hw/display/xlnx_dp: Free FIFOs adding xlnx_dp_finalize()
[qemu/ar7.git] / migration / migration.c
blobca8b97baa5ac8b596f6758cad049c680757b8add
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 "qemu/main-loop.h"
20 #include "migration/blocker.h"
21 #include "exec.h"
22 #include "fd.h"
23 #include "socket.h"
24 #include "sysemu/runstate.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/cpu-throttle.h"
27 #include "rdma.h"
28 #include "ram.h"
29 #include "migration/global_state.h"
30 #include "migration/misc.h"
31 #include "migration.h"
32 #include "savevm.h"
33 #include "qemu-file-channel.h"
34 #include "qemu-file.h"
35 #include "migration/vmstate.h"
36 #include "block/block.h"
37 #include "qapi/error.h"
38 #include "qapi/clone-visitor.h"
39 #include "qapi/qapi-visit-migration.h"
40 #include "qapi/qapi-visit-sockets.h"
41 #include "qapi/qapi-commands-migration.h"
42 #include "qapi/qapi-events-migration.h"
43 #include "qapi/qmp/qerror.h"
44 #include "qapi/qmp/qnull.h"
45 #include "qemu/rcu.h"
46 #include "block.h"
47 #include "postcopy-ram.h"
48 #include "qemu/thread.h"
49 #include "trace.h"
50 #include "exec/target_page.h"
51 #include "io/channel-buffer.h"
52 #include "migration/colo.h"
53 #include "hw/boards.h"
54 #include "hw/qdev-properties.h"
55 #include "hw/qdev-properties-system.h"
56 #include "monitor/monitor.h"
57 #include "net/announce.h"
58 #include "qemu/queue.h"
59 #include "multifd.h"
60 #include "qemu/yank.h"
61 #include "sysemu/cpus.h"
63 #ifdef CONFIG_VFIO
64 #include "hw/vfio/vfio-common.h"
65 #endif
67 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
69 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
70 * data. */
71 #define BUFFER_DELAY 100
72 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
74 /* Time in milliseconds we are allowed to stop the source,
75 * for sending the last part */
76 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
78 /* Maximum migrate downtime set to 2000 seconds */
79 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
80 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
82 /* Default compression thread count */
83 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
84 /* Default decompression thread count, usually decompression is at
85 * least 4 times as fast as compression.*/
86 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
87 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
88 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
89 /* Define default autoconverge cpu throttle migration parameters */
90 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
91 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
92 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
93 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
95 /* Migration XBZRLE default cache size */
96 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
98 /* The delay time (in ms) between two COLO checkpoints */
99 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
100 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
101 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
102 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
103 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
104 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
105 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
107 /* Background transfer rate for postcopy, 0 means unlimited, note
108 * that page requests can still exceed this limit.
110 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
113 * Parameters for self_announce_delay giving a stream of RARP/ARP
114 * packets after migration.
116 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
117 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
118 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
119 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
121 static NotifierList migration_state_notifiers =
122 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
124 /* Messages sent on the return path from destination to source */
125 enum mig_rp_message_type {
126 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
127 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
128 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
130 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
131 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
132 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
133 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
135 MIG_RP_MSG_MAX
138 /* Migration capabilities set */
139 struct MigrateCapsSet {
140 int size; /* Capability set size */
141 MigrationCapability caps[]; /* Variadic array of capabilities */
143 typedef struct MigrateCapsSet MigrateCapsSet;
145 /* Define and initialize MigrateCapsSet */
146 #define INITIALIZE_MIGRATE_CAPS_SET(_name, ...) \
147 MigrateCapsSet _name = { \
148 .size = sizeof((int []) { __VA_ARGS__ }) / sizeof(int), \
149 .caps = { __VA_ARGS__ } \
152 /* Background-snapshot compatibility check list */
153 static const
154 INITIALIZE_MIGRATE_CAPS_SET(check_caps_background_snapshot,
155 MIGRATION_CAPABILITY_POSTCOPY_RAM,
156 MIGRATION_CAPABILITY_DIRTY_BITMAPS,
157 MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME,
158 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE,
159 MIGRATION_CAPABILITY_RETURN_PATH,
160 MIGRATION_CAPABILITY_MULTIFD,
161 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER,
162 MIGRATION_CAPABILITY_AUTO_CONVERGE,
163 MIGRATION_CAPABILITY_RELEASE_RAM,
164 MIGRATION_CAPABILITY_RDMA_PIN_ALL,
165 MIGRATION_CAPABILITY_COMPRESS,
166 MIGRATION_CAPABILITY_XBZRLE,
167 MIGRATION_CAPABILITY_X_COLO,
168 MIGRATION_CAPABILITY_VALIDATE_UUID);
170 /* When we add fault tolerance, we could have several
171 migrations at once. For now we don't need to add
172 dynamic creation of migration */
174 static MigrationState *current_migration;
175 static MigrationIncomingState *current_incoming;
177 static GSList *migration_blockers;
179 static bool migration_object_check(MigrationState *ms, Error **errp);
180 static int migration_maybe_pause(MigrationState *s,
181 int *current_active_state,
182 int new_state);
183 static void migrate_fd_cancel(MigrationState *s);
185 static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp)
187 uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp;
189 return (a > b) - (a < b);
192 void migration_object_init(void)
194 Error *err = NULL;
196 /* This can only be called once. */
197 assert(!current_migration);
198 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
201 * Init the migrate incoming object as well no matter whether
202 * we'll use it or not.
204 assert(!current_incoming);
205 current_incoming = g_new0(MigrationIncomingState, 1);
206 current_incoming->state = MIGRATION_STATUS_NONE;
207 current_incoming->postcopy_remote_fds =
208 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
209 qemu_mutex_init(&current_incoming->rp_mutex);
210 qemu_event_init(&current_incoming->main_thread_load_event, false);
211 qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
212 qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
213 qemu_mutex_init(&current_incoming->page_request_mutex);
214 current_incoming->page_requested = g_tree_new(page_request_addr_cmp);
216 if (!migration_object_check(current_migration, &err)) {
217 error_report_err(err);
218 exit(1);
221 blk_mig_init();
222 ram_mig_init();
223 dirty_bitmap_mig_init();
226 void migration_shutdown(void)
229 * Cancel the current migration - that will (eventually)
230 * stop the migration using this structure
232 migrate_fd_cancel(current_migration);
233 object_unref(OBJECT(current_migration));
236 * Cancel outgoing migration of dirty bitmaps. It should
237 * at least unref used block nodes.
239 dirty_bitmap_mig_cancel_outgoing();
242 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
243 * are non-critical data, and their loss never considered as
244 * something serious.
246 dirty_bitmap_mig_cancel_incoming();
249 /* For outgoing */
250 MigrationState *migrate_get_current(void)
252 /* This can only be called after the object created. */
253 assert(current_migration);
254 return current_migration;
257 MigrationIncomingState *migration_incoming_get_current(void)
259 assert(current_incoming);
260 return current_incoming;
263 void migration_incoming_state_destroy(void)
265 struct MigrationIncomingState *mis = migration_incoming_get_current();
267 if (mis->to_src_file) {
268 /* Tell source that we are done */
269 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
270 qemu_fclose(mis->to_src_file);
271 mis->to_src_file = NULL;
274 if (mis->from_src_file) {
275 qemu_fclose(mis->from_src_file);
276 mis->from_src_file = NULL;
278 if (mis->postcopy_remote_fds) {
279 g_array_free(mis->postcopy_remote_fds, TRUE);
280 mis->postcopy_remote_fds = NULL;
283 qemu_event_reset(&mis->main_thread_load_event);
285 if (mis->page_requested) {
286 g_tree_destroy(mis->page_requested);
287 mis->page_requested = NULL;
290 if (mis->socket_address_list) {
291 qapi_free_SocketAddressList(mis->socket_address_list);
292 mis->socket_address_list = NULL;
295 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
298 static void migrate_generate_event(int new_state)
300 if (migrate_use_events()) {
301 qapi_event_send_migration(new_state);
305 static bool migrate_late_block_activate(void)
307 MigrationState *s;
309 s = migrate_get_current();
311 return s->enabled_capabilities[
312 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
316 * Send a message on the return channel back to the source
317 * of the migration.
319 static int migrate_send_rp_message(MigrationIncomingState *mis,
320 enum mig_rp_message_type message_type,
321 uint16_t len, void *data)
323 int ret = 0;
325 trace_migrate_send_rp_message((int)message_type, len);
326 QEMU_LOCK_GUARD(&mis->rp_mutex);
329 * It's possible that the file handle got lost due to network
330 * failures.
332 if (!mis->to_src_file) {
333 ret = -EIO;
334 return ret;
337 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
338 qemu_put_be16(mis->to_src_file, len);
339 qemu_put_buffer(mis->to_src_file, data, len);
340 qemu_fflush(mis->to_src_file);
342 /* It's possible that qemu file got error during sending */
343 ret = qemu_file_get_error(mis->to_src_file);
345 return ret;
348 /* Request one page from the source VM at the given start address.
349 * rb: the RAMBlock to request the page in
350 * Start: Address offset within the RB
351 * Len: Length in bytes required - must be a multiple of pagesize
353 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
354 RAMBlock *rb, ram_addr_t start)
356 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
357 size_t msglen = 12; /* start + len */
358 size_t len = qemu_ram_pagesize(rb);
359 enum mig_rp_message_type msg_type;
360 const char *rbname;
361 int rbname_len;
363 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
364 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
367 * We maintain the last ramblock that we requested for page. Note that we
368 * don't need locking because this function will only be called within the
369 * postcopy ram fault thread.
371 if (rb != mis->last_rb) {
372 mis->last_rb = rb;
374 rbname = qemu_ram_get_idstr(rb);
375 rbname_len = strlen(rbname);
377 assert(rbname_len < 256);
379 bufc[msglen++] = rbname_len;
380 memcpy(bufc + msglen, rbname, rbname_len);
381 msglen += rbname_len;
382 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
383 } else {
384 msg_type = MIG_RP_MSG_REQ_PAGES;
387 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
390 int migrate_send_rp_req_pages(MigrationIncomingState *mis,
391 RAMBlock *rb, ram_addr_t start, uint64_t haddr)
393 void *aligned = (void *)(uintptr_t)(haddr & (-qemu_ram_pagesize(rb)));
394 bool received = false;
396 WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
397 received = ramblock_recv_bitmap_test_byte_offset(rb, start);
398 if (!received && !g_tree_lookup(mis->page_requested, aligned)) {
400 * The page has not been received, and it's not yet in the page
401 * request list. Queue it. Set the value of element to 1, so that
402 * things like g_tree_lookup() will return TRUE (1) when found.
404 g_tree_insert(mis->page_requested, aligned, (gpointer)1);
405 mis->page_requested_count++;
406 trace_postcopy_page_req_add(aligned, mis->page_requested_count);
411 * If the page is there, skip sending the message. We don't even need the
412 * lock because as long as the page arrived, it'll be there forever.
414 if (received) {
415 return 0;
418 return migrate_send_rp_message_req_pages(mis, rb, start);
421 static bool migration_colo_enabled;
422 bool migration_incoming_colo_enabled(void)
424 return migration_colo_enabled;
427 void migration_incoming_disable_colo(void)
429 ram_block_discard_disable(false);
430 migration_colo_enabled = false;
433 int migration_incoming_enable_colo(void)
435 if (ram_block_discard_disable(true)) {
436 error_report("COLO: cannot disable RAM discard");
437 return -EBUSY;
439 migration_colo_enabled = true;
440 return 0;
443 void migrate_add_address(SocketAddress *address)
445 MigrationIncomingState *mis = migration_incoming_get_current();
447 QAPI_LIST_PREPEND(mis->socket_address_list,
448 QAPI_CLONE(SocketAddress, address));
451 static void qemu_start_incoming_migration(const char *uri, Error **errp)
453 const char *p = NULL;
455 if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
456 return;
459 qapi_event_send_migration(MIGRATION_STATUS_SETUP);
460 if (strstart(uri, "tcp:", &p) ||
461 strstart(uri, "unix:", NULL) ||
462 strstart(uri, "vsock:", NULL)) {
463 socket_start_incoming_migration(p ? p : uri, errp);
464 #ifdef CONFIG_RDMA
465 } else if (strstart(uri, "rdma:", &p)) {
466 rdma_start_incoming_migration(p, errp);
467 #endif
468 } else if (strstart(uri, "exec:", &p)) {
469 exec_start_incoming_migration(p, errp);
470 } else if (strstart(uri, "fd:", &p)) {
471 fd_start_incoming_migration(p, errp);
472 } else {
473 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
474 error_setg(errp, "unknown migration protocol: %s", uri);
478 static void process_incoming_migration_bh(void *opaque)
480 Error *local_err = NULL;
481 MigrationIncomingState *mis = opaque;
483 /* If capability late_block_activate is set:
484 * Only fire up the block code now if we're going to restart the
485 * VM, else 'cont' will do it.
486 * This causes file locking to happen; so we don't want it to happen
487 * unless we really are starting the VM.
489 if (!migrate_late_block_activate() ||
490 (autostart && (!global_state_received() ||
491 global_state_get_runstate() == RUN_STATE_RUNNING))) {
492 /* Make sure all file formats flush their mutable metadata.
493 * If we get an error here, just don't restart the VM yet. */
494 bdrv_invalidate_cache_all(&local_err);
495 if (local_err) {
496 error_report_err(local_err);
497 local_err = NULL;
498 autostart = false;
503 * This must happen after all error conditions are dealt with and
504 * we're sure the VM is going to be running on this host.
506 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
508 if (multifd_load_cleanup(&local_err) != 0) {
509 error_report_err(local_err);
510 autostart = false;
512 /* If global state section was not received or we are in running
513 state, we need to obey autostart. Any other state is set with
514 runstate_set. */
516 dirty_bitmap_mig_before_vm_start();
518 if (!global_state_received() ||
519 global_state_get_runstate() == RUN_STATE_RUNNING) {
520 if (autostart) {
521 vm_start();
522 } else {
523 runstate_set(RUN_STATE_PAUSED);
525 } else if (migration_incoming_colo_enabled()) {
526 migration_incoming_disable_colo();
527 vm_start();
528 } else {
529 runstate_set(global_state_get_runstate());
532 * This must happen after any state changes since as soon as an external
533 * observer sees this event they might start to prod at the VM assuming
534 * it's ready to use.
