pc: fix auto_enable_numa_with_memhp/auto_enable_numa_with_memdev for the 5.0 machine
[qemu.git] / migration / migration.c
1 /*
2 * QEMU live migration
3 *
4 * Copyright IBM, Corp. 2008
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
14 */
15
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "migration/blocker.h"
21 #include "exec.h"
22 #include "fd.h"
23 #include "socket.h"
24 #include "sysemu/runstate.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/cpu-throttle.h"
27 #include "rdma.h"
28 #include "ram.h"
29 #include "migration/global_state.h"
30 #include "migration/misc.h"
31 #include "migration.h"
32 #include "savevm.h"
33 #include "qemu-file-channel.h"
34 #include "qemu-file.h"
35 #include "migration/vmstate.h"
36 #include "block/block.h"
37 #include "qapi/error.h"
38 #include "qapi/clone-visitor.h"
39 #include "qapi/qapi-visit-migration.h"
40 #include "qapi/qapi-visit-sockets.h"
41 #include "qapi/qapi-commands-migration.h"
42 #include "qapi/qapi-events-migration.h"
43 #include "qapi/qmp/qerror.h"
44 #include "qapi/qmp/qnull.h"
45 #include "qemu/rcu.h"
46 #include "block.h"
47 #include "postcopy-ram.h"
48 #include "qemu/thread.h"
49 #include "trace.h"
50 #include "exec/target_page.h"
51 #include "io/channel-buffer.h"
52 #include "migration/colo.h"
53 #include "hw/boards.h"
54 #include "hw/qdev-properties.h"
55 #include "monitor/monitor.h"
56 #include "net/announce.h"
57 #include "qemu/queue.h"
58 #include "multifd.h"
59
60 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
61
62 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
63 * data. */
64 #define BUFFER_DELAY 100
65 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
66
67 /* Time in milliseconds we are allowed to stop the source,
68 * for sending the last part */
69 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
70
71 /* Maximum migrate downtime set to 2000 seconds */
72 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
73 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
74
75 /* Default compression thread count */
76 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
77 /* Default decompression thread count, usually decompression is at
78 * least 4 times as fast as compression.*/
79 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
80 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
81 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
82 /* Define default autoconverge cpu throttle migration parameters */
83 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
84 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
85 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
86 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
87
88 /* Migration XBZRLE default cache size */
89 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
90
91 /* The delay time (in ms) between two COLO checkpoints */
92 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
93 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
94 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
95 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
96 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
97 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
98 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
99
100 /* Background transfer rate for postcopy, 0 means unlimited, note
101 * that page requests can still exceed this limit.
102 */
103 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
104
105 /*
106 * Parameters for self_announce_delay giving a stream of RARP/ARP
107 * packets after migration.
108 */
109 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
110 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
111 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
112 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
113
114 static NotifierList migration_state_notifiers =
115 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
116
117 static bool deferred_incoming;
118
119 /* Messages sent on the return path from destination to source */
120 enum mig_rp_message_type {
121 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
122 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
123 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
124
125 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
126 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
127 MIG_RP_MSG_RECV_BITMAP, /* send recved_bitmap back to source */
128 MIG_RP_MSG_RESUME_ACK, /* tell source that we are ready to resume */
129
130 MIG_RP_MSG_MAX
131 };
132
133 /* When we add fault tolerance, we could have several
134 migrations at once. For now we don't need to add
135 dynamic creation of migration */
136
137 static MigrationState *current_migration;
138 static MigrationIncomingState *current_incoming;
139
140 static bool migration_object_check(MigrationState *ms, Error **errp);
141 static int migration_maybe_pause(MigrationState *s,
142 int *current_active_state,
143 int new_state);
144 static void migrate_fd_cancel(MigrationState *s);
145
146 void migration_object_init(void)
147 {
148 MachineState *ms = MACHINE(qdev_get_machine());
149 Error *err = NULL;
150
151 /* This can only be called once. */
152 assert(!current_migration);
153 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
154
155 /*
156 * Init the migrate incoming object as well no matter whether
157 * we'll use it or not.
158 */
159 assert(!current_incoming);
160 current_incoming = g_new0(MigrationIncomingState, 1);
161 current_incoming->state = MIGRATION_STATUS_NONE;
162 current_incoming->postcopy_remote_fds =
163 g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
164 qemu_mutex_init(&current_incoming->rp_mutex);
165 qemu_event_init(&current_incoming->main_thread_load_event, false);
166 qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
167 qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
168
169 if (!migration_object_check(current_migration, &err)) {
170 error_report_err(err);
171 exit(1);
172 }
173
174 /*
175 * We cannot really do this in migration_instance_init() since at
176 * that time global properties are not yet applied, then this
177 * value will be definitely replaced by something else.
178 */
179 if (ms->enforce_config_section) {
180 current_migration->send_configuration = true;
181 }
182 }
183
184 void migration_shutdown(void)
185 {
186 /*
187 * Cancel the current migration - that will (eventually)
188 * stop the migration using this structure
189 */
190 migrate_fd_cancel(current_migration);
191 object_unref(OBJECT(current_migration));
192
193 /*
194 * Cancel outgoing migration of dirty bitmaps. It should
195 * at least unref used block nodes.
196 */
197 dirty_bitmap_mig_cancel_outgoing();
198
199 /*
200 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
201 * are non-critical data, and their loss never considered as
202 * something serious.
203 */
204 dirty_bitmap_mig_cancel_incoming();
205 }
206
207 /* For outgoing */
208 MigrationState *migrate_get_current(void)
209 {
210 /* This can only be called after the object created. */
211 assert(current_migration);
212 return current_migration;
213 }
214
215 MigrationIncomingState *migration_incoming_get_current(void)
216 {
217 assert(current_incoming);
218 return current_incoming;
219 }
220
221 void migration_incoming_state_destroy(void)
222 {
223 struct MigrationIncomingState *mis = migration_incoming_get_current();
224
225 if (mis->to_src_file) {
226 /* Tell source that we are done */
227 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
228 qemu_fclose(mis->to_src_file);
229 mis->to_src_file = NULL;
230 }
231
232 if (mis->from_src_file) {
233 qemu_fclose(mis->from_src_file);
234 mis->from_src_file = NULL;
235 }
236 if (mis->postcopy_remote_fds) {
237 g_array_free(mis->postcopy_remote_fds, TRUE);
238 mis->postcopy_remote_fds = NULL;
239 }
240
241 if (mis->socket_address_list) {
242 qapi_free_SocketAddressList(mis->socket_address_list);
243 mis->socket_address_list = NULL;
244 }
245
246 qemu_event_destroy(&mis->main_thread_load_event);
247 qemu_sem_destroy(&mis->postcopy_pause_sem_dst);
248 qemu_sem_destroy(&mis->postcopy_pause_sem_fault);
249 qemu_mutex_destroy(&mis->rp_mutex);
250 }
251
252 static void migrate_generate_event(int new_state)
253 {
254 if (migrate_use_events()) {
255 qapi_event_send_migration(new_state);
256 }
257 }
258
259 static bool migrate_late_block_activate(void)
260 {
261 MigrationState *s;
262
263 s = migrate_get_current();
264
265 return s->enabled_capabilities[
266 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
267 }
268
269 /*
270 * Called on -incoming with a defer: uri.
271 * The migration can be started later after any parameters have been
272 * changed.
273 */
274 static void deferred_incoming_migration(Error **errp)
275 {
276 if (deferred_incoming) {
277 error_setg(errp, "Incoming migration already deferred");
278 }
279 deferred_incoming = true;
280 }
281
282 /*
283 * Send a message on the return channel back to the source
284 * of the migration.
285 */
286 static int migrate_send_rp_message(MigrationIncomingState *mis,
287 enum mig_rp_message_type message_type,
288 uint16_t len, void *data)
289 {
290 int ret = 0;
291
292 trace_migrate_send_rp_message((int)message_type, len);
293 qemu_mutex_lock(&mis->rp_mutex);
294
295 /*
296 * It's possible that the file handle got lost due to network
297 * failures.
298 */
299 if (!mis->to_src_file) {
300 ret = -EIO;
301 goto error;
302 }
303
304 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
305 qemu_put_be16(mis->to_src_file, len);
306 qemu_put_buffer(mis->to_src_file, data, len);
307 qemu_fflush(mis->to_src_file);
308
309 /* It's possible that qemu file got error during sending */
310 ret = qemu_file_get_error(mis->to_src_file);
311
312 error:
313 qemu_mutex_unlock(&mis->rp_mutex);
314 return ret;
315 }
316
317 /* Request one page from the source VM at the given start address.
318 * rb: the RAMBlock to request the page in
319 * Start: Address offset within the RB
320 * Len: Length in bytes required - must be a multiple of pagesize
321 */
322 int migrate_send_rp_req_pages(MigrationIncomingState *mis, RAMBlock *rb,
323 ram_addr_t start)
324 {
325 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
326 size_t msglen = 12; /* start + len */
327 size_t len = qemu_ram_pagesize(rb);
328 enum mig_rp_message_type msg_type;
329 const char *rbname;
330 int rbname_len;
331
332 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
333 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
334
335 /*
336 * We maintain the last ramblock that we requested for page. Note that we
337 * don't need locking because this function will only be called within the
338 * postcopy ram fault thread.
339 */
340 if (rb != mis->last_rb) {
341 mis->last_rb = rb;
342
343 rbname = qemu_ram_get_idstr(rb);
344 rbname_len = strlen(rbname);
345
346 assert(rbname_len < 256);
347
348 bufc[msglen++] = rbname_len;
349 memcpy(bufc + msglen, rbname, rbname_len);
350 msglen += rbname_len;
351 msg_type = MIG_RP_MSG_REQ_PAGES_ID;
352 } else {
353 msg_type = MIG_RP_MSG_REQ_PAGES;
354 }
355
356 return migrate_send_rp_message(mis, msg_type, msglen, bufc);
357 }
358
359 static bool migration_colo_enabled;
360 bool migration_incoming_colo_enabled(void)
361 {
362 return migration_colo_enabled;
363 }
364
365 void migration_incoming_disable_colo(void)
366 {
367 ram_block_discard_disable(false);
368 migration_colo_enabled = false;
369 }
370
371 int migration_incoming_enable_colo(void)
372 {
373 if (ram_block_discard_disable(true)) {
374 error_report("COLO: cannot disable RAM discard");
375 return -EBUSY;
376 }
377 migration_colo_enabled = true;
378 return 0;
379 }
380
381 void migrate_add_address(SocketAddress *address)
382 {
383 MigrationIncomingState *mis = migration_incoming_get_current();
384 SocketAddressList *addrs;
385
386 addrs = g_new0(SocketAddressList, 1);
387 addrs->next = mis->socket_address_list;
388 mis->socket_address_list = addrs;
389 addrs->value = QAPI_CLONE(SocketAddress, address);
390 }
391
392 void qemu_start_incoming_migration(const char *uri, Error **errp)
393 {
394 const char *p = NULL;
395
396 qapi_event_send_migration(MIGRATION_STATUS_SETUP);
397 if (!strcmp(uri, "defer")) {
398 deferred_incoming_migration(errp);
399 } else if (strstart(uri, "tcp:", &p) ||
400 strstart(uri, "unix:", NULL) ||
401 strstart(uri, "vsock:", NULL)) {
402 socket_start_incoming_migration(p ? p : uri, errp);
403 #ifdef CONFIG_RDMA
404 } else if (strstart(uri, "rdma:", &p)) {
405 rdma_start_incoming_migration(p, errp);
406 #endif
407 } else if (strstart(uri, "exec:", &p)) {
408 exec_start_incoming_migration(p, errp);
409 } else if (strstart(uri, "fd:", &p)) {
410 fd_start_incoming_migration(p, errp);
411 } else {
412 error_setg(errp, "unknown migration protocol: %s", uri);
413 }
414 }
415
416 static void process_incoming_migration_bh(void *opaque)
417 {
418 Error *local_err = NULL;
419 MigrationIncomingState *mis = opaque;
420
421 /* If capability late_block_activate is set:
422 * Only fire up the block code now if we're going to restart the
423 * VM, else 'cont' will do it.
424 * This causes file locking to happen; so we don't want it to happen
425 * unless we really are starting the VM.
426 */
427 if (!migrate_late_block_activate() ||
428 (autostart && (!global_state_received() ||
429 global_state_get_runstate() == RUN_STATE_RUNNING))) {
430 /* Make sure all file formats flush their mutable metadata.
431 * If we get an error here, just don't restart the VM yet. */
432 bdrv_invalidate_cache_all(&local_err);
433 if (local_err) {
434 error_report_err(local_err);
435 local_err = NULL;
436 autostart = false;
437 }
438 }
439
440 /*
441 * This must happen after all error conditions are dealt with and
442 * we're sure the VM is going to be running on this host.
