Merge remote-tracking branch 'remotes/dgilbert/tags/pull-migration-20200925a' into...
[qemu.git] / migration / migration.c
1 /*
2 * QEMU live migration
3 *
4 * Copyright IBM, Corp. 2008
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
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.
14 */
15
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 "rdma.h"
27 #include "ram.h"
28 #include "migration/global_state.h"
29 #include "migration/misc.h"
30 #include "migration.h"
31 #include "savevm.h"
32 #include "qemu-file-channel.h"
33 #include "qemu-file.h"
34 #include "migration/vmstate.h"
35 #include "block/block.h"
36 #include "qapi/error.h"
37 #include "qapi/clone-visitor.h"
38 #include "qapi/qapi-visit-sockets.h"
39 #include "qapi/qapi-commands-migration.h"
40 #include "qapi/qapi-events-migration.h"
41 #include "qapi/qmp/qerror.h"
42 #include "qapi/qmp/qnull.h"
43 #include "qemu/rcu.h"
44 #include "block.h"
45 #include "postcopy-ram.h"
46 #include "qemu/thread.h"
47 #include "trace.h"
48 #include "exec/target_page.h"
49 #include "io/channel-buffer.h"
50 #include "migration/colo.h"
51 #include "hw/boards.h"
52 #include "hw/qdev-properties.h"
53 #include "monitor/monitor.h"
54 #include "net/announce.h"
55 #include "qemu/queue.h"
56
57 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
58
59 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
60 * data. */
61 #define BUFFER_DELAY 100
62 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
63
64 /* Time in milliseconds we are allowed to stop the source,
65 * for sending the last part */
66 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
67
68 /* Maximum migrate downtime set to 2000 seconds */
69 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
70 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
71
72 /* Default compression thread count */
73 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
74 /* Default decompression thread count, usually decompression is at
75 * least 4 times as fast as compression.*/
76 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
77 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
78 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
79 /* Define default autoconverge cpu throttle migration parameters */
80 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
81 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
82 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
83
84 /* Migration XBZRLE default cache size */
85 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
86
87 /* The delay time (in ms) between two COLO checkpoints */
88 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
89 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
90
91 /* Background transfer rate for postcopy, 0 means unlimited, note
92 * that page requests can still exceed this limit.
93 */
94 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
95
96 /*
97 * Parameters for self_announce_delay giving a stream of RARP/ARP
98 * packets after migration.
99 */
100 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
101 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
102 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
103 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
104
105 static NotifierList migration_state_notifiers =
106 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
107
108 static bool deferred_incoming;
109
110 /* Messages sent on the return path from destination to source */
111 enum mig_rp_message_type {
112 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
113 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
114 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
115
116 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
117 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
118 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
119 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
120
121 MIG_RP_MSG_MAX
122 };
123
124 /* When we add fault tolerance, we could have several
125 migrations at once. For now we don't need to add
126 dynamic creation of migration */
127
128 static MigrationState *current_migration;
129 static MigrationIncomingState *current_incoming;
130
131 static bool migration_object_check(MigrationState *ms, Error **errp);
132 static int migration_maybe_pause(MigrationState *s,
133 int *current_active_state,
134 int new_state);
135 static void migrate_fd_cancel(MigrationState *s);
136
137 void migration_object_init(void)
138 {
139 MachineState *ms = MACHINE(qdev_get_machine());
140 Error *err = NULL;
141
142 /* This can only be called once. */
143 assert(!current_migration);
144 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
145
146 /*
147 * Init the migrate incoming object as well no matter whether
148 * we'll use it or not.
149 */
150 assert(!current_incoming);
151 current_incoming = g_new0(MigrationIncomingState, 1);
152 current_incoming->state = MIGRATION_STATUS_NONE;
153 current_incoming->postcopy_remote_fds =
154 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
155 qemu_mutex_init(&current_incoming->rp_mutex);
156 qemu_event_init(&current_incoming->main_thread_load_event, false);
157 qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
158 qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
159
160 init_dirty_bitmap_incoming_migration();
161
162 if (!migration_object_check(current_migration, &err)) {
163 error_report_err(err);
164 exit(1);
165 }
166
167 /*
168 * We cannot really do this in migration_instance_init() since at
169 * that time global properties are not yet applied, then this
170 * value will be definitely replaced by something else.
171 */
172 if (ms->enforce_config_section) {
173 current_migration->send_configuration = true;
174 }
175 }
176
177 void migration_shutdown(void)
178 {
179 /*
180 * Cancel the current migration - that will (eventually)
181 * stop the migration using this structure
182 */
183 migrate_fd_cancel(current_migration);
184 object_unref(OBJECT(current_migration));
185 }
186
187 /* For outgoing */
188 MigrationState *migrate_get_current(void)
189 {
190 /* This can only be called after the object created. */
191 assert(current_migration);
192 return current_migration;
193 }
194
195 MigrationIncomingState *migration_incoming_get_current(void)
196 {
197 assert(current_incoming);
198 return current_incoming;
199 }
200
201 void migration_incoming_state_destroy(void)
202 {
203 struct MigrationIncomingState *mis = migration_incoming_get_current();
204
205 if (mis->to_src_file) {
206 /* Tell source that we are done */
207 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
208 qemu_fclose(mis->to_src_file);
209 mis->to_src_file = NULL;
210 }
211
212 if (mis->from_src_file) {
213 qemu_fclose(mis->from_src_file);
214 mis->from_src_file = NULL;
215 }
216 if (mis->postcopy_remote_fds) {
217 g_array_free(mis->postcopy_remote_fds, TRUE);
218 mis->postcopy_remote_fds = NULL;
219 }
220
221 qemu_event_reset(&mis->main_thread_load_event);
222
223 if (mis->socket_address_list) {
224 qapi_free_SocketAddressList(mis->socket_address_list);
225 mis->socket_address_list = NULL;
226 }
227 }
228
229 static void migrate_generate_event(int new_state)
230 {
231 if (migrate_use_events()) {
232 qapi_event_send_migration(new_state);
233 }
234 }
235
236 static bool migrate_late_block_activate(void)
237 {
238 MigrationState *s;
239
240 s = migrate_get_current();
241
242 return s->enabled_capabilities[
243 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
244 }
245
246 /*
247 * Called on -incoming with a defer: uri.
248 * The migration can be started later after any parameters have been
249 * changed.
250 */
251 static void deferred_incoming_migration(Error **errp)
252 {
253 if (deferred_incoming) {
254 error_setg(errp, "Incoming migration already deferred");
255 }
256 deferred_incoming = true;
257 }
258
259 /*
260 * Send a message on the return channel back to the source
261 * of the migration.
262 */
263 static int migrate_send_rp_message(MigrationIncomingState *mis,
264 enum mig_rp_message_type message_type,
265 uint16_t len, void *data)
266 {
267 int ret = 0;
268
269 trace_migrate_send_rp_message((int)message_type, len);
270 qemu_mutex_lock(&mis->rp_mutex);
271
272 /*
273 * It's possible that the file handle got lost due to network
274 * failures.
275 */
276 if (!mis->to_src_file) {
277 ret = -EIO;
278 goto error;
279 }
280
281 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
282 qemu_put_be16(mis->to_src_file, len);
283 qemu_put_buffer(mis->to_src_file, data, len);
284 qemu_fflush(mis->to_src_file);
285
286 /* It's possible that qemu file got error during sending */
287 ret = qemu_file_get_error(mis->to_src_file);
288
289 error:
290 qemu_mutex_unlock(&mis->rp_mutex);
291 return ret;
292 }
293
294 /* Request a range of pages from the source VM at the given
295 * start address.
296 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
297 * as the last request (a name must have been given previously)
298 * Start: Address offset within the RB
299 * Len: Length in bytes required - must be a multiple of pagesize
300 */
301 int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
302 ram_addr_t start, size_t len)
303 {
304 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
305 size_t msglen = 12; /* start + len */
306 enum mig_rp_message_type msg_type;
307
308 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
309 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
310
311 if (rbname) {
312 int rbname_len = strlen(rbname);
313 assert(rbname_len < 256);
314
315 bufc[msglen++] = rbname_len;
316 memcpy(bufc + msglen, rbname, rbname_len);
317 msglen += rbname_len;
318 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
319 } else {
320 msg_type = MIG_RP_MSG_REQ_PAGES;
321 }
322
323 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
324 }
325
326 static bool migration_colo_enabled;
327 bool migration_incoming_colo_enabled(void)
328 {
329 return migration_colo_enabled;
330 }
331
332 void migration_incoming_disable_colo(void)
333 {
334 migration_colo_enabled = false;
335 }
336
337 void migration_incoming_enable_colo(void)
338 {
339 migration_colo_enabled = true;
340 }
341
342 void migrate_add_address(SocketAddress *address)
343 {
344 MigrationIncomingState *mis = migration_incoming_get_current();
345 SocketAddressList *addrs;
346
347 addrs = g_new0(SocketAddressList, 1);
348 addrs->next = mis->socket_address_list;
349 mis->socket_address_list = addrs;
350 addrs->value = QAPI_CLONE(SocketAddress, address);
351 }
352
353 void qemu_start_incoming_migration(const char *uri, Error **errp)
354 {
355 const char *p;
356
357 qapi_event_send_migration(MIGRATION_STATUS_SETUP);
358 if (!strcmp(uri, "defer")) {
359 deferred_incoming_migration(errp);
360 } else if (strstart(uri, "tcp:", &p)) {
361 tcp_start_incoming_migration(p, errp);
362 #ifdef CONFIG_RDMA
363 } else if (strstart(uri, "rdma:", &p)) {
364 rdma_start_incoming_migration(p, errp);
365 #endif
366 } else if (strstart(uri, "exec:", &p)) {
367 exec_start_incoming_migration(p, errp);
368 } else if (strstart(uri, "unix:", &p)) {
369 unix_start_incoming_migration(p, errp);
370 } else if (strstart(uri, "fd:", &p)) {
371 fd_start_incoming_migration(p, errp);
372 } else {
373 error_setg(errp, "unknown migration protocol: %s", uri);
374 }
375 }
376
377 static void process_incoming_migration_bh(void *opaque)
378 {
379 Error *local_err = NULL;
380 MigrationIncomingState *mis = opaque;
381
382 /* If capability late_block_activate is set:
383 * Only fire up the block code now if we're going to restart the
384 * VM, else 'cont' will do it.
385 * This causes file locking to happen; so we don't want it to happen
386 * unless we really are starting the VM.
387 */
388 if (!migrate_late_block_activate() ||
389 (autostart && (!global_state_received() ||
390 global_state_get_runstate() == RUN_STATE_RUNNING))) {
391 /* Make sure all file formats flush their mutable metadata.
392 * If we get an error here, just don't restart the VM yet. */
393 bdrv_invalidate_cache_all(&local_err);
394 if (local_err) {
395 error_report_err(local_err);
396 local_err = NULL;
397 autostart = false;
398 }
399 }
400
401 /*
402 * This must happen after all error conditions are dealt with and
403 * we're sure the VM is going to be running on this host.
404 */
405 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
406
407 if (multifd_load_cleanup(&local_err) != 0) {
408 error_report_err(local_err);
409 autostart = false;
410 }
411 /* If global state section was not received or we are in running
412 state, we need to obey autostart. Any other state is set with
413 runstate_set. */
414
415 dirty_bitmap_mig_before_vm_start();
416
417 if (!global_state_received() ||
418 global_state_get_runstate() == RUN_STATE_RUNNING) {
419 if (autostart) {
420 vm_start();
421 } else {
422 runstate_set(RUN_STATE_PAUSED);
423 }
424 } else if (migration_incoming_colo_enabled()) {
425 migration_incoming_disable_colo();
426 vm_start();
427 } else {
428 runstate_set(global_state_get_runstate());
429 }
430 /*
431 * This must happen after any state changes since as soon as an external
432 * observer sees this event they might start to prod at the VM assuming
433 * it's ready to use.
434 */
435 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
436 MIGRATION_STATUS_COMPLETED);
437 qemu_bh_delete(mis->bh);
438 migration_incoming_state_destroy();
439 }
440
441 static void process_incoming_migration_co(void *opaque)
442 {
443 MigrationIncomingState *mis = migration_incoming_get_current();
444 PostcopyState ps;
445 int ret;
446 Error *local_err = NULL;
447
448 assert(mis->from_src_file);
449 mis->migration_incoming_co = qemu_coroutine_self();
450 mis->largest_page_size = qemu_ram_pagesize_largest();
451 postcopy_state_set(POSTCOPY_INCOMING_NONE);
452 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
453 MIGRATION_STATUS_ACTIVE);
454 ret = qemu_loadvm_state(mis->from_src_file);
455
456 ps = postcopy_state_get();
457 trace_process_incoming_migration_co_end(ret, ps);
458 if (ps != POSTCOPY_INCOMING_NONE) {
459 if (ps == POSTCOPY_INCOMING_ADVISE) {
460 /*
461 * Where a migration had postcopy enabled (and thus went to advise)
462 * but managed to complete within the precopy period, we can use
463 * the normal exit.
464 */
465 postcopy_ram_incoming_cleanup(mis);
466 } else if (ret >= 0) {
467 /*
468 * Postcopy was started, cleanup should happen at the end of the
469 * postcopy thread.