536 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
537 MIGRATION_STATUS_COMPLETED);
538 qemu_bh_delete(mis->bh);
539 migration_incoming_state_destroy();
542 static void process_incoming_migration_co(void *opaque)
544 MigrationIncomingState *mis = migration_incoming_get_current();
545 PostcopyState ps;
546 int ret;
547 Error *local_err = NULL;
549 assert(mis->from_src_file);
550 mis->migration_incoming_co = qemu_coroutine_self();
551 mis->largest_page_size = qemu_ram_pagesize_largest();
552 postcopy_state_set(POSTCOPY_INCOMING_NONE);
553 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
554 MIGRATION_STATUS_ACTIVE);
555 ret = qemu_loadvm_state(mis->from_src_file);
557 ps = postcopy_state_get();
558 trace_process_incoming_migration_co_end(ret, ps);
559 if (ps != POSTCOPY_INCOMING_NONE) {
560 if (ps == POSTCOPY_INCOMING_ADVISE) {
562 * Where a migration had postcopy enabled (and thus went to advise)
563 * but managed to complete within the precopy period, we can use
564 * the normal exit.
566 postcopy_ram_incoming_cleanup(mis);
567 } else if (ret >= 0) {
569 * Postcopy was started, cleanup should happen at the end of the
570 * postcopy thread.
572 trace_process_incoming_migration_co_postcopy_end_main();
573 return;
575 /* Else if something went wrong then just fall out of the normal exit */
578 /* we get COLO info, and know if we are in COLO mode */
579 if (!ret && migration_incoming_colo_enabled()) {
580 /* Make sure all file formats flush their mutable metadata */
581 bdrv_invalidate_cache_all(&local_err);
582 if (local_err) {
583 error_report_err(local_err);
584 goto fail;
587 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
588 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
589 mis->have_colo_incoming_thread = true;
590 qemu_coroutine_yield();
592 /* Wait checkpoint incoming thread exit before free resource */
593 qemu_thread_join(&mis->colo_incoming_thread);
594 /* We hold the global iothread lock, so it is safe here */
595 colo_release_ram_cache();
598 if (ret < 0) {
599 error_report("load of migration failed: %s", strerror(-ret));
600 goto fail;
602 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
603 qemu_bh_schedule(mis->bh);
604 mis->migration_incoming_co = NULL;
605 return;
606 fail:
607 local_err = NULL;
608 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
609 MIGRATION_STATUS_FAILED);
610 qemu_fclose(mis->from_src_file);
611 if (multifd_load_cleanup(&local_err) != 0) {
612 error_report_err(local_err);
614 exit(EXIT_FAILURE);
618 * @migration_incoming_setup: Setup incoming migration
620 * Returns 0 for no error or 1 for error
622 * @f: file for main migration channel
623 * @errp: where to put errors
625 static int migration_incoming_setup(QEMUFile *f, Error **errp)
627 MigrationIncomingState *mis = migration_incoming_get_current();
628 Error *local_err = NULL;
630 if (multifd_load_setup(&local_err) != 0) {
631 /* We haven't been able to create multifd threads
632 nothing better to do */
633 error_report_err(local_err);
634 exit(EXIT_FAILURE);
637 if (!mis->from_src_file) {
638 mis->from_src_file = f;
640 qemu_file_set_blocking(f, false);
641 return 0;
644 void migration_incoming_process(void)
646 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
647 qemu_coroutine_enter(co);
650 /* Returns true if recovered from a paused migration, otherwise false */
651 static bool postcopy_try_recover(QEMUFile *f)
653 MigrationIncomingState *mis = migration_incoming_get_current();
655 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
656 /* Resumed from a paused postcopy migration */
658 mis->from_src_file = f;
659 /* Postcopy has standalone thread to do vm load */
660 qemu_file_set_blocking(f, true);
662 /* Re-configure the return path */
663 mis->to_src_file = qemu_file_get_return_path(f);
665 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
666 MIGRATION_STATUS_POSTCOPY_RECOVER);
669 * Here, we only wake up the main loading thread (while the
670 * fault thread will still be waiting), so that we can receive
671 * commands from source now, and answer it if needed. The
672 * fault thread will be woken up afterwards until we are sure
673 * that source is ready to reply to page requests.
675 qemu_sem_post(&mis->postcopy_pause_sem_dst);
676 return true;
679 return false;
682 void migration_fd_process_incoming(QEMUFile *f, Error **errp)
684 Error *local_err = NULL;
686 if (postcopy_try_recover(f)) {
687 return;
690 if (migration_incoming_setup(f, &local_err)) {
691 error_propagate(errp, local_err);
692 return;
694 migration_incoming_process();
697 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
699 MigrationIncomingState *mis = migration_incoming_get_current();
700 Error *local_err = NULL;
701 bool start_migration;
703 if (!mis->from_src_file) {
704 /* The first connection (multifd may have multiple) */
705 QEMUFile *f = qemu_fopen_channel_input(ioc);
707 /* If it's a recovery, we're done */
708 if (postcopy_try_recover(f)) {
709 return;
712 if (migration_incoming_setup(f, &local_err)) {
713 error_propagate(errp, local_err);
714 return;
718 * Common migration only needs one channel, so we can start
719 * right now. Multifd needs more than one channel, we wait.
721 start_migration = !migrate_use_multifd();
722 } else {
723 /* Multiple connections */
724 assert(migrate_use_multifd());
725 start_migration = multifd_recv_new_channel(ioc, &local_err);
726 if (local_err) {
727 error_propagate(errp, local_err);
728 return;
732 if (start_migration) {
733 migration_incoming_process();
738 * @migration_has_all_channels: We have received all channels that we need
740 * Returns true when we have got connections to all the channels that
741 * we need for migration.
743 bool migration_has_all_channels(void)
745 MigrationIncomingState *mis = migration_incoming_get_current();
746 bool all_channels;
748 all_channels = multifd_recv_all_channels_created();
750 return all_channels && mis->from_src_file != NULL;
754 * Send a 'SHUT' message on the return channel with the given value
755 * to indicate that we've finished with the RP. Non-0 value indicates
756 * error.
758 void migrate_send_rp_shut(MigrationIncomingState *mis,
759 uint32_t value)
761 uint32_t buf;
763 buf = cpu_to_be32(value);
764 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
768 * Send a 'PONG' message on the return channel with the given value
769 * (normally in response to a 'PING')
771 void migrate_send_rp_pong(MigrationIncomingState *mis,
772 uint32_t value)
774 uint32_t buf;
776 buf = cpu_to_be32(value);
777 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
780 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
781 char *block_name)
783 char buf[512];
784 int len;
785 int64_t res;
788 * First, we send the header part. It contains only the len of
789 * idstr, and the idstr itself.
791 len = strlen(block_name);
792 buf[0] = len;
793 memcpy(buf + 1, block_name, len);
795 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
796 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
797 __func__);
798 return;
801 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
804 * Next, we dump the received bitmap to the stream.
806 * TODO: currently we are safe since we are the only one that is
807 * using the to_src_file handle (fault thread is still paused),
808 * and it's ok even not taking the mutex. However the best way is
809 * to take the lock before sending the message header, and release
810 * the lock after sending the bitmap.
812 qemu_mutex_lock(&mis->rp_mutex);
813 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
814 qemu_mutex_unlock(&mis->rp_mutex);
816 trace_migrate_send_rp_recv_bitmap(block_name, res);
819 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
821 uint32_t buf;
823 buf = cpu_to_be32(value);
824 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
827 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
829 MigrationCapabilityStatusList *head = NULL, **tail = &head;
830 MigrationCapabilityStatus *caps;
831 MigrationState *s = migrate_get_current();
832 int i;
834 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
835 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
836 if (i == MIGRATION_CAPABILITY_BLOCK) {
837 continue;
839 #endif
840 caps = g_malloc0(sizeof(*caps));
841 caps->capability = i;
842 caps->state = s->enabled_capabilities[i];
843 QAPI_LIST_APPEND(tail, caps);
846 return head;
849 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
851 MigrationParameters *params;
852 MigrationState *s = migrate_get_current();
854 /* TODO use QAPI_CLONE() instead of duplicating it inline */
855 params = g_malloc0(sizeof(*params));
856 params->has_compress_level = true;
857 params->compress_level = s->parameters.compress_level;
858 params->has_compress_threads = true;
859 params->compress_threads = s->parameters.compress_threads;
860 params->has_compress_wait_thread = true;
861 params->compress_wait_thread = s->parameters.compress_wait_thread;
862 params->has_decompress_threads = true;
863 params->decompress_threads = s->parameters.decompress_threads;
864 params->has_throttle_trigger_threshold = true;
865 params->throttle_trigger_threshold = s->parameters.throttle_trigger_threshold;
866 params->has_cpu_throttle_initial = true;
867 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
868 params->has_cpu_throttle_increment = true;
869 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
870 params->has_cpu_throttle_tailslow = true;
871 params->cpu_throttle_tailslow = s->parameters.cpu_throttle_tailslow;
872 params->has_tls_creds = true;
873 params->tls_creds = g_strdup(s->parameters.tls_creds);
874 params->has_tls_hostname = true;
875 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
876 params->has_tls_authz = true;
877 params->tls_authz = g_strdup(s->parameters.tls_authz ?
878 s->parameters.tls_authz : "");
879 params->has_max_bandwidth = true;
880 params->max_bandwidth = s->parameters.max_bandwidth;
881 params->has_downtime_limit = true;
882 params->downtime_limit = s->parameters.downtime_limit;
883 params->has_x_checkpoint_delay = true;
884 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
885 params->has_block_incremental = true;
886 params->block_incremental = s->parameters.block_incremental;
887 params->has_multifd_channels = true;
888 params->multifd_channels = s->parameters.multifd_channels;
889 params->has_multifd_compression = true;
890 params->multifd_compression = s->parameters.multifd_compression;
891 params->has_multifd_zlib_level = true;
892 params->multifd_zlib_level = s->parameters.multifd_zlib_level;
893 params->has_multifd_zstd_level = true;
894 params->multifd_zstd_level = s->parameters.multifd_zstd_level;
895 params->has_xbzrle_cache_size = true;
896 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
897 params->has_max_postcopy_bandwidth = true;
898 params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
899 params->has_max_cpu_throttle = true;
900 params->max_cpu_throttle = s->parameters.max_cpu_throttle;
901 params->has_announce_initial = true;
902 params->announce_initial = s->parameters.announce_initial;
903 params->has_announce_max = true;
904 params->announce_max = s->parameters.announce_max;
905 params->has_announce_rounds = true;
906 params->announce_rounds = s->parameters.announce_rounds;
907 params->has_announce_step = true;
908 params->announce_step = s->parameters.announce_step;
910 if (s->parameters.has_block_bitmap_mapping) {
911 params->has_block_bitmap_mapping = true;
912 params->block_bitmap_mapping =
913 QAPI_CLONE(BitmapMigrationNodeAliasList,
914 s->parameters.block_bitmap_mapping);
917 return params;
920 AnnounceParameters *migrate_announce_params(void)
922 static AnnounceParameters ap;
924 MigrationState *s = migrate_get_current();
926 ap.initial = s->parameters.announce_initial;
927 ap.max = s->parameters.announce_max;
928 ap.rounds = s->parameters.announce_rounds;
929 ap.step = s->parameters.announce_step;
931 return &ap;
935 * Return true if we're already in the middle of a migration
936 * (i.e. any of the active or setup states)
938 bool migration_is_setup_or_active(int state)
940 switch (state) {
941 case MIGRATION_STATUS_ACTIVE:
942 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
943 case MIGRATION_STATUS_POSTCOPY_PAUSED:
944 case MIGRATION_STATUS_POSTCOPY_RECOVER:
945 case MIGRATION_STATUS_SETUP:
946 case MIGRATION_STATUS_PRE_SWITCHOVER:
947 case MIGRATION_STATUS_DEVICE:
948 case MIGRATION_STATUS_WAIT_UNPLUG:
949 case MIGRATION_STATUS_COLO:
950 return true;
952 default:
953 return false;
958 bool migration_is_running(int state)
960 switch (state) {
961 case MIGRATION_STATUS_ACTIVE:
962 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
963 case MIGRATION_STATUS_POSTCOPY_PAUSED:
964 case MIGRATION_STATUS_POSTCOPY_RECOVER:
965 case MIGRATION_STATUS_SETUP:
966 case MIGRATION_STATUS_PRE_SWITCHOVER:
967 case MIGRATION_STATUS_DEVICE:
968 case MIGRATION_STATUS_WAIT_UNPLUG:
969 case MIGRATION_STATUS_CANCELLING:
970 return true;
972 default:
973 return false;
978 static void populate_time_info(MigrationInfo *info, MigrationState *s)
980 info->has_status = true;
981 info->has_setup_time = true;
982 info->setup_time = s->setup_time;
983 if (s->state == MIGRATION_STATUS_COMPLETED) {
984 info->has_total_time = true;
985 info->total_time = s->total_time;
986 info->has_downtime = true;
987 info->downtime = s->downtime;
988 } else {
989 info->has_total_time = true;
990 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
991 s->start_time;
992 info->has_expected_downtime = true;
993 info->expected_downtime = s->expected_downtime;
997 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
999 info->has_ram = true;
1000 info->ram = g_malloc0(sizeof(*info->ram));
1001 info->ram->transferred = ram_counters.transferred;
1002 info->ram->total = ram_bytes_total();
1003 info->ram->duplicate = ram_counters.duplicate;
1004 /* legacy value. It is not used anymore */
1005 info->ram->skipped = 0;
1006 info->ram->normal = ram_counters.normal;
1007 info->ram->normal_bytes = ram_counters.normal *
1008 qemu_target_page_size();
1009 info->ram->mbps = s->mbps;
1010 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
1011 info->ram->postcopy_requests = ram_counters.postcopy_requests;
1012 info->ram->page_size = qemu_target_page_size();
1013 info->ram->multifd_bytes = ram_counters.multifd_bytes;
1014 info->ram->pages_per_second = s->pages_per_second;
1016 if (migrate_use_xbzrle()) {
1017 info->has_xbzrle_cache = true;
1018 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
1019 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
1020 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
1021 info->xbzrle_cache->pages = xbzrle_counters.pages;
1022 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
1023 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
1024 info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate;
1025 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
1028 if (migrate_use_compression()) {
1029 info->has_compression = true;
1030 info->compression = g_malloc0(sizeof(*info->compression));
1031 info->compression->pages = compression_counters.pages;
1032 info->compression->busy = compression_counters.busy;
1033 info->compression->busy_rate = compression_counters.busy_rate;
1034 info->compression->compressed_size =
1035 compression_counters.compressed_size;
1036 info->compression->compression_rate =
1037 compression_counters.compression_rate;
1040 if (cpu_throttle_active()) {
1041 info->has_cpu_throttle_percentage = true;
1042 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
1045 if (s->state != MIGRATION_STATUS_COMPLETED) {
1046 info->ram->remaining = ram_bytes_remaining();
1047 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
1051 static void populate_disk_info(MigrationInfo *info)
1053 if (blk_mig_active()) {
1054 info->has_disk = true;
1055 info->disk = g_malloc0(sizeof(*info->disk));
1056 info->disk->transferred = blk_mig_bytes_transferred();
1057 info->disk->remaining = blk_mig_bytes_remaining();
1058 info->disk->total = blk_mig_bytes_total();
1062 static void populate_vfio_info(MigrationInfo *info)
1064 #ifdef CONFIG_VFIO
1065 if (vfio_mig_active()) {
1066 info->has_vfio = true;
1067 info->vfio = g_malloc0(sizeof(*info->vfio));
1068 info->vfio->transferred = vfio_mig_bytes_transferred();
1070 #endif
1073 static void fill_source_migration_info(MigrationInfo *info)
1075 MigrationState *s = migrate_get_current();
1077 info->blocked = migration_is_blocked(NULL);
1078 info->has_blocked_reasons = info->blocked;
1079 info->blocked_reasons = NULL;
1080 if (info->blocked) {
1081 GSList *cur_blocker = migration_blockers;
1084 * There are two types of reasons a migration might be blocked;
1085 * a) devices marked in VMState as non-migratable, and
1086 * b) Explicit migration blockers
1087 * We need to add both of them here.
