?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
migration/savevm: release gslist after dump_vmstate_json
[qemu/kevin.git] / migration / migration.c
blobc1814a68616a4eb6f297da7a9d3135bd2e19c1e3
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 #include "multifd.h"
57
58 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
59
60 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
61 * data. */
62 #define BUFFER_DELAY 100
63 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
64
65 /* Time in milliseconds we are allowed to stop the source,
66 * for sending the last part */
67 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
68
69 /* Maximum migrate downtime set to 2000 seconds */
70 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
71 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
72
73 /* Default compression thread count */
74 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
75 /* Default decompression thread count, usually decompression is at
76 * least 4 times as fast as compression.*/
77 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
78 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
79 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
80 /* Define default autoconverge cpu throttle migration parameters */
81 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
82 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
83 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
84
85 /* Migration XBZRLE default cache size */
86 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
87
88 /* The delay time (in ms) between two COLO checkpoints */
89 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
90 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
91 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
92 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
93 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
94 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
95 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
96
97 /* Background transfer rate for postcopy, 0 means unlimited, note
98 * that page requests can still exceed this limit.
99 */
100 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
101
102 /*
103 * Parameters for self_announce_delay giving a stream of RARP/ARP
104 * packets after migration.
105 */
106 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
107 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
108 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
109 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
110
111 static NotifierList migration_state_notifiers =
112 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
113
114 static bool deferred_incoming;
115
116 /* Messages sent on the return path from destination to source */
117 enum mig_rp_message_type {
118 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
119 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
120 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
121
122 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
123 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
124 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
125 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
126
127 MIG_RP_MSG_MAX
128 };
129
130 /* When we add fault tolerance, we could have several
131 migrations at once. For now we don't need to add
132 dynamic creation of migration */
133
134 static MigrationState *current_migration;
135 static MigrationIncomingState *current_incoming;
136
137 static bool migration_object_check(MigrationState *ms, Error **errp);
138 static int migration_maybe_pause(MigrationState *s,
139 int *current_active_state,
140 int new_state);
141 static void migrate_fd_cancel(MigrationState *s);
142
143 void migration_object_init(void)
144 {
145 MachineState *ms = MACHINE(qdev_get_machine());
146 Error *err = NULL;
147
148 /* This can only be called once. */
149 assert(!current_migration);
150 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
151
152 /*
153 * Init the migrate incoming object as well no matter whether
154 * we'll use it or not.
155 */
156 assert(!current_incoming);
157 current_incoming = g_new0(MigrationIncomingState, 1);
158 current_incoming->state = MIGRATION_STATUS_NONE;
159 current_incoming->postcopy_remote_fds =
160 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
161 qemu_mutex_init(&current_incoming->rp_mutex);
162 qemu_event_init(&current_incoming->main_thread_load_event, false);
163 qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
164 qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
165
166 init_dirty_bitmap_incoming_migration();
167
168 if (!migration_object_check(current_migration, &err)) {
169 error_report_err(err);
170 exit(1);
171 }
172
173 /*
174 * We cannot really do this in migration_instance_init() since at
175 * that time global properties are not yet applied, then this
176 * value will be definitely replaced by something else.
177 */
178 if (ms->enforce_config_section) {
179 current_migration->send_configuration = true;
180 }
181 }
182
183 void migration_shutdown(void)
184 {
185 /*
186 * Cancel the current migration - that will (eventually)
187 * stop the migration using this structure
188 */
189 migrate_fd_cancel(current_migration);
190 object_unref(OBJECT(current_migration));
191 }
192
193 /* For outgoing */
194 MigrationState *migrate_get_current(void)
195 {
196 /* This can only be called after the object created. */
197 assert(current_migration);
198 return current_migration;
199 }
200
201 MigrationIncomingState *migration_incoming_get_current(void)
202 {
203 assert(current_incoming);
204 return current_incoming;
205 }
206
207 void migration_incoming_state_destroy(void)
208 {
209 struct MigrationIncomingState *mis = migration_incoming_get_current();
210
211 if (mis->to_src_file) {
212 /* Tell source that we are done */
213 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
214 qemu_fclose(mis->to_src_file);
215 mis->to_src_file = NULL;
216 }
217
218 if (mis->from_src_file) {
219 qemu_fclose(mis->from_src_file);
220 mis->from_src_file = NULL;
221 }
222 if (mis->postcopy_remote_fds) {
223 g_array_free(mis->postcopy_remote_fds, TRUE);
224 mis->postcopy_remote_fds = NULL;
225 }
226
227 qemu_event_reset(&mis->main_thread_load_event);
228
229 if (mis->socket_address_list) {
230 qapi_free_SocketAddressList(mis->socket_address_list);
231 mis->socket_address_list = NULL;
232 }
233 }
234
235 static void migrate_generate_event(int new_state)
236 {
237 if (migrate_use_events()) {
238 qapi_event_send_migration(new_state);
239 }
240 }
241
242 static bool migrate_late_block_activate(void)
243 {
244 MigrationState *s;
245
246 s = migrate_get_current();
247
248 return s->enabled_capabilities[
249 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
250 }
251
252 /*
253 * Called on -incoming with a defer: uri.
254 * The migration can be started later after any parameters have been
255 * changed.
256 */
257 static void deferred_incoming_migration(Error **errp)
258 {
259 if (deferred_incoming) {
260 error_setg(errp, "Incoming migration already deferred");
261 }
262 deferred_incoming = true;
263 }
264
265 /*
266 * Send a message on the return channel back to the source
267 * of the migration.
268 */
269 static int migrate_send_rp_message(MigrationIncomingState *mis,
270 enum mig_rp_message_type message_type,
271 uint16_t len, void *data)
272 {
273 int ret = 0;
274
275 trace_migrate_send_rp_message((int)message_type, len);
276 qemu_mutex_lock(&mis->rp_mutex);
277
278 /*
279 * It's possible that the file handle got lost due to network
280 * failures.
281 */
282 if (!mis->to_src_file) {
283 ret = -EIO;
284 goto error;
285 }
286
287 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
288 qemu_put_be16(mis->to_src_file, len);
289 qemu_put_buffer(mis->to_src_file, data, len);
290 qemu_fflush(mis->to_src_file);
291
292 /* It's possible that qemu file got error during sending */
293 ret = qemu_file_get_error(mis->to_src_file);
294
295 error:
296 qemu_mutex_unlock(&mis->rp_mutex);
297 return ret;
298 }
299
300 /* Request a range of pages from the source VM at the given
301 * start address.
302 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
303 * as the last request (a name must have been given previously)
304 * Start: Address offset within the RB
305 * Len: Length in bytes required - must be a multiple of pagesize
306 */
307 int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
308 ram_addr_t start, size_t len)
309 {
310 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
311 size_t msglen = 12; /* start + len */
312 enum mig_rp_message_type msg_type;
313
314 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
315 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
316
317 if (rbname) {
318 int rbname_len = strlen(rbname);
319 assert(rbname_len < 256);
320
321 bufc[msglen++] = rbname_len;
322 memcpy(bufc + msglen, rbname, rbname_len);
323 msglen += rbname_len;
324 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
325 } else {
326 msg_type = MIG_RP_MSG_REQ_PAGES;
327 }
328
329 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
330 }
331
332 static bool migration_colo_enabled;
333 bool migration_incoming_colo_enabled(void)
334 {
335 return migration_colo_enabled;
336 }
337
338 void migration_incoming_disable_colo(void)
339 {
340 migration_colo_enabled = false;
341 }
342
343 void migration_incoming_enable_colo(void)
344 {
345 migration_colo_enabled = true;
346 }
347
348 void migrate_add_address(SocketAddress *address)
349 {
350 MigrationIncomingState *mis = migration_incoming_get_current();
351 SocketAddressList *addrs;
352
353 addrs = g_new0(SocketAddressList, 1);
354 addrs->next = mis->socket_address_list;
355 mis->socket_address_list = addrs;
356 addrs->value = QAPI_CLONE(SocketAddress, address);
357 }
358
359 void qemu_start_incoming_migration(const char *uri, Error **errp)
360 {
361 const char *p;
362
363 qapi_event_send_migration(MIGRATION_STATUS_SETUP);
364 if (!strcmp(uri, "defer")) {
365 deferred_incoming_migration(errp);
366 } else if (strstart(uri, "tcp:", &p)) {
367 tcp_start_incoming_migration(p, errp);
368 #ifdef CONFIG_RDMA
369 } else if (strstart(uri, "rdma:", &p)) {
370 rdma_start_incoming_migration(p, errp);
371 #endif
372 } else if (strstart(uri, "exec:", &p)) {
373 exec_start_incoming_migration(p, errp);
374 } else if (strstart(uri, "unix:", &p)) {
375 unix_start_incoming_migration(p, errp);
376 } else if (strstart(uri, "fd:", &p)) {
377 fd_start_incoming_migration(p, errp);
378 } else {
379 error_setg(errp, "unknown migration protocol: %s", uri);
380 }
381 }
382
383 static void process_incoming_migration_bh(void *opaque)
384 {
385 Error *local_err = NULL;
386 MigrationIncomingState *mis = opaque;
387
388 /* If capability late_block_activate is set:
389 * Only fire up the block code now if we're going to restart the
390 * VM, else 'cont' will do it.
391 * This causes file locking to happen; so we don't want it to happen
392 * unless we really are starting the VM.
393 */
394 if (!migrate_late_block_activate() ||
395 (autostart && (!global_state_received() ||
396 global_state_get_runstate() == RUN_STATE_RUNNING))) {
397 /* Make sure all file formats flush their mutable metadata.
398 * If we get an error here, just don't restart the VM yet. */
399 bdrv_invalidate_cache_all(&local_err);
400 if (local_err) {
401 error_report_err(local_err);
402 local_err = NULL;
403 autostart = false;
404 }
405 }
406
407 /*
408 * This must happen after all error conditions are dealt with and
409 * we're sure the VM is going to be running on this host.
410 */
411 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
412
413 if (multifd_load_cleanup(&local_err) != 0) {
414 error_report_err(local_err);
415 autostart = false;
416 }
417 /* If global state section was not received or we are in running
418 state, we need to obey autostart. Any other state is set with
419 runstate_set. */
420
421 dirty_bitmap_mig_before_vm_start();
422
423 if (!global_state_received() ||
424 global_state_get_runstate() == RUN_STATE_RUNNING) {
425 if (autostart) {
426 vm_start();
427 } else {
428 runstate_set(RUN_STATE_PAUSED);
429 }
430 } else if (migration_incoming_colo_enabled()) {
431 migration_incoming_disable_colo();
432 vm_start();
433 } else {
434 runstate_set(global_state_get_runstate());
435 }
436 /*
437 * This must happen after any state changes since as soon as an external
438 * observer sees this event they might start to prod at the VM assuming
439 * it's ready to use.
440 */
441 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
442 MIGRATION_STATUS_COMPLETED);
443 qemu_bh_delete(mis->bh);
444 migration_incoming_state_destroy();
445 }
446
447 static void process_incoming_migration_co(void *opaque)
448 {
449 MigrationIncomingState *mis = migration_incoming_get_current();
450 PostcopyState ps;
451 int ret;
452 Error *local_err = NULL;
453
454 assert(mis->from_src_file);
455 mis->migration_incoming_co = qemu_coroutine_self();
456 mis->largest_page_size = qemu_ram_pagesize_largest();
457 postcopy_state_set(POSTCOPY_INCOMING_NONE);
458 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
459 MIGRATION_STATUS_ACTIVE);
460 ret = qemu_loadvm_state(mis->from_src_file);
461
462 ps = postcopy_state_get();
463 trace_process_incoming_migration_co_end(ret, ps);
464 if (ps != POSTCOPY_INCOMING_NONE) {
465 if (ps == POSTCOPY_INCOMING_ADVISE) {
466 /*
467 * Where a migration had postcopy enabled (and thus went to advise)
468 * but managed to complete within the precopy period, we can use
469 * the normal exit.
470 */
471 postcopy_ram_incoming_cleanup(mis);
472 } else if (ret >= 0) {
473 /*
474 * Postcopy was started, cleanup should happen at the end of the
475 * postcopy thread.
476 */
477 trace_process_incoming_migration_co_postcopy_end_main();
478 return;
479 }
480 /* Else if something went wrong then just fall out of the normal exit */
481 }
482
483 /* we get COLO info, and know if we are in COLO mode */
484 if (!ret && migration_incoming_colo_enabled()) {
485 /* Make sure all file formats flush their mutable metadata */
486 bdrv_invalidate_cache_all(&local_err);
487 if (local_err) {
488 error_report_err(local_err);
489 goto fail;
490 }
491
492 if (colo_init_ram_cache() < 0) {
493 error_report("Init ram cache failed");
494 goto fail;
495 }
496
497 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
498 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
499 mis->have_colo_incoming_thread = true;
500 qemu_coroutine_yield();
501
502 /* Wait checkpoint incoming thread exit before free resource */
503 qemu_thread_join(&mis->colo_incoming_thread);
504 /* We hold the global iothread lock, so it is safe here */
505 colo_release_ram_cache();
506 }
507
508 if (ret < 0) {
509 error_report("load of migration failed: %s", strerror(-ret));
510 goto fail;
511 }
512 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
513 qemu_bh_schedule(mis->bh);
514 mis->migration_incoming_co = NULL;
515 return;
516 fail:
517 local_err = NULL;
518 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
519 MIGRATION_STATUS_FAILED);
520 qemu_fclose(mis->from_src_file);
521 if (multifd_load_cleanup(&local_err) != 0) {
522 error_report_err(local_err);
523 }
524 exit(EXIT_FAILURE);
525 }
526
527 /**
528 * @migration_incoming_setup: Setup incoming migration
529 *
530 * Returns 0 for no error or 1 for error
531 *
532 * @f: file for main migration channel
533 * @errp: where to put errors
534 */
535 static int migration_incoming_setup(QEMUFile *f, Error **errp)
536 {
537 MigrationIncomingState *mis = migration_incoming_get_current();
538 Error *local_err = NULL;
539
540 if (multifd_load_setup(&local_err) != 0) {
541 /* We haven't been able to create multifd threads
542 nothing better to do */
543 error_report_err(local_err);
544 exit(EXIT_FAILURE);
545 }
546
547 if (!mis->from_src_file) {
548 mis->from_src_file = f;
549 }
550 qemu_file_set_blocking(f, false);
551 return 0;
552 }
553
554 void migration_incoming_process(void)
555 {
556 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
557 qemu_coroutine_enter(co);
558 }
559
560 /* Returns true if recovered from a paused migration, otherwise false */
561 static bool postcopy_try_recover(QEMUFile *f)
562 {
563 MigrationIncomingState *mis = migration_incoming_get_current();
564
565 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
566 /* Resumed from a paused postcopy migration */
567
568 mis->from_src_file = f;
569 /* Postcopy has standalone thread to do vm load */
570 qemu_file_set_blocking(f, true);
571
572 /* Re-configure the return path */
573 mis->to_src_file = qemu_file_get_return_path(f);
574
575 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
576 MIGRATION_STATUS_POSTCOPY_RECOVER);
577
578 /*
579 * Here, we only wake up the main loading thread (while the
580 * fault thread will still be waiting), so that we can receive
581 * commands from source now, and answer it if needed. The
582 * fault thread will be woken up afterwards until we are sure
583 * that source is ready to reply to page requests.
