
2244 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2245 MIGRATION_STATUS_FAILED);
2246 if (restart_block) {
2247 /* A failure happened early enough that we know the destination hasn't
2248 * accessed block devices, so we're safe to recover.
2250 Error *local_err = NULL;
2252 bdrv_activate_all(&local_err);
2254 error_report_err(local_err);
2257 qemu_mutex_unlock_iothread();
2262 * migration_maybe_pause: Pause if required to by
2263 * migrate_pause_before_switchover called with the iothread locked
2264 * Returns: 0 on success
2266 static int migration_maybe_pause(MigrationState *s,
2267 int *current_active_state,
2270 if (!migrate_pause_before_switchover()) {
2274 /* Since leaving this state is not atomic with posting the semaphore
2275 * it's possible that someone could have issued multiple migrate_continue
2276 * and the semaphore is incorrectly positive at this point;
2277 * the docs say it's undefined to reinit a semaphore that's already
2278 * init'd, so use timedwait to eat up any existing posts.
2280 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2281 /* This block intentionally left blank */
2285 * If the migration is cancelled when it is in the completion phase,
2286 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2287 * So we don't need to wait a semaphore, otherwise we would always
2288 * wait for the 'pause_sem' semaphore.
2290 if (s->state != MIGRATION_STATUS_CANCELLING) {
2291 qemu_mutex_unlock_iothread();
2292 migrate_set_state(&s->state, *current_active_state,
2293 MIGRATION_STATUS_PRE_SWITCHOVER);
2294 qemu_sem_wait(&s->pause_sem);
2295 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2297 *current_active_state = new_state;
2298 qemu_mutex_lock_iothread();
2301 return s->state == new_state ? 0 : -EINVAL;
2305 * migration_completion: Used by migration_thread when there's not much left.
2306 * The caller 'breaks' the loop when this returns.
2308 * @s: Current migration state
2310 static void migration_completion(MigrationState *s)
2313 int current_active_state = s->state;
2315 if (s->state == MIGRATION_STATUS_ACTIVE) {
2316 qemu_mutex_lock_iothread();
2317 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2318 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2320 s->vm_old_state = runstate_get();
2321 global_state_store();
2323 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2324 trace_migration_completion_vm_stop(ret);
2326 ret = migration_maybe_pause(s, ¤t_active_state,
2327 MIGRATION_STATUS_DEVICE);
2331 * Inactivate disks except in COLO, and track that we
2332 * have done so in order to remember to reactivate
2333 * them if migration fails or is cancelled.
2335 s->block_inactive = !migrate_colo();
2336 migration_rate_set(RATE_LIMIT_DISABLED);
2337 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2341 qemu_mutex_unlock_iothread();
2346 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2347 trace_migration_completion_postcopy_end();
2349 qemu_mutex_lock_iothread();
2350 qemu_savevm_state_complete_postcopy(s->to_dst_file);
2351 qemu_mutex_unlock_iothread();
2354 * Shutdown the postcopy fast path thread. This is only needed
2355 * when dest QEMU binary is old (7.1/7.2). QEMU 8.0+ doesn't need
2358 if (migrate_postcopy_preempt() && s->preempt_pre_7_2) {
2359 postcopy_preempt_shutdown_file(s);
2362 trace_migration_completion_postcopy_end_after_complete();
2368 * If rp was opened we must clean up the thread before
2369 * cleaning everything else up (since if there are no failures
2370 * it will wait for the destination to send it's status in
2373 if (s->rp_state.rp_thread_created) {
2375 trace_migration_return_path_end_before();
2376 rp_error = await_return_path_close_on_source(s);
2377 trace_migration_return_path_end_after(rp_error);
2383 if (qemu_file_get_error(s->to_dst_file)) {
2384 trace_migration_completion_file_err();
2388 if (migrate_colo() && s->state == MIGRATION_STATUS_ACTIVE) {
2389 /* COLO does not support postcopy */
2390 migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
2391 MIGRATION_STATUS_COLO);
2393 migrate_set_state(&s->state, current_active_state,
2394 MIGRATION_STATUS_COMPLETED);
2400 if (s->block_inactive && (s->state == MIGRATION_STATUS_ACTIVE ||
2401 s->state == MIGRATION_STATUS_DEVICE)) {
2403 * If not doing postcopy, vm_start() will be called: let's
2404 * regain control on images.