1089 qemu_savevm_non_migratable_list(&info->blocked_reasons);
1091 while (cur_blocker) {
1092 QAPI_LIST_PREPEND(info->blocked_reasons,
1093 g_strdup(error_get_pretty(cur_blocker->data)));
1094 cur_blocker = g_slist_next(cur_blocker);
1098 switch (s->state) {
1099 case MIGRATION_STATUS_NONE:
1100 /* no migration has happened ever */
1101 /* do not overwrite destination migration status */
1102 return;
1103 case MIGRATION_STATUS_SETUP:
1104 info->has_status = true;
1105 info->has_total_time = false;
1106 break;
1107 case MIGRATION_STATUS_ACTIVE:
1108 case MIGRATION_STATUS_CANCELLING:
1109 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1110 case MIGRATION_STATUS_PRE_SWITCHOVER:
1111 case MIGRATION_STATUS_DEVICE:
1112 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1113 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1114 /* TODO add some postcopy stats */
1115 populate_time_info(info, s);
1116 populate_ram_info(info, s);
1117 populate_disk_info(info);
1118 populate_vfio_info(info);
1119 break;
1120 case MIGRATION_STATUS_COLO:
1121 info->has_status = true;
1122 /* TODO: display COLO specific information (checkpoint info etc.) */
1123 break;
1124 case MIGRATION_STATUS_COMPLETED:
1125 populate_time_info(info, s);
1126 populate_ram_info(info, s);
1127 populate_vfio_info(info);
1128 break;
1129 case MIGRATION_STATUS_FAILED:
1130 info->has_status = true;
1131 if (s->error) {
1132 info->has_error_desc = true;
1133 info->error_desc = g_strdup(error_get_pretty(s->error));
1135 break;
1136 case MIGRATION_STATUS_CANCELLED:
1137 info->has_status = true;
1138 break;
1139 case MIGRATION_STATUS_WAIT_UNPLUG:
1140 info->has_status = true;
1141 break;
1143 info->status = s->state;
1146 typedef enum WriteTrackingSupport {
1147 WT_SUPPORT_UNKNOWN = 0,
1148 WT_SUPPORT_ABSENT,
1149 WT_SUPPORT_AVAILABLE,
1150 WT_SUPPORT_COMPATIBLE
1151 } WriteTrackingSupport;
1153 static
1154 WriteTrackingSupport migrate_query_write_tracking(void)
1156 /* Check if kernel supports required UFFD features */
1157 if (!ram_write_tracking_available()) {
1158 return WT_SUPPORT_ABSENT;
1161 * Check if current memory configuration is
1162 * compatible with required UFFD features.
1164 if (!ram_write_tracking_compatible()) {
1165 return WT_SUPPORT_AVAILABLE;
1168 return WT_SUPPORT_COMPATIBLE;
1172 * @migration_caps_check - check capability validity
1174 * @cap_list: old capability list, array of bool
1175 * @params: new capabilities to be applied soon
1176 * @errp: set *errp if the check failed, with reason
1178 * Returns true if check passed, otherwise false.
1180 static bool migrate_caps_check(bool *cap_list,
1181 MigrationCapabilityStatusList *params,
1182 Error **errp)
1184 MigrationCapabilityStatusList *cap;
1185 bool old_postcopy_cap;
1186 MigrationIncomingState *mis = migration_incoming_get_current();
1188 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1190 for (cap = params; cap; cap = cap->next) {
1191 cap_list[cap->value->capability] = cap->value->state;
1194 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1195 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
1196 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
1197 "block migration");
1198 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
1199 return false;
1201 #endif
1203 #ifndef CONFIG_REPLICATION
1204 if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
1205 error_setg(errp, "QEMU compiled without replication module"
1206 " can't enable COLO");
1207 error_append_hint(errp, "Please enable replication before COLO.\n");
1208 return false;
1210 #endif
1212 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
1213 /* This check is reasonably expensive, so only when it's being
1214 * set the first time, also it's only the destination that needs
1215 * special support.
1217 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1218 !postcopy_ram_supported_by_host(mis)) {
1219 /* postcopy_ram_supported_by_host will have emitted a more
1220 * detailed message
1222 error_setg(errp, "Postcopy is not supported");
1223 return false;
1226 if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1227 error_setg(errp, "Postcopy is not compatible with ignore-shared");
1228 return false;
1232 if (cap_list[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT]) {
1233 WriteTrackingSupport wt_support;
1234 int idx;
1236 * Check if 'background-snapshot' capability is supported by
1237 * host kernel and compatible with guest memory configuration.
1239 wt_support = migrate_query_write_tracking();
1240 if (wt_support < WT_SUPPORT_AVAILABLE) {
1241 error_setg(errp, "Background-snapshot is not supported by host kernel");
1242 return false;
1244 if (wt_support < WT_SUPPORT_COMPATIBLE) {
1245 error_setg(errp, "Background-snapshot is not compatible "
1246 "with guest memory configuration");
1247 return false;
1251 * Check if there are any migration capabilities
1252 * incompatible with 'background-snapshot'.
1254 for (idx = 0; idx < check_caps_background_snapshot.size; idx++) {
1255 int incomp_cap = check_caps_background_snapshot.caps[idx];
1256 if (cap_list[incomp_cap]) {
1257 error_setg(errp,
1258 "Background-snapshot is not compatible with %s",
1259 MigrationCapability_str(incomp_cap));
1260 return false;
1265 return true;
1268 static void fill_destination_migration_info(MigrationInfo *info)
1270 MigrationIncomingState *mis = migration_incoming_get_current();
1272 if (mis->socket_address_list) {
1273 info->has_socket_address = true;
1274 info->socket_address =
1275 QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1278 switch (mis->state) {
1279 case MIGRATION_STATUS_NONE:
1280 return;
1281 case MIGRATION_STATUS_SETUP:
1282 case MIGRATION_STATUS_CANCELLING:
1283 case MIGRATION_STATUS_CANCELLED:
1284 case MIGRATION_STATUS_ACTIVE:
1285 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1286 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1287 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1288 case MIGRATION_STATUS_FAILED:
1289 case MIGRATION_STATUS_COLO:
1290 info->has_status = true;
1291 break;
1292 case MIGRATION_STATUS_COMPLETED:
1293 info->has_status = true;
1294 fill_destination_postcopy_migration_info(info);
1295 break;
1297 info->status = mis->state;
1300 MigrationInfo *qmp_query_migrate(Error **errp)
1302 MigrationInfo *info = g_malloc0(sizeof(*info));
1304 fill_destination_migration_info(info);
1305 fill_source_migration_info(info);
1307 return info;
1310 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1311 Error **errp)
1313 MigrationState *s = migrate_get_current();
1314 MigrationCapabilityStatusList *cap;
1315 bool cap_list[MIGRATION_CAPABILITY__MAX];
1317 if (migration_is_running(s->state)) {
1318 error_setg(errp, QERR_MIGRATION_ACTIVE);
1319 return;
1322 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1323 if (!migrate_caps_check(cap_list, params, errp)) {
1324 return;
1327 for (cap = params; cap; cap = cap->next) {
1328 s->enabled_capabilities[cap->value->capability] = cap->value->state;
1333 * Check whether the parameters are valid. Error will be put into errp
1334 * (if provided). Return true if valid, otherwise false.