584 */
585 qemu_sem_post(&mis->postcopy_pause_sem_dst);
586 return true;
587 }
588
589 return false;
590 }
591
592 void migration_fd_process_incoming(QEMUFile *f, Error **errp)
593 {
594 Error *local_err = NULL;
595
596 if (postcopy_try_recover(f)) {
597 return;
598 }
599
600 if (migration_incoming_setup(f, &local_err)) {
601 if (local_err) {
602 error_propagate(errp, local_err);
603 }
604 return;
605 }
606 migration_incoming_process();
607 }
608
609 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
610 {
611 MigrationIncomingState *mis = migration_incoming_get_current();
612 Error *local_err = NULL;
613 bool start_migration;
614
615 if (!mis->from_src_file) {
616 /* The first connection (multifd may have multiple) */
617 QEMUFile *f = qemu_fopen_channel_input(ioc);
618
619 /* If it's a recovery, we're done */
620 if (postcopy_try_recover(f)) {
621 return;
622 }
623
624 if (migration_incoming_setup(f, &local_err)) {
625 if (local_err) {
626 error_propagate(errp, local_err);
627 }
628 return;
629 }
630
631 /*
632 * Common migration only needs one channel, so we can start
633 * right now. Multifd needs more than one channel, we wait.
634 */
635 start_migration = !migrate_use_multifd();
636 } else {
637 /* Multiple connections */
638 assert(migrate_use_multifd());
639 start_migration = multifd_recv_new_channel(ioc, &local_err);
640 if (local_err) {
641 error_propagate(errp, local_err);
642 return;
643 }
644 }
645
646 if (start_migration) {
647 migration_incoming_process();
648 }
649 }
650
651 /**
652 * @migration_has_all_channels: We have received all channels that we need
653 *
654 * Returns true when we have got connections to all the channels that
655 * we need for migration.
656 */
657 bool migration_has_all_channels(void)
658 {
659 MigrationIncomingState *mis = migration_incoming_get_current();
660 bool all_channels;
661
662 all_channels = multifd_recv_all_channels_created();
663
664 return all_channels && mis->from_src_file != NULL;
665 }
666
667 /*
668 * Send a 'SHUT' message on the return channel with the given value
669 * to indicate that we've finished with the RP. Non-0 value indicates
670 * error.
671 */
672 void migrate_send_rp_shut(MigrationIncomingState *mis,
673 uint32_t value)
674 {
675 uint32_t buf;
676
677 buf = cpu_to_be32(value);
678 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
679 }
680
681 /*
682 * Send a 'PONG' message on the return channel with the given value
683 * (normally in response to a 'PING')
684 */
685 void migrate_send_rp_pong(MigrationIncomingState *mis,
686 uint32_t value)
687 {
688 uint32_t buf;
689
690 buf = cpu_to_be32(value);
691 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
692 }
693
694 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
695 char *block_name)
696 {
697 char buf[512];
698 int len;
699 int64_t res;
700
701 /*
702 * First, we send the header part. It contains only the len of
703 * idstr, and the idstr itself.
704 */
705 len = strlen(block_name);
706 buf[0] = len;
707 memcpy(buf + 1, block_name, len);
708
709 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
710 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
711 __func__);
712 return;
713 }
714
715 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
716
717 /*
718 * Next, we dump the received bitmap to the stream.
719 *
720 * TODO: currently we are safe since we are the only one that is
721 * using the to_src_file handle (fault thread is still paused),
722 * and it's ok even not taking the mutex. However the best way is
723 * to take the lock before sending the message header, and release
724 * the lock after sending the bitmap.
725 */
726 qemu_mutex_lock(&mis->rp_mutex);
727 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
728 qemu_mutex_unlock(&mis->rp_mutex);
729
730 trace_migrate_send_rp_recv_bitmap(block_name, res);
731 }
732
733 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
734 {
735 uint32_t buf;
736
737 buf = cpu_to_be32(value);
738 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
739 }
740
741 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
742 {
743 MigrationCapabilityStatusList *head = NULL;
744 MigrationCapabilityStatusList *caps;
745 MigrationState *s = migrate_get_current();
746 int i;
747
748 caps = NULL; /* silence compiler warning */
749 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
750 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
751 if (i == MIGRATION_CAPABILITY_BLOCK) {
752 continue;
753 }
754 #endif
755 if (head == NULL) {
756 head = g_malloc0(sizeof(*caps));
757 caps = head;
758 } else {
759 caps->next = g_malloc0(sizeof(*caps));
760 caps = caps->next;
761 }
762 caps->value =
763 g_malloc(sizeof(*caps->value));
764 caps->value->capability = i;
765 caps->value->state = s->enabled_capabilities[i];
766 }
767
768 return head;
769 }
770
771 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
772 {
773 MigrationParameters *params;
774 MigrationState *s = migrate_get_current();
775
776 /* TODO use QAPI_CLONE() instead of duplicating it inline */
777 params = g_malloc0(sizeof(*params));
778 params->has_compress_level = true;
779 params->compress_level = s->parameters.compress_level;
780 params->has_compress_threads = true;
781 params->compress_threads = s->parameters.compress_threads;
782 params->has_compress_wait_thread = true;
783 params->compress_wait_thread = s->parameters.compress_wait_thread;
784 params->has_decompress_threads = true;
785 params->decompress_threads = s->parameters.decompress_threads;
786 params->has_cpu_throttle_initial = true;
787 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
788 params->has_cpu_throttle_increment = true;
789 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
790 params->has_tls_creds = true;
791 params->tls_creds = g_strdup(s->parameters.tls_creds);
792 params->has_tls_hostname = true;
793 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
794 params->has_tls_authz = true;
795 params->tls_authz = g_strdup(s->parameters.tls_authz);
796 params->has_max_bandwidth = true;
797 params->max_bandwidth = s->parameters.max_bandwidth;
798 params->has_downtime_limit = true;
799 params->downtime_limit = s->parameters.downtime_limit;
800 params->has_x_checkpoint_delay = true;
801 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
802 params->has_block_incremental = true;
803 params->block_incremental = s->parameters.block_incremental;
804 params->has_multifd_channels = true;
805 params->multifd_channels = s->parameters.multifd_channels;
806 params->has_multifd_compression = true;
807 params->multifd_compression = s->parameters.multifd_compression;
808 params->has_multifd_zlib_level = true;
809 params->multifd_zlib_level = s->parameters.multifd_zlib_level;
810 params->has_multifd_zstd_level = true;
811 params->multifd_zstd_level = s->parameters.multifd_zstd_level;
812 params->has_xbzrle_cache_size = true;
813 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
814 params->has_max_postcopy_bandwidth = true;
815 params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
816 params->has_max_cpu_throttle = true;
817 params->max_cpu_throttle = s->parameters.max_cpu_throttle;
818 params->has_announce_initial = true;
819 params->announce_initial = s->parameters.announce_initial;
820 params->has_announce_max = true;
821 params->announce_max = s->parameters.announce_max;
822 params->has_announce_rounds = true;
823 params->announce_rounds = s->parameters.announce_rounds;
824 params->has_announce_step = true;
825 params->announce_step = s->parameters.announce_step;
826
827 return params;
828 }
829
830 AnnounceParameters *migrate_announce_params(void)
831 {
832 static AnnounceParameters ap;
833
834 MigrationState *s = migrate_get_current();
835
836 ap.initial = s->parameters.announce_initial;
837 ap.max = s->parameters.announce_max;
838 ap.rounds = s->parameters.announce_rounds;
839 ap.step = s->parameters.announce_step;
840
841 return &ap;
842 }
843
844 /*
845 * Return true if we're already in the middle of a migration
846 * (i.e. any of the active or setup states)
847 */
848 bool migration_is_setup_or_active(int state)
849 {
850 switch (state) {
851 case MIGRATION_STATUS_ACTIVE:
852 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
853 case MIGRATION_STATUS_POSTCOPY_PAUSED:
854 case MIGRATION_STATUS_POSTCOPY_RECOVER:
855 case MIGRATION_STATUS_SETUP:
856 case MIGRATION_STATUS_PRE_SWITCHOVER:
857 case MIGRATION_STATUS_DEVICE:
858 case MIGRATION_STATUS_WAIT_UNPLUG:
859 return true;
860
861 default:
862 return false;
863
864 }
865 }
866
867 bool migration_is_running(int state)
868 {
869 switch (state) {
870 case MIGRATION_STATUS_ACTIVE:
871 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
872 case MIGRATION_STATUS_POSTCOPY_PAUSED:
873 case MIGRATION_STATUS_POSTCOPY_RECOVER:
874 case MIGRATION_STATUS_SETUP:
875 case MIGRATION_STATUS_PRE_SWITCHOVER:
876 case MIGRATION_STATUS_DEVICE:
877 case MIGRATION_STATUS_WAIT_UNPLUG:
878 case MIGRATION_STATUS_CANCELLING:
879 case MIGRATION_STATUS_COLO:
880 return true;
881
882 default:
883 return false;
884
885 }
886 }
887
888 static void populate_time_info(MigrationInfo *info, MigrationState *s)
889 {
890 info->has_status = true;
891 info->has_setup_time = true;
892 info->setup_time = s->setup_time;
893 if (s->state == MIGRATION_STATUS_COMPLETED) {
894 info->has_total_time = true;
895 info->total_time = s->total_time;
896 info->has_downtime = true;
897 info->downtime = s->downtime;
898 } else {
899 info->has_total_time = true;
900 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
901 s->start_time;
902 info->has_expected_downtime = true;
903 info->expected_downtime = s->expected_downtime;
904 }
905 }
906
907 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
908 {
909 info->has_ram = true;
910 info->ram = g_malloc0(sizeof(*info->ram));
911 info->ram->transferred = ram_counters.transferred;
912 info->ram->total = ram_bytes_total();
913 info->ram->duplicate = ram_counters.duplicate;
914 /* legacy value. It is not used anymore */
915 info->ram->skipped = 0;
916 info->ram->normal = ram_counters.normal;
917 info->ram->normal_bytes = ram_counters.normal *
918 qemu_target_page_size();
919 info->ram->mbps = s->mbps;
920 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
921 info->ram->postcopy_requests = ram_counters.postcopy_requests;
922 info->ram->page_size = qemu_target_page_size();
923 info->ram->multifd_bytes = ram_counters.multifd_bytes;
924 info->ram->pages_per_second = s->pages_per_second;
925
926 if (migrate_use_xbzrle()) {
927 info->has_xbzrle_cache = true;
928 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
929 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
930 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
931 info->xbzrle_cache->pages = xbzrle_counters.pages;
932 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
933 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
934 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
935 }
936
937 if (migrate_use_compression()) {
938 info->has_compression = true;
939 info->compression = g_malloc0(sizeof(*info->compression));
940 info->compression->pages = compression_counters.pages;
941 info->compression->busy = compression_counters.busy;
942 info->compression->busy_rate = compression_counters.busy_rate;
943 info->compression->compressed_size =
944 compression_counters.compressed_size;
945 info->compression->compression_rate =
946 compression_counters.compression_rate;
947 }
948
949 if (cpu_throttle_active()) {
950 info->has_cpu_throttle_percentage = true;
951 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
952 }
953
954 if (s->state != MIGRATION_STATUS_COMPLETED) {
955 info->ram->remaining = ram_bytes_remaining();
956 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
957 }
958 }
959
960 static void populate_disk_info(MigrationInfo *info)
961 {
962 if (blk_mig_active()) {
963 info->has_disk = true;
964 info->disk = g_malloc0(sizeof(*info->disk));
965 info->disk->transferred = blk_mig_bytes_transferred();
966 info->disk->remaining = blk_mig_bytes_remaining();
967 info->disk->total = blk_mig_bytes_total();
968 }
969 }
970
971 static void fill_source_migration_info(MigrationInfo *info)
972 {
973 MigrationState *s = migrate_get_current();
974
975 switch (s->state) {
976 case MIGRATION_STATUS_NONE:
977 /* no migration has happened ever */
978 /* do not overwrite destination migration status */
979 return;
980 break;
981 case MIGRATION_STATUS_SETUP:
982 info->has_status = true;
983 info->has_total_time = false;
984 break;
985 case MIGRATION_STATUS_ACTIVE:
986 case MIGRATION_STATUS_CANCELLING:
987 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
988 case MIGRATION_STATUS_PRE_SWITCHOVER:
989 case MIGRATION_STATUS_DEVICE:
990 case MIGRATION_STATUS_POSTCOPY_PAUSED:
991 case MIGRATION_STATUS_POSTCOPY_RECOVER:
992 /* TODO add some postcopy stats */
993 populate_time_info(info, s);
994 populate_ram_info(info, s);
995 populate_disk_info(info);
996 break;
997 case MIGRATION_STATUS_COLO:
998 info->has_status = true;
999 /* TODO: display COLO specific information (checkpoint info etc.) */
1000 break;
1001 case MIGRATION_STATUS_COMPLETED:
1002 populate_time_info(info, s);
1003 populate_ram_info(info, s);
1004 break;
1005 case MIGRATION_STATUS_FAILED:
1006 info->has_status = true;
1007 if (s->error) {
1008 info->has_error_desc = true;
1009 info->error_desc = g_strdup(error_get_pretty(s->error));
1010 }
1011 break;
1012 case MIGRATION_STATUS_CANCELLED:
1013 info->has_status = true;
1014 break;
1015 case MIGRATION_STATUS_WAIT_UNPLUG:
1016 info->has_status = true;
1017 break;
1018 }
1019 info->status = s->state;
1020 }
1021
1022 /**
1023 * @migration_caps_check - check capability validity
1024 *
1025 * @cap_list: old capability list, array of bool
1026 * @params: new capabilities to be applied soon
1027 * @errp: set *errp if the check failed, with reason
1028 *
1029 * Returns true if check passed, otherwise false.