443 */
444 qemu_announce_self(&mis->announce_timer, migrate_announce_params());
445
446 if (multifd_load_cleanup(&local_err) != 0) {
447 error_report_err(local_err);
448 autostart = false;
449 }
450 /* If global state section was not received or we are in running
451 state, we need to obey autostart. Any other state is set with
452 runstate_set. */
453
454 dirty_bitmap_mig_before_vm_start();
455
456 if (!global_state_received() ||
457 global_state_get_runstate() == RUN_STATE_RUNNING) {
458 if (autostart) {
459 vm_start();
460 } else {
461 runstate_set(RUN_STATE_PAUSED);
462 }
463 } else if (migration_incoming_colo_enabled()) {
464 migration_incoming_disable_colo();
465 vm_start();
466 } else {
467 runstate_set(global_state_get_runstate());
468 }
469 /*
470 * This must happen after any state changes since as soon as an external
471 * observer sees this event they might start to prod at the VM assuming
472 * it's ready to use.
473 */
474 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
475 MIGRATION_STATUS_COMPLETED);
476 qemu_bh_delete(mis->bh);
477 migration_incoming_state_destroy();
478 }
479
480 static void process_incoming_migration_co(void *opaque)
481 {
482 MigrationIncomingState *mis = migration_incoming_get_current();
483 PostcopyState ps;
484 int ret;
485 Error *local_err = NULL;
486
487 assert(mis->from_src_file);
488 mis->migration_incoming_co = qemu_coroutine_self();
489 mis->largest_page_size = qemu_ram_pagesize_largest();
490 postcopy_state_set(POSTCOPY_INCOMING_NONE);
491 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
492 MIGRATION_STATUS_ACTIVE);
493 ret = qemu_loadvm_state(mis->from_src_file);
494
495 ps = postcopy_state_get();
496 trace_process_incoming_migration_co_end(ret, ps);
497 if (ps != POSTCOPY_INCOMING_NONE) {
498 if (ps == POSTCOPY_INCOMING_ADVISE) {
499 /*
500 * Where a migration had postcopy enabled (and thus went to advise)
501 * but managed to complete within the precopy period, we can use
502 * the normal exit.
503 */
504 postcopy_ram_incoming_cleanup(mis);
505 } else if (ret >= 0) {
506 /*
507 * Postcopy was started, cleanup should happen at the end of the
508 * postcopy thread.
509 */
510 trace_process_incoming_migration_co_postcopy_end_main();
511 return;
512 }
513 /* Else if something went wrong then just fall out of the normal exit */
514 }
515
516 /* we get COLO info, and know if we are in COLO mode */
517 if (!ret && migration_incoming_colo_enabled()) {
518 /* Make sure all file formats flush their mutable metadata */
519 bdrv_invalidate_cache_all(&local_err);
520 if (local_err) {
521 error_report_err(local_err);
522 goto fail;
523 }
524
525 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
526 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
527 mis->have_colo_incoming_thread = true;
528 qemu_coroutine_yield();
529
530 /* Wait checkpoint incoming thread exit before free resource */
531 qemu_thread_join(&mis->colo_incoming_thread);
532 /* We hold the global iothread lock, so it is safe here */
533 colo_release_ram_cache();
534 }
535
536 if (ret < 0) {
537 error_report("load of migration failed: %s", strerror(-ret));
538 goto fail;
539 }
540 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
541 qemu_bh_schedule(mis->bh);
542 mis->migration_incoming_co = NULL;
543 return;
544 fail:
545 local_err = NULL;
546 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
547 MIGRATION_STATUS_FAILED);
548 qemu_fclose(mis->from_src_file);
549 if (multifd_load_cleanup(&local_err) != 0) {
550 error_report_err(local_err);
551 }
552 exit(EXIT_FAILURE);
553 }
554
555 /**
556 * @migration_incoming_setup: Setup incoming migration
557 *
558 * Returns 0 for no error or 1 for error
559 *
560 * @f: file for main migration channel
561 * @errp: where to put errors
562 */
563 static int migration_incoming_setup(QEMUFile *f, Error **errp)
564 {
565 MigrationIncomingState *mis = migration_incoming_get_current();
566 Error *local_err = NULL;
567
568 if (multifd_load_setup(&local_err) != 0) {
569 /* We haven't been able to create multifd threads
570 nothing better to do */
571 error_report_err(local_err);
572 exit(EXIT_FAILURE);
573 }
574
575 if (!mis->from_src_file) {
576 mis->from_src_file = f;
577 }
578 qemu_file_set_blocking(f, false);
579 return 0;
580 }
581
582 void migration_incoming_process(void)
583 {
584 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
585 qemu_coroutine_enter(co);
586 }
587
588 /* Returns true if recovered from a paused migration, otherwise false */
589 static bool postcopy_try_recover(QEMUFile *f)
590 {
591 MigrationIncomingState *mis = migration_incoming_get_current();
592
593 if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
594 /* Resumed from a paused postcopy migration */
595
596 mis->from_src_file = f;
597 /* Postcopy has standalone thread to do vm load */
598 qemu_file_set_blocking(f, true);
599
600 /* Re-configure the return path */
601 mis->to_src_file = qemu_file_get_return_path(f);
602
603 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
604 MIGRATION_STATUS_POSTCOPY_RECOVER);
605
606 /*
607 * Here, we only wake up the main loading thread (while the
608 * fault thread will still be waiting), so that we can receive
609 * commands from source now, and answer it if needed. The
610 * fault thread will be woken up afterwards until we are sure
611 * that source is ready to reply to page requests.
612 */
613 qemu_sem_post(&mis->postcopy_pause_sem_dst);
614 return true;
615 }
616
617 return false;
618 }
619
620 void migration_fd_process_incoming(QEMUFile *f, Error **errp)
621 {
622 Error *local_err = NULL;
623
624 if (postcopy_try_recover(f)) {
625 return;
626 }
627
628 if (migration_incoming_setup(f, &local_err)) {
629 if (local_err) {
630 error_propagate(errp, local_err);
631 }
632 return;
633 }
634 migration_incoming_process();
635 }
636
637 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
638 {
639 MigrationIncomingState *mis = migration_incoming_get_current();
640 Error *local_err = NULL;
641 bool start_migration;
642
643 if (!mis->from_src_file) {
644 /* The first connection (multifd may have multiple) */
645 QEMUFile *f = qemu_fopen_channel_input(ioc);
646
647 /* If it's a recovery, we're done */
648 if (postcopy_try_recover(f)) {
649 return;
650 }
651
652 if (migration_incoming_setup(f, &local_err)) {
653 if (local_err) {
654 error_propagate(errp, local_err);
655 }
656 return;
657 }
658
659 /*
660 * Common migration only needs one channel, so we can start
661 * right now. Multifd needs more than one channel, we wait.
662 */
663 start_migration = !migrate_use_multifd();
664 } else {
665 /* Multiple connections */
666 assert(migrate_use_multifd());
667 start_migration = multifd_recv_new_channel(ioc, &local_err);
668 if (local_err) {
669 error_propagate(errp, local_err);
670 return;
671 }
672 }
673
674 if (start_migration) {
675 migration_incoming_process();
676 }
677 }
678
679 /**
680 * @migration_has_all_channels: We have received all channels that we need
681 *
682 * Returns true when we have got connections to all the channels that
683 * we need for migration.
684 */
685 bool migration_has_all_channels(void)
686 {
687 MigrationIncomingState *mis = migration_incoming_get_current();
688 bool all_channels;
689
690 all_channels = multifd_recv_all_channels_created();
691
692 return all_channels && mis->from_src_file != NULL;
693 }
694
695 /*
696 * Send a 'SHUT' message on the return channel with the given value
697 * to indicate that we've finished with the RP. Non-0 value indicates
698 * error.
699 */
700 void migrate_send_rp_shut(MigrationIncomingState *mis,
701 uint32_t value)
702 {
703 uint32_t buf;
704
705 buf = cpu_to_be32(value);
706 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
707 }
708
709 /*
710 * Send a 'PONG' message on the return channel with the given value
711 * (normally in response to a 'PING')
712 */
713 void migrate_send_rp_pong(MigrationIncomingState *mis,
714 uint32_t value)
715 {
716 uint32_t buf;
717
718 buf = cpu_to_be32(value);
719 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
720 }
721
722 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
723 char *block_name)
724 {
725 char buf[512];
726 int len;
727 int64_t res;
728
729 /*
730 * First, we send the header part. It contains only the len of
731 * idstr, and the idstr itself.
732 */
733 len = strlen(block_name);
734 buf[0] = len;
735 memcpy(buf + 1, block_name, len);
736
737 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
738 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
739 __func__);
740 return;
741 }
742
743 migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
744
745 /*
746 * Next, we dump the received bitmap to the stream.
747 *
748 * TODO: currently we are safe since we are the only one that is
749 * using the to_src_file handle (fault thread is still paused),
750 * and it's ok even not taking the mutex. However the best way is
751 * to take the lock before sending the message header, and release
752 * the lock after sending the bitmap.
753 */
754 qemu_mutex_lock(&mis->rp_mutex);
755 res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
756 qemu_mutex_unlock(&mis->rp_mutex);
757
758 trace_migrate_send_rp_recv_bitmap(block_name, res);
759 }
760
761 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
762 {
763 uint32_t buf;
764
765 buf = cpu_to_be32(value);
766 migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
767 }
768
769 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
770 {
771 MigrationCapabilityStatusList *head = NULL;
772 MigrationCapabilityStatusList *caps;
773 MigrationState *s = migrate_get_current();
774 int i;
775
776 caps = NULL; /* silence compiler warning */
777 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
778 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
779 if (i == MIGRATION_CAPABILITY_BLOCK) {
780 continue;
781 }
782 #endif
783 if (head == NULL) {
784 head = g_malloc0(sizeof(*caps));
785 caps = head;
786 } else {
787 caps->next = g_malloc0(sizeof(*caps));
788 caps = caps->next;
789 }
790 caps->value =
791 g_malloc(sizeof(*caps->value));
792 caps->value->capability = i;
793 caps->value->state = s->enabled_capabilities[i];
794 }
795
796 return head;
797 }
798
799 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
800 {
801 MigrationParameters *params;
802 MigrationState *s = migrate_get_current();
803
804 /* TODO use QAPI_CLONE() instead of duplicating it inline */
805 params = g_malloc0(sizeof(*params));
806 params->has_compress_level = true;
807 params->compress_level = s->parameters.compress_level;
808 params->has_compress_threads = true;
809 params->compress_threads = s->parameters.compress_threads;
810 params->has_compress_wait_thread = true;
811 params->compress_wait_thread = s->parameters.compress_wait_thread;
812 params->has_decompress_threads = true;
813 params->decompress_threads = s->parameters.decompress_threads;
814 params->has_throttle_trigger_threshold = true;
815 params->throttle_trigger_threshold = s->parameters.throttle_trigger_threshold;
816 params->has_cpu_throttle_initial = true;
817 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
818 params->has_cpu_throttle_increment = true;
819 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
820 params->has_cpu_throttle_tailslow = true;
821 params->cpu_throttle_tailslow = s->parameters.cpu_throttle_tailslow;
822 params->has_tls_creds = true;
823 params->tls_creds = g_strdup(s->parameters.tls_creds);
824 params->has_tls_hostname = true;
825 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
826 params->has_tls_authz = true;
827 params->tls_authz = g_strdup(s->parameters.tls_authz ?