1336 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1338 if (params->has_compress_level &&
1339 (params->compress_level > 9)) {
1340 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1341 "a value between 0 and 9");
1342 return false;
1345 if (params->has_compress_threads && (params->compress_threads < 1)) {
1346 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1347 "compress_threads",
1348 "a value between 1 and 255");
1349 return false;
1352 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1353 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1354 "decompress_threads",
1355 "a value between 1 and 255");
1356 return false;
1359 if (params->has_throttle_trigger_threshold &&
1360 (params->throttle_trigger_threshold < 1 ||
1361 params->throttle_trigger_threshold > 100)) {
1362 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1363 "throttle_trigger_threshold",
1364 "an integer in the range of 1 to 100");
1365 return false;
1368 if (params->has_cpu_throttle_initial &&
1369 (params->cpu_throttle_initial < 1 ||
1370 params->cpu_throttle_initial > 99)) {
1371 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1372 "cpu_throttle_initial",
1373 "an integer in the range of 1 to 99");
1374 return false;
1377 if (params->has_cpu_throttle_increment &&
1378 (params->cpu_throttle_increment < 1 ||
1379 params->cpu_throttle_increment > 99)) {
1380 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1381 "cpu_throttle_increment",
1382 "an integer in the range of 1 to 99");
1383 return false;
1386 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1387 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1388 "max_bandwidth",
1389 "an integer in the range of 0 to "stringify(SIZE_MAX)
1390 " bytes/second");
1391 return false;
1394 if (params->has_downtime_limit &&
1395 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1396 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1397 "downtime_limit",
1398 "an integer in the range of 0 to "
1399 stringify(MAX_MIGRATE_DOWNTIME)" ms");
1400 return false;
1403 /* x_checkpoint_delay is now always positive */
1405 if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1406 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1407 "multifd_channels",
1408 "a value between 1 and 255");
1409 return false;
1412 if (params->has_multifd_zlib_level &&
1413 (params->multifd_zlib_level > 9)) {
1414 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level",
1415 "a value between 0 and 9");
1416 return false;
1419 if (params->has_multifd_zstd_level &&
1420 (params->multifd_zstd_level > 20)) {
1421 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level",
1422 "a value between 0 and 20");
1423 return false;
1426 if (params->has_xbzrle_cache_size &&
1427 (params->xbzrle_cache_size < qemu_target_page_size() ||
1428 !is_power_of_2(params->xbzrle_cache_size))) {
1429 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1430 "xbzrle_cache_size",
1431 "a power of two no less than the target page size");
1432 return false;
1435 if (params->has_max_cpu_throttle &&
1436 (params->max_cpu_throttle < params->cpu_throttle_initial ||
1437 params->max_cpu_throttle > 99)) {
1438 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1439 "max_cpu_throttle",
1440 "an integer in the range of cpu_throttle_initial to 99");
1441 return false;
1444 if (params->has_announce_initial &&
1445 params->announce_initial > 100000) {
1446 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1447 "announce_initial",
1448 "a value between 0 and 100000");
1449 return false;
1451 if (params->has_announce_max &&
1452 params->announce_max > 100000) {
1453 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1454 "announce_max",
1455 "a value between 0 and 100000");
1456 return false;
1458 if (params->has_announce_rounds &&
1459 params->announce_rounds > 1000) {
1460 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1461 "announce_rounds",
1462 "a value between 0 and 1000");
1463 return false;
1465 if (params->has_announce_step &&
1466 (params->announce_step < 1 ||
1467 params->announce_step > 10000)) {
1468 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1469 "announce_step",
1470 "a value between 0 and 10000");
1471 return false;
1474 if (params->has_block_bitmap_mapping &&
1475 !check_dirty_bitmap_mig_alias_map(params->block_bitmap_mapping, errp)) {
1476 error_prepend(errp, "Invalid mapping given for block-bitmap-mapping: ");
1477 return false;
1480 return true;
1483 static void migrate_params_test_apply(MigrateSetParameters *params,
1484 MigrationParameters *dest)
1486 *dest = migrate_get_current()->parameters;
1488 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1490 if (params->has_compress_level) {
1491 dest->compress_level = params->compress_level;
1494 if (params->has_compress_threads) {
1495 dest->compress_threads = params->compress_threads;
1498 if (params->has_compress_wait_thread) {
1499 dest->compress_wait_thread = params->compress_wait_thread;
1502 if (params->has_decompress_threads) {
1503 dest->decompress_threads = params->decompress_threads;
1506 if (params->has_throttle_trigger_threshold) {
1507 dest->throttle_trigger_threshold = params->throttle_trigger_threshold;
1510 if (params->has_cpu_throttle_initial) {
1511 dest->cpu_throttle_initial = params->cpu_throttle_initial;
1514 if (params->has_cpu_throttle_increment) {
1515 dest->cpu_throttle_increment = params->cpu_throttle_increment;
1518 if (params->has_cpu_throttle_tailslow) {
1519 dest->cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1522 if (params->has_tls_creds) {
1523 assert(params->tls_creds->type == QTYPE_QSTRING);
1524 dest->tls_creds = params->tls_creds->u.s;
1527 if (params->has_tls_hostname) {
1528 assert(params->tls_hostname->type == QTYPE_QSTRING);
1529 dest->tls_hostname = params->tls_hostname->u.s;
1532 if (params->has_max_bandwidth) {
1533 dest->max_bandwidth = params->max_bandwidth;
1536 if (params->has_downtime_limit) {
1537 dest->downtime_limit = params->downtime_limit;
1540 if (params->has_x_checkpoint_delay) {
1541 dest->x_checkpoint_delay = params->x_checkpoint_delay;
1544 if (params->has_block_incremental) {
1545 dest->block_incremental = params->block_incremental;
1547 if (params->has_multifd_channels) {
1548 dest->multifd_channels = params->multifd_channels;
1550 if (params->has_multifd_compression) {
1551 dest->multifd_compression = params->multifd_compression;
1553 if (params->has_xbzrle_cache_size) {
1554 dest->xbzrle_cache_size = params->xbzrle_cache_size;
1556 if (params->has_max_postcopy_bandwidth) {
1557 dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1559 if (params->has_max_cpu_throttle) {
1560 dest->max_cpu_throttle = params->max_cpu_throttle;
1562 if (params->has_announce_initial) {
1563 dest->announce_initial = params->announce_initial;
1565 if (params->has_announce_max) {
1566 dest->announce_max = params->announce_max;
1568 if (params->has_announce_rounds) {
1569 dest->announce_rounds = params->announce_rounds;
1571 if (params->has_announce_step) {
1572 dest->announce_step = params->announce_step;
1575 if (params->has_block_bitmap_mapping) {
1576 dest->has_block_bitmap_mapping = true;
1577 dest->block_bitmap_mapping = params->block_bitmap_mapping;
1581 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1583 MigrationState *s = migrate_get_current();
1585 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1587 if (params->has_compress_level) {
1588 s->parameters.compress_level = params->compress_level;
1591 if (params->has_compress_threads) {
1592 s->parameters.compress_threads = params->compress_threads;
1595 if (params->has_compress_wait_thread) {
1596 s->parameters.compress_wait_thread = params->compress_wait_thread;
1599 if (params->has_decompress_threads) {
1600 s->parameters.decompress_threads = params->decompress_threads;
1603 if (params->has_throttle_trigger_threshold) {
1604 s->parameters.throttle_trigger_threshold = params->throttle_trigger_threshold;
1607 if (params->has_cpu_throttle_initial) {
1608 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1611 if (params->has_cpu_throttle_increment) {
1612 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1615 if (params->has_cpu_throttle_tailslow) {
1616 s->parameters.cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1619 if (params->has_tls_creds) {
1620 g_free(s->parameters.tls_creds);
1621 assert(params->tls_creds->type == QTYPE_QSTRING);
1622 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1625 if (params->has_tls_hostname) {
1626 g_free(s->parameters.tls_hostname);
1627 assert(params->tls_hostname->type == QTYPE_QSTRING);
1628 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1631 if (params->has_tls_authz) {
1632 g_free(s->parameters.tls_authz);
1633 assert(params->tls_authz->type == QTYPE_QSTRING);
1634 s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1637 if (params->has_max_bandwidth) {
1638 s->parameters.max_bandwidth = params->max_bandwidth;
1639 if (s->to_dst_file && !migration_in_postcopy()) {
1640 qemu_file_set_rate_limit(s->to_dst_file,
1641 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1645 if (params->has_downtime_limit) {
1646 s->parameters.downtime_limit = params->downtime_limit;
1649 if (params->has_x_checkpoint_delay) {
1650 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1651 if (migration_in_colo_state()) {
1652 colo_checkpoint_notify(s);
1656 if (params->has_block_incremental) {
1657 s->parameters.block_incremental = params->block_incremental;
1659 if (params->has_multifd_channels) {
1660 s->parameters.multifd_channels = params->multifd_channels;
1662 if (params->has_multifd_compression) {
1663 s->parameters.multifd_compression = params->multifd_compression;
1665 if (params->has_xbzrle_cache_size) {
1666 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1667 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1669 if (params->has_max_postcopy_bandwidth) {
1670 s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1671 if (s->to_dst_file && migration_in_postcopy()) {
1672 qemu_file_set_rate_limit(s->to_dst_file,
1673 s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1676 if (params->has_max_cpu_throttle) {
1677 s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1679 if (params->has_announce_initial) {
1680 s->parameters.announce_initial = params->announce_initial;
1682 if (params->has_announce_max) {
1683 s->parameters.announce_max = params->announce_max;
1685 if (params->has_announce_rounds) {
1686 s->parameters.announce_rounds = params->announce_rounds;
1688 if (params->has_announce_step) {
1689 s->parameters.announce_step = params->announce_step;
1692 if (params->has_block_bitmap_mapping) {
1693 qapi_free_BitmapMigrationNodeAliasList(
1694 s->parameters.block_bitmap_mapping);
1696 s->parameters.has_block_bitmap_mapping = true;
1697 s->parameters.block_bitmap_mapping =
1698 QAPI_CLONE(BitmapMigrationNodeAliasList,
1699 params->block_bitmap_mapping);
1703 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1705 MigrationParameters tmp;
1707 /* TODO Rewrite "" to null instead */
1708 if (params->has_tls_creds
1709 && params->tls_creds->type == QTYPE_QNULL) {
1710 qobject_unref(params->tls_creds->u.n);
1711 params->tls_creds->type = QTYPE_QSTRING;
1712 params->tls_creds->u.s = strdup("");
1714 /* TODO Rewrite "" to null instead */
1715 if (params->has_tls_hostname
1716 && params->tls_hostname->type == QTYPE_QNULL) {
1717 qobject_unref(params->tls_hostname->u.n);
1718 params->tls_hostname->type = QTYPE_QSTRING;
1719 params->tls_hostname->u.s = strdup("");
1722 migrate_params_test_apply(params, &tmp);
1724 if (!migrate_params_check(&tmp, errp)) {
1725 /* Invalid parameter */
1726 return;
1729 migrate_params_apply(params, errp);
1733 void qmp_migrate_start_postcopy(Error **errp)
1735 MigrationState *s = migrate_get_current();
1737 if (!migrate_postcopy()) {
1738 error_setg(errp, "Enable postcopy with migrate_set_capability before"
1739 " the start of migration");
1740 return;
1743 if (s->state == MIGRATION_STATUS_NONE) {
1744 error_setg(errp, "Postcopy must be started after migration has been"
1745 " started");
1746 return;
1749 * we don't error if migration has finished since that would be racy
1750 * with issuing this command.
1752 qatomic_set(&s->start_postcopy, true);
1755 /* shared migration helpers */
1757 void migrate_set_state(int *state, int old_state, int new_state)
1759 assert(new_state < MIGRATION_STATUS__MAX);
1760 if (qatomic_cmpxchg(state, old_state, new_state) == old_state) {
1761 trace_migrate_set_state(MigrationStatus_str(new_state));
1762 migrate_generate_event(new_state);
1766 static MigrationCapabilityStatus *migrate_cap_add(MigrationCapability index,
1767 bool state)
1769 MigrationCapabilityStatus *cap;
1771 cap = g_new0(MigrationCapabilityStatus, 1);
1772 cap->capability = index;
1773 cap->state = state;
1775 return cap;
1778 void migrate_set_block_enabled(bool value, Error **errp)
1780 MigrationCapabilityStatusList *cap = NULL;
1782 QAPI_LIST_PREPEND(cap, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK, value));
1783 qmp_migrate_set_capabilities(cap, errp);
1784 qapi_free_MigrationCapabilityStatusList(cap);
1787 static void migrate_set_block_incremental(MigrationState *s, bool value)
1789 s->parameters.block_incremental = value;
1792 static void block_cleanup_parameters(MigrationState *s)
1794 if (s->must_remove_block_options) {
1795 /* setting to false can never fail */
1796 migrate_set_block_enabled(false, &error_abort);
1797 migrate_set_block_incremental(s, false);
1798 s->must_remove_block_options = false;
1802 static void migrate_fd_cleanup(MigrationState *s)
1804 qemu_bh_delete(s->cleanup_bh);
1805 s->cleanup_bh = NULL;
1807 qemu_savevm_state_cleanup();
1809 if (s->to_dst_file) {
1810 QEMUFile *tmp;
1812 trace_migrate_fd_cleanup();
1813 qemu_mutex_unlock_iothread();
1814 if (s->migration_thread_running) {
1815 qemu_thread_join(&s->thread);
1816 s->migration_thread_running = false;
1818 qemu_mutex_lock_iothread();
1820 multifd_save_cleanup();
1821 qemu_mutex_lock(&s->qemu_file_lock);
1822 tmp = s->to_dst_file;
1823 s->to_dst_file = NULL;
1824 qemu_mutex_unlock(&s->qemu_file_lock);
1826 * Close the file handle without the lock to make sure the
1827 * critical section won't block for long.
1829 qemu_fclose(tmp);
1832 assert(!migration_is_active(s));
1834 if (s->state == MIGRATION_STATUS_CANCELLING) {
1835 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1836 MIGRATION_STATUS_CANCELLED);
1839 if (s->error) {
1840 /* It is used on info migrate. We can't free it */
1841 error_report_err(error_copy(s->error));
1843 notifier_list_notify(&migration_state_notifiers, s);
1844 block_cleanup_parameters(s);
1845 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
1848 static void migrate_fd_cleanup_schedule(MigrationState *s)
1851 * Ref the state for bh, because it may be called when
1852 * there're already no other refs
1854 object_ref(OBJECT(s));
1855 qemu_bh_schedule(s->cleanup_bh);
1858 static void migrate_fd_cleanup_bh(void *opaque)
1860 MigrationState *s = opaque;
1861 migrate_fd_cleanup(s);
1862 object_unref(OBJECT(s));
1865 void migrate_set_error(MigrationState *s, const Error *error)
1867 QEMU_LOCK_GUARD(&s->error_mutex);
1868 if (!s->error) {
1869 s->error = error_copy(error);
1873 void migrate_fd_error(MigrationState *s, const Error *error)
1875 trace_migrate_fd_error(error_get_pretty(error));
1876 assert(s->to_dst_file == NULL);
1877 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1878 MIGRATION_STATUS_FAILED);
1879 migrate_set_error(s, error);
1882 static void migrate_fd_cancel(MigrationState *s)
1884 int old_state ;
1885 QEMUFile *f = migrate_get_current()->to_dst_file;
1886 trace_migrate_fd_cancel();
1888 if (s->rp_state.from_dst_file) {
1889 /* shutdown the rp socket, so causing the rp thread to shutdown */
1890 qemu_file_shutdown(s->rp_state.from_dst_file);
1893 do {
1894 old_state = s->state;
1895 if (!migration_is_running(old_state)) {
1896 break;
1898 /* If the migration is paused, kick it out of the pause */
1899 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1900 qemu_sem_post(&s->pause_sem);
1902 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1903 } while (s->state != MIGRATION_STATUS_CANCELLING);
1906 * If we're unlucky the migration code might be stuck somewhere in a
1907 * send/write while the network has failed and is waiting to timeout;
1908 * if we've got shutdown(2) available then we can force it to quit.
1909 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1910 * called in a bh, so there is no race against this cancel.
1912 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1913 qemu_file_shutdown(f);
1915 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1916 Error *local_err = NULL;
1918 bdrv_invalidate_cache_all(&local_err);
1919 if (local_err) {
1920 error_report_err(local_err);
1921 } else {
1922 s->block_inactive = false;
1927 void add_migration_state_change_notifier(Notifier *notify)
1929 notifier_list_add(&migration_state_notifiers, notify);
1932 void remove_migration_state_change_notifier(Notifier *notify)
1934 notifier_remove(notify);
1937 bool migration_in_setup(MigrationState *s)
1939 return s->state == MIGRATION_STATUS_SETUP;
1942 bool migration_has_finished(MigrationState *s)
1944 return s->state == MIGRATION_STATUS_COMPLETED;
1947 bool migration_has_failed(MigrationState *s)
1949 return (s->state == MIGRATION_STATUS_CANCELLED ||
1950 s->state == MIGRATION_STATUS_FAILED);
1953 bool migration_in_postcopy(void)
1955 MigrationState *s = migrate_get_current();
1957 switch (s->state) {
1958 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1959 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1960 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1961 return true;
1962 default:
1963 return false;
1967 bool migration_in_postcopy_after_devices(MigrationState *s)
1969 return migration_in_postcopy() && s->postcopy_after_devices;
1972 bool migration_in_incoming_postcopy(void)
1974 PostcopyState ps = postcopy_state_get();
1976 return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END;
1979 bool migration_is_idle(void)
1981 MigrationState *s = current_migration;
1983 if (!s) {
1984 return true;
1987 switch (s->state) {
1988 case MIGRATION_STATUS_NONE:
1989 case MIGRATION_STATUS_CANCELLED:
1990 case MIGRATION_STATUS_COMPLETED:
1991 case MIGRATION_STATUS_FAILED:
1992 return true;
1993 case MIGRATION_STATUS_SETUP:
1994 case MIGRATION_STATUS_CANCELLING:
1995 case MIGRATION_STATUS_ACTIVE:
1996 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1997 case MIGRATION_STATUS_COLO:
1998 case MIGRATION_STATUS_PRE_SWITCHOVER:
1999 case MIGRATION_STATUS_DEVICE:
2000 case MIGRATION_STATUS_WAIT_UNPLUG:
2001 return false;
2002 case MIGRATION_STATUS__MAX:
2003 g_assert_not_reached();
2006 return false;
2009 bool migration_is_active(MigrationState *s)
2011 return (s->state == MIGRATION_STATUS_ACTIVE ||
2012 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2015 void migrate_init(MigrationState *s)
2018 * Reinitialise all migration state, except
2019 * parameters/capabilities that the user set, and
2020 * locks.