470 */
471 trace_process_incoming_migration_co_postcopy_end_main();
472 return;
473 }
474 /* Else if something went wrong then just fall out of the normal exit */
475 }
476
477 /* we get COLO info, and know if we are in COLO mode */
478 if (!ret && migration_incoming_colo_enabled()) {
479 /* Make sure all file formats flush their mutable metadata */
480 bdrv_invalidate_cache_all(&local_err);
481 if (local_err) {
482 error_report_err(local_err);
483 goto fail;
484 }
485
486 if (colo_init_ram_cache() < 0) {
487 error_report("Init ram cache failed");
488 goto fail;
489 }
490
491 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
492 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
493 mis->have_colo_incoming_thread = true;
494 qemu_coroutine_yield();
495
496 /* Wait checkpoint incoming thread exit before free resource */
497 qemu_thread_join(&mis->colo_incoming_thread);
498 /* We hold the global iothread lock, so it is safe here */
499 colo_release_ram_cache();
500 }
501
502 if (ret < 0) {
503 error_report("load of migration failed: %s", strerror(-ret));
504 goto fail;
505 }
506 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
507 qemu_bh_schedule(mis->bh);
508 mis->migration_incoming_co = NULL;
509 return;
510 fail:
511 local_err = NULL;
512 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
513 MIGRATION_STATUS_FAILED);
514 qemu_fclose(mis->from_src_file);
515 if (multifd_load_cleanup(&local_err) != 0) {
516 error_report_err(local_err);
517 }
518 exit(EXIT_FAILURE);
519 }
520
521 static void migration_incoming_setup(QEMUFile *f)
522 {
523 MigrationIncomingState *mis = migration_incoming_get_current();
524
525 if (multifd_load_setup() != 0) {
526 /* We haven't been able to create multifd threads
527 nothing better to do */
528 exit(EXIT_FAILURE);
529 }
530
531 if (!mis->from_src_file) {
532 mis->from_src_file = f;
533 }
534 qemu_file_set_blocking(f, false);
535 }
536
537 void migration_incoming_process(void)
538 {
539 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
540 qemu_coroutine_enter(co);
541 }
542
543 /* Returns true if recovered from a paused migration, otherwise false */
544 static bool postcopy_try_recover(QEMUFile *f)
545 {
546 MigrationIncomingState *mis = migration_incoming_get_current();
547
548 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
549 /* Resumed from a paused postcopy migration */
550
551 mis->from_src_file = f;
552 /* Postcopy has standalone thread to do vm load */
553 qemu_file_set_blocking(f, true);
554
555 /* Re-configure the return path */
556 mis->to_src_file = qemu_file_get_return_path(f);
557
558 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
559 MIGRATION_STATUS_POSTCOPY_RECOVER);
560
561 /*
562 * Here, we only wake up the main loading thread (while the
563 * fault thread will still be waiting), so that we can receive
564 * commands from source now, and answer it if needed. The
565 * fault thread will be woken up afterwards until we are sure
566 * that source is ready to reply to page requests.
567 */
568 qemu_sem_post(&mis->postcopy_pause_sem_dst);
569 return true;
570 }
571
572 return false;
573 }
574
575 void migration_fd_process_incoming(QEMUFile *f)
576 {
577 if (postcopy_try_recover(f)) {
578 return;
579 }
580
581 migration_incoming_setup(f);
582 migration_incoming_process();
583 }
584
585 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
586 {
587 MigrationIncomingState *mis = migration_incoming_get_current();
588 bool start_migration;
589
590 if (!mis->from_src_file) {
591 /* The first connection (multifd may have multiple) */
592 QEMUFile *f = qemu_fopen_channel_input(ioc);
593
594 /* If it's a recovery, we're done */
595 if (postcopy_try_recover(f)) {
596 return;
597 }
598
599 migration_incoming_setup(f);
600
601 /*
602 * Common migration only needs one channel, so we can start
603 * right now. Multifd needs more than one channel, we wait.
604 */
605 start_migration = !migrate_use_multifd();
606 } else {
607 Error *local_err = NULL;
608 /* Multiple connections */
609 assert(migrate_use_multifd());
610 start_migration = multifd_recv_new_channel(ioc, &local_err);
611 if (local_err) {
612 error_propagate(errp, local_err);
613 return;
614 }
615 }
616
617 if (start_migration) {
618 migration_incoming_process();
619 }
620 }
621
622 /**
623 * @migration_has_all_channels: We have received all channels that we need
624 *
625 * Returns true when we have got connections to all the channels that
626 * we need for migration.
627 */
628 bool migration_has_all_channels(void)
629 {
630 MigrationIncomingState *mis = migration_incoming_get_current();
631 bool all_channels;
632
633 all_channels = multifd_recv_all_channels_created();
634
635 return all_channels && mis->from_src_file != NULL;
636 }
637
638 /*
639 * Send a 'SHUT' message on the return channel with the given value
640 * to indicate that we've finished with the RP. Non-0 value indicates
641 * error.
642 */
643 void migrate_send_rp_shut(MigrationIncomingState *mis,
644 uint32_t value)
645 {
646 uint32_t buf;
647
648 buf = cpu_to_be32(value);
649 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
650 }
651
652 /*
653 * Send a 'PONG' message on the return channel with the given value
654 * (normally in response to a 'PING')
655 */
656 void migrate_send_rp_pong(MigrationIncomingState *mis,
657 uint32_t value)
658 {
659 uint32_t buf;
660
661 buf = cpu_to_be32(value);
662 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
663 }
664
665 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
666 char *block_name)
667 {
668 char buf[512];
669 int len;
670 int64_t res;
671
672 /*
673 * First, we send the header part. It contains only the len of
674 * idstr, and the idstr itself.
675 */
676 len = strlen(block_name);
677 buf[0] = len;
678 memcpy(buf + 1, block_name, len);
679
680 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
681 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
682 __func__);
683 return;
684 }
685
686 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
687
688 /*
689 * Next, we dump the received bitmap to the stream.
690 *
691 * TODO: currently we are safe since we are the only one that is
692 * using the to_src_file handle (fault thread is still paused),
693 * and it's ok even not taking the mutex. However the best way is
694 * to take the lock before sending the message header, and release
695 * the lock after sending the bitmap.
696 */
697 qemu_mutex_lock(&mis->rp_mutex);
698 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
699 qemu_mutex_unlock(&mis->rp_mutex);
700
701 trace_migrate_send_rp_recv_bitmap(block_name, res);
702 }
703
704 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
705 {
706 uint32_t buf;
707
708 buf = cpu_to_be32(value);
709 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
710 }
711
712 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
713 {
714 MigrationCapabilityStatusList *head = NULL;
715 MigrationCapabilityStatusList *caps;
716 MigrationState *s = migrate_get_current();
717 int i;
718
719 caps = NULL; /* silence compiler warning */
720 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
721 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
722 if (i == MIGRATION_CAPABILITY_BLOCK) {
723 continue;
724 }
725 #endif
726 if (head == NULL) {
727 head = g_malloc0(sizeof(*caps));
728 caps = head;
729 } else {
730 caps->next = g_malloc0(sizeof(*caps));
731 caps = caps->next;
732 }
733 caps->value =
734 g_malloc(sizeof(*caps->value));
735 caps->value->capability = i;
736 caps->value->state = s->enabled_capabilities[i];
737 }
738
739 return head;
740 }
741
742 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
743 {
744 MigrationParameters *params;
745 MigrationState *s = migrate_get_current();
746
747 /* TODO use QAPI_CLONE() instead of duplicating it inline */
748 params = g_malloc0(sizeof(*params));
749 params->has_compress_level = true;
750 params->compress_level = s->parameters.compress_level;
751 params->has_compress_threads = true;
752 params->compress_threads = s->parameters.compress_threads;
753 params->has_compress_wait_thread = true;
754 params->compress_wait_thread = s->parameters.compress_wait_thread;
755 params->has_decompress_threads = true;
756 params->decompress_threads = s->parameters.decompress_threads;
757 params->has_cpu_throttle_initial = true;
758 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
759 params->has_cpu_throttle_increment = true;
760 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
761 params->has_tls_creds = true;
762 params->tls_creds = g_strdup(s->parameters.tls_creds);
763 params->has_tls_hostname = true;
764 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
765 params->has_tls_authz = true;
766 params->tls_authz = g_strdup(s->parameters.tls_authz);
767 params->has_max_bandwidth = true;
768 params->max_bandwidth = s->parameters.max_bandwidth;
769 params->has_downtime_limit = true;
770 params->downtime_limit = s->parameters.downtime_limit;
771 params->has_x_checkpoint_delay = true;
772 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
773 params->has_block_incremental = true;
774 params->block_incremental = s->parameters.block_incremental;
775 params->has_multifd_channels = true;
776 params->multifd_channels = s->parameters.multifd_channels;
777 params->has_xbzrle_cache_size = true;
778 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
779 params->has_max_postcopy_bandwidth = true;
780 params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
781 params->has_max_cpu_throttle = true;
782 params->max_cpu_throttle = s->parameters.max_cpu_throttle;
783 params->has_announce_initial = true;
784 params->announce_initial = s->parameters.announce_initial;
785 params->has_announce_max = true;
786 params->announce_max = s->parameters.announce_max;
787 params->has_announce_rounds = true;
788 params->announce_rounds = s->parameters.announce_rounds;
789 params->has_announce_step = true;
790 params->announce_step = s->parameters.announce_step;
791
792 return params;
793 }
794
795 AnnounceParameters *migrate_announce_params(void)
796 {
797 static AnnounceParameters ap;
798
799 MigrationState *s = migrate_get_current();
800
801 ap.initial = s->parameters.announce_initial;
802 ap.max = s->parameters.announce_max;
803 ap.rounds = s->parameters.announce_rounds;
804 ap.step = s->parameters.announce_step;
805
806 return &ap;
807 }
808
809 /*
810 * Return true if we're already in the middle of a migration
811 * (i.e. any of the active or setup states)
812 */
813 bool migration_is_setup_or_active(int state)
814 {
815 switch (state) {
816 case MIGRATION_STATUS_ACTIVE:
817 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
818 case MIGRATION_STATUS_POSTCOPY_PAUSED:
819 case MIGRATION_STATUS_POSTCOPY_RECOVER:
820 case MIGRATION_STATUS_SETUP:
821 case MIGRATION_STATUS_PRE_SWITCHOVER:
822 case MIGRATION_STATUS_DEVICE:
823 case MIGRATION_STATUS_WAIT_UNPLUG:
824 return true;
825
826 default:
827 return false;
828
829 }
830 }
831
832 static void populate_time_info(MigrationInfo *info, MigrationState *s)
833 {
834 info->has_status = true;
835 info->has_setup_time = true;
836 info->setup_time = s->setup_time;
837 if (s->state == MIGRATION_STATUS_COMPLETED) {
838 info->has_total_time = true;
839 info->total_time = s->total_time;
840 info->has_downtime = true;
841 info->downtime = s->downtime;
842 } else {
843 info->has_total_time = true;
844 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
845 s->start_time;
846 info->has_expected_downtime = true;
847 info->expected_downtime = s->expected_downtime;
848 }
849 }
850
851 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
852 {
853 info->has_ram = true;
854 info->ram = g_malloc0(sizeof(*info->ram));
855 info->ram->transferred = ram_counters.transferred;
856 info->ram->total = ram_bytes_total();
857 info->ram->duplicate = ram_counters.duplicate;
858 /* legacy value. It is not used anymore */
859 info->ram->skipped = 0;
860 info->ram->normal = ram_counters.normal;
861 info->ram->normal_bytes = ram_counters.normal *
862 qemu_target_page_size();
863 info->ram->mbps = s->mbps;
864 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
865 info->ram->postcopy_requests = ram_counters.postcopy_requests;
866 info->ram->page_size = qemu_target_page_size();
867 info->ram->multifd_bytes = ram_counters.multifd_bytes;
868 info->ram->pages_per_second = s->pages_per_second;
869
870 if (migrate_use_xbzrle()) {
871 info->has_xbzrle_cache = true;
872 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
873 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
874 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
875 info->xbzrle_cache->pages = xbzrle_counters.pages;
876 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
877 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
878 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
879 }
880
881 if (migrate_use_compression()) {
882 info->has_compression = true;
883 info->compression = g_malloc0(sizeof(*info->compression));
884 info->compression->pages = compression_counters.pages;
885 info->compression->busy = compression_counters.busy;
886 info->compression->busy_rate = compression_counters.busy_rate;
887 info->compression->compressed_size =
888 compression_counters.compressed_size;
889 info->compression->compression_rate =
890 compression_counters.compression_rate;
891 }
892
893 if (cpu_throttle_active()) {
894 info->has_cpu_throttle_percentage = true;
895 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
896 }
897
898 if (s->state != MIGRATION_STATUS_COMPLETED) {
899 info->ram->remaining = ram_bytes_remaining();
900 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
901 }
902 }
903
904 static void populate_disk_info(MigrationInfo *info)
905 {
906 if (blk_mig_active()) {
907 info->has_disk = true;
908 info->disk = g_malloc0(sizeof(*info->disk));
909 info->disk->transferred = blk_mig_bytes_transferred();
910 info->disk->remaining = blk_mig_bytes_remaining();
911 info->disk->total = blk_mig_bytes_total();
912 }
913 }
914
915 static void fill_source_migration_info(MigrationInfo *info)
916 {
917 MigrationState *s = migrate_get_current();
918
919 switch (s->state) {
920 case MIGRATION_STATUS_NONE:
921 /* no migration has happened ever */
922 /* do not overwrite destination migration status */
923 return;
924 break;
925 case MIGRATION_STATUS_SETUP:
926 info->has_status = true;
927 info->has_total_time = false;
928 break;
929 case MIGRATION_STATUS_ACTIVE:
930 case MIGRATION_STATUS_CANCELLING:
931 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
932 case MIGRATION_STATUS_PRE_SWITCHOVER:
933 case MIGRATION_STATUS_DEVICE:
934 case MIGRATION_STATUS_POSTCOPY_PAUSED:
935 case MIGRATION_STATUS_POSTCOPY_RECOVER:
936 /* TODO add some postcopy stats */
937 populate_time_info(info, s);
938 populate_ram_info(info, s);
939 populate_disk_info(info);
940 break;
941 case MIGRATION_STATUS_COLO:
942 info->has_status = true;
943 /* TODO: display COLO specific information (checkpoint info etc.) */
944 break;
945 case MIGRATION_STATUS_COMPLETED:
946 populate_time_info(info, s);
947 populate_ram_info(info, s);
948 break;
949 case MIGRATION_STATUS_FAILED:
950 info->has_status = true;
951 if (s->error) {
952 info->has_error_desc = true;
953 info->error_desc = g_strdup(error_get_pretty(s->error));
954 }
955 break;
956 case MIGRATION_STATUS_CANCELLED:
957 info->has_status = true;
958 break;
959 case MIGRATION_STATUS_WAIT_UNPLUG:
960 info->has_status = true;
961 break;
962 }
963 info->status = s->state;
964 }
965
966 /**
967 * @migration_caps_check - check capability validity
968 *
969 * @cap_list: old capability list, array of bool
970 * @params: new capabilities to be applied soon
971 * @errp: set *errp if the check failed, with reason
972 *
973 * Returns true if check passed, otherwise false.