2406 Error *local_err = NULL;
2408 qemu_mutex_lock_iothread();
2409 bdrv_activate_all(&local_err);
2411 error_report_err(local_err);
2413 s->block_inactive = false;
2415 qemu_mutex_unlock_iothread();
2418 migrate_set_state(&s->state, current_active_state,
2419 MIGRATION_STATUS_FAILED);
2423 * bg_migration_completion: Used by bg_migration_thread when after all the
2424 * RAM has been saved. The caller 'breaks' the loop when this returns.
2426 * @s: Current migration state
2428 static void bg_migration_completion(MigrationState *s)
2430 int current_active_state = s->state;
2432 if (s->state == MIGRATION_STATUS_ACTIVE) {
2434 * By this moment we have RAM content saved into the migration stream.
2435 * The next step is to flush the non-RAM content (device state)
2436 * right after the ram content. The device state has been stored into
2437 * the temporary buffer before RAM saving started.
2439 qemu_put_buffer(s->to_dst_file, s->bioc->data, s->bioc->usage);
2440 qemu_fflush(s->to_dst_file);
2441 } else if (s->state == MIGRATION_STATUS_CANCELLING) {
2445 if (qemu_file_get_error(s->to_dst_file)) {
2446 trace_migration_completion_file_err();
2450 migrate_set_state(&s->state, current_active_state,
2451 MIGRATION_STATUS_COMPLETED);
2455 migrate_set_state(&s->state, current_active_state,
2456 MIGRATION_STATUS_FAILED);
2459 typedef enum MigThrError {
2460 /* No error detected */
2461 MIG_THR_ERR_NONE = 0,
2462 /* Detected error, but resumed successfully */
2463 MIG_THR_ERR_RECOVERED = 1,
2464 /* Detected fatal error, need to exit */
2465 MIG_THR_ERR_FATAL = 2,
2468 static int postcopy_resume_handshake(MigrationState *s)
2470 qemu_savevm_send_postcopy_resume(s->to_dst_file);
2472 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2473 qemu_sem_wait(&s->rp_state.rp_sem);
2476 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2483 /* Return zero if success, or <0 for error */
2484 static int postcopy_do_resume(MigrationState *s)
2489 * Call all the resume_prepare() hooks, so that modules can be
2490 * ready for the migration resume.
2492 ret = qemu_savevm_state_resume_prepare(s);
2494 error_report("%s: resume_prepare() failure detected: %d",
2500 * If preempt is enabled, re-establish the preempt channel. Note that
2501 * we do it after resume prepare to make sure the main channel will be
2502 * created before the preempt channel. E.g. with weak network, the
2503 * dest QEMU may get messed up with the preempt and main channels on
2504 * the order of connection setup. This guarantees the correct order.
2506 ret = postcopy_preempt_establish_channel(s);
2508 error_report("%s: postcopy_preempt_establish_channel(): %d",
2514 * Last handshake with destination on the resume (destination will
2515 * switch to postcopy-active afterwards)
2517 ret = postcopy_resume_handshake(s);
2519 error_report("%s: handshake failed: %d", __func__, ret);
2527 * We don't return until we are in a safe state to continue current
2528 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2529 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2531 static MigThrError postcopy_pause(MigrationState *s)
2533 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2539 * Current channel is possibly broken. Release it. Note that this is
2540 * guaranteed even without lock because to_dst_file should only be
2541 * modified by the migration thread. That also guarantees that the
2542 * unregister of yank is safe too without the lock. It should be safe
2543 * even to be within the qemu_file_lock, but we didn't do that to avoid
2544 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
2545 * the qemu_file_lock critical section as small as possible.