2022 s->cleanup_bh = 0;
2023 s->vm_start_bh = 0;
2024 s->to_dst_file = NULL;
2025 s->state = MIGRATION_STATUS_NONE;
2026 s->rp_state.from_dst_file = NULL;
2027 s->rp_state.error = false;
2028 s->mbps = 0.0;
2029 s->pages_per_second = 0.0;
2030 s->downtime = 0;
2031 s->expected_downtime = 0;
2032 s->setup_time = 0;
2033 s->start_postcopy = false;
2034 s->postcopy_after_devices = false;
2035 s->migration_thread_running = false;
2036 error_free(s->error);
2037 s->error = NULL;
2038 s->hostname = NULL;
2040 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
2042 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2043 s->total_time = 0;
2044 s->vm_was_running = false;
2045 s->iteration_initial_bytes = 0;
2046 s->threshold_size = 0;
2049 int migrate_add_blocker(Error *reason, Error **errp)
2051 if (only_migratable) {
2052 error_propagate_prepend(errp, error_copy(reason),
2053 "disallowing migration blocker "
2054 "(--only-migratable) for: ");
2055 return -EACCES;
2058 if (migration_is_idle()) {
2059 migration_blockers = g_slist_prepend(migration_blockers, reason);
2060 return 0;
2063 error_propagate_prepend(errp, error_copy(reason),
2064 "disallowing migration blocker "
2065 "(migration in progress) for: ");
2066 return -EBUSY;
2069 void migrate_del_blocker(Error *reason)
2071 migration_blockers = g_slist_remove(migration_blockers, reason);
2074 void qmp_migrate_incoming(const char *uri, Error **errp)
2076 Error *local_err = NULL;
2077 static bool once = true;
2079 if (!once) {
2080 error_setg(errp, "The incoming migration has already been started");
2081 return;
2083 if (!runstate_check(RUN_STATE_INMIGRATE)) {
2084 error_setg(errp, "'-incoming' was not specified on the command line");
2085 return;
2088 qemu_start_incoming_migration(uri, &local_err);
2090 if (local_err) {
2091 error_propagate(errp, local_err);
2092 return;
2095 once = false;
2098 void qmp_migrate_recover(const char *uri, Error **errp)
2100 MigrationIncomingState *mis = migration_incoming_get_current();
2102 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
2103 error_setg(errp, "Migrate recover can only be run "
2104 "when postcopy is paused.");
2105 return;
2108 if (qatomic_cmpxchg(&mis->postcopy_recover_triggered,
2109 false, true) == true) {
2110 error_setg(errp, "Migrate recovery is triggered already");
2111 return;
2115 * Note that this call will never start a real migration; it will
2116 * only re-setup the migration stream and poke existing migration
2117 * to continue using that newly established channel.
2119 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
2120 qemu_start_incoming_migration(uri, errp);
2123 void qmp_migrate_pause(Error **errp)
2125 MigrationState *ms = migrate_get_current();
2126 MigrationIncomingState *mis = migration_incoming_get_current();
2127 int ret;
2129 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2130 /* Source side, during postcopy */
2131 qemu_mutex_lock(&ms->qemu_file_lock);
2132 ret = qemu_file_shutdown(ms->to_dst_file);
2133 qemu_mutex_unlock(&ms->qemu_file_lock);
2134 if (ret) {
2135 error_setg(errp, "Failed to pause source migration");
2137 return;
2140 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2141 ret = qemu_file_shutdown(mis->from_src_file);
2142 if (ret) {
2143 error_setg(errp, "Failed to pause destination migration");
2145 return;
2148 error_setg(errp, "migrate-pause is currently only supported "
2149 "during postcopy-active state");
2152 bool migration_is_blocked(Error **errp)
2154 if (qemu_savevm_state_blocked(errp)) {
2155 return true;
2158 if (migration_blockers) {
2159 error_propagate(errp, error_copy(migration_blockers->data));
2160 return true;
2163 return false;
2166 /* Returns true if continue to migrate, or false if error detected */
2167 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
2168 bool resume, Error **errp)
2170 Error *local_err = NULL;
2172 if (resume) {
2173 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
2174 error_setg(errp, "Cannot resume if there is no "
2175 "paused migration");
2176 return false;
2180 * Postcopy recovery won't work well with release-ram
2181 * capability since release-ram will drop the page buffer as
2182 * long as the page is put into the send buffer. So if there
2183 * is a network failure happened, any page buffers that have
2184 * not yet reached the destination VM but have already been
2185 * sent from the source VM will be lost forever. Let's refuse
2186 * the client from resuming such a postcopy migration.
2187 * Luckily release-ram was designed to only be used when src
2188 * and destination VMs are on the same host, so it should be
2189 * fine.
2191 if (migrate_release_ram()) {
2192 error_setg(errp, "Postcopy recovery cannot work "
2193 "when release-ram capability is set");
2194 return false;
2197 /* This is a resume, skip init status */
2198 return true;
2201 if (migration_is_running(s->state)) {
2202 error_setg(errp, QERR_MIGRATION_ACTIVE);
2203 return false;
2206 if (runstate_check(RUN_STATE_INMIGRATE)) {
2207 error_setg(errp, "Guest is waiting for an incoming migration");
2208 return false;
2211 if (runstate_check(RUN_STATE_POSTMIGRATE)) {
2212 error_setg(errp, "Can't migrate the vm that was paused due to "
2213 "previous migration");
2214 return false;
2217 if (migration_is_blocked(errp)) {
2218 return false;
2221 if (blk || blk_inc) {
2222 if (migrate_use_block() || migrate_use_block_incremental()) {
2223 error_setg(errp, "Command options are incompatible with "
2224 "current migration capabilities");
2225 return false;
2227 migrate_set_block_enabled(true, &local_err);
2228 if (local_err) {
2229 error_propagate(errp, local_err);
2230 return false;
2232 s->must_remove_block_options = true;
2235 if (blk_inc) {
2236 migrate_set_block_incremental(s, true);
2239 migrate_init(s);
2241 * set ram_counters memory to zero for a
2242 * new migration
2244 memset(&ram_counters, 0, sizeof(ram_counters));
2246 return true;
2249 void qmp_migrate(const char *uri, bool has_blk, bool blk,
2250 bool has_inc, bool inc, bool has_detach, bool detach,
2251 bool has_resume, bool resume, Error **errp)
2253 Error *local_err = NULL;
2254 MigrationState *s = migrate_get_current();
2255 const char *p = NULL;
2257 if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
2258 has_resume && resume, errp)) {
2259 /* Error detected, put into errp */
2260 return;
2263 if (!(has_resume && resume)) {
2264 if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
2265 return;
2269 if (strstart(uri, "tcp:", &p) ||
2270 strstart(uri, "unix:", NULL) ||
2271 strstart(uri, "vsock:", NULL)) {
2272 socket_start_outgoing_migration(s, p ? p : uri, &local_err);
2273 #ifdef CONFIG_RDMA
2274 } else if (strstart(uri, "rdma:", &p)) {
2275 rdma_start_outgoing_migration(s, p, &local_err);
2276 #endif
2277 } else if (strstart(uri, "exec:", &p)) {
2278 exec_start_outgoing_migration(s, p, &local_err);
2279 } else if (strstart(uri, "fd:", &p)) {
2280 fd_start_outgoing_migration(s, p, &local_err);
2281 } else {
2282 if (!(has_resume && resume)) {
2283 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
2285 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
2286 "a valid migration protocol");
2287 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2288 MIGRATION_STATUS_FAILED);
2289 block_cleanup_parameters(s);
2290 return;
2293 if (local_err) {
2294 if (!(has_resume && resume)) {
2295 yank_unregister_instance(MIGRATION_YANK_INSTANCE);
2297 migrate_fd_error(s, local_err);
2298 error_propagate(errp, local_err);
2299 return;
2303 void qmp_migrate_cancel(Error **errp)
2305 migrate_fd_cancel(migrate_get_current());
2308 void qmp_migrate_continue(MigrationStatus state, Error **errp)
2310 MigrationState *s = migrate_get_current();
2311 if (s->state != state) {
2312 error_setg(errp, "Migration not in expected state: %s",
2313 MigrationStatus_str(s->state));
2314 return;
2316 qemu_sem_post(&s->pause_sem);
2319 bool migrate_release_ram(void)
2321 MigrationState *s;
2323 s = migrate_get_current();
2325 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2328 bool migrate_postcopy_ram(void)
2330 MigrationState *s;
2332 s = migrate_get_current();
2334 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2337 bool migrate_postcopy(void)
2339 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2342 bool migrate_auto_converge(void)
2344 MigrationState *s;
2346 s = migrate_get_current();
2348 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2351 bool migrate_zero_blocks(void)
2353 MigrationState *s;
2355 s = migrate_get_current();
2357 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2360 bool migrate_postcopy_blocktime(void)
2362 MigrationState *s;
2364 s = migrate_get_current();
2366 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2369 bool migrate_use_compression(void)
2371 MigrationState *s;
2373 s = migrate_get_current();
2375 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2378 int migrate_compress_level(void)
2380 MigrationState *s;
2382 s = migrate_get_current();
2384 return s->parameters.compress_level;
2387 int migrate_compress_threads(void)
2389 MigrationState *s;
2391 s = migrate_get_current();
2393 return s->parameters.compress_threads;
2396 int migrate_compress_wait_thread(void)
2398 MigrationState *s;
2400 s = migrate_get_current();
2402 return s->parameters.compress_wait_thread;
2405 int migrate_decompress_threads(void)
2407 MigrationState *s;
2409 s = migrate_get_current();
2411 return s->parameters.decompress_threads;
2414 bool migrate_dirty_bitmaps(void)
2416 MigrationState *s;
2418 s = migrate_get_current();
2420 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2423 bool migrate_ignore_shared(void)
2425 MigrationState *s;
2427 s = migrate_get_current();
2429 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2432 bool migrate_validate_uuid(void)
2434 MigrationState *s;
2436 s = migrate_get_current();
2438 return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
2441 bool migrate_use_events(void)
2443 MigrationState *s;
2445 s = migrate_get_current();
2447 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2450 bool migrate_use_multifd(void)
2452 MigrationState *s;
2454 s = migrate_get_current();
2456 return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2459 bool migrate_pause_before_switchover(void)
2461 MigrationState *s;
2463 s = migrate_get_current();
2465 return s->enabled_capabilities[
2466 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2469 int migrate_multifd_channels(void)
2471 MigrationState *s;
2473 s = migrate_get_current();
2475 return s->parameters.multifd_channels;
2478 MultiFDCompression migrate_multifd_compression(void)
2480 MigrationState *s;
2482 s = migrate_get_current();
2484 return s->parameters.multifd_compression;
2487 int migrate_multifd_zlib_level(void)
2489 MigrationState *s;
2491 s = migrate_get_current();
2493 return s->parameters.multifd_zlib_level;
2496 int migrate_multifd_zstd_level(void)
2498 MigrationState *s;
2500 s = migrate_get_current();
2502 return s->parameters.multifd_zstd_level;
2505 int migrate_use_xbzrle(void)
2507 MigrationState *s;
2509 s = migrate_get_current();
2511 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2514 uint64_t migrate_xbzrle_cache_size(void)
2516 MigrationState *s;
2518 s = migrate_get_current();
2520 return s->parameters.xbzrle_cache_size;
2523 static int64_t migrate_max_postcopy_bandwidth(void)
2525 MigrationState *s;
2527 s = migrate_get_current();
2529 return s->parameters.max_postcopy_bandwidth;
2532 bool migrate_use_block(void)
2534 MigrationState *s;
2536 s = migrate_get_current();
2538 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2541 bool migrate_use_return_path(void)
2543 MigrationState *s;
2545 s = migrate_get_current();
2547 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2550 bool migrate_use_block_incremental(void)
2552 MigrationState *s;
2554 s = migrate_get_current();
2556 return s->parameters.block_incremental;
2559 bool migrate_background_snapshot(void)
2561 MigrationState *s;
2563 s = migrate_get_current();
2565 return s->enabled_capabilities[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT];
2568 /* migration thread support */
2570 * Something bad happened to the RP stream, mark an error
2571 * The caller shall print or trace something to indicate why
2573 static void mark_source_rp_bad(MigrationState *s)
2575 s->rp_state.error = true;
2578 static struct rp_cmd_args {
2579 ssize_t len; /* -1 = variable */
2580 const char *name;
2581 } rp_cmd_args[] = {
2582 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
2583 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
2584 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
2585 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
2586 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
2587 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
2588 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
2589 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
2593 * Process a request for pages received on the return path,
2594 * We're allowed to send more than requested (e.g. to round to our page size)
2595 * and we don't need to send pages that have already been sent.
2597 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2598 ram_addr_t start, size_t len)
2600 long our_host_ps = qemu_real_host_page_size;
2602 trace_migrate_handle_rp_req_pages(rbname, start, len);
2605 * Since we currently insist on matching page sizes, just sanity check
2606 * we're being asked for whole host pages.