974 */
975 static bool migrate_caps_check(bool *cap_list,
976 MigrationCapabilityStatusList *params,
977 Error **errp)
978 {
979 MigrationCapabilityStatusList *cap;
980 bool old_postcopy_cap;
981 MigrationIncomingState *mis = migration_incoming_get_current();
982
983 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
984
985 for (cap = params; cap; cap = cap->next) {
986 cap_list[cap->value->capability] = cap->value->state;
987 }
988
989 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
990 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
991 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
992 "block migration");
993 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
994 return false;
995 }
996 #endif
997
998 #ifndef CONFIG_REPLICATION
999 if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
1000 error_setg(errp, "QEMU compiled without replication module"
1001 " can't enable COLO");
1002 error_append_hint(errp, "Please enable replication before COLO.\n");
1003 return false;
1004 }
1005 #endif
1006
1007 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
1008 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
1009 /* The decompression threads asynchronously write into RAM
1010 * rather than use the atomic copies needed to avoid
1011 * userfaulting. It should be possible to fix the decompression
1012 * threads for compatibility in future.
1013 */
1014 error_setg(errp, "Postcopy is not currently compatible "
1015 "with compression");
1016 return false;
1017 }
1018
1019 /* This check is reasonably expensive, so only when it's being
1020 * set the first time, also it's only the destination that needs
1021 * special support.
1022 */
1023 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1024 !postcopy_ram_supported_by_host(mis)) {
1025 /* postcopy_ram_supported_by_host will have emitted a more
1026 * detailed message
1027 */
1028 error_setg(errp, "Postcopy is not supported");
1029 return false;
1030 }
1031
1032 if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1033 error_setg(errp, "Postcopy is not compatible with ignore-shared");
1034 return false;
1035 }
1036 }
1037
1038 return true;
1039 }
1040
1041 static void fill_destination_migration_info(MigrationInfo *info)
1042 {
1043 MigrationIncomingState *mis = migration_incoming_get_current();
1044
1045 if (mis->socket_address_list) {
1046 info->has_socket_address = true;
1047 info->socket_address =
1048 QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1049 }
1050
1051 switch (mis->state) {
1052 case MIGRATION_STATUS_NONE:
1053 return;
1054 break;
1055 case MIGRATION_STATUS_SETUP:
1056 case MIGRATION_STATUS_CANCELLING:
1057 case MIGRATION_STATUS_CANCELLED:
1058 case MIGRATION_STATUS_ACTIVE:
1059 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1060 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1061 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1062 case MIGRATION_STATUS_FAILED:
1063 case MIGRATION_STATUS_COLO:
1064 info->has_status = true;
1065 break;
1066 case MIGRATION_STATUS_COMPLETED:
1067 info->has_status = true;
1068 fill_destination_postcopy_migration_info(info);
1069 break;
1070 }
1071 info->status = mis->state;
1072 }
1073
1074 MigrationInfo *qmp_query_migrate(Error **errp)
1075 {
1076 MigrationInfo *info = g_malloc0(sizeof(*info));
1077
1078 fill_destination_migration_info(info);
1079 fill_source_migration_info(info);
1080
1081 return info;
1082 }
1083
1084 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1085 Error **errp)
1086 {
1087 MigrationState *s = migrate_get_current();
1088 MigrationCapabilityStatusList *cap;
1089 bool cap_list[MIGRATION_CAPABILITY__MAX];
1090
1091 if (migration_is_setup_or_active(s->state)) {
1092 error_setg(errp, QERR_MIGRATION_ACTIVE);
1093 return;
1094 }
1095
1096 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1097 if (!migrate_caps_check(cap_list, params, errp)) {
1098 return;
1099 }
1100
1101 for (cap = params; cap; cap = cap->next) {
1102 s->enabled_capabilities[cap->value->capability] = cap->value->state;
1103 }
1104 }
1105
1106 /*
1107 * Check whether the parameters are valid. Error will be put into errp
1108 * (if provided). Return true if valid, otherwise false.
1109 */
1110 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1111 {
1112 if (params->has_compress_level &&
1113 (params->compress_level > 9)) {
1114 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1115 "is invalid, it should be in the range of 0 to 9");
1116 return false;
1117 }
1118
1119 if (params->has_compress_threads && (params->compress_threads < 1)) {
1120 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1121 "compress_threads",
1122 "is invalid, it should be in the range of 1 to 255");
1123 return false;
1124 }
1125
1126 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1127 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1128 "decompress_threads",
1129 "is invalid, it should be in the range of 1 to 255");
1130 return false;
1131 }
1132
1133 if (params->has_cpu_throttle_initial &&
1134 (params->cpu_throttle_initial < 1 ||
1135 params->cpu_throttle_initial > 99)) {
1136 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1137 "cpu_throttle_initial",
1138 "an integer in the range of 1 to 99");
1139 return false;
1140 }
1141
1142 if (params->has_cpu_throttle_increment &&
1143 (params->cpu_throttle_increment < 1 ||
1144 params->cpu_throttle_increment > 99)) {
1145 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1146 "cpu_throttle_increment",
1147 "an integer in the range of 1 to 99");
1148 return false;
1149 }
1150
1151 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1152 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
1153 " range of 0 to %zu bytes/second", SIZE_MAX);
1154 return false;
1155 }
1156
1157 if (params->has_downtime_limit &&
1158 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1159 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1160 "the range of 0 to %d milliseconds",
1161 MAX_MIGRATE_DOWNTIME);
1162 return false;
1163 }
1164
1165 /* x_checkpoint_delay is now always positive */
1166
1167 if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1168 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1169 "multifd_channels",
1170 "is invalid, it should be in the range of 1 to 255");
1171 return false;
1172 }
1173
1174 if (params->has_xbzrle_cache_size &&
1175 (params->xbzrle_cache_size < qemu_target_page_size() ||
1176 !is_power_of_2(params->xbzrle_cache_size))) {
1177 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1178 "xbzrle_cache_size",
1179 "is invalid, it should be bigger than target page size"
1180 " and a power of two");
1181 return false;
1182 }
1183
1184 if (params->has_max_cpu_throttle &&
1185 (params->max_cpu_throttle < params->cpu_throttle_initial ||
1186 params->max_cpu_throttle > 99)) {
1187 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1188 "max_cpu_throttle",
1189 "an integer in the range of cpu_throttle_initial to 99");
1190 return false;
1191 }
1192
1193 if (params->has_announce_initial &&
1194 params->announce_initial > 100000) {
1195 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1196 "announce_initial",
1197 "is invalid, it must be less than 100000 ms");
1198 return false;
1199 }
1200 if (params->has_announce_max &&
1201 params->announce_max > 100000) {
1202 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1203 "announce_max",
1204 "is invalid, it must be less than 100000 ms");
1205 return false;
1206 }
1207 if (params->has_announce_rounds &&
1208 params->announce_rounds > 1000) {
1209 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1210 "announce_rounds",
1211 "is invalid, it must be in the range of 0 to 1000");
1212 return false;
1213 }
1214 if (params->has_announce_step &&
1215 (params->announce_step < 1 ||
1216 params->announce_step > 10000)) {
1217 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1218 "announce_step",
1219 "is invalid, it must be in the range of 1 to 10000 ms");
1220 return false;
1221 }
1222 return true;
1223 }
1224
1225 static void migrate_params_test_apply(MigrateSetParameters *params,
1226 MigrationParameters *dest)
1227 {
1228 *dest = migrate_get_current()->parameters;
1229
1230 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1231
1232 if (params->has_compress_level) {
1233 dest->compress_level = params->compress_level;
1234 }
1235
1236 if (params->has_compress_threads) {
1237 dest->compress_threads = params->compress_threads;
1238 }
1239
1240 if (params->has_compress_wait_thread) {
1241 dest->compress_wait_thread = params->compress_wait_thread;
1242 }
1243
1244 if (params->has_decompress_threads) {
1245 dest->decompress_threads = params->decompress_threads;
1246 }
1247
1248 if (params->has_cpu_throttle_initial) {
1249 dest->cpu_throttle_initial = params->cpu_throttle_initial;
1250 }
1251
1252 if (params->has_cpu_throttle_increment) {
1253 dest->cpu_throttle_increment = params->cpu_throttle_increment;
1254 }
1255
1256 if (params->has_tls_creds) {
1257 assert(params->tls_creds->type == QTYPE_QSTRING);
1258 dest->tls_creds = g_strdup(params->tls_creds->u.s);
1259 }
1260
1261 if (params->has_tls_hostname) {
1262 assert(params->tls_hostname->type == QTYPE_QSTRING);
1263 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1264 }
1265
1266 if (params->has_max_bandwidth) {
1267 dest->max_bandwidth = params->max_bandwidth;
1268 }
1269
1270 if (params->has_downtime_limit) {
1271 dest->downtime_limit = params->downtime_limit;
1272 }
1273
1274 if (params->has_x_checkpoint_delay) {
1275 dest->x_checkpoint_delay = params->x_checkpoint_delay;
1276 }
1277
1278 if (params->has_block_incremental) {
1279 dest->block_incremental = params->block_incremental;
1280 }
1281 if (params->has_multifd_channels) {
1282 dest->multifd_channels = params->multifd_channels;
1283 }
1284 if (params->has_xbzrle_cache_size) {
1285 dest->xbzrle_cache_size = params->xbzrle_cache_size;
1286 }
1287 if (params->has_max_postcopy_bandwidth) {
1288 dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1289 }
1290 if (params->has_max_cpu_throttle) {
1291 dest->max_cpu_throttle = params->max_cpu_throttle;
1292 }
1293 if (params->has_announce_initial) {
1294 dest->announce_initial = params->announce_initial;
1295 }
1296 if (params->has_announce_max) {
1297 dest->announce_max = params->announce_max;
1298 }
1299 if (params->has_announce_rounds) {
1300 dest->announce_rounds = params->announce_rounds;
1301 }
1302 if (params->has_announce_step) {
1303 dest->announce_step = params->announce_step;
1304 }
1305 }
1306
1307 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1308 {
1309 MigrationState *s = migrate_get_current();
1310
1311 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1312
1313 if (params->has_compress_level) {
1314 s->parameters.compress_level = params->compress_level;
1315 }
1316
1317 if (params->has_compress_threads) {
1318 s->parameters.compress_threads = params->compress_threads;
1319 }
1320
1321 if (params->has_compress_wait_thread) {
1322 s->parameters.compress_wait_thread = params->compress_wait_thread;
1323 }
1324
1325 if (params->has_decompress_threads) {
1326 s->parameters.decompress_threads = params->decompress_threads;
1327 }
1328
1329 if (params->has_cpu_throttle_initial) {
1330 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1331 }
1332
1333 if (params->has_cpu_throttle_increment) {
1334 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1335 }
1336
1337 if (params->has_tls_creds) {
1338 g_free(s->parameters.tls_creds);
1339 assert(params->tls_creds->type == QTYPE_QSTRING);
1340 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1341 }
1342
1343 if (params->has_tls_hostname) {
1344 g_free(s->parameters.tls_hostname);
1345 assert(params->tls_hostname->type == QTYPE_QSTRING);
1346 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1347 }
1348
1349 if (params->has_tls_authz) {
1350 g_free(s->parameters.tls_authz);
1351 assert(params->tls_authz->type == QTYPE_QSTRING);
1352 s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1353 }
1354
1355 if (params->has_max_bandwidth) {
1356 s->parameters.max_bandwidth = params->max_bandwidth;
1357 if (s->to_dst_file && !migration_in_postcopy()) {
1358 qemu_file_set_rate_limit(s->to_dst_file,
1359 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1360 }
1361 }
1362
1363 if (params->has_downtime_limit) {
1364 s->parameters.downtime_limit = params->downtime_limit;
1365 }
1366
1367 if (params->has_x_checkpoint_delay) {
1368 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1369 if (migration_in_colo_state()) {
1370 colo_checkpoint_notify(s);
1371 }
1372 }
1373
1374 if (params->has_block_incremental) {
1375 s->parameters.block_incremental = params->block_incremental;
1376 }
1377 if (params->has_multifd_channels) {
1378 s->parameters.multifd_channels = params->multifd_channels;
1379 }
1380 if (params->has_xbzrle_cache_size) {
1381 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1382 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1383 }
1384 if (params->has_max_postcopy_bandwidth) {
1385 s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1386 if (s->to_dst_file && migration_in_postcopy()) {
1387 qemu_file_set_rate_limit(s->to_dst_file,
1388 s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1389 }
1390 }
1391 if (params->has_max_cpu_throttle) {
1392 s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1393 }
1394 if (params->has_announce_initial) {
1395 s->parameters.announce_initial = params->announce_initial;
1396 }
1397 if (params->has_announce_max) {
1398 s->parameters.announce_max = params->announce_max;
1399 }
1400 if (params->has_announce_rounds) {
1401 s->parameters.announce_rounds = params->announce_rounds;
1402 }
1403 if (params->has_announce_step) {
1404 s->parameters.announce_step = params->announce_step;
1405 }
1406 }
1407
1408 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1409 {
1410 MigrationParameters tmp;
1411
1412 /* TODO Rewrite "" to null instead */
1413 if (params->has_tls_creds
1414 && params->tls_creds->type == QTYPE_QNULL) {
1415 qobject_unref(params->tls_creds->u.n);
1416 params->tls_creds->type = QTYPE_QSTRING;
1417 params->tls_creds->u.s = strdup("");
1418 }
1419 /* TODO Rewrite "" to null instead */
1420 if (params->has_tls_hostname
1421 && params->tls_hostname->type == QTYPE_QNULL) {
1422 qobject_unref(params->tls_hostname->u.n);
1423 params->tls_hostname->type = QTYPE_QSTRING;
1424 params->tls_hostname->u.s = strdup("");
1425 }
1426
1427 migrate_params_test_apply(params, &tmp);
1428
1429 if (!migrate_params_check(&tmp, errp)) {
1430 /* Invalid parameter */
1431 return;
1432 }
1433
1434 migrate_params_apply(params, errp);
1435 }
1436
1437
1438 void qmp_migrate_start_postcopy(Error **errp)
1439 {
1440 MigrationState *s = migrate_get_current();
1441
1442 if (!migrate_postcopy()) {
1443 error_setg(errp, "Enable postcopy with migrate_set_capability before"
1444 " the start of migration");
1445 return;
1446 }
1447
1448 if (s->state == MIGRATION_STATUS_NONE) {
1449 error_setg(errp, "Postcopy must be started after migration has been"
1450 " started");
1451 return;
1452 }
1453 /*
1454 * we don't error if migration has finished since that would be racy
1455 * with issuing this command.