2547 assert(s->to_dst_file);
2548 migration_ioc_unregister_yank_from_file(s->to_dst_file);
2549 qemu_mutex_lock(&s->qemu_file_lock);
2550 file = s->to_dst_file;
2551 s->to_dst_file = NULL;
2552 qemu_mutex_unlock(&s->qemu_file_lock);
2554 qemu_file_shutdown(file);
2557 migrate_set_state(&s->state, s->state,
2558 MIGRATION_STATUS_POSTCOPY_PAUSED);
2560 error_report("Detected IO failure for postcopy. "
2561 "Migration paused.");
2564 * We wait until things fixed up. Then someone will setup the
2565 * status back for us.
2567 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2568 qemu_sem_wait(&s->postcopy_pause_sem);
2571 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2572 /* Woken up by a recover procedure. Give it a shot */
2575 * Firstly, let's wake up the return path now, with a new
2576 * return path channel.
2578 qemu_sem_post(&s->postcopy_pause_rp_sem);
2580 /* Do the resume logic */
2581 if (postcopy_do_resume(s) == 0) {
2582 /* Let's continue! */
2583 trace_postcopy_pause_continued();
2584 return MIG_THR_ERR_RECOVERED;
2587 * Something wrong happened during the recovery, let's
2588 * pause again. Pause is always better than throwing
2594 /* This is not right... Time to quit. */
2595 return MIG_THR_ERR_FATAL;
2600 static MigThrError migration_detect_error(MigrationState *s)
2603 int state = s->state;
2604 Error *local_error = NULL;
2606 if (state == MIGRATION_STATUS_CANCELLING ||
2607 state == MIGRATION_STATUS_CANCELLED) {
2608 /* End the migration, but don't set the state to failed */
2609 return MIG_THR_ERR_FATAL;
2613 * Try to detect any file errors. Note that postcopy_qemufile_src will
2614 * be NULL when postcopy preempt is not enabled.
2616 ret = qemu_file_get_error_obj_any(s->to_dst_file,
2617 s->postcopy_qemufile_src,
2620 /* Everything is fine */
2621 assert(!local_error);
2622 return MIG_THR_ERR_NONE;
2626 migrate_set_error(s, local_error);
2627 error_free(local_error);
2630 if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret) {
2632 * For postcopy, we allow the network to be down for a
2633 * while. After that, it can be continued by a
2636 return postcopy_pause(s);
2639 * For precopy (or postcopy with error outside IO), we fail
2642 migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
2643 trace_migration_thread_file_err();
2645 /* Time to stop the migration, now. */
2646 return MIG_THR_ERR_FATAL;
2650 static void migration_calculate_complete(MigrationState *s)
2652 uint64_t bytes = migration_transferred_bytes(s->to_dst_file);
2653 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2654 int64_t transfer_time;
2656 s->total_time = end_time - s->start_time;
2659 * It's still not set, so we are precopy migration. For
2660 * postcopy, downtime is calculated during postcopy_start().
2662 s->downtime = end_time - s->downtime_start;
2665 transfer_time = s->total_time - s->setup_time;
2666 if (transfer_time) {
2667 s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
2671 static void update_iteration_initial_status(MigrationState *s)
2674 * Update these three fields at the same time to avoid mismatch info lead
2675 * wrong speed calculation.