2608 if (start & (our_host_ps - 1) ||
2609 (len & (our_host_ps - 1))) {
2610 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2611 " len: %zd", __func__, start, len);
2612 mark_source_rp_bad(ms);
2613 return;
2616 if (ram_save_queue_pages(rbname, start, len)) {
2617 mark_source_rp_bad(ms);
2621 /* Return true to retry, false to quit */
2622 static bool postcopy_pause_return_path_thread(MigrationState *s)
2624 trace_postcopy_pause_return_path();
2626 qemu_sem_wait(&s->postcopy_pause_rp_sem);
2628 trace_postcopy_pause_return_path_continued();
2630 return true;
2633 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2635 RAMBlock *block = qemu_ram_block_by_name(block_name);
2637 if (!block) {
2638 error_report("%s: invalid block name '%s'", __func__, block_name);
2639 return -EINVAL;
2642 /* Fetch the received bitmap and refresh the dirty bitmap */
2643 return ram_dirty_bitmap_reload(s, block);
2646 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2648 trace_source_return_path_thread_resume_ack(value);
2650 if (value != MIGRATION_RESUME_ACK_VALUE) {
2651 error_report("%s: illegal resume_ack value %"PRIu32,
2652 __func__, value);
2653 return -1;
2656 /* Now both sides are active. */
2657 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2658 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2660 /* Notify send thread that time to continue send pages */
2661 qemu_sem_post(&s->rp_state.rp_sem);
2663 return 0;
2667 * Handles messages sent on the return path towards the source VM
2670 static void *source_return_path_thread(void *opaque)
2672 MigrationState *ms = opaque;
2673 QEMUFile *rp = ms->rp_state.from_dst_file;
2674 uint16_t header_len, header_type;
2675 uint8_t buf[512];
2676 uint32_t tmp32, sibling_error;
2677 ram_addr_t start = 0; /* =0 to silence warning */
2678 size_t len = 0, expected_len;
2679 int res;
2681 trace_source_return_path_thread_entry();
2682 rcu_register_thread();
2684 retry:
2685 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2686 migration_is_setup_or_active(ms->state)) {
2687 trace_source_return_path_thread_loop_top();
2688 header_type = qemu_get_be16(rp);
2689 header_len = qemu_get_be16(rp);
2691 if (qemu_file_get_error(rp)) {
2692 mark_source_rp_bad(ms);
2693 goto out;
2696 if (header_type >= MIG_RP_MSG_MAX ||
2697 header_type == MIG_RP_MSG_INVALID) {
2698 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2699 header_type, header_len);
2700 mark_source_rp_bad(ms);
2701 goto out;
2704 if ((rp_cmd_args[header_type].len != -1 &&
2705 header_len != rp_cmd_args[header_type].len) ||
2706 header_len > sizeof(buf)) {
2707 error_report("RP: Received '%s' message (0x%04x) with"
2708 "incorrect length %d expecting %zu",
2709 rp_cmd_args[header_type].name, header_type, header_len,
2710 (size_t)rp_cmd_args[header_type].len);
2711 mark_source_rp_bad(ms);
2712 goto out;
2715 /* We know we've got a valid header by this point */
2716 res = qemu_get_buffer(rp, buf, header_len);
2717 if (res != header_len) {
2718 error_report("RP: Failed reading data for message 0x%04x"
2719 " read %d expected %d",
2720 header_type, res, header_len);
2721 mark_source_rp_bad(ms);
2722 goto out;
2725 /* OK, we have the message and the data */
2726 switch (header_type) {
2727 case MIG_RP_MSG_SHUT:
2728 sibling_error = ldl_be_p(buf);
2729 trace_source_return_path_thread_shut(sibling_error);
2730 if (sibling_error) {
2731 error_report("RP: Sibling indicated error %d", sibling_error);
2732 mark_source_rp_bad(ms);
2735 * We'll let the main thread deal with closing the RP
2736 * we could do a shutdown(2) on it, but we're the only user
2737 * anyway, so there's nothing gained.
2739 goto out;
2741 case MIG_RP_MSG_PONG:
2742 tmp32 = ldl_be_p(buf);
2743 trace_source_return_path_thread_pong(tmp32);
2744 break;
2746 case MIG_RP_MSG_REQ_PAGES:
2747 start = ldq_be_p(buf);
2748 len = ldl_be_p(buf + 8);
2749 migrate_handle_rp_req_pages(ms, NULL, start, len);
2750 break;
2752 case MIG_RP_MSG_REQ_PAGES_ID:
2753 expected_len = 12 + 1; /* header + termination */
2755 if (header_len >= expected_len) {
2756 start = ldq_be_p(buf);
2757 len = ldl_be_p(buf + 8);
2758 /* Now we expect an idstr */
2759 tmp32 = buf[12]; /* Length of the following idstr */
2760 buf[13 + tmp32] = '\0';
2761 expected_len += tmp32;
2763 if (header_len != expected_len) {
2764 error_report("RP: Req_Page_id with length %d expecting %zd",
2765 header_len, expected_len);
2766 mark_source_rp_bad(ms);
2767 goto out;
2769 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2770 break;
2772 case MIG_RP_MSG_RECV_BITMAP:
2773 if (header_len < 1) {
2774 error_report("%s: missing block name", __func__);
2775 mark_source_rp_bad(ms);
2776 goto out;
2778 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2779 buf[buf[0] + 1] = '\0';
2780 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2781 mark_source_rp_bad(ms);
2782 goto out;
2784 break;
2786 case MIG_RP_MSG_RESUME_ACK:
2787 tmp32 = ldl_be_p(buf);
2788 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2789 mark_source_rp_bad(ms);
2790 goto out;
2792 break;
2794 default:
2795 break;
2799 out:
2800 res = qemu_file_get_error(rp);
2801 if (res) {
2802 if (res == -EIO && migration_in_postcopy()) {
2804 * Maybe there is something we can do: it looks like a
2805 * network down issue, and we pause for a recovery.
2807 if (postcopy_pause_return_path_thread(ms)) {
2808 /* Reload rp, reset the rest */
2809 if (rp != ms->rp_state.from_dst_file) {
2810 qemu_fclose(rp);
2811 rp = ms->rp_state.from_dst_file;
2813 ms->rp_state.error = false;
2814 goto retry;
2818 trace_source_return_path_thread_bad_end();
2819 mark_source_rp_bad(ms);
2822 trace_source_return_path_thread_end();
2823 ms->rp_state.from_dst_file = NULL;
2824 qemu_fclose(rp);
2825 rcu_unregister_thread();
2826 return NULL;
2829 static int open_return_path_on_source(MigrationState *ms,
2830 bool create_thread)
2833 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2834 if (!ms->rp_state.from_dst_file) {
2835 return -1;
2838 trace_open_return_path_on_source();
2840 if (!create_thread) {
2841 /* We're done */
2842 return 0;
2845 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2846 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2848 trace_open_return_path_on_source_continue();
2850 return 0;
2853 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2854 static int await_return_path_close_on_source(MigrationState *ms)
2857 * If this is a normal exit then the destination will send a SHUT and the
2858 * rp_thread will exit, however if there's an error we need to cause
2859 * it to exit.
2861 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2863 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2864 * waiting for the destination.
2866 qemu_file_shutdown(ms->rp_state.from_dst_file);
2867 mark_source_rp_bad(ms);
2869 trace_await_return_path_close_on_source_joining();
2870 qemu_thread_join(&ms->rp_state.rp_thread);
2871 trace_await_return_path_close_on_source_close();
2872 return ms->rp_state.error;
2876 * Switch from normal iteration to postcopy
2877 * Returns non-0 on error
2879 static int postcopy_start(MigrationState *ms)
2881 int ret;
2882 QIOChannelBuffer *bioc;
2883 QEMUFile *fb;
2884 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2885 int64_t bandwidth = migrate_max_postcopy_bandwidth();
2886 bool restart_block = false;
2887 int cur_state = MIGRATION_STATUS_ACTIVE;
2888 if (!migrate_pause_before_switchover()) {
2889 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2890 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2893 trace_postcopy_start();
2894 qemu_mutex_lock_iothread();
2895 trace_postcopy_start_set_run();
2897 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2898 global_state_store();
2899 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2900 if (ret < 0) {
2901 goto fail;
2904 ret = migration_maybe_pause(ms, &cur_state,
2905 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2906 if (ret < 0) {
2907 goto fail;
2910 ret = bdrv_inactivate_all();
2911 if (ret < 0) {
2912 goto fail;
2914 restart_block = true;
2917 * Cause any non-postcopiable, but iterative devices to
2918 * send out their final data.
2920 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2923 * in Finish migrate and with the io-lock held everything should
2924 * be quiet, but we've potentially still got dirty pages and we
2925 * need to tell the destination to throw any pages it's already received
2926 * that are dirty
2928 if (migrate_postcopy_ram()) {
2929 if (ram_postcopy_send_discard_bitmap(ms)) {
2930 error_report("postcopy send discard bitmap failed");
2931 goto fail;
2936 * send rest of state - note things that are doing postcopy
2937 * will notice we're in POSTCOPY_ACTIVE and not actually
2938 * wrap their state up here
2940 /* 0 max-postcopy-bandwidth means unlimited */
2941 if (!bandwidth) {
2942 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2943 } else {
2944 qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2946 if (migrate_postcopy_ram()) {
2947 /* Ping just for debugging, helps line traces up */
2948 qemu_savevm_send_ping(ms->to_dst_file, 2);
2952 * While loading the device state we may trigger page transfer
2953 * requests and the fd must be free to process those, and thus
2954 * the destination must read the whole device state off the fd before
2955 * it starts processing it. Unfortunately the ad-hoc migration format
2956 * doesn't allow the destination to know the size to read without fully
2957 * parsing it through each devices load-state code (especially the open
2958 * coded devices that use get/put).
2959 * So we wrap the device state up in a package with a length at the start;
2960 * to do this we use a qemu_buf to hold the whole of the device state.
2962 bioc = qio_channel_buffer_new(4096);
2963 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2964 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2965 object_unref(OBJECT(bioc));
2968 * Make sure the receiver can get incoming pages before we send the rest
2969 * of the state
2971 qemu_savevm_send_postcopy_listen(fb);
2973 qemu_savevm_state_complete_precopy(fb, false, false);
2974 if (migrate_postcopy_ram()) {
2975 qemu_savevm_send_ping(fb, 3);
2978 qemu_savevm_send_postcopy_run(fb);
2980 /* <><> end of stuff going into the package */
2982 /* Last point of recovery; as soon as we send the package the destination
2983 * can open devices and potentially start running.
2984 * Lets just check again we've not got any errors.
2986 ret = qemu_file_get_error(ms->to_dst_file);
2987 if (ret) {
2988 error_report("postcopy_start: Migration stream errored (pre package)");
2989 goto fail_closefb;
2992 restart_block = false;
2994 /* Now send that blob */
2995 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2996 goto fail_closefb;
2998 qemu_fclose(fb);
3000 /* Send a notify to give a chance for anything that needs to happen
3001 * at the transition to postcopy and after the device state; in particular
3002 * spice needs to trigger a transition now
3004 ms->postcopy_after_devices = true;
3005 notifier_list_notify(&migration_state_notifiers, ms);
3007 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
3009 qemu_mutex_unlock_iothread();
3011 if (migrate_postcopy_ram()) {
3013 * Although this ping is just for debug, it could potentially be
3014 * used for getting a better measurement of downtime at the source.
3016 qemu_savevm_send_ping(ms->to_dst_file, 4);
3019 if (migrate_release_ram()) {
3020 ram_postcopy_migrated_memory_release(ms);
3023 ret = qemu_file_get_error(ms->to_dst_file);
3024 if (ret) {
3025 error_report("postcopy_start: Migration stream errored");
3026 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
3027 MIGRATION_STATUS_FAILED);
3030 return ret;
3032 fail_closefb:
3033 qemu_fclose(fb);
3034 fail:
3035 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
3036 MIGRATION_STATUS_FAILED);
3037 if (restart_block) {
3038 /* A failure happened early enough that we know the destination hasn't
3039 * accessed block devices, so we're safe to recover.
3041 Error *local_err = NULL;
3043 bdrv_invalidate_cache_all(&local_err);
3044 if (local_err) {
3045 error_report_err(local_err);
3048 qemu_mutex_unlock_iothread();
3049 return -1;
3053 * migration_maybe_pause: Pause if required to by
3054 * migrate_pause_before_switchover called with the iothread locked
3055 * Returns: 0 on success
3057 static int migration_maybe_pause(MigrationState *s,
3058 int *current_active_state,
3059 int new_state)
3061 if (!migrate_pause_before_switchover()) {
3062 return 0;
3065 /* Since leaving this state is not atomic with posting the semaphore
3066 * it's possible that someone could have issued multiple migrate_continue
3067 * and the semaphore is incorrectly positive at this point;
3068 * the docs say it's undefined to reinit a semaphore that's already
3069 * init'd, so use timedwait to eat up any existing posts.
3071 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
3072 /* This block intentionally left blank */
3076 * If the migration is cancelled when it is in the completion phase,
3077 * the migration state is set to MIGRATION_STATUS_CANCELLING.
3078 * So we don't need to wait a semaphore, otherwise we would always
3079 * wait for the 'pause_sem' semaphore.
3081 if (s->state != MIGRATION_STATUS_CANCELLING) {
3082 qemu_mutex_unlock_iothread();
3083 migrate_set_state(&s->state, *current_active_state,
3084 MIGRATION_STATUS_PRE_SWITCHOVER);
3085 qemu_sem_wait(&s->pause_sem);
3086 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
3087 new_state);
3088 *current_active_state = new_state;
3089 qemu_mutex_lock_iothread();
3092 return s->state == new_state ? 0 : -EINVAL;
3096 * migration_completion: Used by migration_thread when there's not much left.
3097 * The caller 'breaks' the loop when this returns.