828 s->parameters.tls_authz : "");
829 params->has_max_bandwidth = true;
830 params->max_bandwidth = s->parameters.max_bandwidth;
831 params->has_downtime_limit = true;
832 params->downtime_limit = s->parameters.downtime_limit;
833 params->has_x_checkpoint_delay = true;
834 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
835 params->has_block_incremental = true;
836 params->block_incremental = s->parameters.block_incremental;
837 params->has_multifd_channels = true;
838 params->multifd_channels = s->parameters.multifd_channels;
839 params->has_multifd_compression = true;
840 params->multifd_compression = s->parameters.multifd_compression;
841 params->has_multifd_zlib_level = true;
842 params->multifd_zlib_level = s->parameters.multifd_zlib_level;
843 params->has_multifd_zstd_level = true;
844 params->multifd_zstd_level = s->parameters.multifd_zstd_level;
845 params->has_xbzrle_cache_size = true;
846 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
847 params->has_max_postcopy_bandwidth = true;
848 params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
849 params->has_max_cpu_throttle = true;
850 params->max_cpu_throttle = s->parameters.max_cpu_throttle;
851 params->has_announce_initial = true;
852 params->announce_initial = s->parameters.announce_initial;
853 params->has_announce_max = true;
854 params->announce_max = s->parameters.announce_max;
855 params->has_announce_rounds = true;
856 params->announce_rounds = s->parameters.announce_rounds;
857 params->has_announce_step = true;
858 params->announce_step = s->parameters.announce_step;
859
860 if (s->parameters.has_block_bitmap_mapping) {
861 params->has_block_bitmap_mapping = true;
862 params->block_bitmap_mapping =
863 QAPI_CLONE(BitmapMigrationNodeAliasList,
864 s->parameters.block_bitmap_mapping);
865 }
866
867 return params;
868 }
869
870 AnnounceParameters *migrate_announce_params(void)
871 {
872 static AnnounceParameters ap;
873
874 MigrationState *s = migrate_get_current();
875
876 ap.initial = s->parameters.announce_initial;
877 ap.max = s->parameters.announce_max;
878 ap.rounds = s->parameters.announce_rounds;
879 ap.step = s->parameters.announce_step;
880
881 return &ap;
882 }
883
884 /*
885 * Return true if we're already in the middle of a migration
886 * (i.e. any of the active or setup states)
887 */
888 bool migration_is_setup_or_active(int state)
889 {
890 switch (state) {
891 case MIGRATION_STATUS_ACTIVE:
892 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
893 case MIGRATION_STATUS_POSTCOPY_PAUSED:
894 case MIGRATION_STATUS_POSTCOPY_RECOVER:
895 case MIGRATION_STATUS_SETUP:
896 case MIGRATION_STATUS_PRE_SWITCHOVER:
897 case MIGRATION_STATUS_DEVICE:
898 case MIGRATION_STATUS_WAIT_UNPLUG:
899 case MIGRATION_STATUS_COLO:
900 return true;
901
902 default:
903 return false;
904
905 }
906 }
907
908 bool migration_is_running(int state)
909 {
910 switch (state) {
911 case MIGRATION_STATUS_ACTIVE:
912 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
913 case MIGRATION_STATUS_POSTCOPY_PAUSED:
914 case MIGRATION_STATUS_POSTCOPY_RECOVER:
915 case MIGRATION_STATUS_SETUP:
916 case MIGRATION_STATUS_PRE_SWITCHOVER:
917 case MIGRATION_STATUS_DEVICE:
918 case MIGRATION_STATUS_WAIT_UNPLUG:
919 case MIGRATION_STATUS_CANCELLING:
920 return true;
921
922 default:
923 return false;
924
925 }
926 }
927
928 static void populate_time_info(MigrationInfo *info, MigrationState *s)
929 {
930 info->has_status = true;
931 info->has_setup_time = true;
932 info->setup_time = s->setup_time;
933 if (s->state == MIGRATION_STATUS_COMPLETED) {
934 info->has_total_time = true;
935 info->total_time = s->total_time;
936 info->has_downtime = true;
937 info->downtime = s->downtime;
938 } else {
939 info->has_total_time = true;
940 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
941 s->start_time;
942 info->has_expected_downtime = true;
943 info->expected_downtime = s->expected_downtime;
944 }
945 }
946
947 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
948 {
949 info->has_ram = true;
950 info->ram = g_malloc0(sizeof(*info->ram));
951 info->ram->transferred = ram_counters.transferred;
952 info->ram->total = ram_bytes_total();
953 info->ram->duplicate = ram_counters.duplicate;
954 /* legacy value. It is not used anymore */
955 info->ram->skipped = 0;
956 info->ram->normal = ram_counters.normal;
957 info->ram->normal_bytes = ram_counters.normal *
958 qemu_target_page_size();
959 info->ram->mbps = s->mbps;
960 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
961 info->ram->postcopy_requests = ram_counters.postcopy_requests;
962 info->ram->page_size = qemu_target_page_size();
963 info->ram->multifd_bytes = ram_counters.multifd_bytes;
964 info->ram->pages_per_second = s->pages_per_second;
965
966 if (migrate_use_xbzrle()) {
967 info->has_xbzrle_cache = true;
968 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
969 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
970 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
971 info->xbzrle_cache->pages = xbzrle_counters.pages;
972 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
973 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
974 info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate;
975 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
976 }
977
978 if (migrate_use_compression()) {
979 info->has_compression = true;
980 info->compression = g_malloc0(sizeof(*info->compression));
981 info->compression->pages = compression_counters.pages;
982 info->compression->busy = compression_counters.busy;
983 info->compression->busy_rate = compression_counters.busy_rate;
984 info->compression->compressed_size =
985 compression_counters.compressed_size;
986 info->compression->compression_rate =
987 compression_counters.compression_rate;
988 }
989
990 if (cpu_throttle_active()) {
991 info->has_cpu_throttle_percentage = true;
992 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
993 }
994
995 if (s->state != MIGRATION_STATUS_COMPLETED) {
996 info->ram->remaining = ram_bytes_remaining();
997 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
998 }
999 }
1000
1001 static void populate_disk_info(MigrationInfo *info)
1002 {
1003 if (blk_mig_active()) {
1004 info->has_disk = true;
1005 info->disk = g_malloc0(sizeof(*info->disk));
1006 info->disk->transferred = blk_mig_bytes_transferred();
1007 info->disk->remaining = blk_mig_bytes_remaining();
1008 info->disk->total = blk_mig_bytes_total();
1009 }
1010 }
1011
1012 static void fill_source_migration_info(MigrationInfo *info)
1013 {
1014 MigrationState *s = migrate_get_current();
1015
1016 switch (s->state) {
1017 case MIGRATION_STATUS_NONE:
1018 /* no migration has happened ever */
1019 /* do not overwrite destination migration status */
1020 return;
1021 case MIGRATION_STATUS_SETUP:
1022 info->has_status = true;
1023 info->has_total_time = false;
1024 break;
1025 case MIGRATION_STATUS_ACTIVE:
1026 case MIGRATION_STATUS_CANCELLING:
1027 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1028 case MIGRATION_STATUS_PRE_SWITCHOVER:
1029 case MIGRATION_STATUS_DEVICE:
1030 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1031 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1032 /* TODO add some postcopy stats */
1033 populate_time_info(info, s);
1034 populate_ram_info(info, s);
1035 populate_disk_info(info);
1036 break;
1037 case MIGRATION_STATUS_COLO:
1038 info->has_status = true;
1039 /* TODO: display COLO specific information (checkpoint info etc.) */
1040 break;
1041 case MIGRATION_STATUS_COMPLETED:
1042 populate_time_info(info, s);
1043 populate_ram_info(info, s);
1044 break;
1045 case MIGRATION_STATUS_FAILED:
1046 info->has_status = true;
1047 if (s->error) {
1048 info->has_error_desc = true;
1049 info->error_desc = g_strdup(error_get_pretty(s->error));
1050 }
1051 break;
1052 case MIGRATION_STATUS_CANCELLED:
1053 info->has_status = true;
1054 break;
1055 case MIGRATION_STATUS_WAIT_UNPLUG:
1056 info->has_status = true;
1057 break;
1058 }
1059 info->status = s->state;
1060 }
1061
1062 /**
1063 * @migration_caps_check - check capability validity
1064 *
1065 * @cap_list: old capability list, array of bool
1066 * @params: new capabilities to be applied soon
1067 * @errp: set *errp if the check failed, with reason
1068 *
1069 * Returns true if check passed, otherwise false.
1070 */
1071 static bool migrate_caps_check(bool *cap_list,
1072 MigrationCapabilityStatusList *params,
1073 Error **errp)
1074 {
1075 MigrationCapabilityStatusList *cap;
1076 bool old_postcopy_cap;
1077 MigrationIncomingState *mis = migration_incoming_get_current();
1078
1079 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1080
1081 for (cap = params; cap; cap = cap->next) {
1082 cap_list[cap->value->capability] = cap->value->state;
1083 }
1084
1085 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1086 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
1087 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
1088 "block migration");
1089 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
1090 return false;
1091 }
1092 #endif
1093
1094 #ifndef CONFIG_REPLICATION
1095 if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
1096 error_setg(errp, "QEMU compiled without replication module"
1097 " can't enable COLO");
1098 error_append_hint(errp, "Please enable replication before COLO.\n");
1099 return false;
1100 }
1101 #endif
1102
1103 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
1104 /* This check is reasonably expensive, so only when it's being
1105 * set the first time, also it's only the destination that needs
1106 * special support.
1107 */
1108 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1109 !postcopy_ram_supported_by_host(mis)) {
1110 /* postcopy_ram_supported_by_host will have emitted a more
1111 * detailed message
1112 */
1113 error_setg(errp, "Postcopy is not supported");
1114 return false;
1115 }
1116
1117 if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1118 error_setg(errp, "Postcopy is not compatible with ignore-shared");
1119 return false;
1120 }
1121 }
1122
1123 return true;
1124 }
1125
1126 static void fill_destination_migration_info(MigrationInfo *info)
1127 {
1128 MigrationIncomingState *mis = migration_incoming_get_current();
1129
1130 if (mis->socket_address_list) {
1131 info->has_socket_address = true;
1132 info->socket_address =
1133 QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1134 }
1135
1136 switch (mis->state) {
1137 case MIGRATION_STATUS_NONE:
1138 return;
1139 case MIGRATION_STATUS_SETUP:
1140 case MIGRATION_STATUS_CANCELLING:
1141 case MIGRATION_STATUS_CANCELLED:
1142 case MIGRATION_STATUS_ACTIVE:
1143 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1144 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1145 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1146 case MIGRATION_STATUS_FAILED:
1147 case MIGRATION_STATUS_COLO:
1148 info->has_status = true;
1149 break;
1150 case MIGRATION_STATUS_COMPLETED:
1151 info->has_status = true;
1152 fill_destination_postcopy_migration_info(info);
1153 break;
1154 }
1155 info->status = mis->state;
1156 }
1157
1158 MigrationInfo *qmp_query_migrate(Error **errp)
1159 {
1160 MigrationInfo *info = g_malloc0(sizeof(*info));
1161
1162 fill_destination_migration_info(info);
1163 fill_source_migration_info(info);
1164
1165 return info;
1166 }
1167
1168 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1169 Error **errp)
1170 {
1171 MigrationState *s = migrate_get_current();
1172 MigrationCapabilityStatusList *cap;
1173 bool cap_list[MIGRATION_CAPABILITY__MAX];
1174
1175 if (migration_is_running(s->state)) {
1176 error_setg(errp, QERR_MIGRATION_ACTIVE);
1177 return;
1178 }
1179
1180 memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1181 if (!migrate_caps_check(cap_list, params, errp)) {
1182 return;
1183 }
1184
1185 for (cap = params; cap; cap = cap->next) {
1186 s->enabled_capabilities[cap->value->capability] = cap->value->state;
1187 }
1188 }
1189
1190 /*
1191 * Check whether the parameters are valid. Error will be put into errp
1192 * (if provided). Return true if valid, otherwise false.