2677 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2678 s->iteration_initial_bytes = migration_transferred_bytes(s->to_dst_file);
2679 s->iteration_initial_pages = ram_get_total_transferred_pages();
2682 static void migration_update_counters(MigrationState *s,
2683 int64_t current_time)
2685 uint64_t transferred, transferred_pages, time_spent;
2686 uint64_t current_bytes; /* bytes transferred since the beginning */
2689 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
2693 current_bytes = migration_transferred_bytes(s->to_dst_file);
2694 transferred = current_bytes - s->iteration_initial_bytes;
2695 time_spent = current_time - s->iteration_start_time;
2696 bandwidth = (double)transferred / time_spent;
2697 s->threshold_size = bandwidth * migrate_downtime_limit();
2699 s->mbps = (((double) transferred * 8.0) /
2700 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2702 transferred_pages = ram_get_total_transferred_pages() -
2703 s->iteration_initial_pages;
2704 s->pages_per_second = (double) transferred_pages /
2705 (((double) time_spent / 1000.0));
2708 * if we haven't sent anything, we don't want to
2709 * recalculate. 10000 is a small enough number for our purposes
2711 if (stat64_get(&mig_stats.dirty_pages_rate) &&
2712 transferred > 10000) {
2713 s->expected_downtime =
2714 stat64_get(&mig_stats.dirty_bytes_last_sync) / bandwidth;
2717 migration_rate_reset(s->to_dst_file);
2719 update_iteration_initial_status(s);
2721 trace_migrate_transferred(transferred, time_spent,
2722 bandwidth, s->threshold_size);
2725 static bool migration_can_switchover(MigrationState *s)
2727 if (!migrate_switchover_ack()) {
2731 /* No reason to wait for switchover ACK if VM is stopped */
2732 if (!runstate_is_running()) {
2736 return s->switchover_acked;
2739 /* Migration thread iteration status */
2741 MIG_ITERATE_RESUME, /* Resume current iteration */
2742 MIG_ITERATE_SKIP, /* Skip current iteration */
2743 MIG_ITERATE_BREAK, /* Break the loop */
2747 * Return true if continue to the next iteration directly, false
2750 static MigIterateState migration_iteration_run(MigrationState *s)
2752 uint64_t must_precopy, can_postcopy;
2753 Error *local_err = NULL;
2754 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
2755 bool can_switchover = migration_can_switchover(s);
2757 qemu_savevm_state_pending_estimate(&must_precopy, &can_postcopy);
2758 uint64_t pending_size = must_precopy + can_postcopy;
2760 trace_migrate_pending_estimate(pending_size, must_precopy, can_postcopy);
2762 if (must_precopy <= s->threshold_size) {
2763 qemu_savevm_state_pending_exact(&must_precopy, &can_postcopy);
2764 pending_size = must_precopy + can_postcopy;
2765 trace_migrate_pending_exact(pending_size, must_precopy, can_postcopy);
2768 if ((!pending_size || pending_size < s->threshold_size) && can_switchover) {
2769 trace_migration_thread_low_pending(pending_size);
2770 migration_completion(s);
2771 return MIG_ITERATE_BREAK;
2774 /* Still a significant amount to transfer */
2775 if (!in_postcopy && must_precopy <= s->threshold_size && can_switchover &&
2776 qatomic_read(&s->start_postcopy)) {
2777 if (postcopy_start(s, &local_err)) {
2778 migrate_set_error(s, local_err);
2779 error_report_err(local_err);
2781 return MIG_ITERATE_SKIP;
2784 /* Just another iteration step */
2785 qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
2786 return MIG_ITERATE_RESUME;
2789 static void migration_iteration_finish(MigrationState *s)
2791 /* If we enabled cpu throttling for auto-converge, turn it off. */
2792 cpu_throttle_stop();
2794 qemu_mutex_lock_iothread();
2796 case MIGRATION_STATUS_COMPLETED:
2797 migration_calculate_complete(s);
2798 runstate_set(RUN_STATE_POSTMIGRATE);
2800 case MIGRATION_STATUS_COLO:
2801 assert(migrate_colo());
2802 migrate_start_colo_process(s);
2803 s->vm_old_state = RUN_STATE_RUNNING;
2805 case MIGRATION_STATUS_FAILED:
2806 case MIGRATION_STATUS_CANCELLED:
2807 case MIGRATION_STATUS_CANCELLING:
2808 if (s->vm_old_state == RUN_STATE_RUNNING) {
2809 if (!runstate_check(RUN_STATE_SHUTDOWN)) {
2813 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2814 runstate_set(s->vm_old_state);
2820 /* Should not reach here, but if so, forgive the VM. */
2821 error_report("%s: Unknown ending state %d", __func__, s->state);
2824 migrate_fd_cleanup_schedule(s);
2825 qemu_mutex_unlock_iothread();
2828 static void bg_migration_iteration_finish(MigrationState *s)
2831 * Stop tracking RAM writes - un-protect memory, un-register UFFD
2832 * memory ranges, flush kernel wait queues and wake up threads
2833 * waiting for write fault to be resolved.