1030 */
1031 static bool migrate_caps_check(bool *cap_list,
1032 MigrationCapabilityStatusList *params,
1033 Error **errp)
1034 {
1035 MigrationCapabilityStatusList *cap;
1036 bool old_postcopy_cap;
1037 MigrationIncomingState *mis = migration_incoming_get_current();
1038
1039 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1040
1041 for (cap = params; cap; cap = cap->next) {
1042 cap_list[cap->value->capability] = cap->value->state;
1043 }
1044
1045 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1046 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
1047 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
1048 "block migration");
1049 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
1050 return false;
1051 }
1052 #endif
1053
1054 #ifndef CONFIG_REPLICATION
1055 if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
1056 error_setg(errp, "QEMU compiled without replication module"
1057 " can't enable COLO");
1058 error_append_hint(errp, "Please enable replication before COLO.\n");
1059 return false;
1060 }
1061 #endif
1062
1063 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
1064 /* This check is reasonably expensive, so only when it's being
1065 * set the first time, also it's only the destination that needs
1066 * special support.
1067 */
1068 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1069 !postcopy_ram_supported_by_host(mis)) {
1070 /* postcopy_ram_supported_by_host will have emitted a more
1071 * detailed message
1072 */
1073 error_setg(errp, "Postcopy is not supported");
1074 return false;
1075 }
1076
1077 if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1078 error_setg(errp, "Postcopy is not compatible with ignore-shared");
1079 return false;
1080 }
1081 }
1082
1083 return true;
1084 }
1085
1086 static void fill_destination_migration_info(MigrationInfo *info)
1087 {
1088 MigrationIncomingState *mis = migration_incoming_get_current();
1089
1090 if (mis->socket_address_list) {
1091 info->has_socket_address = true;
1092 info->socket_address =
1093 QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1094 }
1095
1096 switch (mis->state) {
1097 case MIGRATION_STATUS_NONE:
1098 return;
1099 break;
1100 case MIGRATION_STATUS_SETUP:
1101 case MIGRATION_STATUS_CANCELLING:
1102 case MIGRATION_STATUS_CANCELLED:
1103 case MIGRATION_STATUS_ACTIVE:
1104 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1105 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1106 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1107 case MIGRATION_STATUS_FAILED:
1108 case MIGRATION_STATUS_COLO:
1109 info->has_status = true;
1110 break;
1111 case MIGRATION_STATUS_COMPLETED:
1112 info->has_status = true;
1113 fill_destination_postcopy_migration_info(info);
1114 break;
1115 }
1116 info->status = mis->state;
1117 }
1118
1119 MigrationInfo *qmp_query_migrate(Error **errp)
1120 {
1121 MigrationInfo *info = g_malloc0(sizeof(*info));
1122
1123 fill_destination_migration_info(info);
1124 fill_source_migration_info(info);
1125
1126 return info;
1127 }
1128
1129 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1130 Error **errp)
1131 {
1132 MigrationState *s = migrate_get_current();
1133 MigrationCapabilityStatusList *cap;
1134 bool cap_list[MIGRATION_CAPABILITY__MAX];
1135
1136 if (migration_is_running(s->state)) {
1137 error_setg(errp, QERR_MIGRATION_ACTIVE);
1138 return;
1139 }
1140
1141 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1142 if (!migrate_caps_check(cap_list, params, errp)) {
1143 return;
1144 }
1145
1146 for (cap = params; cap; cap = cap->next) {
1147 s->enabled_capabilities[cap->value->capability] = cap->value->state;
1148 }
1149 }
1150
1151 /*
1152 * Check whether the parameters are valid. Error will be put into errp
1153 * (if provided). Return true if valid, otherwise false.
1154 */
1155 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1156 {
1157 if (params->has_compress_level &&
1158 (params->compress_level > 9)) {
1159 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1160 "is invalid, it should be in the range of 0 to 9");
1161 return false;
1162 }
1163
1164 if (params->has_compress_threads && (params->compress_threads < 1)) {
1165 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1166 "compress_threads",
1167 "is invalid, it should be in the range of 1 to 255");
1168 return false;
1169 }
1170
1171 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1172 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1173 "decompress_threads",
1174 "is invalid, it should be in the range of 1 to 255");
1175 return false;
1176 }
1177
1178 if (params->has_cpu_throttle_initial &&
1179 (params->cpu_throttle_initial < 1 ||
1180 params->cpu_throttle_initial > 99)) {
1181 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1182 "cpu_throttle_initial",
1183 "an integer in the range of 1 to 99");
1184 return false;
1185 }
1186
1187 if (params->has_cpu_throttle_increment &&
1188 (params->cpu_throttle_increment < 1 ||
1189 params->cpu_throttle_increment > 99)) {
1190 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1191 "cpu_throttle_increment",
1192 "an integer in the range of 1 to 99");
1193 return false;
1194 }
1195
1196 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1197 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
1198 " range of 0 to %zu bytes/second", SIZE_MAX);
1199 return false;
1200 }
1201
1202 if (params->has_downtime_limit &&
1203 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1204 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1205 "the range of 0 to %d milliseconds",
1206 MAX_MIGRATE_DOWNTIME);
1207 return false;
1208 }
1209
1210 /* x_checkpoint_delay is now always positive */
1211
1212 if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1213 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1214 "multifd_channels",
1215 "is invalid, it should be in the range of 1 to 255");
1216 return false;
1217 }
1218
1219 if (params->has_multifd_zlib_level &&
1220 (params->multifd_zlib_level > 9)) {
1221 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level",
1222 "is invalid, it should be in the range of 0 to 9");
1223 return false;
1224 }
1225
1226 if (params->has_multifd_zstd_level &&
1227 (params->multifd_zstd_level > 20)) {
1228 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level",
1229 "is invalid, it should be in the range of 0 to 20");
1230 return false;
1231 }
1232
1233 if (params->has_xbzrle_cache_size &&
1234 (params->xbzrle_cache_size < qemu_target_page_size() ||
1235 !is_power_of_2(params->xbzrle_cache_size))) {
1236 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1237 "xbzrle_cache_size",
1238 "is invalid, it should be bigger than target page size"
1239 " and a power of two");
1240 return false;
1241 }
1242
1243 if (params->has_max_cpu_throttle &&
1244 (params->max_cpu_throttle < params->cpu_throttle_initial ||
1245 params->max_cpu_throttle > 99)) {
1246 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1247 "max_cpu_throttle",
1248 "an integer in the range of cpu_throttle_initial to 99");
1249 return false;
1250 }
1251
1252 if (params->has_announce_initial &&
1253 params->announce_initial > 100000) {
1254 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1255 "announce_initial",
1256 "is invalid, it must be less than 100000 ms");
1257 return false;
1258 }
1259 if (params->has_announce_max &&
1260 params->announce_max > 100000) {
1261 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1262 "announce_max",
1263 "is invalid, it must be less than 100000 ms");
1264 return false;
1265 }
1266 if (params->has_announce_rounds &&
1267 params->announce_rounds > 1000) {
1268 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1269 "announce_rounds",
1270 "is invalid, it must be in the range of 0 to 1000");
1271 return false;
1272 }
1273 if (params->has_announce_step &&
1274 (params->announce_step < 1 ||
1275 params->announce_step > 10000)) {
1276 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1277 "announce_step",
1278 "is invalid, it must be in the range of 1 to 10000 ms");
1279 return false;
1280 }
1281 return true;
1282 }
1283
1284 static void migrate_params_test_apply(MigrateSetParameters *params,
1285 MigrationParameters *dest)
1286 {
1287 *dest = migrate_get_current()->parameters;
1288
1289 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1290
1291 if (params->has_compress_level) {
1292 dest->compress_level = params->compress_level;
1293 }
1294
1295 if (params->has_compress_threads) {
1296 dest->compress_threads = params->compress_threads;
1297 }
1298
1299 if (params->has_compress_wait_thread) {
1300 dest->compress_wait_thread = params->compress_wait_thread;
1301 }
1302
1303 if (params->has_decompress_threads) {
1304 dest->decompress_threads = params->decompress_threads;
1305 }
1306
1307 if (params->has_cpu_throttle_initial) {
1308 dest->cpu_throttle_initial = params->cpu_throttle_initial;
1309 }
1310
1311 if (params->has_cpu_throttle_increment) {
1312 dest->cpu_throttle_increment = params->cpu_throttle_increment;
1313 }
1314
1315 if (params->has_tls_creds) {
1316 assert(params->tls_creds->type == QTYPE_QSTRING);
1317 dest->tls_creds = g_strdup(params->tls_creds->u.s);
1318 }
1319
1320 if (params->has_tls_hostname) {
1321 assert(params->tls_hostname->type == QTYPE_QSTRING);
1322 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1323 }
1324
1325 if (params->has_max_bandwidth) {
1326 dest->max_bandwidth = params->max_bandwidth;
1327 }
1328
1329 if (params->has_downtime_limit) {
1330 dest->downtime_limit = params->downtime_limit;
1331 }
1332
1333 if (params->has_x_checkpoint_delay) {
1334 dest->x_checkpoint_delay = params->x_checkpoint_delay;
1335 }
1336
1337 if (params->has_block_incremental) {
1338 dest->block_incremental = params->block_incremental;
1339 }
1340 if (params->has_multifd_channels) {
1341 dest->multifd_channels = params->multifd_channels;
1342 }
1343 if (params->has_multifd_compression) {
1344 dest->multifd_compression = params->multifd_compression;
1345 }
1346 if (params->has_xbzrle_cache_size) {
1347 dest->xbzrle_cache_size = params->xbzrle_cache_size;
1348 }
1349 if (params->has_max_postcopy_bandwidth) {
1350 dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1351 }
1352 if (params->has_max_cpu_throttle) {
1353 dest->max_cpu_throttle = params->max_cpu_throttle;
1354 }
1355 if (params->has_announce_initial) {
1356 dest->announce_initial = params->announce_initial;
1357 }
1358 if (params->has_announce_max) {
1359 dest->announce_max = params->announce_max;
1360 }
1361 if (params->has_announce_rounds) {
1362 dest->announce_rounds = params->announce_rounds;
1363 }
1364 if (params->has_announce_step) {
1365 dest->announce_step = params->announce_step;
1366 }
1367 }
1368
1369 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1370 {
1371 MigrationState *s = migrate_get_current();
1372
1373 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1374
1375 if (params->has_compress_level) {
1376 s->parameters.compress_level = params->compress_level;
1377 }
1378
1379 if (params->has_compress_threads) {
1380 s->parameters.compress_threads = params->compress_threads;
1381 }
1382
1383 if (params->has_compress_wait_thread) {
1384 s->parameters.compress_wait_thread = params->compress_wait_thread;
1385 }
1386
1387 if (params->has_decompress_threads) {
1388 s->parameters.decompress_threads = params->decompress_threads;
1389 }
1390
1391 if (params->has_cpu_throttle_initial) {
1392 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1393 }
1394
1395 if (params->has_cpu_throttle_increment) {
1396 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1397 }
1398
1399 if (params->has_tls_creds) {
1400 g_free(s->parameters.tls_creds);
1401 assert(params->tls_creds->type == QTYPE_QSTRING);
1402 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1403 }
1404
1405 if (params->has_tls_hostname) {
1406 g_free(s->parameters.tls_hostname);
1407 assert(params->tls_hostname->type == QTYPE_QSTRING);
1408 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1409 }
1410
1411 if (params->has_tls_authz) {
1412 g_free(s->parameters.tls_authz);
1413 assert(params->tls_authz->type == QTYPE_QSTRING);
1414 s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1415 }
1416
1417 if (params->has_max_bandwidth) {
1418 s->parameters.max_bandwidth = params->max_bandwidth;
1419 if (s->to_dst_file && !migration_in_postcopy()) {
1420 qemu_file_set_rate_limit(s->to_dst_file,
1421 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1422 }
1423 }
1424
1425 if (params->has_downtime_limit) {
1426 s->parameters.downtime_limit = params->downtime_limit;
1427 }
1428
1429 if (params->has_x_checkpoint_delay) {
1430 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1431 if (migration_in_colo_state()) {
1432 colo_checkpoint_notify(s);
1433 }
1434 }
1435
1436 if (params->has_block_incremental) {
1437 s->parameters.block_incremental = params->block_incremental;
1438 }
1439 if (params->has_multifd_channels) {
1440 s->parameters.multifd_channels = params->multifd_channels;
1441 }
1442 if (params->has_multifd_compression) {
1443 s->parameters.multifd_compression = params->multifd_compression;
1444 }
1445 if (params->has_xbzrle_cache_size) {
1446 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1447 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1448 }
1449 if (params->has_max_postcopy_bandwidth) {
1450 s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1451 if (s->to_dst_file && migration_in_postcopy()) {
1452 qemu_file_set_rate_limit(s->to_dst_file,
1453 s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1454 }
1455 }
1456 if (params->has_max_cpu_throttle) {
1457 s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1458 }
1459 if (params->has_announce_initial) {
1460 s->parameters.announce_initial = params->announce_initial;
1461 }
1462 if (params->has_announce_max) {
1463 s->parameters.announce_max = params->announce_max;
1464 }
1465 if (params->has_announce_rounds) {
1466 s->parameters.announce_rounds = params->announce_rounds;
1467 }
1468 if (params->has_announce_step) {
1469 s->parameters.announce_step = params->announce_step;
1470 }
1471 }
1472
1473 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1474 {
1475 MigrationParameters tmp;
1476
1477 /* TODO Rewrite "" to null instead */
1478 if (params->has_tls_creds
1479 && params->tls_creds->type == QTYPE_QNULL) {
1480 qobject_unref(params->tls_creds->u.n);
1481 params->tls_creds->type = QTYPE_QSTRING;
1482 params->tls_creds->u.s = strdup("");
1483 }
1484 /* TODO Rewrite "" to null instead */
1485 if (params->has_tls_hostname
1486 && params->tls_hostname->type == QTYPE_QNULL) {
1487 qobject_unref(params->tls_hostname->u.n);
1488 params->tls_hostname->type = QTYPE_QSTRING;
1489 params->tls_hostname->u.s = strdup("");
1490 }
1491
1492 migrate_params_test_apply(params, &tmp);
1493
1494 if (!migrate_params_check(&tmp, errp)) {
1495 /* Invalid parameter */
1496 return;
1497 }
1498
1499 migrate_params_apply(params, errp);
1500 }
1501
1502
1503 void qmp_migrate_start_postcopy(Error **errp)
1504 {
1505 MigrationState *s = migrate_get_current();
1506
1507 if (!migrate_postcopy()) {
1508 error_setg(errp, "Enable postcopy with migrate_set_capability before"
1509 " the start of migration");
1510 return;
1511 }
1512
1513 if (s->state == MIGRATION_STATUS_NONE) {
1514 error_setg(errp, "Postcopy must be started after migration has been"
1515 " started");
1516 return;
1517 }
1518 /*
1519 * we don't error if migration has finished since that would be racy
1520 * with issuing this command.