1193 */
1194 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1195 {
1196 if (params->has_compress_level &&
1197 (params->compress_level > 9)) {
1198 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1199 "is invalid, it should be in the range of 0 to 9");
1200 return false;
1201 }
1202
1203 if (params->has_compress_threads && (params->compress_threads < 1)) {
1204 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1205 "compress_threads",
1206 "is invalid, it should be in the range of 1 to 255");
1207 return false;
1208 }
1209
1210 if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1211 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1212 "decompress_threads",
1213 "is invalid, it should be in the range of 1 to 255");
1214 return false;
1215 }
1216
1217 if (params->has_throttle_trigger_threshold &&
1218 (params->throttle_trigger_threshold < 1 ||
1219 params->throttle_trigger_threshold > 100)) {
1220 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1221 "throttle_trigger_threshold",
1222 "an integer in the range of 1 to 100");
1223 return false;
1224 }
1225
1226 if (params->has_cpu_throttle_initial &&
1227 (params->cpu_throttle_initial < 1 ||
1228 params->cpu_throttle_initial > 99)) {
1229 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1230 "cpu_throttle_initial",
1231 "an integer in the range of 1 to 99");
1232 return false;
1233 }
1234
1235 if (params->has_cpu_throttle_increment &&
1236 (params->cpu_throttle_increment < 1 ||
1237 params->cpu_throttle_increment > 99)) {
1238 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1239 "cpu_throttle_increment",
1240 "an integer in the range of 1 to 99");
1241 return false;
1242 }
1243
1244 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1245 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1246 "max_bandwidth",
1247 "an integer in the range of 0 to "stringify(SIZE_MAX)
1248 " bytes/second");
1249 return false;
1250 }
1251
1252 if (params->has_downtime_limit &&
1253 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1254 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1255 "downtime_limit",
1256 "an integer in the range of 0 to "
1257 stringify(MAX_MIGRATE_DOWNTIME)" ms");
1258 return false;
1259 }
1260
1261 /* x_checkpoint_delay is now always positive */
1262
1263 if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1264 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1265 "multifd_channels",
1266 "is invalid, it should be in the range of 1 to 255");
1267 return false;
1268 }
1269
1270 if (params->has_multifd_zlib_level &&
1271 (params->multifd_zlib_level > 9)) {
1272 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level",
1273 "is invalid, it should be in the range of 0 to 9");
1274 return false;
1275 }
1276
1277 if (params->has_multifd_zstd_level &&
1278 (params->multifd_zstd_level > 20)) {
1279 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level",
1280 "is invalid, it should be in the range of 0 to 20");
1281 return false;
1282 }
1283
1284 if (params->has_xbzrle_cache_size &&
1285 (params->xbzrle_cache_size < qemu_target_page_size() ||
1286 !is_power_of_2(params->xbzrle_cache_size))) {
1287 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1288 "xbzrle_cache_size",
1289 "is invalid, it should be bigger than target page size"
1290 " and a power of 2");
1291 return false;
1292 }
1293
1294 if (params->has_max_cpu_throttle &&
1295 (params->max_cpu_throttle < params->cpu_throttle_initial ||
1296 params->max_cpu_throttle > 99)) {
1297 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1298 "max_cpu_throttle",
1299 "an integer in the range of cpu_throttle_initial to 99");
1300 return false;
1301 }
1302
1303 if (params->has_announce_initial &&
1304 params->announce_initial > 100000) {
1305 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1306 "announce_initial",
1307 "is invalid, it must be less than 100000 ms");
1308 return false;
1309 }
1310 if (params->has_announce_max &&
1311 params->announce_max > 100000) {
1312 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1313 "announce_max",
1314 "is invalid, it must be less than 100000 ms");
1315 return false;
1316 }
1317 if (params->has_announce_rounds &&
1318 params->announce_rounds > 1000) {
1319 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1320 "announce_rounds",
1321 "is invalid, it must be in the range of 0 to 1000");
1322 return false;
1323 }
1324 if (params->has_announce_step &&
1325 (params->announce_step < 1 ||
1326 params->announce_step > 10000)) {
1327 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1328 "announce_step",
1329 "is invalid, it must be in the range of 1 to 10000 ms");
1330 return false;
1331 }
1332
1333 if (params->has_block_bitmap_mapping &&
1334 !check_dirty_bitmap_mig_alias_map(params->block_bitmap_mapping, errp)) {
1335 error_prepend(errp, "Invalid mapping given for block-bitmap-mapping: ");
1336 return false;
1337 }
1338
1339 return true;
1340 }
1341
1342 static void migrate_params_test_apply(MigrateSetParameters *params,
1343 MigrationParameters *dest)
1344 {
1345 *dest = migrate_get_current()->parameters;
1346
1347 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1348
1349 if (params->has_compress_level) {
1350 dest->compress_level = params->compress_level;
1351 }
1352
1353 if (params->has_compress_threads) {
1354 dest->compress_threads = params->compress_threads;
1355 }
1356
1357 if (params->has_compress_wait_thread) {
1358 dest->compress_wait_thread = params->compress_wait_thread;
1359 }
1360
1361 if (params->has_decompress_threads) {
1362 dest->decompress_threads = params->decompress_threads;
1363 }
1364
1365 if (params->has_throttle_trigger_threshold) {
1366 dest->throttle_trigger_threshold = params->throttle_trigger_threshold;
1367 }
1368
1369 if (params->has_cpu_throttle_initial) {
1370 dest->cpu_throttle_initial = params->cpu_throttle_initial;
1371 }
1372
1373 if (params->has_cpu_throttle_increment) {
1374 dest->cpu_throttle_increment = params->cpu_throttle_increment;
1375 }
1376
1377 if (params->has_cpu_throttle_tailslow) {
1378 dest->cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1379 }
1380
1381 if (params->has_tls_creds) {
1382 assert(params->tls_creds->type == QTYPE_QSTRING);
1383 dest->tls_creds = params->tls_creds->u.s;
1384 }
1385
1386 if (params->has_tls_hostname) {
1387 assert(params->tls_hostname->type == QTYPE_QSTRING);
1388 dest->tls_hostname = params->tls_hostname->u.s;
1389 }
1390
1391 if (params->has_max_bandwidth) {
1392 dest->max_bandwidth = params->max_bandwidth;
1393 }
1394
1395 if (params->has_downtime_limit) {
1396 dest->downtime_limit = params->downtime_limit;
1397 }
1398
1399 if (params->has_x_checkpoint_delay) {
1400 dest->x_checkpoint_delay = params->x_checkpoint_delay;
1401 }
1402
1403 if (params->has_block_incremental) {
1404 dest->block_incremental = params->block_incremental;
1405 }
1406 if (params->has_multifd_channels) {
1407 dest->multifd_channels = params->multifd_channels;
1408 }
1409 if (params->has_multifd_compression) {
1410 dest->multifd_compression = params->multifd_compression;
1411 }
1412 if (params->has_xbzrle_cache_size) {
1413 dest->xbzrle_cache_size = params->xbzrle_cache_size;
1414 }
1415 if (params->has_max_postcopy_bandwidth) {
1416 dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1417 }
1418 if (params->has_max_cpu_throttle) {
1419 dest->max_cpu_throttle = params->max_cpu_throttle;
1420 }
1421 if (params->has_announce_initial) {
1422 dest->announce_initial = params->announce_initial;
1423 }
1424 if (params->has_announce_max) {
1425 dest->announce_max = params->announce_max;
1426 }
1427 if (params->has_announce_rounds) {
1428 dest->announce_rounds = params->announce_rounds;
1429 }
1430 if (params->has_announce_step) {
1431 dest->announce_step = params->announce_step;
1432 }
1433
1434 if (params->has_block_bitmap_mapping) {
1435 dest->has_block_bitmap_mapping = true;
1436 dest->block_bitmap_mapping = params->block_bitmap_mapping;
1437 }
1438 }
1439
1440 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1441 {
1442 MigrationState *s = migrate_get_current();
1443
1444 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1445
1446 if (params->has_compress_level) {
1447 s->parameters.compress_level = params->compress_level;
1448 }
1449
1450 if (params->has_compress_threads) {
1451 s->parameters.compress_threads = params->compress_threads;
1452 }
1453
1454 if (params->has_compress_wait_thread) {
1455 s->parameters.compress_wait_thread = params->compress_wait_thread;
1456 }
1457
1458 if (params->has_decompress_threads) {
1459 s->parameters.decompress_threads = params->decompress_threads;
1460 }
1461
1462 if (params->has_throttle_trigger_threshold) {
1463 s->parameters.throttle_trigger_threshold = params->throttle_trigger_threshold;
1464 }
1465
1466 if (params->has_cpu_throttle_initial) {
1467 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1468 }
1469
1470 if (params->has_cpu_throttle_increment) {
1471 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1472 }
1473
1474 if (params->has_cpu_throttle_tailslow) {
1475 s->parameters.cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1476 }
1477
1478 if (params->has_tls_creds) {
1479 g_free(s->parameters.tls_creds);
1480 assert(params->tls_creds->type == QTYPE_QSTRING);
1481 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1482 }
1483
1484 if (params->has_tls_hostname) {
1485 g_free(s->parameters.tls_hostname);
1486 assert(params->tls_hostname->type == QTYPE_QSTRING);
1487 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1488 }
1489
1490 if (params->has_tls_authz) {
1491 g_free(s->parameters.tls_authz);
1492 assert(params->tls_authz->type == QTYPE_QSTRING);
1493 s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1494 }
1495
1496 if (params->has_max_bandwidth) {
1497 s->parameters.max_bandwidth = params->max_bandwidth;
1498 if (s->to_dst_file && !migration_in_postcopy()) {
1499 qemu_file_set_rate_limit(s->to_dst_file,
1500 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1501 }
1502 }
1503
1504 if (params->has_downtime_limit) {
1505 s->parameters.downtime_limit = params->downtime_limit;
1506 }
1507
1508 if (params->has_x_checkpoint_delay) {
1509 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1510 if (migration_in_colo_state()) {
1511 colo_checkpoint_notify(s);
1512 }
1513 }
1514
1515 if (params->has_block_incremental) {
1516 s->parameters.block_incremental = params->block_incremental;
1517 }
1518 if (params->has_multifd_channels) {
1519 s->parameters.multifd_channels = params->multifd_channels;
1520 }
1521 if (params->has_multifd_compression) {
1522 s->parameters.multifd_compression = params->multifd_compression;
1523 }
1524 if (params->has_xbzrle_cache_size) {
1525 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1526 xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1527 }
1528 if (params->has_max_postcopy_bandwidth) {
1529 s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1530 if (s->to_dst_file && migration_in_postcopy()) {
1531 qemu_file_set_rate_limit(s->to_dst_file,
1532 s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1533 }
1534 }
1535 if (params->has_max_cpu_throttle) {
1536 s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1537 }
1538 if (params->has_announce_initial) {
1539 s->parameters.announce_initial = params->announce_initial;
1540 }
1541 if (params->has_announce_max) {
1542 s->parameters.announce_max = params->announce_max;
1543 }
1544 if (params->has_announce_rounds) {
1545 s->parameters.announce_rounds = params->announce_rounds;
1546 }
1547 if (params->has_announce_step) {
1548 s->parameters.announce_step = params->announce_step;
1549 }
1550
1551 if (params->has_block_bitmap_mapping) {
1552 qapi_free_BitmapMigrationNodeAliasList(
1553 s->parameters.block_bitmap_mapping);
1554
1555 s->parameters.has_block_bitmap_mapping = true;
1556 s->parameters.block_bitmap_mapping =
1557 QAPI_CLONE(BitmapMigrationNodeAliasList,
1558 params->block_bitmap_mapping);
1559 }
1560 }
1561
1562 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1563 {
1564 MigrationParameters tmp;
1565
1566 /* TODO Rewrite "" to null instead */
1567 if (params->has_tls_creds
1568 && params->tls_creds->type == QTYPE_QNULL) {
1569 qobject_unref(params->tls_creds->u.n);
1570 params->tls_creds->type = QTYPE_QSTRING;
1571 params->tls_creds->u.s = strdup("");
1572 }
1573 /* TODO Rewrite "" to null instead */
1574 if (params->has_tls_hostname
1575 && params->tls_hostname->type == QTYPE_QNULL) {
1576 qobject_unref(params->tls_hostname->u.n);
1577 params->tls_hostname->type = QTYPE_QSTRING;
1578 params->tls_hostname->u.s = strdup("");
1579 }
1580
1581 migrate_params_test_apply(params, &tmp);
1582
1583 if (!migrate_params_check(&tmp, errp)) {
1584 /* Invalid parameter */
1585 return;
1586 }
1587
1588 migrate_params_apply(params, errp);
1589 }
1590
1591
1592 void qmp_migrate_start_postcopy(Error **errp)
1593 {
1594 MigrationState *s = migrate_get_current();
1595
1596 if (!migrate_postcopy()) {
1597 error_setg(errp, "Enable postcopy with migrate_set_capability before"
1598 " the start of migration");
1599 return;
1600 }
1601
1602 if (s->state == MIGRATION_STATUS_NONE) {
1603 error_setg(errp, "Postcopy must be started after migration has been"
1604 " started");
1605 return;
1606 }
1607 /*
1608 * we don't error if migration has finished since that would be racy
1609 * with issuing this command.
1610 */
1611 qatomic_set(&s->start_postcopy, true);
1612 }
1613
1614 /* shared migration helpers */
1615
1616 void migrate_set_state(int *state, int old_state, int new_state)
1617 {
1618 assert(new_state < MIGRATION_STATUS__MAX);
1619 if (qatomic_cmpxchg(state, old_state, new_state) == old_state) {
1620 trace_migrate_set_state(MigrationStatus_str(new_state));
1621 migrate_generate_event(new_state);
1622 }
1623 }
1624
1625 static MigrationCapabilityStatusList *migrate_cap_add(
1626 MigrationCapabilityStatusList *list,
1627 MigrationCapability index,
1628 bool state)
1629 {
1630 MigrationCapabilityStatusList *cap;
1631
1632 cap = g_new0(MigrationCapabilityStatusList, 1);
1633 cap->value = g_new0(MigrationCapabilityStatus, 1);
1634 cap->value->capability = index;
1635 cap->value->state = state;
1636 cap->next = list;
1637
1638 return cap;
1639 }
1640
1641 void migrate_set_block_enabled(bool value, Error **errp)
1642 {
1643 MigrationCapabilityStatusList *cap;
1644
1645 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1646 qmp_migrate_set_capabilities(cap, errp);
1647 qapi_free_MigrationCapabilityStatusList(cap);
1648 }
1649
1650 static void migrate_set_block_incremental(MigrationState *s, bool value)
1651 {
1652 s->parameters.block_incremental = value;
1653 }
1654
1655 static void block_cleanup_parameters(MigrationState *s)
1656 {
1657 if (s->must_remove_block_options) {
1658 /* setting to false can never fail */
1659 migrate_set_block_enabled(false, &error_abort);
1660 migrate_set_block_incremental(s, false);
1661 s->must_remove_block_options = false;
1662 }
1663 }
1664
1665 static void migrate_fd_cleanup(MigrationState *s)
1666 {
1667 qemu_bh_delete(s->cleanup_bh);
1668 s->cleanup_bh = NULL;
1669
1670 qemu_savevm_state_cleanup();
1671
1672 if (s->to_dst_file) {
1673 QEMUFile *tmp;
1674
1675 trace_migrate_fd_cleanup();
1676 qemu_mutex_unlock_iothread();
1677 if (s->migration_thread_running) {
1678 qemu_thread_join(&s->thread);
1679 s->migration_thread_running = false;
1680 }
1681 qemu_mutex_lock_iothread();
1682
1683 multifd_save_cleanup();
1684 qemu_mutex_lock(&s->qemu_file_lock);
1685 tmp = s->to_dst_file;
1686 s->to_dst_file = NULL;
1687 qemu_mutex_unlock(&s->qemu_file_lock);
1688 /*
1689 * Close the file handle without the lock to make sure the
1690 * critical section won't block for long.