2835 ram_write_tracking_stop();
2837 qemu_mutex_lock_iothread();
2839 case MIGRATION_STATUS_COMPLETED:
2840 migration_calculate_complete(s);
2843 case MIGRATION_STATUS_ACTIVE:
2844 case MIGRATION_STATUS_FAILED:
2845 case MIGRATION_STATUS_CANCELLED:
2846 case MIGRATION_STATUS_CANCELLING:
2850 /* Should not reach here, but if so, forgive the VM. */
2851 error_report("%s: Unknown ending state %d", __func__, s->state);
2855 migrate_fd_cleanup_schedule(s);
2856 qemu_mutex_unlock_iothread();
2860 * Return true if continue to the next iteration directly, false
2863 static MigIterateState bg_migration_iteration_run(MigrationState *s)
2867 res = qemu_savevm_state_iterate(s->to_dst_file, false);
2869 bg_migration_completion(s);
2870 return MIG_ITERATE_BREAK;
2873 return MIG_ITERATE_RESUME;
2876 void migration_make_urgent_request(void)
2878 qemu_sem_post(&migrate_get_current()->rate_limit_sem);
2881 void migration_consume_urgent_request(void)
2883 qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
2886 /* Returns true if the rate limiting was broken by an urgent request */
2887 bool migration_rate_limit(void)
2889 int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2890 MigrationState *s = migrate_get_current();
2892 bool urgent = false;
2893 migration_update_counters(s, now);
2894 if (migration_rate_exceeded(s->to_dst_file)) {
2896 if (qemu_file_get_error(s->to_dst_file)) {
2900 * Wait for a delay to do rate limiting OR
2901 * something urgent to post the semaphore.
2903 int ms = s->iteration_start_time + BUFFER_DELAY - now;
2904 trace_migration_rate_limit_pre(ms);
2905 if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
2907 * We were woken by one or more urgent things but
2908 * the timedwait will have consumed one of them.
2909 * The service routine for the urgent wake will dec
2910 * the semaphore itself for each item it consumes,
2911 * so add this one we just eat back.
2913 qemu_sem_post(&s->rate_limit_sem);
2916 trace_migration_rate_limit_post(urgent);
2922 * if failover devices are present, wait they are completely
2926 static void qemu_savevm_wait_unplug(MigrationState *s, int old_state,
2929 if (qemu_savevm_state_guest_unplug_pending()) {
2930 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_WAIT_UNPLUG);
2932 while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
2933 qemu_savevm_state_guest_unplug_pending()) {
2934 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
2936 if (s->state != MIGRATION_STATUS_WAIT_UNPLUG) {
2937 int timeout = 120; /* 30 seconds */
2939 * migration has been canceled
2940 * but as we have started an unplug we must wait the end
2941 * to be able to plug back the card
2943 while (timeout-- && qemu_savevm_state_guest_unplug_pending()) {
2944 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
2946 if (qemu_savevm_state_guest_unplug_pending() &&
2948 warn_report("migration: partially unplugged device on "
2953 migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG, new_state);
2955 migrate_set_state(&s->state, old_state, new_state);
2960 * Master migration thread on the source VM.
2961 * It drives the migration and pumps the data down the outgoing channel.
2963 static void *migration_thread(void *opaque)
2965 MigrationState *s = opaque;
2966 MigrationThread *thread = NULL;
2967 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
2968 MigThrError thr_error;
2969 bool urgent = false;
2971 thread = migration_threads_add("live_migration", qemu_get_thread_id());
2973 rcu_register_thread();
2975 object_ref(OBJECT(s));
2976 update_iteration_initial_status(s);
2978 qemu_savevm_state_header(s->to_dst_file);
2981 * If we opened the return path, we need to make sure dst has it
2984 if (s->rp_state.rp_thread_created) {
2985 /* Now tell the dest that it should open its end so it can reply */
2986 qemu_savevm_send_open_return_path(s->to_dst_file);
2988 /* And do a ping that will make stuff easier to debug */
2989 qemu_savevm_send_ping(s->to_dst_file, 1);
2992 if (migrate_postcopy()) {
2994 * Tell the destination that we *might* want to do postcopy later;
2995 * if the other end can't do postcopy it should fail now, nice and
2998 qemu_savevm_send_postcopy_advise(s->to_dst_file);
3001 if (migrate_colo()) {
3002 /* Notify migration destination that we enable COLO */
3003 qemu_savevm_send_colo_enable(s->to_dst_file);
3006 qemu_savevm_state_setup(s->to_dst_file);
3008 qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP,
3009 MIGRATION_STATUS_ACTIVE);
3011 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3013 trace_migration_thread_setup_complete();
3015 while (migration_is_active(s)) {
3016 if (urgent || !migration_rate_exceeded(s->to_dst_file)) {
3017 MigIterateState iter_state = migration_iteration_run(s);
3018 if (iter_state == MIG_ITERATE_SKIP) {
3020 } else if (iter_state == MIG_ITERATE_BREAK) {
3026 * Try to detect any kind of failures, and see whether we
3027 * should stop the migration now.