1521 */
1522 atomic_set(&s->start_postcopy, true);
1523 }
1524
1525 /* shared migration helpers */
1526
1527 void migrate_set_state(int *state, int old_state, int new_state)
1528 {
1529 assert(new_state < MIGRATION_STATUS__MAX);
1530 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1531 trace_migrate_set_state(MigrationStatus_str(new_state));
1532 migrate_generate_event(new_state);
1533 }
1534 }
1535
1536 static MigrationCapabilityStatusList *migrate_cap_add(
1537 MigrationCapabilityStatusList *list,
1538 MigrationCapability index,
1539 bool state)
1540 {
1541 MigrationCapabilityStatusList *cap;
1542
1543 cap = g_new0(MigrationCapabilityStatusList, 1);
1544 cap->value = g_new0(MigrationCapabilityStatus, 1);
1545 cap->value->capability = index;
1546 cap->value->state = state;
1547 cap->next = list;
1548
1549 return cap;
1550 }
1551
1552 void migrate_set_block_enabled(bool value, Error **errp)
1553 {
1554 MigrationCapabilityStatusList *cap;
1555
1556 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1557 qmp_migrate_set_capabilities(cap, errp);
1558 qapi_free_MigrationCapabilityStatusList(cap);
1559 }
1560
1561 static void migrate_set_block_incremental(MigrationState *s, bool value)
1562 {
1563 s->parameters.block_incremental = value;
1564 }
1565
1566 static void block_cleanup_parameters(MigrationState *s)
1567 {
1568 if (s->must_remove_block_options) {
1569 /* setting to false can never fail */
1570 migrate_set_block_enabled(false, &error_abort);
1571 migrate_set_block_incremental(s, false);
1572 s->must_remove_block_options = false;
1573 }
1574 }
1575
1576 static void migrate_fd_cleanup(MigrationState *s)
1577 {
1578 qemu_bh_delete(s->cleanup_bh);
1579 s->cleanup_bh = NULL;
1580
1581 qemu_savevm_state_cleanup();
1582
1583 if (s->to_dst_file) {
1584 QEMUFile *tmp;
1585
1586 trace_migrate_fd_cleanup();
1587 qemu_mutex_unlock_iothread();
1588 if (s->migration_thread_running) {
1589 qemu_thread_join(&s->thread);
1590 s->migration_thread_running = false;
1591 }
1592 qemu_mutex_lock_iothread();
1593
1594 multifd_save_cleanup();
1595 qemu_mutex_lock(&s->qemu_file_lock);
1596 tmp = s->to_dst_file;
1597 s->to_dst_file = NULL;
1598 qemu_mutex_unlock(&s->qemu_file_lock);
1599 /*
1600 * Close the file handle without the lock to make sure the
1601 * critical section won't block for long.
1602 */
1603 qemu_fclose(tmp);
1604 }
1605
1606 assert(!migration_is_active(s));
1607
1608 if (s->state == MIGRATION_STATUS_CANCELLING) {
1609 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1610 MIGRATION_STATUS_CANCELLED);
1611 }
1612
1613 if (s->error) {
1614 /* It is used on info migrate. We can't free it */
1615 error_report_err(error_copy(s->error));
1616 }
1617 notifier_list_notify(&migration_state_notifiers, s);
1618 block_cleanup_parameters(s);
1619 }
1620
1621 static void migrate_fd_cleanup_schedule(MigrationState *s)
1622 {
1623 /*
1624 * Ref the state for bh, because it may be called when
1625 * there're already no other refs
1626 */
1627 object_ref(OBJECT(s));
1628 qemu_bh_schedule(s->cleanup_bh);
1629 }
1630
1631 static void migrate_fd_cleanup_bh(void *opaque)
1632 {
1633 MigrationState *s = opaque;
1634 migrate_fd_cleanup(s);
1635 object_unref(OBJECT(s));
1636 }
1637
1638 void migrate_set_error(MigrationState *s, const Error *error)
1639 {
1640 qemu_mutex_lock(&s->error_mutex);
1641 if (!s->error) {
1642 s->error = error_copy(error);
1643 }
1644 qemu_mutex_unlock(&s->error_mutex);
1645 }
1646
1647 void migrate_fd_error(MigrationState *s, const Error *error)
1648 {
1649 trace_migrate_fd_error(error_get_pretty(error));
1650 assert(s->to_dst_file == NULL);
1651 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1652 MIGRATION_STATUS_FAILED);
1653 migrate_set_error(s, error);
1654 }
1655
1656 static void migrate_fd_cancel(MigrationState *s)
1657 {
1658 int old_state ;
1659 QEMUFile *f = migrate_get_current()->to_dst_file;
1660 trace_migrate_fd_cancel();
1661
1662 if (s->rp_state.from_dst_file) {
1663 /* shutdown the rp socket, so causing the rp thread to shutdown */
1664 qemu_file_shutdown(s->rp_state.from_dst_file);
1665 }
1666
1667 do {
1668 old_state = s->state;
1669 if (!migration_is_running(old_state)) {
1670 break;
1671 }
1672 /* If the migration is paused, kick it out of the pause */
1673 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1674 qemu_sem_post(&s->pause_sem);
1675 }
1676 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1677 } while (s->state != MIGRATION_STATUS_CANCELLING);
1678
1679 /*
1680 * If we're unlucky the migration code might be stuck somewhere in a
1681 * send/write while the network has failed and is waiting to timeout;
1682 * if we've got shutdown(2) available then we can force it to quit.
1683 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1684 * called in a bh, so there is no race against this cancel.
1685 */
1686 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1687 qemu_file_shutdown(f);
1688 }
1689 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1690 Error *local_err = NULL;
1691
1692 bdrv_invalidate_cache_all(&local_err);
1693 if (local_err) {
1694 error_report_err(local_err);
1695 } else {
1696 s->block_inactive = false;
1697 }
1698 }
1699 }
1700
1701 void add_migration_state_change_notifier(Notifier *notify)
1702 {
1703 notifier_list_add(&migration_state_notifiers, notify);
1704 }
1705
1706 void remove_migration_state_change_notifier(Notifier *notify)
1707 {
1708 notifier_remove(notify);
1709 }
1710
1711 bool migration_in_setup(MigrationState *s)
1712 {
1713 return s->state == MIGRATION_STATUS_SETUP;
1714 }
1715
1716 bool migration_has_finished(MigrationState *s)
1717 {
1718 return s->state == MIGRATION_STATUS_COMPLETED;
1719 }
1720
1721 bool migration_has_failed(MigrationState *s)
1722 {
1723 return (s->state == MIGRATION_STATUS_CANCELLED ||
1724 s->state == MIGRATION_STATUS_FAILED);
1725 }
1726
1727 bool migration_in_postcopy(void)
1728 {
1729 MigrationState *s = migrate_get_current();
1730
1731 switch (s->state) {
1732 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1733 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1734 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1735 return true;
1736 default:
1737 return false;
1738 }
1739 }
1740
1741 bool migration_in_postcopy_after_devices(MigrationState *s)
1742 {
1743 return migration_in_postcopy() && s->postcopy_after_devices;
1744 }
1745
1746 bool migration_is_idle(void)
1747 {
1748 MigrationState *s = current_migration;
1749
1750 if (!s) {
1751 return true;
1752 }
1753
1754 switch (s->state) {
1755 case MIGRATION_STATUS_NONE:
1756 case MIGRATION_STATUS_CANCELLED:
1757 case MIGRATION_STATUS_COMPLETED:
1758 case MIGRATION_STATUS_FAILED:
1759 return true;
1760 case MIGRATION_STATUS_SETUP:
1761 case MIGRATION_STATUS_CANCELLING:
1762 case MIGRATION_STATUS_ACTIVE:
1763 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1764 case MIGRATION_STATUS_COLO:
1765 case MIGRATION_STATUS_PRE_SWITCHOVER:
1766 case MIGRATION_STATUS_DEVICE:
1767 case MIGRATION_STATUS_WAIT_UNPLUG:
1768 return false;
1769 case MIGRATION_STATUS__MAX:
1770 g_assert_not_reached();
1771 }
1772
1773 return false;
1774 }
1775
1776 bool migration_is_active(MigrationState *s)
1777 {
1778 return (s->state == MIGRATION_STATUS_ACTIVE ||
1779 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1780 }
1781
1782 void migrate_init(MigrationState *s)
1783 {
1784 /*
1785 * Reinitialise all migration state, except
1786 * parameters/capabilities that the user set, and
1787 * locks.
1788 */
1789 s->cleanup_bh = 0;
1790 s->to_dst_file = NULL;
1791 s->state = MIGRATION_STATUS_NONE;
1792 s->rp_state.from_dst_file = NULL;
1793 s->rp_state.error = false;
1794 s->mbps = 0.0;
1795 s->pages_per_second = 0.0;
1796 s->downtime = 0;
1797 s->expected_downtime = 0;
1798 s->setup_time = 0;
1799 s->start_postcopy = false;
1800 s->postcopy_after_devices = false;
1801 s->migration_thread_running = false;
1802 error_free(s->error);
1803 s->error = NULL;
1804
1805 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1806
1807 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1808 s->total_time = 0;
1809 s->vm_was_running = false;
1810 s->iteration_initial_bytes = 0;
1811 s->threshold_size = 0;
1812 }
1813
1814 static GSList *migration_blockers;
1815
1816 int migrate_add_blocker(Error *reason, Error **errp)
1817 {
1818 if (only_migratable) {
1819 error_propagate_prepend(errp, error_copy(reason),
1820 "disallowing migration blocker "
1821 "(--only-migratable) for: ");
1822 return -EACCES;
1823 }
1824
1825 if (migration_is_idle()) {
1826 migration_blockers = g_slist_prepend(migration_blockers, reason);
1827 return 0;
1828 }
1829
1830 error_propagate_prepend(errp, error_copy(reason),
1831 "disallowing migration blocker "
1832 "(migration in progress) for: ");
1833 return -EBUSY;
1834 }
1835
1836 void migrate_del_blocker(Error *reason)
1837 {
1838 migration_blockers = g_slist_remove(migration_blockers, reason);
1839 }
1840
1841 void qmp_migrate_incoming(const char *uri, Error **errp)
1842 {
1843 Error *local_err = NULL;
1844 static bool once = true;
1845
1846 if (!deferred_incoming) {
1847 error_setg(errp, "For use with '-incoming defer'");
1848 return;
1849 }
1850 if (!once) {
1851 error_setg(errp, "The incoming migration has already been started");
1852 return;
1853 }
1854
1855 qemu_start_incoming_migration(uri, &local_err);
1856
1857 if (local_err) {
1858 error_propagate(errp, local_err);
1859 return;
1860 }
1861
1862 once = false;
1863 }
1864
1865 void qmp_migrate_recover(const char *uri, Error **errp)
1866 {
1867 MigrationIncomingState *mis = migration_incoming_get_current();
1868
1869 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1870 error_setg(errp, "Migrate recover can only be run "
1871 "when postcopy is paused.");
1872 return;
1873 }
1874
1875 if (atomic_cmpxchg(&mis->postcopy_recover_triggered,
1876 false, true) == true) {
1877 error_setg(errp, "Migrate recovery is triggered already");
1878 return;
1879 }
1880
1881 /*
1882 * Note that this call will never start a real migration; it will
1883 * only re-setup the migration stream and poke existing migration
1884 * to continue using that newly established channel.