1691 */
1692 qemu_fclose(tmp);
1693 }
1694
1695 assert(!migration_is_active(s));
1696
1697 if (s->state == MIGRATION_STATUS_CANCELLING) {
1698 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1699 MIGRATION_STATUS_CANCELLED);
1700 }
1701
1702 if (s->error) {
1703 /* It is used on info migrate. We can't free it */
1704 error_report_err(error_copy(s->error));
1705 }
1706 notifier_list_notify(&migration_state_notifiers, s);
1707 block_cleanup_parameters(s);
1708 }
1709
1710 static void migrate_fd_cleanup_schedule(MigrationState *s)
1711 {
1712 /*
1713 * Ref the state for bh, because it may be called when
1714 * there're already no other refs
1715 */
1716 object_ref(OBJECT(s));
1717 qemu_bh_schedule(s->cleanup_bh);
1718 }
1719
1720 static void migrate_fd_cleanup_bh(void *opaque)
1721 {
1722 MigrationState *s = opaque;
1723 migrate_fd_cleanup(s);
1724 object_unref(OBJECT(s));
1725 }
1726
1727 void migrate_set_error(MigrationState *s, const Error *error)
1728 {
1729 QEMU_LOCK_GUARD(&s->error_mutex);
1730 if (!s->error) {
1731 s->error = error_copy(error);
1732 }
1733 }
1734
1735 void migrate_fd_error(MigrationState *s, const Error *error)
1736 {
1737 trace_migrate_fd_error(error_get_pretty(error));
1738 assert(s->to_dst_file == NULL);
1739 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1740 MIGRATION_STATUS_FAILED);
1741 migrate_set_error(s, error);
1742 }
1743
1744 static void migrate_fd_cancel(MigrationState *s)
1745 {
1746 int old_state ;
1747 QEMUFile *f = migrate_get_current()->to_dst_file;
1748 trace_migrate_fd_cancel();
1749
1750 if (s->rp_state.from_dst_file) {
1751 /* shutdown the rp socket, so causing the rp thread to shutdown */
1752 qemu_file_shutdown(s->rp_state.from_dst_file);
1753 }
1754
1755 do {
1756 old_state = s->state;
1757 if (!migration_is_running(old_state)) {
1758 break;
1759 }
1760 /* If the migration is paused, kick it out of the pause */
1761 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1762 qemu_sem_post(&s->pause_sem);
1763 }
1764 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1765 } while (s->state != MIGRATION_STATUS_CANCELLING);
1766
1767 /*
1768 * If we're unlucky the migration code might be stuck somewhere in a
1769 * send/write while the network has failed and is waiting to timeout;
1770 * if we've got shutdown(2) available then we can force it to quit.
1771 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1772 * called in a bh, so there is no race against this cancel.
1773 */
1774 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1775 qemu_file_shutdown(f);
1776 }
1777 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1778 Error *local_err = NULL;
1779
1780 bdrv_invalidate_cache_all(&local_err);
1781 if (local_err) {
1782 error_report_err(local_err);
1783 } else {
1784 s->block_inactive = false;
1785 }
1786 }
1787 }
1788
1789 void add_migration_state_change_notifier(Notifier *notify)
1790 {
1791 notifier_list_add(&migration_state_notifiers, notify);
1792 }
1793
1794 void remove_migration_state_change_notifier(Notifier *notify)
1795 {
1796 notifier_remove(notify);
1797 }
1798
1799 bool migration_in_setup(MigrationState *s)
1800 {
1801 return s->state == MIGRATION_STATUS_SETUP;
1802 }
1803
1804 bool migration_has_finished(MigrationState *s)
1805 {
1806 return s->state == MIGRATION_STATUS_COMPLETED;
1807 }
1808
1809 bool migration_has_failed(MigrationState *s)
1810 {
1811 return (s->state == MIGRATION_STATUS_CANCELLED ||
1812 s->state == MIGRATION_STATUS_FAILED);
1813 }
1814
1815 bool migration_in_postcopy(void)
1816 {
1817 MigrationState *s = migrate_get_current();
1818
1819 switch (s->state) {
1820 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1821 case MIGRATION_STATUS_POSTCOPY_PAUSED:
1822 case MIGRATION_STATUS_POSTCOPY_RECOVER:
1823 return true;
1824 default:
1825 return false;
1826 }
1827 }
1828
1829 bool migration_in_postcopy_after_devices(MigrationState *s)
1830 {
1831 return migration_in_postcopy() && s->postcopy_after_devices;
1832 }
1833
1834 bool migration_in_incoming_postcopy(void)
1835 {
1836 PostcopyState ps = postcopy_state_get();
1837
1838 return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END;
1839 }
1840
1841 bool migration_is_idle(void)
1842 {
1843 MigrationState *s = current_migration;
1844
1845 if (!s) {
1846 return true;
1847 }
1848
1849 switch (s->state) {
1850 case MIGRATION_STATUS_NONE:
1851 case MIGRATION_STATUS_CANCELLED:
1852 case MIGRATION_STATUS_COMPLETED:
1853 case MIGRATION_STATUS_FAILED:
1854 return true;
1855 case MIGRATION_STATUS_SETUP:
1856 case MIGRATION_STATUS_CANCELLING:
1857 case MIGRATION_STATUS_ACTIVE:
1858 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1859 case MIGRATION_STATUS_COLO:
1860 case MIGRATION_STATUS_PRE_SWITCHOVER:
1861 case MIGRATION_STATUS_DEVICE:
1862 case MIGRATION_STATUS_WAIT_UNPLUG:
1863 return false;
1864 case MIGRATION_STATUS__MAX:
1865 g_assert_not_reached();
1866 }
1867
1868 return false;
1869 }
1870
1871 bool migration_is_active(MigrationState *s)
1872 {
1873 return (s->state == MIGRATION_STATUS_ACTIVE ||
1874 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1875 }
1876
1877 void migrate_init(MigrationState *s)
1878 {
1879 /*
1880 * Reinitialise all migration state, except
1881 * parameters/capabilities that the user set, and
1882 * locks.
1883 */
1884 s->cleanup_bh = 0;
1885 s->to_dst_file = NULL;
1886 s->state = MIGRATION_STATUS_NONE;
1887 s->rp_state.from_dst_file = NULL;
1888 s->rp_state.error = false;
1889 s->mbps = 0.0;
1890 s->pages_per_second = 0.0;
1891 s->downtime = 0;
1892 s->expected_downtime = 0;
1893 s->setup_time = 0;
1894 s->start_postcopy = false;
1895 s->postcopy_after_devices = false;
1896 s->migration_thread_running = false;
1897 error_free(s->error);
1898 s->error = NULL;
1899 s->hostname = NULL;
1900
1901 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1902
1903 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1904 s->total_time = 0;
1905 s->vm_was_running = false;
1906 s->iteration_initial_bytes = 0;
1907 s->threshold_size = 0;
1908 }
1909
1910 static GSList *migration_blockers;
1911
1912 int migrate_add_blocker(Error *reason, Error **errp)
1913 {
1914 if (only_migratable) {
1915 error_propagate_prepend(errp, error_copy(reason),
1916 "disallowing migration blocker "
1917 "(--only-migratable) for: ");
1918 return -EACCES;
1919 }
1920
1921 if (migration_is_idle()) {
1922 migration_blockers = g_slist_prepend(migration_blockers, reason);
1923 return 0;
1924 }
1925
1926 error_propagate_prepend(errp, error_copy(reason),
1927 "disallowing migration blocker "
1928 "(migration in progress) for: ");
1929 return -EBUSY;
1930 }
1931
1932 void migrate_del_blocker(Error *reason)
1933 {
1934 migration_blockers = g_slist_remove(migration_blockers, reason);
1935 }
1936
1937 void qmp_migrate_incoming(const char *uri, Error **errp)
1938 {
1939 Error *local_err = NULL;
1940 static bool once = true;
1941
1942 if (!deferred_incoming) {
1943 error_setg(errp, "For use with '-incoming defer'");
1944 return;
1945 }
1946 if (!once) {
1947 error_setg(errp, "The incoming migration has already been started");
1948 return;
1949 }
1950
1951 qemu_start_incoming_migration(uri, &local_err);
1952
1953 if (local_err) {
1954 error_propagate(errp, local_err);
1955 return;
1956 }
1957
1958 once = false;
1959 }
1960
1961 void qmp_migrate_recover(const char *uri, Error **errp)
1962 {
1963 MigrationIncomingState *mis = migration_incoming_get_current();
1964
1965 if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1966 error_setg(errp, "Migrate recover can only be run "
1967 "when postcopy is paused.");
1968 return;
1969 }
1970
1971 if (qatomic_cmpxchg(&mis->postcopy_recover_triggered,
1972 false, true) == true) {
1973 error_setg(errp, "Migrate recovery is triggered already");
1974 return;
1975 }
1976
1977 /*
1978 * Note that this call will never start a real migration; it will
1979 * only re-setup the migration stream and poke existing migration
1980 * to continue using that newly established channel.
1981 */
1982 qemu_start_incoming_migration(uri, errp);
1983 }
1984
1985 void qmp_migrate_pause(Error **errp)
1986 {
1987 MigrationState *ms = migrate_get_current();
1988 MigrationIncomingState *mis = migration_incoming_get_current();
1989 int ret;
1990
1991 if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1992 /* Source side, during postcopy */
1993 qemu_mutex_lock(&ms->qemu_file_lock);
1994 ret = qemu_file_shutdown(ms->to_dst_file);
1995 qemu_mutex_unlock(&ms->qemu_file_lock);
1996 if (ret) {
1997 error_setg(errp, "Failed to pause source migration");
1998 }
1999 return;
2000 }
2001
2002 if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2003 ret = qemu_file_shutdown(mis->from_src_file);
2004 if (ret) {
2005 error_setg(errp, "Failed to pause destination migration");
2006 }
2007 return;
2008 }
2009
2010 error_setg(errp, "migrate-pause is currently only supported "
2011 "during postcopy-active state");
2012 }
2013
2014 bool migration_is_blocked(Error **errp)
2015 {
2016 if (qemu_savevm_state_blocked(errp)) {
2017 return true;
2018 }
2019
2020 if (migration_blockers) {
2021 error_propagate(errp, error_copy(migration_blockers->data));
2022 return true;
2023 }
2024
2025 return false;
2026 }
2027
2028 /* Returns true if continue to migrate, or false if error detected */
2029 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
2030 bool resume, Error **errp)
2031 {
2032 Error *local_err = NULL;
2033
2034 if (resume) {
2035 if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
2036 error_setg(errp, "Cannot resume if there is no "
2037 "paused migration");
2038 return false;
2039 }
2040
2041 /*
2042 * Postcopy recovery won't work well with release-ram
2043 * capability since release-ram will drop the page buffer as
2044 * long as the page is put into the send buffer. So if there
2045 * is a network failure happened, any page buffers that have
2046 * not yet reached the destination VM but have already been
2047 * sent from the source VM will be lost forever. Let's refuse
2048 * the client from resuming such a postcopy migration.
2049 * Luckily release-ram was designed to only be used when src
2050 * and destination VMs are on the same host, so it should be
2051 * fine.