3029 thr_error = migration_detect_error(s);
3030 if (thr_error == MIG_THR_ERR_FATAL) {
3031 /* Stop migration */
3033 } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3035 * Just recovered from a e.g. network failure, reset all
3036 * the local variables. This is important to avoid
3037 * breaking transferred_bytes and bandwidth calculation
3039 update_iteration_initial_status(s);
3042 urgent = migration_rate_limit();
3045 trace_migration_thread_after_loop();
3046 migration_iteration_finish(s);
3047 object_unref(OBJECT(s));
3048 rcu_unregister_thread();
3049 migration_threads_remove(thread);
3053 static void bg_migration_vm_start_bh(void *opaque)
3055 MigrationState *s = opaque;
3057 qemu_bh_delete(s->vm_start_bh);
3058 s->vm_start_bh = NULL;
3061 s->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - s->downtime_start;
3065 * Background snapshot thread, based on live migration code.
3066 * This is an alternative implementation of live migration mechanism
3067 * introduced specifically to support background snapshots.
3069 * It takes advantage of userfault_fd write protection mechanism introduced
3070 * in v5.7 kernel. Compared to existing dirty page logging migration much
3071 * lesser stream traffic is produced resulting in smaller snapshot images,
3072 * simply cause of no page duplicates can get into the stream.
3074 * Another key point is that generated vmstate stream reflects machine state
3075 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
3076 * mechanism, which effectively results in that saved snapshot is the state of VM
3077 * at the end of the process.
3079 static void *bg_migration_thread(void *opaque)
3081 MigrationState *s = opaque;
3082 int64_t setup_start;
3083 MigThrError thr_error;
3085 bool early_fail = true;
3087 rcu_register_thread();
3088 object_ref(OBJECT(s));
3090 migration_rate_set(RATE_LIMIT_DISABLED);
3092 setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3094 * We want to save vmstate for the moment when migration has been
3095 * initiated but also we want to save RAM content while VM is running.
3096 * The RAM content should appear first in the vmstate. So, we first
3097 * stash the non-RAM part of the vmstate to the temporary buffer,
3098 * then write RAM part of the vmstate to the migration stream
3099 * with vCPUs running and, finally, write stashed non-RAM part of
3100 * the vmstate from the buffer to the migration stream.
3102 s->bioc = qio_channel_buffer_new(512 * 1024);
3103 qio_channel_set_name(QIO_CHANNEL(s->bioc), "vmstate-buffer");
3104 fb = qemu_file_new_output(QIO_CHANNEL(s->bioc));
3105 object_unref(OBJECT(s->bioc));
3107 update_iteration_initial_status(s);
3110 * Prepare for tracking memory writes with UFFD-WP - populate
3111 * RAM pages before protecting.
3114 ram_write_tracking_prepare();
3117 qemu_savevm_state_header(s->to_dst_file);
3118 qemu_savevm_state_setup(s->to_dst_file);
3120 qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP,
3121 MIGRATION_STATUS_ACTIVE);
3123 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3125 trace_migration_thread_setup_complete();
3126 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3128 qemu_mutex_lock_iothread();
3131 * If VM is currently in suspended state, then, to make a valid runstate
3132 * transition in vm_stop_force_state() we need to wakeup it up.