1456 */
1457 atomic_set(&s->start_postcopy, true);
1458 }
1459
1460 /* shared migration helpers */
1461
1462 void migrate_set_state(int *state, int old_state, int new_state)
1463 {
1464 assert(new_state < MIGRATION_STATUS__MAX);
1465 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1466 trace_migrate_set_state(MigrationStatus_str(new_state));
1467 migrate_generate_event(new_state);
1468 }
1469 }
1470
1471 static MigrationCapabilityStatusList *migrate_cap_add(
1472 MigrationCapabilityStatusList *list,
1473 MigrationCapability index,
1474 bool state)
1475 {
1476 MigrationCapabilityStatusList *cap;
1477
1478 cap = g_new0(MigrationCapabilityStatusList, 1);
1479 cap->value = g_new0(MigrationCapabilityStatus, 1);
1480 cap->value->capability = index;
1481 cap->value->state = state;
1482 cap->next = list;
1483
1484 return cap;
1485 }
1486
1487 void migrate_set_block_enabled(bool value, Error **errp)
1488 {
1489 MigrationCapabilityStatusList *cap;
1490
1491 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1492 qmp_migrate_set_capabilities(cap, errp);
1493 qapi_free_MigrationCapabilityStatusList(cap);
1494 }
1495
1496 static void migrate_set_block_incremental(MigrationState *s, bool value)
1497 {
1498 s->parameters.block_incremental = value;
1499 }
1500
1501 static void block_cleanup_parameters(MigrationState *s)
1502 {
1503 if (s->must_remove_block_options) {
1504 /* setting to false can never fail */
1505 migrate_set_block_enabled(false, &error_abort);
1506 migrate_set_block_incremental(s, false);
1507 s->must_remove_block_options = false;
1508 }
1509 }
1510
1511 static void migrate_fd_cleanup(MigrationState *s)
1512 {
1513 qemu_bh_delete(s->cleanup_bh);
1514 s->cleanup_bh = NULL;
1515
1516 qemu_savevm_state_cleanup();
1517
1518 if (s->to_dst_file) {
1519 QEMUFile *tmp;
1520
1521 trace_migrate_fd_cleanup();
1522 qemu_mutex_unlock_iothread();
1523 if (s->migration_thread_running) {
1524 qemu_thread_join(&s->thread);
1525 s->migration_thread_running = false;
1526 }
1527 qemu_mutex_lock_iothread();
1528
1529 multifd_save_cleanup();
1530 qemu_mutex_lock(&s->qemu_file_lock);
1531 tmp = s->to_dst_file;
1532 s->to_dst_file = NULL;
1533 qemu_mutex_unlock(&s->qemu_file_lock);
1534 /*
1535 * Close the file handle without the lock to make sure the
1536 * critical section won't block for long.
1537 */
1538 qemu_fclose(tmp);
1539 }
1540
1541 assert(!migration_is_active(s));
1542
1543 if (s->state == MIGRATION_STATUS_CANCELLING) {
1544 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1545 MIGRATION_STATUS_CANCELLED);
1546 }
1547
1548 if (s->error) {
1549 /* It is used on info migrate. We can't free it */
1550 error_report_err(error_copy(s->error));
1551 }
1552 notifier_list_notify(&migration_state_notifiers, s);
1553 block_cleanup_parameters(s);
1554 }
1555
1556 static void migrate_fd_cleanup_schedule(MigrationState *s)
1557 {
1558 /*
1559 * Ref the state for bh, because it may be called when
1560 * there're already no other refs
1561 */
1562 object_ref(OBJECT(s));
1563 qemu_bh_schedule(s->cleanup_bh);
1564 }
1565
1566 static void migrate_fd_cleanup_bh(void *opaque)
1567 {
1568 MigrationState *s = opaque;
1569 migrate_fd_cleanup(s);
1570 object_unref(OBJECT(s));
1571 }
1572
1573 void migrate_set_error(MigrationState *s, const Error *error)
1574 {
1575 qemu_mutex_lock(&s->error_mutex);
1576 if (!s->error) {
1577 s->error = error_copy(error);
1578 }
1579 qemu_mutex_unlock(&s->error_mutex);
1580 }
1581
1582 void migrate_fd_error(MigrationState *s, const Error *error)
1583 {
1584 trace_migrate_fd_error(error_get_pretty(error));
1585 assert(s->to_dst_file == NULL);
1586 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1587 MIGRATION_STATUS_FAILED);
1588 migrate_set_error(s, error);
1589 }
1590
1591 static void migrate_fd_cancel(MigrationState *s)
1592 {
1593 int old_state ;
1594 QEMUFile *f = migrate_get_current()->to_dst_file;
1595 trace_migrate_fd_cancel();
1596
1597 if (s->rp_state.from_dst_file) {
1598 /* shutdown the rp socket, so causing the rp thread to shutdown */
1599 qemu_file_shutdown(s->rp_state.from_dst_file);
1600 }
1601
1602 do {
1603 old_state = s->state;
1604 if (!migration_is_setup_or_active(old_state)) {
1605 break;
1606 }
1607 /* If the migration is paused, kick it out of the pause */
1608 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1609 qemu_sem_post(&s->pause_sem);
1610 }
1611 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1612 } while (s->state != MIGRATION_STATUS_CANCELLING);
1613
1614 /*
1615 * If we're unlucky the migration code might be stuck somewhere in a
1616 * send/write while the network has failed and is waiting to timeout;
1617 * if we've got shutdown(2) available then we can force it to quit.
1618 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1619 * called in a bh, so there is no race against this cancel.
1620 */
1621 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1622 qemu_file_shutdown(f);
1623 }
1624 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1625 Error *local_err = NULL;
1626
1627 bdrv_invalidate_cache_all(&local_err);
1628 if (local_err) {
1629 error_report_err(local_err);
1630 } else {
1631 s->block_inactive = false;
1632 }
1633 }
1634 }
1635
1636 void add_migration_state_change_notifier(Notifier *notify)
1637 {
1638 notifier_list_add(&migration_state_notifiers, notify);
1639 }
1640
1641 void remove_migration_state_change_notifier(Notifier *notify)
1642 {
1643 notifier_remove(notify);
1644 }
1645
1646 bool migration_in_setup(MigrationState *s)
1647 {
1648 return s->state == MIGRATION_STATUS_SETUP;
1649 }
1650
1651 bool migration_has_finished(MigrationState *s)
1652 {
1653 return s->state == MIGRATION_STATUS_COMPLETED;
1654 }
1655
1656 bool migration_has_failed(MigrationState *s)
1657 {
1658 return (s->state == MIGRATION_STATUS_CANCELLED ||
1659 s->state == MIGRATION_STATUS_FAILED);
1660 }
1661
1662 bool migration_in_postcopy(void)
1663 {
1664 MigrationState *s = migrate_get_current();
1665
1666 switch (s->state) {
1667 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1668 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1669 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1670 return true;
1671 default:
1672 return false;
1673 }
1674 }
1675
1676 bool migration_in_postcopy_after_devices(MigrationState *s)
1677 {
1678 return migration_in_postcopy() && s->postcopy_after_devices;
1679 }
1680
1681 bool migration_is_idle(void)
1682 {
1683 MigrationState *s = current_migration;
1684
1685 if (!s) {
1686 return true;
1687 }
1688
1689 switch (s->state) {
1690 case MIGRATION_STATUS_NONE:
1691 case MIGRATION_STATUS_CANCELLED:
1692 case MIGRATION_STATUS_COMPLETED:
1693 case MIGRATION_STATUS_FAILED:
1694 return true;
1695 case MIGRATION_STATUS_SETUP:
1696 case MIGRATION_STATUS_CANCELLING:
1697 case MIGRATION_STATUS_ACTIVE:
1698 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1699 case MIGRATION_STATUS_COLO:
1700 case MIGRATION_STATUS_PRE_SWITCHOVER:
1701 case MIGRATION_STATUS_DEVICE:
1702 case MIGRATION_STATUS_WAIT_UNPLUG:
1703 return false;
1704 case MIGRATION_STATUS__MAX:
1705 g_assert_not_reached();
1706 }
1707
1708 return false;
1709 }
1710
1711 bool migration_is_active(MigrationState *s)
1712 {
1713 return (s->state == MIGRATION_STATUS_ACTIVE ||
1714 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1715 }
1716
1717 void migrate_init(MigrationState *s)
1718 {
1719 /*
1720 * Reinitialise all migration state, except
1721 * parameters/capabilities that the user set, and
1722 * locks.
1723 */
1724 s->cleanup_bh = 0;
1725 s->to_dst_file = NULL;
1726 s->state = MIGRATION_STATUS_NONE;
1727 s->rp_state.from_dst_file = NULL;
1728 s->rp_state.error = false;
1729 s->mbps = 0.0;
1730 s->pages_per_second = 0.0;
1731 s->downtime = 0;
1732 s->expected_downtime = 0;
1733 s->setup_time = 0;
1734 s->start_postcopy = false;
1735 s->postcopy_after_devices = false;
1736 s->migration_thread_running = false;
1737 error_free(s->error);
1738 s->error = NULL;
1739
1740 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1741
1742 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1743 s->total_time = 0;
1744 s->vm_was_running = false;
1745 s->iteration_initial_bytes = 0;
1746 s->threshold_size = 0;
1747 }
1748
1749 static GSList *migration_blockers;
1750
1751 int migrate_add_blocker(Error *reason, Error **errp)
1752 {
1753 if (only_migratable) {
1754 error_propagate_prepend(errp, error_copy(reason),
1755 "disallowing migration blocker "
1756 "(--only-migratable) for: ");
1757 return -EACCES;
1758 }
1759
1760 if (migration_is_idle()) {
1761 migration_blockers = g_slist_prepend(migration_blockers, reason);
1762 return 0;
1763 }
1764
1765 error_propagate_prepend(errp, error_copy(reason),
1766 "disallowing migration blocker "
1767 "(migration in progress) for: ");
1768 return -EBUSY;
1769 }
1770
1771 void migrate_del_blocker(Error *reason)
1772 {
1773 migration_blockers = g_slist_remove(migration_blockers, reason);
1774 }
1775
1776 void qmp_migrate_incoming(const char *uri, Error **errp)
1777 {
1778 Error *local_err = NULL;
1779 static bool once = true;
1780
1781 if (!deferred_incoming) {
1782 error_setg(errp, "For use with '-incoming defer'");
1783 return;
1784 }
1785 if (!once) {
1786 error_setg(errp, "The incoming migration has already been started");
1787 }
1788
1789 qemu_start_incoming_migration(uri, &local_err);
1790
1791 if (local_err) {
1792 error_propagate(errp, local_err);
1793 return;
1794 }
1795
1796 once = false;
1797 }
1798
1799 void qmp_migrate_recover(const char *uri, Error **errp)
1800 {
1801 MigrationIncomingState *mis = migration_incoming_get_current();
1802
1803 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1804 error_setg(errp, "Migrate recover can only be run "
1805 "when postcopy is paused.");
1806 return;
1807 }
1808
1809 if (atomic_cmpxchg(&mis->postcopy_recover_triggered,
1810 false, true) == true) {
1811 error_setg(errp, "Migrate recovery is triggered already");
1812 return;
1813 }
1814
1815 /*
1816 * Note that this call will never start a real migration; it will
1817 * only re-setup the migration stream and poke existing migration
1818 * to continue using that newly established channel.