2052 */
2053 if (migrate_release_ram()) {
2054 error_setg(errp, "Postcopy recovery cannot work "
2055 "when release-ram capability is set");
2056 return false;
2057 }
2058
2059 /* This is a resume, skip init status */
2060 return true;
2061 }
2062
2063 if (migration_is_running(s->state)) {
2064 error_setg(errp, QERR_MIGRATION_ACTIVE);
2065 return false;
2066 }
2067
2068 if (runstate_check(RUN_STATE_INMIGRATE)) {
2069 error_setg(errp, "Guest is waiting for an incoming migration");
2070 return false;
2071 }
2072
2073 if (migration_is_blocked(errp)) {
2074 return false;
2075 }
2076
2077 if (blk || blk_inc) {
2078 if (migrate_use_block() || migrate_use_block_incremental()) {
2079 error_setg(errp, "Command options are incompatible with "
2080 "current migration capabilities");
2081 return false;
2082 }
2083 migrate_set_block_enabled(true, &local_err);
2084 if (local_err) {
2085 error_propagate(errp, local_err);
2086 return false;
2087 }
2088 s->must_remove_block_options = true;
2089 }
2090
2091 if (blk_inc) {
2092 migrate_set_block_incremental(s, true);
2093 }
2094
2095 migrate_init(s);
2096 /*
2097 * set ram_counters memory to zero for a
2098 * new migration
2099 */
2100 memset(&ram_counters, 0, sizeof(ram_counters));
2101
2102 return true;
2103 }
2104
2105 void qmp_migrate(const char *uri, bool has_blk, bool blk,
2106 bool has_inc, bool inc, bool has_detach, bool detach,
2107 bool has_resume, bool resume, Error **errp)
2108 {
2109 Error *local_err = NULL;
2110 MigrationState *s = migrate_get_current();
2111 const char *p = NULL;
2112
2113 if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
2114 has_resume && resume, errp)) {
2115 /* Error detected, put into errp */
2116 return;
2117 }
2118
2119 if (strstart(uri, "tcp:", &p) ||
2120 strstart(uri, "unix:", NULL) ||
2121 strstart(uri, "vsock:", NULL)) {
2122 socket_start_outgoing_migration(s, p ? p : uri, &local_err);
2123 #ifdef CONFIG_RDMA
2124 } else if (strstart(uri, "rdma:", &p)) {
2125 rdma_start_outgoing_migration(s, p, &local_err);
2126 #endif
2127 } else if (strstart(uri, "exec:", &p)) {
2128 exec_start_outgoing_migration(s, p, &local_err);
2129 } else if (strstart(uri, "fd:", &p)) {
2130 fd_start_outgoing_migration(s, p, &local_err);
2131 } else {
2132 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
2133 "a valid migration protocol");
2134 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2135 MIGRATION_STATUS_FAILED);
2136 block_cleanup_parameters(s);
2137 return;
2138 }
2139
2140 if (local_err) {
2141 migrate_fd_error(s, local_err);
2142 error_propagate(errp, local_err);
2143 return;
2144 }
2145 }
2146
2147 void qmp_migrate_cancel(Error **errp)
2148 {
2149 migrate_fd_cancel(migrate_get_current());
2150 }
2151
2152 void qmp_migrate_continue(MigrationStatus state, Error **errp)
2153 {
2154 MigrationState *s = migrate_get_current();
2155 if (s->state != state) {
2156 error_setg(errp, "Migration not in expected state: %s",
2157 MigrationStatus_str(s->state));
2158 return;
2159 }
2160 qemu_sem_post(&s->pause_sem);
2161 }
2162
2163 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
2164 {
2165 MigrateSetParameters p = {
2166 .has_xbzrle_cache_size = true,
2167 .xbzrle_cache_size = value,
2168 };
2169
2170 qmp_migrate_set_parameters(&p, errp);
2171 }
2172
2173 int64_t qmp_query_migrate_cache_size(Error **errp)
2174 {
2175 return migrate_xbzrle_cache_size();
2176 }
2177
2178 void qmp_migrate_set_speed(int64_t value, Error **errp)
2179 {
2180 MigrateSetParameters p = {
2181 .has_max_bandwidth = true,
2182 .max_bandwidth = value,
2183 };
2184
2185 qmp_migrate_set_parameters(&p, errp);
2186 }
2187
2188 void qmp_migrate_set_downtime(double value, Error **errp)
2189 {
2190 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
2191 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
2192 "downtime_limit",
2193 "an integer in the range of 0 to "
2194 stringify(MAX_MIGRATE_DOWNTIME_SECONDS)" seconds");
2195 return;
2196 }
2197
2198 value *= 1000; /* Convert to milliseconds */
2199
2200 MigrateSetParameters p = {
2201 .has_downtime_limit = true,
2202 .downtime_limit = (int64_t)value,
2203 };
2204
2205 qmp_migrate_set_parameters(&p, errp);
2206 }
2207
2208 bool migrate_release_ram(void)
2209 {
2210 MigrationState *s;
2211
2212 s = migrate_get_current();
2213
2214 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2215 }
2216
2217 bool migrate_postcopy_ram(void)
2218 {
2219 MigrationState *s;
2220
2221 s = migrate_get_current();
2222
2223 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2224 }
2225
2226 bool migrate_postcopy(void)
2227 {
2228 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2229 }
2230
2231 bool migrate_auto_converge(void)
2232 {
2233 MigrationState *s;
2234
2235 s = migrate_get_current();
2236
2237 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2238 }
2239
2240 bool migrate_zero_blocks(void)
2241 {
2242 MigrationState *s;
2243
2244 s = migrate_get_current();
2245
2246 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2247 }
2248
2249 bool migrate_postcopy_blocktime(void)
2250 {
2251 MigrationState *s;
2252
2253 s = migrate_get_current();
2254
2255 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2256 }
2257
2258 bool migrate_use_compression(void)
2259 {
2260 MigrationState *s;
2261
2262 s = migrate_get_current();
2263
2264 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2265 }
2266
2267 int migrate_compress_level(void)
2268 {
2269 MigrationState *s;
2270
2271 s = migrate_get_current();
2272
2273 return s->parameters.compress_level;
2274 }
2275
2276 int migrate_compress_threads(void)
2277 {
2278 MigrationState *s;
2279
2280 s = migrate_get_current();
2281
2282 return s->parameters.compress_threads;
2283 }
2284
2285 int migrate_compress_wait_thread(void)
2286 {
2287 MigrationState *s;
2288
2289 s = migrate_get_current();
2290
2291 return s->parameters.compress_wait_thread;
2292 }
2293
2294 int migrate_decompress_threads(void)
2295 {
2296 MigrationState *s;
2297
2298 s = migrate_get_current();
2299
2300 return s->parameters.decompress_threads;
2301 }
2302
2303 bool migrate_dirty_bitmaps(void)
2304 {
2305 MigrationState *s;
2306
2307 s = migrate_get_current();
2308
2309 return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2310 }
2311
2312 bool migrate_ignore_shared(void)
2313 {
2314 MigrationState *s;
2315
2316 s = migrate_get_current();
2317
2318 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2319 }
2320
2321 bool migrate_validate_uuid(void)
2322 {
2323 MigrationState *s;
2324
2325 s = migrate_get_current();
2326
2327 return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
2328 }
2329
2330 bool migrate_use_events(void)
2331 {
2332 MigrationState *s;
2333
2334 s = migrate_get_current();
2335
2336 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2337 }
2338
2339 bool migrate_use_multifd(void)
2340 {
2341 MigrationState *s;
2342
2343 s = migrate_get_current();
2344
2345 return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2346 }
2347
2348 bool migrate_pause_before_switchover(void)
2349 {
2350 MigrationState *s;
2351
2352 s = migrate_get_current();
2353
2354 return s->enabled_capabilities[
2355 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2356 }
2357
2358 int migrate_multifd_channels(void)
2359 {
2360 MigrationState *s;
2361
2362 s = migrate_get_current();
2363
2364 return s->parameters.multifd_channels;
2365 }
2366
2367 MultiFDCompression migrate_multifd_compression(void)
2368 {
2369 MigrationState *s;
2370
2371 s = migrate_get_current();
2372
2373 return s->parameters.multifd_compression;
2374 }
2375
2376 int migrate_multifd_zlib_level(void)
2377 {
2378 MigrationState *s;
2379
2380 s = migrate_get_current();
2381
2382 return s->parameters.multifd_zlib_level;
2383 }
2384
2385 int migrate_multifd_zstd_level(void)
2386 {
2387 MigrationState *s;
2388
2389 s = migrate_get_current();
2390
2391 return s->parameters.multifd_zstd_level;
2392 }
2393
2394 int migrate_use_xbzrle(void)
2395 {
2396 MigrationState *s;
2397
2398 s = migrate_get_current();
2399
2400 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2401 }
2402
2403 int64_t migrate_xbzrle_cache_size(void)
2404 {
2405 MigrationState *s;
2406
2407 s = migrate_get_current();
2408
2409 return s->parameters.xbzrle_cache_size;
2410 }
2411
2412 static int64_t migrate_max_postcopy_bandwidth(void)
2413 {
2414 MigrationState *s;
2415
2416 s = migrate_get_current();
2417
2418 return s->parameters.max_postcopy_bandwidth;
2419 }
2420
2421 bool migrate_use_block(void)
2422 {
2423 MigrationState *s;
2424
2425 s = migrate_get_current();
2426
2427 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2428 }
2429
2430 bool migrate_use_return_path(void)
2431 {
2432 MigrationState *s;
2433
2434 s = migrate_get_current();
2435
2436 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2437 }
2438
2439 bool migrate_use_block_incremental(void)
2440 {
2441 MigrationState *s;
2442
2443 s = migrate_get_current();
2444
2445 return s->parameters.block_incremental;
2446 }
2447
2448 /* migration thread support */
2449 /*
2450 * Something bad happened to the RP stream, mark an error
2451 * The caller shall print or trace something to indicate why
2452 */
2453 static void mark_source_rp_bad(MigrationState *s)
2454 {
2455 s->rp_state.error = true;
2456 }
2457
2458 static struct rp_cmd_args {
2459 ssize_t len; /* -1 = variable */
2460 const char *name;
2461 } rp_cmd_args[] = {
2462 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
2463 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
2464 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
2465 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
2466 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
2467 [MIG_RP_MSG_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" },
2468 [MIG_RP_MSG_RESUME_ACK] = { .len = 4, .name = "RESUME_ACK" },
2469 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
2470 };
2471
2472 /*
2473 * Process a request for pages received on the return path,
2474 * We're allowed to send more than requested (e.g. to round to our page size)
2475 * and we don't need to send pages that have already been sent.
2476 */
2477 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2478 ram_addr_t start, size_t len)
2479 {
2480 long our_host_ps = qemu_real_host_page_size;
2481
2482 trace_migrate_handle_rp_req_pages(rbname, start, len);
2483
2484 /*
2485 * Since we currently insist on matching page sizes, just sanity check
2486 * we're being asked for whole host pages.
2487 */
2488 if (start & (our_host_ps-1) ||
2489 (len & (our_host_ps-1))) {
2490 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2491 " len: %zd", __func__, start, len);
2492 mark_source_rp_bad(ms);
2493 return;
2494 }
2495
2496 if (ram_save_queue_pages(rbname, start, len)) {
2497 mark_source_rp_bad(ms);
2498 }
2499 }
2500
2501 /* Return true to retry, false to quit */
2502 static bool postcopy_pause_return_path_thread(MigrationState *s)
2503 {
2504 trace_postcopy_pause_return_path();
2505
2506 qemu_sem_wait(&s->postcopy_pause_rp_sem);
2507
2508 trace_postcopy_pause_return_path_continued();
2509
2510 return true;
2511 }
2512
2513 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2514 {
2515 RAMBlock *block = qemu_ram_block_by_name(block_name);
2516
2517 if (!block) {
2518 error_report("%s: invalid block name '%s'", __func__, block_name);
2519 return -EINVAL;
2520 }
2521
2522 /* Fetch the received bitmap and refresh the dirty bitmap */
2523 return ram_dirty_bitmap_reload(s, block);
2524 }
2525
2526 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2527 {
2528 trace_source_return_path_thread_resume_ack(value);
2529
2530 if (value != MIGRATION_RESUME_ACK_VALUE) {
2531 error_report("%s: illegal resume_ack value %"PRIu32,
2532 __func__, value);
2533 return -1;
2534 }
2535
2536 /* Now both sides are active. */
2537 migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2538 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2539
2540 /* Notify send thread that time to continue send pages */
2541 qemu_sem_post(&s->rp_state.rp_sem);
2542
2543 return 0;
2544 }
2545
2546 /*
2547 * Handles messages sent on the return path towards the source VM
2548 *
2549 */
2550 static void *source_return_path_thread(void *opaque)
2551 {
2552 MigrationState *ms = opaque;
2553 QEMUFile *rp = ms->rp_state.from_dst_file;
2554 uint16_t header_len, header_type;
2555 uint8_t buf[512];
2556 uint32_t tmp32, sibling_error;
2557 ram_addr_t start = 0; /* =0 to silence warning */
2558 size_t len = 0, expected_len;
2559 int res;
2560
2561 trace_source_return_path_thread_entry();
2562 rcu_register_thread();
2563
2564 retry:
2565 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2566 migration_is_setup_or_active(ms->state)) {
2567 trace_source_return_path_thread_loop_top();
2568 header_type = qemu_get_be16(rp);
2569 header_len = qemu_get_be16(rp);
2570
2571 if (qemu_file_get_error(rp)) {
2572 mark_source_rp_bad(ms);
2573 goto out;
2574 }
2575
2576 if (header_type >= MIG_RP_MSG_MAX ||
2577 header_type == MIG_RP_MSG_INVALID) {
2578 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2579 header_type, header_len);
2580 mark_source_rp_bad(ms);
2581 goto out;
2582 }
2583
2584 if ((rp_cmd_args[header_type].len != -1 &&
2585 header_len != rp_cmd_args[header_type].len) ||
2586 header_len > sizeof(buf)) {
2587 error_report("RP: Received '%s' message (0x%04x) with"
2588 "incorrect length %d expecting %zu",
2589 rp_cmd_args[header_type].name, header_type, header_len,
2590 (size_t)rp_cmd_args[header_type].len);
2591 mark_source_rp_bad(ms);
2592 goto out;
2593 }
2594
2595 /* We know we've got a valid header by this point */
2596 res = qemu_get_buffer(rp, buf, header_len);
2597 if (res != header_len) {
2598 error_report("RP: Failed reading data for message 0x%04x"
2599 " read %d expected %d",
2600 header_type, res, header_len);
2601 mark_source_rp_bad(ms);
2602 goto out;
2603 }
2604
2605 /* OK, we have the message and the data */
2606 switch (header_type) {
2607 case MIG_RP_MSG_SHUT:
2608 sibling_error = ldl_be_p(buf);
2609 trace_source_return_path_thread_shut(sibling_error);
2610 if (sibling_error) {
2611 error_report("RP: Sibling indicated error %d", sibling_error);
2612 mark_source_rp_bad(ms);
2613 }
2614 /*
2615 * We'll let the main thread deal with closing the RP
2616 * we could do a shutdown(2) on it, but we're the only user
2617 * anyway, so there's nothing gained.