3134 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
3135 s->vm_old_state = runstate_get();
3137 global_state_store();
3138 /* Forcibly stop VM before saving state of vCPUs and devices */
3139 if (vm_stop_force_state(RUN_STATE_PAUSED)) {
3143 * Put vCPUs in sync with shadow context structures, then
3144 * save their state to channel-buffer along with devices.
3146 cpu_synchronize_all_states();
3147 if (qemu_savevm_state_complete_precopy_non_iterable(fb, false, false)) {
3151 * Since we are going to get non-iterable state data directly
3152 * from s->bioc->data, explicit flush is needed here.
3156 /* Now initialize UFFD context and start tracking RAM writes */
3157 if (ram_write_tracking_start()) {
3163 * Start VM from BH handler to avoid write-fault lock here.
3164 * UFFD-WP protection for the whole RAM is already enabled so
3165 * calling VM state change notifiers from vm_start() would initiate
3166 * writes to virtio VQs memory which is in write-protected region.
3168 s->vm_start_bh = qemu_bh_new(bg_migration_vm_start_bh, s);
3169 qemu_bh_schedule(s->vm_start_bh);
3171 qemu_mutex_unlock_iothread();
3173 while (migration_is_active(s)) {
3174 MigIterateState iter_state = bg_migration_iteration_run(s);
3175 if (iter_state == MIG_ITERATE_SKIP) {
3177 } else if (iter_state == MIG_ITERATE_BREAK) {
3182 * Try to detect any kind of failures, and see whether we
3183 * should stop the migration now.
3185 thr_error = migration_detect_error(s);
3186 if (thr_error == MIG_THR_ERR_FATAL) {
3187 /* Stop migration */
3191 migration_update_counters(s, qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
3194 trace_migration_thread_after_loop();
3198 migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
3199 MIGRATION_STATUS_FAILED);
3200 qemu_mutex_unlock_iothread();
3203 bg_migration_iteration_finish(s);
3206 object_unref(OBJECT(s));
3207 rcu_unregister_thread();
3212 void migrate_fd_connect(MigrationState *s, Error *error_in)
3214 Error *local_err = NULL;
3215 uint64_t rate_limit;
3216 bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
3219 * If there's a previous error, free it and prepare for another one.
3220 * Meanwhile if migration completes successfully, there won't have an error
3221 * dumped when calling migrate_fd_cleanup().
3223 migrate_error_free(s);
3225 s->expected_downtime = migrate_downtime_limit();
3227 assert(s->cleanup_bh);
3229 assert(!s->cleanup_bh);
3230 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s);
3233 migrate_fd_error(s, error_in);
3236 * Don't do cleanup for resume if channel is invalid, but only dump
3237 * the error. We wait for another channel connect from the user.
3238 * The error_report still gives HMP user a hint on what failed.
3239 * It's normally done in migrate_fd_cleanup(), but call it here
3242 error_report_err(error_copy(s->error));
3244 migrate_fd_cleanup(s);
3250 /* This is a resumed migration */
3251 rate_limit = migrate_max_postcopy_bandwidth();
3253 /* This is a fresh new migration */
3254 rate_limit = migrate_max_bandwidth();
3256 /* Notify before starting migration thread */
3257 notifier_list_notify(&migration_state_notifiers, s);
3260 migration_rate_set(rate_limit);
3261 qemu_file_set_blocking(s->to_dst_file, true);
3264 * Open the return path. For postcopy, it is used exclusively. For
3265 * precopy, only if user specified "return-path" capability would
3266 * QEMU uses the return path.
3268 if (migrate_postcopy_ram() || migrate_return_path()) {
3269 if (open_return_path_on_source(s, !resume)) {
3270 error_setg(&local_err, "Unable to open return-path for postcopy");
3271 migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3272 migrate_set_error(s, local_err);
3273 error_report_err(local_err);
3274 migrate_fd_cleanup(s);
3280 * This needs to be done before resuming a postcopy. Note: for newer
3281 * QEMUs we will delay the channel creation until postcopy_start(), to
3282 * avoid disorder of channel creations.