1819 */
1820 qemu_start_incoming_migration(uri, errp);
1821 }
1822
1823 void qmp_migrate_pause(Error **errp)
1824 {
1825 MigrationState *ms = migrate_get_current();
1826 MigrationIncomingState *mis = migration_incoming_get_current();
1827 int ret;
1828
1829 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1830 /* Source side, during postcopy */
1831 qemu_mutex_lock(&ms->qemu_file_lock);
1832 ret = qemu_file_shutdown(ms->to_dst_file);
1833 qemu_mutex_unlock(&ms->qemu_file_lock);
1834 if (ret) {
1835 error_setg(errp, "Failed to pause source migration");
1836 }
1837 return;
1838 }
1839
1840 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1841 ret = qemu_file_shutdown(mis->from_src_file);
1842 if (ret) {
1843 error_setg(errp, "Failed to pause destination migration");
1844 }
1845 return;
1846 }
1847
1848 error_setg(errp, "migrate-pause is currently only supported "
1849 "during postcopy-active state");
1850 }
1851
1852 bool migration_is_blocked(Error **errp)
1853 {
1854 if (qemu_savevm_state_blocked(errp)) {
1855 return true;
1856 }
1857
1858 if (migration_blockers) {
1859 error_propagate(errp, error_copy(migration_blockers->data));
1860 return true;
1861 }
1862
1863 return false;
1864 }
1865
1866 /* Returns true if continue to migrate, or false if error detected */
1867 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
1868 bool resume, Error **errp)
1869 {
1870 Error *local_err = NULL;
1871
1872 if (resume) {
1873 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1874 error_setg(errp, "Cannot resume if there is no "
1875 "paused migration");
1876 return false;
1877 }
1878
1879 /*
1880 * Postcopy recovery won't work well with release-ram
1881 * capability since release-ram will drop the page buffer as
1882 * long as the page is put into the send buffer. So if there
1883 * is a network failure happened, any page buffers that have
1884 * not yet reached the destination VM but have already been
1885 * sent from the source VM will be lost forever. Let's refuse
1886 * the client from resuming such a postcopy migration.
1887 * Luckily release-ram was designed to only be used when src
1888 * and destination VMs are on the same host, so it should be
1889 * fine.
1890 */
1891 if (migrate_release_ram()) {
1892 error_setg(errp, "Postcopy recovery cannot work "
1893 "when release-ram capability is set");
1894 return false;
1895 }
1896
1897 /* This is a resume, skip init status */
1898 return true;
1899 }
1900
1901 if (migration_is_setup_or_active(s->state) ||
1902 s->state == MIGRATION_STATUS_CANCELLING ||
1903 s->state == MIGRATION_STATUS_COLO) {
1904 error_setg(errp, QERR_MIGRATION_ACTIVE);
1905 return false;
1906 }
1907
1908 if (runstate_check(RUN_STATE_INMIGRATE)) {
1909 error_setg(errp, "Guest is waiting for an incoming migration");
1910 return false;
1911 }
1912
1913 if (migration_is_blocked(errp)) {
1914 return false;
1915 }
1916
1917 if (blk || blk_inc) {
1918 if (migrate_use_block() || migrate_use_block_incremental()) {
1919 error_setg(errp, "Command options are incompatible with "
1920 "current migration capabilities");
1921 return false;
1922 }
1923 migrate_set_block_enabled(true, &local_err);
1924 if (local_err) {
1925 error_propagate(errp, local_err);
1926 return false;
1927 }
1928 s->must_remove_block_options = true;
1929 }
1930
1931 if (blk_inc) {
1932 migrate_set_block_incremental(s, true);
1933 }
1934
1935 migrate_init(s);
1936 /*
1937 * set ram_counters memory to zero for a
1938 * new migration
1939 */
1940 memset(&ram_counters, 0, sizeof(ram_counters));
1941
1942 return true;
1943 }
1944
1945 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1946 bool has_inc, bool inc, bool has_detach, bool detach,
1947 bool has_resume, bool resume, Error **errp)
1948 {
1949 Error *local_err = NULL;
1950 MigrationState *s = migrate_get_current();
1951 const char *p;
1952
1953 if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
1954 has_resume && resume, errp)) {
1955 /* Error detected, put into errp */
1956 return;
1957 }
1958
1959 if (strstart(uri, "tcp:", &p)) {
1960 tcp_start_outgoing_migration(s, p, &local_err);
1961 #ifdef CONFIG_RDMA
1962 } else if (strstart(uri, "rdma:", &p)) {
1963 rdma_start_outgoing_migration(s, p, &local_err);
1964 #endif
1965 } else if (strstart(uri, "exec:", &p)) {
1966 exec_start_outgoing_migration(s, p, &local_err);
1967 } else if (strstart(uri, "unix:", &p)) {
1968 unix_start_outgoing_migration(s, p, &local_err);
1969 } else if (strstart(uri, "fd:", &p)) {
1970 fd_start_outgoing_migration(s, p, &local_err);
1971 } else {
1972 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1973 "a valid migration protocol");
1974 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1975 MIGRATION_STATUS_FAILED);
1976 block_cleanup_parameters(s);
1977 return;
1978 }
1979
1980 if (local_err) {
1981 migrate_fd_error(s, local_err);
1982 error_propagate(errp, local_err);
1983 return;
1984 }
1985 }
1986
1987 void qmp_migrate_cancel(Error **errp)
1988 {
1989 migrate_fd_cancel(migrate_get_current());
1990 }
1991
1992 void qmp_migrate_continue(MigrationStatus state, Error **errp)
1993 {
1994 MigrationState *s = migrate_get_current();
1995 if (s->state != state) {
1996 error_setg(errp, "Migration not in expected state: %s",
1997 MigrationStatus_str(s->state));
1998 return;
1999 }
2000 qemu_sem_post(&s->pause_sem);
2001 }
2002
2003 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
2004 {
2005 MigrateSetParameters p = {
2006 .has_xbzrle_cache_size = true,
2007 .xbzrle_cache_size = value,
2008 };
2009
2010 qmp_migrate_set_parameters(&p, errp);
2011 }
2012
2013 int64_t qmp_query_migrate_cache_size(Error **errp)
2014 {
2015 return migrate_xbzrle_cache_size();
2016 }
2017
2018 void qmp_migrate_set_speed(int64_t value, Error **errp)
2019 {
2020 MigrateSetParameters p = {
2021 .has_max_bandwidth = true,
2022 .max_bandwidth = value,
2023 };
2024
2025 qmp_migrate_set_parameters(&p, errp);
2026 }
2027
2028 void qmp_migrate_set_downtime(double value, Error **errp)
2029 {
2030 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
2031 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
2032 "the range of 0 to %d seconds",
2033 MAX_MIGRATE_DOWNTIME_SECONDS);
2034 return;
2035 }
2036
2037 value *= 1000; /* Convert to milliseconds */
2038 value = MAX(0, MIN(INT64_MAX, value));
2039
2040 MigrateSetParameters p = {
2041 .has_downtime_limit = true,
2042 .downtime_limit = value,
2043 };
2044
2045 qmp_migrate_set_parameters(&p, errp);
2046 }
2047
2048 bool migrate_release_ram(void)
2049 {
2050 MigrationState *s;
2051
2052 s = migrate_get_current();
2053
2054 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2055 }
2056
2057 bool migrate_postcopy_ram(void)
2058 {
2059 MigrationState *s;
2060
2061 s = migrate_get_current();
2062
2063 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2064 }
2065
2066 bool migrate_postcopy(void)
2067 {
2068 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2069 }
2070
2071 bool migrate_auto_converge(void)
2072 {
2073 MigrationState *s;
2074
2075 s = migrate_get_current();
2076
2077 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2078 }
2079
2080 bool migrate_zero_blocks(void)
2081 {
2082 MigrationState *s;
2083
2084 s = migrate_get_current();
2085
2086 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2087 }
2088
2089 bool migrate_postcopy_blocktime(void)
2090 {
2091 MigrationState *s;
2092
2093 s = migrate_get_current();
2094
2095 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2096 }
2097
2098 bool migrate_use_compression(void)
2099 {
2100 MigrationState *s;
2101
2102 s = migrate_get_current();
2103
2104 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2105 }
2106
2107 int migrate_compress_level(void)
2108 {
2109 MigrationState *s;
2110
2111 s = migrate_get_current();
2112
2113 return s->parameters.compress_level;
2114 }
2115
2116 int migrate_compress_threads(void)
2117 {
2118 MigrationState *s;
2119
2120 s = migrate_get_current();
2121
2122 return s->parameters.compress_threads;
2123 }
2124
2125 int migrate_compress_wait_thread(void)
2126 {
2127 MigrationState *s;
2128
2129 s = migrate_get_current();
2130
2131 return s->parameters.compress_wait_thread;
2132 }
2133
2134 int migrate_decompress_threads(void)
2135 {
2136 MigrationState *s;
2137
2138 s = migrate_get_current();
2139
2140 return s->parameters.decompress_threads;
2141 }
2142
2143 bool migrate_dirty_bitmaps(void)
2144 {
2145 MigrationState *s;
2146
2147 s = migrate_get_current();
2148
2149 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2150 }
2151
2152 bool migrate_ignore_shared(void)
2153 {
2154 MigrationState *s;
2155
2156 s = migrate_get_current();
2157
2158 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2159 }
2160
2161 bool migrate_validate_uuid(void)
2162 {
2163 MigrationState *s;
2164
2165 s = migrate_get_current();
2166
2167 return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
2168 }
2169
2170 bool migrate_use_events(void)
2171 {
2172 MigrationState *s;
2173
2174 s = migrate_get_current();
2175
2176 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2177 }
2178
2179 bool migrate_use_multifd(void)
2180 {
2181 MigrationState *s;
2182
2183 s = migrate_get_current();
2184
2185 return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2186 }
2187
2188 bool migrate_pause_before_switchover(void)
2189 {
2190 MigrationState *s;
2191
2192 s = migrate_get_current();
2193
2194 return s->enabled_capabilities[
2195 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2196 }
2197
2198 int migrate_multifd_channels(void)
2199 {
2200 MigrationState *s;
2201
2202 s = migrate_get_current();
2203
2204 return s->parameters.multifd_channels;
2205 }
2206
2207 int migrate_use_xbzrle(void)
2208 {
2209 MigrationState *s;
2210
2211 s = migrate_get_current();
2212
2213 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2214 }
2215
2216 int64_t migrate_xbzrle_cache_size(void)
2217 {
2218 MigrationState *s;
2219
2220 s = migrate_get_current();
2221
2222 return s->parameters.xbzrle_cache_size;
2223 }
2224
2225 static int64_t migrate_max_postcopy_bandwidth(void)
2226 {
2227 MigrationState *s;
2228
2229 s = migrate_get_current();
2230
2231 return s->parameters.max_postcopy_bandwidth;
2232 }
2233
2234 bool migrate_use_block(void)
2235 {
2236 MigrationState *s;
2237
2238 s = migrate_get_current();
2239
2240 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2241 }
2242
2243 bool migrate_use_return_path(void)
2244 {
2245 MigrationState *s;
2246
2247 s = migrate_get_current();
2248
2249 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2250 }
2251
2252 bool migrate_use_block_incremental(void)
2253 {
2254 MigrationState *s;
2255
2256 s = migrate_get_current();
2257
2258 return s->parameters.block_incremental;
2259 }
2260
2261 /* migration thread support */
2262 /*
2263 * Something bad happened to the RP stream, mark an error
2264 * The caller shall print or trace something to indicate why
2265 */
2266 static void mark_source_rp_bad(MigrationState *s)
2267 {
2268 s->rp_state.error = true;
2269 }
2270
2271 static struct rp_cmd_args {
2272 ssize_t len; /* -1 = variable */
2273 const char *name;
2274 } rp_cmd_args[] = {
2275 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
2276 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
2277 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
2278 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
2279 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
2280 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
2281 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
2282 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
2283 };
2284
2285 /*
2286 * Process a request for pages received on the return path,
2287 * We're allowed to send more than requested (e.g. to round to our page size)
2288 * and we don't need to send pages that have already been sent.
2289 */
2290 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2291 ram_addr_t start, size_t len)
2292 {
2293 long our_host_ps = qemu_real_host_page_size;
2294
2295 trace_migrate_handle_rp_req_pages(rbname, start, len);
2296
2297 /*
2298 * Since we currently insist on matching page sizes, just sanity check
2299 * we're being asked for whole host pages.