1885 */
1886 qemu_start_incoming_migration(uri, errp);
1887 }
1888
1889 void qmp_migrate_pause(Error **errp)
1890 {
1891 MigrationState *ms = migrate_get_current();
1892 MigrationIncomingState *mis = migration_incoming_get_current();
1893 int ret;
1894
1895 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1896 /* Source side, during postcopy */
1897 qemu_mutex_lock(&ms->qemu_file_lock);
1898 ret = qemu_file_shutdown(ms->to_dst_file);
1899 qemu_mutex_unlock(&ms->qemu_file_lock);
1900 if (ret) {
1901 error_setg(errp, "Failed to pause source migration");
1902 }
1903 return;
1904 }
1905
1906 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1907 ret = qemu_file_shutdown(mis->from_src_file);
1908 if (ret) {
1909 error_setg(errp, "Failed to pause destination migration");
1910 }
1911 return;
1912 }
1913
1914 error_setg(errp, "migrate-pause is currently only supported "
1915 "during postcopy-active state");
1916 }
1917
1918 bool migration_is_blocked(Error **errp)
1919 {
1920 if (qemu_savevm_state_blocked(errp)) {
1921 return true;
1922 }
1923
1924 if (migration_blockers) {
1925 error_propagate(errp, error_copy(migration_blockers->data));
1926 return true;
1927 }
1928
1929 return false;
1930 }
1931
1932 /* Returns true if continue to migrate, or false if error detected */
1933 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
1934 bool resume, Error **errp)
1935 {
1936 Error *local_err = NULL;
1937
1938 if (resume) {
1939 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1940 error_setg(errp, "Cannot resume if there is no "
1941 "paused migration");
1942 return false;
1943 }
1944
1945 /*
1946 * Postcopy recovery won't work well with release-ram
1947 * capability since release-ram will drop the page buffer as
1948 * long as the page is put into the send buffer. So if there
1949 * is a network failure happened, any page buffers that have
1950 * not yet reached the destination VM but have already been
1951 * sent from the source VM will be lost forever. Let's refuse
1952 * the client from resuming such a postcopy migration.
1953 * Luckily release-ram was designed to only be used when src
1954 * and destination VMs are on the same host, so it should be
1955 * fine.
1956 */
1957 if (migrate_release_ram()) {
1958 error_setg(errp, "Postcopy recovery cannot work "
1959 "when release-ram capability is set");
1960 return false;
1961 }
1962
1963 /* This is a resume, skip init status */
1964 return true;
1965 }
1966
1967 if (migration_is_running(s->state)) {
1968 error_setg(errp, QERR_MIGRATION_ACTIVE);
1969 return false;
1970 }
1971
1972 if (runstate_check(RUN_STATE_INMIGRATE)) {
1973 error_setg(errp, "Guest is waiting for an incoming migration");
1974 return false;
1975 }
1976
1977 if (migration_is_blocked(errp)) {
1978 return false;
1979 }
1980
1981 if (blk || blk_inc) {
1982 if (migrate_use_block() || migrate_use_block_incremental()) {
1983 error_setg(errp, "Command options are incompatible with "
1984 "current migration capabilities");
1985 return false;
1986 }
1987 migrate_set_block_enabled(true, &local_err);
1988 if (local_err) {
1989 error_propagate(errp, local_err);
1990 return false;
1991 }
1992 s->must_remove_block_options = true;
1993 }
1994
1995 if (blk_inc) {
1996 migrate_set_block_incremental(s, true);
1997 }
1998
1999 migrate_init(s);
2000 /*
2001 * set ram_counters memory to zero for a
2002 * new migration
2003 */
2004 memset(&ram_counters, 0, sizeof(ram_counters));
2005
2006 return true;
2007 }
2008
2009 void qmp_migrate(const char *uri, bool has_blk, bool blk,
2010 bool has_inc, bool inc, bool has_detach, bool detach,
2011 bool has_resume, bool resume, Error **errp)
2012 {
2013 Error *local_err = NULL;
2014 MigrationState *s = migrate_get_current();
2015 const char *p;
2016
2017 if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
2018 has_resume && resume, errp)) {
2019 /* Error detected, put into errp */
2020 return;
2021 }
2022
2023 if (strstart(uri, "tcp:", &p)) {
2024 tcp_start_outgoing_migration(s, p, &local_err);
2025 #ifdef CONFIG_RDMA
2026 } else if (strstart(uri, "rdma:", &p)) {
2027 rdma_start_outgoing_migration(s, p, &local_err);
2028 #endif
2029 } else if (strstart(uri, "exec:", &p)) {
2030 exec_start_outgoing_migration(s, p, &local_err);
2031 } else if (strstart(uri, "unix:", &p)) {
2032 unix_start_outgoing_migration(s, p, &local_err);
2033 } else if (strstart(uri, "fd:", &p)) {
2034 fd_start_outgoing_migration(s, p, &local_err);
2035 } else {
2036 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
2037 "a valid migration protocol");
2038 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2039 MIGRATION_STATUS_FAILED);
2040 block_cleanup_parameters(s);
2041 return;
2042 }
2043
2044 if (local_err) {
2045 migrate_fd_error(s, local_err);
2046 error_propagate(errp, local_err);
2047 return;
2048 }
2049 }
2050
2051 void qmp_migrate_cancel(Error **errp)
2052 {
2053 migrate_fd_cancel(migrate_get_current());
2054 }
2055
2056 void qmp_migrate_continue(MigrationStatus state, Error **errp)
2057 {
2058 MigrationState *s = migrate_get_current();
2059 if (s->state != state) {
2060 error_setg(errp, "Migration not in expected state: %s",
2061 MigrationStatus_str(s->state));
2062 return;
2063 }
2064 qemu_sem_post(&s->pause_sem);
2065 }
2066
2067 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
2068 {
2069 MigrateSetParameters p = {
2070 .has_xbzrle_cache_size = true,
2071 .xbzrle_cache_size = value,
2072 };
2073
2074 qmp_migrate_set_parameters(&p, errp);
2075 }
2076
2077 int64_t qmp_query_migrate_cache_size(Error **errp)
2078 {
2079 return migrate_xbzrle_cache_size();
2080 }
2081
2082 void qmp_migrate_set_speed(int64_t value, Error **errp)
2083 {
2084 MigrateSetParameters p = {
2085 .has_max_bandwidth = true,
2086 .max_bandwidth = value,
2087 };
2088
2089 qmp_migrate_set_parameters(&p, errp);
2090 }
2091
2092 void qmp_migrate_set_downtime(double value, Error **errp)
2093 {
2094 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
2095 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
2096 "the range of 0 to %d seconds",
2097 MAX_MIGRATE_DOWNTIME_SECONDS);
2098 return;
2099 }
2100
2101 value *= 1000; /* Convert to milliseconds */
2102
2103 MigrateSetParameters p = {
2104 .has_downtime_limit = true,
2105 .downtime_limit = (int64_t)value,
2106 };
2107
2108 qmp_migrate_set_parameters(&p, errp);
2109 }
2110
2111 bool migrate_release_ram(void)
2112 {
2113 MigrationState *s;
2114
2115 s = migrate_get_current();
2116
2117 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2118 }
2119
2120 bool migrate_postcopy_ram(void)
2121 {
2122 MigrationState *s;
2123
2124 s = migrate_get_current();
2125
2126 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2127 }
2128
2129 bool migrate_postcopy(void)
2130 {
2131 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2132 }
2133
2134 bool migrate_auto_converge(void)
2135 {
2136 MigrationState *s;
2137
2138 s = migrate_get_current();
2139
2140 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2141 }
2142
2143 bool migrate_zero_blocks(void)
2144 {
2145 MigrationState *s;
2146
2147 s = migrate_get_current();
2148
2149 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2150 }
2151
2152 bool migrate_postcopy_blocktime(void)
2153 {
2154 MigrationState *s;
2155
2156 s = migrate_get_current();
2157
2158 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2159 }
2160
2161 bool migrate_use_compression(void)
2162 {
2163 MigrationState *s;
2164
2165 s = migrate_get_current();
2166
2167 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2168 }
2169
2170 int migrate_compress_level(void)
2171 {
2172 MigrationState *s;
2173
2174 s = migrate_get_current();
2175
2176 return s->parameters.compress_level;
2177 }
2178
2179 int migrate_compress_threads(void)
2180 {
2181 MigrationState *s;
2182
2183 s = migrate_get_current();
2184
2185 return s->parameters.compress_threads;
2186 }
2187
2188 int migrate_compress_wait_thread(void)
2189 {
2190 MigrationState *s;
2191
2192 s = migrate_get_current();
2193
2194 return s->parameters.compress_wait_thread;
2195 }
2196
2197 int migrate_decompress_threads(void)
2198 {
2199 MigrationState *s;
2200
2201 s = migrate_get_current();
2202
2203 return s->parameters.decompress_threads;
2204 }
2205
2206 bool migrate_dirty_bitmaps(void)
2207 {
2208 MigrationState *s;
2209
2210 s = migrate_get_current();
2211
2212 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2213 }
2214
2215 bool migrate_ignore_shared(void)
2216 {
2217 MigrationState *s;
2218
2219 s = migrate_get_current();
2220
2221 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2222 }
2223
2224 bool migrate_validate_uuid(void)
2225 {
2226 MigrationState *s;
2227
2228 s = migrate_get_current();
2229
2230 return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
2231 }
2232
2233 bool migrate_use_events(void)
2234 {
2235 MigrationState *s;
2236
2237 s = migrate_get_current();
2238
2239 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2240 }
2241
2242 bool migrate_use_multifd(void)
2243 {
2244 MigrationState *s;
2245
2246 s = migrate_get_current();
2247
2248 return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2249 }
2250
2251 bool migrate_pause_before_switchover(void)
2252 {
2253 MigrationState *s;
2254
2255 s = migrate_get_current();
2256
2257 return s->enabled_capabilities[
2258 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2259 }
2260
2261 int migrate_multifd_channels(void)
2262 {
2263 MigrationState *s;
2264
2265 s = migrate_get_current();
2266
2267 return s->parameters.multifd_channels;
2268 }
2269
2270 MultiFDCompression migrate_multifd_compression(void)
2271 {
2272 MigrationState *s;
2273
2274 s = migrate_get_current();
2275
2276 return s->parameters.multifd_compression;
2277 }
2278
2279 int migrate_multifd_zlib_level(void)
2280 {
2281 MigrationState *s;
2282
2283 s = migrate_get_current();
2284
2285 return s->parameters.multifd_zlib_level;
2286 }
2287
2288 int migrate_multifd_zstd_level(void)
2289 {
2290 MigrationState *s;
2291
2292 s = migrate_get_current();
2293
2294 return s->parameters.multifd_zstd_level;
2295 }
2296
2297 int migrate_use_xbzrle(void)
2298 {
2299 MigrationState *s;
2300
2301 s = migrate_get_current();
2302
2303 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2304 }
2305
2306 int64_t migrate_xbzrle_cache_size(void)
2307 {
2308 MigrationState *s;
2309
2310 s = migrate_get_current();
2311
2312 return s->parameters.xbzrle_cache_size;
2313 }
2314
2315 static int64_t migrate_max_postcopy_bandwidth(void)
2316 {
2317 MigrationState *s;
2318
2319 s = migrate_get_current();
2320
2321 return s->parameters.max_postcopy_bandwidth;
2322 }
2323
2324 bool migrate_use_block(void)
2325 {
2326 MigrationState *s;
2327
2328 s = migrate_get_current();
2329
2330 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2331 }
2332
2333 bool migrate_use_return_path(void)
2334 {
2335 MigrationState *s;
2336
2337 s = migrate_get_current();
2338
2339 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2340 }
2341
2342 bool migrate_use_block_incremental(void)
2343 {
2344 MigrationState *s;
2345
2346 s = migrate_get_current();
2347
2348 return s->parameters.block_incremental;
2349 }
2350
2351 /* migration thread support */
2352 /*
2353 * Something bad happened to the RP stream, mark an error
2354 * The caller shall print or trace something to indicate why
2355 */
2356 static void mark_source_rp_bad(MigrationState *s)
2357 {
2358 s->rp_state.error = true;
2359 }
2360
2361 static struct rp_cmd_args {
2362 ssize_t len; /* -1 = variable */
2363 const char *name;
2364 } rp_cmd_args[] = {
2365 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
2366 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
2367 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
2368 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
2369 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
2370 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
2371 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
2372 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
2373 };
2374
2375 /*
2376 * Process a request for pages received on the return path,
2377 * We're allowed to send more than requested (e.g. to round to our page size)
2378 * and we don't need to send pages that have already been sent.
2379 */
2380 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2381 ram_addr_t start, size_t len)
2382 {
2383 long our_host_ps = qemu_real_host_page_size;
2384
2385 trace_migrate_handle_rp_req_pages(rbname, start, len);
2386
2387 /*
2388 * Since we currently insist on matching page sizes, just sanity check
2389 * we're being asked for whole host pages.