2618 */
2619 goto out;
2620
2621 case MIG_RP_MSG_PONG:
2622 tmp32 = ldl_be_p(buf);
2623 trace_source_return_path_thread_pong(tmp32);
2624 break;
2625
2626 case MIG_RP_MSG_REQ_PAGES:
2627 start = ldq_be_p(buf);
2628 len = ldl_be_p(buf + 8);
2629 migrate_handle_rp_req_pages(ms, NULL, start, len);
2630 break;
2631
2632 case MIG_RP_MSG_REQ_PAGES_ID:
2633 expected_len = 12 + 1; /* header + termination */
2634
2635 if (header_len >= expected_len) {
2636 start = ldq_be_p(buf);
2637 len = ldl_be_p(buf + 8);
2638 /* Now we expect an idstr */
2639 tmp32 = buf[12]; /* Length of the following idstr */
2640 buf[13 + tmp32] = '\0';
2641 expected_len += tmp32;
2642 }
2643 if (header_len != expected_len) {
2644 error_report("RP: Req_Page_id with length %d expecting %zd",
2645 header_len, expected_len);
2646 mark_source_rp_bad(ms);
2647 goto out;
2648 }
2649 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2650 break;
2651
2652 case MIG_RP_MSG_RECV_BITMAP:
2653 if (header_len < 1) {
2654 error_report("%s: missing block name", __func__);
2655 mark_source_rp_bad(ms);
2656 goto out;
2657 }
2658 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2659 buf[buf[0] + 1] = '\0';
2660 if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2661 mark_source_rp_bad(ms);
2662 goto out;
2663 }
2664 break;
2665
2666 case MIG_RP_MSG_RESUME_ACK:
2667 tmp32 = ldl_be_p(buf);
2668 if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2669 mark_source_rp_bad(ms);
2670 goto out;
2671 }
2672 break;
2673
2674 default:
2675 break;
2676 }
2677 }
2678
2679 out:
2680 res = qemu_file_get_error(rp);
2681 if (res) {
2682 if (res == -EIO && migration_in_postcopy()) {
2683 /*
2684 * Maybe there is something we can do: it looks like a
2685 * network down issue, and we pause for a recovery.
2686 */
2687 if (postcopy_pause_return_path_thread(ms)) {
2688 /* Reload rp, reset the rest */
2689 if (rp != ms->rp_state.from_dst_file) {
2690 qemu_fclose(rp);
2691 rp = ms->rp_state.from_dst_file;
2692 }
2693 ms->rp_state.error = false;
2694 goto retry;
2695 }
2696 }
2697
2698 trace_source_return_path_thread_bad_end();
2699 mark_source_rp_bad(ms);
2700 }
2701
2702 trace_source_return_path_thread_end();
2703 ms->rp_state.from_dst_file = NULL;
2704 qemu_fclose(rp);
2705 rcu_unregister_thread();
2706 return NULL;
2707 }
2708
2709 static int open_return_path_on_source(MigrationState *ms,
2710 bool create_thread)
2711 {
2712
2713 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2714 if (!ms->rp_state.from_dst_file) {
2715 return -1;
2716 }
2717
2718 trace_open_return_path_on_source();
2719
2720 if (!create_thread) {
2721 /* We're done */
2722 return 0;
2723 }
2724
2725 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2726 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2727
2728 trace_open_return_path_on_source_continue();
2729
2730 return 0;
2731 }
2732
2733 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2734 static int await_return_path_close_on_source(MigrationState *ms)
2735 {
2736 /*
2737 * If this is a normal exit then the destination will send a SHUT and the
2738 * rp_thread will exit, however if there's an error we need to cause
2739 * it to exit.
2740 */
2741 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2742 /*
2743 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2744 * waiting for the destination.
2745 */
2746 qemu_file_shutdown(ms->rp_state.from_dst_file);
2747 mark_source_rp_bad(ms);
2748 }
2749 trace_await_return_path_close_on_source_joining();
2750 qemu_thread_join(&ms->rp_state.rp_thread);
2751 trace_await_return_path_close_on_source_close();
2752 return ms->rp_state.error;
2753 }
2754
2755 /*
2756 * Switch from normal iteration to postcopy
2757 * Returns non-0 on error
2758 */
2759 static int postcopy_start(MigrationState *ms)
2760 {
2761 int ret;
2762 QIOChannelBuffer *bioc;
2763 QEMUFile *fb;
2764 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2765 int64_t bandwidth = migrate_max_postcopy_bandwidth();
2766 bool restart_block = false;
2767 int cur_state = MIGRATION_STATUS_ACTIVE;
2768 if (!migrate_pause_before_switchover()) {
2769 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2770 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2771 }
2772
2773 trace_postcopy_start();
2774 qemu_mutex_lock_iothread();
2775 trace_postcopy_start_set_run();
2776
2777 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2778 global_state_store();
2779 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2780 if (ret < 0) {
2781 goto fail;
2782 }
2783
2784 ret = migration_maybe_pause(ms, &cur_state,
2785 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2786 if (ret < 0) {
2787 goto fail;
2788 }
2789
2790 ret = bdrv_inactivate_all();
2791 if (ret < 0) {
2792 goto fail;
2793 }
2794 restart_block = true;
2795
2796 /*
2797 * Cause any non-postcopiable, but iterative devices to
2798 * send out their final data.
2799 */
2800 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2801
2802 /*
2803 * in Finish migrate and with the io-lock held everything should
2804 * be quiet, but we've potentially still got dirty pages and we
2805 * need to tell the destination to throw any pages it's already received
2806 * that are dirty
2807 */
2808 if (migrate_postcopy_ram()) {
2809 if (ram_postcopy_send_discard_bitmap(ms)) {
2810 error_report("postcopy send discard bitmap failed");
2811 goto fail;
2812 }
2813 }
2814
2815 /*
2816 * send rest of state - note things that are doing postcopy
2817 * will notice we're in POSTCOPY_ACTIVE and not actually
2818 * wrap their state up here
2819 */
2820 /* 0 max-postcopy-bandwidth means unlimited */
2821 if (!bandwidth) {
2822 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2823 } else {
2824 qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2825 }
2826 if (migrate_postcopy_ram()) {
2827 /* Ping just for debugging, helps line traces up */
2828 qemu_savevm_send_ping(ms->to_dst_file, 2);
2829 }
2830
2831 /*
2832 * While loading the device state we may trigger page transfer
2833 * requests and the fd must be free to process those, and thus
2834 * the destination must read the whole device state off the fd before
2835 * it starts processing it. Unfortunately the ad-hoc migration format
2836 * doesn't allow the destination to know the size to read without fully
2837 * parsing it through each devices load-state code (especially the open
2838 * coded devices that use get/put).
2839 * So we wrap the device state up in a package with a length at the start;
2840 * to do this we use a qemu_buf to hold the whole of the device state.
2841 */
2842 bioc = qio_channel_buffer_new(4096);
2843 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2844 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2845 object_unref(OBJECT(bioc));
2846
2847 /*
2848 * Make sure the receiver can get incoming pages before we send the rest
2849 * of the state
2850 */
2851 qemu_savevm_send_postcopy_listen(fb);
2852
2853 qemu_savevm_state_complete_precopy(fb, false, false);
2854 if (migrate_postcopy_ram()) {
2855 qemu_savevm_send_ping(fb, 3);
2856 }
2857
2858 qemu_savevm_send_postcopy_run(fb);
2859
2860 /* <><> end of stuff going into the package */
2861
2862 /* Last point of recovery; as soon as we send the package the destination
2863 * can open devices and potentially start running.
2864 * Lets just check again we've not got any errors.
2865 */
2866 ret = qemu_file_get_error(ms->to_dst_file);
2867 if (ret) {
2868 error_report("postcopy_start: Migration stream errored (pre package)");
2869 goto fail_closefb;
2870 }
2871
2872 restart_block = false;
2873
2874 /* Now send that blob */
2875 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2876 goto fail_closefb;
2877 }
2878 qemu_fclose(fb);
2879
2880 /* Send a notify to give a chance for anything that needs to happen
2881 * at the transition to postcopy and after the device state; in particular
2882 * spice needs to trigger a transition now
2883 */
2884 ms->postcopy_after_devices = true;
2885 notifier_list_notify(&migration_state_notifiers, ms);
2886
2887 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2888
2889 qemu_mutex_unlock_iothread();
2890
2891 if (migrate_postcopy_ram()) {
2892 /*
2893 * Although this ping is just for debug, it could potentially be
2894 * used for getting a better measurement of downtime at the source.
2895 */
2896 qemu_savevm_send_ping(ms->to_dst_file, 4);
2897 }
2898
2899 if (migrate_release_ram()) {
2900 ram_postcopy_migrated_memory_release(ms);
2901 }
2902
2903 ret = qemu_file_get_error(ms->to_dst_file);
2904 if (ret) {
2905 error_report("postcopy_start: Migration stream errored");
2906 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2907 MIGRATION_STATUS_FAILED);
2908 }
2909
2910 return ret;
2911
2912 fail_closefb:
2913 qemu_fclose(fb);
2914 fail:
2915 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2916 MIGRATION_STATUS_FAILED);
2917 if (restart_block) {
2918 /* A failure happened early enough that we know the destination hasn't
2919 * accessed block devices, so we're safe to recover.
2920 */
2921 Error *local_err = NULL;
2922
2923 bdrv_invalidate_cache_all(&local_err);
2924 if (local_err) {
2925 error_report_err(local_err);
2926 }
2927 }
2928 qemu_mutex_unlock_iothread();
2929 return -1;
2930 }
2931
2932 /**
2933 * migration_maybe_pause: Pause if required to by
2934 * migrate_pause_before_switchover called with the iothread locked
2935 * Returns: 0 on success
2936 */
2937 static int migration_maybe_pause(MigrationState *s,
2938 int *current_active_state,
2939 int new_state)
2940 {
2941 if (!migrate_pause_before_switchover()) {
2942 return 0;
2943 }
2944
2945 /* Since leaving this state is not atomic with posting the semaphore
2946 * it's possible that someone could have issued multiple migrate_continue
2947 * and the semaphore is incorrectly positive at this point;
2948 * the docs say it's undefined to reinit a semaphore that's already
2949 * init'd, so use timedwait to eat up any existing posts.
2950 */
2951 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2952 /* This block intentionally left blank */
2953 }
2954
2955 /*
2956 * If the migration is cancelled when it is in the completion phase,
2957 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2958 * So we don't need to wait a semaphore, otherwise we would always
2959 * wait for the 'pause_sem' semaphore.
2960 */
2961 if (s->state != MIGRATION_STATUS_CANCELLING) {
2962 qemu_mutex_unlock_iothread();
2963 migrate_set_state(&s->state, *current_active_state,
2964 MIGRATION_STATUS_PRE_SWITCHOVER);
2965 qemu_sem_wait(&s->pause_sem);
2966 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2967 new_state);
2968 *current_active_state = new_state;
2969 qemu_mutex_lock_iothread();
2970 }
2971
2972 return s->state == new_state ? 0 : -EINVAL;
2973 }
2974
2975 /**
2976 * migration_completion: Used by migration_thread when there's not much left.
2977 * The caller 'breaks' the loop when this returns.
2978 *
2979 * @s: Current migration state
2980 */
2981 static void migration_completion(MigrationState *s)
2982 {
2983 int ret;
2984 int current_active_state = s->state;
2985
2986 if (s->state == MIGRATION_STATUS_ACTIVE) {
2987 qemu_mutex_lock_iothread();
2988 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2989 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2990 s->vm_was_running = runstate_is_running();
2991 ret = global_state_store();
2992
2993 if (!ret) {
2994 bool inactivate = !migrate_colo_enabled();
2995 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2996 if (ret >= 0) {
2997 ret = migration_maybe_pause(s, &current_active_state,
2998 MIGRATION_STATUS_DEVICE);
2999 }
3000 if (ret >= 0) {
3001 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
3002 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
3003 inactivate);
3004 }
3005 if (inactivate && ret >= 0) {
3006 s->block_inactive = true;
3007 }
3008 }
3009 qemu_mutex_unlock_iothread();
3010
3011 if (ret < 0) {
3012 goto fail;
3013 }
3014 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3015 trace_migration_completion_postcopy_end();
3016
3017 qemu_savevm_state_complete_postcopy(s->to_dst_file);
3018 trace_migration_completion_postcopy_end_after_complete();
3019 }
3020
3021 /*
3022 * If rp was opened we must clean up the thread before
3023 * cleaning everything else up (since if there are no failures
3024 * it will wait for the destination to send it's status in
3025 * a SHUT command).