2300 */
2301 if (start & (our_host_ps-1) ||
2302 (len & (our_host_ps-1))) {
2303 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2304 " len: %zd", __func__, start, len);
2305 mark_source_rp_bad(ms);
2306 return;
2307 }
2308
2309 if (ram_save_queue_pages(rbname, start, len)) {
2310 mark_source_rp_bad(ms);
2311 }
2312 }
2313
2314 /* Return true to retry, false to quit */
2315 static bool postcopy_pause_return_path_thread(MigrationState *s)
2316 {
2317 trace_postcopy_pause_return_path();
2318
2319 qemu_sem_wait(&s->postcopy_pause_rp_sem);
2320
2321 trace_postcopy_pause_return_path_continued();
2322
2323 return true;
2324 }
2325
2326 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2327 {
2328 RAMBlock *block = qemu_ram_block_by_name(block_name);
2329
2330 if (!block) {
2331 error_report("%s: invalid block name '%s'", __func__, block_name);
2332 return -EINVAL;
2333 }
2334
2335 /* Fetch the received bitmap and refresh the dirty bitmap */
2336 return ram_dirty_bitmap_reload(s, block);
2337 }
2338
2339 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2340 {
2341 trace_source_return_path_thread_resume_ack(value);
2342
2343 if (value != MIGRATION_RESUME_ACK_VALUE) {
2344 error_report("%s: illegal resume_ack value %"PRIu32,
2345 __func__, value);
2346 return -1;
2347 }
2348
2349 /* Now both sides are active. */
2350 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2351 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2352
2353 /* Notify send thread that time to continue send pages */
2354 qemu_sem_post(&s->rp_state.rp_sem);
2355
2356 return 0;
2357 }
2358
2359 /*
2360 * Handles messages sent on the return path towards the source VM
2361 *
2362 */
2363 static void *source_return_path_thread(void *opaque)
2364 {
2365 MigrationState *ms = opaque;
2366 QEMUFile *rp = ms->rp_state.from_dst_file;
2367 uint16_t header_len, header_type;
2368 uint8_t buf[512];
2369 uint32_t tmp32, sibling_error;
2370 ram_addr_t start = 0; /* =0 to silence warning */
2371 size_t len = 0, expected_len;
2372 int res;
2373
2374 trace_source_return_path_thread_entry();
2375 rcu_register_thread();
2376
2377 retry:
2378 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2379 migration_is_setup_or_active(ms->state)) {
2380 trace_source_return_path_thread_loop_top();
2381 header_type = qemu_get_be16(rp);
2382 header_len = qemu_get_be16(rp);
2383
2384 if (qemu_file_get_error(rp)) {
2385 mark_source_rp_bad(ms);
2386 goto out;
2387 }
2388
2389 if (header_type >= MIG_RP_MSG_MAX ||
2390 header_type == MIG_RP_MSG_INVALID) {
2391 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2392 header_type, header_len);
2393 mark_source_rp_bad(ms);
2394 goto out;
2395 }
2396
2397 if ((rp_cmd_args[header_type].len != -1 &&
2398 header_len != rp_cmd_args[header_type].len) ||
2399 header_len > sizeof(buf)) {
2400 error_report("RP: Received '%s' message (0x%04x) with"
2401 "incorrect length %d expecting %zu",
2402 rp_cmd_args[header_type].name, header_type, header_len,
2403 (size_t)rp_cmd_args[header_type].len);
2404 mark_source_rp_bad(ms);
2405 goto out;
2406 }
2407
2408 /* We know we've got a valid header by this point */
2409 res = qemu_get_buffer(rp, buf, header_len);
2410 if (res != header_len) {
2411 error_report("RP: Failed reading data for message 0x%04x"
2412 " read %d expected %d",
2413 header_type, res, header_len);
2414 mark_source_rp_bad(ms);
2415 goto out;
2416 }
2417
2418 /* OK, we have the message and the data */
2419 switch (header_type) {
2420 case MIG_RP_MSG_SHUT:
2421 sibling_error = ldl_be_p(buf);
2422 trace_source_return_path_thread_shut(sibling_error);
2423 if (sibling_error) {
2424 error_report("RP: Sibling indicated error %d", sibling_error);
2425 mark_source_rp_bad(ms);
2426 }
2427 /*
2428 * We'll let the main thread deal with closing the RP
2429 * we could do a shutdown(2) on it, but we're the only user
2430 * anyway, so there's nothing gained.
2431 */
2432 goto out;
2433
2434 case MIG_RP_MSG_PONG:
2435 tmp32 = ldl_be_p(buf);
2436 trace_source_return_path_thread_pong(tmp32);
2437 break;
2438
2439 case MIG_RP_MSG_REQ_PAGES:
2440 start = ldq_be_p(buf);
2441 len = ldl_be_p(buf + 8);
2442 migrate_handle_rp_req_pages(ms, NULL, start, len);
2443 break;
2444
2445 case MIG_RP_MSG_REQ_PAGES_ID:
2446 expected_len = 12 + 1; /* header + termination */
2447
2448 if (header_len >= expected_len) {
2449 start = ldq_be_p(buf);
2450 len = ldl_be_p(buf + 8);
2451 /* Now we expect an idstr */
2452 tmp32 = buf[12]; /* Length of the following idstr */
2453 buf[13 + tmp32] = '\0';
2454 expected_len += tmp32;
2455 }
2456 if (header_len != expected_len) {
2457 error_report("RP: Req_Page_id with length %d expecting %zd",
2458 header_len, expected_len);
2459 mark_source_rp_bad(ms);
2460 goto out;
2461 }
2462 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2463 break;
2464
2465 case MIG_RP_MSG_RECV_BITMAP:
2466 if (header_len < 1) {
2467 error_report("%s: missing block name", __func__);
2468 mark_source_rp_bad(ms);
2469 goto out;
2470 }
2471 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2472 buf[buf[0] + 1] = '\0';
2473 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2474 mark_source_rp_bad(ms);
2475 goto out;
2476 }
2477 break;
2478
2479 case MIG_RP_MSG_RESUME_ACK:
2480 tmp32 = ldl_be_p(buf);
2481 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2482 mark_source_rp_bad(ms);
2483 goto out;
2484 }
2485 break;
2486
2487 default:
2488 break;
2489 }
2490 }
2491
2492 out:
2493 res = qemu_file_get_error(rp);
2494 if (res) {
2495 if (res == -EIO && migration_in_postcopy()) {
2496 /*
2497 * Maybe there is something we can do: it looks like a
2498 * network down issue, and we pause for a recovery.
2499 */
2500 if (postcopy_pause_return_path_thread(ms)) {
2501 /* Reload rp, reset the rest */
2502 if (rp != ms->rp_state.from_dst_file) {
2503 qemu_fclose(rp);
2504 rp = ms->rp_state.from_dst_file;
2505 }
2506 ms->rp_state.error = false;
2507 goto retry;
2508 }
2509 }
2510
2511 trace_source_return_path_thread_bad_end();
2512 mark_source_rp_bad(ms);
2513 }
2514
2515 trace_source_return_path_thread_end();
2516 ms->rp_state.from_dst_file = NULL;
2517 qemu_fclose(rp);
2518 rcu_unregister_thread();
2519 return NULL;
2520 }
2521
2522 static int open_return_path_on_source(MigrationState *ms,
2523 bool create_thread)
2524 {
2525
2526 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2527 if (!ms->rp_state.from_dst_file) {
2528 return -1;
2529 }
2530
2531 trace_open_return_path_on_source();
2532
2533 if (!create_thread) {
2534 /* We're done */
2535 return 0;
2536 }
2537
2538 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2539 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2540
2541 trace_open_return_path_on_source_continue();
2542
2543 return 0;
2544 }
2545
2546 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2547 static int await_return_path_close_on_source(MigrationState *ms)
2548 {
2549 /*
2550 * If this is a normal exit then the destination will send a SHUT and the
2551 * rp_thread will exit, however if there's an error we need to cause
2552 * it to exit.
2553 */
2554 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2555 /*
2556 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2557 * waiting for the destination.
2558 */
2559 qemu_file_shutdown(ms->rp_state.from_dst_file);
2560 mark_source_rp_bad(ms);
2561 }
2562 trace_await_return_path_close_on_source_joining();
2563 qemu_thread_join(&ms->rp_state.rp_thread);
2564 trace_await_return_path_close_on_source_close();
2565 return ms->rp_state.error;
2566 }
2567
2568 /*
2569 * Switch from normal iteration to postcopy
2570 * Returns non-0 on error
2571 */
2572 static int postcopy_start(MigrationState *ms)
2573 {
2574 int ret;
2575 QIOChannelBuffer *bioc;
2576 QEMUFile *fb;
2577 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2578 int64_t bandwidth = migrate_max_postcopy_bandwidth();
2579 bool restart_block = false;
2580 int cur_state = MIGRATION_STATUS_ACTIVE;
2581 if (!migrate_pause_before_switchover()) {
2582 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2583 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2584 }
2585
2586 trace_postcopy_start();
2587 qemu_mutex_lock_iothread();
2588 trace_postcopy_start_set_run();
2589
2590 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2591 global_state_store();
2592 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2593 if (ret < 0) {
2594 goto fail;
2595 }
2596
2597 ret = migration_maybe_pause(ms, &cur_state,
2598 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2599 if (ret < 0) {
2600 goto fail;
2601 }
2602
2603 ret = bdrv_inactivate_all();
2604 if (ret < 0) {
2605 goto fail;
2606 }
2607 restart_block = true;
2608
2609 /*
2610 * Cause any non-postcopiable, but iterative devices to
2611 * send out their final data.
2612 */
2613 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2614
2615 /*
2616 * in Finish migrate and with the io-lock held everything should
2617 * be quiet, but we've potentially still got dirty pages and we
2618 * need to tell the destination to throw any pages it's already received
2619 * that are dirty
2620 */
2621 if (migrate_postcopy_ram()) {
2622 if (ram_postcopy_send_discard_bitmap(ms)) {
2623 error_report("postcopy send discard bitmap failed");
2624 goto fail;
2625 }
2626 }
2627
2628 /*
2629 * send rest of state - note things that are doing postcopy
2630 * will notice we're in POSTCOPY_ACTIVE and not actually
2631 * wrap their state up here
2632 */
2633 /* 0 max-postcopy-bandwidth means unlimited */
2634 if (!bandwidth) {
2635 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2636 } else {
2637 qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2638 }
2639 if (migrate_postcopy_ram()) {
2640 /* Ping just for debugging, helps line traces up */
2641 qemu_savevm_send_ping(ms->to_dst_file, 2);
2642 }
2643
2644 /*
2645 * While loading the device state we may trigger page transfer
2646 * requests and the fd must be free to process those, and thus
2647 * the destination must read the whole device state off the fd before
2648 * it starts processing it. Unfortunately the ad-hoc migration format
2649 * doesn't allow the destination to know the size to read without fully
2650 * parsing it through each devices load-state code (especially the open
2651 * coded devices that use get/put).
2652 * So we wrap the device state up in a package with a length at the start;
2653 * to do this we use a qemu_buf to hold the whole of the device state.
2654 */
2655 bioc = qio_channel_buffer_new(4096);
2656 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2657 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2658 object_unref(OBJECT(bioc));
2659
2660 /*
2661 * Make sure the receiver can get incoming pages before we send the rest
2662 * of the state
2663 */
2664 qemu_savevm_send_postcopy_listen(fb);
2665
2666 qemu_savevm_state_complete_precopy(fb, false, false);
2667 if (migrate_postcopy_ram()) {
2668 qemu_savevm_send_ping(fb, 3);
2669 }
2670
2671 qemu_savevm_send_postcopy_run(fb);
2672
2673 /* <><> end of stuff going into the package */
2674
2675 /* Last point of recovery; as soon as we send the package the destination
2676 * can open devices and potentially start running.
2677 * Lets just check again we've not got any errors.
2678 */
2679 ret = qemu_file_get_error(ms->to_dst_file);
2680 if (ret) {
2681 error_report("postcopy_start: Migration stream errored (pre package)");
2682 goto fail_closefb;
2683 }
2684
2685 restart_block = false;
2686
2687 /* Now send that blob */
2688 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2689 goto fail_closefb;
2690 }
2691 qemu_fclose(fb);
2692
2693 /* Send a notify to give a chance for anything that needs to happen
2694 * at the transition to postcopy and after the device state; in particular
2695 * spice needs to trigger a transition now
2696 */
2697 ms->postcopy_after_devices = true;
2698 notifier_list_notify(&migration_state_notifiers, ms);
2699
2700 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2701
2702 qemu_mutex_unlock_iothread();
2703
2704 if (migrate_postcopy_ram()) {
2705 /*
2706 * Although this ping is just for debug, it could potentially be
2707 * used for getting a better measurement of downtime at the source.
2708 */
2709 qemu_savevm_send_ping(ms->to_dst_file, 4);
2710 }
2711
2712 if (migrate_release_ram()) {
2713 ram_postcopy_migrated_memory_release(ms);
2714 }
2715
2716 ret = qemu_file_get_error(ms->to_dst_file);
2717 if (ret) {
2718 error_report("postcopy_start: Migration stream errored");
2719 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2720 MIGRATION_STATUS_FAILED);
2721 }
2722
2723 return ret;
2724
2725 fail_closefb:
2726 qemu_fclose(fb);
2727 fail:
2728 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2729 MIGRATION_STATUS_FAILED);
2730 if (restart_block) {
2731 /* A failure happened early enough that we know the destination hasn't
2732 * accessed block devices, so we're safe to recover.
2733 */
2734 Error *local_err = NULL;
2735
2736 bdrv_invalidate_cache_all(&local_err);
2737 if (local_err) {
2738 error_report_err(local_err);
2739 }
2740 }
2741 qemu_mutex_unlock_iothread();
2742 return -1;
2743 }
2744
2745 /**
2746 * migration_maybe_pause: Pause if required to by
2747 * migrate_pause_before_switchover called with the iothread locked
2748 * Returns: 0 on success
2749 */
2750 static int migration_maybe_pause(MigrationState *s,
2751 int *current_active_state,
2752 int new_state)
2753 {
2754 if (!migrate_pause_before_switchover()) {
2755 return 0;
2756 }
2757
2758 /* Since leaving this state is not atomic with posting the semaphore
2759 * it's possible that someone could have issued multiple migrate_continue
2760 * and the semaphore is incorrectly positive at this point;
2761 * the docs say it's undefined to reinit a semaphore that's already
2762 * init'd, so use timedwait to eat up any existing posts.
2763 */
2764 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2765 /* This block intentionally left blank */
2766 }
2767
2768 qemu_mutex_unlock_iothread();
2769 migrate_set_state(&s->state, *current_active_state,
2770 MIGRATION_STATUS_PRE_SWITCHOVER);
2771 qemu_sem_wait(&s->pause_sem);
2772 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2773 new_state);
2774 *current_active_state = new_state;
2775 qemu_mutex_lock_iothread();
2776
2777 return s->state == new_state ? 0 : -EINVAL;
2778 }
2779
2780 /**
2781 * migration_completion: Used by migration_thread when there's not much left.
2782 * The caller 'breaks' the loop when this returns.