2390 */
2391 if (start & (our_host_ps-1) ||
2392 (len & (our_host_ps-1))) {
2393 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2394 " len: %zd", __func__, start, len);
2395 mark_source_rp_bad(ms);
2396 return;
2397 }
2398
2399 if (ram_save_queue_pages(rbname, start, len)) {
2400 mark_source_rp_bad(ms);
2401 }
2402 }
2403
2404 /* Return true to retry, false to quit */
2405 static bool postcopy_pause_return_path_thread(MigrationState *s)
2406 {
2407 trace_postcopy_pause_return_path();
2408
2409 qemu_sem_wait(&s->postcopy_pause_rp_sem);
2410
2411 trace_postcopy_pause_return_path_continued();
2412
2413 return true;
2414 }
2415
2416 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2417 {
2418 RAMBlock *block = qemu_ram_block_by_name(block_name);
2419
2420 if (!block) {
2421 error_report("%s: invalid block name '%s'", __func__, block_name);
2422 return -EINVAL;
2423 }
2424
2425 /* Fetch the received bitmap and refresh the dirty bitmap */
2426 return ram_dirty_bitmap_reload(s, block);
2427 }
2428
2429 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2430 {
2431 trace_source_return_path_thread_resume_ack(value);
2432
2433 if (value != MIGRATION_RESUME_ACK_VALUE) {
2434 error_report("%s: illegal resume_ack value %"PRIu32,
2435 __func__, value);
2436 return -1;
2437 }
2438
2439 /* Now both sides are active. */
2440 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2441 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2442
2443 /* Notify send thread that time to continue send pages */
2444 qemu_sem_post(&s->rp_state.rp_sem);
2445
2446 return 0;
2447 }
2448
2449 /*
2450 * Handles messages sent on the return path towards the source VM
2451 *
2452 */
2453 static void *source_return_path_thread(void *opaque)
2454 {
2455 MigrationState *ms = opaque;
2456 QEMUFile *rp = ms->rp_state.from_dst_file;
2457 uint16_t header_len, header_type;
2458 uint8_t buf[512];
2459 uint32_t tmp32, sibling_error;
2460 ram_addr_t start = 0; /* =0 to silence warning */
2461 size_t len = 0, expected_len;
2462 int res;
2463
2464 trace_source_return_path_thread_entry();
2465 rcu_register_thread();
2466
2467 retry:
2468 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2469 migration_is_setup_or_active(ms->state)) {
2470 trace_source_return_path_thread_loop_top();
2471 header_type = qemu_get_be16(rp);
2472 header_len = qemu_get_be16(rp);
2473
2474 if (qemu_file_get_error(rp)) {
2475 mark_source_rp_bad(ms);
2476 goto out;
2477 }
2478
2479 if (header_type >= MIG_RP_MSG_MAX ||
2480 header_type == MIG_RP_MSG_INVALID) {
2481 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2482 header_type, header_len);
2483 mark_source_rp_bad(ms);
2484 goto out;
2485 }
2486
2487 if ((rp_cmd_args[header_type].len != -1 &&
2488 header_len != rp_cmd_args[header_type].len) ||
2489 header_len > sizeof(buf)) {
2490 error_report("RP: Received '%s' message (0x%04x) with"
2491 "incorrect length %d expecting %zu",
2492 rp_cmd_args[header_type].name, header_type, header_len,
2493 (size_t)rp_cmd_args[header_type].len);
2494 mark_source_rp_bad(ms);
2495 goto out;
2496 }
2497
2498 /* We know we've got a valid header by this point */
2499 res = qemu_get_buffer(rp, buf, header_len);
2500 if (res != header_len) {
2501 error_report("RP: Failed reading data for message 0x%04x"
2502 " read %d expected %d",
2503 header_type, res, header_len);
2504 mark_source_rp_bad(ms);
2505 goto out;
2506 }
2507
2508 /* OK, we have the message and the data */
2509 switch (header_type) {
2510 case MIG_RP_MSG_SHUT:
2511 sibling_error = ldl_be_p(buf);
2512 trace_source_return_path_thread_shut(sibling_error);
2513 if (sibling_error) {
2514 error_report("RP: Sibling indicated error %d", sibling_error);
2515 mark_source_rp_bad(ms);
2516 }
2517 /*
2518 * We'll let the main thread deal with closing the RP
2519 * we could do a shutdown(2) on it, but we're the only user
2520 * anyway, so there's nothing gained.
2521 */
2522 goto out;
2523
2524 case MIG_RP_MSG_PONG:
2525 tmp32 = ldl_be_p(buf);
2526 trace_source_return_path_thread_pong(tmp32);
2527 break;
2528
2529 case MIG_RP_MSG_REQ_PAGES:
2530 start = ldq_be_p(buf);
2531 len = ldl_be_p(buf + 8);
2532 migrate_handle_rp_req_pages(ms, NULL, start, len);
2533 break;
2534
2535 case MIG_RP_MSG_REQ_PAGES_ID:
2536 expected_len = 12 + 1; /* header + termination */
2537
2538 if (header_len >= expected_len) {
2539 start = ldq_be_p(buf);
2540 len = ldl_be_p(buf + 8);
2541 /* Now we expect an idstr */
2542 tmp32 = buf[12]; /* Length of the following idstr */
2543 buf[13 + tmp32] = '\0';
2544 expected_len += tmp32;
2545 }
2546 if (header_len != expected_len) {
2547 error_report("RP: Req_Page_id with length %d expecting %zd",
2548 header_len, expected_len);
2549 mark_source_rp_bad(ms);
2550 goto out;
2551 }
2552 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2553 break;
2554
2555 case MIG_RP_MSG_RECV_BITMAP:
2556 if (header_len < 1) {
2557 error_report("%s: missing block name", __func__);
2558 mark_source_rp_bad(ms);
2559 goto out;
2560 }
2561 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2562 buf[buf[0] + 1] = '\0';
2563 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2564 mark_source_rp_bad(ms);
2565 goto out;
2566 }
2567 break;
2568
2569 case MIG_RP_MSG_RESUME_ACK:
2570 tmp32 = ldl_be_p(buf);
2571 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2572 mark_source_rp_bad(ms);
2573 goto out;
2574 }
2575 break;
2576
2577 default:
2578 break;
2579 }
2580 }
2581
2582 out:
2583 res = qemu_file_get_error(rp);
2584 if (res) {
2585 if (res == -EIO && migration_in_postcopy()) {
2586 /*
2587 * Maybe there is something we can do: it looks like a
2588 * network down issue, and we pause for a recovery.
2589 */
2590 if (postcopy_pause_return_path_thread(ms)) {
2591 /* Reload rp, reset the rest */
2592 if (rp != ms->rp_state.from_dst_file) {
2593 qemu_fclose(rp);
2594 rp = ms->rp_state.from_dst_file;
2595 }
2596 ms->rp_state.error = false;
2597 goto retry;
2598 }
2599 }
2600
2601 trace_source_return_path_thread_bad_end();
2602 mark_source_rp_bad(ms);
2603 }
2604
2605 trace_source_return_path_thread_end();
2606 ms->rp_state.from_dst_file = NULL;
2607 qemu_fclose(rp);
2608 rcu_unregister_thread();
2609 return NULL;
2610 }
2611
2612 static int open_return_path_on_source(MigrationState *ms,
2613 bool create_thread)
2614 {
2615
2616 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2617 if (!ms->rp_state.from_dst_file) {
2618 return -1;
2619 }
2620
2621 trace_open_return_path_on_source();
2622
2623 if (!create_thread) {
2624 /* We're done */
2625 return 0;
2626 }
2627
2628 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2629 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2630
2631 trace_open_return_path_on_source_continue();
2632
2633 return 0;
2634 }
2635
2636 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2637 static int await_return_path_close_on_source(MigrationState *ms)
2638 {
2639 /*
2640 * If this is a normal exit then the destination will send a SHUT and the
2641 * rp_thread will exit, however if there's an error we need to cause
2642 * it to exit.
2643 */
2644 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2645 /*
2646 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2647 * waiting for the destination.
2648 */
2649 qemu_file_shutdown(ms->rp_state.from_dst_file);
2650 mark_source_rp_bad(ms);
2651 }
2652 trace_await_return_path_close_on_source_joining();
2653 qemu_thread_join(&ms->rp_state.rp_thread);
2654 trace_await_return_path_close_on_source_close();
2655 return ms->rp_state.error;
2656 }
2657
2658 /*
2659 * Switch from normal iteration to postcopy
2660 * Returns non-0 on error
2661 */
2662 static int postcopy_start(MigrationState *ms)
2663 {
2664 int ret;
2665 QIOChannelBuffer *bioc;
2666 QEMUFile *fb;
2667 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2668 int64_t bandwidth = migrate_max_postcopy_bandwidth();
2669 bool restart_block = false;
2670 int cur_state = MIGRATION_STATUS_ACTIVE;
2671 if (!migrate_pause_before_switchover()) {
2672 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2673 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2674 }
2675
2676 trace_postcopy_start();
2677 qemu_mutex_lock_iothread();
2678 trace_postcopy_start_set_run();
2679
2680 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2681 global_state_store();
2682 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2683 if (ret < 0) {
2684 goto fail;
2685 }
2686
2687 ret = migration_maybe_pause(ms, &cur_state,
2688 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2689 if (ret < 0) {
2690 goto fail;
2691 }
2692
2693 ret = bdrv_inactivate_all();
2694 if (ret < 0) {
2695 goto fail;
2696 }
2697 restart_block = true;
2698
2699 /*
2700 * Cause any non-postcopiable, but iterative devices to
2701 * send out their final data.
2702 */
2703 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2704
2705 /*
2706 * in Finish migrate and with the io-lock held everything should
2707 * be quiet, but we've potentially still got dirty pages and we
2708 * need to tell the destination to throw any pages it's already received
2709 * that are dirty
2710 */
2711 if (migrate_postcopy_ram()) {
2712 if (ram_postcopy_send_discard_bitmap(ms)) {
2713 error_report("postcopy send discard bitmap failed");
2714 goto fail;
2715 }
2716 }
2717
2718 /*
2719 * send rest of state - note things that are doing postcopy
2720 * will notice we're in POSTCOPY_ACTIVE and not actually
2721 * wrap their state up here
2722 */
2723 /* 0 max-postcopy-bandwidth means unlimited */
2724 if (!bandwidth) {
2725 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2726 } else {
2727 qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2728 }
2729 if (migrate_postcopy_ram()) {
2730 /* Ping just for debugging, helps line traces up */
2731 qemu_savevm_send_ping(ms->to_dst_file, 2);
2732 }
2733
2734 /*
2735 * While loading the device state we may trigger page transfer
2736 * requests and the fd must be free to process those, and thus
2737 * the destination must read the whole device state off the fd before
2738 * it starts processing it. Unfortunately the ad-hoc migration format
2739 * doesn't allow the destination to know the size to read without fully
2740 * parsing it through each devices load-state code (especially the open
2741 * coded devices that use get/put).
2742 * So we wrap the device state up in a package with a length at the start;
2743 * to do this we use a qemu_buf to hold the whole of the device state.
2744 */
2745 bioc = qio_channel_buffer_new(4096);
2746 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2747 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2748 object_unref(OBJECT(bioc));
2749
2750 /*
2751 * Make sure the receiver can get incoming pages before we send the rest
2752 * of the state
2753 */
2754 qemu_savevm_send_postcopy_listen(fb);
2755
2756 qemu_savevm_state_complete_precopy(fb, false, false);
2757 if (migrate_postcopy_ram()) {
2758 qemu_savevm_send_ping(fb, 3);
2759 }
2760
2761 qemu_savevm_send_postcopy_run(fb);
2762
2763 /* <><> end of stuff going into the package */
2764
2765 /* Last point of recovery; as soon as we send the package the destination
2766 * can open devices and potentially start running.
2767 * Lets just check again we've not got any errors.
2768 */
2769 ret = qemu_file_get_error(ms->to_dst_file);
2770 if (ret) {
2771 error_report("postcopy_start: Migration stream errored (pre package)");
2772 goto fail_closefb;
2773 }
2774
2775 restart_block = false;
2776
2777 /* Now send that blob */
2778 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2779 goto fail_closefb;
2780 }
2781 qemu_fclose(fb);
2782
2783 /* Send a notify to give a chance for anything that needs to happen
2784 * at the transition to postcopy and after the device state; in particular
2785 * spice needs to trigger a transition now
2786 */
2787 ms->postcopy_after_devices = true;
2788 notifier_list_notify(&migration_state_notifiers, ms);
2789
2790 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2791
2792 qemu_mutex_unlock_iothread();
2793
2794 if (migrate_postcopy_ram()) {
2795 /*
2796 * Although this ping is just for debug, it could potentially be
2797 * used for getting a better measurement of downtime at the source.
2798 */
2799 qemu_savevm_send_ping(ms->to_dst_file, 4);
2800 }
2801
2802 if (migrate_release_ram()) {
2803 ram_postcopy_migrated_memory_release(ms);
2804 }
2805
2806 ret = qemu_file_get_error(ms->to_dst_file);
2807 if (ret) {
2808 error_report("postcopy_start: Migration stream errored");
2809 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2810 MIGRATION_STATUS_FAILED);
2811 }
2812
2813 return ret;
2814
2815 fail_closefb:
2816 qemu_fclose(fb);
2817 fail:
2818 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2819 MIGRATION_STATUS_FAILED);
2820 if (restart_block) {
2821 /* A failure happened early enough that we know the destination hasn't
2822 * accessed block devices, so we're safe to recover.
2823 */
2824 Error *local_err = NULL;
2825
2826 bdrv_invalidate_cache_all(&local_err);
2827 if (local_err) {
2828 error_report_err(local_err);
2829 }
2830 }
2831 qemu_mutex_unlock_iothread();
2832 return -1;
2833 }
2834
2835 /**
2836 * migration_maybe_pause: Pause if required to by
2837 * migrate_pause_before_switchover called with the iothread locked
2838 * Returns: 0 on success
2839 */
2840 static int migration_maybe_pause(MigrationState *s,
2841 int *current_active_state,
2842 int new_state)
2843 {
2844 if (!migrate_pause_before_switchover()) {
2845 return 0;
2846 }
2847
2848 /* Since leaving this state is not atomic with posting the semaphore
2849 * it's possible that someone could have issued multiple migrate_continue
2850 * and the semaphore is incorrectly positive at this point;
2851 * the docs say it's undefined to reinit a semaphore that's already
2852 * init'd, so use timedwait to eat up any existing posts.
2853 */
2854 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2855 /* This block intentionally left blank */
2856 }
2857
2858 /*
2859 * If the migration is cancelled when it is in the completion phase,
2860 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2861 * So we don't need to wait a semaphore, otherwise we would always
2862 * wait for the 'pause_sem' semaphore.
2863 */
2864 if (s->state != MIGRATION_STATUS_CANCELLING) {
2865 qemu_mutex_unlock_iothread();
2866 migrate_set_state(&s->state, *current_active_state,
2867 MIGRATION_STATUS_PRE_SWITCHOVER);
2868 qemu_sem_wait(&s->pause_sem);
2869 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2870 new_state);
2871 *current_active_state = new_state;
2872 qemu_mutex_lock_iothread();
2873 }
2874
2875 return s->state == new_state ? 0 : -EINVAL;
2876 }
2877
2878 /**
2879 * migration_completion: Used by migration_thread when there's not much left.
2880 * The caller 'breaks' the loop when this returns.