3026 */
3027 if (s->rp_state.from_dst_file) {
3028 int rp_error;
3029 trace_migration_return_path_end_before();
3030 rp_error = await_return_path_close_on_source(s);
3031 trace_migration_return_path_end_after(rp_error);
3032 if (rp_error) {
3033 goto fail_invalidate;
3034 }
3035 }
3036
3037 if (qemu_file_get_error(s->to_dst_file)) {
3038 trace_migration_completion_file_err();
3039 goto fail_invalidate;
3040 }
3041
3042 if (!migrate_colo_enabled()) {
3043 migrate_set_state(&s->state, current_active_state,
3044 MIGRATION_STATUS_COMPLETED);
3045 }
3046
3047 return;
3048
3049 fail_invalidate:
3050 /* If not doing postcopy, vm_start() will be called: let's regain
3051 * control on images.
3052 */
3053 if (s->state == MIGRATION_STATUS_ACTIVE ||
3054 s->state == MIGRATION_STATUS_DEVICE) {
3055 Error *local_err = NULL;
3056
3057 qemu_mutex_lock_iothread();
3058 bdrv_invalidate_cache_all(&local_err);
3059 if (local_err) {
3060 error_report_err(local_err);
3061 } else {
3062 s->block_inactive = false;
3063 }
3064 qemu_mutex_unlock_iothread();
3065 }
3066
3067 fail:
3068 migrate_set_state(&s->state, current_active_state,
3069 MIGRATION_STATUS_FAILED);
3070 }
3071
3072 bool migrate_colo_enabled(void)
3073 {
3074 MigrationState *s = migrate_get_current();
3075 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
3076 }
3077
3078 typedef enum MigThrError {
3079 /* No error detected */
3080 MIG_THR_ERR_NONE = 0,
3081 /* Detected error, but resumed successfully */
3082 MIG_THR_ERR_RECOVERED = 1,
3083 /* Detected fatal error, need to exit */
3084 MIG_THR_ERR_FATAL = 2,
3085 } MigThrError;
3086
3087 static int postcopy_resume_handshake(MigrationState *s)
3088 {
3089 qemu_savevm_send_postcopy_resume(s->to_dst_file);
3090
3091 while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3092 qemu_sem_wait(&s->rp_state.rp_sem);
3093 }
3094
3095 if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3096 return 0;
3097 }
3098
3099 return -1;
3100 }
3101
3102 /* Return zero if success, or <0 for error */
3103 static int postcopy_do_resume(MigrationState *s)
3104 {
3105 int ret;
3106
3107 /*
3108 * Call all the resume_prepare() hooks, so that modules can be
3109 * ready for the migration resume.
3110 */
3111 ret = qemu_savevm_state_resume_prepare(s);
3112 if (ret) {
3113 error_report("%s: resume_prepare() failure detected: %d",
3114 __func__, ret);
3115 return ret;
3116 }
3117
3118 /*
3119 * Last handshake with destination on the resume (destination will
3120 * switch to postcopy-active afterwards)
3121 */
3122 ret = postcopy_resume_handshake(s);
3123 if (ret) {
3124 error_report("%s: handshake failed: %d", __func__, ret);
3125 return ret;
3126 }
3127
3128 return 0;
3129 }
3130
3131 /*
3132 * We don't return until we are in a safe state to continue current
3133 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3134 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3135 */
3136 static MigThrError postcopy_pause(MigrationState *s)
3137 {
3138 assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
3139
3140 while (true) {
3141 QEMUFile *file;
3142
3143 migrate_set_state(&s->state, s->state,
3144 MIGRATION_STATUS_POSTCOPY_PAUSED);
3145
3146 /* Current channel is possibly broken. Release it. */
3147 assert(s->to_dst_file);
3148 qemu_mutex_lock(&s->qemu_file_lock);
3149 file = s->to_dst_file;
3150 s->to_dst_file = NULL;
3151 qemu_mutex_unlock(&s->qemu_file_lock);
3152
3153 qemu_file_shutdown(file);
3154 qemu_fclose(file);
3155
3156 error_report("Detected IO failure for postcopy. "
3157 "Migration paused.");
3158
3159 /*
3160 * We wait until things fixed up. Then someone will setup the
3161 * status back for us.
3162 */
3163 while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
3164 qemu_sem_wait(&s->postcopy_pause_sem);
3165 }
3166
3167 if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3168 /* Woken up by a recover procedure. Give it a shot */
3169
3170 /*
3171 * Firstly, let's wake up the return path now, with a new
3172 * return path channel.
3173 */
3174 qemu_sem_post(&s->postcopy_pause_rp_sem);
3175
3176 /* Do the resume logic */
3177 if (postcopy_do_resume(s) == 0) {
3178 /* Let's continue! */
3179 trace_postcopy_pause_continued();
3180 return MIG_THR_ERR_RECOVERED;
3181 } else {
3182 /*
3183 * Something wrong happened during the recovery, let's
3184 * pause again. Pause is always better than throwing
3185 * data away.
3186 */
3187 continue;
3188 }
3189 } else {
3190 /* This is not right... Time to quit. */
3191 return MIG_THR_ERR_FATAL;
3192 }
3193 }
3194 }
3195
3196 static MigThrError migration_detect_error(MigrationState *s)
3197 {
3198 int ret;
3199 int state = s->state;
3200 Error *local_error = NULL;
3201
3202 if (state == MIGRATION_STATUS_CANCELLING ||
3203 state == MIGRATION_STATUS_CANCELLED) {
3204 /* End the migration, but don't set the state to failed */
3205 return MIG_THR_ERR_FATAL;
3206 }
3207
3208 /* Try to detect any file errors */
3209 ret = qemu_file_get_error_obj(s->to_dst_file, &local_error);
3210 if (!ret) {
3211 /* Everything is fine */
3212 assert(!local_error);
3213 return MIG_THR_ERR_NONE;
3214 }
3215
3216 if (local_error) {
3217 migrate_set_error(s, local_error);
3218 error_free(local_error);
3219 }
3220
3221 if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
3222 /*
3223 * For postcopy, we allow the network to be down for a
3224 * while. After that, it can be continued by a
3225 * recovery phase.
3226 */
3227 return postcopy_pause(s);
3228 } else {
3229 /*
3230 * For precopy (or postcopy with error outside IO), we fail
3231 * with no time.
3232 */
3233 migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
3234 trace_migration_thread_file_err();
3235
3236 /* Time to stop the migration, now. */
3237 return MIG_THR_ERR_FATAL;
3238 }
3239 }
3240
3241 /* How many bytes have we transferred since the beginning of the migration */
3242 static uint64_t migration_total_bytes(MigrationState *s)
3243 {
3244 return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3245 }
3246
3247 static void migration_calculate_complete(MigrationState *s)
3248 {
3249 uint64_t bytes = migration_total_bytes(s);
3250 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3251 int64_t transfer_time;
3252
3253 s->total_time = end_time - s->start_time;
3254 if (!s->downtime) {
3255 /*
3256 * It's still not set, so we are precopy migration. For
3257 * postcopy, downtime is calculated during postcopy_start().
3258 */
3259 s->downtime = end_time - s->downtime_start;
3260 }
3261
3262 transfer_time = s->total_time - s->setup_time;
3263 if (transfer_time) {
3264 s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3265 }
3266 }
3267
3268 static void update_iteration_initial_status(MigrationState *s)
3269 {
3270 /*
3271 * Update these three fields at the same time to avoid mismatch info lead
3272 * wrong speed calculation.
3273 */
3274 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3275 s->iteration_initial_bytes = migration_total_bytes(s);
3276 s->iteration_initial_pages = ram_get_total_transferred_pages();
3277 }
3278
3279 static void migration_update_counters(MigrationState *s,
3280 int64_t current_time)
3281 {
3282 uint64_t transferred, transferred_pages, time_spent;
3283 uint64_t current_bytes; /* bytes transferred since the beginning */
3284 double bandwidth;
3285
3286 if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3287 return;
3288 }
3289
3290 current_bytes = migration_total_bytes(s);
3291 transferred = current_bytes - s->iteration_initial_bytes;
3292 time_spent = current_time - s->iteration_start_time;
3293 bandwidth = (double)transferred / time_spent;
3294 s->threshold_size = bandwidth * s->parameters.downtime_limit;
3295
3296 s->mbps = (((double) transferred * 8.0) /
3297 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3298
3299 transferred_pages = ram_get_total_transferred_pages() -
3300 s->iteration_initial_pages;
3301 s->pages_per_second = (double) transferred_pages /
3302 (((double) time_spent / 1000.0));
3303
3304 /*
3305 * if we haven't sent anything, we don't want to
3306 * recalculate. 10000 is a small enough number for our purposes
3307 */
3308 if (ram_counters.dirty_pages_rate && transferred > 10000) {
3309 s->expected_downtime = ram_counters.remaining / bandwidth;
3310 }
3311
3312 qemu_file_reset_rate_limit(s->to_dst_file);
3313
3314 update_iteration_initial_status(s);
3315
3316 trace_migrate_transferred(transferred, time_spent,
3317 bandwidth, s->threshold_size);
3318 }
3319
3320 /* Migration thread iteration status */
3321 typedef enum {
3322 MIG_ITERATE_RESUME, /* Resume current iteration */
3323 MIG_ITERATE_SKIP, /* Skip current iteration */
3324 MIG_ITERATE_BREAK, /* Break the loop */
3325 } MigIterateState;
3326
3327 /*
3328 * Return true if continue to the next iteration directly, false
3329 * otherwise.
3330 */
3331 static MigIterateState migration_iteration_run(MigrationState *s)
3332 {
3333 uint64_t pending_size, pend_pre, pend_compat, pend_post;
3334 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3335
3336 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3337 &pend_compat, &pend_post);
3338 pending_size = pend_pre + pend_compat + pend_post;
3339
3340 trace_migrate_pending(pending_size, s->threshold_size,
3341 pend_pre, pend_compat, pend_post);
3342
3343 if (pending_size && pending_size >= s->threshold_size) {
3344 /* Still a significant amount to transfer */
3345 if (!in_postcopy && pend_pre <= s->threshold_size &&
3346 qatomic_read(&s->start_postcopy)) {
3347 if (postcopy_start(s)) {
3348 error_report("%s: postcopy failed to start", __func__);
3349 }
3350 return MIG_ITERATE_SKIP;
3351 }
3352 /* Just another iteration step */
3353 qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
3354 } else {
3355 trace_migration_thread_low_pending(pending_size);
3356 migration_completion(s);
3357 return MIG_ITERATE_BREAK;
3358 }
3359
3360 return MIG_ITERATE_RESUME;
3361 }
3362
3363 static void migration_iteration_finish(MigrationState *s)
3364 {
3365 /* If we enabled cpu throttling for auto-converge, turn it off. */
3366 cpu_throttle_stop();
3367
3368 qemu_mutex_lock_iothread();
3369 switch (s->state) {
3370 case MIGRATION_STATUS_COMPLETED:
3371 migration_calculate_complete(s);
3372 runstate_set(RUN_STATE_POSTMIGRATE);
3373 break;
3374
3375 case MIGRATION_STATUS_ACTIVE:
3376 /*
3377 * We should really assert here, but since it's during
3378 * migration, let's try to reduce the usage of assertions.
3379 */
3380 if (!migrate_colo_enabled()) {
3381 error_report("%s: critical error: calling COLO code without "
3382 "COLO enabled", __func__);
3383 }
3384 migrate_start_colo_process(s);
3385 /*
3386 * Fixme: we will run VM in COLO no matter its old running state.
3387 * After exited COLO, we will keep running.
3388 */
3389 s->vm_was_running = true;
3390 /* Fallthrough */
3391 case MIGRATION_STATUS_FAILED:
3392 case MIGRATION_STATUS_CANCELLED:
3393 case MIGRATION_STATUS_CANCELLING:
3394 if (s->vm_was_running) {
3395 vm_start();
3396 } else {
3397 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3398 runstate_set(RUN_STATE_POSTMIGRATE);
3399 }
3400 }
3401 break;
3402
3403 default:
3404 /* Should not reach here, but if so, forgive the VM. */
3405 error_report("%s: Unknown ending state %d", __func__, s->state);
3406 break;
3407 }
3408 migrate_fd_cleanup_schedule(s);
3409 qemu_mutex_unlock_iothread();
3410 }
3411
3412 void migration_make_urgent_request(void)
3413 {
3414 qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3415 }
3416
3417 void migration_consume_urgent_request(void)
3418 {
3419 qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3420 }
3421
3422 /* Returns true if the rate limiting was broken by an urgent request */
3423 bool migration_rate_limit(void)
3424 {
3425 int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3426 MigrationState *s = migrate_get_current();
3427
3428 bool urgent = false;
3429 migration_update_counters(s, now);
3430 if (qemu_file_rate_limit(s->to_dst_file)) {
3431
3432 if (qemu_file_get_error(s->to_dst_file)) {
3433 return false;
3434 }
3435 /*
3436 * Wait for a delay to do rate limiting OR
3437 * something urgent to post the semaphore.
3438 */
3439 int ms = s->iteration_start_time + BUFFER_DELAY - now;
3440 trace_migration_rate_limit_pre(ms);
3441 if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3442 /*