2783 *
2784 * @s: Current migration state
2785 */
2786 static void migration_completion(MigrationState *s)
2787 {
2788 int ret;
2789 int current_active_state = s->state;
2790
2791 if (s->state == MIGRATION_STATUS_ACTIVE) {
2792 qemu_mutex_lock_iothread();
2793 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2794 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2795 s->vm_was_running = runstate_is_running();
2796 ret = global_state_store();
2797
2798 if (!ret) {
2799 bool inactivate = !migrate_colo_enabled();
2800 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2801 if (ret >= 0) {
2802 ret = migration_maybe_pause(s, &current_active_state,
2803 MIGRATION_STATUS_DEVICE);
2804 }
2805 if (ret >= 0) {
2806 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2807 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2808 inactivate);
2809 }
2810 if (inactivate && ret >= 0) {
2811 s->block_inactive = true;
2812 }
2813 }
2814 qemu_mutex_unlock_iothread();
2815
2816 if (ret < 0) {
2817 goto fail;
2818 }
2819 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2820 trace_migration_completion_postcopy_end();
2821
2822 qemu_savevm_state_complete_postcopy(s->to_dst_file);
2823 trace_migration_completion_postcopy_end_after_complete();
2824 }
2825
2826 /*
2827 * If rp was opened we must clean up the thread before
2828 * cleaning everything else up (since if there are no failures
2829 * it will wait for the destination to send it's status in
2830 * a SHUT command).
2831 */
2832 if (s->rp_state.from_dst_file) {
2833 int rp_error;
2834 trace_migration_return_path_end_before();
2835 rp_error = await_return_path_close_on_source(s);
2836 trace_migration_return_path_end_after(rp_error);
2837 if (rp_error) {
2838 goto fail_invalidate;
2839 }
2840 }
2841
2842 if (qemu_file_get_error(s->to_dst_file)) {
2843 trace_migration_completion_file_err();
2844 goto fail_invalidate;
2845 }
2846
2847 if (!migrate_colo_enabled()) {
2848 migrate_set_state(&s->state, current_active_state,
2849 MIGRATION_STATUS_COMPLETED);
2850 }
2851
2852 return;
2853
2854 fail_invalidate:
2855 /* If not doing postcopy, vm_start() will be called: let's regain
2856 * control on images.
2857 */
2858 if (s->state == MIGRATION_STATUS_ACTIVE ||
2859 s->state == MIGRATION_STATUS_DEVICE) {
2860 Error *local_err = NULL;
2861
2862 qemu_mutex_lock_iothread();
2863 bdrv_invalidate_cache_all(&local_err);
2864 if (local_err) {
2865 error_report_err(local_err);
2866 } else {
2867 s->block_inactive = false;
2868 }
2869 qemu_mutex_unlock_iothread();
2870 }
2871
2872 fail:
2873 migrate_set_state(&s->state, current_active_state,
2874 MIGRATION_STATUS_FAILED);
2875 }
2876
2877 bool migrate_colo_enabled(void)
2878 {
2879 MigrationState *s = migrate_get_current();
2880 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2881 }
2882
2883 typedef enum MigThrError {
2884 /* No error detected */
2885 MIG_THR_ERR_NONE = 0,
2886 /* Detected error, but resumed successfully */
2887 MIG_THR_ERR_RECOVERED = 1,
2888 /* Detected fatal error, need to exit */
2889 MIG_THR_ERR_FATAL = 2,
2890 } MigThrError;
2891
2892 static int postcopy_resume_handshake(MigrationState *s)
2893 {
2894 qemu_savevm_send_postcopy_resume(s->to_dst_file);
2895
2896 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2897 qemu_sem_wait(&s->rp_state.rp_sem);
2898 }
2899
2900 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2901 return 0;
2902 }
2903
2904 return -1;
2905 }
2906
2907 /* Return zero if success, or <0 for error */
2908 static int postcopy_do_resume(MigrationState *s)
2909 {
2910 int ret;
2911
2912 /*
2913 * Call all the resume_prepare() hooks, so that modules can be
2914 * ready for the migration resume.
2915 */
2916 ret = qemu_savevm_state_resume_prepare(s);
2917 if (ret) {
2918 error_report("%s: resume_prepare() failure detected: %d",
2919 __func__, ret);
2920 return ret;
2921 }
2922
2923 /*
2924 * Last handshake with destination on the resume (destination will
2925 * switch to postcopy-active afterwards)
2926 */
2927 ret = postcopy_resume_handshake(s);
2928 if (ret) {
2929 error_report("%s: handshake failed: %d", __func__, ret);
2930 return ret;
2931 }
2932
2933 return 0;
2934 }
2935
2936 /*
2937 * We don't return until we are in a safe state to continue current
2938 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2939 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2940 */
2941 static MigThrError postcopy_pause(MigrationState *s)
2942 {
2943 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2944
2945 while (true) {
2946 QEMUFile *file;
2947
2948 migrate_set_state(&s->state, s->state,
2949 MIGRATION_STATUS_POSTCOPY_PAUSED);
2950
2951 /* Current channel is possibly broken. Release it. */
2952 assert(s->to_dst_file);
2953 qemu_mutex_lock(&s->qemu_file_lock);
2954 file = s->to_dst_file;
2955 s->to_dst_file = NULL;
2956 qemu_mutex_unlock(&s->qemu_file_lock);
2957
2958 qemu_file_shutdown(file);
2959 qemu_fclose(file);
2960
2961 error_report("Detected IO failure for postcopy. "
2962 "Migration paused.");
2963
2964 /*
2965 * We wait until things fixed up. Then someone will setup the
2966 * status back for us.
2967 */
2968 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2969 qemu_sem_wait(&s->postcopy_pause_sem);
2970 }
2971
2972 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2973 /* Woken up by a recover procedure. Give it a shot */
2974
2975 /*
2976 * Firstly, let's wake up the return path now, with a new
2977 * return path channel.
2978 */
2979 qemu_sem_post(&s->postcopy_pause_rp_sem);
2980
2981 /* Do the resume logic */
2982 if (postcopy_do_resume(s) == 0) {
2983 /* Let's continue! */
2984 trace_postcopy_pause_continued();
2985 return MIG_THR_ERR_RECOVERED;
2986 } else {
2987 /*
2988 * Something wrong happened during the recovery, let's
2989 * pause again. Pause is always better than throwing
2990 * data away.
2991 */
2992 continue;
2993 }
2994 } else {
2995 /* This is not right... Time to quit. */
2996 return MIG_THR_ERR_FATAL;
2997 }
2998 }
2999 }
3000
3001 static MigThrError migration_detect_error(MigrationState *s)
3002 {
3003 int ret;
3004 int state = s->state;
3005 Error *local_error = NULL;
3006
3007 if (state == MIGRATION_STATUS_CANCELLING ||
3008 state == MIGRATION_STATUS_CANCELLED) {
3009 /* End the migration, but don't set the state to failed */
3010 return MIG_THR_ERR_FATAL;
3011 }
3012
3013 /* Try to detect any file errors */
3014 ret = qemu_file_get_error_obj(s->to_dst_file, &local_error);
3015 if (!ret) {
3016 /* Everything is fine */
3017 assert(!local_error);
3018 return MIG_THR_ERR_NONE;
3019 }
3020
3021 if (local_error) {
3022 migrate_set_error(s, local_error);
3023 error_free(local_error);
3024 }
3025
3026 if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
3027 /*
3028 * For postcopy, we allow the network to be down for a
3029 * while. After that, it can be continued by a
3030 * recovery phase.
3031 */
3032 return postcopy_pause(s);
3033 } else {
3034 /*
3035 * For precopy (or postcopy with error outside IO), we fail
3036 * with no time.
3037 */
3038 migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
3039 trace_migration_thread_file_err();
3040
3041 /* Time to stop the migration, now. */
3042 return MIG_THR_ERR_FATAL;
3043 }
3044 }
3045
3046 /* How many bytes have we transferred since the beginning of the migration */
3047 static uint64_t migration_total_bytes(MigrationState *s)
3048 {
3049 return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3050 }
3051
3052 static void migration_calculate_complete(MigrationState *s)
3053 {
3054 uint64_t bytes = migration_total_bytes(s);
3055 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3056 int64_t transfer_time;
3057
3058 s->total_time = end_time - s->start_time;
3059 if (!s->downtime) {
3060 /*
3061 * It's still not set, so we are precopy migration. For
3062 * postcopy, downtime is calculated during postcopy_start().
3063 */
3064 s->downtime = end_time - s->downtime_start;
3065 }
3066
3067 transfer_time = s->total_time - s->setup_time;
3068 if (transfer_time) {
3069 s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3070 }
3071 }
3072
3073 static void update_iteration_initial_status(MigrationState *s)
3074 {
3075 /*
3076 * Update these three fields at the same time to avoid mismatch info lead
3077 * wrong speed calculation.
3078 */
3079 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3080 s->iteration_initial_bytes = migration_total_bytes(s);
3081 s->iteration_initial_pages = ram_get_total_transferred_pages();
3082 }
3083
3084 static void migration_update_counters(MigrationState *s,
3085 int64_t current_time)
3086 {
3087 uint64_t transferred, transferred_pages, time_spent;
3088 uint64_t current_bytes; /* bytes transferred since the beginning */
3089 double bandwidth;
3090
3091 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3092 return;
3093 }
3094
3095 current_bytes = migration_total_bytes(s);
3096 transferred = current_bytes - s->iteration_initial_bytes;
3097 time_spent = current_time - s->iteration_start_time;
3098 bandwidth = (double)transferred / time_spent;
3099 s->threshold_size = bandwidth * s->parameters.downtime_limit;
3100
3101 s->mbps = (((double) transferred * 8.0) /
3102 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3103
3104 transferred_pages = ram_get_total_transferred_pages() -
3105 s->iteration_initial_pages;
3106 s->pages_per_second = (double) transferred_pages /
3107 (((double) time_spent / 1000.0));
3108
3109 /*
3110 * if we haven't sent anything, we don't want to
3111 * recalculate. 10000 is a small enough number for our purposes
3112 */
3113 if (ram_counters.dirty_pages_rate && transferred > 10000) {
3114 s->expected_downtime = ram_counters.remaining / bandwidth;
3115 }
3116
3117 qemu_file_reset_rate_limit(s->to_dst_file);
3118
3119 update_iteration_initial_status(s);
3120
3121 trace_migrate_transferred(transferred, time_spent,
3122 bandwidth, s->threshold_size);
3123 }
3124
3125 /* Migration thread iteration status */
3126 typedef enum {
3127 MIG_ITERATE_RESUME, /* Resume current iteration */
3128 MIG_ITERATE_SKIP, /* Skip current iteration */
3129 MIG_ITERATE_BREAK, /* Break the loop */
3130 } MigIterateState;
3131
3132 /*
3133 * Return true if continue to the next iteration directly, false
3134 * otherwise.
3135 */
3136 static MigIterateState migration_iteration_run(MigrationState *s)
3137 {
3138 uint64_t pending_size, pend_pre, pend_compat, pend_post;
3139 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3140
3141 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3142 &pend_compat, &pend_post);
3143 pending_size = pend_pre + pend_compat + pend_post;
3144
3145 trace_migrate_pending(pending_size, s->threshold_size,
3146 pend_pre, pend_compat, pend_post);
3147
3148 if (pending_size && pending_size >= s->threshold_size) {
3149 /* Still a significant amount to transfer */
3150 if (!in_postcopy && pend_pre <= s->threshold_size &&
3151 atomic_read(&s->start_postcopy)) {
3152 if (postcopy_start(s)) {
3153 error_report("%s: postcopy failed to start", __func__);
3154 }
3155 return MIG_ITERATE_SKIP;
3156 }
3157 /* Just another iteration step */
3158 qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
3159 } else {
3160 trace_migration_thread_low_pending(pending_size);
3161 migration_completion(s);
3162 return MIG_ITERATE_BREAK;
3163 }
3164
3165 return MIG_ITERATE_RESUME;
3166 }
3167
3168 static void migration_iteration_finish(MigrationState *s)
3169 {
3170 /* If we enabled cpu throttling for auto-converge, turn it off. */
3171 cpu_throttle_stop();
3172
3173 qemu_mutex_lock_iothread();
3174 switch (s->state) {
3175 case MIGRATION_STATUS_COMPLETED:
3176 migration_calculate_complete(s);
3177 runstate_set(RUN_STATE_POSTMIGRATE);
3178 break;
3179
3180 case MIGRATION_STATUS_ACTIVE:
3181 /*
3182 * We should really assert here, but since it's during
3183 * migration, let's try to reduce the usage of assertions.
3184 */
3185 if (!migrate_colo_enabled()) {
3186 error_report("%s: critical error: calling COLO code without "
3187 "COLO enabled", __func__);
3188 }
3189 migrate_start_colo_process(s);
3190 /*
3191 * Fixme: we will run VM in COLO no matter its old running state.
3192 * After exited COLO, we will keep running.
3193 */
3194 s->vm_was_running = true;
3195 /* Fallthrough */
3196 case MIGRATION_STATUS_FAILED:
3197 case MIGRATION_STATUS_CANCELLED:
3198 case MIGRATION_STATUS_CANCELLING:
3199 if (s->vm_was_running) {
3200 vm_start();
3201 } else {
3202 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3203 runstate_set(RUN_STATE_POSTMIGRATE);
3204 }
3205 }
3206 break;
3207
3208 default:
3209 /* Should not reach here, but if so, forgive the VM. */
3210 error_report("%s: Unknown ending state %d", __func__, s->state);
3211 break;
3212 }
3213 migrate_fd_cleanup_schedule(s);
3214 qemu_mutex_unlock_iothread();
3215 }
3216
3217 void migration_make_urgent_request(void)
3218 {
3219 qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3220 }
3221
3222 void migration_consume_urgent_request(void)
3223 {
3224 qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3225 }
3226
3227 /*
3228 * Master migration thread on the source VM.