2881 *
2882 * @s: Current migration state
2883 */
2884 static void migration_completion(MigrationState *s)
2885 {
2886 int ret;
2887 int current_active_state = s->state;
2888
2889 if (s->state == MIGRATION_STATUS_ACTIVE) {
2890 qemu_mutex_lock_iothread();
2891 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2892 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2893 s->vm_was_running = runstate_is_running();
2894 ret = global_state_store();
2895
2896 if (!ret) {
2897 bool inactivate = !migrate_colo_enabled();
2898 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2899 if (ret >= 0) {
2900 ret = migration_maybe_pause(s, &current_active_state,
2901 MIGRATION_STATUS_DEVICE);
2902 }
2903 if (ret >= 0) {
2904 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2905 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2906 inactivate);
2907 }
2908 if (inactivate && ret >= 0) {
2909 s->block_inactive = true;
2910 }
2911 }
2912 qemu_mutex_unlock_iothread();
2913
2914 if (ret < 0) {
2915 goto fail;
2916 }
2917 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2918 trace_migration_completion_postcopy_end();
2919
2920 qemu_savevm_state_complete_postcopy(s->to_dst_file);
2921 trace_migration_completion_postcopy_end_after_complete();
2922 }
2923
2924 /*
2925 * If rp was opened we must clean up the thread before
2926 * cleaning everything else up (since if there are no failures
2927 * it will wait for the destination to send it's status in
2928 * a SHUT command).
2929 */
2930 if (s->rp_state.from_dst_file) {
2931 int rp_error;
2932 trace_migration_return_path_end_before();
2933 rp_error = await_return_path_close_on_source(s);
2934 trace_migration_return_path_end_after(rp_error);
2935 if (rp_error) {
2936 goto fail_invalidate;
2937 }
2938 }
2939
2940 if (qemu_file_get_error(s->to_dst_file)) {
2941 trace_migration_completion_file_err();
2942 goto fail_invalidate;
2943 }
2944
2945 if (!migrate_colo_enabled()) {
2946 migrate_set_state(&s->state, current_active_state,
2947 MIGRATION_STATUS_COMPLETED);
2948 }
2949
2950 return;
2951
2952 fail_invalidate:
2953 /* If not doing postcopy, vm_start() will be called: let's regain
2954 * control on images.
2955 */
2956 if (s->state == MIGRATION_STATUS_ACTIVE ||
2957 s->state == MIGRATION_STATUS_DEVICE) {
2958 Error *local_err = NULL;
2959
2960 qemu_mutex_lock_iothread();
2961 bdrv_invalidate_cache_all(&local_err);
2962 if (local_err) {
2963 error_report_err(local_err);
2964 } else {
2965 s->block_inactive = false;
2966 }
2967 qemu_mutex_unlock_iothread();
2968 }
2969
2970 fail:
2971 migrate_set_state(&s->state, current_active_state,
2972 MIGRATION_STATUS_FAILED);
2973 }
2974
2975 bool migrate_colo_enabled(void)
2976 {
2977 MigrationState *s = migrate_get_current();
2978 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2979 }
2980
2981 typedef enum MigThrError {
2982 /* No error detected */
2983 MIG_THR_ERR_NONE = 0,
2984 /* Detected error, but resumed successfully */
2985 MIG_THR_ERR_RECOVERED = 1,
2986 /* Detected fatal error, need to exit */
2987 MIG_THR_ERR_FATAL = 2,
2988 } MigThrError;
2989
2990 static int postcopy_resume_handshake(MigrationState *s)
2991 {
2992 qemu_savevm_send_postcopy_resume(s->to_dst_file);
2993
2994 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2995 qemu_sem_wait(&s->rp_state.rp_sem);
2996 }
2997
2998 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2999 return 0;
3000 }
3001
3002 return -1;
3003 }
3004
3005 /* Return zero if success, or <0 for error */
3006 static int postcopy_do_resume(MigrationState *s)
3007 {
3008 int ret;
3009
3010 /*
3011 * Call all the resume_prepare() hooks, so that modules can be
3012 * ready for the migration resume.
3013 */
3014 ret = qemu_savevm_state_resume_prepare(s);
3015 if (ret) {
3016 error_report("%s: resume_prepare() failure detected: %d",
3017 __func__, ret);
3018 return ret;
3019 }
3020
3021 /*
3022 * Last handshake with destination on the resume (destination will
3023 * switch to postcopy-active afterwards)
3024 */
3025 ret = postcopy_resume_handshake(s);
3026 if (ret) {
3027 error_report("%s: handshake failed: %d", __func__, ret);
3028 return ret;
3029 }
3030
3031 return 0;
3032 }
3033
3034 /*
3035 * We don't return until we are in a safe state to continue current
3036 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3037 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3038 */
3039 static MigThrError postcopy_pause(MigrationState *s)
3040 {
3041 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
3042
3043 while (true) {
3044 QEMUFile *file;
3045
3046 migrate_set_state(&s->state, s->state,
3047 MIGRATION_STATUS_POSTCOPY_PAUSED);
3048
3049 /* Current channel is possibly broken. Release it. */
3050 assert(s->to_dst_file);
3051 qemu_mutex_lock(&s->qemu_file_lock);
3052 file = s->to_dst_file;
3053 s->to_dst_file = NULL;
3054 qemu_mutex_unlock(&s->qemu_file_lock);
3055
3056 qemu_file_shutdown(file);
3057 qemu_fclose(file);
3058
3059 error_report("Detected IO failure for postcopy. "
3060 "Migration paused.");
3061
3062 /*
3063 * We wait until things fixed up. Then someone will setup the
3064 * status back for us.
3065 */
3066 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
3067 qemu_sem_wait(&s->postcopy_pause_sem);
3068 }
3069
3070 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3071 /* Woken up by a recover procedure. Give it a shot */
3072
3073 /*
3074 * Firstly, let's wake up the return path now, with a new
3075 * return path channel.
3076 */
3077 qemu_sem_post(&s->postcopy_pause_rp_sem);
3078
3079 /* Do the resume logic */
3080 if (postcopy_do_resume(s) == 0) {
3081 /* Let's continue! */
3082 trace_postcopy_pause_continued();
3083 return MIG_THR_ERR_RECOVERED;
3084 } else {
3085 /*
3086 * Something wrong happened during the recovery, let's
3087 * pause again. Pause is always better than throwing
3088 * data away.
3089 */
3090 continue;
3091 }
3092 } else {
3093 /* This is not right... Time to quit. */
3094 return MIG_THR_ERR_FATAL;
3095 }
3096 }
3097 }
3098
3099 static MigThrError migration_detect_error(MigrationState *s)
3100 {
3101 int ret;
3102 int state = s->state;
3103 Error *local_error = NULL;
3104
3105 if (state == MIGRATION_STATUS_CANCELLING ||
3106 state == MIGRATION_STATUS_CANCELLED) {
3107 /* End the migration, but don't set the state to failed */
3108 return MIG_THR_ERR_FATAL;
3109 }
3110
3111 /* Try to detect any file errors */
3112 ret = qemu_file_get_error_obj(s->to_dst_file, &local_error);
3113 if (!ret) {
3114 /* Everything is fine */
3115 assert(!local_error);
3116 return MIG_THR_ERR_NONE;
3117 }
3118
3119 if (local_error) {
3120 migrate_set_error(s, local_error);
3121 error_free(local_error);
3122 }
3123
3124 if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
3125 /*
3126 * For postcopy, we allow the network to be down for a
3127 * while. After that, it can be continued by a
3128 * recovery phase.
3129 */
3130 return postcopy_pause(s);
3131 } else {
3132 /*
3133 * For precopy (or postcopy with error outside IO), we fail
3134 * with no time.
3135 */
3136 migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
3137 trace_migration_thread_file_err();
3138
3139 /* Time to stop the migration, now. */
3140 return MIG_THR_ERR_FATAL;
3141 }
3142 }
3143
3144 /* How many bytes have we transferred since the beginning of the migration */
3145 static uint64_t migration_total_bytes(MigrationState *s)
3146 {
3147 return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3148 }
3149
3150 static void migration_calculate_complete(MigrationState *s)
3151 {
3152 uint64_t bytes = migration_total_bytes(s);
3153 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3154 int64_t transfer_time;
3155
3156 s->total_time = end_time - s->start_time;
3157 if (!s->downtime) {
3158 /*
3159 * It's still not set, so we are precopy migration. For
3160 * postcopy, downtime is calculated during postcopy_start().
3161 */
3162 s->downtime = end_time - s->downtime_start;
3163 }
3164
3165 transfer_time = s->total_time - s->setup_time;
3166 if (transfer_time) {
3167 s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3168 }
3169 }
3170
3171 static void update_iteration_initial_status(MigrationState *s)
3172 {
3173 /*
3174 * Update these three fields at the same time to avoid mismatch info lead
3175 * wrong speed calculation.
3176 */
3177 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3178 s->iteration_initial_bytes = migration_total_bytes(s);
3179 s->iteration_initial_pages = ram_get_total_transferred_pages();
3180 }
3181
3182 static void migration_update_counters(MigrationState *s,
3183 int64_t current_time)
3184 {
3185 uint64_t transferred, transferred_pages, time_spent;
3186 uint64_t current_bytes; /* bytes transferred since the beginning */
3187 double bandwidth;
3188
3189 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3190 return;
3191 }
3192
3193 current_bytes = migration_total_bytes(s);
3194 transferred = current_bytes - s->iteration_initial_bytes;
3195 time_spent = current_time - s->iteration_start_time;
3196 bandwidth = (double)transferred / time_spent;
3197 s->threshold_size = bandwidth * s->parameters.downtime_limit;
3198
3199 s->mbps = (((double) transferred * 8.0) /
3200 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3201
3202 transferred_pages = ram_get_total_transferred_pages() -
3203 s->iteration_initial_pages;
3204 s->pages_per_second = (double) transferred_pages /
3205 (((double) time_spent / 1000.0));
3206
3207 /*
3208 * if we haven't sent anything, we don't want to
3209 * recalculate. 10000 is a small enough number for our purposes
3210 */
3211 if (ram_counters.dirty_pages_rate && transferred > 10000) {
3212 s->expected_downtime = ram_counters.remaining / bandwidth;
3213 }
3214
3215 qemu_file_reset_rate_limit(s->to_dst_file);
3216
3217 update_iteration_initial_status(s);
3218
3219 trace_migrate_transferred(transferred, time_spent,
3220 bandwidth, s->threshold_size);
3221 }
3222
3223 /* Migration thread iteration status */
3224 typedef enum {
3225 MIG_ITERATE_RESUME, /* Resume current iteration */
3226 MIG_ITERATE_SKIP, /* Skip current iteration */
3227 MIG_ITERATE_BREAK, /* Break the loop */
3228 } MigIterateState;
3229
3230 /*
3231 * Return true if continue to the next iteration directly, false
3232 * otherwise.
3233 */
3234 static MigIterateState migration_iteration_run(MigrationState *s)
3235 {
3236 uint64_t pending_size, pend_pre, pend_compat, pend_post;
3237 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3238
3239 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3240 &pend_compat, &pend_post);
3241 pending_size = pend_pre + pend_compat + pend_post;
3242
3243 trace_migrate_pending(pending_size, s->threshold_size,
3244 pend_pre, pend_compat, pend_post);
3245
3246 if (pending_size && pending_size >= s->threshold_size) {
3247 /* Still a significant amount to transfer */
3248 if (!in_postcopy && pend_pre <= s->threshold_size &&
3249 atomic_read(&s->start_postcopy)) {
3250 if (postcopy_start(s)) {
3251 error_report("%s: postcopy failed to start", __func__);
3252 }
3253 return MIG_ITERATE_SKIP;
3254 }
3255 /* Just another iteration step */
3256 qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
3257 } else {
3258 trace_migration_thread_low_pending(pending_size);
3259 migration_completion(s);
3260 return MIG_ITERATE_BREAK;
3261 }
3262
3263 return MIG_ITERATE_RESUME;
3264 }
3265
3266 static void migration_iteration_finish(MigrationState *s)
3267 {
3268 /* If we enabled cpu throttling for auto-converge, turn it off. */
3269 cpu_throttle_stop();
3270
3271 qemu_mutex_lock_iothread();
3272 switch (s->state) {
3273 case MIGRATION_STATUS_COMPLETED:
3274 migration_calculate_complete(s);
3275 runstate_set(RUN_STATE_POSTMIGRATE);
3276 break;
3277
3278 case MIGRATION_STATUS_ACTIVE:
3279 /*
3280 * We should really assert here, but since it's during
3281 * migration, let's try to reduce the usage of assertions.
3282 */
3283 if (!migrate_colo_enabled()) {
3284 error_report("%s: critical error: calling COLO code without "
3285 "COLO enabled", __func__);
3286 }
3287 migrate_start_colo_process(s);
3288 /*
3289 * Fixme: we will run VM in COLO no matter its old running state.
3290 * After exited COLO, we will keep running.
3291 */
3292 s->vm_was_running = true;
3293 /* Fallthrough */
3294 case MIGRATION_STATUS_FAILED:
3295 case MIGRATION_STATUS_CANCELLED:
3296 case MIGRATION_STATUS_CANCELLING:
3297 if (s->vm_was_running) {
3298 vm_start();
3299 } else {
3300 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3301 runstate_set(RUN_STATE_POSTMIGRATE);
3302 }
3303 }
3304 break;
3305
3306 default:
3307 /* Should not reach here, but if so, forgive the VM. */
3308 error_report("%s: Unknown ending state %d", __func__, s->state);
3309 break;
3310 }
3311 migrate_fd_cleanup_schedule(s);
3312 qemu_mutex_unlock_iothread();
3313 }
3314
3315 void migration_make_urgent_request(void)
3316 {
3317 qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3318 }
3319
3320 void migration_consume_urgent_request(void)
3321 {
3322 qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3323 }
3324
3325 /* Returns true if the rate limiting was broken by an urgent request */
3326 bool migration_rate_limit(void)
3327 {
3328 int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3329 MigrationState *s = migrate_get_current();
3330
3331 bool urgent = false;
3332 migration_update_counters(s, now);
3333 if (qemu_file_rate_limit(s->to_dst_file)) {
3334 /*
3335 * Wait for a delay to do rate limiting OR
3336 * something urgent to post the semaphore.