3099 * @s: Current migration state
3101 static void migration_completion(MigrationState *s)
3103 int ret;
3104 int current_active_state = s->state;
3106 if (s->state == MIGRATION_STATUS_ACTIVE) {
3107 qemu_mutex_lock_iothread();
3108 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3109 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
3110 s->vm_was_running = runstate_is_running();
3111 ret = global_state_store();
3113 if (!ret) {
3114 bool inactivate = !migrate_colo_enabled();
3115 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
3116 if (ret >= 0) {
3117 ret = migration_maybe_pause(s, &current_active_state,
3118 MIGRATION_STATUS_DEVICE);
3120 if (ret >= 0) {
3121 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
3122 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
3123 inactivate);
3125 if (inactivate && ret >= 0) {
3126 s->block_inactive = true;
3129 qemu_mutex_unlock_iothread();
3131 if (ret < 0) {
3132 goto fail;
3134 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3135 trace_migration_completion_postcopy_end();
3137 qemu_savevm_state_complete_postcopy(s->to_dst_file);
3138 trace_migration_completion_postcopy_end_after_complete();
3139 } else if (s->state == MIGRATION_STATUS_CANCELLING) {
3140 goto fail;
3144 * If rp was opened we must clean up the thread before
3145 * cleaning everything else up (since if there are no failures
3146 * it will wait for the destination to send it's status in
3147 * a SHUT command).
3149 if (s->rp_state.from_dst_file) {
3150 int rp_error;
3151 trace_migration_return_path_end_before();
3152 rp_error = await_return_path_close_on_source(s);
3153 trace_migration_return_path_end_after(rp_error);
3154 if (rp_error) {
3155 goto fail_invalidate;
3159 if (qemu_file_get_error(s->to_dst_file)) {
3160 trace_migration_completion_file_err();
3161 goto fail_invalidate;
3164 if (!migrate_colo_enabled()) {
3165 migrate_set_state(&s->state, current_active_state,
3166 MIGRATION_STATUS_COMPLETED);
3169 return;
3171 fail_invalidate:
3172 /* If not doing postcopy, vm_start() will be called: let's regain
3173 * control on images.
3175 if (s->state == MIGRATION_STATUS_ACTIVE ||
3176 s->state == MIGRATION_STATUS_DEVICE) {
3177 Error *local_err = NULL;
3179 qemu_mutex_lock_iothread();
3180 bdrv_invalidate_cache_all(&local_err);
3181 if (local_err) {
3182 error_report_err(local_err);
3183 } else {
3184 s->block_inactive = false;
3186 qemu_mutex_unlock_iothread();
3189 fail:
3190 migrate_set_state(&s->state, current_active_state,
3191 MIGRATION_STATUS_FAILED);
3195 * bg_migration_completion: Used by bg_migration_thread when after all the
3196 * RAM has been saved. The caller 'breaks' the loop when this returns.
3198 * @s: Current migration state
3200 static void bg_migration_completion(MigrationState *s)
3202 int current_active_state = s->state;
3205 * Stop tracking RAM writes - un-protect memory, un-register UFFD
3206 * memory ranges, flush kernel wait queues and wake up threads
3207 * waiting for write fault to be resolved.
3209 ram_write_tracking_stop();
3211 if (s->state == MIGRATION_STATUS_ACTIVE) {
3213 * By this moment we have RAM content saved into the migration stream.
3214 * The next step is to flush the non-RAM content (device state)
3215 * right after the ram content. The device state has been stored into
3216 * the temporary buffer before RAM saving started.
3218 qemu_put_buffer(s->to_dst_file, s->bioc->data, s->bioc->usage);
3219 qemu_fflush(s->to_dst_file);
3220 } else if (s->state == MIGRATION_STATUS_CANCELLING) {
3221 goto fail;
3224 if (qemu_file_get_error(s->to_dst_file)) {
3225 trace_migration_completion_file_err();
3226 goto fail;
3229 migrate_set_state(&s->state, current_active_state,
3230 MIGRATION_STATUS_COMPLETED);
3231 return;
3233 fail:
3234 migrate_set_state(&s->state, current_active_state,
3235 MIGRATION_STATUS_FAILED);
3238 bool migrate_colo_enabled(void)
3240 MigrationState *s = migrate_get_current();
3241 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
3244 typedef enum MigThrError {
3245 /* No error detected */
3246 MIG_THR_ERR_NONE = 0,
3247 /* Detected error, but resumed successfully */
3248 MIG_THR_ERR_RECOVERED = 1,
3249 /* Detected fatal error, need to exit */
3250 MIG_THR_ERR_FATAL = 2,
3251 } MigThrError;
3253 static int postcopy_resume_handshake(MigrationState *s)
3255 qemu_savevm_send_postcopy_resume(s->to_dst_file);
3257 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3258 qemu_sem_wait(&s->rp_state.rp_sem);
3261 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3262 return 0;
3265 return -1;
3268 /* Return zero if success, or <0 for error */
3269 static int postcopy_do_resume(MigrationState *s)
3271 int ret;
3274 * Call all the resume_prepare() hooks, so that modules can be
3275 * ready for the migration resume.
3277 ret = qemu_savevm_state_resume_prepare(s);
3278 if (ret) {
3279 error_report("%s: resume_prepare() failure detected: %d",
3280 __func__, ret);
3281 return ret;
3285 * Last handshake with destination on the resume (destination will
3286 * switch to postcopy-active afterwards)
3288 ret = postcopy_resume_handshake(s);
3289 if (ret) {
3290 error_report("%s: handshake failed: %d", __func__, ret);
3291 return ret;
3294 return 0;
3298 * We don't return until we are in a safe state to continue current
3299 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3300 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3302 static MigThrError postcopy_pause(MigrationState *s)
3304 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
3306 while (true) {
3307 QEMUFile *file;
3309 /* Current channel is possibly broken. Release it. */
3310 assert(s->to_dst_file);
3311 qemu_mutex_lock(&s->qemu_file_lock);
3312 file = s->to_dst_file;
3313 s->to_dst_file = NULL;
3314 qemu_mutex_unlock(&s->qemu_file_lock);
3316 qemu_file_shutdown(file);
3317 qemu_fclose(file);
3319 migrate_set_state(&s->state, s->state,
3320 MIGRATION_STATUS_POSTCOPY_PAUSED);
3322 error_report("Detected IO failure for postcopy. "
3323 "Migration paused.");
3326 * We wait until things fixed up. Then someone will setup the
3327 * status back for us.
3329 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
3330 qemu_sem_wait(&s->postcopy_pause_sem);
3333 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3334 /* Woken up by a recover procedure. Give it a shot */
3337 * Firstly, let's wake up the return path now, with a new
3338 * return path channel.
3340 qemu_sem_post(&s->postcopy_pause_rp_sem);
3342 /* Do the resume logic */
3343 if (postcopy_do_resume(s) == 0) {
3344 /* Let's continue! */
3345 trace_postcopy_pause_continued();
3346 return MIG_THR_ERR_RECOVERED;
3347 } else {
3349 * Something wrong happened during the recovery, let's
3350 * pause again. Pause is always better than throwing
3351 * data away.
3353 continue;
3355 } else {
3356 /* This is not right... Time to quit. */
3357 return MIG_THR_ERR_FATAL;
3362 static MigThrError migration_detect_error(MigrationState *s)
3364 int ret;
3365 int state = s->state;
3366 Error *local_error = NULL;
3368 if (state == MIGRATION_STATUS_CANCELLING ||
3369 state == MIGRATION_STATUS_CANCELLED) {
3370 /* End the migration, but don't set the state to failed */
3371 return MIG_THR_ERR_FATAL;
3374 /* Try to detect any file errors */
3375 ret = qemu_file_get_error_obj(s->to_dst_file, &local_error);
3376 if (!ret) {
3377 /* Everything is fine */
3378 assert(!local_error);
3379 return MIG_THR_ERR_NONE;
3382 if (local_error) {
3383 migrate_set_error(s, local_error);
3384 error_free(local_error);
3387 if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
3389 * For postcopy, we allow the network to be down for a
3390 * while. After that, it can be continued by a
3391 * recovery phase.
3393 return postcopy_pause(s);
3394 } else {
3396 * For precopy (or postcopy with error outside IO), we fail
3397 * with no time.
3399 migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
3400 trace_migration_thread_file_err();
3402 /* Time to stop the migration, now. */
3403 return MIG_THR_ERR_FATAL;
3407 /* How many bytes have we transferred since the beginning of the migration */
3408 static uint64_t migration_total_bytes(MigrationState *s)
3410 return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3413 static void migration_calculate_complete(MigrationState *s)
3415 uint64_t bytes = migration_total_bytes(s);
3416 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3417 int64_t transfer_time;
3419 s->total_time = end_time - s->start_time;
3420 if (!s->downtime) {
3422 * It's still not set, so we are precopy migration. For
3423 * postcopy, downtime is calculated during postcopy_start().
3425 s->downtime = end_time - s->downtime_start;
3428 transfer_time = s->total_time - s->setup_time;
3429 if (transfer_time) {
3430 s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3434 static void update_iteration_initial_status(MigrationState *s)
3437 * Update these three fields at the same time to avoid mismatch info lead
3438 * wrong speed calculation.
3440 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3441 s->iteration_initial_bytes = migration_total_bytes(s);
3442 s->iteration_initial_pages = ram_get_total_transferred_pages();
3445 static void migration_update_counters(MigrationState *s,
3446 int64_t current_time)
3448 uint64_t transferred, transferred_pages, time_spent;
3449 uint64_t current_bytes; /* bytes transferred since the beginning */
3450 double bandwidth;
3452 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3453 return;
3456 current_bytes = migration_total_bytes(s);
3457 transferred = current_bytes - s->iteration_initial_bytes;
3458 time_spent = current_time - s->iteration_start_time;
3459 bandwidth = (double)transferred / time_spent;
3460 s->threshold_size = bandwidth * s->parameters.downtime_limit;
3462 s->mbps = (((double) transferred * 8.0) /
3463 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3465 transferred_pages = ram_get_total_transferred_pages() -
3466 s->iteration_initial_pages;
3467 s->pages_per_second = (double) transferred_pages /
3468 (((double) time_spent / 1000.0));
3471 * if we haven't sent anything, we don't want to
3472 * recalculate. 10000 is a small enough number for our purposes
3474 if (ram_counters.dirty_pages_rate && transferred > 10000) {
3475 s->expected_downtime = ram_counters.remaining / bandwidth;
3478 qemu_file_reset_rate_limit(s->to_dst_file);
3480 update_iteration_initial_status(s);
3482 trace_migrate_transferred(transferred, time_spent,
3483 bandwidth, s->threshold_size);
3486 /* Migration thread iteration status */
3487 typedef enum {
3488 MIG_ITERATE_RESUME, /* Resume current iteration */
3489 MIG_ITERATE_SKIP, /* Skip current iteration */
3490 MIG_ITERATE_BREAK, /* Break the loop */
3491 } MigIterateState;
3494 * Return true if continue to the next iteration directly, false
3495 * otherwise.
3497 static MigIterateState migration_iteration_run(MigrationState *s)
3499 uint64_t pending_size, pend_pre, pend_compat, pend_post;
3500 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3502 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3503 &pend_compat, &pend_post);
3504 pending_size = pend_pre + pend_compat + pend_post;
3506 trace_migrate_pending(pending_size, s->threshold_size,
3507 pend_pre, pend_compat, pend_post);
3509 if (pending_size && pending_size >= s->threshold_size) {
3510 /* Still a significant amount to transfer */
3511 if (!in_postcopy && pend_pre <= s->threshold_size &&
3512 qatomic_read(&s->start_postcopy)) {
3513 if (postcopy_start(s)) {
3514 error_report("%s: postcopy failed to start", __func__);
3516 return MIG_ITERATE_SKIP;
3518 /* Just another iteration step */
3519 qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
3520 } else {
3521 trace_migration_thread_low_pending(pending_size);
3522 migration_completion(s);
3523 return MIG_ITERATE_BREAK;
3526 return MIG_ITERATE_RESUME;
3529 static void migration_iteration_finish(MigrationState *s)
3531 /* If we enabled cpu throttling for auto-converge, turn it off. */
3532 cpu_throttle_stop();
3534 qemu_mutex_lock_iothread();
3535 switch (s->state) {
3536 case MIGRATION_STATUS_COMPLETED:
3537 migration_calculate_complete(s);
3538 runstate_set(RUN_STATE_POSTMIGRATE);
3539 break;
3541 case MIGRATION_STATUS_ACTIVE:
3543 * We should really assert here, but since it's during
3544 * migration, let's try to reduce the usage of assertions.
3546 if (!migrate_colo_enabled()) {
3547 error_report("%s: critical error: calling COLO code without "
3548 "COLO enabled", __func__);
3550 migrate_start_colo_process(s);
3552 * Fixme: we will run VM in COLO no matter its old running state.
3553 * After exited COLO, we will keep running.
3555 s->vm_was_running = true;
3556 /* Fallthrough */
3557 case MIGRATION_STATUS_FAILED:
3558 case MIGRATION_STATUS_CANCELLED:
3559 case MIGRATION_STATUS_CANCELLING:
3560 if (s->vm_was_running) {
3561 vm_start();
3562 } else {
3563 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3564 runstate_set(RUN_STATE_POSTMIGRATE);
3567 break;
3569 default:
3570 /* Should not reach here, but if so, forgive the VM. */
3571 error_report("%s: Unknown ending state %d", __func__, s->state);
3572 break;
3574 migrate_fd_cleanup_schedule(s);
3575 qemu_mutex_unlock_iothread();
3578 static void bg_migration_iteration_finish(MigrationState *s)
3580 qemu_mutex_lock_iothread();
3581 switch (s->state) {
3582 case MIGRATION_STATUS_COMPLETED:
3583 migration_calculate_complete(s);
3584 break;
3586 case MIGRATION_STATUS_ACTIVE:
3587 case MIGRATION_STATUS_FAILED:
3588 case MIGRATION_STATUS_CANCELLED:
3589 case MIGRATION_STATUS_CANCELLING:
3590 break;
3592 default:
3593 /* Should not reach here, but if so, forgive the VM. */
3594 error_report("%s: Unknown ending state %d", __func__, s->state);
3595 break;
3598 migrate_fd_cleanup_schedule(s);
3599 qemu_mutex_unlock_iothread();
3603 * Return true if continue to the next iteration directly, false
3604 * otherwise.