3229 * It drives the migration and pumps the data down the outgoing channel.
3230 */
3231 static void *migration_thread(void *opaque)
3232 {
3233 MigrationState *s = opaque;
3234 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3235 MigThrError thr_error;
3236 bool urgent = false;
3237
3238 rcu_register_thread();
3239
3240 object_ref(OBJECT(s));
3241 update_iteration_initial_status(s);
3242
3243 qemu_savevm_state_header(s->to_dst_file);
3244
3245 /*
3246 * If we opened the return path, we need to make sure dst has it
3247 * opened as well.
3248 */
3249 if (s->rp_state.from_dst_file) {
3250 /* Now tell the dest that it should open its end so it can reply */
3251 qemu_savevm_send_open_return_path(s->to_dst_file);
3252
3253 /* And do a ping that will make stuff easier to debug */
3254 qemu_savevm_send_ping(s->to_dst_file, 1);
3255 }
3256
3257 if (migrate_postcopy()) {
3258 /*
3259 * Tell the destination that we *might* want to do postcopy later;
3260 * if the other end can't do postcopy it should fail now, nice and
3261 * early.
3262 */
3263 qemu_savevm_send_postcopy_advise(s->to_dst_file);
3264 }
3265
3266 if (migrate_colo_enabled()) {
3267 /* Notify migration destination that we enable COLO */
3268 qemu_savevm_send_colo_enable(s->to_dst_file);
3269 }
3270
3271 qemu_savevm_state_setup(s->to_dst_file);
3272
3273 if (qemu_savevm_nr_failover_devices()) {
3274 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3275 MIGRATION_STATUS_WAIT_UNPLUG);
3276
3277 while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3278 qemu_savevm_state_guest_unplug_pending()) {
3279 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3280 }
3281
3282 migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG,
3283 MIGRATION_STATUS_ACTIVE);
3284 }
3285
3286 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3287 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3288 MIGRATION_STATUS_ACTIVE);
3289
3290 trace_migration_thread_setup_complete();
3291
3292 while (migration_is_active(s)) {
3293 int64_t current_time;
3294
3295 if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3296 MigIterateState iter_state = migration_iteration_run(s);
3297 if (iter_state == MIG_ITERATE_SKIP) {
3298 continue;
3299 } else if (iter_state == MIG_ITERATE_BREAK) {
3300 break;
3301 }
3302 }
3303
3304 /*
3305 * Try to detect any kind of failures, and see whether we
3306 * should stop the migration now.
3307 */
3308 thr_error = migration_detect_error(s);
3309 if (thr_error == MIG_THR_ERR_FATAL) {
3310 /* Stop migration */
3311 break;
3312 } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3313 /*
3314 * Just recovered from a e.g. network failure, reset all
3315 * the local variables. This is important to avoid
3316 * breaking transferred_bytes and bandwidth calculation
3317 */
3318 update_iteration_initial_status(s);
3319 }
3320
3321 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3322
3323 migration_update_counters(s, current_time);
3324
3325 urgent = false;
3326 if (qemu_file_rate_limit(s->to_dst_file)) {
3327 /* Wait for a delay to do rate limiting OR
3328 * something urgent to post the semaphore.
3329 */
3330 int ms = s->iteration_start_time + BUFFER_DELAY - current_time;
3331 trace_migration_thread_ratelimit_pre(ms);
3332 if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3333 /* We were worken by one or more urgent things but
3334 * the timedwait will have consumed one of them.
3335 * The service routine for the urgent wake will dec
3336 * the semaphore itself for each item it consumes,
3337 * so add this one we just eat back.
3338 */
3339 qemu_sem_post(&s->rate_limit_sem);
3340 urgent = true;
3341 }
3342 trace_migration_thread_ratelimit_post(urgent);
3343 }
3344 }
3345
3346 trace_migration_thread_after_loop();
3347 migration_iteration_finish(s);
3348 object_unref(OBJECT(s));
3349 rcu_unregister_thread();
3350 return NULL;
3351 }
3352
3353 void migrate_fd_connect(MigrationState *s, Error *error_in)
3354 {
3355 int64_t rate_limit;
3356 bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
3357
3358 s->expected_downtime = s->parameters.downtime_limit;
3359 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s);
3360 if (error_in) {
3361 migrate_fd_error(s, error_in);
3362 migrate_fd_cleanup(s);
3363 return;
3364 }
3365
3366 if (resume) {
3367 /* This is a resumed migration */
3368 rate_limit = s->parameters.max_postcopy_bandwidth /
3369 XFER_LIMIT_RATIO;
3370 } else {
3371 /* This is a fresh new migration */
3372 rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
3373
3374 /* Notify before starting migration thread */
3375 notifier_list_notify(&migration_state_notifiers, s);
3376 }
3377
3378 qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
3379 qemu_file_set_blocking(s->to_dst_file, true);
3380
3381 /*
3382 * Open the return path. For postcopy, it is used exclusively. For
3383 * precopy, only if user specified "return-path" capability would
3384 * QEMU uses the return path.
3385 */
3386 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3387 if (open_return_path_on_source(s, !resume)) {
3388 error_report("Unable to open return-path for postcopy");
3389 migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3390 migrate_fd_cleanup(s);
3391 return;
3392 }
3393 }
3394
3395 if (resume) {
3396 /* Wakeup the main migration thread to do the recovery */
3397 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3398 MIGRATION_STATUS_POSTCOPY_RECOVER);
3399 qemu_sem_post(&s->postcopy_pause_sem);
3400 return;
3401 }
3402
3403 if (multifd_save_setup() != 0) {
3404 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3405 MIGRATION_STATUS_FAILED);
3406 migrate_fd_cleanup(s);
3407 return;
3408 }
3409 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
3410 QEMU_THREAD_JOINABLE);
3411 s->migration_thread_running = true;
3412 }
3413
3414 void migration_global_dump(Monitor *mon)
3415 {
3416 MigrationState *ms = migrate_get_current();
3417
3418 monitor_printf(mon, "globals:\n");
3419 monitor_printf(mon, "store-global-state: %s\n",
3420 ms->store_global_state ? "on" : "off");
3421 monitor_printf(mon, "only-migratable: %s\n",
3422 only_migratable ? "on" : "off");
3423 monitor_printf(mon, "send-configuration: %s\n",
3424 ms->send_configuration ? "on" : "off");
3425 monitor_printf(mon, "send-section-footer: %s\n",
3426 ms->send_section_footer ? "on" : "off");
3427 monitor_printf(mon, "decompress-error-check: %s\n",
3428 ms->decompress_error_check ? "on" : "off");
3429 monitor_printf(mon, "clear-bitmap-shift: %u\n",
3430 ms->clear_bitmap_shift);
3431 }
3432
3433 #define DEFINE_PROP_MIG_CAP(name, x) \
3434 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3435
3436 static Property migration_properties[] = {
3437 DEFINE_PROP_BOOL("store-global-state", MigrationState,
3438 store_global_state, true),
3439 DEFINE_PROP_BOOL("send-configuration", MigrationState,
3440 send_configuration, true),
3441 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
3442 send_section_footer, true),
3443 DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
3444 decompress_error_check, true),
3445 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState,
3446 clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT),
3447
3448 /* Migration parameters */
3449 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
3450 parameters.compress_level,
3451 DEFAULT_MIGRATE_COMPRESS_LEVEL),
3452 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
3453 parameters.compress_threads,
3454 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
3455 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
3456 parameters.compress_wait_thread, true),
3457 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
3458 parameters.decompress_threads,
3459 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
3460 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
3461 parameters.cpu_throttle_initial,
3462 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
3463 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
3464 parameters.cpu_throttle_increment,
3465 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
3466 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
3467 parameters.max_bandwidth, MAX_THROTTLE),
3468 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
3469 parameters.downtime_limit,
3470 DEFAULT_MIGRATE_SET_DOWNTIME),
3471 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
3472 parameters.x_checkpoint_delay,
3473 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
3474 DEFINE_PROP_UINT8("multifd-channels", MigrationState,
3475 parameters.multifd_channels,
3476 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
3477 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
3478 parameters.xbzrle_cache_size,
3479 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
3480 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
3481 parameters.max_postcopy_bandwidth,
3482 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
3483 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState,
3484 parameters.max_cpu_throttle,
3485 DEFAULT_MIGRATE_MAX_CPU_THROTTLE),
3486 DEFINE_PROP_SIZE("announce-initial", MigrationState,
3487 parameters.announce_initial,
3488 DEFAULT_MIGRATE_ANNOUNCE_INITIAL),
3489 DEFINE_PROP_SIZE("announce-max", MigrationState,
3490 parameters.announce_max,
3491 DEFAULT_MIGRATE_ANNOUNCE_MAX),
3492 DEFINE_PROP_SIZE("announce-rounds", MigrationState,
3493 parameters.announce_rounds,
3494 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS),
3495 DEFINE_PROP_SIZE("announce-step", MigrationState,
3496 parameters.announce_step,
3497 DEFAULT_MIGRATE_ANNOUNCE_STEP),
3498
3499 /* Migration capabilities */
3500 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
3501 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
3502 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
3503 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
3504 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
3505 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
3506 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
3507 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
3508 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
3509 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
3510 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
3511 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD),
3512
3513 DEFINE_PROP_END_OF_LIST(),
3514 };
3515
3516 static void migration_class_init(ObjectClass *klass, void *data)
3517 {
3518 DeviceClass *dc = DEVICE_CLASS(klass);
3519
3520 dc->user_creatable = false;
3521 dc->props = migration_properties;
3522 }
3523
3524 static void migration_instance_finalize(Object *obj)
3525 {
3526 MigrationState *ms = MIGRATION_OBJ(obj);
3527 MigrationParameters *params = &ms->parameters;
3528
3529 qemu_mutex_destroy(&ms->error_mutex);
3530 qemu_mutex_destroy(&ms->qemu_file_lock);
3531 g_free(params->tls_hostname);
3532 g_free(params->tls_creds);
3533 qemu_sem_destroy(&ms->wait_unplug_sem);
3534 qemu_sem_destroy(&ms->rate_limit_sem);
3535 qemu_sem_destroy(&ms->pause_sem);
3536 qemu_sem_destroy(&ms->postcopy_pause_sem);
3537 qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
3538 qemu_sem_destroy(&ms->rp_state.rp_sem);
3539 error_free(ms->error);
3540 }
3541
3542 static void migration_instance_init(Object *obj)
3543 {
3544 MigrationState *ms = MIGRATION_OBJ(obj);
3545 MigrationParameters *params = &ms->parameters;
3546
3547 ms->state = MIGRATION_STATUS_NONE;
3548 ms->mbps = -1;
3549 ms->pages_per_second = -1;
3550 qemu_sem_init(&ms->pause_sem, 0);
3551 qemu_mutex_init(&ms->error_mutex);
3552
3553 params->tls_hostname = g_strdup("");
3554 params->tls_creds = g_strdup("");
3555
3556 /* Set has_* up only for parameter checks */
3557 params->has_compress_level = true;
3558 params->has_compress_threads = true;
3559 params->has_decompress_threads = true;
3560 params->has_cpu_throttle_initial = true;
3561 params->has_cpu_throttle_increment = true;
3562 params->has_max_bandwidth = true;
3563 params->has_downtime_limit = true;
3564 params->has_x_checkpoint_delay = true;
3565 params->has_block_incremental = true;
3566 params->has_multifd_channels = true;
3567 params->has_xbzrle_cache_size = true;
3568 params->has_max_postcopy_bandwidth = true;
3569 params->has_max_cpu_throttle = true;
3570 params->has_announce_initial = true;
3571 params->has_announce_max = true;
3572 params->has_announce_rounds = true;
3573 params->has_announce_step = true;
3574
3575 qemu_sem_init(&ms->postcopy_pause_sem, 0);
3576 qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
3577 qemu_sem_init(&ms->rp_state.rp_sem, 0);
3578 qemu_sem_init(&ms->rate_limit_sem, 0);
3579 qemu_sem_init(&ms->wait_unplug_sem, 0);
3580 qemu_mutex_init(&ms->qemu_file_lock);
3581 }
3582
3583 /*
3584 * Return true if check pass, false otherwise. Error will be put
3585 * inside errp if provided.
3586 */
3587 static bool migration_object_check(MigrationState *ms, Error **errp)
3588 {
3589 MigrationCapabilityStatusList *head = NULL;
3590 /* Assuming all off */
3591 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
3592 int i;
3593
3594 if (!migrate_params_check(&ms->parameters, errp)) {
3595 return false;
3596 }
3597
3598 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
3599 if (ms->enabled_capabilities[i]) {
3600 head = migrate_cap_add(head, i, true);
3601 }
3602 }
3603
3604 ret = migrate_caps_check(cap_list, head, errp);
3605
3606 /* It works with head == NULL */
3607 qapi_free_MigrationCapabilityStatusList(head);
3608
3609 return ret;
3610 }
3611
3612 static const TypeInfo migration_type = {
3613 .name = TYPE_MIGRATION,
3614 /*
3615 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3616 * not created using qdev_create(), it is not attached to the qdev
3617 * device tree, and it is never realized.
3618 *
3619 * TODO: Make this TYPE_OBJECT once QOM provides something like
3620 * TYPE_DEVICE's "-global" properties.
3621 */
3622 .parent = TYPE_DEVICE,
3623 .class_init = migration_class_init,
3624 .class_size = sizeof(MigrationClass),
3625 .instance_size = sizeof(MigrationState),
3626 .instance_init = migration_instance_init,
3627 .instance_finalize = migration_instance_finalize,
3628 };
3629
3630 static void register_migration_types(void)
3631 {
3632 type_register_static(&migration_type);
3633 }
3634
3635 type_init(register_migration_types);