3337 */
3338 int ms = s->iteration_start_time + BUFFER_DELAY - now;
3339 trace_migration_rate_limit_pre(ms);
3340 if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3341 /*
3342 * We were woken by one or more urgent things but
3343 * the timedwait will have consumed one of them.
3344 * The service routine for the urgent wake will dec
3345 * the semaphore itself for each item it consumes,
3346 * so add this one we just eat back.
3347 */
3348 qemu_sem_post(&s->rate_limit_sem);
3349 urgent = true;
3350 }
3351 trace_migration_rate_limit_post(urgent);
3352 }
3353 return urgent;
3354 }
3355
3356 /*
3357 * Master migration thread on the source VM.
3358 * It drives the migration and pumps the data down the outgoing channel.
3359 */
3360 static void *migration_thread(void *opaque)
3361 {
3362 MigrationState *s = opaque;
3363 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3364 MigThrError thr_error;
3365 bool urgent = false;
3366
3367 rcu_register_thread();
3368
3369 object_ref(OBJECT(s));
3370 update_iteration_initial_status(s);
3371
3372 qemu_savevm_state_header(s->to_dst_file);
3373
3374 /*
3375 * If we opened the return path, we need to make sure dst has it
3376 * opened as well.
3377 */
3378 if (s->rp_state.from_dst_file) {
3379 /* Now tell the dest that it should open its end so it can reply */
3380 qemu_savevm_send_open_return_path(s->to_dst_file);
3381
3382 /* And do a ping that will make stuff easier to debug */
3383 qemu_savevm_send_ping(s->to_dst_file, 1);
3384 }
3385
3386 if (migrate_postcopy()) {
3387 /*
3388 * Tell the destination that we *might* want to do postcopy later;
3389 * if the other end can't do postcopy it should fail now, nice and
3390 * early.
3391 */
3392 qemu_savevm_send_postcopy_advise(s->to_dst_file);
3393 }
3394
3395 if (migrate_colo_enabled()) {
3396 /* Notify migration destination that we enable COLO */
3397 qemu_savevm_send_colo_enable(s->to_dst_file);
3398 }
3399
3400 qemu_savevm_state_setup(s->to_dst_file);
3401
3402 if (qemu_savevm_state_guest_unplug_pending()) {
3403 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3404 MIGRATION_STATUS_WAIT_UNPLUG);
3405
3406 while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3407 qemu_savevm_state_guest_unplug_pending()) {
3408 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3409 }
3410
3411 migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG,
3412 MIGRATION_STATUS_ACTIVE);
3413 }
3414
3415 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3416 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3417 MIGRATION_STATUS_ACTIVE);
3418
3419 trace_migration_thread_setup_complete();
3420
3421 while (migration_is_active(s)) {
3422 if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3423 MigIterateState iter_state = migration_iteration_run(s);
3424 if (iter_state == MIG_ITERATE_SKIP) {
3425 continue;
3426 } else if (iter_state == MIG_ITERATE_BREAK) {
3427 break;
3428 }
3429 }
3430
3431 /*
3432 * Try to detect any kind of failures, and see whether we
3433 * should stop the migration now.
3434 */
3435 thr_error = migration_detect_error(s);
3436 if (thr_error == MIG_THR_ERR_FATAL) {
3437 /* Stop migration */
3438 break;
3439 } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3440 /*
3441 * Just recovered from a e.g. network failure, reset all
3442 * the local variables. This is important to avoid
3443 * breaking transferred_bytes and bandwidth calculation
3444 */
3445 update_iteration_initial_status(s);
3446 }
3447
3448 urgent = migration_rate_limit();
3449 }
3450
3451 trace_migration_thread_after_loop();
3452 migration_iteration_finish(s);
3453 object_unref(OBJECT(s));
3454 rcu_unregister_thread();
3455 return NULL;
3456 }
3457
3458 void migrate_fd_connect(MigrationState *s, Error *error_in)
3459 {
3460 Error *local_err = NULL;
3461 int64_t rate_limit;
3462 bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
3463
3464 s->expected_downtime = s->parameters.downtime_limit;
3465 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s);
3466 if (error_in) {
3467 migrate_fd_error(s, error_in);
3468 migrate_fd_cleanup(s);
3469 return;
3470 }
3471
3472 if (resume) {
3473 /* This is a resumed migration */
3474 rate_limit = s->parameters.max_postcopy_bandwidth /
3475 XFER_LIMIT_RATIO;
3476 } else {
3477 /* This is a fresh new migration */
3478 rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
3479
3480 /* Notify before starting migration thread */
3481 notifier_list_notify(&migration_state_notifiers, s);
3482 }
3483
3484 qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
3485 qemu_file_set_blocking(s->to_dst_file, true);
3486
3487 /*
3488 * Open the return path. For postcopy, it is used exclusively. For
3489 * precopy, only if user specified "return-path" capability would
3490 * QEMU uses the return path.
3491 */
3492 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3493 if (open_return_path_on_source(s, !resume)) {
3494 error_report("Unable to open return-path for postcopy");
3495 migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3496 migrate_fd_cleanup(s);
3497 return;
3498 }
3499 }
3500
3501 if (resume) {
3502 /* Wakeup the main migration thread to do the recovery */
3503 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3504 MIGRATION_STATUS_POSTCOPY_RECOVER);
3505 qemu_sem_post(&s->postcopy_pause_sem);
3506 return;
3507 }
3508
3509 if (multifd_save_setup(&local_err) != 0) {
3510 error_report_err(local_err);
3511 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3512 MIGRATION_STATUS_FAILED);
3513 migrate_fd_cleanup(s);
3514 return;
3515 }
3516 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
3517 QEMU_THREAD_JOINABLE);
3518 s->migration_thread_running = true;
3519 }
3520
3521 void migration_global_dump(Monitor *mon)
3522 {
3523 MigrationState *ms = migrate_get_current();
3524
3525 monitor_printf(mon, "globals:\n");
3526 monitor_printf(mon, "store-global-state: %s\n",
3527 ms->store_global_state ? "on" : "off");
3528 monitor_printf(mon, "only-migratable: %s\n",
3529 only_migratable ? "on" : "off");
3530 monitor_printf(mon, "send-configuration: %s\n",
3531 ms->send_configuration ? "on" : "off");
3532 monitor_printf(mon, "send-section-footer: %s\n",
3533 ms->send_section_footer ? "on" : "off");
3534 monitor_printf(mon, "decompress-error-check: %s\n",
3535 ms->decompress_error_check ? "on" : "off");
3536 monitor_printf(mon, "clear-bitmap-shift: %u\n",
3537 ms->clear_bitmap_shift);
3538 }
3539
3540 #define DEFINE_PROP_MIG_CAP(name, x) \
3541 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3542
3543 static Property migration_properties[] = {
3544 DEFINE_PROP_BOOL("store-global-state", MigrationState,
3545 store_global_state, true),
3546 DEFINE_PROP_BOOL("send-configuration", MigrationState,
3547 send_configuration, true),
3548 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
3549 send_section_footer, true),
3550 DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
3551 decompress_error_check, true),
3552 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState,
3553 clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT),
3554
3555 /* Migration parameters */
3556 DEFINE_PROP_UINT8("x-compress-level", MigrationState,
3557 parameters.compress_level,
3558 DEFAULT_MIGRATE_COMPRESS_LEVEL),
3559 DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
3560 parameters.compress_threads,
3561 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
3562 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
3563 parameters.compress_wait_thread, true),
3564 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
3565 parameters.decompress_threads,
3566 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
3567 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
3568 parameters.cpu_throttle_initial,
3569 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
3570 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
3571 parameters.cpu_throttle_increment,
3572 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
3573 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
3574 parameters.max_bandwidth, MAX_THROTTLE),
3575 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
3576 parameters.downtime_limit,
3577 DEFAULT_MIGRATE_SET_DOWNTIME),
3578 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
3579 parameters.x_checkpoint_delay,
3580 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
3581 DEFINE_PROP_UINT8("multifd-channels", MigrationState,
3582 parameters.multifd_channels,
3583 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
3584 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState,
3585 parameters.multifd_compression,
3586 DEFAULT_MIGRATE_MULTIFD_COMPRESSION),
3587 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState,
3588 parameters.multifd_zlib_level,
3589 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL),
3590 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState,
3591 parameters.multifd_zstd_level,
3592 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL),
3593 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
3594 parameters.xbzrle_cache_size,
3595 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
3596 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
3597 parameters.max_postcopy_bandwidth,
3598 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
3599 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState,
3600 parameters.max_cpu_throttle,
3601 DEFAULT_MIGRATE_MAX_CPU_THROTTLE),
3602 DEFINE_PROP_SIZE("announce-initial", MigrationState,
3603 parameters.announce_initial,
3604 DEFAULT_MIGRATE_ANNOUNCE_INITIAL),
3605 DEFINE_PROP_SIZE("announce-max", MigrationState,
3606 parameters.announce_max,
3607 DEFAULT_MIGRATE_ANNOUNCE_MAX),
3608 DEFINE_PROP_SIZE("announce-rounds", MigrationState,
3609 parameters.announce_rounds,
3610 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS),
3611 DEFINE_PROP_SIZE("announce-step", MigrationState,
3612 parameters.announce_step,
3613 DEFAULT_MIGRATE_ANNOUNCE_STEP),
3614
3615 /* Migration capabilities */
3616 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
3617 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
3618 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
3619 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
3620 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
3621 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
3622 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
3623 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
3624 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
3625 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
3626 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
3627 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD),
3628
3629 DEFINE_PROP_END_OF_LIST(),
3630 };
3631
3632 static void migration_class_init(ObjectClass *klass, void *data)
3633 {
3634 DeviceClass *dc = DEVICE_CLASS(klass);
3635
3636 dc->user_creatable = false;
3637 device_class_set_props(dc, migration_properties);
3638 }
3639
3640 static void migration_instance_finalize(Object *obj)
3641 {
3642 MigrationState *ms = MIGRATION_OBJ(obj);
3643 MigrationParameters *params = &ms->parameters;
3644
3645 qemu_mutex_destroy(&ms->error_mutex);
3646 qemu_mutex_destroy(&ms->qemu_file_lock);
3647 g_free(params->tls_hostname);
3648 g_free(params->tls_creds);
3649 qemu_sem_destroy(&ms->wait_unplug_sem);
3650 qemu_sem_destroy(&ms->rate_limit_sem);
3651 qemu_sem_destroy(&ms->pause_sem);
3652 qemu_sem_destroy(&ms->postcopy_pause_sem);
3653 qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
3654 qemu_sem_destroy(&ms->rp_state.rp_sem);
3655 error_free(ms->error);
3656 }
3657
3658 static void migration_instance_init(Object *obj)
3659 {
3660 MigrationState *ms = MIGRATION_OBJ(obj);
3661 MigrationParameters *params = &ms->parameters;
3662
3663 ms->state = MIGRATION_STATUS_NONE;
3664 ms->mbps = -1;
3665 ms->pages_per_second = -1;
3666 qemu_sem_init(&ms->pause_sem, 0);
3667 qemu_mutex_init(&ms->error_mutex);
3668
3669 params->tls_hostname = g_strdup("");
3670 params->tls_creds = g_strdup("");
3671
3672 /* Set has_* up only for parameter checks */
3673 params->has_compress_level = true;
3674 params->has_compress_threads = true;
3675 params->has_decompress_threads = true;
3676 params->has_cpu_throttle_initial = true;
3677 params->has_cpu_throttle_increment = true;
3678 params->has_max_bandwidth = true;
3679 params->has_downtime_limit = true;
3680 params->has_x_checkpoint_delay = true;
3681 params->has_block_incremental = true;
3682 params->has_multifd_channels = true;
3683 params->has_multifd_compression = true;
3684 params->has_multifd_zlib_level = true;
3685 params->has_multifd_zstd_level = true;
3686 params->has_xbzrle_cache_size = true;
3687 params->has_max_postcopy_bandwidth = true;
3688 params->has_max_cpu_throttle = true;
3689 params->has_announce_initial = true;
3690 params->has_announce_max = true;
3691 params->has_announce_rounds = true;
3692 params->has_announce_step = true;
3693
3694 qemu_sem_init(&ms->postcopy_pause_sem, 0);
3695 qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
3696 qemu_sem_init(&ms->rp_state.rp_sem, 0);
3697 qemu_sem_init(&ms->rate_limit_sem, 0);
3698 qemu_sem_init(&ms->wait_unplug_sem, 0);
3699 qemu_mutex_init(&ms->qemu_file_lock);
3700 }
3701
3702 /*
3703 * Return true if check pass, false otherwise. Error will be put
3704 * inside errp if provided.
3705 */
3706 static bool migration_object_check(MigrationState *ms, Error **errp)
3707 {
3708 MigrationCapabilityStatusList *head = NULL;
3709 /* Assuming all off */
3710 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
3711 int i;
3712
3713 if (!migrate_params_check(&ms->parameters, errp)) {
3714 return false;
3715 }
3716
3717 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
3718 if (ms->enabled_capabilities[i]) {
3719 head = migrate_cap_add(head, i, true);
3720 }
3721 }
3722
3723 ret = migrate_caps_check(cap_list, head, errp);
3724
3725 /* It works with head == NULL */
3726 qapi_free_MigrationCapabilityStatusList(head);
3727
3728 return ret;
3729 }
3730
3731 static const TypeInfo migration_type = {
3732 .name = TYPE_MIGRATION,
3733 /*
3734 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3735 * not created using qdev_create(), it is not attached to the qdev
3736 * device tree, and it is never realized.
3737 *
3738 * TODO: Make this TYPE_OBJECT once QOM provides something like
3739 * TYPE_DEVICE's "-global" properties.
3740 */
3741 .parent = TYPE_DEVICE,
3742 .class_init = migration_class_init,
3743 .class_size = sizeof(MigrationClass),
3744 .instance_size = sizeof(MigrationState),
3745 .instance_init = migration_instance_init,
3746 .instance_finalize = migration_instance_finalize,
3747 };
3748
3749 static void register_migration_types(void)
3750 {
3751 type_register_static(&migration_type);
3752 }
3753
3754 type_init(register_migration_types);
Random patches