3284 if (migrate_postcopy_preempt() && s->preempt_pre_7_2) {
3285 postcopy_preempt_setup(s);
3289 /* Wakeup the main migration thread to do the recovery */
3290 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3291 MIGRATION_STATUS_POSTCOPY_RECOVER);
3292 qemu_sem_post(&s->postcopy_pause_sem);
3296 if (multifd_save_setup(&local_err) != 0) {
3297 migrate_set_error(s, local_err);
3298 error_report_err(local_err);
3299 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3300 MIGRATION_STATUS_FAILED);
3301 migrate_fd_cleanup(s);
3305 if (migrate_background_snapshot()) {
3306 qemu_thread_create(&s->thread, "bg_snapshot",
3307 bg_migration_thread, s, QEMU_THREAD_JOINABLE);
3309 qemu_thread_create(&s->thread, "live_migration",
3310 migration_thread, s, QEMU_THREAD_JOINABLE);
3312 s->migration_thread_running = true;
3315 static void migration_class_init(ObjectClass *klass, void *data)
3317 DeviceClass *dc = DEVICE_CLASS(klass);
3319 dc->user_creatable = false;
3320 device_class_set_props(dc, migration_properties);
3323 static void migration_instance_finalize(Object *obj)
3325 MigrationState *ms = MIGRATION_OBJ(obj);
3327 qemu_mutex_destroy(&ms->error_mutex);
3328 qemu_mutex_destroy(&ms->qemu_file_lock);
3329 qemu_sem_destroy(&ms->wait_unplug_sem);
3330 qemu_sem_destroy(&ms->rate_limit_sem);
3331 qemu_sem_destroy(&ms->pause_sem);
3332 qemu_sem_destroy(&ms->postcopy_pause_sem);
3333 qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
3334 qemu_sem_destroy(&ms->rp_state.rp_sem);
3335 qemu_sem_destroy(&ms->rp_state.rp_pong_acks);
3336 qemu_sem_destroy(&ms->postcopy_qemufile_src_sem);
3337 error_free(ms->error);
3340 static void migration_instance_init(Object *obj)
3342 MigrationState *ms = MIGRATION_OBJ(obj);
3344 ms->state = MIGRATION_STATUS_NONE;
3346 ms->pages_per_second = -1;
3347 qemu_sem_init(&ms->pause_sem, 0);
3348 qemu_mutex_init(&ms->error_mutex);
3350 migrate_params_init(&ms->parameters);
3352 qemu_sem_init(&ms->postcopy_pause_sem, 0);
3353 qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
3354 qemu_sem_init(&ms->rp_state.rp_sem, 0);
3355 qemu_sem_init(&ms->rp_state.rp_pong_acks, 0);
3356 qemu_sem_init(&ms->rate_limit_sem, 0);
3357 qemu_sem_init(&ms->wait_unplug_sem, 0);
3358 qemu_sem_init(&ms->postcopy_qemufile_src_sem, 0);
3359 qemu_mutex_init(&ms->qemu_file_lock);
3363 * Return true if check pass, false otherwise. Error will be put
3364 * inside errp if provided.
3366 static bool migration_object_check(MigrationState *ms, Error **errp)
3368 /* Assuming all off */
3369 bool old_caps[MIGRATION_CAPABILITY__MAX] = { 0 };
3371 if (!migrate_params_check(&ms->parameters, errp)) {
3375 return migrate_caps_check(old_caps, ms->capabilities, errp);
3378 static const TypeInfo migration_type = {
3379 .name = TYPE_MIGRATION,
3381 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3382 * not created using qdev_new(), it is not attached to the qdev
3383 * device tree, and it is never realized.
3385 * TODO: Make this TYPE_OBJECT once QOM provides something like
3386 * TYPE_DEVICE's "-global" properties.
3388 .parent = TYPE_DEVICE,
3389 .class_init = migration_class_init,
3390 .class_size = sizeof(MigrationClass),
3391 .instance_size = sizeof(MigrationState),
3392 .instance_init = migration_instance_init,
3393 .instance_finalize = migration_instance_finalize,
3396 static void register_migration_types(void)
3398 type_register_static(&migration_type);
3401 type_init(register_migration_types);