3606 static MigIterateState bg_migration_iteration_run(MigrationState *s)
3608 int res;
3610 res = qemu_savevm_state_iterate(s->to_dst_file, false);
3611 if (res > 0) {
3612 bg_migration_completion(s);
3613 return MIG_ITERATE_BREAK;
3616 return MIG_ITERATE_RESUME;
3619 void migration_make_urgent_request(void)
3621 qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3624 void migration_consume_urgent_request(void)
3626 qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3629 /* Returns true if the rate limiting was broken by an urgent request */
3630 bool migration_rate_limit(void)
3632 int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3633 MigrationState *s = migrate_get_current();
3635 bool urgent = false;
3636 migration_update_counters(s, now);
3637 if (qemu_file_rate_limit(s->to_dst_file)) {
3639 if (qemu_file_get_error(s->to_dst_file)) {
3640 return false;
3643 * Wait for a delay to do rate limiting OR
3644 * something urgent to post the semaphore.
3646 int ms = s->iteration_start_time + BUFFER_DELAY - now;
3647 trace_migration_rate_limit_pre(ms);
3648 if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3650 * We were woken by one or more urgent things but
3651 * the timedwait will have consumed one of them.
3652 * The service routine for the urgent wake will dec
3653 * the semaphore itself for each item it consumes,
3654 * so add this one we just eat back.
3656 qemu_sem_post(&s->rate_limit_sem);
3657 urgent = true;
3659 trace_migration_rate_limit_post(urgent);
3661 return urgent;
3665 * Master migration thread on the source VM.
3666 * It drives the migration and pumps the data down the outgoing channel.
3668 static void *migration_thread(void *opaque)
3670 MigrationState *s = opaque;
3671 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3672 MigThrError thr_error;
3673 bool urgent = false;
3675 rcu_register_thread();
3677 object_ref(OBJECT(s));
3678 update_iteration_initial_status(s);
3680 qemu_savevm_state_header(s->to_dst_file);
3683 * If we opened the return path, we need to make sure dst has it
3684 * opened as well.
3686 if (s->rp_state.from_dst_file) {
3687 /* Now tell the dest that it should open its end so it can reply */
3688 qemu_savevm_send_open_return_path(s->to_dst_file);
3690 /* And do a ping that will make stuff easier to debug */
3691 qemu_savevm_send_ping(s->to_dst_file, 1);
3694 if (migrate_postcopy()) {
3696 * Tell the destination that we *might* want to do postcopy later;
3697 * if the other end can't do postcopy it should fail now, nice and
3698 * early.
3700 qemu_savevm_send_postcopy_advise(s->to_dst_file);
3703 if (migrate_colo_enabled()) {
3704 /* Notify migration destination that we enable COLO */
3705 qemu_savevm_send_colo_enable(s->to_dst_file);
3708 qemu_savevm_state_setup(s->to_dst_file);
3710 if (qemu_savevm_state_guest_unplug_pending()) {
3711 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3712 MIGRATION_STATUS_WAIT_UNPLUG);
3714 while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3715 qemu_savevm_state_guest_unplug_pending()) {
3716 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3719 migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG,
3720 MIGRATION_STATUS_ACTIVE);
3723 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3724 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3725 MIGRATION_STATUS_ACTIVE);
3727 trace_migration_thread_setup_complete();
3729 while (migration_is_active(s)) {
3730 if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3731 MigIterateState iter_state = migration_iteration_run(s);
3732 if (iter_state == MIG_ITERATE_SKIP) {
3733 continue;
3734 } else if (iter_state == MIG_ITERATE_BREAK) {
3735 break;
3740 * Try to detect any kind of failures, and see whether we
3741 * should stop the migration now.
3743 thr_error = migration_detect_error(s);
3744 if (thr_error == MIG_THR_ERR_FATAL) {
3745 /* Stop migration */
3746 break;
3747 } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3749 * Just recovered from a e.g. network failure, reset all
3750 * the local variables. This is important to avoid
3751 * breaking transferred_bytes and bandwidth calculation
3753 update_iteration_initial_status(s);
3756 urgent = migration_rate_limit();
3759 trace_migration_thread_after_loop();
3760 migration_iteration_finish(s);
3761 object_unref(OBJECT(s));
3762 rcu_unregister_thread();
3763 return NULL;
3766 static void bg_migration_vm_start_bh(void *opaque)
3768 MigrationState *s = opaque;
3770 qemu_bh_delete(s->vm_start_bh);
3771 s->vm_start_bh = NULL;
3773 vm_start();
3774 s->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - s->downtime_start;
3778 * Background snapshot thread, based on live migration code.
3779 * This is an alternative implementation of live migration mechanism
3780 * introduced specifically to support background snapshots.
3782 * It takes advantage of userfault_fd write protection mechanism introduced
3783 * in v5.7 kernel. Compared to existing dirty page logging migration much
3784 * lesser stream traffic is produced resulting in smaller snapshot images,
3785 * simply cause of no page duplicates can get into the stream.
3787 * Another key point is that generated vmstate stream reflects machine state
3788 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
3789 * mechanism, which effectively results in that saved snapshot is the state of VM
3790 * at the end of the process.
3792 static void *bg_migration_thread(void *opaque)
3794 MigrationState *s = opaque;
3795 int64_t setup_start;
3796 MigThrError thr_error;
3797 QEMUFile *fb;
3798 bool early_fail = true;
3800 rcu_register_thread();
3801 object_ref(OBJECT(s));
3803 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
3805 setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3807 * We want to save vmstate for the moment when migration has been
3808 * initiated but also we want to save RAM content while VM is running.
3809 * The RAM content should appear first in the vmstate. So, we first
3810 * stash the non-RAM part of the vmstate to the temporary buffer,
3811 * then write RAM part of the vmstate to the migration stream
3812 * with vCPUs running and, finally, write stashed non-RAM part of
3813 * the vmstate from the buffer to the migration stream.
3815 s->bioc = qio_channel_buffer_new(128 * 1024);
3816 qio_channel_set_name(QIO_CHANNEL(s->bioc), "vmstate-buffer");
3817 fb = qemu_fopen_channel_output(QIO_CHANNEL(s->bioc));
3818 object_unref(OBJECT(s->bioc));
3820 update_iteration_initial_status(s);
3822 qemu_savevm_state_header(s->to_dst_file);
3823 qemu_savevm_state_setup(s->to_dst_file);
3825 if (qemu_savevm_state_guest_unplug_pending()) {
3826 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3827 MIGRATION_STATUS_WAIT_UNPLUG);
3829 while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3830 qemu_savevm_state_guest_unplug_pending()) {
3831 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3834 migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG,
3835 MIGRATION_STATUS_ACTIVE);
3836 } else {
3837 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3838 MIGRATION_STATUS_ACTIVE);
3840 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3842 trace_migration_thread_setup_complete();
3843 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3845 qemu_mutex_lock_iothread();
3848 * If VM is currently in suspended state, then, to make a valid runstate
3849 * transition in vm_stop_force_state() we need to wakeup it up.
3851 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
3852 s->vm_was_running = runstate_is_running();
3854 if (global_state_store()) {
3855 goto fail;
3857 /* Forcibly stop VM before saving state of vCPUs and devices */
3858 if (vm_stop_force_state(RUN_STATE_PAUSED)) {
3859 goto fail;
3862 * Put vCPUs in sync with shadow context structures, then
3863 * save their state to channel-buffer along with devices.
3865 cpu_synchronize_all_states();
3866 if (qemu_savevm_state_complete_precopy_non_iterable(fb, false, false)) {
3867 goto fail;
3869 /* Now initialize UFFD context and start tracking RAM writes */
3870 if (ram_write_tracking_start()) {
3871 goto fail;
3873 early_fail = false;
3876 * Start VM from BH handler to avoid write-fault lock here.
3877 * UFFD-WP protection for the whole RAM is already enabled so
3878 * calling VM state change notifiers from vm_start() would initiate
3879 * writes to virtio VQs memory which is in write-protected region.
3881 s->vm_start_bh = qemu_bh_new(bg_migration_vm_start_bh, s);
3882 qemu_bh_schedule(s->vm_start_bh);
3884 qemu_mutex_unlock_iothread();
3886 while (migration_is_active(s)) {
3887 MigIterateState iter_state = bg_migration_iteration_run(s);
3888 if (iter_state == MIG_ITERATE_SKIP) {
3889 continue;
3890 } else if (iter_state == MIG_ITERATE_BREAK) {
3891 break;
3895 * Try to detect any kind of failures, and see whether we
3896 * should stop the migration now.
3898 thr_error = migration_detect_error(s);
3899 if (thr_error == MIG_THR_ERR_FATAL) {
3900 /* Stop migration */
3901 break;
3904 migration_update_counters(s, qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
3907 trace_migration_thread_after_loop();
3909 fail:
3910 if (early_fail) {
3911 migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
3912 MIGRATION_STATUS_FAILED);
3913 qemu_mutex_unlock_iothread();
3916 bg_migration_iteration_finish(s);
3918 qemu_fclose(fb);
3919 object_unref(OBJECT(s));
3920 rcu_unregister_thread();
3922 return NULL;
3925 void migrate_fd_connect(MigrationState *s, Error *error_in)
3927 Error *local_err = NULL;
3928 int64_t rate_limit;
3929 bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
3931 s->expected_downtime = s->parameters.downtime_limit;
3932 if (resume) {
3933 assert(s->cleanup_bh);
3934 } else {
3935 assert(!s->cleanup_bh);
3936 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s);
3938 if (error_in) {
3939 migrate_fd_error(s, error_in);
3940 migrate_fd_cleanup(s);
3941 return;
3944 if (resume) {
3945 /* This is a resumed migration */
3946 rate_limit = s->parameters.max_postcopy_bandwidth /
3947 XFER_LIMIT_RATIO;
3948 } else {
3949 /* This is a fresh new migration */
3950 rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
3952 /* Notify before starting migration thread */
3953 notifier_list_notify(&migration_state_notifiers, s);
3956 qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
3957 qemu_file_set_blocking(s->to_dst_file, true);
3960 * Open the return path. For postcopy, it is used exclusively. For
3961 * precopy, only if user specified "return-path" capability would
3962 * QEMU uses the return path.
3964 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3965 if (open_return_path_on_source(s, !resume)) {
3966 error_report("Unable to open return-path for postcopy");
3967 migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3968 migrate_fd_cleanup(s);
3969 return;
3973 if (resume) {
3974 /* Wakeup the main migration thread to do the recovery */
3975 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3976 MIGRATION_STATUS_POSTCOPY_RECOVER);
3977 qemu_sem_post(&s->postcopy_pause_sem);
3978 return;
3981 if (multifd_save_setup(&local_err) != 0) {
3982 error_report_err(local_err);
3983 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3984 MIGRATION_STATUS_FAILED);
3985 migrate_fd_cleanup(s);
3986 return;
3989 if (migrate_background_snapshot()) {
3990 qemu_thread_create(&s->thread, "bg_snapshot",
3991 bg_migration_thread, s, QEMU_THREAD_JOINABLE);
3992 } else {
3993 qemu_thread_create(&s->thread, "live_migration",
3994 migration_thread, s, QEMU_THREAD_JOINABLE);
3996 s->migration_thread_running = true;
3999 void migration_global_dump(Monitor *mon)
4001 MigrationState *ms = migrate_get_current();
4003 monitor_printf(mon, "globals:\n");
4004 monitor_printf(mon, "store-global-state: %s\n",
4005 ms->store_global_state ? "on" : "off");
4006 monitor_printf(mon, "only-migratable: %s\n",
4007 only_migratable ? "on" : "off");
4008 monitor_printf(mon, "send-configuration: %s\n",
4009 ms->send_configuration ? "on" : "off");
4010 monitor_printf(mon, "send-section-footer: %s\n",
4011 ms->send_section_footer ? "on" : "off");
4012 monitor_printf(mon, "decompress-error-check: %s\n",
4013 ms->decompress_error_check ? "on" : "off");
4014 monitor_printf(mon, "clear-bitmap-shift: %u\n",
4015 ms->clear_bitmap_shift);
4018 #define DEFINE_PROP_MIG_CAP(name, x) \
4019 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
4021 static Property migration_properties[] = {
4022 DEFINE_PROP_BOOL("store-global-state", MigrationState,
4023 store_global_state, true),
4024 DEFINE_PROP_BOOL("send-configuration", MigrationState,
4025 send_configuration, true),
4026 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
4027 send_section_footer, true),
4028 DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
4029 decompress_error_check, true),
4030 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState,
4031 clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT),
4033 /* Migration parameters */
4034 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
4035 parameters.compress_level,
4036 DEFAULT_MIGRATE_COMPRESS_LEVEL),
4037 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
4038 parameters.compress_threads,
4039 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
4040 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
4041 parameters.compress_wait_thread, true),
4042 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
4043 parameters.decompress_threads,
4044 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
4045 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState,
4046 parameters.throttle_trigger_threshold,
4047 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD),
4048 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
4049 parameters.cpu_throttle_initial,
4050 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
4051 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
4052 parameters.cpu_throttle_increment,
4053 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
4054 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState,
4055 parameters.cpu_throttle_tailslow, false),
4056 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
4057 parameters.max_bandwidth, MAX_THROTTLE),
4058 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
4059 parameters.downtime_limit,
4060 DEFAULT_MIGRATE_SET_DOWNTIME),
4061 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
4062 parameters.x_checkpoint_delay,
4063 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
4064 DEFINE_PROP_UINT8("multifd-channels", MigrationState,
4065 parameters.multifd_channels,
4066 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
4067 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState,
4068 parameters.multifd_compression,
4069 DEFAULT_MIGRATE_MULTIFD_COMPRESSION),
4070 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState,
4071 parameters.multifd_zlib_level,
4072 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL),
4073 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState,
4074 parameters.multifd_zstd_level,
4075 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL),
4076 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
4077 parameters.xbzrle_cache_size,
4078 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
4079 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,